138953821 an-empirical-study-of-stock-market

29
Homework Help https://www.homeworkping.com/ Research Paper help https://www.homeworkping.com/ Online Tutoring https://www.homeworkping.com/ click here for freelancing tutoring sites An Empirical Study of Macroeconomic Factors and Stock Market: An Indian Perspective Saurabh Yadav EDHEC Business School Master's in Risk and Investment Management [email protected] June 26, 2012 EDHEC Business School Abstract This thesis is an empirical study of relationship between Indian stock markets and macro economy. There is a huge literature about such kind of empirical studies but mostly on US/UK stock markets and macroeconomic indicators. This study is similar to many of the earlier studies in some aspects, so it uses econometric tools used in earlier studies but at the same time this study di_erentiates itself from other studies in the sense it uses Indian markets and macroeconomic data for analysing the relationship and it also tries to analyse the impact of global economy on the Indian

Transcript of 138953821 an-empirical-study-of-stock-market

Homework Help

httpswwwhomeworkpingcom

Research Paper help

httpswwwhomeworkpingcom

Online Tutoring

httpswwwhomeworkpingcom

click here for freelancing tutoring sites

An Empirical Study ofMacroeconomic Factors and StockMarket An Indian PerspectiveSaurabh YadavEDHEC Business SchoolMasters in Risk and Investment ManagementsaurabhyadavedheccomJune 26 2012EDHEC Business SchoolAbstractThis thesis is an empirical study of relationship between Indian stockmarkets and macro economy There is a huge literature about such kind ofempirical studies but mostly on USUK stock markets and macroeconomicindicators This study is similar to many of the earlier studies in someaspects so it uses econometric tools used in earlier studies but at the sametime this study di_erentiates itself from other studies in the sense it usesIndian markets and macroeconomic data for analysing the relationshipand it also tries to analyse the impact of global economy on the Indianmarkets The period that will be used for the study will be from 1990to 2011 We have chosen this period as it represents big regulatory andstructural changes in Indian economy So an analysis of this period canprovide us with insights to how some regulatory and structural changesimpact the economy and asset prices in that country In this study we willuse Unit root tests cointegration Ljung-Box Q test and multivariate VARanalysis for analysing each macro economic and asset prices time seriesindividually and to build a model that can analyse the impact of one overthe other Also we will conduct Grangers Causality test and Impulseresponse analysis between Stock market and macro economic indicatorsto analyze the impact of macro economic newsshocks on India Stock

index (BSE)EDHEC Business School 4AcknowledgmentI am thankful to Professor Robert Kimmel for his comments and guidanceon the subject He has been a constant source of inspiration and a good men-tor from whom I learned a lot I am also grateful to Stoyan Stoyanov MarcRakotomalala Aishwarya Iyer Wen lei Lixia Loh for some great insights intothe subject Their timely comments and suggestions on empirical tests helpedme improve the statistical signi_cance of my tests I thank EDHEC Risk In-stitute for allowing me to use their resources to get the data from various dataproviders In the end ill like to thank my parents and my sister for constantsupport and motivation without which it would have been impossible to climbthis arduous pathRegardsSaurabh YADAVEDHEC Business School 5CONTENTSContents1 Introduction 72 Literature Review 93 Data 1331 Description of Macroeconomic Indicators 1332 Description of Stock Market Indices 144 Methodology 1541 Construction of Time Series 1542 Unit Root Test and Stationarity 15421 Mathematical representation of Stationary series and unitroot test 16422 Augmented Dickey Fuller Unit Root Test 1743 Testing Long Term Relationships 18431 Johansen test for Cointegration 1844 Impulse Response 195 Results 216 Conclusions 247 Graphs and Tables 2571 Graphs of Time series 2572 Graphs of Time Series - Di_erenced 2973 Correlograms of Time series 3374 Tables 40741 Table for Unit root test of Time series 40742 Tables for Unit root test of Di_erenced time series 40743 Tables for Residual based test of cointegration 40744 Johansen cointegration test 43745 Impulse response tests 46746 Granger causality test between IP and BSE 498 Bibliography 50EDHEC Business School 61 INTRODUCTION1 IntroductionIn the past few decades there has been a growing interest among academiciansand practitioners about the relationship between macroeconomic variables andasset prices mainly stocks and house prices In a good and expanding economyprices of stocks are supposed to increase as there is an increase in expectationof large future cash ows pro_ts for the companies and various role playersin the economy Similarly during a bad or downward spiralling economy theexpectation of large future cash ows and pro_ts decrease and consequently theprice of stocks decreaseStock markets are representative of economy of a country and investors belief

They are able to capture macro economic movements in the economy as well asidiosyncratic factors related to each company or industry As Stock prices arereal time and are more frequent than macroeconomic releases they are betterreector of changes in domestic and global economy and can predict the move-ment of macroeconomic indicators In other words stock markets are a leadingindicator of the economyMarkets respond to di_erent macroeconomic indicators in di_erent ways Theresponse of Stock markets to any macroeconomic news is dependent on how thenews will e_ect the pro_ts and interest rates The price of the stock accordingto the Discounted Cash Flow formula isPt =Div1(1 + r1)1 +Div2(1 + r2)2 + +Divt(1 + rt)t (1)As both dividends and interest rates enter into the formula for value of astock the reaction of stock price to a macro news will depend on how the newse_ect the discounting factor ( Interest rates ) and future pro_ts of the com-panies Macro economic factors that project brighter times and more pro_tsfor the companies like increasing Industrial production Increasing M1 moneysupply good consumer con_dence levels will have a positive e_ect on the stockprices Whereas macro news that point to economic recession or slow growthlike decreasing Industrial production coupled with Rising interest rates Risein ination rise in unemployment etc will have a downward e_ect on stockpricesFirst people to do an empirical study on this subject were Eugene Fama andKenneth French In their 1981 paper Stock returns Real activity Inationand money they analysed the relationship between stock returns real activityination and money supply using macro economic data After that study therehas been a barrage of studies on relationship between stock returns and macroeconomic factors based on US and UK data Another important paper pub-lished on this research was by ChenRoll and Ross (1986) who analysed whetherinnovations in the macroeconomic variables are risks that are awarded in thestock markets They found that macroeconomic variables like spread betweenlong and short interest rates expected and unexpected ination Industrial pro-duction are some of the factors that are awarded by the markets Further theArbitrage pricing theory (APT) of Ross (1976) posits relation between stockprices and certain macro-economic variables In the last decade or so the focusfor these kind of studies have started to shift from developed world economies todeveloping world economies As developing world economies have shown signsEDHEC Business School 71 INTRODUCTIONof huge growth potential and leading the economies globally out of recessionsthis motivates us to research on developing markets like India Such a studywill help us to _nd the relation between stock market and macroeconomic indi-cators and give a new insight to foreign investors academicianspolicy makerstraders and domestic investorsThis study is important in a sense it provides an insight to how are Indian stockmarkets are related to its macroeconomic variables and global macromicro eco-nomic factors This study will also help us in analysing whether the Indian stockmarkets have become coupled to global factors or are they still dominated bydomestic economic factorsThe focus of this study is on relation between Indian stock market representedby BSE Sensex and domestic macroeconomic factors and global factors repre-sented by Standard and Poors 500 Index This study builds on earlier studiesdone in this area but also open some new doors for further research It is sim-ilar to some earlier studies in a respect that it uses data macro and micro

factors and econometrics tools used in previous studies but at the same time itdi_erentiates itself from earlier studies in a sense that it is done on a marketthat is still developing Also the time period used in the analysis is a periodwhere Indian market has undergone lot of regulatory changes that has createda structural change in the market Further in this study Ill analyse whetherthe Indian markets are driven mainly by Domestic factors or do global factorshave more inuence on Indian markets To analyse the impact of internationalfactors Ill use Standard and Poors 500 Index and USDINR exchange rate as asubstitute of global factors and to model domestic demand Ill use macro factorslike Industrial production M1 money supply Consumer Price Index and Pro-ducer price Index The outline of the thesis is as followings Section 2 providesa literature review of the studies done earlier in this area Section 3 provides adetailed description of the data used in the study Section 4 provides a detaileddescription of the methodology and various econometric tools that will be usedin the study Section 5 provides the results of the study and Section 6 providesthe conclusion of the studyEDHEC Business School 82 LITERATURE REVIEW2 Literature ReviewMany studies and researchers have tried to _nd factors that can explain stockreturns The most famous and earliest model is the Capital Asset Pricing Model(CAPM) developed by Sharpe (1964) Lintner (1965) Mossin (1967) and Black(1972) The concept of this single factor model is developed from diversi_-cation introduced by Markowitz (1952) In CAPM model the expected stockreturns can be explained with the help of Risk free rate and one risk factorMarket CAPM says that the systematic risk can be captured by sensitivenessof each stock to change in overall market which is measured by Beta Accordingto CAPM the market factor is the only factor determining the stock returnsCAPM was a revolutionary model It changed the way people looked at thestock returns as something that is vary arbitrary As it is very easy to under-stand and use CAPM is very popular as the model used to determine the stockreturn in most of _nance textbooks and used by many practitioners in stockmarketHowever the numerous set of assumptions made in deriving CAPM made itinconsistent with the real world and led to criticism of CAPM To overcomethe limitations and assumptions made in CAPM many scholars came up withmulti- factor models like Fama-French three factor model APT model etc InFama-French model they try to explain stock returns with help of three factorsmarketsmall minus big and value minus growth the model was able to explainthe returns based on these risk factors for some time before it failed Therehave been many studies on failure of Fama-French model and markets where itis not applicableThe macroeconomic models of explaining stock returns started with APT (Ar-bitrage Pricing Theory) by Ross (1976) which was later re_ned by Roll andRoss (1980) APT is a multi-factor model and claims that the stock return canbe explained by unexpected changes or shocks in multiple factors ChenRolland Ross (1986) perform the empirical study for APT model and identify thatsurprise or shock in macroeconomic variables can explain the stock return sig-ni_cantly The variables used in their study are Industrial production indexdefault risk premium that can measure the con_dence of investors and changein yield curve that can be measured by term premiumThe study of macroeconomic factors in explaining stock returns have been pop-ular since then Stock price is present value of all discounted future cash owsIf a _rm is performing well then the expectation of large future cash ows risesand consequently the stock price rises On the other hand if a _rm is performingbad for couple of years then the expectation of big future cash ows decreaseand in turn the stock price fall This is a micro and idiosyncratic explanation ofstock prices and returns But the future cash ows of a stock does not dependsolely on the companys performance or pro_tsloss The systematic factor can

have a huge impact on the cash ows of not only one but many companies Thesystematic factor here refers to macro economic variables The state of Macroeconomic conditions lead to changes in Monetary and regulatory policies by thegovernment and which in turn a_ects the stock prices For example a countrywith good economic conditions represented by its Industrial production indexGDP CPI Interest rates will create an environment that is conducive for thegrowth of companies by lowering borrowing rates and other open market opera-tions So all macroeconomic factors that can inuence future cash ows or theEDHEC Business School 92 LITERATURE REVIEWdiscount rate by which the cash ows are discounted should have an inuenceon the stock priceMany researcher have studies the relationship between stock prices and macroeconomic variables and tried to explain the a_ect of one over the other Fama(1981) tries to establish a relationship between stock returns real activity ina-tion and money In his paper he _nds that Stock returns have positive relationwith real output and money supply but a negative relation with ination Heexplains that negative relation between stock returns and ination is induced bynegative relation between real output approximated by Industrial productionand ination This negative relationship between ination and real activityis explained by money demand theory and quantity theory of money Fama(1990) explains that measuring the total return variation explained by shocksto expected cash ows time-varying expected returns and shocks to expectedreturns is one way to judge the rationality of stock prices In his paper he_nds that growth rates of production used to proxy for shocks to expected cashows explain 45 of return variance ChenRoll and Ross (1986) explored therelationship between a set of economic variables and their systematic inuenceon stock market returns They found that Industrial production changes inrisk premium twists in yield curve had strong relationship and impact on stockreturns A somewhat weaker e_ect was found for measures of unanticipatedination and changes in expected ination during periods when these variableswere highly volatile They concluded that stock returns were exposed to sys-tematic economic news that they are priced in accordance to their exposuresand that the news can be measured as innovation in state variables Chen(1991) found that state variables that are priced are those that can forecastchanges in the investment and consumption opportunity set According to hisresearch default spread the term spread the one-month T-Bill rate the laggedindustrial production growth rate and the dividend-price ration are importantdeterminants of future stock market returns Bulmash and Trivoli (1991) showthe e_ect of business cycle movements on the relationship between stock returnsand money growthAn interesting paper in this _eld of research is by Fama (1990) and Schwert(1990) In the paper they claim that there are three explanations for the stronglink between stock prices and real economic activityFirst information about the future real activity may be reected in stockprices well before it occurs|this is essentially the notion that stock pricesare a leading indicator for the well-being of the economy Second changesin discount rates may a_ect stock prices and real investment similarly butthe output from real investment doesnt appear for some time after it ismade Third changes in stock prices are changes in wealth and thiscan a_ect the demand for consumption and investment goods [Schwert(1990)p1237]Campbell and Ammer (1993) use a VAR approach to model the simulta-neous interactions between the stock and bond markets since most previousworks do not address the channels through which the macroeconomic activityinuences the stock prices One example could be that industrial productioncould be linked to changing expectations of future cash ows (Balvers at al1990) On the other hand interest rate innovations could be the driving factorEDHEC Business School 102 LITERATURE REVIEW

in determining both industrial production (due to change in investment) andstock prices (due to change in the discounted present value of future cash ows)A VAR analysis can distinguish these possibilities Mukherjee and Naka (1995)show a long-term relationship between the Japanese stock price and real macroe-conomic variables Dr Nishat (2004) studies the long term association amongmacroeconomic variables like money supply CPIIPI and foreign exchange rateand stock markets in Pakistan The results show that there are causal relation-ship among the stock price and macroeconomic variables He uses data from1974 to 2004 in his study As most of the _nancial time series are non station-ary in levels he uses unit root technique to make data stationary Fazal Hussianand Tariq Massod (2001) used variables like investment GDP and consumptionemploying Grangers causality test to _nd relationship between macro factorsand stock markets They show that at two lags all macroeconomic variableshave highly signi_cant e_ect on stock prices James et al (1985) use a VARMAanalysis for investigating relationship between macro economy and stock mar-ket Using VARMA analysis for _nding causal relationship between factors isa better technique as the procedure does not preclude any causal structure apriori since it allows feedback among variables Thus the VARMA approachallow whatever causal relationship exist to emerge from the data They _ndlinkages between real activity and stock returns and real activity and inationAlso they _nd that stock returns signal changes in the monetary base Sincestock returns also signal changes in expected real activity this suggests a linkbetween the money supply and expected real activity that is consistent with themoney supply explanation o_ered by Geske and RollIn recent years the focus of these kind of studies have shifted from developedeconomies to developing economies As developing economies are the economiesthat see a lot of structural and monetary policy changes an analysis of relation-ship between macro and micro can provide new insights Also one can analysethe e_ects of monetary policies on the asset prices especially on stock pricesTangjitprom (2012) study of macroeconomic factors like unemployment rateinterest rate ination rate and exchange rate and stock market of Thailand con-cludes that macroeconomic factors signi_cantly explain stock returns He also_nds that for Thailand unemployment rate and ination rate are insigni_cant todetermine the stock returns The reason he provides is that the unemploymentrate and ination rate are not timely and there could be some lags before thedata becomes available Also Grangers test to examine lead-lag relationshipamong the factors reveal that only few macroeconomic variables could predictthe future stock returns whereas the stock returns can predict most of futuremacro economic variables This implies that performance of stock markets canbe a leading indicator for future macroeconomic conditions Ali (2011) study ofimpact of macro and micro factors on stock returns reveals that ination andforeign remittance have negative inuence and industrial production index havepositive impact on stock markets Also he didnt found any Grangers Causal-ity between stock markets and any of the explanatory variables This lack ofGrangers causality reveals the evidence of informationally ine_cient marketsAli uses a multivariate regression analysis on standard OLD formula for estimat-ing the relationship Hosseini et al (2011) tested the relationship between stockmarkets and four macro economic variables namely crude oil prices Money sup-ply Industrial production and ination rate in China and India They used aperiod of 1999 to 2009 for analysis As most of the economic time series have unitEDHEC Business School 112 LITERATURE REVIEWroot they _rst used the Augmented Dickey Fuller unit root test and found theunderlying series to be non-stationary at levels but stationary after in di_erenceAlso the use of Jhonson-Juselius (1990) Multivariate cointegration and VectorError Correction model technique indicate that there are both long and shortrun linkages between macroeconomic variable and stock market index in each ofthe two countries Their analysis shows that in long run the impact of increasein prices of crude oil for China is positive but for India is negative In terms

of money supply the impact on Indian stock market is negative but for Chinathere is a positive impact The e_ect of Industrial production is negative onlyin China In addition the e_ect of increases in ination on these stock marketsis positive in both countries Wickremasinghe (2006) analysed the relationshipbetween stock prices and macroeconomic variables in Sri Lanka He used theUnit root tests Jhonsons test Error-correction model variance decomposi-tion and impulse response to analyse the relationships His _ndings indicatethat there is both long term and short term causal relationship between stockprices and macroeconomic variables in Sri Lanka The result indicate that thestock prices can be predicted from certain macroeconomic variables and henceviolate the validity of the semi-strong version of e_cient market hypothesisAhmed (2008) investigates the causal relationship between Indian macroeco-nomic factors like Industrial Production Exports Foreign direct investmentMoney supply exchange rate interest rate and stock market indices NSE NiftyIndex and BSE Sensex For _nding the long term relationship he applies Jo-hansens cointegration and Toda and Yamamoto Granger Causality tests Foranalysing the Impulse response and variance decomposition he uses bivariateVAR His _ndings reveal that stock prices in India lead macroeconomic activityexcept movement in interest rate Interest rate seem to lead the stock priceThe study also reveals that movement of stock prices is not only the outcomeof behaviour of key macro economic variables but it is also one of the causesof movement in other macro dimensions in the economy An important paperby Bilson et al (2001) argues that emerging markets local factors are moreimportant than global factors They _nd that for emerging markets are at leastpartially segmented from global capital markets The global factors are proxiedby world market returns and local factors by set of macro economic variableslike money supply prices real activity and exchange rate Some evidence isfound that local factors are signi_cant in their association with emerging equitymarket returns above than that explained by the world factor When they usea larger set of variables the explanatory power of the model improves substan-tially such that they are able to explain a large amount of return variation formost emerging marketsEDHEC Business School 123 DATA3 Data31 Description of Macroeconomic IndicatorsOne of the biggest problems when conducting a research with macroeconomicdata is the frequency of the data Most of the macroeconomic indicator timeseries are yearlyquarterly or monthly time series This low frequency of themacroeconomic indicators results in very few data points for conducting a anal-ysis that is robust A possible cure for the problem is to use longer time periodsto incorporate more data points for macroeconomic variables But anotherproblem that we face when we look at the macroeconomic indicators for Asiancountries is reporting of the data For most of the Asian countries the macroe-conomic data doesnt have a long history and same can be said about historyof Indian macroeconomic variables So in this research we have used a timeperiod for which we can _nd data for most of the macroeconomic indicators Inthis paper we use a time period of 20 years starting from 1990 to 2011 Thistime period in Indian economy is representative of many structural and mone-tary policy changes like liberalization of India markets Also as the time periodis long it gives us enough data point for each macroeconomic factors to do arobust empirical analysisWhen one starts to build a model of interaction between macro and micro eco-nomic factors one dominant and important question one faces is among themyriad of macro indicators available for an economy which factors to chooseto incorporate in the model If one chooses macroeconomic factors that arehighly correlated among themselves then the power of test results decrease asit may result in a model where the macro indicators are able to explain mostof the movement of micro factors but the macro factors may not be relevant

To circumvent this problem we use variables that have been tested in earlierresearches and that have been proven to have e_ect on stock markets I alsotest a few macro factors that have some _nancial theory behind them that con-nect them to stock markets Ali (2011) Wickremasinghe (2006) Bilson etal(2001) and Bailey (1996) _nd that Industrial production CPI exchange rateM1 money supply GDP are few of the macro economic factors that can signi_-cantly explain stock returns Sahu(2011) Ahmed(2008) Tripathy(2011) studyon Indian markets speci_cally show that Industrial Production Exchange rateInation index are macro economic indicators that have a strong positive ornegative relationship with the stock markets So in our study we test 5 macroeconomic variables namely M1 money supply Consumer and Producer price In-dex Industrial production Exchange rate The time period for these indicatorsis from 1990-2011 The data for Ination indices Industrial production andexchange rate has been pulled from Bloombergc and Datastreamc The datahas been processed for errors and missing values Data for M1 money supplyhas been pulled from RBI website For most of the indices like ination andIndustrial production index the base year has been changed to 1990 Also assome of the indices are in levels and some in actual _gures (M1 money supply)we convert all of the indicators to level form (starting at 100 in 1990)EDHEC Business School 133 DATA32 Description of Stock Market IndicesCompared to Macro Indicators stock market data is relatively easy to _nd andhas considerably long history Also the stock market data is a real time data soit has a very high frequency of seconds Here in our analysis we will make use ofBSE (Bombay Stock Exchange) as representation of Indian markets and SP500(Standard and Poors 500 Index) as representation of global factors BSE is amarket cap-weighted of 30 stocks It is the oldest Index in the Asian markets(established in 1875) and have had a long history We choose this index as it isthe Index that represent the most liquid and traded stocks of the Indian stockmarket Also the index is most traded index in India and a good representationof trade prices of the stocks Even in terms of an orderly growth much beforethe actual legislations were enacted BSE Limited had formulated a compre-hensive set of Rules and Regulations for the securities market It had also laiddown best practices which were adopted subsequently by 23 stock exchangeswhich were set up after India gained its independence Our choice of SP500 isbased on the fact that it has a long history and many researchers have usedthis index as a good proxy representation of global markets and economic con-ditions We will take the monthly returns of each of the indices from 1990-2011in accordance with data frequency of macro economic variables Also as theindices have di_erent levels at beginning of 1990 we rebase both the indices tobase year of 1990 starting at a level of 100EDHEC Business School 144 METHODOLOGY4 Methodology41 Construction of Time SeriesThe _rst step in constructing an econometric model is constructing time seriesall of which are in same units Most of the time series used in our analysis are indi_erent formats For example CPI PPI BSE Index SP500 are in levels M1money supply USDINR exchange rate is in absolute current format Industrialproduction is in absolute production levels So _rst we convert all of the giventime series to level The way we construct time series in levels is _rstly takingthe initial data point of each time series as 100 We then _nd the percentagechange from one period to the next one for each time series using a continuouscompounding assumption (taking a natural log of change in values) In math-ematical terms it can be stated as Assume the original Index value at time tto be It and at time t+1 to be It + 1 Then we can compute the new rebasedindex by formulaRIt+1 = RIt _ (1 + ln(It+1=It))

whereRIt= Rebased Index at time tRIt+1=Rebased Index at time t+1We can use these rebased indices in building and testing our econometric model42 Unit Root Test and StationarityUnit root test is to _nd whether the series is stationary or non-stationary Astrictly stationary process is one where for any t1 t2 tt 2Z any k 2Z andT=12Fyt1 yt2 yt3 ytT

(y1 yT ) = Fyt1+k yt2+k yt3+k ytT+k

(y1 yT )where F represents joint distribution function of the set of random variablesIt can also be stated that the probability measure of sequence of yt is same asyt+k for all k In other words a series is stationary if the distribution of its valueremain the same as time progresses Similar to the concept of strict stationaryis weakly stationary process A weakly stationary process is one which has aconstant mean variance and autocovariance structure Stationary is a necessarycondition for a time series to be tested in regression A non-stationary seriescan have several problems like1 The shocks given to the series would not die of gradually resulting inincrease of variance as time passes2 If the series is non stationary then it can lead to spurious regressions If twoseries are generated independent of each other then if one is regressed onother it will result in very low R2 values But if two series are trending overtime then a regression of one over the other will give high R2 even thoughthe series may be unrelated to each other So if normal regressions toolsEDHEC Business School 154 METHODOLOGYare used on non stationary data then it may result in good but valuelessresults3 If the variables employed in a regression model are not stationary thenit can be proved that the standard assumptions for asymptotic analysiswill not be valid In other words the usual t-ratios will not follow at-distribution and the F-statistic will not follow an F-distribution and soonStationarity is a desirable condition for any time series so that it can be usedin regressions and give meaningful result that have some value to test for sta-tionarity a quick and dirty way is looking at the autocorrelation and partialcorrelation function of the series If the series is stationary then the autocorre-lation function should die o_ gradually after few lags and the partial correlationfunction will me non zero for some lags and zero thereafter Also we can usethe Ljung-Box test for testing that all m of _k autocorrelation coe_cients arezero using Q-statistic given by formulaQ = T(T + 2)_mk=1_k2T 1048576 k_ _2where T = Sample size and m = Maximum lag lengthThe lag length selection can be based on di_erent Information Criteria likeAkaikes Information criteria (AIC) Schwarzs Bayesian information criteria(SBIC) Hannan-Quinn criterion (HQIC) Mathematically di_erent criteria arerepresented asAIC = ln(_2) + 2kTSBIC = ln(_2) + kT lnTHQIC = ln(_2) + 2kT ln(ln(T))

For a better test for stationarity we use augmented Dickey fuller Unit roottest on each time series separately Augmented Dickey Fuller test is test ofnull hypothesis that the time series contains a unit roots against a alternativehypothesis that the series is stationary421 Mathematical representation of Stationary series and unit roottestAssume a variable Y whose structure can be given by AR process with no driftequationyt = _1yt10485761 + _2yt10485762 + _3yt10485763 + + _nyt1048576n + ut (2)where ut is the residual at time t Using a Lag operator L we can write eq(1)asyt = _1L1yt + _2L2yt + _3L3yt + + _nLnyt + ut (3)EDHEC Business School 164 METHODOLOGYRearranging eqn (2) we getyt 1048576 _1L1yt 1048576 _2L2yt 1048576 _3L3yt + 1048576 _nLnyt = ut (4)yt(1 1048576 _1L1 1048576 _2L2 1048576 _3L3 + 1048576 _nLn) = ut (5)or_(L)yt = ut (6)The time series is stationary if we can write eqn(5) in formyt = _(L)10485761ut (7)with _(L)10485761 converging to zero It means the autocorrelation function woulddecline as lag length is increased If eqn (6) is expanded to a MA(1) processthe coe_cients of residuals should decrease such that the the residuals that thee_ect of residuals decrease with increase in lags SO if the process is stationarythe coe_cients of residuals will converge to zero and for non-stationary seriesthey will and converge to zero and will have long term e_ect The condition fortesting of unit root for an AR process is that the roots of eqn(6) or Charac-teristic equation should lie outside unit circle422 Augmented Dickey Fuller Unit Root TestConsider an AR(1) process of variable Yyt = _yt10485761 + ut (8)Subtracting yt10485761 from both sides of eqn(7) we get_y = (_ 1048576 1)yt10485761 + ut (9)Eqn(8) is the test equation for Dickey Fuller test For Dickey-Fuller Unit roottestNull Hypothesis The value of _ is equal to 1 or value of _10485761 is equal to 0 vsAlternate Hypothesis The value of _ is less than one or value of _ 1048576 1 is lessthan zero Augmented Dickey-Fuller test is similar to normal Dickey-Fuller testsexcept it takes the lag structure of more than one into account_y = yt10485761 +Xpi=1_i_yt1048576i + ut (10)If the series has one or more unit root it is said to be integrated of order nwhere n is the number of unit roots of the characteristic equation To makethese time series stationary they needs to be di_erenced Mathematically ifyt _ I (n) (11)then_(d) yt _ I (0) (12)To make our time-series stationary we will use the natural log returns of theseseries in the analysisEDHEC Business School 174 METHODOLOGY43 Testing Long Term RelationshipsEngle and Granger (1987) in their seminal paper described cointegration whichforms the basis for testing for long term relationship between variables Accord-ing to Engle and Granger two variables are cointegrated if they are integratedprocess in their natural form (of the same order) but a weighted combination

of the variables can be found such that the combined new variable is integratedof order less than the order of individual time series Mathematically assumeyt to be a k X 1 vector of variables then the components are cointegrated orintegrated of order (db) if1 All components of yt are I(d)2 There is at least one vector of coe_cients _ such that_0

yt _ I (d 1048576 b) (13)As most of the _nancial time series are integrated of order one we will restrictourselves to case d=b=1 Two or more variables are said to be cointegrated ifthere exist a linear combination of these variables that is stationary Many ofthe series are non-stationary but move together over time which implies twoseries are bound by some common force or factor in long run We will test forcointegration by a residual-based approach and Johansens VAR methodResidual Based approach Consider a modelyt = _1 + _2x2t + _3x3t + + ut (14)where yt x2t x3t are all integrated of order N Now if the residual of this re-gression ut is stationary then we can say that the variables are cointegrated elsethere exist no long term relationship between the variables To test the resid-ual for stationarity we will run Augmented Dickey-Fuller tests on the residualsUnder the Null hypothesis the residual are integrated of order one or more andunder alternate hypothesis the residuals are I(0)431 Johansen test for CointegrationJohansen test for cointegration presents a better model for testing multiplecointegration among multiple variables The Residual based approach can only_nd atmost one cointegration and can be tested for a model with two variablesEven if more than two variables are present in the equation that are cointegratedthe Residual based approach will give only one cointegration SO we will useJhoansen VAR based cointegration for testing more than one cointegrationSuppose that a set of g variables are under consideration that are I(1) andwhich are thought to be cointegrated A VAR with k lags containing thesevariables could be set upyt = _1yt10485761 + _2yt10485762 + _ _ _ + _kyt1048576k + ut (15)g _ 1 g _ g g _ 1 g _ g g _ 1 g _ g g _ 1 g _ 1EDHEC Business School 184 METHODOLOGYIn order to use the Johansen test the VAR above should be turned into avector error correction model of form_yt = _yt1048576k + 1_yt10485761 + 2_yt10485762 + _ _ _ + k10485761_yt1048576(k10485761) + ut (16)where _ = (_ki=1_i) 1048576 Ig and i = (_ij=1_j) 1048576 IgThe Johansens test centers around testing the _ matrix which is the matrixthat represents the long term cointegration between the variables The test fornumber of cointegration is calculated by looking at the rank of the _ matrixthrough its eigenvalues The rank of the matrix is equal to number of roots(eigenvalues) _i of the matrix that are di_erent from zero The roots should beless than 1 in absolute value and positive If the variables are not cointegratedthe rank of the matrix will not be signi_cantly di_erent from zero ie _i _ 0There are two test statistics for Johansen test _tracer and _max_trace (r) = 1048576TPgi=r+1 ln(1 1048576 _ _i)and_max(r r + 1) = 1048576Tln(1 1048576 _r_+1)_trace is a test statistic for joint test where the null hypothesis is that thenumber of cointegration vector is less than or equal to r against an alternativethat there are more than r_max conducts another separate test on eigenvalues and has null hypothesis that

the number of cointegrating vector is r against r+144 Impulse ResponseOnce we have determined whether the variables have long term relationship ornot we can form a multivariate VAR model for the variables A multivariateVAR model between g variables is a model where the current value of a variabledepend on di_ernt combinations of the previous k values of all the variables anderror terms A general representation of the model can beyBSEt = _ + _BSEyBSE + _IP yIP + CPIyCPI + _M1yM1 + _SP500ySP500 + u1t(17)where all the coe_cients except _ are g _ k matrices and all variables y are k_ 1 matricesOnce we have formed a model like this we can use the model for Impulse re-sponse A VAR(p) model can be written as a linear fuction of the past innova-tions that isrt = _ + at + 1at10485761 + 2at10485762 + (18)where _ = [_(1)]10485761_0 provided that the inverse exists and the coe_cient ma-trices i can be obtained by equating the coe_cients of Bi in the equation(I 1048576 _1B 1048576 1048576 _PBP )(I + 1B + 2B2 + ) = I (19)EDHEC Business School 194 METHODOLOGYwhere I is the Identity martix This is a moving average representation of rtwith the coe_cient matrix i being the impact of the past innovation at1048576i onrt Equivalently i is the e_ect of at on the future observation rt+i Therefore i is often referred to as the Impulse Response Function of rt For our impulseresponse we will use equation of variables in _rst di_ernce form like_BSEt = _t +Xki=0_11(i)_BSEt1048576i +Xkj=1_12(j)_MIt1048576j + _BSEt (20)_MIt = _t +Xki=0_21(i)_MIt1048576i +Xkj=1_22(j)_BSEt1048576j + _MIt (21)Grangers causality and Blocks F test of a VAR model will suggest which ofthe variables have statistically signi_cant impacts on the future values of othervariables in the system But F-test results cannot explain the sign of the re-lationship nor how long these e_ects require to take place Such informationwill however be given by an examination of the VARs impulse responses andvariance decompositions Impulse response is a technique that trace out theresponsiveness of the dependent variable in the VAR to shocks of each of theother variables So for each variable from each equation separately we will applya unit shock to the error and trace the e_ects upon the VAR system over timeBy using the impulse response technique we can determine how responsive isthe BSE stock index to Indian macro indicators and SP500 This will help usdetermine whether the BSE index is more reactive to domestic news or globalnewsEDHEC Business School 205 RESULTS5 ResultsBefore we use the time series for VAR analysis or cointegration tests we need todetermine whether the series are Stationary or not If the series are stationaryin levels we can use them directly else we need to use the di_erenced time seriesOne way to look for autocorrelation or integrated process is to see the graphsof the various time series used Section 71 shows the graphs of variables we

use for our analysis As we can see from the graphs all of the time series havea trend in long run which points to an integrated process As a second stepwe plot the graphs of di_erenced time series in Section 52 We can see thatthe di_erenced graphs in Section 72 dont show a long term trend and crossthe X-axis frequently This is usually a property of I(1) processes So we checkthe series for autocorrelations at di_erent lag lengths Section 73 shows cor-relograms graph autocorrelation coe_cient partial autocorrelation coe_cientQ-Stat and p-value for various time series up to 36 lags As can be seen in thetables the Q-stat for all lags is zero and we can reject the joint null hypothesisthat all the autocorrelations up to 36 lags are zero Table 741 shows that ifwe conduct a Unit root test on levels of the series we _nd that all the 7 seriesare integrated as we cannot reject the t-stat for unit root at 1 level But ifwe conduct the same test on di_erenced values of the series we _nd that we canreject the null hypothesis of unit root at 1 signi_cance level for all the seriesexcept CPI This tells us that all the series are I(1) as there _rst di_erence seriesare I(0)As our series are I(1) we will work with index levels of time series to determineif there exist one or more cointegrating relationships between the series Tablesin subsection 743 are based on residual approach where we run a regression ofBSE and various macroeconomic indicators and test the residuals for unit rootusing Augmented Dickey-Fuller test As we assume the two series are cointe-grated we conduct the test with no trend and intercept If the two series arecointegrated then the errors should not have any trend or intercept We see thatwe can reject the null hypothesis of unit root at 1 signi_cance for CPIIP M1We can reject the null of unit root for PPI at 5 and for SP500 and USDINRwe cant reject the null hypothesis of unit root at even 5 level This pointsto the fact that BSE has a strong long term relationship with IP M1 moneysupply CPI at 1 level with IP M1 CPI PPI at 5 signi_cance level AlsoBSE has no long term relationship with SP500 and USD INR exchange rateTo test for multiple cointegrating relationship we now employ a Johansen VARbased cointegration test The results of the test are displayed in subsection744 The _rst panel of the test results displays the value of _trace and_maxof Johansen test with di_erent assumptions about intercept and trend We cansee from this panel that when we consider a functional form of intercept and noTrend we have atleast and atmost three cointegrating relationships The secondpanel of the results display the value of information criteria for lag lengths Formost of the models we see that Akalike criteria points to a lag of three andSchwarz criteria points to a lag of one To estimate the cointegrating model wechoose the model with intercept and no trend and run a cointegration testTestresults are shown in Table 2 of subsection 744 At 5 signi_cance level wecan reject the null of atmost two cointegrating factors for _trace and same for_max Now to test which all variables have a long tern relationship we perform aRestricted cointegration with vector error correction model As we had alreadyEDHEC Business School 215 RESULTSseen in our residual based test of cointegration that BSE has no cointegratingrelationship with SP500 and USDINR we create a restricted cointegration modelwhere we set coe_cients of SP500 and USDINR as zero The test results aredisplayed in Table 3 of subsection 744 In this case as there are two restrictionsthe test statistic follow _2 with two degrees of freedom We can see that thep-value for the test is 1333 which tells us that the restrictions are supportedby data at 10 level of signi_cance So we can conclude that the BSE has along term relationship with CPIIPPPIM1 money supply but has no long termrelationship with SP500 and USDINR exchange rate One interpretation of thisresult can be that the Indian stock market represented here by BSE Sensexmoves more in accordance with domestic factors like Industrial production M1money supply Consumer price index and Producer Price index than with globalfactors or in other words as BSE is representation of largest market cap Indiancompanies we can say that the biggest companies in India are ones that are

more dependent on domestic demand rather than exports This result presentsan opportunity for international investors to diversify their portfolio by invest-ing in BSE Sensex as it is decoupled with global markets and macroeconomicfactorsWe use A bivariate Vector Autoregression (BVAR) technique to analyze thedynamic interaction between real asset prices and macro economy VAR ispreferred method to study Macroeconomy and asset prices where variables en-dogenously e_ect each otherWe begin with a bivariate VAR with no restriction Asset prices and instru-ments are allowed to respond to each other freely For paired variables withcointegration relationship VAR is performed at levels whilst for those that arenot cointegrated VAR is performed at _rst di_erence Constant term is ignoredwith loss of generality We use the Bivariate Autoregression analysis for bothimpulse response and Grangers causality testsImpulse response results are displayed in subsection 745 From _rst graph ofimpulse response of BSE to USDINR we can see that USDINR has a negativeimpact on BSE As impulse response is response of BSE to shocks given to US-DINR we can see that a positive shock or unexpected appreciation INR valuewrt USD will have a negative e_ect on BSE for few lags and will disappearafter few lags If we look at the constituents of BSE Index over time we seethat most of the time some of its constituent are companies that thrive on ex-ports Some of the biggest Market-Cap in India are companies in service sectorlike Infosys TCS etc that are hugely dependent on services provided to clientsfrom Europe and US So an appreciation of INR compared to USD makesthese _rms costlier for the global clients and in turn reduces the income of thesecompanies As the _rms revenue pro_t decreases the value of the stock alsodecreases that in turn a_ects the returns of BSE SensexSecond graph (betwen BSE and SP500) shows that increase in SP500 has a pos-itive e_ect on BSE as higher returns of SP500 indicate strong global economywhich in turn results in higher trade between countries The positive responseof BSE to one unit shock to SP500 indicates a spillover e_ect of global factorson Indian economy but the response is weak as can be seen from the graphMoving forward response of BSE to shocks in M1 money supply CPI PPImake economic sense As for M1 money supply one unit shock means increasein M1 money supply This increase in money supply allows companies to bor-row more money from banks at lower rates which they can use for investingEDHEC Business School 225 RESULTSin pro_table projects and generating larger cash ows For Ination indicatorsone unit shock means increase in ination This increase in ination results inhigher costs for the companies that in turn reduces their pro_t margins and asa result value of stocksBy looking at the graphs we can also see that shocks to Indian macroeconomicindicators creates stronger response by BSE as compared to global factors likeSP500 or USDINR This indicates that BSE Index is driven by companies thatdepend hugely on domestic demand rather than exports Response of BSE toshocks to Industrial Production are contradictory to theory In theory an in-crease in industrial production should result in positive response from BSE butour analysis shows the other way A possible reason for this response could bethat industrial production time series is seasonal as can be seen from the graphSo there is a possibility of a leadlag relationship between the two variablesTo test for possibility of leadlag relationship we run a Grangers causality testbetween BSE and IP The result in section 646 shows that at a lag lengthof 4 we can reject the Null hypothesis of BSE does not Granger cause IP at1 signi_cance level This proves that BSE is a leading indicator of industrialproduction and there exist a leadlag relationship between the two indicatorsEDHEC Business School 236 CONCLUSIONS6 Conclusions

In this paper I tested the relations between Indian stock market represented byBSE and domestic and global macro economic factors The research concludesthat the India stock markets are mainly driven by domestic demand and theinuence of global macro factors on the stock market is weak I also tested forGranger causality between BSE and IP and found that BSE is a leading indicatorof Industrial production and can help in predicting the industrial climate inIndiaThe research is insightful for investors and professionals who are looking forinvestment opportunities to diversify their risks As Indian stock markets aremore dependent on domestic factors one can invest in Indian indices and stocksto diversify their risks gained through investing in US and European stocksThe paper opens new doors for research in this _eld One can use variancedecomposition technique to see how much variance of BSE can be explained myvarious domestic and global macro factors Also one can use di_erent globalfactors like sovereign CDS spreads T-Bill rates a composite indicator of globaleconomy for further research on interaction between Indian stock market andglobal economyOne can also research on how various global macroeconomicnews a_ects India stock markets and for how long the e_ects persistsEDHEC Business School 247 GRAPHS AND TABLES7 Graphs and Tables71 Graphs of Time seriesEDHEC Business School 257 GRAPHS AND TABLESEDHEC Business School 267 GRAPHS AND TABLESEDHEC Business School 277 GRAPHS AND TABLESEDHEC Business School 287 GRAPHS AND TABLES72 Graphs of Time Series - Di_erencedEDHEC Business School 297 GRAPHS AND TABLESEDHEC Business School 307 GRAPHS AND TABLESEDHEC Business School 317 GRAPHS AND TABLESEDHEC Business School 327 GRAPHS AND TABLES73 Correlograms of Time seriesBSEEDHEC Business School 337 GRAPHS AND TABLESIPEDHEC Business School 347 GRAPHS AND TABLESSP500EDHEC Business School 357 GRAPHS AND TABLESUSDINREDHEC Business School 367 GRAPHS AND TABLESCPIEDHEC Business School 377 GRAPHS AND TABLESPPIEDHEC Business School 387 GRAPHS AND TABLES

M1EDHEC Business School 397 GRAPHS AND TABLES74 Tables741 Table for Unit root test of Time seriesVariables T-Stat p-valueBSE -2671 2495 SP500 -1315 8818 CPI -1909 6466 IP -1669 899 M1 -2420 3679 PPI -3353 601 USDINR -2955 1469 742 Tables for Unit root test of Di_erenced time seriesVariables T-Stat p-valueBSE -13848 000 SP500 -14832 000 CPI -3344 140 IP -3865 027 M1 -3867 026 PPI -9656 000 USDINR -13701 000 743 Tables for Residual based test of cointegrationTable 1BSE - CPIt-Statistic ProbADF test statistic -2622676 087Test critical values 1 level -25738185 level -19420410 level -1615891Table 2BSE - IPt-Statistic ProbADF test statistic -3738802 002Test critical values 1 level -25745135 level -194213610 level -1615828EDHEC Business School 407 GRAPHS AND TABLESTable 3BSE - M1t-Statistic ProbADF test statistic -2875518 041Test critical values 1 level -25737845 level -194203510 level -1615894Table 4BSE - PPIt-Statistic ProbADF test statistic -2399055 162Test critical values 1 level -25737845 level -194203510 level -1615894Table 5BSE - SP500t-Statistic ProbADF test statistic -1427184 1430Test critical values 1 level -25737845 level -1942035

10 level -1615894EDHEC Business School 417 GRAPHS AND TABLESTable 6BSE - USDINRt-Statistic ProbADF test statistic -1659522 917Test critical values 1 level -25738185 level -19420410 level -1615891EDHEC Business School 427 GRAPHS AND TABLES744 Johansen cointegration testEDHEC Business School 437 GRAPHS AND TABLESTable 2EDHEC Business School 447 GRAPHS AND TABLESTable 3EDHEC Business School 457 GRAPHS AND TABLES745 Impulse response testsEDHEC Business School 467 GRAPHS AND TABLESEDHEC Business School 477 GRAPHS AND TABLESEDHEC Business School 487 GRAPHS AND TABLES746 Granger causality test between IP and BSEEDHEC Business School 498 BIBLIOGRAPHY8 BibliographyEugene F Fama Ination Output and Money Journal of Business 1982Eugene F Fama Stock Returns Real activity and Money The American Eco-nomic Review 1981Eugene F Fama Stock Returns Expected Returns and Real activity Journal ofFinance 1990Pal and Mittal Impact of macroeconomic indicators in Indian capital marketsJournal of Risk Finance 2011Shahid Ahmed Aggregate Economic Variables and Stock Markets in India In-ternational Research Journal of Finance and Economics 2008Sahu and Dhiman Correlation and Causality between Stock Market and MacroEconomic Variables in India An Empirical Study 2010 International Confer-ence on E-Business and Economics 2011Mohammad Bayezid Ali Impact of Micro Variables on Emerging Stock MarketReturn A case on Dhaka Stock Exchange (DSE) Interdisciplinary Journal ofResearch in Business 2011Napphon Tangjitprom Macroeconomic Factors of Emerging Stock Market Theevidence from Thailand International Journal of Finance and Research 2012Sayed Mehdi Hosseini The Role of Macroeconomic Variables on Stock MarketIndex in China and India International Journal of Economics and Finance2011John Y Campbell Pitfalls and Opportunities What Macroeconomists shouldknow about Unit Roots NBER Working Papers 1991Hacker and Hatemi The properties of Procedures Dealing with Uncertainityabout Intercept and Deterministic Trend in Unit Root Testing CESIS Elec-tronic Working Papers 2010Elder and Kennedy Testing for Unit Roots What should Students be TaughtNasseh and Strauss Stock Prices and domestic and international macroeco-

nomic activity a cointegration approach The Quarterly Review of Economicsand Finance 2000Engle and Granger Co-Integration and Error Correction Representation Es-timation and Testing Econometrica 1987Eugene F Fama Stock Returns Real Activity Ination and Money 1981American Economic AssociationNaliniprave Tripathy Causal Relationship between Macro-Economic Indicatorsand Stock Market in India Asian Journal of Finance and Accounting 2011Rogalski and Vinso Stock Returns Money Supply and the Direction of Causal-ity The Journal of Finance 1977James et al A VARMA Analysis of the Causal Relations Among Stock Re-turns Real output and Nominal Interest Rates 1985 The Journal of FinanceBailey and Chung Risk and return in the Philippine Equity market A multi-factor exploration Paci_c-Basin Finance Journal 1996Nai-Fu Chen Financial Investment opportunities and the Macroeconomy TheJournal of Finance 1991GB Wickremasinghe Macroeconomic forces and stock prices Some empiricalevidence from an emerging stock markets University of Wollongong 2006EDHEC Business School 508 BIBLIOGRAPHYYao Juo and Loh On Chinas Monetary Policy and Asset Prices University ofNottingham- China policy Institute 2011Bilson et al Selecting macroeconomic variables as explanatory factors of emerg-ing stock market returns Paci_c-Basin Finance Journal 2001CHen Roll and Ross Economic forces and the Stock Markets The Journal ofBusiness 1986William H Greene Econometric Analysis 6th Edition Pearson InternationalEditionRuey Tsay Analysis of Financial Time seriesChris Brooks Introductory Econometrics for Finance Cambridge PublicationsEDHEC Business School 51

index (BSE)EDHEC Business School 4AcknowledgmentI am thankful to Professor Robert Kimmel for his comments and guidanceon the subject He has been a constant source of inspiration and a good men-tor from whom I learned a lot I am also grateful to Stoyan Stoyanov MarcRakotomalala Aishwarya Iyer Wen lei Lixia Loh for some great insights intothe subject Their timely comments and suggestions on empirical tests helpedme improve the statistical signi_cance of my tests I thank EDHEC Risk In-stitute for allowing me to use their resources to get the data from various dataproviders In the end ill like to thank my parents and my sister for constantsupport and motivation without which it would have been impossible to climbthis arduous pathRegardsSaurabh YADAVEDHEC Business School 5CONTENTSContents1 Introduction 72 Literature Review 93 Data 1331 Description of Macroeconomic Indicators 1332 Description of Stock Market Indices 144 Methodology 1541 Construction of Time Series 1542 Unit Root Test and Stationarity 15421 Mathematical representation of Stationary series and unitroot test 16422 Augmented Dickey Fuller Unit Root Test 1743 Testing Long Term Relationships 18431 Johansen test for Cointegration 1844 Impulse Response 195 Results 216 Conclusions 247 Graphs and Tables 2571 Graphs of Time series 2572 Graphs of Time Series - Di_erenced 2973 Correlograms of Time series 3374 Tables 40741 Table for Unit root test of Time series 40742 Tables for Unit root test of Di_erenced time series 40743 Tables for Residual based test of cointegration 40744 Johansen cointegration test 43745 Impulse response tests 46746 Granger causality test between IP and BSE 498 Bibliography 50EDHEC Business School 61 INTRODUCTION1 IntroductionIn the past few decades there has been a growing interest among academiciansand practitioners about the relationship between macroeconomic variables andasset prices mainly stocks and house prices In a good and expanding economyprices of stocks are supposed to increase as there is an increase in expectationof large future cash ows pro_ts for the companies and various role playersin the economy Similarly during a bad or downward spiralling economy theexpectation of large future cash ows and pro_ts decrease and consequently theprice of stocks decreaseStock markets are representative of economy of a country and investors belief

They are able to capture macro economic movements in the economy as well asidiosyncratic factors related to each company or industry As Stock prices arereal time and are more frequent than macroeconomic releases they are betterreector of changes in domestic and global economy and can predict the move-ment of macroeconomic indicators In other words stock markets are a leadingindicator of the economyMarkets respond to di_erent macroeconomic indicators in di_erent ways Theresponse of Stock markets to any macroeconomic news is dependent on how thenews will e_ect the pro_ts and interest rates The price of the stock accordingto the Discounted Cash Flow formula isPt =Div1(1 + r1)1 +Div2(1 + r2)2 + +Divt(1 + rt)t (1)As both dividends and interest rates enter into the formula for value of astock the reaction of stock price to a macro news will depend on how the newse_ect the discounting factor ( Interest rates ) and future pro_ts of the com-panies Macro economic factors that project brighter times and more pro_tsfor the companies like increasing Industrial production Increasing M1 moneysupply good consumer con_dence levels will have a positive e_ect on the stockprices Whereas macro news that point to economic recession or slow growthlike decreasing Industrial production coupled with Rising interest rates Risein ination rise in unemployment etc will have a downward e_ect on stockpricesFirst people to do an empirical study on this subject were Eugene Fama andKenneth French In their 1981 paper Stock returns Real activity Inationand money they analysed the relationship between stock returns real activityination and money supply using macro economic data After that study therehas been a barrage of studies on relationship between stock returns and macroeconomic factors based on US and UK data Another important paper pub-lished on this research was by ChenRoll and Ross (1986) who analysed whetherinnovations in the macroeconomic variables are risks that are awarded in thestock markets They found that macroeconomic variables like spread betweenlong and short interest rates expected and unexpected ination Industrial pro-duction are some of the factors that are awarded by the markets Further theArbitrage pricing theory (APT) of Ross (1976) posits relation between stockprices and certain macro-economic variables In the last decade or so the focusfor these kind of studies have started to shift from developed world economies todeveloping world economies As developing world economies have shown signsEDHEC Business School 71 INTRODUCTIONof huge growth potential and leading the economies globally out of recessionsthis motivates us to research on developing markets like India Such a studywill help us to _nd the relation between stock market and macroeconomic indi-cators and give a new insight to foreign investors academicianspolicy makerstraders and domestic investorsThis study is important in a sense it provides an insight to how are Indian stockmarkets are related to its macroeconomic variables and global macromicro eco-nomic factors This study will also help us in analysing whether the Indian stockmarkets have become coupled to global factors or are they still dominated bydomestic economic factorsThe focus of this study is on relation between Indian stock market representedby BSE Sensex and domestic macroeconomic factors and global factors repre-sented by Standard and Poors 500 Index This study builds on earlier studiesdone in this area but also open some new doors for further research It is sim-ilar to some earlier studies in a respect that it uses data macro and micro

factors and econometrics tools used in previous studies but at the same time itdi_erentiates itself from earlier studies in a sense that it is done on a marketthat is still developing Also the time period used in the analysis is a periodwhere Indian market has undergone lot of regulatory changes that has createda structural change in the market Further in this study Ill analyse whetherthe Indian markets are driven mainly by Domestic factors or do global factorshave more inuence on Indian markets To analyse the impact of internationalfactors Ill use Standard and Poors 500 Index and USDINR exchange rate as asubstitute of global factors and to model domestic demand Ill use macro factorslike Industrial production M1 money supply Consumer Price Index and Pro-ducer price Index The outline of the thesis is as followings Section 2 providesa literature review of the studies done earlier in this area Section 3 provides adetailed description of the data used in the study Section 4 provides a detaileddescription of the methodology and various econometric tools that will be usedin the study Section 5 provides the results of the study and Section 6 providesthe conclusion of the studyEDHEC Business School 82 LITERATURE REVIEW2 Literature ReviewMany studies and researchers have tried to _nd factors that can explain stockreturns The most famous and earliest model is the Capital Asset Pricing Model(CAPM) developed by Sharpe (1964) Lintner (1965) Mossin (1967) and Black(1972) The concept of this single factor model is developed from diversi_-cation introduced by Markowitz (1952) In CAPM model the expected stockreturns can be explained with the help of Risk free rate and one risk factorMarket CAPM says that the systematic risk can be captured by sensitivenessof each stock to change in overall market which is measured by Beta Accordingto CAPM the market factor is the only factor determining the stock returnsCAPM was a revolutionary model It changed the way people looked at thestock returns as something that is vary arbitrary As it is very easy to under-stand and use CAPM is very popular as the model used to determine the stockreturn in most of _nance textbooks and used by many practitioners in stockmarketHowever the numerous set of assumptions made in deriving CAPM made itinconsistent with the real world and led to criticism of CAPM To overcomethe limitations and assumptions made in CAPM many scholars came up withmulti- factor models like Fama-French three factor model APT model etc InFama-French model they try to explain stock returns with help of three factorsmarketsmall minus big and value minus growth the model was able to explainthe returns based on these risk factors for some time before it failed Therehave been many studies on failure of Fama-French model and markets where itis not applicableThe macroeconomic models of explaining stock returns started with APT (Ar-bitrage Pricing Theory) by Ross (1976) which was later re_ned by Roll andRoss (1980) APT is a multi-factor model and claims that the stock return canbe explained by unexpected changes or shocks in multiple factors ChenRolland Ross (1986) perform the empirical study for APT model and identify thatsurprise or shock in macroeconomic variables can explain the stock return sig-ni_cantly The variables used in their study are Industrial production indexdefault risk premium that can measure the con_dence of investors and changein yield curve that can be measured by term premiumThe study of macroeconomic factors in explaining stock returns have been pop-ular since then Stock price is present value of all discounted future cash owsIf a _rm is performing well then the expectation of large future cash ows risesand consequently the stock price rises On the other hand if a _rm is performingbad for couple of years then the expectation of big future cash ows decreaseand in turn the stock price fall This is a micro and idiosyncratic explanation ofstock prices and returns But the future cash ows of a stock does not dependsolely on the companys performance or pro_tsloss The systematic factor can

have a huge impact on the cash ows of not only one but many companies Thesystematic factor here refers to macro economic variables The state of Macroeconomic conditions lead to changes in Monetary and regulatory policies by thegovernment and which in turn a_ects the stock prices For example a countrywith good economic conditions represented by its Industrial production indexGDP CPI Interest rates will create an environment that is conducive for thegrowth of companies by lowering borrowing rates and other open market opera-tions So all macroeconomic factors that can inuence future cash ows or theEDHEC Business School 92 LITERATURE REVIEWdiscount rate by which the cash ows are discounted should have an inuenceon the stock priceMany researcher have studies the relationship between stock prices and macroeconomic variables and tried to explain the a_ect of one over the other Fama(1981) tries to establish a relationship between stock returns real activity ina-tion and money In his paper he _nds that Stock returns have positive relationwith real output and money supply but a negative relation with ination Heexplains that negative relation between stock returns and ination is induced bynegative relation between real output approximated by Industrial productionand ination This negative relationship between ination and real activityis explained by money demand theory and quantity theory of money Fama(1990) explains that measuring the total return variation explained by shocksto expected cash ows time-varying expected returns and shocks to expectedreturns is one way to judge the rationality of stock prices In his paper he_nds that growth rates of production used to proxy for shocks to expected cashows explain 45 of return variance ChenRoll and Ross (1986) explored therelationship between a set of economic variables and their systematic inuenceon stock market returns They found that Industrial production changes inrisk premium twists in yield curve had strong relationship and impact on stockreturns A somewhat weaker e_ect was found for measures of unanticipatedination and changes in expected ination during periods when these variableswere highly volatile They concluded that stock returns were exposed to sys-tematic economic news that they are priced in accordance to their exposuresand that the news can be measured as innovation in state variables Chen(1991) found that state variables that are priced are those that can forecastchanges in the investment and consumption opportunity set According to hisresearch default spread the term spread the one-month T-Bill rate the laggedindustrial production growth rate and the dividend-price ration are importantdeterminants of future stock market returns Bulmash and Trivoli (1991) showthe e_ect of business cycle movements on the relationship between stock returnsand money growthAn interesting paper in this _eld of research is by Fama (1990) and Schwert(1990) In the paper they claim that there are three explanations for the stronglink between stock prices and real economic activityFirst information about the future real activity may be reected in stockprices well before it occurs|this is essentially the notion that stock pricesare a leading indicator for the well-being of the economy Second changesin discount rates may a_ect stock prices and real investment similarly butthe output from real investment doesnt appear for some time after it ismade Third changes in stock prices are changes in wealth and thiscan a_ect the demand for consumption and investment goods [Schwert(1990)p1237]Campbell and Ammer (1993) use a VAR approach to model the simulta-neous interactions between the stock and bond markets since most previousworks do not address the channels through which the macroeconomic activityinuences the stock prices One example could be that industrial productioncould be linked to changing expectations of future cash ows (Balvers at al1990) On the other hand interest rate innovations could be the driving factorEDHEC Business School 102 LITERATURE REVIEW

in determining both industrial production (due to change in investment) andstock prices (due to change in the discounted present value of future cash ows)A VAR analysis can distinguish these possibilities Mukherjee and Naka (1995)show a long-term relationship between the Japanese stock price and real macroe-conomic variables Dr Nishat (2004) studies the long term association amongmacroeconomic variables like money supply CPIIPI and foreign exchange rateand stock markets in Pakistan The results show that there are causal relation-ship among the stock price and macroeconomic variables He uses data from1974 to 2004 in his study As most of the _nancial time series are non station-ary in levels he uses unit root technique to make data stationary Fazal Hussianand Tariq Massod (2001) used variables like investment GDP and consumptionemploying Grangers causality test to _nd relationship between macro factorsand stock markets They show that at two lags all macroeconomic variableshave highly signi_cant e_ect on stock prices James et al (1985) use a VARMAanalysis for investigating relationship between macro economy and stock mar-ket Using VARMA analysis for _nding causal relationship between factors isa better technique as the procedure does not preclude any causal structure apriori since it allows feedback among variables Thus the VARMA approachallow whatever causal relationship exist to emerge from the data They _ndlinkages between real activity and stock returns and real activity and inationAlso they _nd that stock returns signal changes in the monetary base Sincestock returns also signal changes in expected real activity this suggests a linkbetween the money supply and expected real activity that is consistent with themoney supply explanation o_ered by Geske and RollIn recent years the focus of these kind of studies have shifted from developedeconomies to developing economies As developing economies are the economiesthat see a lot of structural and monetary policy changes an analysis of relation-ship between macro and micro can provide new insights Also one can analysethe e_ects of monetary policies on the asset prices especially on stock pricesTangjitprom (2012) study of macroeconomic factors like unemployment rateinterest rate ination rate and exchange rate and stock market of Thailand con-cludes that macroeconomic factors signi_cantly explain stock returns He also_nds that for Thailand unemployment rate and ination rate are insigni_cant todetermine the stock returns The reason he provides is that the unemploymentrate and ination rate are not timely and there could be some lags before thedata becomes available Also Grangers test to examine lead-lag relationshipamong the factors reveal that only few macroeconomic variables could predictthe future stock returns whereas the stock returns can predict most of futuremacro economic variables This implies that performance of stock markets canbe a leading indicator for future macroeconomic conditions Ali (2011) study ofimpact of macro and micro factors on stock returns reveals that ination andforeign remittance have negative inuence and industrial production index havepositive impact on stock markets Also he didnt found any Grangers Causal-ity between stock markets and any of the explanatory variables This lack ofGrangers causality reveals the evidence of informationally ine_cient marketsAli uses a multivariate regression analysis on standard OLD formula for estimat-ing the relationship Hosseini et al (2011) tested the relationship between stockmarkets and four macro economic variables namely crude oil prices Money sup-ply Industrial production and ination rate in China and India They used aperiod of 1999 to 2009 for analysis As most of the economic time series have unitEDHEC Business School 112 LITERATURE REVIEWroot they _rst used the Augmented Dickey Fuller unit root test and found theunderlying series to be non-stationary at levels but stationary after in di_erenceAlso the use of Jhonson-Juselius (1990) Multivariate cointegration and VectorError Correction model technique indicate that there are both long and shortrun linkages between macroeconomic variable and stock market index in each ofthe two countries Their analysis shows that in long run the impact of increasein prices of crude oil for China is positive but for India is negative In terms

of money supply the impact on Indian stock market is negative but for Chinathere is a positive impact The e_ect of Industrial production is negative onlyin China In addition the e_ect of increases in ination on these stock marketsis positive in both countries Wickremasinghe (2006) analysed the relationshipbetween stock prices and macroeconomic variables in Sri Lanka He used theUnit root tests Jhonsons test Error-correction model variance decomposi-tion and impulse response to analyse the relationships His _ndings indicatethat there is both long term and short term causal relationship between stockprices and macroeconomic variables in Sri Lanka The result indicate that thestock prices can be predicted from certain macroeconomic variables and henceviolate the validity of the semi-strong version of e_cient market hypothesisAhmed (2008) investigates the causal relationship between Indian macroeco-nomic factors like Industrial Production Exports Foreign direct investmentMoney supply exchange rate interest rate and stock market indices NSE NiftyIndex and BSE Sensex For _nding the long term relationship he applies Jo-hansens cointegration and Toda and Yamamoto Granger Causality tests Foranalysing the Impulse response and variance decomposition he uses bivariateVAR His _ndings reveal that stock prices in India lead macroeconomic activityexcept movement in interest rate Interest rate seem to lead the stock priceThe study also reveals that movement of stock prices is not only the outcomeof behaviour of key macro economic variables but it is also one of the causesof movement in other macro dimensions in the economy An important paperby Bilson et al (2001) argues that emerging markets local factors are moreimportant than global factors They _nd that for emerging markets are at leastpartially segmented from global capital markets The global factors are proxiedby world market returns and local factors by set of macro economic variableslike money supply prices real activity and exchange rate Some evidence isfound that local factors are signi_cant in their association with emerging equitymarket returns above than that explained by the world factor When they usea larger set of variables the explanatory power of the model improves substan-tially such that they are able to explain a large amount of return variation formost emerging marketsEDHEC Business School 123 DATA3 Data31 Description of Macroeconomic IndicatorsOne of the biggest problems when conducting a research with macroeconomicdata is the frequency of the data Most of the macroeconomic indicator timeseries are yearlyquarterly or monthly time series This low frequency of themacroeconomic indicators results in very few data points for conducting a anal-ysis that is robust A possible cure for the problem is to use longer time periodsto incorporate more data points for macroeconomic variables But anotherproblem that we face when we look at the macroeconomic indicators for Asiancountries is reporting of the data For most of the Asian countries the macroe-conomic data doesnt have a long history and same can be said about historyof Indian macroeconomic variables So in this research we have used a timeperiod for which we can _nd data for most of the macroeconomic indicators Inthis paper we use a time period of 20 years starting from 1990 to 2011 Thistime period in Indian economy is representative of many structural and mone-tary policy changes like liberalization of India markets Also as the time periodis long it gives us enough data point for each macroeconomic factors to do arobust empirical analysisWhen one starts to build a model of interaction between macro and micro eco-nomic factors one dominant and important question one faces is among themyriad of macro indicators available for an economy which factors to chooseto incorporate in the model If one chooses macroeconomic factors that arehighly correlated among themselves then the power of test results decrease asit may result in a model where the macro indicators are able to explain mostof the movement of micro factors but the macro factors may not be relevant

To circumvent this problem we use variables that have been tested in earlierresearches and that have been proven to have e_ect on stock markets I alsotest a few macro factors that have some _nancial theory behind them that con-nect them to stock markets Ali (2011) Wickremasinghe (2006) Bilson etal(2001) and Bailey (1996) _nd that Industrial production CPI exchange rateM1 money supply GDP are few of the macro economic factors that can signi_-cantly explain stock returns Sahu(2011) Ahmed(2008) Tripathy(2011) studyon Indian markets speci_cally show that Industrial Production Exchange rateInation index are macro economic indicators that have a strong positive ornegative relationship with the stock markets So in our study we test 5 macroeconomic variables namely M1 money supply Consumer and Producer price In-dex Industrial production Exchange rate The time period for these indicatorsis from 1990-2011 The data for Ination indices Industrial production andexchange rate has been pulled from Bloombergc and Datastreamc The datahas been processed for errors and missing values Data for M1 money supplyhas been pulled from RBI website For most of the indices like ination andIndustrial production index the base year has been changed to 1990 Also assome of the indices are in levels and some in actual _gures (M1 money supply)we convert all of the indicators to level form (starting at 100 in 1990)EDHEC Business School 133 DATA32 Description of Stock Market IndicesCompared to Macro Indicators stock market data is relatively easy to _nd andhas considerably long history Also the stock market data is a real time data soit has a very high frequency of seconds Here in our analysis we will make use ofBSE (Bombay Stock Exchange) as representation of Indian markets and SP500(Standard and Poors 500 Index) as representation of global factors BSE is amarket cap-weighted of 30 stocks It is the oldest Index in the Asian markets(established in 1875) and have had a long history We choose this index as it isthe Index that represent the most liquid and traded stocks of the Indian stockmarket Also the index is most traded index in India and a good representationof trade prices of the stocks Even in terms of an orderly growth much beforethe actual legislations were enacted BSE Limited had formulated a compre-hensive set of Rules and Regulations for the securities market It had also laiddown best practices which were adopted subsequently by 23 stock exchangeswhich were set up after India gained its independence Our choice of SP500 isbased on the fact that it has a long history and many researchers have usedthis index as a good proxy representation of global markets and economic con-ditions We will take the monthly returns of each of the indices from 1990-2011in accordance with data frequency of macro economic variables Also as theindices have di_erent levels at beginning of 1990 we rebase both the indices tobase year of 1990 starting at a level of 100EDHEC Business School 144 METHODOLOGY4 Methodology41 Construction of Time SeriesThe _rst step in constructing an econometric model is constructing time seriesall of which are in same units Most of the time series used in our analysis are indi_erent formats For example CPI PPI BSE Index SP500 are in levels M1money supply USDINR exchange rate is in absolute current format Industrialproduction is in absolute production levels So _rst we convert all of the giventime series to level The way we construct time series in levels is _rstly takingthe initial data point of each time series as 100 We then _nd the percentagechange from one period to the next one for each time series using a continuouscompounding assumption (taking a natural log of change in values) In math-ematical terms it can be stated as Assume the original Index value at time tto be It and at time t+1 to be It + 1 Then we can compute the new rebasedindex by formulaRIt+1 = RIt _ (1 + ln(It+1=It))

whereRIt= Rebased Index at time tRIt+1=Rebased Index at time t+1We can use these rebased indices in building and testing our econometric model42 Unit Root Test and StationarityUnit root test is to _nd whether the series is stationary or non-stationary Astrictly stationary process is one where for any t1 t2 tt 2Z any k 2Z andT=12Fyt1 yt2 yt3 ytT

(y1 yT ) = Fyt1+k yt2+k yt3+k ytT+k

(y1 yT )where F represents joint distribution function of the set of random variablesIt can also be stated that the probability measure of sequence of yt is same asyt+k for all k In other words a series is stationary if the distribution of its valueremain the same as time progresses Similar to the concept of strict stationaryis weakly stationary process A weakly stationary process is one which has aconstant mean variance and autocovariance structure Stationary is a necessarycondition for a time series to be tested in regression A non-stationary seriescan have several problems like1 The shocks given to the series would not die of gradually resulting inincrease of variance as time passes2 If the series is non stationary then it can lead to spurious regressions If twoseries are generated independent of each other then if one is regressed onother it will result in very low R2 values But if two series are trending overtime then a regression of one over the other will give high R2 even thoughthe series may be unrelated to each other So if normal regressions toolsEDHEC Business School 154 METHODOLOGYare used on non stationary data then it may result in good but valuelessresults3 If the variables employed in a regression model are not stationary thenit can be proved that the standard assumptions for asymptotic analysiswill not be valid In other words the usual t-ratios will not follow at-distribution and the F-statistic will not follow an F-distribution and soonStationarity is a desirable condition for any time series so that it can be usedin regressions and give meaningful result that have some value to test for sta-tionarity a quick and dirty way is looking at the autocorrelation and partialcorrelation function of the series If the series is stationary then the autocorre-lation function should die o_ gradually after few lags and the partial correlationfunction will me non zero for some lags and zero thereafter Also we can usethe Ljung-Box test for testing that all m of _k autocorrelation coe_cients arezero using Q-statistic given by formulaQ = T(T + 2)_mk=1_k2T 1048576 k_ _2where T = Sample size and m = Maximum lag lengthThe lag length selection can be based on di_erent Information Criteria likeAkaikes Information criteria (AIC) Schwarzs Bayesian information criteria(SBIC) Hannan-Quinn criterion (HQIC) Mathematically di_erent criteria arerepresented asAIC = ln(_2) + 2kTSBIC = ln(_2) + kT lnTHQIC = ln(_2) + 2kT ln(ln(T))

For a better test for stationarity we use augmented Dickey fuller Unit roottest on each time series separately Augmented Dickey Fuller test is test ofnull hypothesis that the time series contains a unit roots against a alternativehypothesis that the series is stationary421 Mathematical representation of Stationary series and unit roottestAssume a variable Y whose structure can be given by AR process with no driftequationyt = _1yt10485761 + _2yt10485762 + _3yt10485763 + + _nyt1048576n + ut (2)where ut is the residual at time t Using a Lag operator L we can write eq(1)asyt = _1L1yt + _2L2yt + _3L3yt + + _nLnyt + ut (3)EDHEC Business School 164 METHODOLOGYRearranging eqn (2) we getyt 1048576 _1L1yt 1048576 _2L2yt 1048576 _3L3yt + 1048576 _nLnyt = ut (4)yt(1 1048576 _1L1 1048576 _2L2 1048576 _3L3 + 1048576 _nLn) = ut (5)or_(L)yt = ut (6)The time series is stationary if we can write eqn(5) in formyt = _(L)10485761ut (7)with _(L)10485761 converging to zero It means the autocorrelation function woulddecline as lag length is increased If eqn (6) is expanded to a MA(1) processthe coe_cients of residuals should decrease such that the the residuals that thee_ect of residuals decrease with increase in lags SO if the process is stationarythe coe_cients of residuals will converge to zero and for non-stationary seriesthey will and converge to zero and will have long term e_ect The condition fortesting of unit root for an AR process is that the roots of eqn(6) or Charac-teristic equation should lie outside unit circle422 Augmented Dickey Fuller Unit Root TestConsider an AR(1) process of variable Yyt = _yt10485761 + ut (8)Subtracting yt10485761 from both sides of eqn(7) we get_y = (_ 1048576 1)yt10485761 + ut (9)Eqn(8) is the test equation for Dickey Fuller test For Dickey-Fuller Unit roottestNull Hypothesis The value of _ is equal to 1 or value of _10485761 is equal to 0 vsAlternate Hypothesis The value of _ is less than one or value of _ 1048576 1 is lessthan zero Augmented Dickey-Fuller test is similar to normal Dickey-Fuller testsexcept it takes the lag structure of more than one into account_y = yt10485761 +Xpi=1_i_yt1048576i + ut (10)If the series has one or more unit root it is said to be integrated of order nwhere n is the number of unit roots of the characteristic equation To makethese time series stationary they needs to be di_erenced Mathematically ifyt _ I (n) (11)then_(d) yt _ I (0) (12)To make our time-series stationary we will use the natural log returns of theseseries in the analysisEDHEC Business School 174 METHODOLOGY43 Testing Long Term RelationshipsEngle and Granger (1987) in their seminal paper described cointegration whichforms the basis for testing for long term relationship between variables Accord-ing to Engle and Granger two variables are cointegrated if they are integratedprocess in their natural form (of the same order) but a weighted combination

of the variables can be found such that the combined new variable is integratedof order less than the order of individual time series Mathematically assumeyt to be a k X 1 vector of variables then the components are cointegrated orintegrated of order (db) if1 All components of yt are I(d)2 There is at least one vector of coe_cients _ such that_0

yt _ I (d 1048576 b) (13)As most of the _nancial time series are integrated of order one we will restrictourselves to case d=b=1 Two or more variables are said to be cointegrated ifthere exist a linear combination of these variables that is stationary Many ofthe series are non-stationary but move together over time which implies twoseries are bound by some common force or factor in long run We will test forcointegration by a residual-based approach and Johansens VAR methodResidual Based approach Consider a modelyt = _1 + _2x2t + _3x3t + + ut (14)where yt x2t x3t are all integrated of order N Now if the residual of this re-gression ut is stationary then we can say that the variables are cointegrated elsethere exist no long term relationship between the variables To test the resid-ual for stationarity we will run Augmented Dickey-Fuller tests on the residualsUnder the Null hypothesis the residual are integrated of order one or more andunder alternate hypothesis the residuals are I(0)431 Johansen test for CointegrationJohansen test for cointegration presents a better model for testing multiplecointegration among multiple variables The Residual based approach can only_nd atmost one cointegration and can be tested for a model with two variablesEven if more than two variables are present in the equation that are cointegratedthe Residual based approach will give only one cointegration SO we will useJhoansen VAR based cointegration for testing more than one cointegrationSuppose that a set of g variables are under consideration that are I(1) andwhich are thought to be cointegrated A VAR with k lags containing thesevariables could be set upyt = _1yt10485761 + _2yt10485762 + _ _ _ + _kyt1048576k + ut (15)g _ 1 g _ g g _ 1 g _ g g _ 1 g _ g g _ 1 g _ 1EDHEC Business School 184 METHODOLOGYIn order to use the Johansen test the VAR above should be turned into avector error correction model of form_yt = _yt1048576k + 1_yt10485761 + 2_yt10485762 + _ _ _ + k10485761_yt1048576(k10485761) + ut (16)where _ = (_ki=1_i) 1048576 Ig and i = (_ij=1_j) 1048576 IgThe Johansens test centers around testing the _ matrix which is the matrixthat represents the long term cointegration between the variables The test fornumber of cointegration is calculated by looking at the rank of the _ matrixthrough its eigenvalues The rank of the matrix is equal to number of roots(eigenvalues) _i of the matrix that are di_erent from zero The roots should beless than 1 in absolute value and positive If the variables are not cointegratedthe rank of the matrix will not be signi_cantly di_erent from zero ie _i _ 0There are two test statistics for Johansen test _tracer and _max_trace (r) = 1048576TPgi=r+1 ln(1 1048576 _ _i)and_max(r r + 1) = 1048576Tln(1 1048576 _r_+1)_trace is a test statistic for joint test where the null hypothesis is that thenumber of cointegration vector is less than or equal to r against an alternativethat there are more than r_max conducts another separate test on eigenvalues and has null hypothesis that

the number of cointegrating vector is r against r+144 Impulse ResponseOnce we have determined whether the variables have long term relationship ornot we can form a multivariate VAR model for the variables A multivariateVAR model between g variables is a model where the current value of a variabledepend on di_ernt combinations of the previous k values of all the variables anderror terms A general representation of the model can beyBSEt = _ + _BSEyBSE + _IP yIP + CPIyCPI + _M1yM1 + _SP500ySP500 + u1t(17)where all the coe_cients except _ are g _ k matrices and all variables y are k_ 1 matricesOnce we have formed a model like this we can use the model for Impulse re-sponse A VAR(p) model can be written as a linear fuction of the past innova-tions that isrt = _ + at + 1at10485761 + 2at10485762 + (18)where _ = [_(1)]10485761_0 provided that the inverse exists and the coe_cient ma-trices i can be obtained by equating the coe_cients of Bi in the equation(I 1048576 _1B 1048576 1048576 _PBP )(I + 1B + 2B2 + ) = I (19)EDHEC Business School 194 METHODOLOGYwhere I is the Identity martix This is a moving average representation of rtwith the coe_cient matrix i being the impact of the past innovation at1048576i onrt Equivalently i is the e_ect of at on the future observation rt+i Therefore i is often referred to as the Impulse Response Function of rt For our impulseresponse we will use equation of variables in _rst di_ernce form like_BSEt = _t +Xki=0_11(i)_BSEt1048576i +Xkj=1_12(j)_MIt1048576j + _BSEt (20)_MIt = _t +Xki=0_21(i)_MIt1048576i +Xkj=1_22(j)_BSEt1048576j + _MIt (21)Grangers causality and Blocks F test of a VAR model will suggest which ofthe variables have statistically signi_cant impacts on the future values of othervariables in the system But F-test results cannot explain the sign of the re-lationship nor how long these e_ects require to take place Such informationwill however be given by an examination of the VARs impulse responses andvariance decompositions Impulse response is a technique that trace out theresponsiveness of the dependent variable in the VAR to shocks of each of theother variables So for each variable from each equation separately we will applya unit shock to the error and trace the e_ects upon the VAR system over timeBy using the impulse response technique we can determine how responsive isthe BSE stock index to Indian macro indicators and SP500 This will help usdetermine whether the BSE index is more reactive to domestic news or globalnewsEDHEC Business School 205 RESULTS5 ResultsBefore we use the time series for VAR analysis or cointegration tests we need todetermine whether the series are Stationary or not If the series are stationaryin levels we can use them directly else we need to use the di_erenced time seriesOne way to look for autocorrelation or integrated process is to see the graphsof the various time series used Section 71 shows the graphs of variables we

use for our analysis As we can see from the graphs all of the time series havea trend in long run which points to an integrated process As a second stepwe plot the graphs of di_erenced time series in Section 52 We can see thatthe di_erenced graphs in Section 72 dont show a long term trend and crossthe X-axis frequently This is usually a property of I(1) processes So we checkthe series for autocorrelations at di_erent lag lengths Section 73 shows cor-relograms graph autocorrelation coe_cient partial autocorrelation coe_cientQ-Stat and p-value for various time series up to 36 lags As can be seen in thetables the Q-stat for all lags is zero and we can reject the joint null hypothesisthat all the autocorrelations up to 36 lags are zero Table 741 shows that ifwe conduct a Unit root test on levels of the series we _nd that all the 7 seriesare integrated as we cannot reject the t-stat for unit root at 1 level But ifwe conduct the same test on di_erenced values of the series we _nd that we canreject the null hypothesis of unit root at 1 signi_cance level for all the seriesexcept CPI This tells us that all the series are I(1) as there _rst di_erence seriesare I(0)As our series are I(1) we will work with index levels of time series to determineif there exist one or more cointegrating relationships between the series Tablesin subsection 743 are based on residual approach where we run a regression ofBSE and various macroeconomic indicators and test the residuals for unit rootusing Augmented Dickey-Fuller test As we assume the two series are cointe-grated we conduct the test with no trend and intercept If the two series arecointegrated then the errors should not have any trend or intercept We see thatwe can reject the null hypothesis of unit root at 1 signi_cance for CPIIP M1We can reject the null of unit root for PPI at 5 and for SP500 and USDINRwe cant reject the null hypothesis of unit root at even 5 level This pointsto the fact that BSE has a strong long term relationship with IP M1 moneysupply CPI at 1 level with IP M1 CPI PPI at 5 signi_cance level AlsoBSE has no long term relationship with SP500 and USD INR exchange rateTo test for multiple cointegrating relationship we now employ a Johansen VARbased cointegration test The results of the test are displayed in subsection744 The _rst panel of the test results displays the value of _trace and_maxof Johansen test with di_erent assumptions about intercept and trend We cansee from this panel that when we consider a functional form of intercept and noTrend we have atleast and atmost three cointegrating relationships The secondpanel of the results display the value of information criteria for lag lengths Formost of the models we see that Akalike criteria points to a lag of three andSchwarz criteria points to a lag of one To estimate the cointegrating model wechoose the model with intercept and no trend and run a cointegration testTestresults are shown in Table 2 of subsection 744 At 5 signi_cance level wecan reject the null of atmost two cointegrating factors for _trace and same for_max Now to test which all variables have a long tern relationship we perform aRestricted cointegration with vector error correction model As we had alreadyEDHEC Business School 215 RESULTSseen in our residual based test of cointegration that BSE has no cointegratingrelationship with SP500 and USDINR we create a restricted cointegration modelwhere we set coe_cients of SP500 and USDINR as zero The test results aredisplayed in Table 3 of subsection 744 In this case as there are two restrictionsthe test statistic follow _2 with two degrees of freedom We can see that thep-value for the test is 1333 which tells us that the restrictions are supportedby data at 10 level of signi_cance So we can conclude that the BSE has along term relationship with CPIIPPPIM1 money supply but has no long termrelationship with SP500 and USDINR exchange rate One interpretation of thisresult can be that the Indian stock market represented here by BSE Sensexmoves more in accordance with domestic factors like Industrial production M1money supply Consumer price index and Producer Price index than with globalfactors or in other words as BSE is representation of largest market cap Indiancompanies we can say that the biggest companies in India are ones that are

more dependent on domestic demand rather than exports This result presentsan opportunity for international investors to diversify their portfolio by invest-ing in BSE Sensex as it is decoupled with global markets and macroeconomicfactorsWe use A bivariate Vector Autoregression (BVAR) technique to analyze thedynamic interaction between real asset prices and macro economy VAR ispreferred method to study Macroeconomy and asset prices where variables en-dogenously e_ect each otherWe begin with a bivariate VAR with no restriction Asset prices and instru-ments are allowed to respond to each other freely For paired variables withcointegration relationship VAR is performed at levels whilst for those that arenot cointegrated VAR is performed at _rst di_erence Constant term is ignoredwith loss of generality We use the Bivariate Autoregression analysis for bothimpulse response and Grangers causality testsImpulse response results are displayed in subsection 745 From _rst graph ofimpulse response of BSE to USDINR we can see that USDINR has a negativeimpact on BSE As impulse response is response of BSE to shocks given to US-DINR we can see that a positive shock or unexpected appreciation INR valuewrt USD will have a negative e_ect on BSE for few lags and will disappearafter few lags If we look at the constituents of BSE Index over time we seethat most of the time some of its constituent are companies that thrive on ex-ports Some of the biggest Market-Cap in India are companies in service sectorlike Infosys TCS etc that are hugely dependent on services provided to clientsfrom Europe and US So an appreciation of INR compared to USD makesthese _rms costlier for the global clients and in turn reduces the income of thesecompanies As the _rms revenue pro_t decreases the value of the stock alsodecreases that in turn a_ects the returns of BSE SensexSecond graph (betwen BSE and SP500) shows that increase in SP500 has a pos-itive e_ect on BSE as higher returns of SP500 indicate strong global economywhich in turn results in higher trade between countries The positive responseof BSE to one unit shock to SP500 indicates a spillover e_ect of global factorson Indian economy but the response is weak as can be seen from the graphMoving forward response of BSE to shocks in M1 money supply CPI PPImake economic sense As for M1 money supply one unit shock means increasein M1 money supply This increase in money supply allows companies to bor-row more money from banks at lower rates which they can use for investingEDHEC Business School 225 RESULTSin pro_table projects and generating larger cash ows For Ination indicatorsone unit shock means increase in ination This increase in ination results inhigher costs for the companies that in turn reduces their pro_t margins and asa result value of stocksBy looking at the graphs we can also see that shocks to Indian macroeconomicindicators creates stronger response by BSE as compared to global factors likeSP500 or USDINR This indicates that BSE Index is driven by companies thatdepend hugely on domestic demand rather than exports Response of BSE toshocks to Industrial Production are contradictory to theory In theory an in-crease in industrial production should result in positive response from BSE butour analysis shows the other way A possible reason for this response could bethat industrial production time series is seasonal as can be seen from the graphSo there is a possibility of a leadlag relationship between the two variablesTo test for possibility of leadlag relationship we run a Grangers causality testbetween BSE and IP The result in section 646 shows that at a lag lengthof 4 we can reject the Null hypothesis of BSE does not Granger cause IP at1 signi_cance level This proves that BSE is a leading indicator of industrialproduction and there exist a leadlag relationship between the two indicatorsEDHEC Business School 236 CONCLUSIONS6 Conclusions

In this paper I tested the relations between Indian stock market represented byBSE and domestic and global macro economic factors The research concludesthat the India stock markets are mainly driven by domestic demand and theinuence of global macro factors on the stock market is weak I also tested forGranger causality between BSE and IP and found that BSE is a leading indicatorof Industrial production and can help in predicting the industrial climate inIndiaThe research is insightful for investors and professionals who are looking forinvestment opportunities to diversify their risks As Indian stock markets aremore dependent on domestic factors one can invest in Indian indices and stocksto diversify their risks gained through investing in US and European stocksThe paper opens new doors for research in this _eld One can use variancedecomposition technique to see how much variance of BSE can be explained myvarious domestic and global macro factors Also one can use di_erent globalfactors like sovereign CDS spreads T-Bill rates a composite indicator of globaleconomy for further research on interaction between Indian stock market andglobal economyOne can also research on how various global macroeconomicnews a_ects India stock markets and for how long the e_ects persistsEDHEC Business School 247 GRAPHS AND TABLES7 Graphs and Tables71 Graphs of Time seriesEDHEC Business School 257 GRAPHS AND TABLESEDHEC Business School 267 GRAPHS AND TABLESEDHEC Business School 277 GRAPHS AND TABLESEDHEC Business School 287 GRAPHS AND TABLES72 Graphs of Time Series - Di_erencedEDHEC Business School 297 GRAPHS AND TABLESEDHEC Business School 307 GRAPHS AND TABLESEDHEC Business School 317 GRAPHS AND TABLESEDHEC Business School 327 GRAPHS AND TABLES73 Correlograms of Time seriesBSEEDHEC Business School 337 GRAPHS AND TABLESIPEDHEC Business School 347 GRAPHS AND TABLESSP500EDHEC Business School 357 GRAPHS AND TABLESUSDINREDHEC Business School 367 GRAPHS AND TABLESCPIEDHEC Business School 377 GRAPHS AND TABLESPPIEDHEC Business School 387 GRAPHS AND TABLES

M1EDHEC Business School 397 GRAPHS AND TABLES74 Tables741 Table for Unit root test of Time seriesVariables T-Stat p-valueBSE -2671 2495 SP500 -1315 8818 CPI -1909 6466 IP -1669 899 M1 -2420 3679 PPI -3353 601 USDINR -2955 1469 742 Tables for Unit root test of Di_erenced time seriesVariables T-Stat p-valueBSE -13848 000 SP500 -14832 000 CPI -3344 140 IP -3865 027 M1 -3867 026 PPI -9656 000 USDINR -13701 000 743 Tables for Residual based test of cointegrationTable 1BSE - CPIt-Statistic ProbADF test statistic -2622676 087Test critical values 1 level -25738185 level -19420410 level -1615891Table 2BSE - IPt-Statistic ProbADF test statistic -3738802 002Test critical values 1 level -25745135 level -194213610 level -1615828EDHEC Business School 407 GRAPHS AND TABLESTable 3BSE - M1t-Statistic ProbADF test statistic -2875518 041Test critical values 1 level -25737845 level -194203510 level -1615894Table 4BSE - PPIt-Statistic ProbADF test statistic -2399055 162Test critical values 1 level -25737845 level -194203510 level -1615894Table 5BSE - SP500t-Statistic ProbADF test statistic -1427184 1430Test critical values 1 level -25737845 level -1942035

10 level -1615894EDHEC Business School 417 GRAPHS AND TABLESTable 6BSE - USDINRt-Statistic ProbADF test statistic -1659522 917Test critical values 1 level -25738185 level -19420410 level -1615891EDHEC Business School 427 GRAPHS AND TABLES744 Johansen cointegration testEDHEC Business School 437 GRAPHS AND TABLESTable 2EDHEC Business School 447 GRAPHS AND TABLESTable 3EDHEC Business School 457 GRAPHS AND TABLES745 Impulse response testsEDHEC Business School 467 GRAPHS AND TABLESEDHEC Business School 477 GRAPHS AND TABLESEDHEC Business School 487 GRAPHS AND TABLES746 Granger causality test between IP and BSEEDHEC Business School 498 BIBLIOGRAPHY8 BibliographyEugene F Fama Ination Output and Money Journal of Business 1982Eugene F Fama Stock Returns Real activity and Money The American Eco-nomic Review 1981Eugene F Fama Stock Returns Expected Returns and Real activity Journal ofFinance 1990Pal and Mittal Impact of macroeconomic indicators in Indian capital marketsJournal of Risk Finance 2011Shahid Ahmed Aggregate Economic Variables and Stock Markets in India In-ternational Research Journal of Finance and Economics 2008Sahu and Dhiman Correlation and Causality between Stock Market and MacroEconomic Variables in India An Empirical Study 2010 International Confer-ence on E-Business and Economics 2011Mohammad Bayezid Ali Impact of Micro Variables on Emerging Stock MarketReturn A case on Dhaka Stock Exchange (DSE) Interdisciplinary Journal ofResearch in Business 2011Napphon Tangjitprom Macroeconomic Factors of Emerging Stock Market Theevidence from Thailand International Journal of Finance and Research 2012Sayed Mehdi Hosseini The Role of Macroeconomic Variables on Stock MarketIndex in China and India International Journal of Economics and Finance2011John Y Campbell Pitfalls and Opportunities What Macroeconomists shouldknow about Unit Roots NBER Working Papers 1991Hacker and Hatemi The properties of Procedures Dealing with Uncertainityabout Intercept and Deterministic Trend in Unit Root Testing CESIS Elec-tronic Working Papers 2010Elder and Kennedy Testing for Unit Roots What should Students be TaughtNasseh and Strauss Stock Prices and domestic and international macroeco-

nomic activity a cointegration approach The Quarterly Review of Economicsand Finance 2000Engle and Granger Co-Integration and Error Correction Representation Es-timation and Testing Econometrica 1987Eugene F Fama Stock Returns Real Activity Ination and Money 1981American Economic AssociationNaliniprave Tripathy Causal Relationship between Macro-Economic Indicatorsand Stock Market in India Asian Journal of Finance and Accounting 2011Rogalski and Vinso Stock Returns Money Supply and the Direction of Causal-ity The Journal of Finance 1977James et al A VARMA Analysis of the Causal Relations Among Stock Re-turns Real output and Nominal Interest Rates 1985 The Journal of FinanceBailey and Chung Risk and return in the Philippine Equity market A multi-factor exploration Paci_c-Basin Finance Journal 1996Nai-Fu Chen Financial Investment opportunities and the Macroeconomy TheJournal of Finance 1991GB Wickremasinghe Macroeconomic forces and stock prices Some empiricalevidence from an emerging stock markets University of Wollongong 2006EDHEC Business School 508 BIBLIOGRAPHYYao Juo and Loh On Chinas Monetary Policy and Asset Prices University ofNottingham- China policy Institute 2011Bilson et al Selecting macroeconomic variables as explanatory factors of emerg-ing stock market returns Paci_c-Basin Finance Journal 2001CHen Roll and Ross Economic forces and the Stock Markets The Journal ofBusiness 1986William H Greene Econometric Analysis 6th Edition Pearson InternationalEditionRuey Tsay Analysis of Financial Time seriesChris Brooks Introductory Econometrics for Finance Cambridge PublicationsEDHEC Business School 51

They are able to capture macro economic movements in the economy as well asidiosyncratic factors related to each company or industry As Stock prices arereal time and are more frequent than macroeconomic releases they are betterreector of changes in domestic and global economy and can predict the move-ment of macroeconomic indicators In other words stock markets are a leadingindicator of the economyMarkets respond to di_erent macroeconomic indicators in di_erent ways Theresponse of Stock markets to any macroeconomic news is dependent on how thenews will e_ect the pro_ts and interest rates The price of the stock accordingto the Discounted Cash Flow formula isPt =Div1(1 + r1)1 +Div2(1 + r2)2 + +Divt(1 + rt)t (1)As both dividends and interest rates enter into the formula for value of astock the reaction of stock price to a macro news will depend on how the newse_ect the discounting factor ( Interest rates ) and future pro_ts of the com-panies Macro economic factors that project brighter times and more pro_tsfor the companies like increasing Industrial production Increasing M1 moneysupply good consumer con_dence levels will have a positive e_ect on the stockprices Whereas macro news that point to economic recession or slow growthlike decreasing Industrial production coupled with Rising interest rates Risein ination rise in unemployment etc will have a downward e_ect on stockpricesFirst people to do an empirical study on this subject were Eugene Fama andKenneth French In their 1981 paper Stock returns Real activity Inationand money they analysed the relationship between stock returns real activityination and money supply using macro economic data After that study therehas been a barrage of studies on relationship between stock returns and macroeconomic factors based on US and UK data Another important paper pub-lished on this research was by ChenRoll and Ross (1986) who analysed whetherinnovations in the macroeconomic variables are risks that are awarded in thestock markets They found that macroeconomic variables like spread betweenlong and short interest rates expected and unexpected ination Industrial pro-duction are some of the factors that are awarded by the markets Further theArbitrage pricing theory (APT) of Ross (1976) posits relation between stockprices and certain macro-economic variables In the last decade or so the focusfor these kind of studies have started to shift from developed world economies todeveloping world economies As developing world economies have shown signsEDHEC Business School 71 INTRODUCTIONof huge growth potential and leading the economies globally out of recessionsthis motivates us to research on developing markets like India Such a studywill help us to _nd the relation between stock market and macroeconomic indi-cators and give a new insight to foreign investors academicianspolicy makerstraders and domestic investorsThis study is important in a sense it provides an insight to how are Indian stockmarkets are related to its macroeconomic variables and global macromicro eco-nomic factors This study will also help us in analysing whether the Indian stockmarkets have become coupled to global factors or are they still dominated bydomestic economic factorsThe focus of this study is on relation between Indian stock market representedby BSE Sensex and domestic macroeconomic factors and global factors repre-sented by Standard and Poors 500 Index This study builds on earlier studiesdone in this area but also open some new doors for further research It is sim-ilar to some earlier studies in a respect that it uses data macro and micro

factors and econometrics tools used in previous studies but at the same time itdi_erentiates itself from earlier studies in a sense that it is done on a marketthat is still developing Also the time period used in the analysis is a periodwhere Indian market has undergone lot of regulatory changes that has createda structural change in the market Further in this study Ill analyse whetherthe Indian markets are driven mainly by Domestic factors or do global factorshave more inuence on Indian markets To analyse the impact of internationalfactors Ill use Standard and Poors 500 Index and USDINR exchange rate as asubstitute of global factors and to model domestic demand Ill use macro factorslike Industrial production M1 money supply Consumer Price Index and Pro-ducer price Index The outline of the thesis is as followings Section 2 providesa literature review of the studies done earlier in this area Section 3 provides adetailed description of the data used in the study Section 4 provides a detaileddescription of the methodology and various econometric tools that will be usedin the study Section 5 provides the results of the study and Section 6 providesthe conclusion of the studyEDHEC Business School 82 LITERATURE REVIEW2 Literature ReviewMany studies and researchers have tried to _nd factors that can explain stockreturns The most famous and earliest model is the Capital Asset Pricing Model(CAPM) developed by Sharpe (1964) Lintner (1965) Mossin (1967) and Black(1972) The concept of this single factor model is developed from diversi_-cation introduced by Markowitz (1952) In CAPM model the expected stockreturns can be explained with the help of Risk free rate and one risk factorMarket CAPM says that the systematic risk can be captured by sensitivenessof each stock to change in overall market which is measured by Beta Accordingto CAPM the market factor is the only factor determining the stock returnsCAPM was a revolutionary model It changed the way people looked at thestock returns as something that is vary arbitrary As it is very easy to under-stand and use CAPM is very popular as the model used to determine the stockreturn in most of _nance textbooks and used by many practitioners in stockmarketHowever the numerous set of assumptions made in deriving CAPM made itinconsistent with the real world and led to criticism of CAPM To overcomethe limitations and assumptions made in CAPM many scholars came up withmulti- factor models like Fama-French three factor model APT model etc InFama-French model they try to explain stock returns with help of three factorsmarketsmall minus big and value minus growth the model was able to explainthe returns based on these risk factors for some time before it failed Therehave been many studies on failure of Fama-French model and markets where itis not applicableThe macroeconomic models of explaining stock returns started with APT (Ar-bitrage Pricing Theory) by Ross (1976) which was later re_ned by Roll andRoss (1980) APT is a multi-factor model and claims that the stock return canbe explained by unexpected changes or shocks in multiple factors ChenRolland Ross (1986) perform the empirical study for APT model and identify thatsurprise or shock in macroeconomic variables can explain the stock return sig-ni_cantly The variables used in their study are Industrial production indexdefault risk premium that can measure the con_dence of investors and changein yield curve that can be measured by term premiumThe study of macroeconomic factors in explaining stock returns have been pop-ular since then Stock price is present value of all discounted future cash owsIf a _rm is performing well then the expectation of large future cash ows risesand consequently the stock price rises On the other hand if a _rm is performingbad for couple of years then the expectation of big future cash ows decreaseand in turn the stock price fall This is a micro and idiosyncratic explanation ofstock prices and returns But the future cash ows of a stock does not dependsolely on the companys performance or pro_tsloss The systematic factor can

have a huge impact on the cash ows of not only one but many companies Thesystematic factor here refers to macro economic variables The state of Macroeconomic conditions lead to changes in Monetary and regulatory policies by thegovernment and which in turn a_ects the stock prices For example a countrywith good economic conditions represented by its Industrial production indexGDP CPI Interest rates will create an environment that is conducive for thegrowth of companies by lowering borrowing rates and other open market opera-tions So all macroeconomic factors that can inuence future cash ows or theEDHEC Business School 92 LITERATURE REVIEWdiscount rate by which the cash ows are discounted should have an inuenceon the stock priceMany researcher have studies the relationship between stock prices and macroeconomic variables and tried to explain the a_ect of one over the other Fama(1981) tries to establish a relationship between stock returns real activity ina-tion and money In his paper he _nds that Stock returns have positive relationwith real output and money supply but a negative relation with ination Heexplains that negative relation between stock returns and ination is induced bynegative relation between real output approximated by Industrial productionand ination This negative relationship between ination and real activityis explained by money demand theory and quantity theory of money Fama(1990) explains that measuring the total return variation explained by shocksto expected cash ows time-varying expected returns and shocks to expectedreturns is one way to judge the rationality of stock prices In his paper he_nds that growth rates of production used to proxy for shocks to expected cashows explain 45 of return variance ChenRoll and Ross (1986) explored therelationship between a set of economic variables and their systematic inuenceon stock market returns They found that Industrial production changes inrisk premium twists in yield curve had strong relationship and impact on stockreturns A somewhat weaker e_ect was found for measures of unanticipatedination and changes in expected ination during periods when these variableswere highly volatile They concluded that stock returns were exposed to sys-tematic economic news that they are priced in accordance to their exposuresand that the news can be measured as innovation in state variables Chen(1991) found that state variables that are priced are those that can forecastchanges in the investment and consumption opportunity set According to hisresearch default spread the term spread the one-month T-Bill rate the laggedindustrial production growth rate and the dividend-price ration are importantdeterminants of future stock market returns Bulmash and Trivoli (1991) showthe e_ect of business cycle movements on the relationship between stock returnsand money growthAn interesting paper in this _eld of research is by Fama (1990) and Schwert(1990) In the paper they claim that there are three explanations for the stronglink between stock prices and real economic activityFirst information about the future real activity may be reected in stockprices well before it occurs|this is essentially the notion that stock pricesare a leading indicator for the well-being of the economy Second changesin discount rates may a_ect stock prices and real investment similarly butthe output from real investment doesnt appear for some time after it ismade Third changes in stock prices are changes in wealth and thiscan a_ect the demand for consumption and investment goods [Schwert(1990)p1237]Campbell and Ammer (1993) use a VAR approach to model the simulta-neous interactions between the stock and bond markets since most previousworks do not address the channels through which the macroeconomic activityinuences the stock prices One example could be that industrial productioncould be linked to changing expectations of future cash ows (Balvers at al1990) On the other hand interest rate innovations could be the driving factorEDHEC Business School 102 LITERATURE REVIEW

in determining both industrial production (due to change in investment) andstock prices (due to change in the discounted present value of future cash ows)A VAR analysis can distinguish these possibilities Mukherjee and Naka (1995)show a long-term relationship between the Japanese stock price and real macroe-conomic variables Dr Nishat (2004) studies the long term association amongmacroeconomic variables like money supply CPIIPI and foreign exchange rateand stock markets in Pakistan The results show that there are causal relation-ship among the stock price and macroeconomic variables He uses data from1974 to 2004 in his study As most of the _nancial time series are non station-ary in levels he uses unit root technique to make data stationary Fazal Hussianand Tariq Massod (2001) used variables like investment GDP and consumptionemploying Grangers causality test to _nd relationship between macro factorsand stock markets They show that at two lags all macroeconomic variableshave highly signi_cant e_ect on stock prices James et al (1985) use a VARMAanalysis for investigating relationship between macro economy and stock mar-ket Using VARMA analysis for _nding causal relationship between factors isa better technique as the procedure does not preclude any causal structure apriori since it allows feedback among variables Thus the VARMA approachallow whatever causal relationship exist to emerge from the data They _ndlinkages between real activity and stock returns and real activity and inationAlso they _nd that stock returns signal changes in the monetary base Sincestock returns also signal changes in expected real activity this suggests a linkbetween the money supply and expected real activity that is consistent with themoney supply explanation o_ered by Geske and RollIn recent years the focus of these kind of studies have shifted from developedeconomies to developing economies As developing economies are the economiesthat see a lot of structural and monetary policy changes an analysis of relation-ship between macro and micro can provide new insights Also one can analysethe e_ects of monetary policies on the asset prices especially on stock pricesTangjitprom (2012) study of macroeconomic factors like unemployment rateinterest rate ination rate and exchange rate and stock market of Thailand con-cludes that macroeconomic factors signi_cantly explain stock returns He also_nds that for Thailand unemployment rate and ination rate are insigni_cant todetermine the stock returns The reason he provides is that the unemploymentrate and ination rate are not timely and there could be some lags before thedata becomes available Also Grangers test to examine lead-lag relationshipamong the factors reveal that only few macroeconomic variables could predictthe future stock returns whereas the stock returns can predict most of futuremacro economic variables This implies that performance of stock markets canbe a leading indicator for future macroeconomic conditions Ali (2011) study ofimpact of macro and micro factors on stock returns reveals that ination andforeign remittance have negative inuence and industrial production index havepositive impact on stock markets Also he didnt found any Grangers Causal-ity between stock markets and any of the explanatory variables This lack ofGrangers causality reveals the evidence of informationally ine_cient marketsAli uses a multivariate regression analysis on standard OLD formula for estimat-ing the relationship Hosseini et al (2011) tested the relationship between stockmarkets and four macro economic variables namely crude oil prices Money sup-ply Industrial production and ination rate in China and India They used aperiod of 1999 to 2009 for analysis As most of the economic time series have unitEDHEC Business School 112 LITERATURE REVIEWroot they _rst used the Augmented Dickey Fuller unit root test and found theunderlying series to be non-stationary at levels but stationary after in di_erenceAlso the use of Jhonson-Juselius (1990) Multivariate cointegration and VectorError Correction model technique indicate that there are both long and shortrun linkages between macroeconomic variable and stock market index in each ofthe two countries Their analysis shows that in long run the impact of increasein prices of crude oil for China is positive but for India is negative In terms

of money supply the impact on Indian stock market is negative but for Chinathere is a positive impact The e_ect of Industrial production is negative onlyin China In addition the e_ect of increases in ination on these stock marketsis positive in both countries Wickremasinghe (2006) analysed the relationshipbetween stock prices and macroeconomic variables in Sri Lanka He used theUnit root tests Jhonsons test Error-correction model variance decomposi-tion and impulse response to analyse the relationships His _ndings indicatethat there is both long term and short term causal relationship between stockprices and macroeconomic variables in Sri Lanka The result indicate that thestock prices can be predicted from certain macroeconomic variables and henceviolate the validity of the semi-strong version of e_cient market hypothesisAhmed (2008) investigates the causal relationship between Indian macroeco-nomic factors like Industrial Production Exports Foreign direct investmentMoney supply exchange rate interest rate and stock market indices NSE NiftyIndex and BSE Sensex For _nding the long term relationship he applies Jo-hansens cointegration and Toda and Yamamoto Granger Causality tests Foranalysing the Impulse response and variance decomposition he uses bivariateVAR His _ndings reveal that stock prices in India lead macroeconomic activityexcept movement in interest rate Interest rate seem to lead the stock priceThe study also reveals that movement of stock prices is not only the outcomeof behaviour of key macro economic variables but it is also one of the causesof movement in other macro dimensions in the economy An important paperby Bilson et al (2001) argues that emerging markets local factors are moreimportant than global factors They _nd that for emerging markets are at leastpartially segmented from global capital markets The global factors are proxiedby world market returns and local factors by set of macro economic variableslike money supply prices real activity and exchange rate Some evidence isfound that local factors are signi_cant in their association with emerging equitymarket returns above than that explained by the world factor When they usea larger set of variables the explanatory power of the model improves substan-tially such that they are able to explain a large amount of return variation formost emerging marketsEDHEC Business School 123 DATA3 Data31 Description of Macroeconomic IndicatorsOne of the biggest problems when conducting a research with macroeconomicdata is the frequency of the data Most of the macroeconomic indicator timeseries are yearlyquarterly or monthly time series This low frequency of themacroeconomic indicators results in very few data points for conducting a anal-ysis that is robust A possible cure for the problem is to use longer time periodsto incorporate more data points for macroeconomic variables But anotherproblem that we face when we look at the macroeconomic indicators for Asiancountries is reporting of the data For most of the Asian countries the macroe-conomic data doesnt have a long history and same can be said about historyof Indian macroeconomic variables So in this research we have used a timeperiod for which we can _nd data for most of the macroeconomic indicators Inthis paper we use a time period of 20 years starting from 1990 to 2011 Thistime period in Indian economy is representative of many structural and mone-tary policy changes like liberalization of India markets Also as the time periodis long it gives us enough data point for each macroeconomic factors to do arobust empirical analysisWhen one starts to build a model of interaction between macro and micro eco-nomic factors one dominant and important question one faces is among themyriad of macro indicators available for an economy which factors to chooseto incorporate in the model If one chooses macroeconomic factors that arehighly correlated among themselves then the power of test results decrease asit may result in a model where the macro indicators are able to explain mostof the movement of micro factors but the macro factors may not be relevant

To circumvent this problem we use variables that have been tested in earlierresearches and that have been proven to have e_ect on stock markets I alsotest a few macro factors that have some _nancial theory behind them that con-nect them to stock markets Ali (2011) Wickremasinghe (2006) Bilson etal(2001) and Bailey (1996) _nd that Industrial production CPI exchange rateM1 money supply GDP are few of the macro economic factors that can signi_-cantly explain stock returns Sahu(2011) Ahmed(2008) Tripathy(2011) studyon Indian markets speci_cally show that Industrial Production Exchange rateInation index are macro economic indicators that have a strong positive ornegative relationship with the stock markets So in our study we test 5 macroeconomic variables namely M1 money supply Consumer and Producer price In-dex Industrial production Exchange rate The time period for these indicatorsis from 1990-2011 The data for Ination indices Industrial production andexchange rate has been pulled from Bloombergc and Datastreamc The datahas been processed for errors and missing values Data for M1 money supplyhas been pulled from RBI website For most of the indices like ination andIndustrial production index the base year has been changed to 1990 Also assome of the indices are in levels and some in actual _gures (M1 money supply)we convert all of the indicators to level form (starting at 100 in 1990)EDHEC Business School 133 DATA32 Description of Stock Market IndicesCompared to Macro Indicators stock market data is relatively easy to _nd andhas considerably long history Also the stock market data is a real time data soit has a very high frequency of seconds Here in our analysis we will make use ofBSE (Bombay Stock Exchange) as representation of Indian markets and SP500(Standard and Poors 500 Index) as representation of global factors BSE is amarket cap-weighted of 30 stocks It is the oldest Index in the Asian markets(established in 1875) and have had a long history We choose this index as it isthe Index that represent the most liquid and traded stocks of the Indian stockmarket Also the index is most traded index in India and a good representationof trade prices of the stocks Even in terms of an orderly growth much beforethe actual legislations were enacted BSE Limited had formulated a compre-hensive set of Rules and Regulations for the securities market It had also laiddown best practices which were adopted subsequently by 23 stock exchangeswhich were set up after India gained its independence Our choice of SP500 isbased on the fact that it has a long history and many researchers have usedthis index as a good proxy representation of global markets and economic con-ditions We will take the monthly returns of each of the indices from 1990-2011in accordance with data frequency of macro economic variables Also as theindices have di_erent levels at beginning of 1990 we rebase both the indices tobase year of 1990 starting at a level of 100EDHEC Business School 144 METHODOLOGY4 Methodology41 Construction of Time SeriesThe _rst step in constructing an econometric model is constructing time seriesall of which are in same units Most of the time series used in our analysis are indi_erent formats For example CPI PPI BSE Index SP500 are in levels M1money supply USDINR exchange rate is in absolute current format Industrialproduction is in absolute production levels So _rst we convert all of the giventime series to level The way we construct time series in levels is _rstly takingthe initial data point of each time series as 100 We then _nd the percentagechange from one period to the next one for each time series using a continuouscompounding assumption (taking a natural log of change in values) In math-ematical terms it can be stated as Assume the original Index value at time tto be It and at time t+1 to be It + 1 Then we can compute the new rebasedindex by formulaRIt+1 = RIt _ (1 + ln(It+1=It))

whereRIt= Rebased Index at time tRIt+1=Rebased Index at time t+1We can use these rebased indices in building and testing our econometric model42 Unit Root Test and StationarityUnit root test is to _nd whether the series is stationary or non-stationary Astrictly stationary process is one where for any t1 t2 tt 2Z any k 2Z andT=12Fyt1 yt2 yt3 ytT

(y1 yT ) = Fyt1+k yt2+k yt3+k ytT+k

(y1 yT )where F represents joint distribution function of the set of random variablesIt can also be stated that the probability measure of sequence of yt is same asyt+k for all k In other words a series is stationary if the distribution of its valueremain the same as time progresses Similar to the concept of strict stationaryis weakly stationary process A weakly stationary process is one which has aconstant mean variance and autocovariance structure Stationary is a necessarycondition for a time series to be tested in regression A non-stationary seriescan have several problems like1 The shocks given to the series would not die of gradually resulting inincrease of variance as time passes2 If the series is non stationary then it can lead to spurious regressions If twoseries are generated independent of each other then if one is regressed onother it will result in very low R2 values But if two series are trending overtime then a regression of one over the other will give high R2 even thoughthe series may be unrelated to each other So if normal regressions toolsEDHEC Business School 154 METHODOLOGYare used on non stationary data then it may result in good but valuelessresults3 If the variables employed in a regression model are not stationary thenit can be proved that the standard assumptions for asymptotic analysiswill not be valid In other words the usual t-ratios will not follow at-distribution and the F-statistic will not follow an F-distribution and soonStationarity is a desirable condition for any time series so that it can be usedin regressions and give meaningful result that have some value to test for sta-tionarity a quick and dirty way is looking at the autocorrelation and partialcorrelation function of the series If the series is stationary then the autocorre-lation function should die o_ gradually after few lags and the partial correlationfunction will me non zero for some lags and zero thereafter Also we can usethe Ljung-Box test for testing that all m of _k autocorrelation coe_cients arezero using Q-statistic given by formulaQ = T(T + 2)_mk=1_k2T 1048576 k_ _2where T = Sample size and m = Maximum lag lengthThe lag length selection can be based on di_erent Information Criteria likeAkaikes Information criteria (AIC) Schwarzs Bayesian information criteria(SBIC) Hannan-Quinn criterion (HQIC) Mathematically di_erent criteria arerepresented asAIC = ln(_2) + 2kTSBIC = ln(_2) + kT lnTHQIC = ln(_2) + 2kT ln(ln(T))

For a better test for stationarity we use augmented Dickey fuller Unit roottest on each time series separately Augmented Dickey Fuller test is test ofnull hypothesis that the time series contains a unit roots against a alternativehypothesis that the series is stationary421 Mathematical representation of Stationary series and unit roottestAssume a variable Y whose structure can be given by AR process with no driftequationyt = _1yt10485761 + _2yt10485762 + _3yt10485763 + + _nyt1048576n + ut (2)where ut is the residual at time t Using a Lag operator L we can write eq(1)asyt = _1L1yt + _2L2yt + _3L3yt + + _nLnyt + ut (3)EDHEC Business School 164 METHODOLOGYRearranging eqn (2) we getyt 1048576 _1L1yt 1048576 _2L2yt 1048576 _3L3yt + 1048576 _nLnyt = ut (4)yt(1 1048576 _1L1 1048576 _2L2 1048576 _3L3 + 1048576 _nLn) = ut (5)or_(L)yt = ut (6)The time series is stationary if we can write eqn(5) in formyt = _(L)10485761ut (7)with _(L)10485761 converging to zero It means the autocorrelation function woulddecline as lag length is increased If eqn (6) is expanded to a MA(1) processthe coe_cients of residuals should decrease such that the the residuals that thee_ect of residuals decrease with increase in lags SO if the process is stationarythe coe_cients of residuals will converge to zero and for non-stationary seriesthey will and converge to zero and will have long term e_ect The condition fortesting of unit root for an AR process is that the roots of eqn(6) or Charac-teristic equation should lie outside unit circle422 Augmented Dickey Fuller Unit Root TestConsider an AR(1) process of variable Yyt = _yt10485761 + ut (8)Subtracting yt10485761 from both sides of eqn(7) we get_y = (_ 1048576 1)yt10485761 + ut (9)Eqn(8) is the test equation for Dickey Fuller test For Dickey-Fuller Unit roottestNull Hypothesis The value of _ is equal to 1 or value of _10485761 is equal to 0 vsAlternate Hypothesis The value of _ is less than one or value of _ 1048576 1 is lessthan zero Augmented Dickey-Fuller test is similar to normal Dickey-Fuller testsexcept it takes the lag structure of more than one into account_y = yt10485761 +Xpi=1_i_yt1048576i + ut (10)If the series has one or more unit root it is said to be integrated of order nwhere n is the number of unit roots of the characteristic equation To makethese time series stationary they needs to be di_erenced Mathematically ifyt _ I (n) (11)then_(d) yt _ I (0) (12)To make our time-series stationary we will use the natural log returns of theseseries in the analysisEDHEC Business School 174 METHODOLOGY43 Testing Long Term RelationshipsEngle and Granger (1987) in their seminal paper described cointegration whichforms the basis for testing for long term relationship between variables Accord-ing to Engle and Granger two variables are cointegrated if they are integratedprocess in their natural form (of the same order) but a weighted combination

of the variables can be found such that the combined new variable is integratedof order less than the order of individual time series Mathematically assumeyt to be a k X 1 vector of variables then the components are cointegrated orintegrated of order (db) if1 All components of yt are I(d)2 There is at least one vector of coe_cients _ such that_0

yt _ I (d 1048576 b) (13)As most of the _nancial time series are integrated of order one we will restrictourselves to case d=b=1 Two or more variables are said to be cointegrated ifthere exist a linear combination of these variables that is stationary Many ofthe series are non-stationary but move together over time which implies twoseries are bound by some common force or factor in long run We will test forcointegration by a residual-based approach and Johansens VAR methodResidual Based approach Consider a modelyt = _1 + _2x2t + _3x3t + + ut (14)where yt x2t x3t are all integrated of order N Now if the residual of this re-gression ut is stationary then we can say that the variables are cointegrated elsethere exist no long term relationship between the variables To test the resid-ual for stationarity we will run Augmented Dickey-Fuller tests on the residualsUnder the Null hypothesis the residual are integrated of order one or more andunder alternate hypothesis the residuals are I(0)431 Johansen test for CointegrationJohansen test for cointegration presents a better model for testing multiplecointegration among multiple variables The Residual based approach can only_nd atmost one cointegration and can be tested for a model with two variablesEven if more than two variables are present in the equation that are cointegratedthe Residual based approach will give only one cointegration SO we will useJhoansen VAR based cointegration for testing more than one cointegrationSuppose that a set of g variables are under consideration that are I(1) andwhich are thought to be cointegrated A VAR with k lags containing thesevariables could be set upyt = _1yt10485761 + _2yt10485762 + _ _ _ + _kyt1048576k + ut (15)g _ 1 g _ g g _ 1 g _ g g _ 1 g _ g g _ 1 g _ 1EDHEC Business School 184 METHODOLOGYIn order to use the Johansen test the VAR above should be turned into avector error correction model of form_yt = _yt1048576k + 1_yt10485761 + 2_yt10485762 + _ _ _ + k10485761_yt1048576(k10485761) + ut (16)where _ = (_ki=1_i) 1048576 Ig and i = (_ij=1_j) 1048576 IgThe Johansens test centers around testing the _ matrix which is the matrixthat represents the long term cointegration between the variables The test fornumber of cointegration is calculated by looking at the rank of the _ matrixthrough its eigenvalues The rank of the matrix is equal to number of roots(eigenvalues) _i of the matrix that are di_erent from zero The roots should beless than 1 in absolute value and positive If the variables are not cointegratedthe rank of the matrix will not be signi_cantly di_erent from zero ie _i _ 0There are two test statistics for Johansen test _tracer and _max_trace (r) = 1048576TPgi=r+1 ln(1 1048576 _ _i)and_max(r r + 1) = 1048576Tln(1 1048576 _r_+1)_trace is a test statistic for joint test where the null hypothesis is that thenumber of cointegration vector is less than or equal to r against an alternativethat there are more than r_max conducts another separate test on eigenvalues and has null hypothesis that

the number of cointegrating vector is r against r+144 Impulse ResponseOnce we have determined whether the variables have long term relationship ornot we can form a multivariate VAR model for the variables A multivariateVAR model between g variables is a model where the current value of a variabledepend on di_ernt combinations of the previous k values of all the variables anderror terms A general representation of the model can beyBSEt = _ + _BSEyBSE + _IP yIP + CPIyCPI + _M1yM1 + _SP500ySP500 + u1t(17)where all the coe_cients except _ are g _ k matrices and all variables y are k_ 1 matricesOnce we have formed a model like this we can use the model for Impulse re-sponse A VAR(p) model can be written as a linear fuction of the past innova-tions that isrt = _ + at + 1at10485761 + 2at10485762 + (18)where _ = [_(1)]10485761_0 provided that the inverse exists and the coe_cient ma-trices i can be obtained by equating the coe_cients of Bi in the equation(I 1048576 _1B 1048576 1048576 _PBP )(I + 1B + 2B2 + ) = I (19)EDHEC Business School 194 METHODOLOGYwhere I is the Identity martix This is a moving average representation of rtwith the coe_cient matrix i being the impact of the past innovation at1048576i onrt Equivalently i is the e_ect of at on the future observation rt+i Therefore i is often referred to as the Impulse Response Function of rt For our impulseresponse we will use equation of variables in _rst di_ernce form like_BSEt = _t +Xki=0_11(i)_BSEt1048576i +Xkj=1_12(j)_MIt1048576j + _BSEt (20)_MIt = _t +Xki=0_21(i)_MIt1048576i +Xkj=1_22(j)_BSEt1048576j + _MIt (21)Grangers causality and Blocks F test of a VAR model will suggest which ofthe variables have statistically signi_cant impacts on the future values of othervariables in the system But F-test results cannot explain the sign of the re-lationship nor how long these e_ects require to take place Such informationwill however be given by an examination of the VARs impulse responses andvariance decompositions Impulse response is a technique that trace out theresponsiveness of the dependent variable in the VAR to shocks of each of theother variables So for each variable from each equation separately we will applya unit shock to the error and trace the e_ects upon the VAR system over timeBy using the impulse response technique we can determine how responsive isthe BSE stock index to Indian macro indicators and SP500 This will help usdetermine whether the BSE index is more reactive to domestic news or globalnewsEDHEC Business School 205 RESULTS5 ResultsBefore we use the time series for VAR analysis or cointegration tests we need todetermine whether the series are Stationary or not If the series are stationaryin levels we can use them directly else we need to use the di_erenced time seriesOne way to look for autocorrelation or integrated process is to see the graphsof the various time series used Section 71 shows the graphs of variables we

use for our analysis As we can see from the graphs all of the time series havea trend in long run which points to an integrated process As a second stepwe plot the graphs of di_erenced time series in Section 52 We can see thatthe di_erenced graphs in Section 72 dont show a long term trend and crossthe X-axis frequently This is usually a property of I(1) processes So we checkthe series for autocorrelations at di_erent lag lengths Section 73 shows cor-relograms graph autocorrelation coe_cient partial autocorrelation coe_cientQ-Stat and p-value for various time series up to 36 lags As can be seen in thetables the Q-stat for all lags is zero and we can reject the joint null hypothesisthat all the autocorrelations up to 36 lags are zero Table 741 shows that ifwe conduct a Unit root test on levels of the series we _nd that all the 7 seriesare integrated as we cannot reject the t-stat for unit root at 1 level But ifwe conduct the same test on di_erenced values of the series we _nd that we canreject the null hypothesis of unit root at 1 signi_cance level for all the seriesexcept CPI This tells us that all the series are I(1) as there _rst di_erence seriesare I(0)As our series are I(1) we will work with index levels of time series to determineif there exist one or more cointegrating relationships between the series Tablesin subsection 743 are based on residual approach where we run a regression ofBSE and various macroeconomic indicators and test the residuals for unit rootusing Augmented Dickey-Fuller test As we assume the two series are cointe-grated we conduct the test with no trend and intercept If the two series arecointegrated then the errors should not have any trend or intercept We see thatwe can reject the null hypothesis of unit root at 1 signi_cance for CPIIP M1We can reject the null of unit root for PPI at 5 and for SP500 and USDINRwe cant reject the null hypothesis of unit root at even 5 level This pointsto the fact that BSE has a strong long term relationship with IP M1 moneysupply CPI at 1 level with IP M1 CPI PPI at 5 signi_cance level AlsoBSE has no long term relationship with SP500 and USD INR exchange rateTo test for multiple cointegrating relationship we now employ a Johansen VARbased cointegration test The results of the test are displayed in subsection744 The _rst panel of the test results displays the value of _trace and_maxof Johansen test with di_erent assumptions about intercept and trend We cansee from this panel that when we consider a functional form of intercept and noTrend we have atleast and atmost three cointegrating relationships The secondpanel of the results display the value of information criteria for lag lengths Formost of the models we see that Akalike criteria points to a lag of three andSchwarz criteria points to a lag of one To estimate the cointegrating model wechoose the model with intercept and no trend and run a cointegration testTestresults are shown in Table 2 of subsection 744 At 5 signi_cance level wecan reject the null of atmost two cointegrating factors for _trace and same for_max Now to test which all variables have a long tern relationship we perform aRestricted cointegration with vector error correction model As we had alreadyEDHEC Business School 215 RESULTSseen in our residual based test of cointegration that BSE has no cointegratingrelationship with SP500 and USDINR we create a restricted cointegration modelwhere we set coe_cients of SP500 and USDINR as zero The test results aredisplayed in Table 3 of subsection 744 In this case as there are two restrictionsthe test statistic follow _2 with two degrees of freedom We can see that thep-value for the test is 1333 which tells us that the restrictions are supportedby data at 10 level of signi_cance So we can conclude that the BSE has along term relationship with CPIIPPPIM1 money supply but has no long termrelationship with SP500 and USDINR exchange rate One interpretation of thisresult can be that the Indian stock market represented here by BSE Sensexmoves more in accordance with domestic factors like Industrial production M1money supply Consumer price index and Producer Price index than with globalfactors or in other words as BSE is representation of largest market cap Indiancompanies we can say that the biggest companies in India are ones that are

more dependent on domestic demand rather than exports This result presentsan opportunity for international investors to diversify their portfolio by invest-ing in BSE Sensex as it is decoupled with global markets and macroeconomicfactorsWe use A bivariate Vector Autoregression (BVAR) technique to analyze thedynamic interaction between real asset prices and macro economy VAR ispreferred method to study Macroeconomy and asset prices where variables en-dogenously e_ect each otherWe begin with a bivariate VAR with no restriction Asset prices and instru-ments are allowed to respond to each other freely For paired variables withcointegration relationship VAR is performed at levels whilst for those that arenot cointegrated VAR is performed at _rst di_erence Constant term is ignoredwith loss of generality We use the Bivariate Autoregression analysis for bothimpulse response and Grangers causality testsImpulse response results are displayed in subsection 745 From _rst graph ofimpulse response of BSE to USDINR we can see that USDINR has a negativeimpact on BSE As impulse response is response of BSE to shocks given to US-DINR we can see that a positive shock or unexpected appreciation INR valuewrt USD will have a negative e_ect on BSE for few lags and will disappearafter few lags If we look at the constituents of BSE Index over time we seethat most of the time some of its constituent are companies that thrive on ex-ports Some of the biggest Market-Cap in India are companies in service sectorlike Infosys TCS etc that are hugely dependent on services provided to clientsfrom Europe and US So an appreciation of INR compared to USD makesthese _rms costlier for the global clients and in turn reduces the income of thesecompanies As the _rms revenue pro_t decreases the value of the stock alsodecreases that in turn a_ects the returns of BSE SensexSecond graph (betwen BSE and SP500) shows that increase in SP500 has a pos-itive e_ect on BSE as higher returns of SP500 indicate strong global economywhich in turn results in higher trade between countries The positive responseof BSE to one unit shock to SP500 indicates a spillover e_ect of global factorson Indian economy but the response is weak as can be seen from the graphMoving forward response of BSE to shocks in M1 money supply CPI PPImake economic sense As for M1 money supply one unit shock means increasein M1 money supply This increase in money supply allows companies to bor-row more money from banks at lower rates which they can use for investingEDHEC Business School 225 RESULTSin pro_table projects and generating larger cash ows For Ination indicatorsone unit shock means increase in ination This increase in ination results inhigher costs for the companies that in turn reduces their pro_t margins and asa result value of stocksBy looking at the graphs we can also see that shocks to Indian macroeconomicindicators creates stronger response by BSE as compared to global factors likeSP500 or USDINR This indicates that BSE Index is driven by companies thatdepend hugely on domestic demand rather than exports Response of BSE toshocks to Industrial Production are contradictory to theory In theory an in-crease in industrial production should result in positive response from BSE butour analysis shows the other way A possible reason for this response could bethat industrial production time series is seasonal as can be seen from the graphSo there is a possibility of a leadlag relationship between the two variablesTo test for possibility of leadlag relationship we run a Grangers causality testbetween BSE and IP The result in section 646 shows that at a lag lengthof 4 we can reject the Null hypothesis of BSE does not Granger cause IP at1 signi_cance level This proves that BSE is a leading indicator of industrialproduction and there exist a leadlag relationship between the two indicatorsEDHEC Business School 236 CONCLUSIONS6 Conclusions

In this paper I tested the relations between Indian stock market represented byBSE and domestic and global macro economic factors The research concludesthat the India stock markets are mainly driven by domestic demand and theinuence of global macro factors on the stock market is weak I also tested forGranger causality between BSE and IP and found that BSE is a leading indicatorof Industrial production and can help in predicting the industrial climate inIndiaThe research is insightful for investors and professionals who are looking forinvestment opportunities to diversify their risks As Indian stock markets aremore dependent on domestic factors one can invest in Indian indices and stocksto diversify their risks gained through investing in US and European stocksThe paper opens new doors for research in this _eld One can use variancedecomposition technique to see how much variance of BSE can be explained myvarious domestic and global macro factors Also one can use di_erent globalfactors like sovereign CDS spreads T-Bill rates a composite indicator of globaleconomy for further research on interaction between Indian stock market andglobal economyOne can also research on how various global macroeconomicnews a_ects India stock markets and for how long the e_ects persistsEDHEC Business School 247 GRAPHS AND TABLES7 Graphs and Tables71 Graphs of Time seriesEDHEC Business School 257 GRAPHS AND TABLESEDHEC Business School 267 GRAPHS AND TABLESEDHEC Business School 277 GRAPHS AND TABLESEDHEC Business School 287 GRAPHS AND TABLES72 Graphs of Time Series - Di_erencedEDHEC Business School 297 GRAPHS AND TABLESEDHEC Business School 307 GRAPHS AND TABLESEDHEC Business School 317 GRAPHS AND TABLESEDHEC Business School 327 GRAPHS AND TABLES73 Correlograms of Time seriesBSEEDHEC Business School 337 GRAPHS AND TABLESIPEDHEC Business School 347 GRAPHS AND TABLESSP500EDHEC Business School 357 GRAPHS AND TABLESUSDINREDHEC Business School 367 GRAPHS AND TABLESCPIEDHEC Business School 377 GRAPHS AND TABLESPPIEDHEC Business School 387 GRAPHS AND TABLES

M1EDHEC Business School 397 GRAPHS AND TABLES74 Tables741 Table for Unit root test of Time seriesVariables T-Stat p-valueBSE -2671 2495 SP500 -1315 8818 CPI -1909 6466 IP -1669 899 M1 -2420 3679 PPI -3353 601 USDINR -2955 1469 742 Tables for Unit root test of Di_erenced time seriesVariables T-Stat p-valueBSE -13848 000 SP500 -14832 000 CPI -3344 140 IP -3865 027 M1 -3867 026 PPI -9656 000 USDINR -13701 000 743 Tables for Residual based test of cointegrationTable 1BSE - CPIt-Statistic ProbADF test statistic -2622676 087Test critical values 1 level -25738185 level -19420410 level -1615891Table 2BSE - IPt-Statistic ProbADF test statistic -3738802 002Test critical values 1 level -25745135 level -194213610 level -1615828EDHEC Business School 407 GRAPHS AND TABLESTable 3BSE - M1t-Statistic ProbADF test statistic -2875518 041Test critical values 1 level -25737845 level -194203510 level -1615894Table 4BSE - PPIt-Statistic ProbADF test statistic -2399055 162Test critical values 1 level -25737845 level -194203510 level -1615894Table 5BSE - SP500t-Statistic ProbADF test statistic -1427184 1430Test critical values 1 level -25737845 level -1942035

10 level -1615894EDHEC Business School 417 GRAPHS AND TABLESTable 6BSE - USDINRt-Statistic ProbADF test statistic -1659522 917Test critical values 1 level -25738185 level -19420410 level -1615891EDHEC Business School 427 GRAPHS AND TABLES744 Johansen cointegration testEDHEC Business School 437 GRAPHS AND TABLESTable 2EDHEC Business School 447 GRAPHS AND TABLESTable 3EDHEC Business School 457 GRAPHS AND TABLES745 Impulse response testsEDHEC Business School 467 GRAPHS AND TABLESEDHEC Business School 477 GRAPHS AND TABLESEDHEC Business School 487 GRAPHS AND TABLES746 Granger causality test between IP and BSEEDHEC Business School 498 BIBLIOGRAPHY8 BibliographyEugene F Fama Ination Output and Money Journal of Business 1982Eugene F Fama Stock Returns Real activity and Money The American Eco-nomic Review 1981Eugene F Fama Stock Returns Expected Returns and Real activity Journal ofFinance 1990Pal and Mittal Impact of macroeconomic indicators in Indian capital marketsJournal of Risk Finance 2011Shahid Ahmed Aggregate Economic Variables and Stock Markets in India In-ternational Research Journal of Finance and Economics 2008Sahu and Dhiman Correlation and Causality between Stock Market and MacroEconomic Variables in India An Empirical Study 2010 International Confer-ence on E-Business and Economics 2011Mohammad Bayezid Ali Impact of Micro Variables on Emerging Stock MarketReturn A case on Dhaka Stock Exchange (DSE) Interdisciplinary Journal ofResearch in Business 2011Napphon Tangjitprom Macroeconomic Factors of Emerging Stock Market Theevidence from Thailand International Journal of Finance and Research 2012Sayed Mehdi Hosseini The Role of Macroeconomic Variables on Stock MarketIndex in China and India International Journal of Economics and Finance2011John Y Campbell Pitfalls and Opportunities What Macroeconomists shouldknow about Unit Roots NBER Working Papers 1991Hacker and Hatemi The properties of Procedures Dealing with Uncertainityabout Intercept and Deterministic Trend in Unit Root Testing CESIS Elec-tronic Working Papers 2010Elder and Kennedy Testing for Unit Roots What should Students be TaughtNasseh and Strauss Stock Prices and domestic and international macroeco-

nomic activity a cointegration approach The Quarterly Review of Economicsand Finance 2000Engle and Granger Co-Integration and Error Correction Representation Es-timation and Testing Econometrica 1987Eugene F Fama Stock Returns Real Activity Ination and Money 1981American Economic AssociationNaliniprave Tripathy Causal Relationship between Macro-Economic Indicatorsand Stock Market in India Asian Journal of Finance and Accounting 2011Rogalski and Vinso Stock Returns Money Supply and the Direction of Causal-ity The Journal of Finance 1977James et al A VARMA Analysis of the Causal Relations Among Stock Re-turns Real output and Nominal Interest Rates 1985 The Journal of FinanceBailey and Chung Risk and return in the Philippine Equity market A multi-factor exploration Paci_c-Basin Finance Journal 1996Nai-Fu Chen Financial Investment opportunities and the Macroeconomy TheJournal of Finance 1991GB Wickremasinghe Macroeconomic forces and stock prices Some empiricalevidence from an emerging stock markets University of Wollongong 2006EDHEC Business School 508 BIBLIOGRAPHYYao Juo and Loh On Chinas Monetary Policy and Asset Prices University ofNottingham- China policy Institute 2011Bilson et al Selecting macroeconomic variables as explanatory factors of emerg-ing stock market returns Paci_c-Basin Finance Journal 2001CHen Roll and Ross Economic forces and the Stock Markets The Journal ofBusiness 1986William H Greene Econometric Analysis 6th Edition Pearson InternationalEditionRuey Tsay Analysis of Financial Time seriesChris Brooks Introductory Econometrics for Finance Cambridge PublicationsEDHEC Business School 51

factors and econometrics tools used in previous studies but at the same time itdi_erentiates itself from earlier studies in a sense that it is done on a marketthat is still developing Also the time period used in the analysis is a periodwhere Indian market has undergone lot of regulatory changes that has createda structural change in the market Further in this study Ill analyse whetherthe Indian markets are driven mainly by Domestic factors or do global factorshave more inuence on Indian markets To analyse the impact of internationalfactors Ill use Standard and Poors 500 Index and USDINR exchange rate as asubstitute of global factors and to model domestic demand Ill use macro factorslike Industrial production M1 money supply Consumer Price Index and Pro-ducer price Index The outline of the thesis is as followings Section 2 providesa literature review of the studies done earlier in this area Section 3 provides adetailed description of the data used in the study Section 4 provides a detaileddescription of the methodology and various econometric tools that will be usedin the study Section 5 provides the results of the study and Section 6 providesthe conclusion of the studyEDHEC Business School 82 LITERATURE REVIEW2 Literature ReviewMany studies and researchers have tried to _nd factors that can explain stockreturns The most famous and earliest model is the Capital Asset Pricing Model(CAPM) developed by Sharpe (1964) Lintner (1965) Mossin (1967) and Black(1972) The concept of this single factor model is developed from diversi_-cation introduced by Markowitz (1952) In CAPM model the expected stockreturns can be explained with the help of Risk free rate and one risk factorMarket CAPM says that the systematic risk can be captured by sensitivenessof each stock to change in overall market which is measured by Beta Accordingto CAPM the market factor is the only factor determining the stock returnsCAPM was a revolutionary model It changed the way people looked at thestock returns as something that is vary arbitrary As it is very easy to under-stand and use CAPM is very popular as the model used to determine the stockreturn in most of _nance textbooks and used by many practitioners in stockmarketHowever the numerous set of assumptions made in deriving CAPM made itinconsistent with the real world and led to criticism of CAPM To overcomethe limitations and assumptions made in CAPM many scholars came up withmulti- factor models like Fama-French three factor model APT model etc InFama-French model they try to explain stock returns with help of three factorsmarketsmall minus big and value minus growth the model was able to explainthe returns based on these risk factors for some time before it failed Therehave been many studies on failure of Fama-French model and markets where itis not applicableThe macroeconomic models of explaining stock returns started with APT (Ar-bitrage Pricing Theory) by Ross (1976) which was later re_ned by Roll andRoss (1980) APT is a multi-factor model and claims that the stock return canbe explained by unexpected changes or shocks in multiple factors ChenRolland Ross (1986) perform the empirical study for APT model and identify thatsurprise or shock in macroeconomic variables can explain the stock return sig-ni_cantly The variables used in their study are Industrial production indexdefault risk premium that can measure the con_dence of investors and changein yield curve that can be measured by term premiumThe study of macroeconomic factors in explaining stock returns have been pop-ular since then Stock price is present value of all discounted future cash owsIf a _rm is performing well then the expectation of large future cash ows risesand consequently the stock price rises On the other hand if a _rm is performingbad for couple of years then the expectation of big future cash ows decreaseand in turn the stock price fall This is a micro and idiosyncratic explanation ofstock prices and returns But the future cash ows of a stock does not dependsolely on the companys performance or pro_tsloss The systematic factor can

have a huge impact on the cash ows of not only one but many companies Thesystematic factor here refers to macro economic variables The state of Macroeconomic conditions lead to changes in Monetary and regulatory policies by thegovernment and which in turn a_ects the stock prices For example a countrywith good economic conditions represented by its Industrial production indexGDP CPI Interest rates will create an environment that is conducive for thegrowth of companies by lowering borrowing rates and other open market opera-tions So all macroeconomic factors that can inuence future cash ows or theEDHEC Business School 92 LITERATURE REVIEWdiscount rate by which the cash ows are discounted should have an inuenceon the stock priceMany researcher have studies the relationship between stock prices and macroeconomic variables and tried to explain the a_ect of one over the other Fama(1981) tries to establish a relationship between stock returns real activity ina-tion and money In his paper he _nds that Stock returns have positive relationwith real output and money supply but a negative relation with ination Heexplains that negative relation between stock returns and ination is induced bynegative relation between real output approximated by Industrial productionand ination This negative relationship between ination and real activityis explained by money demand theory and quantity theory of money Fama(1990) explains that measuring the total return variation explained by shocksto expected cash ows time-varying expected returns and shocks to expectedreturns is one way to judge the rationality of stock prices In his paper he_nds that growth rates of production used to proxy for shocks to expected cashows explain 45 of return variance ChenRoll and Ross (1986) explored therelationship between a set of economic variables and their systematic inuenceon stock market returns They found that Industrial production changes inrisk premium twists in yield curve had strong relationship and impact on stockreturns A somewhat weaker e_ect was found for measures of unanticipatedination and changes in expected ination during periods when these variableswere highly volatile They concluded that stock returns were exposed to sys-tematic economic news that they are priced in accordance to their exposuresand that the news can be measured as innovation in state variables Chen(1991) found that state variables that are priced are those that can forecastchanges in the investment and consumption opportunity set According to hisresearch default spread the term spread the one-month T-Bill rate the laggedindustrial production growth rate and the dividend-price ration are importantdeterminants of future stock market returns Bulmash and Trivoli (1991) showthe e_ect of business cycle movements on the relationship between stock returnsand money growthAn interesting paper in this _eld of research is by Fama (1990) and Schwert(1990) In the paper they claim that there are three explanations for the stronglink between stock prices and real economic activityFirst information about the future real activity may be reected in stockprices well before it occurs|this is essentially the notion that stock pricesare a leading indicator for the well-being of the economy Second changesin discount rates may a_ect stock prices and real investment similarly butthe output from real investment doesnt appear for some time after it ismade Third changes in stock prices are changes in wealth and thiscan a_ect the demand for consumption and investment goods [Schwert(1990)p1237]Campbell and Ammer (1993) use a VAR approach to model the simulta-neous interactions between the stock and bond markets since most previousworks do not address the channels through which the macroeconomic activityinuences the stock prices One example could be that industrial productioncould be linked to changing expectations of future cash ows (Balvers at al1990) On the other hand interest rate innovations could be the driving factorEDHEC Business School 102 LITERATURE REVIEW

in determining both industrial production (due to change in investment) andstock prices (due to change in the discounted present value of future cash ows)A VAR analysis can distinguish these possibilities Mukherjee and Naka (1995)show a long-term relationship between the Japanese stock price and real macroe-conomic variables Dr Nishat (2004) studies the long term association amongmacroeconomic variables like money supply CPIIPI and foreign exchange rateand stock markets in Pakistan The results show that there are causal relation-ship among the stock price and macroeconomic variables He uses data from1974 to 2004 in his study As most of the _nancial time series are non station-ary in levels he uses unit root technique to make data stationary Fazal Hussianand Tariq Massod (2001) used variables like investment GDP and consumptionemploying Grangers causality test to _nd relationship between macro factorsand stock markets They show that at two lags all macroeconomic variableshave highly signi_cant e_ect on stock prices James et al (1985) use a VARMAanalysis for investigating relationship between macro economy and stock mar-ket Using VARMA analysis for _nding causal relationship between factors isa better technique as the procedure does not preclude any causal structure apriori since it allows feedback among variables Thus the VARMA approachallow whatever causal relationship exist to emerge from the data They _ndlinkages between real activity and stock returns and real activity and inationAlso they _nd that stock returns signal changes in the monetary base Sincestock returns also signal changes in expected real activity this suggests a linkbetween the money supply and expected real activity that is consistent with themoney supply explanation o_ered by Geske and RollIn recent years the focus of these kind of studies have shifted from developedeconomies to developing economies As developing economies are the economiesthat see a lot of structural and monetary policy changes an analysis of relation-ship between macro and micro can provide new insights Also one can analysethe e_ects of monetary policies on the asset prices especially on stock pricesTangjitprom (2012) study of macroeconomic factors like unemployment rateinterest rate ination rate and exchange rate and stock market of Thailand con-cludes that macroeconomic factors signi_cantly explain stock returns He also_nds that for Thailand unemployment rate and ination rate are insigni_cant todetermine the stock returns The reason he provides is that the unemploymentrate and ination rate are not timely and there could be some lags before thedata becomes available Also Grangers test to examine lead-lag relationshipamong the factors reveal that only few macroeconomic variables could predictthe future stock returns whereas the stock returns can predict most of futuremacro economic variables This implies that performance of stock markets canbe a leading indicator for future macroeconomic conditions Ali (2011) study ofimpact of macro and micro factors on stock returns reveals that ination andforeign remittance have negative inuence and industrial production index havepositive impact on stock markets Also he didnt found any Grangers Causal-ity between stock markets and any of the explanatory variables This lack ofGrangers causality reveals the evidence of informationally ine_cient marketsAli uses a multivariate regression analysis on standard OLD formula for estimat-ing the relationship Hosseini et al (2011) tested the relationship between stockmarkets and four macro economic variables namely crude oil prices Money sup-ply Industrial production and ination rate in China and India They used aperiod of 1999 to 2009 for analysis As most of the economic time series have unitEDHEC Business School 112 LITERATURE REVIEWroot they _rst used the Augmented Dickey Fuller unit root test and found theunderlying series to be non-stationary at levels but stationary after in di_erenceAlso the use of Jhonson-Juselius (1990) Multivariate cointegration and VectorError Correction model technique indicate that there are both long and shortrun linkages between macroeconomic variable and stock market index in each ofthe two countries Their analysis shows that in long run the impact of increasein prices of crude oil for China is positive but for India is negative In terms

of money supply the impact on Indian stock market is negative but for Chinathere is a positive impact The e_ect of Industrial production is negative onlyin China In addition the e_ect of increases in ination on these stock marketsis positive in both countries Wickremasinghe (2006) analysed the relationshipbetween stock prices and macroeconomic variables in Sri Lanka He used theUnit root tests Jhonsons test Error-correction model variance decomposi-tion and impulse response to analyse the relationships His _ndings indicatethat there is both long term and short term causal relationship between stockprices and macroeconomic variables in Sri Lanka The result indicate that thestock prices can be predicted from certain macroeconomic variables and henceviolate the validity of the semi-strong version of e_cient market hypothesisAhmed (2008) investigates the causal relationship between Indian macroeco-nomic factors like Industrial Production Exports Foreign direct investmentMoney supply exchange rate interest rate and stock market indices NSE NiftyIndex and BSE Sensex For _nding the long term relationship he applies Jo-hansens cointegration and Toda and Yamamoto Granger Causality tests Foranalysing the Impulse response and variance decomposition he uses bivariateVAR His _ndings reveal that stock prices in India lead macroeconomic activityexcept movement in interest rate Interest rate seem to lead the stock priceThe study also reveals that movement of stock prices is not only the outcomeof behaviour of key macro economic variables but it is also one of the causesof movement in other macro dimensions in the economy An important paperby Bilson et al (2001) argues that emerging markets local factors are moreimportant than global factors They _nd that for emerging markets are at leastpartially segmented from global capital markets The global factors are proxiedby world market returns and local factors by set of macro economic variableslike money supply prices real activity and exchange rate Some evidence isfound that local factors are signi_cant in their association with emerging equitymarket returns above than that explained by the world factor When they usea larger set of variables the explanatory power of the model improves substan-tially such that they are able to explain a large amount of return variation formost emerging marketsEDHEC Business School 123 DATA3 Data31 Description of Macroeconomic IndicatorsOne of the biggest problems when conducting a research with macroeconomicdata is the frequency of the data Most of the macroeconomic indicator timeseries are yearlyquarterly or monthly time series This low frequency of themacroeconomic indicators results in very few data points for conducting a anal-ysis that is robust A possible cure for the problem is to use longer time periodsto incorporate more data points for macroeconomic variables But anotherproblem that we face when we look at the macroeconomic indicators for Asiancountries is reporting of the data For most of the Asian countries the macroe-conomic data doesnt have a long history and same can be said about historyof Indian macroeconomic variables So in this research we have used a timeperiod for which we can _nd data for most of the macroeconomic indicators Inthis paper we use a time period of 20 years starting from 1990 to 2011 Thistime period in Indian economy is representative of many structural and mone-tary policy changes like liberalization of India markets Also as the time periodis long it gives us enough data point for each macroeconomic factors to do arobust empirical analysisWhen one starts to build a model of interaction between macro and micro eco-nomic factors one dominant and important question one faces is among themyriad of macro indicators available for an economy which factors to chooseto incorporate in the model If one chooses macroeconomic factors that arehighly correlated among themselves then the power of test results decrease asit may result in a model where the macro indicators are able to explain mostof the movement of micro factors but the macro factors may not be relevant

To circumvent this problem we use variables that have been tested in earlierresearches and that have been proven to have e_ect on stock markets I alsotest a few macro factors that have some _nancial theory behind them that con-nect them to stock markets Ali (2011) Wickremasinghe (2006) Bilson etal(2001) and Bailey (1996) _nd that Industrial production CPI exchange rateM1 money supply GDP are few of the macro economic factors that can signi_-cantly explain stock returns Sahu(2011) Ahmed(2008) Tripathy(2011) studyon Indian markets speci_cally show that Industrial Production Exchange rateInation index are macro economic indicators that have a strong positive ornegative relationship with the stock markets So in our study we test 5 macroeconomic variables namely M1 money supply Consumer and Producer price In-dex Industrial production Exchange rate The time period for these indicatorsis from 1990-2011 The data for Ination indices Industrial production andexchange rate has been pulled from Bloombergc and Datastreamc The datahas been processed for errors and missing values Data for M1 money supplyhas been pulled from RBI website For most of the indices like ination andIndustrial production index the base year has been changed to 1990 Also assome of the indices are in levels and some in actual _gures (M1 money supply)we convert all of the indicators to level form (starting at 100 in 1990)EDHEC Business School 133 DATA32 Description of Stock Market IndicesCompared to Macro Indicators stock market data is relatively easy to _nd andhas considerably long history Also the stock market data is a real time data soit has a very high frequency of seconds Here in our analysis we will make use ofBSE (Bombay Stock Exchange) as representation of Indian markets and SP500(Standard and Poors 500 Index) as representation of global factors BSE is amarket cap-weighted of 30 stocks It is the oldest Index in the Asian markets(established in 1875) and have had a long history We choose this index as it isthe Index that represent the most liquid and traded stocks of the Indian stockmarket Also the index is most traded index in India and a good representationof trade prices of the stocks Even in terms of an orderly growth much beforethe actual legislations were enacted BSE Limited had formulated a compre-hensive set of Rules and Regulations for the securities market It had also laiddown best practices which were adopted subsequently by 23 stock exchangeswhich were set up after India gained its independence Our choice of SP500 isbased on the fact that it has a long history and many researchers have usedthis index as a good proxy representation of global markets and economic con-ditions We will take the monthly returns of each of the indices from 1990-2011in accordance with data frequency of macro economic variables Also as theindices have di_erent levels at beginning of 1990 we rebase both the indices tobase year of 1990 starting at a level of 100EDHEC Business School 144 METHODOLOGY4 Methodology41 Construction of Time SeriesThe _rst step in constructing an econometric model is constructing time seriesall of which are in same units Most of the time series used in our analysis are indi_erent formats For example CPI PPI BSE Index SP500 are in levels M1money supply USDINR exchange rate is in absolute current format Industrialproduction is in absolute production levels So _rst we convert all of the giventime series to level The way we construct time series in levels is _rstly takingthe initial data point of each time series as 100 We then _nd the percentagechange from one period to the next one for each time series using a continuouscompounding assumption (taking a natural log of change in values) In math-ematical terms it can be stated as Assume the original Index value at time tto be It and at time t+1 to be It + 1 Then we can compute the new rebasedindex by formulaRIt+1 = RIt _ (1 + ln(It+1=It))

whereRIt= Rebased Index at time tRIt+1=Rebased Index at time t+1We can use these rebased indices in building and testing our econometric model42 Unit Root Test and StationarityUnit root test is to _nd whether the series is stationary or non-stationary Astrictly stationary process is one where for any t1 t2 tt 2Z any k 2Z andT=12Fyt1 yt2 yt3 ytT

(y1 yT ) = Fyt1+k yt2+k yt3+k ytT+k

(y1 yT )where F represents joint distribution function of the set of random variablesIt can also be stated that the probability measure of sequence of yt is same asyt+k for all k In other words a series is stationary if the distribution of its valueremain the same as time progresses Similar to the concept of strict stationaryis weakly stationary process A weakly stationary process is one which has aconstant mean variance and autocovariance structure Stationary is a necessarycondition for a time series to be tested in regression A non-stationary seriescan have several problems like1 The shocks given to the series would not die of gradually resulting inincrease of variance as time passes2 If the series is non stationary then it can lead to spurious regressions If twoseries are generated independent of each other then if one is regressed onother it will result in very low R2 values But if two series are trending overtime then a regression of one over the other will give high R2 even thoughthe series may be unrelated to each other So if normal regressions toolsEDHEC Business School 154 METHODOLOGYare used on non stationary data then it may result in good but valuelessresults3 If the variables employed in a regression model are not stationary thenit can be proved that the standard assumptions for asymptotic analysiswill not be valid In other words the usual t-ratios will not follow at-distribution and the F-statistic will not follow an F-distribution and soonStationarity is a desirable condition for any time series so that it can be usedin regressions and give meaningful result that have some value to test for sta-tionarity a quick and dirty way is looking at the autocorrelation and partialcorrelation function of the series If the series is stationary then the autocorre-lation function should die o_ gradually after few lags and the partial correlationfunction will me non zero for some lags and zero thereafter Also we can usethe Ljung-Box test for testing that all m of _k autocorrelation coe_cients arezero using Q-statistic given by formulaQ = T(T + 2)_mk=1_k2T 1048576 k_ _2where T = Sample size and m = Maximum lag lengthThe lag length selection can be based on di_erent Information Criteria likeAkaikes Information criteria (AIC) Schwarzs Bayesian information criteria(SBIC) Hannan-Quinn criterion (HQIC) Mathematically di_erent criteria arerepresented asAIC = ln(_2) + 2kTSBIC = ln(_2) + kT lnTHQIC = ln(_2) + 2kT ln(ln(T))

For a better test for stationarity we use augmented Dickey fuller Unit roottest on each time series separately Augmented Dickey Fuller test is test ofnull hypothesis that the time series contains a unit roots against a alternativehypothesis that the series is stationary421 Mathematical representation of Stationary series and unit roottestAssume a variable Y whose structure can be given by AR process with no driftequationyt = _1yt10485761 + _2yt10485762 + _3yt10485763 + + _nyt1048576n + ut (2)where ut is the residual at time t Using a Lag operator L we can write eq(1)asyt = _1L1yt + _2L2yt + _3L3yt + + _nLnyt + ut (3)EDHEC Business School 164 METHODOLOGYRearranging eqn (2) we getyt 1048576 _1L1yt 1048576 _2L2yt 1048576 _3L3yt + 1048576 _nLnyt = ut (4)yt(1 1048576 _1L1 1048576 _2L2 1048576 _3L3 + 1048576 _nLn) = ut (5)or_(L)yt = ut (6)The time series is stationary if we can write eqn(5) in formyt = _(L)10485761ut (7)with _(L)10485761 converging to zero It means the autocorrelation function woulddecline as lag length is increased If eqn (6) is expanded to a MA(1) processthe coe_cients of residuals should decrease such that the the residuals that thee_ect of residuals decrease with increase in lags SO if the process is stationarythe coe_cients of residuals will converge to zero and for non-stationary seriesthey will and converge to zero and will have long term e_ect The condition fortesting of unit root for an AR process is that the roots of eqn(6) or Charac-teristic equation should lie outside unit circle422 Augmented Dickey Fuller Unit Root TestConsider an AR(1) process of variable Yyt = _yt10485761 + ut (8)Subtracting yt10485761 from both sides of eqn(7) we get_y = (_ 1048576 1)yt10485761 + ut (9)Eqn(8) is the test equation for Dickey Fuller test For Dickey-Fuller Unit roottestNull Hypothesis The value of _ is equal to 1 or value of _10485761 is equal to 0 vsAlternate Hypothesis The value of _ is less than one or value of _ 1048576 1 is lessthan zero Augmented Dickey-Fuller test is similar to normal Dickey-Fuller testsexcept it takes the lag structure of more than one into account_y = yt10485761 +Xpi=1_i_yt1048576i + ut (10)If the series has one or more unit root it is said to be integrated of order nwhere n is the number of unit roots of the characteristic equation To makethese time series stationary they needs to be di_erenced Mathematically ifyt _ I (n) (11)then_(d) yt _ I (0) (12)To make our time-series stationary we will use the natural log returns of theseseries in the analysisEDHEC Business School 174 METHODOLOGY43 Testing Long Term RelationshipsEngle and Granger (1987) in their seminal paper described cointegration whichforms the basis for testing for long term relationship between variables Accord-ing to Engle and Granger two variables are cointegrated if they are integratedprocess in their natural form (of the same order) but a weighted combination

of the variables can be found such that the combined new variable is integratedof order less than the order of individual time series Mathematically assumeyt to be a k X 1 vector of variables then the components are cointegrated orintegrated of order (db) if1 All components of yt are I(d)2 There is at least one vector of coe_cients _ such that_0

yt _ I (d 1048576 b) (13)As most of the _nancial time series are integrated of order one we will restrictourselves to case d=b=1 Two or more variables are said to be cointegrated ifthere exist a linear combination of these variables that is stationary Many ofthe series are non-stationary but move together over time which implies twoseries are bound by some common force or factor in long run We will test forcointegration by a residual-based approach and Johansens VAR methodResidual Based approach Consider a modelyt = _1 + _2x2t + _3x3t + + ut (14)where yt x2t x3t are all integrated of order N Now if the residual of this re-gression ut is stationary then we can say that the variables are cointegrated elsethere exist no long term relationship between the variables To test the resid-ual for stationarity we will run Augmented Dickey-Fuller tests on the residualsUnder the Null hypothesis the residual are integrated of order one or more andunder alternate hypothesis the residuals are I(0)431 Johansen test for CointegrationJohansen test for cointegration presents a better model for testing multiplecointegration among multiple variables The Residual based approach can only_nd atmost one cointegration and can be tested for a model with two variablesEven if more than two variables are present in the equation that are cointegratedthe Residual based approach will give only one cointegration SO we will useJhoansen VAR based cointegration for testing more than one cointegrationSuppose that a set of g variables are under consideration that are I(1) andwhich are thought to be cointegrated A VAR with k lags containing thesevariables could be set upyt = _1yt10485761 + _2yt10485762 + _ _ _ + _kyt1048576k + ut (15)g _ 1 g _ g g _ 1 g _ g g _ 1 g _ g g _ 1 g _ 1EDHEC Business School 184 METHODOLOGYIn order to use the Johansen test the VAR above should be turned into avector error correction model of form_yt = _yt1048576k + 1_yt10485761 + 2_yt10485762 + _ _ _ + k10485761_yt1048576(k10485761) + ut (16)where _ = (_ki=1_i) 1048576 Ig and i = (_ij=1_j) 1048576 IgThe Johansens test centers around testing the _ matrix which is the matrixthat represents the long term cointegration between the variables The test fornumber of cointegration is calculated by looking at the rank of the _ matrixthrough its eigenvalues The rank of the matrix is equal to number of roots(eigenvalues) _i of the matrix that are di_erent from zero The roots should beless than 1 in absolute value and positive If the variables are not cointegratedthe rank of the matrix will not be signi_cantly di_erent from zero ie _i _ 0There are two test statistics for Johansen test _tracer and _max_trace (r) = 1048576TPgi=r+1 ln(1 1048576 _ _i)and_max(r r + 1) = 1048576Tln(1 1048576 _r_+1)_trace is a test statistic for joint test where the null hypothesis is that thenumber of cointegration vector is less than or equal to r against an alternativethat there are more than r_max conducts another separate test on eigenvalues and has null hypothesis that

the number of cointegrating vector is r against r+144 Impulse ResponseOnce we have determined whether the variables have long term relationship ornot we can form a multivariate VAR model for the variables A multivariateVAR model between g variables is a model where the current value of a variabledepend on di_ernt combinations of the previous k values of all the variables anderror terms A general representation of the model can beyBSEt = _ + _BSEyBSE + _IP yIP + CPIyCPI + _M1yM1 + _SP500ySP500 + u1t(17)where all the coe_cients except _ are g _ k matrices and all variables y are k_ 1 matricesOnce we have formed a model like this we can use the model for Impulse re-sponse A VAR(p) model can be written as a linear fuction of the past innova-tions that isrt = _ + at + 1at10485761 + 2at10485762 + (18)where _ = [_(1)]10485761_0 provided that the inverse exists and the coe_cient ma-trices i can be obtained by equating the coe_cients of Bi in the equation(I 1048576 _1B 1048576 1048576 _PBP )(I + 1B + 2B2 + ) = I (19)EDHEC Business School 194 METHODOLOGYwhere I is the Identity martix This is a moving average representation of rtwith the coe_cient matrix i being the impact of the past innovation at1048576i onrt Equivalently i is the e_ect of at on the future observation rt+i Therefore i is often referred to as the Impulse Response Function of rt For our impulseresponse we will use equation of variables in _rst di_ernce form like_BSEt = _t +Xki=0_11(i)_BSEt1048576i +Xkj=1_12(j)_MIt1048576j + _BSEt (20)_MIt = _t +Xki=0_21(i)_MIt1048576i +Xkj=1_22(j)_BSEt1048576j + _MIt (21)Grangers causality and Blocks F test of a VAR model will suggest which ofthe variables have statistically signi_cant impacts on the future values of othervariables in the system But F-test results cannot explain the sign of the re-lationship nor how long these e_ects require to take place Such informationwill however be given by an examination of the VARs impulse responses andvariance decompositions Impulse response is a technique that trace out theresponsiveness of the dependent variable in the VAR to shocks of each of theother variables So for each variable from each equation separately we will applya unit shock to the error and trace the e_ects upon the VAR system over timeBy using the impulse response technique we can determine how responsive isthe BSE stock index to Indian macro indicators and SP500 This will help usdetermine whether the BSE index is more reactive to domestic news or globalnewsEDHEC Business School 205 RESULTS5 ResultsBefore we use the time series for VAR analysis or cointegration tests we need todetermine whether the series are Stationary or not If the series are stationaryin levels we can use them directly else we need to use the di_erenced time seriesOne way to look for autocorrelation or integrated process is to see the graphsof the various time series used Section 71 shows the graphs of variables we

use for our analysis As we can see from the graphs all of the time series havea trend in long run which points to an integrated process As a second stepwe plot the graphs of di_erenced time series in Section 52 We can see thatthe di_erenced graphs in Section 72 dont show a long term trend and crossthe X-axis frequently This is usually a property of I(1) processes So we checkthe series for autocorrelations at di_erent lag lengths Section 73 shows cor-relograms graph autocorrelation coe_cient partial autocorrelation coe_cientQ-Stat and p-value for various time series up to 36 lags As can be seen in thetables the Q-stat for all lags is zero and we can reject the joint null hypothesisthat all the autocorrelations up to 36 lags are zero Table 741 shows that ifwe conduct a Unit root test on levels of the series we _nd that all the 7 seriesare integrated as we cannot reject the t-stat for unit root at 1 level But ifwe conduct the same test on di_erenced values of the series we _nd that we canreject the null hypothesis of unit root at 1 signi_cance level for all the seriesexcept CPI This tells us that all the series are I(1) as there _rst di_erence seriesare I(0)As our series are I(1) we will work with index levels of time series to determineif there exist one or more cointegrating relationships between the series Tablesin subsection 743 are based on residual approach where we run a regression ofBSE and various macroeconomic indicators and test the residuals for unit rootusing Augmented Dickey-Fuller test As we assume the two series are cointe-grated we conduct the test with no trend and intercept If the two series arecointegrated then the errors should not have any trend or intercept We see thatwe can reject the null hypothesis of unit root at 1 signi_cance for CPIIP M1We can reject the null of unit root for PPI at 5 and for SP500 and USDINRwe cant reject the null hypothesis of unit root at even 5 level This pointsto the fact that BSE has a strong long term relationship with IP M1 moneysupply CPI at 1 level with IP M1 CPI PPI at 5 signi_cance level AlsoBSE has no long term relationship with SP500 and USD INR exchange rateTo test for multiple cointegrating relationship we now employ a Johansen VARbased cointegration test The results of the test are displayed in subsection744 The _rst panel of the test results displays the value of _trace and_maxof Johansen test with di_erent assumptions about intercept and trend We cansee from this panel that when we consider a functional form of intercept and noTrend we have atleast and atmost three cointegrating relationships The secondpanel of the results display the value of information criteria for lag lengths Formost of the models we see that Akalike criteria points to a lag of three andSchwarz criteria points to a lag of one To estimate the cointegrating model wechoose the model with intercept and no trend and run a cointegration testTestresults are shown in Table 2 of subsection 744 At 5 signi_cance level wecan reject the null of atmost two cointegrating factors for _trace and same for_max Now to test which all variables have a long tern relationship we perform aRestricted cointegration with vector error correction model As we had alreadyEDHEC Business School 215 RESULTSseen in our residual based test of cointegration that BSE has no cointegratingrelationship with SP500 and USDINR we create a restricted cointegration modelwhere we set coe_cients of SP500 and USDINR as zero The test results aredisplayed in Table 3 of subsection 744 In this case as there are two restrictionsthe test statistic follow _2 with two degrees of freedom We can see that thep-value for the test is 1333 which tells us that the restrictions are supportedby data at 10 level of signi_cance So we can conclude that the BSE has along term relationship with CPIIPPPIM1 money supply but has no long termrelationship with SP500 and USDINR exchange rate One interpretation of thisresult can be that the Indian stock market represented here by BSE Sensexmoves more in accordance with domestic factors like Industrial production M1money supply Consumer price index and Producer Price index than with globalfactors or in other words as BSE is representation of largest market cap Indiancompanies we can say that the biggest companies in India are ones that are

more dependent on domestic demand rather than exports This result presentsan opportunity for international investors to diversify their portfolio by invest-ing in BSE Sensex as it is decoupled with global markets and macroeconomicfactorsWe use A bivariate Vector Autoregression (BVAR) technique to analyze thedynamic interaction between real asset prices and macro economy VAR ispreferred method to study Macroeconomy and asset prices where variables en-dogenously e_ect each otherWe begin with a bivariate VAR with no restriction Asset prices and instru-ments are allowed to respond to each other freely For paired variables withcointegration relationship VAR is performed at levels whilst for those that arenot cointegrated VAR is performed at _rst di_erence Constant term is ignoredwith loss of generality We use the Bivariate Autoregression analysis for bothimpulse response and Grangers causality testsImpulse response results are displayed in subsection 745 From _rst graph ofimpulse response of BSE to USDINR we can see that USDINR has a negativeimpact on BSE As impulse response is response of BSE to shocks given to US-DINR we can see that a positive shock or unexpected appreciation INR valuewrt USD will have a negative e_ect on BSE for few lags and will disappearafter few lags If we look at the constituents of BSE Index over time we seethat most of the time some of its constituent are companies that thrive on ex-ports Some of the biggest Market-Cap in India are companies in service sectorlike Infosys TCS etc that are hugely dependent on services provided to clientsfrom Europe and US So an appreciation of INR compared to USD makesthese _rms costlier for the global clients and in turn reduces the income of thesecompanies As the _rms revenue pro_t decreases the value of the stock alsodecreases that in turn a_ects the returns of BSE SensexSecond graph (betwen BSE and SP500) shows that increase in SP500 has a pos-itive e_ect on BSE as higher returns of SP500 indicate strong global economywhich in turn results in higher trade between countries The positive responseof BSE to one unit shock to SP500 indicates a spillover e_ect of global factorson Indian economy but the response is weak as can be seen from the graphMoving forward response of BSE to shocks in M1 money supply CPI PPImake economic sense As for M1 money supply one unit shock means increasein M1 money supply This increase in money supply allows companies to bor-row more money from banks at lower rates which they can use for investingEDHEC Business School 225 RESULTSin pro_table projects and generating larger cash ows For Ination indicatorsone unit shock means increase in ination This increase in ination results inhigher costs for the companies that in turn reduces their pro_t margins and asa result value of stocksBy looking at the graphs we can also see that shocks to Indian macroeconomicindicators creates stronger response by BSE as compared to global factors likeSP500 or USDINR This indicates that BSE Index is driven by companies thatdepend hugely on domestic demand rather than exports Response of BSE toshocks to Industrial Production are contradictory to theory In theory an in-crease in industrial production should result in positive response from BSE butour analysis shows the other way A possible reason for this response could bethat industrial production time series is seasonal as can be seen from the graphSo there is a possibility of a leadlag relationship between the two variablesTo test for possibility of leadlag relationship we run a Grangers causality testbetween BSE and IP The result in section 646 shows that at a lag lengthof 4 we can reject the Null hypothesis of BSE does not Granger cause IP at1 signi_cance level This proves that BSE is a leading indicator of industrialproduction and there exist a leadlag relationship between the two indicatorsEDHEC Business School 236 CONCLUSIONS6 Conclusions

In this paper I tested the relations between Indian stock market represented byBSE and domestic and global macro economic factors The research concludesthat the India stock markets are mainly driven by domestic demand and theinuence of global macro factors on the stock market is weak I also tested forGranger causality between BSE and IP and found that BSE is a leading indicatorof Industrial production and can help in predicting the industrial climate inIndiaThe research is insightful for investors and professionals who are looking forinvestment opportunities to diversify their risks As Indian stock markets aremore dependent on domestic factors one can invest in Indian indices and stocksto diversify their risks gained through investing in US and European stocksThe paper opens new doors for research in this _eld One can use variancedecomposition technique to see how much variance of BSE can be explained myvarious domestic and global macro factors Also one can use di_erent globalfactors like sovereign CDS spreads T-Bill rates a composite indicator of globaleconomy for further research on interaction between Indian stock market andglobal economyOne can also research on how various global macroeconomicnews a_ects India stock markets and for how long the e_ects persistsEDHEC Business School 247 GRAPHS AND TABLES7 Graphs and Tables71 Graphs of Time seriesEDHEC Business School 257 GRAPHS AND TABLESEDHEC Business School 267 GRAPHS AND TABLESEDHEC Business School 277 GRAPHS AND TABLESEDHEC Business School 287 GRAPHS AND TABLES72 Graphs of Time Series - Di_erencedEDHEC Business School 297 GRAPHS AND TABLESEDHEC Business School 307 GRAPHS AND TABLESEDHEC Business School 317 GRAPHS AND TABLESEDHEC Business School 327 GRAPHS AND TABLES73 Correlograms of Time seriesBSEEDHEC Business School 337 GRAPHS AND TABLESIPEDHEC Business School 347 GRAPHS AND TABLESSP500EDHEC Business School 357 GRAPHS AND TABLESUSDINREDHEC Business School 367 GRAPHS AND TABLESCPIEDHEC Business School 377 GRAPHS AND TABLESPPIEDHEC Business School 387 GRAPHS AND TABLES

M1EDHEC Business School 397 GRAPHS AND TABLES74 Tables741 Table for Unit root test of Time seriesVariables T-Stat p-valueBSE -2671 2495 SP500 -1315 8818 CPI -1909 6466 IP -1669 899 M1 -2420 3679 PPI -3353 601 USDINR -2955 1469 742 Tables for Unit root test of Di_erenced time seriesVariables T-Stat p-valueBSE -13848 000 SP500 -14832 000 CPI -3344 140 IP -3865 027 M1 -3867 026 PPI -9656 000 USDINR -13701 000 743 Tables for Residual based test of cointegrationTable 1BSE - CPIt-Statistic ProbADF test statistic -2622676 087Test critical values 1 level -25738185 level -19420410 level -1615891Table 2BSE - IPt-Statistic ProbADF test statistic -3738802 002Test critical values 1 level -25745135 level -194213610 level -1615828EDHEC Business School 407 GRAPHS AND TABLESTable 3BSE - M1t-Statistic ProbADF test statistic -2875518 041Test critical values 1 level -25737845 level -194203510 level -1615894Table 4BSE - PPIt-Statistic ProbADF test statistic -2399055 162Test critical values 1 level -25737845 level -194203510 level -1615894Table 5BSE - SP500t-Statistic ProbADF test statistic -1427184 1430Test critical values 1 level -25737845 level -1942035

10 level -1615894EDHEC Business School 417 GRAPHS AND TABLESTable 6BSE - USDINRt-Statistic ProbADF test statistic -1659522 917Test critical values 1 level -25738185 level -19420410 level -1615891EDHEC Business School 427 GRAPHS AND TABLES744 Johansen cointegration testEDHEC Business School 437 GRAPHS AND TABLESTable 2EDHEC Business School 447 GRAPHS AND TABLESTable 3EDHEC Business School 457 GRAPHS AND TABLES745 Impulse response testsEDHEC Business School 467 GRAPHS AND TABLESEDHEC Business School 477 GRAPHS AND TABLESEDHEC Business School 487 GRAPHS AND TABLES746 Granger causality test between IP and BSEEDHEC Business School 498 BIBLIOGRAPHY8 BibliographyEugene F Fama Ination Output and Money Journal of Business 1982Eugene F Fama Stock Returns Real activity and Money The American Eco-nomic Review 1981Eugene F Fama Stock Returns Expected Returns and Real activity Journal ofFinance 1990Pal and Mittal Impact of macroeconomic indicators in Indian capital marketsJournal of Risk Finance 2011Shahid Ahmed Aggregate Economic Variables and Stock Markets in India In-ternational Research Journal of Finance and Economics 2008Sahu and Dhiman Correlation and Causality between Stock Market and MacroEconomic Variables in India An Empirical Study 2010 International Confer-ence on E-Business and Economics 2011Mohammad Bayezid Ali Impact of Micro Variables on Emerging Stock MarketReturn A case on Dhaka Stock Exchange (DSE) Interdisciplinary Journal ofResearch in Business 2011Napphon Tangjitprom Macroeconomic Factors of Emerging Stock Market Theevidence from Thailand International Journal of Finance and Research 2012Sayed Mehdi Hosseini The Role of Macroeconomic Variables on Stock MarketIndex in China and India International Journal of Economics and Finance2011John Y Campbell Pitfalls and Opportunities What Macroeconomists shouldknow about Unit Roots NBER Working Papers 1991Hacker and Hatemi The properties of Procedures Dealing with Uncertainityabout Intercept and Deterministic Trend in Unit Root Testing CESIS Elec-tronic Working Papers 2010Elder and Kennedy Testing for Unit Roots What should Students be TaughtNasseh and Strauss Stock Prices and domestic and international macroeco-

nomic activity a cointegration approach The Quarterly Review of Economicsand Finance 2000Engle and Granger Co-Integration and Error Correction Representation Es-timation and Testing Econometrica 1987Eugene F Fama Stock Returns Real Activity Ination and Money 1981American Economic AssociationNaliniprave Tripathy Causal Relationship between Macro-Economic Indicatorsand Stock Market in India Asian Journal of Finance and Accounting 2011Rogalski and Vinso Stock Returns Money Supply and the Direction of Causal-ity The Journal of Finance 1977James et al A VARMA Analysis of the Causal Relations Among Stock Re-turns Real output and Nominal Interest Rates 1985 The Journal of FinanceBailey and Chung Risk and return in the Philippine Equity market A multi-factor exploration Paci_c-Basin Finance Journal 1996Nai-Fu Chen Financial Investment opportunities and the Macroeconomy TheJournal of Finance 1991GB Wickremasinghe Macroeconomic forces and stock prices Some empiricalevidence from an emerging stock markets University of Wollongong 2006EDHEC Business School 508 BIBLIOGRAPHYYao Juo and Loh On Chinas Monetary Policy and Asset Prices University ofNottingham- China policy Institute 2011Bilson et al Selecting macroeconomic variables as explanatory factors of emerg-ing stock market returns Paci_c-Basin Finance Journal 2001CHen Roll and Ross Economic forces and the Stock Markets The Journal ofBusiness 1986William H Greene Econometric Analysis 6th Edition Pearson InternationalEditionRuey Tsay Analysis of Financial Time seriesChris Brooks Introductory Econometrics for Finance Cambridge PublicationsEDHEC Business School 51

have a huge impact on the cash ows of not only one but many companies Thesystematic factor here refers to macro economic variables The state of Macroeconomic conditions lead to changes in Monetary and regulatory policies by thegovernment and which in turn a_ects the stock prices For example a countrywith good economic conditions represented by its Industrial production indexGDP CPI Interest rates will create an environment that is conducive for thegrowth of companies by lowering borrowing rates and other open market opera-tions So all macroeconomic factors that can inuence future cash ows or theEDHEC Business School 92 LITERATURE REVIEWdiscount rate by which the cash ows are discounted should have an inuenceon the stock priceMany researcher have studies the relationship between stock prices and macroeconomic variables and tried to explain the a_ect of one over the other Fama(1981) tries to establish a relationship between stock returns real activity ina-tion and money In his paper he _nds that Stock returns have positive relationwith real output and money supply but a negative relation with ination Heexplains that negative relation between stock returns and ination is induced bynegative relation between real output approximated by Industrial productionand ination This negative relationship between ination and real activityis explained by money demand theory and quantity theory of money Fama(1990) explains that measuring the total return variation explained by shocksto expected cash ows time-varying expected returns and shocks to expectedreturns is one way to judge the rationality of stock prices In his paper he_nds that growth rates of production used to proxy for shocks to expected cashows explain 45 of return variance ChenRoll and Ross (1986) explored therelationship between a set of economic variables and their systematic inuenceon stock market returns They found that Industrial production changes inrisk premium twists in yield curve had strong relationship and impact on stockreturns A somewhat weaker e_ect was found for measures of unanticipatedination and changes in expected ination during periods when these variableswere highly volatile They concluded that stock returns were exposed to sys-tematic economic news that they are priced in accordance to their exposuresand that the news can be measured as innovation in state variables Chen(1991) found that state variables that are priced are those that can forecastchanges in the investment and consumption opportunity set According to hisresearch default spread the term spread the one-month T-Bill rate the laggedindustrial production growth rate and the dividend-price ration are importantdeterminants of future stock market returns Bulmash and Trivoli (1991) showthe e_ect of business cycle movements on the relationship between stock returnsand money growthAn interesting paper in this _eld of research is by Fama (1990) and Schwert(1990) In the paper they claim that there are three explanations for the stronglink between stock prices and real economic activityFirst information about the future real activity may be reected in stockprices well before it occurs|this is essentially the notion that stock pricesare a leading indicator for the well-being of the economy Second changesin discount rates may a_ect stock prices and real investment similarly butthe output from real investment doesnt appear for some time after it ismade Third changes in stock prices are changes in wealth and thiscan a_ect the demand for consumption and investment goods [Schwert(1990)p1237]Campbell and Ammer (1993) use a VAR approach to model the simulta-neous interactions between the stock and bond markets since most previousworks do not address the channels through which the macroeconomic activityinuences the stock prices One example could be that industrial productioncould be linked to changing expectations of future cash ows (Balvers at al1990) On the other hand interest rate innovations could be the driving factorEDHEC Business School 102 LITERATURE REVIEW

in determining both industrial production (due to change in investment) andstock prices (due to change in the discounted present value of future cash ows)A VAR analysis can distinguish these possibilities Mukherjee and Naka (1995)show a long-term relationship between the Japanese stock price and real macroe-conomic variables Dr Nishat (2004) studies the long term association amongmacroeconomic variables like money supply CPIIPI and foreign exchange rateand stock markets in Pakistan The results show that there are causal relation-ship among the stock price and macroeconomic variables He uses data from1974 to 2004 in his study As most of the _nancial time series are non station-ary in levels he uses unit root technique to make data stationary Fazal Hussianand Tariq Massod (2001) used variables like investment GDP and consumptionemploying Grangers causality test to _nd relationship between macro factorsand stock markets They show that at two lags all macroeconomic variableshave highly signi_cant e_ect on stock prices James et al (1985) use a VARMAanalysis for investigating relationship between macro economy and stock mar-ket Using VARMA analysis for _nding causal relationship between factors isa better technique as the procedure does not preclude any causal structure apriori since it allows feedback among variables Thus the VARMA approachallow whatever causal relationship exist to emerge from the data They _ndlinkages between real activity and stock returns and real activity and inationAlso they _nd that stock returns signal changes in the monetary base Sincestock returns also signal changes in expected real activity this suggests a linkbetween the money supply and expected real activity that is consistent with themoney supply explanation o_ered by Geske and RollIn recent years the focus of these kind of studies have shifted from developedeconomies to developing economies As developing economies are the economiesthat see a lot of structural and monetary policy changes an analysis of relation-ship between macro and micro can provide new insights Also one can analysethe e_ects of monetary policies on the asset prices especially on stock pricesTangjitprom (2012) study of macroeconomic factors like unemployment rateinterest rate ination rate and exchange rate and stock market of Thailand con-cludes that macroeconomic factors signi_cantly explain stock returns He also_nds that for Thailand unemployment rate and ination rate are insigni_cant todetermine the stock returns The reason he provides is that the unemploymentrate and ination rate are not timely and there could be some lags before thedata becomes available Also Grangers test to examine lead-lag relationshipamong the factors reveal that only few macroeconomic variables could predictthe future stock returns whereas the stock returns can predict most of futuremacro economic variables This implies that performance of stock markets canbe a leading indicator for future macroeconomic conditions Ali (2011) study ofimpact of macro and micro factors on stock returns reveals that ination andforeign remittance have negative inuence and industrial production index havepositive impact on stock markets Also he didnt found any Grangers Causal-ity between stock markets and any of the explanatory variables This lack ofGrangers causality reveals the evidence of informationally ine_cient marketsAli uses a multivariate regression analysis on standard OLD formula for estimat-ing the relationship Hosseini et al (2011) tested the relationship between stockmarkets and four macro economic variables namely crude oil prices Money sup-ply Industrial production and ination rate in China and India They used aperiod of 1999 to 2009 for analysis As most of the economic time series have unitEDHEC Business School 112 LITERATURE REVIEWroot they _rst used the Augmented Dickey Fuller unit root test and found theunderlying series to be non-stationary at levels but stationary after in di_erenceAlso the use of Jhonson-Juselius (1990) Multivariate cointegration and VectorError Correction model technique indicate that there are both long and shortrun linkages between macroeconomic variable and stock market index in each ofthe two countries Their analysis shows that in long run the impact of increasein prices of crude oil for China is positive but for India is negative In terms

of money supply the impact on Indian stock market is negative but for Chinathere is a positive impact The e_ect of Industrial production is negative onlyin China In addition the e_ect of increases in ination on these stock marketsis positive in both countries Wickremasinghe (2006) analysed the relationshipbetween stock prices and macroeconomic variables in Sri Lanka He used theUnit root tests Jhonsons test Error-correction model variance decomposi-tion and impulse response to analyse the relationships His _ndings indicatethat there is both long term and short term causal relationship between stockprices and macroeconomic variables in Sri Lanka The result indicate that thestock prices can be predicted from certain macroeconomic variables and henceviolate the validity of the semi-strong version of e_cient market hypothesisAhmed (2008) investigates the causal relationship between Indian macroeco-nomic factors like Industrial Production Exports Foreign direct investmentMoney supply exchange rate interest rate and stock market indices NSE NiftyIndex and BSE Sensex For _nding the long term relationship he applies Jo-hansens cointegration and Toda and Yamamoto Granger Causality tests Foranalysing the Impulse response and variance decomposition he uses bivariateVAR His _ndings reveal that stock prices in India lead macroeconomic activityexcept movement in interest rate Interest rate seem to lead the stock priceThe study also reveals that movement of stock prices is not only the outcomeof behaviour of key macro economic variables but it is also one of the causesof movement in other macro dimensions in the economy An important paperby Bilson et al (2001) argues that emerging markets local factors are moreimportant than global factors They _nd that for emerging markets are at leastpartially segmented from global capital markets The global factors are proxiedby world market returns and local factors by set of macro economic variableslike money supply prices real activity and exchange rate Some evidence isfound that local factors are signi_cant in their association with emerging equitymarket returns above than that explained by the world factor When they usea larger set of variables the explanatory power of the model improves substan-tially such that they are able to explain a large amount of return variation formost emerging marketsEDHEC Business School 123 DATA3 Data31 Description of Macroeconomic IndicatorsOne of the biggest problems when conducting a research with macroeconomicdata is the frequency of the data Most of the macroeconomic indicator timeseries are yearlyquarterly or monthly time series This low frequency of themacroeconomic indicators results in very few data points for conducting a anal-ysis that is robust A possible cure for the problem is to use longer time periodsto incorporate more data points for macroeconomic variables But anotherproblem that we face when we look at the macroeconomic indicators for Asiancountries is reporting of the data For most of the Asian countries the macroe-conomic data doesnt have a long history and same can be said about historyof Indian macroeconomic variables So in this research we have used a timeperiod for which we can _nd data for most of the macroeconomic indicators Inthis paper we use a time period of 20 years starting from 1990 to 2011 Thistime period in Indian economy is representative of many structural and mone-tary policy changes like liberalization of India markets Also as the time periodis long it gives us enough data point for each macroeconomic factors to do arobust empirical analysisWhen one starts to build a model of interaction between macro and micro eco-nomic factors one dominant and important question one faces is among themyriad of macro indicators available for an economy which factors to chooseto incorporate in the model If one chooses macroeconomic factors that arehighly correlated among themselves then the power of test results decrease asit may result in a model where the macro indicators are able to explain mostof the movement of micro factors but the macro factors may not be relevant

To circumvent this problem we use variables that have been tested in earlierresearches and that have been proven to have e_ect on stock markets I alsotest a few macro factors that have some _nancial theory behind them that con-nect them to stock markets Ali (2011) Wickremasinghe (2006) Bilson etal(2001) and Bailey (1996) _nd that Industrial production CPI exchange rateM1 money supply GDP are few of the macro economic factors that can signi_-cantly explain stock returns Sahu(2011) Ahmed(2008) Tripathy(2011) studyon Indian markets speci_cally show that Industrial Production Exchange rateInation index are macro economic indicators that have a strong positive ornegative relationship with the stock markets So in our study we test 5 macroeconomic variables namely M1 money supply Consumer and Producer price In-dex Industrial production Exchange rate The time period for these indicatorsis from 1990-2011 The data for Ination indices Industrial production andexchange rate has been pulled from Bloombergc and Datastreamc The datahas been processed for errors and missing values Data for M1 money supplyhas been pulled from RBI website For most of the indices like ination andIndustrial production index the base year has been changed to 1990 Also assome of the indices are in levels and some in actual _gures (M1 money supply)we convert all of the indicators to level form (starting at 100 in 1990)EDHEC Business School 133 DATA32 Description of Stock Market IndicesCompared to Macro Indicators stock market data is relatively easy to _nd andhas considerably long history Also the stock market data is a real time data soit has a very high frequency of seconds Here in our analysis we will make use ofBSE (Bombay Stock Exchange) as representation of Indian markets and SP500(Standard and Poors 500 Index) as representation of global factors BSE is amarket cap-weighted of 30 stocks It is the oldest Index in the Asian markets(established in 1875) and have had a long history We choose this index as it isthe Index that represent the most liquid and traded stocks of the Indian stockmarket Also the index is most traded index in India and a good representationof trade prices of the stocks Even in terms of an orderly growth much beforethe actual legislations were enacted BSE Limited had formulated a compre-hensive set of Rules and Regulations for the securities market It had also laiddown best practices which were adopted subsequently by 23 stock exchangeswhich were set up after India gained its independence Our choice of SP500 isbased on the fact that it has a long history and many researchers have usedthis index as a good proxy representation of global markets and economic con-ditions We will take the monthly returns of each of the indices from 1990-2011in accordance with data frequency of macro economic variables Also as theindices have di_erent levels at beginning of 1990 we rebase both the indices tobase year of 1990 starting at a level of 100EDHEC Business School 144 METHODOLOGY4 Methodology41 Construction of Time SeriesThe _rst step in constructing an econometric model is constructing time seriesall of which are in same units Most of the time series used in our analysis are indi_erent formats For example CPI PPI BSE Index SP500 are in levels M1money supply USDINR exchange rate is in absolute current format Industrialproduction is in absolute production levels So _rst we convert all of the giventime series to level The way we construct time series in levels is _rstly takingthe initial data point of each time series as 100 We then _nd the percentagechange from one period to the next one for each time series using a continuouscompounding assumption (taking a natural log of change in values) In math-ematical terms it can be stated as Assume the original Index value at time tto be It and at time t+1 to be It + 1 Then we can compute the new rebasedindex by formulaRIt+1 = RIt _ (1 + ln(It+1=It))

whereRIt= Rebased Index at time tRIt+1=Rebased Index at time t+1We can use these rebased indices in building and testing our econometric model42 Unit Root Test and StationarityUnit root test is to _nd whether the series is stationary or non-stationary Astrictly stationary process is one where for any t1 t2 tt 2Z any k 2Z andT=12Fyt1 yt2 yt3 ytT

(y1 yT ) = Fyt1+k yt2+k yt3+k ytT+k

(y1 yT )where F represents joint distribution function of the set of random variablesIt can also be stated that the probability measure of sequence of yt is same asyt+k for all k In other words a series is stationary if the distribution of its valueremain the same as time progresses Similar to the concept of strict stationaryis weakly stationary process A weakly stationary process is one which has aconstant mean variance and autocovariance structure Stationary is a necessarycondition for a time series to be tested in regression A non-stationary seriescan have several problems like1 The shocks given to the series would not die of gradually resulting inincrease of variance as time passes2 If the series is non stationary then it can lead to spurious regressions If twoseries are generated independent of each other then if one is regressed onother it will result in very low R2 values But if two series are trending overtime then a regression of one over the other will give high R2 even thoughthe series may be unrelated to each other So if normal regressions toolsEDHEC Business School 154 METHODOLOGYare used on non stationary data then it may result in good but valuelessresults3 If the variables employed in a regression model are not stationary thenit can be proved that the standard assumptions for asymptotic analysiswill not be valid In other words the usual t-ratios will not follow at-distribution and the F-statistic will not follow an F-distribution and soonStationarity is a desirable condition for any time series so that it can be usedin regressions and give meaningful result that have some value to test for sta-tionarity a quick and dirty way is looking at the autocorrelation and partialcorrelation function of the series If the series is stationary then the autocorre-lation function should die o_ gradually after few lags and the partial correlationfunction will me non zero for some lags and zero thereafter Also we can usethe Ljung-Box test for testing that all m of _k autocorrelation coe_cients arezero using Q-statistic given by formulaQ = T(T + 2)_mk=1_k2T 1048576 k_ _2where T = Sample size and m = Maximum lag lengthThe lag length selection can be based on di_erent Information Criteria likeAkaikes Information criteria (AIC) Schwarzs Bayesian information criteria(SBIC) Hannan-Quinn criterion (HQIC) Mathematically di_erent criteria arerepresented asAIC = ln(_2) + 2kTSBIC = ln(_2) + kT lnTHQIC = ln(_2) + 2kT ln(ln(T))

For a better test for stationarity we use augmented Dickey fuller Unit roottest on each time series separately Augmented Dickey Fuller test is test ofnull hypothesis that the time series contains a unit roots against a alternativehypothesis that the series is stationary421 Mathematical representation of Stationary series and unit roottestAssume a variable Y whose structure can be given by AR process with no driftequationyt = _1yt10485761 + _2yt10485762 + _3yt10485763 + + _nyt1048576n + ut (2)where ut is the residual at time t Using a Lag operator L we can write eq(1)asyt = _1L1yt + _2L2yt + _3L3yt + + _nLnyt + ut (3)EDHEC Business School 164 METHODOLOGYRearranging eqn (2) we getyt 1048576 _1L1yt 1048576 _2L2yt 1048576 _3L3yt + 1048576 _nLnyt = ut (4)yt(1 1048576 _1L1 1048576 _2L2 1048576 _3L3 + 1048576 _nLn) = ut (5)or_(L)yt = ut (6)The time series is stationary if we can write eqn(5) in formyt = _(L)10485761ut (7)with _(L)10485761 converging to zero It means the autocorrelation function woulddecline as lag length is increased If eqn (6) is expanded to a MA(1) processthe coe_cients of residuals should decrease such that the the residuals that thee_ect of residuals decrease with increase in lags SO if the process is stationarythe coe_cients of residuals will converge to zero and for non-stationary seriesthey will and converge to zero and will have long term e_ect The condition fortesting of unit root for an AR process is that the roots of eqn(6) or Charac-teristic equation should lie outside unit circle422 Augmented Dickey Fuller Unit Root TestConsider an AR(1) process of variable Yyt = _yt10485761 + ut (8)Subtracting yt10485761 from both sides of eqn(7) we get_y = (_ 1048576 1)yt10485761 + ut (9)Eqn(8) is the test equation for Dickey Fuller test For Dickey-Fuller Unit roottestNull Hypothesis The value of _ is equal to 1 or value of _10485761 is equal to 0 vsAlternate Hypothesis The value of _ is less than one or value of _ 1048576 1 is lessthan zero Augmented Dickey-Fuller test is similar to normal Dickey-Fuller testsexcept it takes the lag structure of more than one into account_y = yt10485761 +Xpi=1_i_yt1048576i + ut (10)If the series has one or more unit root it is said to be integrated of order nwhere n is the number of unit roots of the characteristic equation To makethese time series stationary they needs to be di_erenced Mathematically ifyt _ I (n) (11)then_(d) yt _ I (0) (12)To make our time-series stationary we will use the natural log returns of theseseries in the analysisEDHEC Business School 174 METHODOLOGY43 Testing Long Term RelationshipsEngle and Granger (1987) in their seminal paper described cointegration whichforms the basis for testing for long term relationship between variables Accord-ing to Engle and Granger two variables are cointegrated if they are integratedprocess in their natural form (of the same order) but a weighted combination

of the variables can be found such that the combined new variable is integratedof order less than the order of individual time series Mathematically assumeyt to be a k X 1 vector of variables then the components are cointegrated orintegrated of order (db) if1 All components of yt are I(d)2 There is at least one vector of coe_cients _ such that_0

yt _ I (d 1048576 b) (13)As most of the _nancial time series are integrated of order one we will restrictourselves to case d=b=1 Two or more variables are said to be cointegrated ifthere exist a linear combination of these variables that is stationary Many ofthe series are non-stationary but move together over time which implies twoseries are bound by some common force or factor in long run We will test forcointegration by a residual-based approach and Johansens VAR methodResidual Based approach Consider a modelyt = _1 + _2x2t + _3x3t + + ut (14)where yt x2t x3t are all integrated of order N Now if the residual of this re-gression ut is stationary then we can say that the variables are cointegrated elsethere exist no long term relationship between the variables To test the resid-ual for stationarity we will run Augmented Dickey-Fuller tests on the residualsUnder the Null hypothesis the residual are integrated of order one or more andunder alternate hypothesis the residuals are I(0)431 Johansen test for CointegrationJohansen test for cointegration presents a better model for testing multiplecointegration among multiple variables The Residual based approach can only_nd atmost one cointegration and can be tested for a model with two variablesEven if more than two variables are present in the equation that are cointegratedthe Residual based approach will give only one cointegration SO we will useJhoansen VAR based cointegration for testing more than one cointegrationSuppose that a set of g variables are under consideration that are I(1) andwhich are thought to be cointegrated A VAR with k lags containing thesevariables could be set upyt = _1yt10485761 + _2yt10485762 + _ _ _ + _kyt1048576k + ut (15)g _ 1 g _ g g _ 1 g _ g g _ 1 g _ g g _ 1 g _ 1EDHEC Business School 184 METHODOLOGYIn order to use the Johansen test the VAR above should be turned into avector error correction model of form_yt = _yt1048576k + 1_yt10485761 + 2_yt10485762 + _ _ _ + k10485761_yt1048576(k10485761) + ut (16)where _ = (_ki=1_i) 1048576 Ig and i = (_ij=1_j) 1048576 IgThe Johansens test centers around testing the _ matrix which is the matrixthat represents the long term cointegration between the variables The test fornumber of cointegration is calculated by looking at the rank of the _ matrixthrough its eigenvalues The rank of the matrix is equal to number of roots(eigenvalues) _i of the matrix that are di_erent from zero The roots should beless than 1 in absolute value and positive If the variables are not cointegratedthe rank of the matrix will not be signi_cantly di_erent from zero ie _i _ 0There are two test statistics for Johansen test _tracer and _max_trace (r) = 1048576TPgi=r+1 ln(1 1048576 _ _i)and_max(r r + 1) = 1048576Tln(1 1048576 _r_+1)_trace is a test statistic for joint test where the null hypothesis is that thenumber of cointegration vector is less than or equal to r against an alternativethat there are more than r_max conducts another separate test on eigenvalues and has null hypothesis that

the number of cointegrating vector is r against r+144 Impulse ResponseOnce we have determined whether the variables have long term relationship ornot we can form a multivariate VAR model for the variables A multivariateVAR model between g variables is a model where the current value of a variabledepend on di_ernt combinations of the previous k values of all the variables anderror terms A general representation of the model can beyBSEt = _ + _BSEyBSE + _IP yIP + CPIyCPI + _M1yM1 + _SP500ySP500 + u1t(17)where all the coe_cients except _ are g _ k matrices and all variables y are k_ 1 matricesOnce we have formed a model like this we can use the model for Impulse re-sponse A VAR(p) model can be written as a linear fuction of the past innova-tions that isrt = _ + at + 1at10485761 + 2at10485762 + (18)where _ = [_(1)]10485761_0 provided that the inverse exists and the coe_cient ma-trices i can be obtained by equating the coe_cients of Bi in the equation(I 1048576 _1B 1048576 1048576 _PBP )(I + 1B + 2B2 + ) = I (19)EDHEC Business School 194 METHODOLOGYwhere I is the Identity martix This is a moving average representation of rtwith the coe_cient matrix i being the impact of the past innovation at1048576i onrt Equivalently i is the e_ect of at on the future observation rt+i Therefore i is often referred to as the Impulse Response Function of rt For our impulseresponse we will use equation of variables in _rst di_ernce form like_BSEt = _t +Xki=0_11(i)_BSEt1048576i +Xkj=1_12(j)_MIt1048576j + _BSEt (20)_MIt = _t +Xki=0_21(i)_MIt1048576i +Xkj=1_22(j)_BSEt1048576j + _MIt (21)Grangers causality and Blocks F test of a VAR model will suggest which ofthe variables have statistically signi_cant impacts on the future values of othervariables in the system But F-test results cannot explain the sign of the re-lationship nor how long these e_ects require to take place Such informationwill however be given by an examination of the VARs impulse responses andvariance decompositions Impulse response is a technique that trace out theresponsiveness of the dependent variable in the VAR to shocks of each of theother variables So for each variable from each equation separately we will applya unit shock to the error and trace the e_ects upon the VAR system over timeBy using the impulse response technique we can determine how responsive isthe BSE stock index to Indian macro indicators and SP500 This will help usdetermine whether the BSE index is more reactive to domestic news or globalnewsEDHEC Business School 205 RESULTS5 ResultsBefore we use the time series for VAR analysis or cointegration tests we need todetermine whether the series are Stationary or not If the series are stationaryin levels we can use them directly else we need to use the di_erenced time seriesOne way to look for autocorrelation or integrated process is to see the graphsof the various time series used Section 71 shows the graphs of variables we

use for our analysis As we can see from the graphs all of the time series havea trend in long run which points to an integrated process As a second stepwe plot the graphs of di_erenced time series in Section 52 We can see thatthe di_erenced graphs in Section 72 dont show a long term trend and crossthe X-axis frequently This is usually a property of I(1) processes So we checkthe series for autocorrelations at di_erent lag lengths Section 73 shows cor-relograms graph autocorrelation coe_cient partial autocorrelation coe_cientQ-Stat and p-value for various time series up to 36 lags As can be seen in thetables the Q-stat for all lags is zero and we can reject the joint null hypothesisthat all the autocorrelations up to 36 lags are zero Table 741 shows that ifwe conduct a Unit root test on levels of the series we _nd that all the 7 seriesare integrated as we cannot reject the t-stat for unit root at 1 level But ifwe conduct the same test on di_erenced values of the series we _nd that we canreject the null hypothesis of unit root at 1 signi_cance level for all the seriesexcept CPI This tells us that all the series are I(1) as there _rst di_erence seriesare I(0)As our series are I(1) we will work with index levels of time series to determineif there exist one or more cointegrating relationships between the series Tablesin subsection 743 are based on residual approach where we run a regression ofBSE and various macroeconomic indicators and test the residuals for unit rootusing Augmented Dickey-Fuller test As we assume the two series are cointe-grated we conduct the test with no trend and intercept If the two series arecointegrated then the errors should not have any trend or intercept We see thatwe can reject the null hypothesis of unit root at 1 signi_cance for CPIIP M1We can reject the null of unit root for PPI at 5 and for SP500 and USDINRwe cant reject the null hypothesis of unit root at even 5 level This pointsto the fact that BSE has a strong long term relationship with IP M1 moneysupply CPI at 1 level with IP M1 CPI PPI at 5 signi_cance level AlsoBSE has no long term relationship with SP500 and USD INR exchange rateTo test for multiple cointegrating relationship we now employ a Johansen VARbased cointegration test The results of the test are displayed in subsection744 The _rst panel of the test results displays the value of _trace and_maxof Johansen test with di_erent assumptions about intercept and trend We cansee from this panel that when we consider a functional form of intercept and noTrend we have atleast and atmost three cointegrating relationships The secondpanel of the results display the value of information criteria for lag lengths Formost of the models we see that Akalike criteria points to a lag of three andSchwarz criteria points to a lag of one To estimate the cointegrating model wechoose the model with intercept and no trend and run a cointegration testTestresults are shown in Table 2 of subsection 744 At 5 signi_cance level wecan reject the null of atmost two cointegrating factors for _trace and same for_max Now to test which all variables have a long tern relationship we perform aRestricted cointegration with vector error correction model As we had alreadyEDHEC Business School 215 RESULTSseen in our residual based test of cointegration that BSE has no cointegratingrelationship with SP500 and USDINR we create a restricted cointegration modelwhere we set coe_cients of SP500 and USDINR as zero The test results aredisplayed in Table 3 of subsection 744 In this case as there are two restrictionsthe test statistic follow _2 with two degrees of freedom We can see that thep-value for the test is 1333 which tells us that the restrictions are supportedby data at 10 level of signi_cance So we can conclude that the BSE has along term relationship with CPIIPPPIM1 money supply but has no long termrelationship with SP500 and USDINR exchange rate One interpretation of thisresult can be that the Indian stock market represented here by BSE Sensexmoves more in accordance with domestic factors like Industrial production M1money supply Consumer price index and Producer Price index than with globalfactors or in other words as BSE is representation of largest market cap Indiancompanies we can say that the biggest companies in India are ones that are

more dependent on domestic demand rather than exports This result presentsan opportunity for international investors to diversify their portfolio by invest-ing in BSE Sensex as it is decoupled with global markets and macroeconomicfactorsWe use A bivariate Vector Autoregression (BVAR) technique to analyze thedynamic interaction between real asset prices and macro economy VAR ispreferred method to study Macroeconomy and asset prices where variables en-dogenously e_ect each otherWe begin with a bivariate VAR with no restriction Asset prices and instru-ments are allowed to respond to each other freely For paired variables withcointegration relationship VAR is performed at levels whilst for those that arenot cointegrated VAR is performed at _rst di_erence Constant term is ignoredwith loss of generality We use the Bivariate Autoregression analysis for bothimpulse response and Grangers causality testsImpulse response results are displayed in subsection 745 From _rst graph ofimpulse response of BSE to USDINR we can see that USDINR has a negativeimpact on BSE As impulse response is response of BSE to shocks given to US-DINR we can see that a positive shock or unexpected appreciation INR valuewrt USD will have a negative e_ect on BSE for few lags and will disappearafter few lags If we look at the constituents of BSE Index over time we seethat most of the time some of its constituent are companies that thrive on ex-ports Some of the biggest Market-Cap in India are companies in service sectorlike Infosys TCS etc that are hugely dependent on services provided to clientsfrom Europe and US So an appreciation of INR compared to USD makesthese _rms costlier for the global clients and in turn reduces the income of thesecompanies As the _rms revenue pro_t decreases the value of the stock alsodecreases that in turn a_ects the returns of BSE SensexSecond graph (betwen BSE and SP500) shows that increase in SP500 has a pos-itive e_ect on BSE as higher returns of SP500 indicate strong global economywhich in turn results in higher trade between countries The positive responseof BSE to one unit shock to SP500 indicates a spillover e_ect of global factorson Indian economy but the response is weak as can be seen from the graphMoving forward response of BSE to shocks in M1 money supply CPI PPImake economic sense As for M1 money supply one unit shock means increasein M1 money supply This increase in money supply allows companies to bor-row more money from banks at lower rates which they can use for investingEDHEC Business School 225 RESULTSin pro_table projects and generating larger cash ows For Ination indicatorsone unit shock means increase in ination This increase in ination results inhigher costs for the companies that in turn reduces their pro_t margins and asa result value of stocksBy looking at the graphs we can also see that shocks to Indian macroeconomicindicators creates stronger response by BSE as compared to global factors likeSP500 or USDINR This indicates that BSE Index is driven by companies thatdepend hugely on domestic demand rather than exports Response of BSE toshocks to Industrial Production are contradictory to theory In theory an in-crease in industrial production should result in positive response from BSE butour analysis shows the other way A possible reason for this response could bethat industrial production time series is seasonal as can be seen from the graphSo there is a possibility of a leadlag relationship between the two variablesTo test for possibility of leadlag relationship we run a Grangers causality testbetween BSE and IP The result in section 646 shows that at a lag lengthof 4 we can reject the Null hypothesis of BSE does not Granger cause IP at1 signi_cance level This proves that BSE is a leading indicator of industrialproduction and there exist a leadlag relationship between the two indicatorsEDHEC Business School 236 CONCLUSIONS6 Conclusions

In this paper I tested the relations between Indian stock market represented byBSE and domestic and global macro economic factors The research concludesthat the India stock markets are mainly driven by domestic demand and theinuence of global macro factors on the stock market is weak I also tested forGranger causality between BSE and IP and found that BSE is a leading indicatorof Industrial production and can help in predicting the industrial climate inIndiaThe research is insightful for investors and professionals who are looking forinvestment opportunities to diversify their risks As Indian stock markets aremore dependent on domestic factors one can invest in Indian indices and stocksto diversify their risks gained through investing in US and European stocksThe paper opens new doors for research in this _eld One can use variancedecomposition technique to see how much variance of BSE can be explained myvarious domestic and global macro factors Also one can use di_erent globalfactors like sovereign CDS spreads T-Bill rates a composite indicator of globaleconomy for further research on interaction between Indian stock market andglobal economyOne can also research on how various global macroeconomicnews a_ects India stock markets and for how long the e_ects persistsEDHEC Business School 247 GRAPHS AND TABLES7 Graphs and Tables71 Graphs of Time seriesEDHEC Business School 257 GRAPHS AND TABLESEDHEC Business School 267 GRAPHS AND TABLESEDHEC Business School 277 GRAPHS AND TABLESEDHEC Business School 287 GRAPHS AND TABLES72 Graphs of Time Series - Di_erencedEDHEC Business School 297 GRAPHS AND TABLESEDHEC Business School 307 GRAPHS AND TABLESEDHEC Business School 317 GRAPHS AND TABLESEDHEC Business School 327 GRAPHS AND TABLES73 Correlograms of Time seriesBSEEDHEC Business School 337 GRAPHS AND TABLESIPEDHEC Business School 347 GRAPHS AND TABLESSP500EDHEC Business School 357 GRAPHS AND TABLESUSDINREDHEC Business School 367 GRAPHS AND TABLESCPIEDHEC Business School 377 GRAPHS AND TABLESPPIEDHEC Business School 387 GRAPHS AND TABLES

M1EDHEC Business School 397 GRAPHS AND TABLES74 Tables741 Table for Unit root test of Time seriesVariables T-Stat p-valueBSE -2671 2495 SP500 -1315 8818 CPI -1909 6466 IP -1669 899 M1 -2420 3679 PPI -3353 601 USDINR -2955 1469 742 Tables for Unit root test of Di_erenced time seriesVariables T-Stat p-valueBSE -13848 000 SP500 -14832 000 CPI -3344 140 IP -3865 027 M1 -3867 026 PPI -9656 000 USDINR -13701 000 743 Tables for Residual based test of cointegrationTable 1BSE - CPIt-Statistic ProbADF test statistic -2622676 087Test critical values 1 level -25738185 level -19420410 level -1615891Table 2BSE - IPt-Statistic ProbADF test statistic -3738802 002Test critical values 1 level -25745135 level -194213610 level -1615828EDHEC Business School 407 GRAPHS AND TABLESTable 3BSE - M1t-Statistic ProbADF test statistic -2875518 041Test critical values 1 level -25737845 level -194203510 level -1615894Table 4BSE - PPIt-Statistic ProbADF test statistic -2399055 162Test critical values 1 level -25737845 level -194203510 level -1615894Table 5BSE - SP500t-Statistic ProbADF test statistic -1427184 1430Test critical values 1 level -25737845 level -1942035

10 level -1615894EDHEC Business School 417 GRAPHS AND TABLESTable 6BSE - USDINRt-Statistic ProbADF test statistic -1659522 917Test critical values 1 level -25738185 level -19420410 level -1615891EDHEC Business School 427 GRAPHS AND TABLES744 Johansen cointegration testEDHEC Business School 437 GRAPHS AND TABLESTable 2EDHEC Business School 447 GRAPHS AND TABLESTable 3EDHEC Business School 457 GRAPHS AND TABLES745 Impulse response testsEDHEC Business School 467 GRAPHS AND TABLESEDHEC Business School 477 GRAPHS AND TABLESEDHEC Business School 487 GRAPHS AND TABLES746 Granger causality test between IP and BSEEDHEC Business School 498 BIBLIOGRAPHY8 BibliographyEugene F Fama Ination Output and Money Journal of Business 1982Eugene F Fama Stock Returns Real activity and Money The American Eco-nomic Review 1981Eugene F Fama Stock Returns Expected Returns and Real activity Journal ofFinance 1990Pal and Mittal Impact of macroeconomic indicators in Indian capital marketsJournal of Risk Finance 2011Shahid Ahmed Aggregate Economic Variables and Stock Markets in India In-ternational Research Journal of Finance and Economics 2008Sahu and Dhiman Correlation and Causality between Stock Market and MacroEconomic Variables in India An Empirical Study 2010 International Confer-ence on E-Business and Economics 2011Mohammad Bayezid Ali Impact of Micro Variables on Emerging Stock MarketReturn A case on Dhaka Stock Exchange (DSE) Interdisciplinary Journal ofResearch in Business 2011Napphon Tangjitprom Macroeconomic Factors of Emerging Stock Market Theevidence from Thailand International Journal of Finance and Research 2012Sayed Mehdi Hosseini The Role of Macroeconomic Variables on Stock MarketIndex in China and India International Journal of Economics and Finance2011John Y Campbell Pitfalls and Opportunities What Macroeconomists shouldknow about Unit Roots NBER Working Papers 1991Hacker and Hatemi The properties of Procedures Dealing with Uncertainityabout Intercept and Deterministic Trend in Unit Root Testing CESIS Elec-tronic Working Papers 2010Elder and Kennedy Testing for Unit Roots What should Students be TaughtNasseh and Strauss Stock Prices and domestic and international macroeco-

nomic activity a cointegration approach The Quarterly Review of Economicsand Finance 2000Engle and Granger Co-Integration and Error Correction Representation Es-timation and Testing Econometrica 1987Eugene F Fama Stock Returns Real Activity Ination and Money 1981American Economic AssociationNaliniprave Tripathy Causal Relationship between Macro-Economic Indicatorsand Stock Market in India Asian Journal of Finance and Accounting 2011Rogalski and Vinso Stock Returns Money Supply and the Direction of Causal-ity The Journal of Finance 1977James et al A VARMA Analysis of the Causal Relations Among Stock Re-turns Real output and Nominal Interest Rates 1985 The Journal of FinanceBailey and Chung Risk and return in the Philippine Equity market A multi-factor exploration Paci_c-Basin Finance Journal 1996Nai-Fu Chen Financial Investment opportunities and the Macroeconomy TheJournal of Finance 1991GB Wickremasinghe Macroeconomic forces and stock prices Some empiricalevidence from an emerging stock markets University of Wollongong 2006EDHEC Business School 508 BIBLIOGRAPHYYao Juo and Loh On Chinas Monetary Policy and Asset Prices University ofNottingham- China policy Institute 2011Bilson et al Selecting macroeconomic variables as explanatory factors of emerg-ing stock market returns Paci_c-Basin Finance Journal 2001CHen Roll and Ross Economic forces and the Stock Markets The Journal ofBusiness 1986William H Greene Econometric Analysis 6th Edition Pearson InternationalEditionRuey Tsay Analysis of Financial Time seriesChris Brooks Introductory Econometrics for Finance Cambridge PublicationsEDHEC Business School 51

in determining both industrial production (due to change in investment) andstock prices (due to change in the discounted present value of future cash ows)A VAR analysis can distinguish these possibilities Mukherjee and Naka (1995)show a long-term relationship between the Japanese stock price and real macroe-conomic variables Dr Nishat (2004) studies the long term association amongmacroeconomic variables like money supply CPIIPI and foreign exchange rateand stock markets in Pakistan The results show that there are causal relation-ship among the stock price and macroeconomic variables He uses data from1974 to 2004 in his study As most of the _nancial time series are non station-ary in levels he uses unit root technique to make data stationary Fazal Hussianand Tariq Massod (2001) used variables like investment GDP and consumptionemploying Grangers causality test to _nd relationship between macro factorsand stock markets They show that at two lags all macroeconomic variableshave highly signi_cant e_ect on stock prices James et al (1985) use a VARMAanalysis for investigating relationship between macro economy and stock mar-ket Using VARMA analysis for _nding causal relationship between factors isa better technique as the procedure does not preclude any causal structure apriori since it allows feedback among variables Thus the VARMA approachallow whatever causal relationship exist to emerge from the data They _ndlinkages between real activity and stock returns and real activity and inationAlso they _nd that stock returns signal changes in the monetary base Sincestock returns also signal changes in expected real activity this suggests a linkbetween the money supply and expected real activity that is consistent with themoney supply explanation o_ered by Geske and RollIn recent years the focus of these kind of studies have shifted from developedeconomies to developing economies As developing economies are the economiesthat see a lot of structural and monetary policy changes an analysis of relation-ship between macro and micro can provide new insights Also one can analysethe e_ects of monetary policies on the asset prices especially on stock pricesTangjitprom (2012) study of macroeconomic factors like unemployment rateinterest rate ination rate and exchange rate and stock market of Thailand con-cludes that macroeconomic factors signi_cantly explain stock returns He also_nds that for Thailand unemployment rate and ination rate are insigni_cant todetermine the stock returns The reason he provides is that the unemploymentrate and ination rate are not timely and there could be some lags before thedata becomes available Also Grangers test to examine lead-lag relationshipamong the factors reveal that only few macroeconomic variables could predictthe future stock returns whereas the stock returns can predict most of futuremacro economic variables This implies that performance of stock markets canbe a leading indicator for future macroeconomic conditions Ali (2011) study ofimpact of macro and micro factors on stock returns reveals that ination andforeign remittance have negative inuence and industrial production index havepositive impact on stock markets Also he didnt found any Grangers Causal-ity between stock markets and any of the explanatory variables This lack ofGrangers causality reveals the evidence of informationally ine_cient marketsAli uses a multivariate regression analysis on standard OLD formula for estimat-ing the relationship Hosseini et al (2011) tested the relationship between stockmarkets and four macro economic variables namely crude oil prices Money sup-ply Industrial production and ination rate in China and India They used aperiod of 1999 to 2009 for analysis As most of the economic time series have unitEDHEC Business School 112 LITERATURE REVIEWroot they _rst used the Augmented Dickey Fuller unit root test and found theunderlying series to be non-stationary at levels but stationary after in di_erenceAlso the use of Jhonson-Juselius (1990) Multivariate cointegration and VectorError Correction model technique indicate that there are both long and shortrun linkages between macroeconomic variable and stock market index in each ofthe two countries Their analysis shows that in long run the impact of increasein prices of crude oil for China is positive but for India is negative In terms

of money supply the impact on Indian stock market is negative but for Chinathere is a positive impact The e_ect of Industrial production is negative onlyin China In addition the e_ect of increases in ination on these stock marketsis positive in both countries Wickremasinghe (2006) analysed the relationshipbetween stock prices and macroeconomic variables in Sri Lanka He used theUnit root tests Jhonsons test Error-correction model variance decomposi-tion and impulse response to analyse the relationships His _ndings indicatethat there is both long term and short term causal relationship between stockprices and macroeconomic variables in Sri Lanka The result indicate that thestock prices can be predicted from certain macroeconomic variables and henceviolate the validity of the semi-strong version of e_cient market hypothesisAhmed (2008) investigates the causal relationship between Indian macroeco-nomic factors like Industrial Production Exports Foreign direct investmentMoney supply exchange rate interest rate and stock market indices NSE NiftyIndex and BSE Sensex For _nding the long term relationship he applies Jo-hansens cointegration and Toda and Yamamoto Granger Causality tests Foranalysing the Impulse response and variance decomposition he uses bivariateVAR His _ndings reveal that stock prices in India lead macroeconomic activityexcept movement in interest rate Interest rate seem to lead the stock priceThe study also reveals that movement of stock prices is not only the outcomeof behaviour of key macro economic variables but it is also one of the causesof movement in other macro dimensions in the economy An important paperby Bilson et al (2001) argues that emerging markets local factors are moreimportant than global factors They _nd that for emerging markets are at leastpartially segmented from global capital markets The global factors are proxiedby world market returns and local factors by set of macro economic variableslike money supply prices real activity and exchange rate Some evidence isfound that local factors are signi_cant in their association with emerging equitymarket returns above than that explained by the world factor When they usea larger set of variables the explanatory power of the model improves substan-tially such that they are able to explain a large amount of return variation formost emerging marketsEDHEC Business School 123 DATA3 Data31 Description of Macroeconomic IndicatorsOne of the biggest problems when conducting a research with macroeconomicdata is the frequency of the data Most of the macroeconomic indicator timeseries are yearlyquarterly or monthly time series This low frequency of themacroeconomic indicators results in very few data points for conducting a anal-ysis that is robust A possible cure for the problem is to use longer time periodsto incorporate more data points for macroeconomic variables But anotherproblem that we face when we look at the macroeconomic indicators for Asiancountries is reporting of the data For most of the Asian countries the macroe-conomic data doesnt have a long history and same can be said about historyof Indian macroeconomic variables So in this research we have used a timeperiod for which we can _nd data for most of the macroeconomic indicators Inthis paper we use a time period of 20 years starting from 1990 to 2011 Thistime period in Indian economy is representative of many structural and mone-tary policy changes like liberalization of India markets Also as the time periodis long it gives us enough data point for each macroeconomic factors to do arobust empirical analysisWhen one starts to build a model of interaction between macro and micro eco-nomic factors one dominant and important question one faces is among themyriad of macro indicators available for an economy which factors to chooseto incorporate in the model If one chooses macroeconomic factors that arehighly correlated among themselves then the power of test results decrease asit may result in a model where the macro indicators are able to explain mostof the movement of micro factors but the macro factors may not be relevant

To circumvent this problem we use variables that have been tested in earlierresearches and that have been proven to have e_ect on stock markets I alsotest a few macro factors that have some _nancial theory behind them that con-nect them to stock markets Ali (2011) Wickremasinghe (2006) Bilson etal(2001) and Bailey (1996) _nd that Industrial production CPI exchange rateM1 money supply GDP are few of the macro economic factors that can signi_-cantly explain stock returns Sahu(2011) Ahmed(2008) Tripathy(2011) studyon Indian markets speci_cally show that Industrial Production Exchange rateInation index are macro economic indicators that have a strong positive ornegative relationship with the stock markets So in our study we test 5 macroeconomic variables namely M1 money supply Consumer and Producer price In-dex Industrial production Exchange rate The time period for these indicatorsis from 1990-2011 The data for Ination indices Industrial production andexchange rate has been pulled from Bloombergc and Datastreamc The datahas been processed for errors and missing values Data for M1 money supplyhas been pulled from RBI website For most of the indices like ination andIndustrial production index the base year has been changed to 1990 Also assome of the indices are in levels and some in actual _gures (M1 money supply)we convert all of the indicators to level form (starting at 100 in 1990)EDHEC Business School 133 DATA32 Description of Stock Market IndicesCompared to Macro Indicators stock market data is relatively easy to _nd andhas considerably long history Also the stock market data is a real time data soit has a very high frequency of seconds Here in our analysis we will make use ofBSE (Bombay Stock Exchange) as representation of Indian markets and SP500(Standard and Poors 500 Index) as representation of global factors BSE is amarket cap-weighted of 30 stocks It is the oldest Index in the Asian markets(established in 1875) and have had a long history We choose this index as it isthe Index that represent the most liquid and traded stocks of the Indian stockmarket Also the index is most traded index in India and a good representationof trade prices of the stocks Even in terms of an orderly growth much beforethe actual legislations were enacted BSE Limited had formulated a compre-hensive set of Rules and Regulations for the securities market It had also laiddown best practices which were adopted subsequently by 23 stock exchangeswhich were set up after India gained its independence Our choice of SP500 isbased on the fact that it has a long history and many researchers have usedthis index as a good proxy representation of global markets and economic con-ditions We will take the monthly returns of each of the indices from 1990-2011in accordance with data frequency of macro economic variables Also as theindices have di_erent levels at beginning of 1990 we rebase both the indices tobase year of 1990 starting at a level of 100EDHEC Business School 144 METHODOLOGY4 Methodology41 Construction of Time SeriesThe _rst step in constructing an econometric model is constructing time seriesall of which are in same units Most of the time series used in our analysis are indi_erent formats For example CPI PPI BSE Index SP500 are in levels M1money supply USDINR exchange rate is in absolute current format Industrialproduction is in absolute production levels So _rst we convert all of the giventime series to level The way we construct time series in levels is _rstly takingthe initial data point of each time series as 100 We then _nd the percentagechange from one period to the next one for each time series using a continuouscompounding assumption (taking a natural log of change in values) In math-ematical terms it can be stated as Assume the original Index value at time tto be It and at time t+1 to be It + 1 Then we can compute the new rebasedindex by formulaRIt+1 = RIt _ (1 + ln(It+1=It))

whereRIt= Rebased Index at time tRIt+1=Rebased Index at time t+1We can use these rebased indices in building and testing our econometric model42 Unit Root Test and StationarityUnit root test is to _nd whether the series is stationary or non-stationary Astrictly stationary process is one where for any t1 t2 tt 2Z any k 2Z andT=12Fyt1 yt2 yt3 ytT

(y1 yT ) = Fyt1+k yt2+k yt3+k ytT+k

(y1 yT )where F represents joint distribution function of the set of random variablesIt can also be stated that the probability measure of sequence of yt is same asyt+k for all k In other words a series is stationary if the distribution of its valueremain the same as time progresses Similar to the concept of strict stationaryis weakly stationary process A weakly stationary process is one which has aconstant mean variance and autocovariance structure Stationary is a necessarycondition for a time series to be tested in regression A non-stationary seriescan have several problems like1 The shocks given to the series would not die of gradually resulting inincrease of variance as time passes2 If the series is non stationary then it can lead to spurious regressions If twoseries are generated independent of each other then if one is regressed onother it will result in very low R2 values But if two series are trending overtime then a regression of one over the other will give high R2 even thoughthe series may be unrelated to each other So if normal regressions toolsEDHEC Business School 154 METHODOLOGYare used on non stationary data then it may result in good but valuelessresults3 If the variables employed in a regression model are not stationary thenit can be proved that the standard assumptions for asymptotic analysiswill not be valid In other words the usual t-ratios will not follow at-distribution and the F-statistic will not follow an F-distribution and soonStationarity is a desirable condition for any time series so that it can be usedin regressions and give meaningful result that have some value to test for sta-tionarity a quick and dirty way is looking at the autocorrelation and partialcorrelation function of the series If the series is stationary then the autocorre-lation function should die o_ gradually after few lags and the partial correlationfunction will me non zero for some lags and zero thereafter Also we can usethe Ljung-Box test for testing that all m of _k autocorrelation coe_cients arezero using Q-statistic given by formulaQ = T(T + 2)_mk=1_k2T 1048576 k_ _2where T = Sample size and m = Maximum lag lengthThe lag length selection can be based on di_erent Information Criteria likeAkaikes Information criteria (AIC) Schwarzs Bayesian information criteria(SBIC) Hannan-Quinn criterion (HQIC) Mathematically di_erent criteria arerepresented asAIC = ln(_2) + 2kTSBIC = ln(_2) + kT lnTHQIC = ln(_2) + 2kT ln(ln(T))

For a better test for stationarity we use augmented Dickey fuller Unit roottest on each time series separately Augmented Dickey Fuller test is test ofnull hypothesis that the time series contains a unit roots against a alternativehypothesis that the series is stationary421 Mathematical representation of Stationary series and unit roottestAssume a variable Y whose structure can be given by AR process with no driftequationyt = _1yt10485761 + _2yt10485762 + _3yt10485763 + + _nyt1048576n + ut (2)where ut is the residual at time t Using a Lag operator L we can write eq(1)asyt = _1L1yt + _2L2yt + _3L3yt + + _nLnyt + ut (3)EDHEC Business School 164 METHODOLOGYRearranging eqn (2) we getyt 1048576 _1L1yt 1048576 _2L2yt 1048576 _3L3yt + 1048576 _nLnyt = ut (4)yt(1 1048576 _1L1 1048576 _2L2 1048576 _3L3 + 1048576 _nLn) = ut (5)or_(L)yt = ut (6)The time series is stationary if we can write eqn(5) in formyt = _(L)10485761ut (7)with _(L)10485761 converging to zero It means the autocorrelation function woulddecline as lag length is increased If eqn (6) is expanded to a MA(1) processthe coe_cients of residuals should decrease such that the the residuals that thee_ect of residuals decrease with increase in lags SO if the process is stationarythe coe_cients of residuals will converge to zero and for non-stationary seriesthey will and converge to zero and will have long term e_ect The condition fortesting of unit root for an AR process is that the roots of eqn(6) or Charac-teristic equation should lie outside unit circle422 Augmented Dickey Fuller Unit Root TestConsider an AR(1) process of variable Yyt = _yt10485761 + ut (8)Subtracting yt10485761 from both sides of eqn(7) we get_y = (_ 1048576 1)yt10485761 + ut (9)Eqn(8) is the test equation for Dickey Fuller test For Dickey-Fuller Unit roottestNull Hypothesis The value of _ is equal to 1 or value of _10485761 is equal to 0 vsAlternate Hypothesis The value of _ is less than one or value of _ 1048576 1 is lessthan zero Augmented Dickey-Fuller test is similar to normal Dickey-Fuller testsexcept it takes the lag structure of more than one into account_y = yt10485761 +Xpi=1_i_yt1048576i + ut (10)If the series has one or more unit root it is said to be integrated of order nwhere n is the number of unit roots of the characteristic equation To makethese time series stationary they needs to be di_erenced Mathematically ifyt _ I (n) (11)then_(d) yt _ I (0) (12)To make our time-series stationary we will use the natural log returns of theseseries in the analysisEDHEC Business School 174 METHODOLOGY43 Testing Long Term RelationshipsEngle and Granger (1987) in their seminal paper described cointegration whichforms the basis for testing for long term relationship between variables Accord-ing to Engle and Granger two variables are cointegrated if they are integratedprocess in their natural form (of the same order) but a weighted combination

of the variables can be found such that the combined new variable is integratedof order less than the order of individual time series Mathematically assumeyt to be a k X 1 vector of variables then the components are cointegrated orintegrated of order (db) if1 All components of yt are I(d)2 There is at least one vector of coe_cients _ such that_0

yt _ I (d 1048576 b) (13)As most of the _nancial time series are integrated of order one we will restrictourselves to case d=b=1 Two or more variables are said to be cointegrated ifthere exist a linear combination of these variables that is stationary Many ofthe series are non-stationary but move together over time which implies twoseries are bound by some common force or factor in long run We will test forcointegration by a residual-based approach and Johansens VAR methodResidual Based approach Consider a modelyt = _1 + _2x2t + _3x3t + + ut (14)where yt x2t x3t are all integrated of order N Now if the residual of this re-gression ut is stationary then we can say that the variables are cointegrated elsethere exist no long term relationship between the variables To test the resid-ual for stationarity we will run Augmented Dickey-Fuller tests on the residualsUnder the Null hypothesis the residual are integrated of order one or more andunder alternate hypothesis the residuals are I(0)431 Johansen test for CointegrationJohansen test for cointegration presents a better model for testing multiplecointegration among multiple variables The Residual based approach can only_nd atmost one cointegration and can be tested for a model with two variablesEven if more than two variables are present in the equation that are cointegratedthe Residual based approach will give only one cointegration SO we will useJhoansen VAR based cointegration for testing more than one cointegrationSuppose that a set of g variables are under consideration that are I(1) andwhich are thought to be cointegrated A VAR with k lags containing thesevariables could be set upyt = _1yt10485761 + _2yt10485762 + _ _ _ + _kyt1048576k + ut (15)g _ 1 g _ g g _ 1 g _ g g _ 1 g _ g g _ 1 g _ 1EDHEC Business School 184 METHODOLOGYIn order to use the Johansen test the VAR above should be turned into avector error correction model of form_yt = _yt1048576k + 1_yt10485761 + 2_yt10485762 + _ _ _ + k10485761_yt1048576(k10485761) + ut (16)where _ = (_ki=1_i) 1048576 Ig and i = (_ij=1_j) 1048576 IgThe Johansens test centers around testing the _ matrix which is the matrixthat represents the long term cointegration between the variables The test fornumber of cointegration is calculated by looking at the rank of the _ matrixthrough its eigenvalues The rank of the matrix is equal to number of roots(eigenvalues) _i of the matrix that are di_erent from zero The roots should beless than 1 in absolute value and positive If the variables are not cointegratedthe rank of the matrix will not be signi_cantly di_erent from zero ie _i _ 0There are two test statistics for Johansen test _tracer and _max_trace (r) = 1048576TPgi=r+1 ln(1 1048576 _ _i)and_max(r r + 1) = 1048576Tln(1 1048576 _r_+1)_trace is a test statistic for joint test where the null hypothesis is that thenumber of cointegration vector is less than or equal to r against an alternativethat there are more than r_max conducts another separate test on eigenvalues and has null hypothesis that

the number of cointegrating vector is r against r+144 Impulse ResponseOnce we have determined whether the variables have long term relationship ornot we can form a multivariate VAR model for the variables A multivariateVAR model between g variables is a model where the current value of a variabledepend on di_ernt combinations of the previous k values of all the variables anderror terms A general representation of the model can beyBSEt = _ + _BSEyBSE + _IP yIP + CPIyCPI + _M1yM1 + _SP500ySP500 + u1t(17)where all the coe_cients except _ are g _ k matrices and all variables y are k_ 1 matricesOnce we have formed a model like this we can use the model for Impulse re-sponse A VAR(p) model can be written as a linear fuction of the past innova-tions that isrt = _ + at + 1at10485761 + 2at10485762 + (18)where _ = [_(1)]10485761_0 provided that the inverse exists and the coe_cient ma-trices i can be obtained by equating the coe_cients of Bi in the equation(I 1048576 _1B 1048576 1048576 _PBP )(I + 1B + 2B2 + ) = I (19)EDHEC Business School 194 METHODOLOGYwhere I is the Identity martix This is a moving average representation of rtwith the coe_cient matrix i being the impact of the past innovation at1048576i onrt Equivalently i is the e_ect of at on the future observation rt+i Therefore i is often referred to as the Impulse Response Function of rt For our impulseresponse we will use equation of variables in _rst di_ernce form like_BSEt = _t +Xki=0_11(i)_BSEt1048576i +Xkj=1_12(j)_MIt1048576j + _BSEt (20)_MIt = _t +Xki=0_21(i)_MIt1048576i +Xkj=1_22(j)_BSEt1048576j + _MIt (21)Grangers causality and Blocks F test of a VAR model will suggest which ofthe variables have statistically signi_cant impacts on the future values of othervariables in the system But F-test results cannot explain the sign of the re-lationship nor how long these e_ects require to take place Such informationwill however be given by an examination of the VARs impulse responses andvariance decompositions Impulse response is a technique that trace out theresponsiveness of the dependent variable in the VAR to shocks of each of theother variables So for each variable from each equation separately we will applya unit shock to the error and trace the e_ects upon the VAR system over timeBy using the impulse response technique we can determine how responsive isthe BSE stock index to Indian macro indicators and SP500 This will help usdetermine whether the BSE index is more reactive to domestic news or globalnewsEDHEC Business School 205 RESULTS5 ResultsBefore we use the time series for VAR analysis or cointegration tests we need todetermine whether the series are Stationary or not If the series are stationaryin levels we can use them directly else we need to use the di_erenced time seriesOne way to look for autocorrelation or integrated process is to see the graphsof the various time series used Section 71 shows the graphs of variables we

use for our analysis As we can see from the graphs all of the time series havea trend in long run which points to an integrated process As a second stepwe plot the graphs of di_erenced time series in Section 52 We can see thatthe di_erenced graphs in Section 72 dont show a long term trend and crossthe X-axis frequently This is usually a property of I(1) processes So we checkthe series for autocorrelations at di_erent lag lengths Section 73 shows cor-relograms graph autocorrelation coe_cient partial autocorrelation coe_cientQ-Stat and p-value for various time series up to 36 lags As can be seen in thetables the Q-stat for all lags is zero and we can reject the joint null hypothesisthat all the autocorrelations up to 36 lags are zero Table 741 shows that ifwe conduct a Unit root test on levels of the series we _nd that all the 7 seriesare integrated as we cannot reject the t-stat for unit root at 1 level But ifwe conduct the same test on di_erenced values of the series we _nd that we canreject the null hypothesis of unit root at 1 signi_cance level for all the seriesexcept CPI This tells us that all the series are I(1) as there _rst di_erence seriesare I(0)As our series are I(1) we will work with index levels of time series to determineif there exist one or more cointegrating relationships between the series Tablesin subsection 743 are based on residual approach where we run a regression ofBSE and various macroeconomic indicators and test the residuals for unit rootusing Augmented Dickey-Fuller test As we assume the two series are cointe-grated we conduct the test with no trend and intercept If the two series arecointegrated then the errors should not have any trend or intercept We see thatwe can reject the null hypothesis of unit root at 1 signi_cance for CPIIP M1We can reject the null of unit root for PPI at 5 and for SP500 and USDINRwe cant reject the null hypothesis of unit root at even 5 level This pointsto the fact that BSE has a strong long term relationship with IP M1 moneysupply CPI at 1 level with IP M1 CPI PPI at 5 signi_cance level AlsoBSE has no long term relationship with SP500 and USD INR exchange rateTo test for multiple cointegrating relationship we now employ a Johansen VARbased cointegration test The results of the test are displayed in subsection744 The _rst panel of the test results displays the value of _trace and_maxof Johansen test with di_erent assumptions about intercept and trend We cansee from this panel that when we consider a functional form of intercept and noTrend we have atleast and atmost three cointegrating relationships The secondpanel of the results display the value of information criteria for lag lengths Formost of the models we see that Akalike criteria points to a lag of three andSchwarz criteria points to a lag of one To estimate the cointegrating model wechoose the model with intercept and no trend and run a cointegration testTestresults are shown in Table 2 of subsection 744 At 5 signi_cance level wecan reject the null of atmost two cointegrating factors for _trace and same for_max Now to test which all variables have a long tern relationship we perform aRestricted cointegration with vector error correction model As we had alreadyEDHEC Business School 215 RESULTSseen in our residual based test of cointegration that BSE has no cointegratingrelationship with SP500 and USDINR we create a restricted cointegration modelwhere we set coe_cients of SP500 and USDINR as zero The test results aredisplayed in Table 3 of subsection 744 In this case as there are two restrictionsthe test statistic follow _2 with two degrees of freedom We can see that thep-value for the test is 1333 which tells us that the restrictions are supportedby data at 10 level of signi_cance So we can conclude that the BSE has along term relationship with CPIIPPPIM1 money supply but has no long termrelationship with SP500 and USDINR exchange rate One interpretation of thisresult can be that the Indian stock market represented here by BSE Sensexmoves more in accordance with domestic factors like Industrial production M1money supply Consumer price index and Producer Price index than with globalfactors or in other words as BSE is representation of largest market cap Indiancompanies we can say that the biggest companies in India are ones that are

more dependent on domestic demand rather than exports This result presentsan opportunity for international investors to diversify their portfolio by invest-ing in BSE Sensex as it is decoupled with global markets and macroeconomicfactorsWe use A bivariate Vector Autoregression (BVAR) technique to analyze thedynamic interaction between real asset prices and macro economy VAR ispreferred method to study Macroeconomy and asset prices where variables en-dogenously e_ect each otherWe begin with a bivariate VAR with no restriction Asset prices and instru-ments are allowed to respond to each other freely For paired variables withcointegration relationship VAR is performed at levels whilst for those that arenot cointegrated VAR is performed at _rst di_erence Constant term is ignoredwith loss of generality We use the Bivariate Autoregression analysis for bothimpulse response and Grangers causality testsImpulse response results are displayed in subsection 745 From _rst graph ofimpulse response of BSE to USDINR we can see that USDINR has a negativeimpact on BSE As impulse response is response of BSE to shocks given to US-DINR we can see that a positive shock or unexpected appreciation INR valuewrt USD will have a negative e_ect on BSE for few lags and will disappearafter few lags If we look at the constituents of BSE Index over time we seethat most of the time some of its constituent are companies that thrive on ex-ports Some of the biggest Market-Cap in India are companies in service sectorlike Infosys TCS etc that are hugely dependent on services provided to clientsfrom Europe and US So an appreciation of INR compared to USD makesthese _rms costlier for the global clients and in turn reduces the income of thesecompanies As the _rms revenue pro_t decreases the value of the stock alsodecreases that in turn a_ects the returns of BSE SensexSecond graph (betwen BSE and SP500) shows that increase in SP500 has a pos-itive e_ect on BSE as higher returns of SP500 indicate strong global economywhich in turn results in higher trade between countries The positive responseof BSE to one unit shock to SP500 indicates a spillover e_ect of global factorson Indian economy but the response is weak as can be seen from the graphMoving forward response of BSE to shocks in M1 money supply CPI PPImake economic sense As for M1 money supply one unit shock means increasein M1 money supply This increase in money supply allows companies to bor-row more money from banks at lower rates which they can use for investingEDHEC Business School 225 RESULTSin pro_table projects and generating larger cash ows For Ination indicatorsone unit shock means increase in ination This increase in ination results inhigher costs for the companies that in turn reduces their pro_t margins and asa result value of stocksBy looking at the graphs we can also see that shocks to Indian macroeconomicindicators creates stronger response by BSE as compared to global factors likeSP500 or USDINR This indicates that BSE Index is driven by companies thatdepend hugely on domestic demand rather than exports Response of BSE toshocks to Industrial Production are contradictory to theory In theory an in-crease in industrial production should result in positive response from BSE butour analysis shows the other way A possible reason for this response could bethat industrial production time series is seasonal as can be seen from the graphSo there is a possibility of a leadlag relationship between the two variablesTo test for possibility of leadlag relationship we run a Grangers causality testbetween BSE and IP The result in section 646 shows that at a lag lengthof 4 we can reject the Null hypothesis of BSE does not Granger cause IP at1 signi_cance level This proves that BSE is a leading indicator of industrialproduction and there exist a leadlag relationship between the two indicatorsEDHEC Business School 236 CONCLUSIONS6 Conclusions

In this paper I tested the relations between Indian stock market represented byBSE and domestic and global macro economic factors The research concludesthat the India stock markets are mainly driven by domestic demand and theinuence of global macro factors on the stock market is weak I also tested forGranger causality between BSE and IP and found that BSE is a leading indicatorof Industrial production and can help in predicting the industrial climate inIndiaThe research is insightful for investors and professionals who are looking forinvestment opportunities to diversify their risks As Indian stock markets aremore dependent on domestic factors one can invest in Indian indices and stocksto diversify their risks gained through investing in US and European stocksThe paper opens new doors for research in this _eld One can use variancedecomposition technique to see how much variance of BSE can be explained myvarious domestic and global macro factors Also one can use di_erent globalfactors like sovereign CDS spreads T-Bill rates a composite indicator of globaleconomy for further research on interaction between Indian stock market andglobal economyOne can also research on how various global macroeconomicnews a_ects India stock markets and for how long the e_ects persistsEDHEC Business School 247 GRAPHS AND TABLES7 Graphs and Tables71 Graphs of Time seriesEDHEC Business School 257 GRAPHS AND TABLESEDHEC Business School 267 GRAPHS AND TABLESEDHEC Business School 277 GRAPHS AND TABLESEDHEC Business School 287 GRAPHS AND TABLES72 Graphs of Time Series - Di_erencedEDHEC Business School 297 GRAPHS AND TABLESEDHEC Business School 307 GRAPHS AND TABLESEDHEC Business School 317 GRAPHS AND TABLESEDHEC Business School 327 GRAPHS AND TABLES73 Correlograms of Time seriesBSEEDHEC Business School 337 GRAPHS AND TABLESIPEDHEC Business School 347 GRAPHS AND TABLESSP500EDHEC Business School 357 GRAPHS AND TABLESUSDINREDHEC Business School 367 GRAPHS AND TABLESCPIEDHEC Business School 377 GRAPHS AND TABLESPPIEDHEC Business School 387 GRAPHS AND TABLES

M1EDHEC Business School 397 GRAPHS AND TABLES74 Tables741 Table for Unit root test of Time seriesVariables T-Stat p-valueBSE -2671 2495 SP500 -1315 8818 CPI -1909 6466 IP -1669 899 M1 -2420 3679 PPI -3353 601 USDINR -2955 1469 742 Tables for Unit root test of Di_erenced time seriesVariables T-Stat p-valueBSE -13848 000 SP500 -14832 000 CPI -3344 140 IP -3865 027 M1 -3867 026 PPI -9656 000 USDINR -13701 000 743 Tables for Residual based test of cointegrationTable 1BSE - CPIt-Statistic ProbADF test statistic -2622676 087Test critical values 1 level -25738185 level -19420410 level -1615891Table 2BSE - IPt-Statistic ProbADF test statistic -3738802 002Test critical values 1 level -25745135 level -194213610 level -1615828EDHEC Business School 407 GRAPHS AND TABLESTable 3BSE - M1t-Statistic ProbADF test statistic -2875518 041Test critical values 1 level -25737845 level -194203510 level -1615894Table 4BSE - PPIt-Statistic ProbADF test statistic -2399055 162Test critical values 1 level -25737845 level -194203510 level -1615894Table 5BSE - SP500t-Statistic ProbADF test statistic -1427184 1430Test critical values 1 level -25737845 level -1942035

10 level -1615894EDHEC Business School 417 GRAPHS AND TABLESTable 6BSE - USDINRt-Statistic ProbADF test statistic -1659522 917Test critical values 1 level -25738185 level -19420410 level -1615891EDHEC Business School 427 GRAPHS AND TABLES744 Johansen cointegration testEDHEC Business School 437 GRAPHS AND TABLESTable 2EDHEC Business School 447 GRAPHS AND TABLESTable 3EDHEC Business School 457 GRAPHS AND TABLES745 Impulse response testsEDHEC Business School 467 GRAPHS AND TABLESEDHEC Business School 477 GRAPHS AND TABLESEDHEC Business School 487 GRAPHS AND TABLES746 Granger causality test between IP and BSEEDHEC Business School 498 BIBLIOGRAPHY8 BibliographyEugene F Fama Ination Output and Money Journal of Business 1982Eugene F Fama Stock Returns Real activity and Money The American Eco-nomic Review 1981Eugene F Fama Stock Returns Expected Returns and Real activity Journal ofFinance 1990Pal and Mittal Impact of macroeconomic indicators in Indian capital marketsJournal of Risk Finance 2011Shahid Ahmed Aggregate Economic Variables and Stock Markets in India In-ternational Research Journal of Finance and Economics 2008Sahu and Dhiman Correlation and Causality between Stock Market and MacroEconomic Variables in India An Empirical Study 2010 International Confer-ence on E-Business and Economics 2011Mohammad Bayezid Ali Impact of Micro Variables on Emerging Stock MarketReturn A case on Dhaka Stock Exchange (DSE) Interdisciplinary Journal ofResearch in Business 2011Napphon Tangjitprom Macroeconomic Factors of Emerging Stock Market Theevidence from Thailand International Journal of Finance and Research 2012Sayed Mehdi Hosseini The Role of Macroeconomic Variables on Stock MarketIndex in China and India International Journal of Economics and Finance2011John Y Campbell Pitfalls and Opportunities What Macroeconomists shouldknow about Unit Roots NBER Working Papers 1991Hacker and Hatemi The properties of Procedures Dealing with Uncertainityabout Intercept and Deterministic Trend in Unit Root Testing CESIS Elec-tronic Working Papers 2010Elder and Kennedy Testing for Unit Roots What should Students be TaughtNasseh and Strauss Stock Prices and domestic and international macroeco-

nomic activity a cointegration approach The Quarterly Review of Economicsand Finance 2000Engle and Granger Co-Integration and Error Correction Representation Es-timation and Testing Econometrica 1987Eugene F Fama Stock Returns Real Activity Ination and Money 1981American Economic AssociationNaliniprave Tripathy Causal Relationship between Macro-Economic Indicatorsand Stock Market in India Asian Journal of Finance and Accounting 2011Rogalski and Vinso Stock Returns Money Supply and the Direction of Causal-ity The Journal of Finance 1977James et al A VARMA Analysis of the Causal Relations Among Stock Re-turns Real output and Nominal Interest Rates 1985 The Journal of FinanceBailey and Chung Risk and return in the Philippine Equity market A multi-factor exploration Paci_c-Basin Finance Journal 1996Nai-Fu Chen Financial Investment opportunities and the Macroeconomy TheJournal of Finance 1991GB Wickremasinghe Macroeconomic forces and stock prices Some empiricalevidence from an emerging stock markets University of Wollongong 2006EDHEC Business School 508 BIBLIOGRAPHYYao Juo and Loh On Chinas Monetary Policy and Asset Prices University ofNottingham- China policy Institute 2011Bilson et al Selecting macroeconomic variables as explanatory factors of emerg-ing stock market returns Paci_c-Basin Finance Journal 2001CHen Roll and Ross Economic forces and the Stock Markets The Journal ofBusiness 1986William H Greene Econometric Analysis 6th Edition Pearson InternationalEditionRuey Tsay Analysis of Financial Time seriesChris Brooks Introductory Econometrics for Finance Cambridge PublicationsEDHEC Business School 51

of money supply the impact on Indian stock market is negative but for Chinathere is a positive impact The e_ect of Industrial production is negative onlyin China In addition the e_ect of increases in ination on these stock marketsis positive in both countries Wickremasinghe (2006) analysed the relationshipbetween stock prices and macroeconomic variables in Sri Lanka He used theUnit root tests Jhonsons test Error-correction model variance decomposi-tion and impulse response to analyse the relationships His _ndings indicatethat there is both long term and short term causal relationship between stockprices and macroeconomic variables in Sri Lanka The result indicate that thestock prices can be predicted from certain macroeconomic variables and henceviolate the validity of the semi-strong version of e_cient market hypothesisAhmed (2008) investigates the causal relationship between Indian macroeco-nomic factors like Industrial Production Exports Foreign direct investmentMoney supply exchange rate interest rate and stock market indices NSE NiftyIndex and BSE Sensex For _nding the long term relationship he applies Jo-hansens cointegration and Toda and Yamamoto Granger Causality tests Foranalysing the Impulse response and variance decomposition he uses bivariateVAR His _ndings reveal that stock prices in India lead macroeconomic activityexcept movement in interest rate Interest rate seem to lead the stock priceThe study also reveals that movement of stock prices is not only the outcomeof behaviour of key macro economic variables but it is also one of the causesof movement in other macro dimensions in the economy An important paperby Bilson et al (2001) argues that emerging markets local factors are moreimportant than global factors They _nd that for emerging markets are at leastpartially segmented from global capital markets The global factors are proxiedby world market returns and local factors by set of macro economic variableslike money supply prices real activity and exchange rate Some evidence isfound that local factors are signi_cant in their association with emerging equitymarket returns above than that explained by the world factor When they usea larger set of variables the explanatory power of the model improves substan-tially such that they are able to explain a large amount of return variation formost emerging marketsEDHEC Business School 123 DATA3 Data31 Description of Macroeconomic IndicatorsOne of the biggest problems when conducting a research with macroeconomicdata is the frequency of the data Most of the macroeconomic indicator timeseries are yearlyquarterly or monthly time series This low frequency of themacroeconomic indicators results in very few data points for conducting a anal-ysis that is robust A possible cure for the problem is to use longer time periodsto incorporate more data points for macroeconomic variables But anotherproblem that we face when we look at the macroeconomic indicators for Asiancountries is reporting of the data For most of the Asian countries the macroe-conomic data doesnt have a long history and same can be said about historyof Indian macroeconomic variables So in this research we have used a timeperiod for which we can _nd data for most of the macroeconomic indicators Inthis paper we use a time period of 20 years starting from 1990 to 2011 Thistime period in Indian economy is representative of many structural and mone-tary policy changes like liberalization of India markets Also as the time periodis long it gives us enough data point for each macroeconomic factors to do arobust empirical analysisWhen one starts to build a model of interaction between macro and micro eco-nomic factors one dominant and important question one faces is among themyriad of macro indicators available for an economy which factors to chooseto incorporate in the model If one chooses macroeconomic factors that arehighly correlated among themselves then the power of test results decrease asit may result in a model where the macro indicators are able to explain mostof the movement of micro factors but the macro factors may not be relevant

To circumvent this problem we use variables that have been tested in earlierresearches and that have been proven to have e_ect on stock markets I alsotest a few macro factors that have some _nancial theory behind them that con-nect them to stock markets Ali (2011) Wickremasinghe (2006) Bilson etal(2001) and Bailey (1996) _nd that Industrial production CPI exchange rateM1 money supply GDP are few of the macro economic factors that can signi_-cantly explain stock returns Sahu(2011) Ahmed(2008) Tripathy(2011) studyon Indian markets speci_cally show that Industrial Production Exchange rateInation index are macro economic indicators that have a strong positive ornegative relationship with the stock markets So in our study we test 5 macroeconomic variables namely M1 money supply Consumer and Producer price In-dex Industrial production Exchange rate The time period for these indicatorsis from 1990-2011 The data for Ination indices Industrial production andexchange rate has been pulled from Bloombergc and Datastreamc The datahas been processed for errors and missing values Data for M1 money supplyhas been pulled from RBI website For most of the indices like ination andIndustrial production index the base year has been changed to 1990 Also assome of the indices are in levels and some in actual _gures (M1 money supply)we convert all of the indicators to level form (starting at 100 in 1990)EDHEC Business School 133 DATA32 Description of Stock Market IndicesCompared to Macro Indicators stock market data is relatively easy to _nd andhas considerably long history Also the stock market data is a real time data soit has a very high frequency of seconds Here in our analysis we will make use ofBSE (Bombay Stock Exchange) as representation of Indian markets and SP500(Standard and Poors 500 Index) as representation of global factors BSE is amarket cap-weighted of 30 stocks It is the oldest Index in the Asian markets(established in 1875) and have had a long history We choose this index as it isthe Index that represent the most liquid and traded stocks of the Indian stockmarket Also the index is most traded index in India and a good representationof trade prices of the stocks Even in terms of an orderly growth much beforethe actual legislations were enacted BSE Limited had formulated a compre-hensive set of Rules and Regulations for the securities market It had also laiddown best practices which were adopted subsequently by 23 stock exchangeswhich were set up after India gained its independence Our choice of SP500 isbased on the fact that it has a long history and many researchers have usedthis index as a good proxy representation of global markets and economic con-ditions We will take the monthly returns of each of the indices from 1990-2011in accordance with data frequency of macro economic variables Also as theindices have di_erent levels at beginning of 1990 we rebase both the indices tobase year of 1990 starting at a level of 100EDHEC Business School 144 METHODOLOGY4 Methodology41 Construction of Time SeriesThe _rst step in constructing an econometric model is constructing time seriesall of which are in same units Most of the time series used in our analysis are indi_erent formats For example CPI PPI BSE Index SP500 are in levels M1money supply USDINR exchange rate is in absolute current format Industrialproduction is in absolute production levels So _rst we convert all of the giventime series to level The way we construct time series in levels is _rstly takingthe initial data point of each time series as 100 We then _nd the percentagechange from one period to the next one for each time series using a continuouscompounding assumption (taking a natural log of change in values) In math-ematical terms it can be stated as Assume the original Index value at time tto be It and at time t+1 to be It + 1 Then we can compute the new rebasedindex by formulaRIt+1 = RIt _ (1 + ln(It+1=It))

whereRIt= Rebased Index at time tRIt+1=Rebased Index at time t+1We can use these rebased indices in building and testing our econometric model42 Unit Root Test and StationarityUnit root test is to _nd whether the series is stationary or non-stationary Astrictly stationary process is one where for any t1 t2 tt 2Z any k 2Z andT=12Fyt1 yt2 yt3 ytT

(y1 yT ) = Fyt1+k yt2+k yt3+k ytT+k

(y1 yT )where F represents joint distribution function of the set of random variablesIt can also be stated that the probability measure of sequence of yt is same asyt+k for all k In other words a series is stationary if the distribution of its valueremain the same as time progresses Similar to the concept of strict stationaryis weakly stationary process A weakly stationary process is one which has aconstant mean variance and autocovariance structure Stationary is a necessarycondition for a time series to be tested in regression A non-stationary seriescan have several problems like1 The shocks given to the series would not die of gradually resulting inincrease of variance as time passes2 If the series is non stationary then it can lead to spurious regressions If twoseries are generated independent of each other then if one is regressed onother it will result in very low R2 values But if two series are trending overtime then a regression of one over the other will give high R2 even thoughthe series may be unrelated to each other So if normal regressions toolsEDHEC Business School 154 METHODOLOGYare used on non stationary data then it may result in good but valuelessresults3 If the variables employed in a regression model are not stationary thenit can be proved that the standard assumptions for asymptotic analysiswill not be valid In other words the usual t-ratios will not follow at-distribution and the F-statistic will not follow an F-distribution and soonStationarity is a desirable condition for any time series so that it can be usedin regressions and give meaningful result that have some value to test for sta-tionarity a quick and dirty way is looking at the autocorrelation and partialcorrelation function of the series If the series is stationary then the autocorre-lation function should die o_ gradually after few lags and the partial correlationfunction will me non zero for some lags and zero thereafter Also we can usethe Ljung-Box test for testing that all m of _k autocorrelation coe_cients arezero using Q-statistic given by formulaQ = T(T + 2)_mk=1_k2T 1048576 k_ _2where T = Sample size and m = Maximum lag lengthThe lag length selection can be based on di_erent Information Criteria likeAkaikes Information criteria (AIC) Schwarzs Bayesian information criteria(SBIC) Hannan-Quinn criterion (HQIC) Mathematically di_erent criteria arerepresented asAIC = ln(_2) + 2kTSBIC = ln(_2) + kT lnTHQIC = ln(_2) + 2kT ln(ln(T))

For a better test for stationarity we use augmented Dickey fuller Unit roottest on each time series separately Augmented Dickey Fuller test is test ofnull hypothesis that the time series contains a unit roots against a alternativehypothesis that the series is stationary421 Mathematical representation of Stationary series and unit roottestAssume a variable Y whose structure can be given by AR process with no driftequationyt = _1yt10485761 + _2yt10485762 + _3yt10485763 + + _nyt1048576n + ut (2)where ut is the residual at time t Using a Lag operator L we can write eq(1)asyt = _1L1yt + _2L2yt + _3L3yt + + _nLnyt + ut (3)EDHEC Business School 164 METHODOLOGYRearranging eqn (2) we getyt 1048576 _1L1yt 1048576 _2L2yt 1048576 _3L3yt + 1048576 _nLnyt = ut (4)yt(1 1048576 _1L1 1048576 _2L2 1048576 _3L3 + 1048576 _nLn) = ut (5)or_(L)yt = ut (6)The time series is stationary if we can write eqn(5) in formyt = _(L)10485761ut (7)with _(L)10485761 converging to zero It means the autocorrelation function woulddecline as lag length is increased If eqn (6) is expanded to a MA(1) processthe coe_cients of residuals should decrease such that the the residuals that thee_ect of residuals decrease with increase in lags SO if the process is stationarythe coe_cients of residuals will converge to zero and for non-stationary seriesthey will and converge to zero and will have long term e_ect The condition fortesting of unit root for an AR process is that the roots of eqn(6) or Charac-teristic equation should lie outside unit circle422 Augmented Dickey Fuller Unit Root TestConsider an AR(1) process of variable Yyt = _yt10485761 + ut (8)Subtracting yt10485761 from both sides of eqn(7) we get_y = (_ 1048576 1)yt10485761 + ut (9)Eqn(8) is the test equation for Dickey Fuller test For Dickey-Fuller Unit roottestNull Hypothesis The value of _ is equal to 1 or value of _10485761 is equal to 0 vsAlternate Hypothesis The value of _ is less than one or value of _ 1048576 1 is lessthan zero Augmented Dickey-Fuller test is similar to normal Dickey-Fuller testsexcept it takes the lag structure of more than one into account_y = yt10485761 +Xpi=1_i_yt1048576i + ut (10)If the series has one or more unit root it is said to be integrated of order nwhere n is the number of unit roots of the characteristic equation To makethese time series stationary they needs to be di_erenced Mathematically ifyt _ I (n) (11)then_(d) yt _ I (0) (12)To make our time-series stationary we will use the natural log returns of theseseries in the analysisEDHEC Business School 174 METHODOLOGY43 Testing Long Term RelationshipsEngle and Granger (1987) in their seminal paper described cointegration whichforms the basis for testing for long term relationship between variables Accord-ing to Engle and Granger two variables are cointegrated if they are integratedprocess in their natural form (of the same order) but a weighted combination

of the variables can be found such that the combined new variable is integratedof order less than the order of individual time series Mathematically assumeyt to be a k X 1 vector of variables then the components are cointegrated orintegrated of order (db) if1 All components of yt are I(d)2 There is at least one vector of coe_cients _ such that_0

yt _ I (d 1048576 b) (13)As most of the _nancial time series are integrated of order one we will restrictourselves to case d=b=1 Two or more variables are said to be cointegrated ifthere exist a linear combination of these variables that is stationary Many ofthe series are non-stationary but move together over time which implies twoseries are bound by some common force or factor in long run We will test forcointegration by a residual-based approach and Johansens VAR methodResidual Based approach Consider a modelyt = _1 + _2x2t + _3x3t + + ut (14)where yt x2t x3t are all integrated of order N Now if the residual of this re-gression ut is stationary then we can say that the variables are cointegrated elsethere exist no long term relationship between the variables To test the resid-ual for stationarity we will run Augmented Dickey-Fuller tests on the residualsUnder the Null hypothesis the residual are integrated of order one or more andunder alternate hypothesis the residuals are I(0)431 Johansen test for CointegrationJohansen test for cointegration presents a better model for testing multiplecointegration among multiple variables The Residual based approach can only_nd atmost one cointegration and can be tested for a model with two variablesEven if more than two variables are present in the equation that are cointegratedthe Residual based approach will give only one cointegration SO we will useJhoansen VAR based cointegration for testing more than one cointegrationSuppose that a set of g variables are under consideration that are I(1) andwhich are thought to be cointegrated A VAR with k lags containing thesevariables could be set upyt = _1yt10485761 + _2yt10485762 + _ _ _ + _kyt1048576k + ut (15)g _ 1 g _ g g _ 1 g _ g g _ 1 g _ g g _ 1 g _ 1EDHEC Business School 184 METHODOLOGYIn order to use the Johansen test the VAR above should be turned into avector error correction model of form_yt = _yt1048576k + 1_yt10485761 + 2_yt10485762 + _ _ _ + k10485761_yt1048576(k10485761) + ut (16)where _ = (_ki=1_i) 1048576 Ig and i = (_ij=1_j) 1048576 IgThe Johansens test centers around testing the _ matrix which is the matrixthat represents the long term cointegration between the variables The test fornumber of cointegration is calculated by looking at the rank of the _ matrixthrough its eigenvalues The rank of the matrix is equal to number of roots(eigenvalues) _i of the matrix that are di_erent from zero The roots should beless than 1 in absolute value and positive If the variables are not cointegratedthe rank of the matrix will not be signi_cantly di_erent from zero ie _i _ 0There are two test statistics for Johansen test _tracer and _max_trace (r) = 1048576TPgi=r+1 ln(1 1048576 _ _i)and_max(r r + 1) = 1048576Tln(1 1048576 _r_+1)_trace is a test statistic for joint test where the null hypothesis is that thenumber of cointegration vector is less than or equal to r against an alternativethat there are more than r_max conducts another separate test on eigenvalues and has null hypothesis that

the number of cointegrating vector is r against r+144 Impulse ResponseOnce we have determined whether the variables have long term relationship ornot we can form a multivariate VAR model for the variables A multivariateVAR model between g variables is a model where the current value of a variabledepend on di_ernt combinations of the previous k values of all the variables anderror terms A general representation of the model can beyBSEt = _ + _BSEyBSE + _IP yIP + CPIyCPI + _M1yM1 + _SP500ySP500 + u1t(17)where all the coe_cients except _ are g _ k matrices and all variables y are k_ 1 matricesOnce we have formed a model like this we can use the model for Impulse re-sponse A VAR(p) model can be written as a linear fuction of the past innova-tions that isrt = _ + at + 1at10485761 + 2at10485762 + (18)where _ = [_(1)]10485761_0 provided that the inverse exists and the coe_cient ma-trices i can be obtained by equating the coe_cients of Bi in the equation(I 1048576 _1B 1048576 1048576 _PBP )(I + 1B + 2B2 + ) = I (19)EDHEC Business School 194 METHODOLOGYwhere I is the Identity martix This is a moving average representation of rtwith the coe_cient matrix i being the impact of the past innovation at1048576i onrt Equivalently i is the e_ect of at on the future observation rt+i Therefore i is often referred to as the Impulse Response Function of rt For our impulseresponse we will use equation of variables in _rst di_ernce form like_BSEt = _t +Xki=0_11(i)_BSEt1048576i +Xkj=1_12(j)_MIt1048576j + _BSEt (20)_MIt = _t +Xki=0_21(i)_MIt1048576i +Xkj=1_22(j)_BSEt1048576j + _MIt (21)Grangers causality and Blocks F test of a VAR model will suggest which ofthe variables have statistically signi_cant impacts on the future values of othervariables in the system But F-test results cannot explain the sign of the re-lationship nor how long these e_ects require to take place Such informationwill however be given by an examination of the VARs impulse responses andvariance decompositions Impulse response is a technique that trace out theresponsiveness of the dependent variable in the VAR to shocks of each of theother variables So for each variable from each equation separately we will applya unit shock to the error and trace the e_ects upon the VAR system over timeBy using the impulse response technique we can determine how responsive isthe BSE stock index to Indian macro indicators and SP500 This will help usdetermine whether the BSE index is more reactive to domestic news or globalnewsEDHEC Business School 205 RESULTS5 ResultsBefore we use the time series for VAR analysis or cointegration tests we need todetermine whether the series are Stationary or not If the series are stationaryin levels we can use them directly else we need to use the di_erenced time seriesOne way to look for autocorrelation or integrated process is to see the graphsof the various time series used Section 71 shows the graphs of variables we

use for our analysis As we can see from the graphs all of the time series havea trend in long run which points to an integrated process As a second stepwe plot the graphs of di_erenced time series in Section 52 We can see thatthe di_erenced graphs in Section 72 dont show a long term trend and crossthe X-axis frequently This is usually a property of I(1) processes So we checkthe series for autocorrelations at di_erent lag lengths Section 73 shows cor-relograms graph autocorrelation coe_cient partial autocorrelation coe_cientQ-Stat and p-value for various time series up to 36 lags As can be seen in thetables the Q-stat for all lags is zero and we can reject the joint null hypothesisthat all the autocorrelations up to 36 lags are zero Table 741 shows that ifwe conduct a Unit root test on levels of the series we _nd that all the 7 seriesare integrated as we cannot reject the t-stat for unit root at 1 level But ifwe conduct the same test on di_erenced values of the series we _nd that we canreject the null hypothesis of unit root at 1 signi_cance level for all the seriesexcept CPI This tells us that all the series are I(1) as there _rst di_erence seriesare I(0)As our series are I(1) we will work with index levels of time series to determineif there exist one or more cointegrating relationships between the series Tablesin subsection 743 are based on residual approach where we run a regression ofBSE and various macroeconomic indicators and test the residuals for unit rootusing Augmented Dickey-Fuller test As we assume the two series are cointe-grated we conduct the test with no trend and intercept If the two series arecointegrated then the errors should not have any trend or intercept We see thatwe can reject the null hypothesis of unit root at 1 signi_cance for CPIIP M1We can reject the null of unit root for PPI at 5 and for SP500 and USDINRwe cant reject the null hypothesis of unit root at even 5 level This pointsto the fact that BSE has a strong long term relationship with IP M1 moneysupply CPI at 1 level with IP M1 CPI PPI at 5 signi_cance level AlsoBSE has no long term relationship with SP500 and USD INR exchange rateTo test for multiple cointegrating relationship we now employ a Johansen VARbased cointegration test The results of the test are displayed in subsection744 The _rst panel of the test results displays the value of _trace and_maxof Johansen test with di_erent assumptions about intercept and trend We cansee from this panel that when we consider a functional form of intercept and noTrend we have atleast and atmost three cointegrating relationships The secondpanel of the results display the value of information criteria for lag lengths Formost of the models we see that Akalike criteria points to a lag of three andSchwarz criteria points to a lag of one To estimate the cointegrating model wechoose the model with intercept and no trend and run a cointegration testTestresults are shown in Table 2 of subsection 744 At 5 signi_cance level wecan reject the null of atmost two cointegrating factors for _trace and same for_max Now to test which all variables have a long tern relationship we perform aRestricted cointegration with vector error correction model As we had alreadyEDHEC Business School 215 RESULTSseen in our residual based test of cointegration that BSE has no cointegratingrelationship with SP500 and USDINR we create a restricted cointegration modelwhere we set coe_cients of SP500 and USDINR as zero The test results aredisplayed in Table 3 of subsection 744 In this case as there are two restrictionsthe test statistic follow _2 with two degrees of freedom We can see that thep-value for the test is 1333 which tells us that the restrictions are supportedby data at 10 level of signi_cance So we can conclude that the BSE has along term relationship with CPIIPPPIM1 money supply but has no long termrelationship with SP500 and USDINR exchange rate One interpretation of thisresult can be that the Indian stock market represented here by BSE Sensexmoves more in accordance with domestic factors like Industrial production M1money supply Consumer price index and Producer Price index than with globalfactors or in other words as BSE is representation of largest market cap Indiancompanies we can say that the biggest companies in India are ones that are

more dependent on domestic demand rather than exports This result presentsan opportunity for international investors to diversify their portfolio by invest-ing in BSE Sensex as it is decoupled with global markets and macroeconomicfactorsWe use A bivariate Vector Autoregression (BVAR) technique to analyze thedynamic interaction between real asset prices and macro economy VAR ispreferred method to study Macroeconomy and asset prices where variables en-dogenously e_ect each otherWe begin with a bivariate VAR with no restriction Asset prices and instru-ments are allowed to respond to each other freely For paired variables withcointegration relationship VAR is performed at levels whilst for those that arenot cointegrated VAR is performed at _rst di_erence Constant term is ignoredwith loss of generality We use the Bivariate Autoregression analysis for bothimpulse response and Grangers causality testsImpulse response results are displayed in subsection 745 From _rst graph ofimpulse response of BSE to USDINR we can see that USDINR has a negativeimpact on BSE As impulse response is response of BSE to shocks given to US-DINR we can see that a positive shock or unexpected appreciation INR valuewrt USD will have a negative e_ect on BSE for few lags and will disappearafter few lags If we look at the constituents of BSE Index over time we seethat most of the time some of its constituent are companies that thrive on ex-ports Some of the biggest Market-Cap in India are companies in service sectorlike Infosys TCS etc that are hugely dependent on services provided to clientsfrom Europe and US So an appreciation of INR compared to USD makesthese _rms costlier for the global clients and in turn reduces the income of thesecompanies As the _rms revenue pro_t decreases the value of the stock alsodecreases that in turn a_ects the returns of BSE SensexSecond graph (betwen BSE and SP500) shows that increase in SP500 has a pos-itive e_ect on BSE as higher returns of SP500 indicate strong global economywhich in turn results in higher trade between countries The positive responseof BSE to one unit shock to SP500 indicates a spillover e_ect of global factorson Indian economy but the response is weak as can be seen from the graphMoving forward response of BSE to shocks in M1 money supply CPI PPImake economic sense As for M1 money supply one unit shock means increasein M1 money supply This increase in money supply allows companies to bor-row more money from banks at lower rates which they can use for investingEDHEC Business School 225 RESULTSin pro_table projects and generating larger cash ows For Ination indicatorsone unit shock means increase in ination This increase in ination results inhigher costs for the companies that in turn reduces their pro_t margins and asa result value of stocksBy looking at the graphs we can also see that shocks to Indian macroeconomicindicators creates stronger response by BSE as compared to global factors likeSP500 or USDINR This indicates that BSE Index is driven by companies thatdepend hugely on domestic demand rather than exports Response of BSE toshocks to Industrial Production are contradictory to theory In theory an in-crease in industrial production should result in positive response from BSE butour analysis shows the other way A possible reason for this response could bethat industrial production time series is seasonal as can be seen from the graphSo there is a possibility of a leadlag relationship between the two variablesTo test for possibility of leadlag relationship we run a Grangers causality testbetween BSE and IP The result in section 646 shows that at a lag lengthof 4 we can reject the Null hypothesis of BSE does not Granger cause IP at1 signi_cance level This proves that BSE is a leading indicator of industrialproduction and there exist a leadlag relationship between the two indicatorsEDHEC Business School 236 CONCLUSIONS6 Conclusions

In this paper I tested the relations between Indian stock market represented byBSE and domestic and global macro economic factors The research concludesthat the India stock markets are mainly driven by domestic demand and theinuence of global macro factors on the stock market is weak I also tested forGranger causality between BSE and IP and found that BSE is a leading indicatorof Industrial production and can help in predicting the industrial climate inIndiaThe research is insightful for investors and professionals who are looking forinvestment opportunities to diversify their risks As Indian stock markets aremore dependent on domestic factors one can invest in Indian indices and stocksto diversify their risks gained through investing in US and European stocksThe paper opens new doors for research in this _eld One can use variancedecomposition technique to see how much variance of BSE can be explained myvarious domestic and global macro factors Also one can use di_erent globalfactors like sovereign CDS spreads T-Bill rates a composite indicator of globaleconomy for further research on interaction between Indian stock market andglobal economyOne can also research on how various global macroeconomicnews a_ects India stock markets and for how long the e_ects persistsEDHEC Business School 247 GRAPHS AND TABLES7 Graphs and Tables71 Graphs of Time seriesEDHEC Business School 257 GRAPHS AND TABLESEDHEC Business School 267 GRAPHS AND TABLESEDHEC Business School 277 GRAPHS AND TABLESEDHEC Business School 287 GRAPHS AND TABLES72 Graphs of Time Series - Di_erencedEDHEC Business School 297 GRAPHS AND TABLESEDHEC Business School 307 GRAPHS AND TABLESEDHEC Business School 317 GRAPHS AND TABLESEDHEC Business School 327 GRAPHS AND TABLES73 Correlograms of Time seriesBSEEDHEC Business School 337 GRAPHS AND TABLESIPEDHEC Business School 347 GRAPHS AND TABLESSP500EDHEC Business School 357 GRAPHS AND TABLESUSDINREDHEC Business School 367 GRAPHS AND TABLESCPIEDHEC Business School 377 GRAPHS AND TABLESPPIEDHEC Business School 387 GRAPHS AND TABLES

M1EDHEC Business School 397 GRAPHS AND TABLES74 Tables741 Table for Unit root test of Time seriesVariables T-Stat p-valueBSE -2671 2495 SP500 -1315 8818 CPI -1909 6466 IP -1669 899 M1 -2420 3679 PPI -3353 601 USDINR -2955 1469 742 Tables for Unit root test of Di_erenced time seriesVariables T-Stat p-valueBSE -13848 000 SP500 -14832 000 CPI -3344 140 IP -3865 027 M1 -3867 026 PPI -9656 000 USDINR -13701 000 743 Tables for Residual based test of cointegrationTable 1BSE - CPIt-Statistic ProbADF test statistic -2622676 087Test critical values 1 level -25738185 level -19420410 level -1615891Table 2BSE - IPt-Statistic ProbADF test statistic -3738802 002Test critical values 1 level -25745135 level -194213610 level -1615828EDHEC Business School 407 GRAPHS AND TABLESTable 3BSE - M1t-Statistic ProbADF test statistic -2875518 041Test critical values 1 level -25737845 level -194203510 level -1615894Table 4BSE - PPIt-Statistic ProbADF test statistic -2399055 162Test critical values 1 level -25737845 level -194203510 level -1615894Table 5BSE - SP500t-Statistic ProbADF test statistic -1427184 1430Test critical values 1 level -25737845 level -1942035

10 level -1615894EDHEC Business School 417 GRAPHS AND TABLESTable 6BSE - USDINRt-Statistic ProbADF test statistic -1659522 917Test critical values 1 level -25738185 level -19420410 level -1615891EDHEC Business School 427 GRAPHS AND TABLES744 Johansen cointegration testEDHEC Business School 437 GRAPHS AND TABLESTable 2EDHEC Business School 447 GRAPHS AND TABLESTable 3EDHEC Business School 457 GRAPHS AND TABLES745 Impulse response testsEDHEC Business School 467 GRAPHS AND TABLESEDHEC Business School 477 GRAPHS AND TABLESEDHEC Business School 487 GRAPHS AND TABLES746 Granger causality test between IP and BSEEDHEC Business School 498 BIBLIOGRAPHY8 BibliographyEugene F Fama Ination Output and Money Journal of Business 1982Eugene F Fama Stock Returns Real activity and Money The American Eco-nomic Review 1981Eugene F Fama Stock Returns Expected Returns and Real activity Journal ofFinance 1990Pal and Mittal Impact of macroeconomic indicators in Indian capital marketsJournal of Risk Finance 2011Shahid Ahmed Aggregate Economic Variables and Stock Markets in India In-ternational Research Journal of Finance and Economics 2008Sahu and Dhiman Correlation and Causality between Stock Market and MacroEconomic Variables in India An Empirical Study 2010 International Confer-ence on E-Business and Economics 2011Mohammad Bayezid Ali Impact of Micro Variables on Emerging Stock MarketReturn A case on Dhaka Stock Exchange (DSE) Interdisciplinary Journal ofResearch in Business 2011Napphon Tangjitprom Macroeconomic Factors of Emerging Stock Market Theevidence from Thailand International Journal of Finance and Research 2012Sayed Mehdi Hosseini The Role of Macroeconomic Variables on Stock MarketIndex in China and India International Journal of Economics and Finance2011John Y Campbell Pitfalls and Opportunities What Macroeconomists shouldknow about Unit Roots NBER Working Papers 1991Hacker and Hatemi The properties of Procedures Dealing with Uncertainityabout Intercept and Deterministic Trend in Unit Root Testing CESIS Elec-tronic Working Papers 2010Elder and Kennedy Testing for Unit Roots What should Students be TaughtNasseh and Strauss Stock Prices and domestic and international macroeco-

nomic activity a cointegration approach The Quarterly Review of Economicsand Finance 2000Engle and Granger Co-Integration and Error Correction Representation Es-timation and Testing Econometrica 1987Eugene F Fama Stock Returns Real Activity Ination and Money 1981American Economic AssociationNaliniprave Tripathy Causal Relationship between Macro-Economic Indicatorsand Stock Market in India Asian Journal of Finance and Accounting 2011Rogalski and Vinso Stock Returns Money Supply and the Direction of Causal-ity The Journal of Finance 1977James et al A VARMA Analysis of the Causal Relations Among Stock Re-turns Real output and Nominal Interest Rates 1985 The Journal of FinanceBailey and Chung Risk and return in the Philippine Equity market A multi-factor exploration Paci_c-Basin Finance Journal 1996Nai-Fu Chen Financial Investment opportunities and the Macroeconomy TheJournal of Finance 1991GB Wickremasinghe Macroeconomic forces and stock prices Some empiricalevidence from an emerging stock markets University of Wollongong 2006EDHEC Business School 508 BIBLIOGRAPHYYao Juo and Loh On Chinas Monetary Policy and Asset Prices University ofNottingham- China policy Institute 2011Bilson et al Selecting macroeconomic variables as explanatory factors of emerg-ing stock market returns Paci_c-Basin Finance Journal 2001CHen Roll and Ross Economic forces and the Stock Markets The Journal ofBusiness 1986William H Greene Econometric Analysis 6th Edition Pearson InternationalEditionRuey Tsay Analysis of Financial Time seriesChris Brooks Introductory Econometrics for Finance Cambridge PublicationsEDHEC Business School 51

To circumvent this problem we use variables that have been tested in earlierresearches and that have been proven to have e_ect on stock markets I alsotest a few macro factors that have some _nancial theory behind them that con-nect them to stock markets Ali (2011) Wickremasinghe (2006) Bilson etal(2001) and Bailey (1996) _nd that Industrial production CPI exchange rateM1 money supply GDP are few of the macro economic factors that can signi_-cantly explain stock returns Sahu(2011) Ahmed(2008) Tripathy(2011) studyon Indian markets speci_cally show that Industrial Production Exchange rateInation index are macro economic indicators that have a strong positive ornegative relationship with the stock markets So in our study we test 5 macroeconomic variables namely M1 money supply Consumer and Producer price In-dex Industrial production Exchange rate The time period for these indicatorsis from 1990-2011 The data for Ination indices Industrial production andexchange rate has been pulled from Bloombergc and Datastreamc The datahas been processed for errors and missing values Data for M1 money supplyhas been pulled from RBI website For most of the indices like ination andIndustrial production index the base year has been changed to 1990 Also assome of the indices are in levels and some in actual _gures (M1 money supply)we convert all of the indicators to level form (starting at 100 in 1990)EDHEC Business School 133 DATA32 Description of Stock Market IndicesCompared to Macro Indicators stock market data is relatively easy to _nd andhas considerably long history Also the stock market data is a real time data soit has a very high frequency of seconds Here in our analysis we will make use ofBSE (Bombay Stock Exchange) as representation of Indian markets and SP500(Standard and Poors 500 Index) as representation of global factors BSE is amarket cap-weighted of 30 stocks It is the oldest Index in the Asian markets(established in 1875) and have had a long history We choose this index as it isthe Index that represent the most liquid and traded stocks of the Indian stockmarket Also the index is most traded index in India and a good representationof trade prices of the stocks Even in terms of an orderly growth much beforethe actual legislations were enacted BSE Limited had formulated a compre-hensive set of Rules and Regulations for the securities market It had also laiddown best practices which were adopted subsequently by 23 stock exchangeswhich were set up after India gained its independence Our choice of SP500 isbased on the fact that it has a long history and many researchers have usedthis index as a good proxy representation of global markets and economic con-ditions We will take the monthly returns of each of the indices from 1990-2011in accordance with data frequency of macro economic variables Also as theindices have di_erent levels at beginning of 1990 we rebase both the indices tobase year of 1990 starting at a level of 100EDHEC Business School 144 METHODOLOGY4 Methodology41 Construction of Time SeriesThe _rst step in constructing an econometric model is constructing time seriesall of which are in same units Most of the time series used in our analysis are indi_erent formats For example CPI PPI BSE Index SP500 are in levels M1money supply USDINR exchange rate is in absolute current format Industrialproduction is in absolute production levels So _rst we convert all of the giventime series to level The way we construct time series in levels is _rstly takingthe initial data point of each time series as 100 We then _nd the percentagechange from one period to the next one for each time series using a continuouscompounding assumption (taking a natural log of change in values) In math-ematical terms it can be stated as Assume the original Index value at time tto be It and at time t+1 to be It + 1 Then we can compute the new rebasedindex by formulaRIt+1 = RIt _ (1 + ln(It+1=It))

whereRIt= Rebased Index at time tRIt+1=Rebased Index at time t+1We can use these rebased indices in building and testing our econometric model42 Unit Root Test and StationarityUnit root test is to _nd whether the series is stationary or non-stationary Astrictly stationary process is one where for any t1 t2 tt 2Z any k 2Z andT=12Fyt1 yt2 yt3 ytT

(y1 yT ) = Fyt1+k yt2+k yt3+k ytT+k

(y1 yT )where F represents joint distribution function of the set of random variablesIt can also be stated that the probability measure of sequence of yt is same asyt+k for all k In other words a series is stationary if the distribution of its valueremain the same as time progresses Similar to the concept of strict stationaryis weakly stationary process A weakly stationary process is one which has aconstant mean variance and autocovariance structure Stationary is a necessarycondition for a time series to be tested in regression A non-stationary seriescan have several problems like1 The shocks given to the series would not die of gradually resulting inincrease of variance as time passes2 If the series is non stationary then it can lead to spurious regressions If twoseries are generated independent of each other then if one is regressed onother it will result in very low R2 values But if two series are trending overtime then a regression of one over the other will give high R2 even thoughthe series may be unrelated to each other So if normal regressions toolsEDHEC Business School 154 METHODOLOGYare used on non stationary data then it may result in good but valuelessresults3 If the variables employed in a regression model are not stationary thenit can be proved that the standard assumptions for asymptotic analysiswill not be valid In other words the usual t-ratios will not follow at-distribution and the F-statistic will not follow an F-distribution and soonStationarity is a desirable condition for any time series so that it can be usedin regressions and give meaningful result that have some value to test for sta-tionarity a quick and dirty way is looking at the autocorrelation and partialcorrelation function of the series If the series is stationary then the autocorre-lation function should die o_ gradually after few lags and the partial correlationfunction will me non zero for some lags and zero thereafter Also we can usethe Ljung-Box test for testing that all m of _k autocorrelation coe_cients arezero using Q-statistic given by formulaQ = T(T + 2)_mk=1_k2T 1048576 k_ _2where T = Sample size and m = Maximum lag lengthThe lag length selection can be based on di_erent Information Criteria likeAkaikes Information criteria (AIC) Schwarzs Bayesian information criteria(SBIC) Hannan-Quinn criterion (HQIC) Mathematically di_erent criteria arerepresented asAIC = ln(_2) + 2kTSBIC = ln(_2) + kT lnTHQIC = ln(_2) + 2kT ln(ln(T))

For a better test for stationarity we use augmented Dickey fuller Unit roottest on each time series separately Augmented Dickey Fuller test is test ofnull hypothesis that the time series contains a unit roots against a alternativehypothesis that the series is stationary421 Mathematical representation of Stationary series and unit roottestAssume a variable Y whose structure can be given by AR process with no driftequationyt = _1yt10485761 + _2yt10485762 + _3yt10485763 + + _nyt1048576n + ut (2)where ut is the residual at time t Using a Lag operator L we can write eq(1)asyt = _1L1yt + _2L2yt + _3L3yt + + _nLnyt + ut (3)EDHEC Business School 164 METHODOLOGYRearranging eqn (2) we getyt 1048576 _1L1yt 1048576 _2L2yt 1048576 _3L3yt + 1048576 _nLnyt = ut (4)yt(1 1048576 _1L1 1048576 _2L2 1048576 _3L3 + 1048576 _nLn) = ut (5)or_(L)yt = ut (6)The time series is stationary if we can write eqn(5) in formyt = _(L)10485761ut (7)with _(L)10485761 converging to zero It means the autocorrelation function woulddecline as lag length is increased If eqn (6) is expanded to a MA(1) processthe coe_cients of residuals should decrease such that the the residuals that thee_ect of residuals decrease with increase in lags SO if the process is stationarythe coe_cients of residuals will converge to zero and for non-stationary seriesthey will and converge to zero and will have long term e_ect The condition fortesting of unit root for an AR process is that the roots of eqn(6) or Charac-teristic equation should lie outside unit circle422 Augmented Dickey Fuller Unit Root TestConsider an AR(1) process of variable Yyt = _yt10485761 + ut (8)Subtracting yt10485761 from both sides of eqn(7) we get_y = (_ 1048576 1)yt10485761 + ut (9)Eqn(8) is the test equation for Dickey Fuller test For Dickey-Fuller Unit roottestNull Hypothesis The value of _ is equal to 1 or value of _10485761 is equal to 0 vsAlternate Hypothesis The value of _ is less than one or value of _ 1048576 1 is lessthan zero Augmented Dickey-Fuller test is similar to normal Dickey-Fuller testsexcept it takes the lag structure of more than one into account_y = yt10485761 +Xpi=1_i_yt1048576i + ut (10)If the series has one or more unit root it is said to be integrated of order nwhere n is the number of unit roots of the characteristic equation To makethese time series stationary they needs to be di_erenced Mathematically ifyt _ I (n) (11)then_(d) yt _ I (0) (12)To make our time-series stationary we will use the natural log returns of theseseries in the analysisEDHEC Business School 174 METHODOLOGY43 Testing Long Term RelationshipsEngle and Granger (1987) in their seminal paper described cointegration whichforms the basis for testing for long term relationship between variables Accord-ing to Engle and Granger two variables are cointegrated if they are integratedprocess in their natural form (of the same order) but a weighted combination

of the variables can be found such that the combined new variable is integratedof order less than the order of individual time series Mathematically assumeyt to be a k X 1 vector of variables then the components are cointegrated orintegrated of order (db) if1 All components of yt are I(d)2 There is at least one vector of coe_cients _ such that_0

yt _ I (d 1048576 b) (13)As most of the _nancial time series are integrated of order one we will restrictourselves to case d=b=1 Two or more variables are said to be cointegrated ifthere exist a linear combination of these variables that is stationary Many ofthe series are non-stationary but move together over time which implies twoseries are bound by some common force or factor in long run We will test forcointegration by a residual-based approach and Johansens VAR methodResidual Based approach Consider a modelyt = _1 + _2x2t + _3x3t + + ut (14)where yt x2t x3t are all integrated of order N Now if the residual of this re-gression ut is stationary then we can say that the variables are cointegrated elsethere exist no long term relationship between the variables To test the resid-ual for stationarity we will run Augmented Dickey-Fuller tests on the residualsUnder the Null hypothesis the residual are integrated of order one or more andunder alternate hypothesis the residuals are I(0)431 Johansen test for CointegrationJohansen test for cointegration presents a better model for testing multiplecointegration among multiple variables The Residual based approach can only_nd atmost one cointegration and can be tested for a model with two variablesEven if more than two variables are present in the equation that are cointegratedthe Residual based approach will give only one cointegration SO we will useJhoansen VAR based cointegration for testing more than one cointegrationSuppose that a set of g variables are under consideration that are I(1) andwhich are thought to be cointegrated A VAR with k lags containing thesevariables could be set upyt = _1yt10485761 + _2yt10485762 + _ _ _ + _kyt1048576k + ut (15)g _ 1 g _ g g _ 1 g _ g g _ 1 g _ g g _ 1 g _ 1EDHEC Business School 184 METHODOLOGYIn order to use the Johansen test the VAR above should be turned into avector error correction model of form_yt = _yt1048576k + 1_yt10485761 + 2_yt10485762 + _ _ _ + k10485761_yt1048576(k10485761) + ut (16)where _ = (_ki=1_i) 1048576 Ig and i = (_ij=1_j) 1048576 IgThe Johansens test centers around testing the _ matrix which is the matrixthat represents the long term cointegration between the variables The test fornumber of cointegration is calculated by looking at the rank of the _ matrixthrough its eigenvalues The rank of the matrix is equal to number of roots(eigenvalues) _i of the matrix that are di_erent from zero The roots should beless than 1 in absolute value and positive If the variables are not cointegratedthe rank of the matrix will not be signi_cantly di_erent from zero ie _i _ 0There are two test statistics for Johansen test _tracer and _max_trace (r) = 1048576TPgi=r+1 ln(1 1048576 _ _i)and_max(r r + 1) = 1048576Tln(1 1048576 _r_+1)_trace is a test statistic for joint test where the null hypothesis is that thenumber of cointegration vector is less than or equal to r against an alternativethat there are more than r_max conducts another separate test on eigenvalues and has null hypothesis that

the number of cointegrating vector is r against r+144 Impulse ResponseOnce we have determined whether the variables have long term relationship ornot we can form a multivariate VAR model for the variables A multivariateVAR model between g variables is a model where the current value of a variabledepend on di_ernt combinations of the previous k values of all the variables anderror terms A general representation of the model can beyBSEt = _ + _BSEyBSE + _IP yIP + CPIyCPI + _M1yM1 + _SP500ySP500 + u1t(17)where all the coe_cients except _ are g _ k matrices and all variables y are k_ 1 matricesOnce we have formed a model like this we can use the model for Impulse re-sponse A VAR(p) model can be written as a linear fuction of the past innova-tions that isrt = _ + at + 1at10485761 + 2at10485762 + (18)where _ = [_(1)]10485761_0 provided that the inverse exists and the coe_cient ma-trices i can be obtained by equating the coe_cients of Bi in the equation(I 1048576 _1B 1048576 1048576 _PBP )(I + 1B + 2B2 + ) = I (19)EDHEC Business School 194 METHODOLOGYwhere I is the Identity martix This is a moving average representation of rtwith the coe_cient matrix i being the impact of the past innovation at1048576i onrt Equivalently i is the e_ect of at on the future observation rt+i Therefore i is often referred to as the Impulse Response Function of rt For our impulseresponse we will use equation of variables in _rst di_ernce form like_BSEt = _t +Xki=0_11(i)_BSEt1048576i +Xkj=1_12(j)_MIt1048576j + _BSEt (20)_MIt = _t +Xki=0_21(i)_MIt1048576i +Xkj=1_22(j)_BSEt1048576j + _MIt (21)Grangers causality and Blocks F test of a VAR model will suggest which ofthe variables have statistically signi_cant impacts on the future values of othervariables in the system But F-test results cannot explain the sign of the re-lationship nor how long these e_ects require to take place Such informationwill however be given by an examination of the VARs impulse responses andvariance decompositions Impulse response is a technique that trace out theresponsiveness of the dependent variable in the VAR to shocks of each of theother variables So for each variable from each equation separately we will applya unit shock to the error and trace the e_ects upon the VAR system over timeBy using the impulse response technique we can determine how responsive isthe BSE stock index to Indian macro indicators and SP500 This will help usdetermine whether the BSE index is more reactive to domestic news or globalnewsEDHEC Business School 205 RESULTS5 ResultsBefore we use the time series for VAR analysis or cointegration tests we need todetermine whether the series are Stationary or not If the series are stationaryin levels we can use them directly else we need to use the di_erenced time seriesOne way to look for autocorrelation or integrated process is to see the graphsof the various time series used Section 71 shows the graphs of variables we

use for our analysis As we can see from the graphs all of the time series havea trend in long run which points to an integrated process As a second stepwe plot the graphs of di_erenced time series in Section 52 We can see thatthe di_erenced graphs in Section 72 dont show a long term trend and crossthe X-axis frequently This is usually a property of I(1) processes So we checkthe series for autocorrelations at di_erent lag lengths Section 73 shows cor-relograms graph autocorrelation coe_cient partial autocorrelation coe_cientQ-Stat and p-value for various time series up to 36 lags As can be seen in thetables the Q-stat for all lags is zero and we can reject the joint null hypothesisthat all the autocorrelations up to 36 lags are zero Table 741 shows that ifwe conduct a Unit root test on levels of the series we _nd that all the 7 seriesare integrated as we cannot reject the t-stat for unit root at 1 level But ifwe conduct the same test on di_erenced values of the series we _nd that we canreject the null hypothesis of unit root at 1 signi_cance level for all the seriesexcept CPI This tells us that all the series are I(1) as there _rst di_erence seriesare I(0)As our series are I(1) we will work with index levels of time series to determineif there exist one or more cointegrating relationships between the series Tablesin subsection 743 are based on residual approach where we run a regression ofBSE and various macroeconomic indicators and test the residuals for unit rootusing Augmented Dickey-Fuller test As we assume the two series are cointe-grated we conduct the test with no trend and intercept If the two series arecointegrated then the errors should not have any trend or intercept We see thatwe can reject the null hypothesis of unit root at 1 signi_cance for CPIIP M1We can reject the null of unit root for PPI at 5 and for SP500 and USDINRwe cant reject the null hypothesis of unit root at even 5 level This pointsto the fact that BSE has a strong long term relationship with IP M1 moneysupply CPI at 1 level with IP M1 CPI PPI at 5 signi_cance level AlsoBSE has no long term relationship with SP500 and USD INR exchange rateTo test for multiple cointegrating relationship we now employ a Johansen VARbased cointegration test The results of the test are displayed in subsection744 The _rst panel of the test results displays the value of _trace and_maxof Johansen test with di_erent assumptions about intercept and trend We cansee from this panel that when we consider a functional form of intercept and noTrend we have atleast and atmost three cointegrating relationships The secondpanel of the results display the value of information criteria for lag lengths Formost of the models we see that Akalike criteria points to a lag of three andSchwarz criteria points to a lag of one To estimate the cointegrating model wechoose the model with intercept and no trend and run a cointegration testTestresults are shown in Table 2 of subsection 744 At 5 signi_cance level wecan reject the null of atmost two cointegrating factors for _trace and same for_max Now to test which all variables have a long tern relationship we perform aRestricted cointegration with vector error correction model As we had alreadyEDHEC Business School 215 RESULTSseen in our residual based test of cointegration that BSE has no cointegratingrelationship with SP500 and USDINR we create a restricted cointegration modelwhere we set coe_cients of SP500 and USDINR as zero The test results aredisplayed in Table 3 of subsection 744 In this case as there are two restrictionsthe test statistic follow _2 with two degrees of freedom We can see that thep-value for the test is 1333 which tells us that the restrictions are supportedby data at 10 level of signi_cance So we can conclude that the BSE has along term relationship with CPIIPPPIM1 money supply but has no long termrelationship with SP500 and USDINR exchange rate One interpretation of thisresult can be that the Indian stock market represented here by BSE Sensexmoves more in accordance with domestic factors like Industrial production M1money supply Consumer price index and Producer Price index than with globalfactors or in other words as BSE is representation of largest market cap Indiancompanies we can say that the biggest companies in India are ones that are

more dependent on domestic demand rather than exports This result presentsan opportunity for international investors to diversify their portfolio by invest-ing in BSE Sensex as it is decoupled with global markets and macroeconomicfactorsWe use A bivariate Vector Autoregression (BVAR) technique to analyze thedynamic interaction between real asset prices and macro economy VAR ispreferred method to study Macroeconomy and asset prices where variables en-dogenously e_ect each otherWe begin with a bivariate VAR with no restriction Asset prices and instru-ments are allowed to respond to each other freely For paired variables withcointegration relationship VAR is performed at levels whilst for those that arenot cointegrated VAR is performed at _rst di_erence Constant term is ignoredwith loss of generality We use the Bivariate Autoregression analysis for bothimpulse response and Grangers causality testsImpulse response results are displayed in subsection 745 From _rst graph ofimpulse response of BSE to USDINR we can see that USDINR has a negativeimpact on BSE As impulse response is response of BSE to shocks given to US-DINR we can see that a positive shock or unexpected appreciation INR valuewrt USD will have a negative e_ect on BSE for few lags and will disappearafter few lags If we look at the constituents of BSE Index over time we seethat most of the time some of its constituent are companies that thrive on ex-ports Some of the biggest Market-Cap in India are companies in service sectorlike Infosys TCS etc that are hugely dependent on services provided to clientsfrom Europe and US So an appreciation of INR compared to USD makesthese _rms costlier for the global clients and in turn reduces the income of thesecompanies As the _rms revenue pro_t decreases the value of the stock alsodecreases that in turn a_ects the returns of BSE SensexSecond graph (betwen BSE and SP500) shows that increase in SP500 has a pos-itive e_ect on BSE as higher returns of SP500 indicate strong global economywhich in turn results in higher trade between countries The positive responseof BSE to one unit shock to SP500 indicates a spillover e_ect of global factorson Indian economy but the response is weak as can be seen from the graphMoving forward response of BSE to shocks in M1 money supply CPI PPImake economic sense As for M1 money supply one unit shock means increasein M1 money supply This increase in money supply allows companies to bor-row more money from banks at lower rates which they can use for investingEDHEC Business School 225 RESULTSin pro_table projects and generating larger cash ows For Ination indicatorsone unit shock means increase in ination This increase in ination results inhigher costs for the companies that in turn reduces their pro_t margins and asa result value of stocksBy looking at the graphs we can also see that shocks to Indian macroeconomicindicators creates stronger response by BSE as compared to global factors likeSP500 or USDINR This indicates that BSE Index is driven by companies thatdepend hugely on domestic demand rather than exports Response of BSE toshocks to Industrial Production are contradictory to theory In theory an in-crease in industrial production should result in positive response from BSE butour analysis shows the other way A possible reason for this response could bethat industrial production time series is seasonal as can be seen from the graphSo there is a possibility of a leadlag relationship between the two variablesTo test for possibility of leadlag relationship we run a Grangers causality testbetween BSE and IP The result in section 646 shows that at a lag lengthof 4 we can reject the Null hypothesis of BSE does not Granger cause IP at1 signi_cance level This proves that BSE is a leading indicator of industrialproduction and there exist a leadlag relationship between the two indicatorsEDHEC Business School 236 CONCLUSIONS6 Conclusions

In this paper I tested the relations between Indian stock market represented byBSE and domestic and global macro economic factors The research concludesthat the India stock markets are mainly driven by domestic demand and theinuence of global macro factors on the stock market is weak I also tested forGranger causality between BSE and IP and found that BSE is a leading indicatorof Industrial production and can help in predicting the industrial climate inIndiaThe research is insightful for investors and professionals who are looking forinvestment opportunities to diversify their risks As Indian stock markets aremore dependent on domestic factors one can invest in Indian indices and stocksto diversify their risks gained through investing in US and European stocksThe paper opens new doors for research in this _eld One can use variancedecomposition technique to see how much variance of BSE can be explained myvarious domestic and global macro factors Also one can use di_erent globalfactors like sovereign CDS spreads T-Bill rates a composite indicator of globaleconomy for further research on interaction between Indian stock market andglobal economyOne can also research on how various global macroeconomicnews a_ects India stock markets and for how long the e_ects persistsEDHEC Business School 247 GRAPHS AND TABLES7 Graphs and Tables71 Graphs of Time seriesEDHEC Business School 257 GRAPHS AND TABLESEDHEC Business School 267 GRAPHS AND TABLESEDHEC Business School 277 GRAPHS AND TABLESEDHEC Business School 287 GRAPHS AND TABLES72 Graphs of Time Series - Di_erencedEDHEC Business School 297 GRAPHS AND TABLESEDHEC Business School 307 GRAPHS AND TABLESEDHEC Business School 317 GRAPHS AND TABLESEDHEC Business School 327 GRAPHS AND TABLES73 Correlograms of Time seriesBSEEDHEC Business School 337 GRAPHS AND TABLESIPEDHEC Business School 347 GRAPHS AND TABLESSP500EDHEC Business School 357 GRAPHS AND TABLESUSDINREDHEC Business School 367 GRAPHS AND TABLESCPIEDHEC Business School 377 GRAPHS AND TABLESPPIEDHEC Business School 387 GRAPHS AND TABLES

M1EDHEC Business School 397 GRAPHS AND TABLES74 Tables741 Table for Unit root test of Time seriesVariables T-Stat p-valueBSE -2671 2495 SP500 -1315 8818 CPI -1909 6466 IP -1669 899 M1 -2420 3679 PPI -3353 601 USDINR -2955 1469 742 Tables for Unit root test of Di_erenced time seriesVariables T-Stat p-valueBSE -13848 000 SP500 -14832 000 CPI -3344 140 IP -3865 027 M1 -3867 026 PPI -9656 000 USDINR -13701 000 743 Tables for Residual based test of cointegrationTable 1BSE - CPIt-Statistic ProbADF test statistic -2622676 087Test critical values 1 level -25738185 level -19420410 level -1615891Table 2BSE - IPt-Statistic ProbADF test statistic -3738802 002Test critical values 1 level -25745135 level -194213610 level -1615828EDHEC Business School 407 GRAPHS AND TABLESTable 3BSE - M1t-Statistic ProbADF test statistic -2875518 041Test critical values 1 level -25737845 level -194203510 level -1615894Table 4BSE - PPIt-Statistic ProbADF test statistic -2399055 162Test critical values 1 level -25737845 level -194203510 level -1615894Table 5BSE - SP500t-Statistic ProbADF test statistic -1427184 1430Test critical values 1 level -25737845 level -1942035

10 level -1615894EDHEC Business School 417 GRAPHS AND TABLESTable 6BSE - USDINRt-Statistic ProbADF test statistic -1659522 917Test critical values 1 level -25738185 level -19420410 level -1615891EDHEC Business School 427 GRAPHS AND TABLES744 Johansen cointegration testEDHEC Business School 437 GRAPHS AND TABLESTable 2EDHEC Business School 447 GRAPHS AND TABLESTable 3EDHEC Business School 457 GRAPHS AND TABLES745 Impulse response testsEDHEC Business School 467 GRAPHS AND TABLESEDHEC Business School 477 GRAPHS AND TABLESEDHEC Business School 487 GRAPHS AND TABLES746 Granger causality test between IP and BSEEDHEC Business School 498 BIBLIOGRAPHY8 BibliographyEugene F Fama Ination Output and Money Journal of Business 1982Eugene F Fama Stock Returns Real activity and Money The American Eco-nomic Review 1981Eugene F Fama Stock Returns Expected Returns and Real activity Journal ofFinance 1990Pal and Mittal Impact of macroeconomic indicators in Indian capital marketsJournal of Risk Finance 2011Shahid Ahmed Aggregate Economic Variables and Stock Markets in India In-ternational Research Journal of Finance and Economics 2008Sahu and Dhiman Correlation and Causality between Stock Market and MacroEconomic Variables in India An Empirical Study 2010 International Confer-ence on E-Business and Economics 2011Mohammad Bayezid Ali Impact of Micro Variables on Emerging Stock MarketReturn A case on Dhaka Stock Exchange (DSE) Interdisciplinary Journal ofResearch in Business 2011Napphon Tangjitprom Macroeconomic Factors of Emerging Stock Market Theevidence from Thailand International Journal of Finance and Research 2012Sayed Mehdi Hosseini The Role of Macroeconomic Variables on Stock MarketIndex in China and India International Journal of Economics and Finance2011John Y Campbell Pitfalls and Opportunities What Macroeconomists shouldknow about Unit Roots NBER Working Papers 1991Hacker and Hatemi The properties of Procedures Dealing with Uncertainityabout Intercept and Deterministic Trend in Unit Root Testing CESIS Elec-tronic Working Papers 2010Elder and Kennedy Testing for Unit Roots What should Students be TaughtNasseh and Strauss Stock Prices and domestic and international macroeco-

nomic activity a cointegration approach The Quarterly Review of Economicsand Finance 2000Engle and Granger Co-Integration and Error Correction Representation Es-timation and Testing Econometrica 1987Eugene F Fama Stock Returns Real Activity Ination and Money 1981American Economic AssociationNaliniprave Tripathy Causal Relationship between Macro-Economic Indicatorsand Stock Market in India Asian Journal of Finance and Accounting 2011Rogalski and Vinso Stock Returns Money Supply and the Direction of Causal-ity The Journal of Finance 1977James et al A VARMA Analysis of the Causal Relations Among Stock Re-turns Real output and Nominal Interest Rates 1985 The Journal of FinanceBailey and Chung Risk and return in the Philippine Equity market A multi-factor exploration Paci_c-Basin Finance Journal 1996Nai-Fu Chen Financial Investment opportunities and the Macroeconomy TheJournal of Finance 1991GB Wickremasinghe Macroeconomic forces and stock prices Some empiricalevidence from an emerging stock markets University of Wollongong 2006EDHEC Business School 508 BIBLIOGRAPHYYao Juo and Loh On Chinas Monetary Policy and Asset Prices University ofNottingham- China policy Institute 2011Bilson et al Selecting macroeconomic variables as explanatory factors of emerg-ing stock market returns Paci_c-Basin Finance Journal 2001CHen Roll and Ross Economic forces and the Stock Markets The Journal ofBusiness 1986William H Greene Econometric Analysis 6th Edition Pearson InternationalEditionRuey Tsay Analysis of Financial Time seriesChris Brooks Introductory Econometrics for Finance Cambridge PublicationsEDHEC Business School 51

whereRIt= Rebased Index at time tRIt+1=Rebased Index at time t+1We can use these rebased indices in building and testing our econometric model42 Unit Root Test and StationarityUnit root test is to _nd whether the series is stationary or non-stationary Astrictly stationary process is one where for any t1 t2 tt 2Z any k 2Z andT=12Fyt1 yt2 yt3 ytT

(y1 yT ) = Fyt1+k yt2+k yt3+k ytT+k

(y1 yT )where F represents joint distribution function of the set of random variablesIt can also be stated that the probability measure of sequence of yt is same asyt+k for all k In other words a series is stationary if the distribution of its valueremain the same as time progresses Similar to the concept of strict stationaryis weakly stationary process A weakly stationary process is one which has aconstant mean variance and autocovariance structure Stationary is a necessarycondition for a time series to be tested in regression A non-stationary seriescan have several problems like1 The shocks given to the series would not die of gradually resulting inincrease of variance as time passes2 If the series is non stationary then it can lead to spurious regressions If twoseries are generated independent of each other then if one is regressed onother it will result in very low R2 values But if two series are trending overtime then a regression of one over the other will give high R2 even thoughthe series may be unrelated to each other So if normal regressions toolsEDHEC Business School 154 METHODOLOGYare used on non stationary data then it may result in good but valuelessresults3 If the variables employed in a regression model are not stationary thenit can be proved that the standard assumptions for asymptotic analysiswill not be valid In other words the usual t-ratios will not follow at-distribution and the F-statistic will not follow an F-distribution and soonStationarity is a desirable condition for any time series so that it can be usedin regressions and give meaningful result that have some value to test for sta-tionarity a quick and dirty way is looking at the autocorrelation and partialcorrelation function of the series If the series is stationary then the autocorre-lation function should die o_ gradually after few lags and the partial correlationfunction will me non zero for some lags and zero thereafter Also we can usethe Ljung-Box test for testing that all m of _k autocorrelation coe_cients arezero using Q-statistic given by formulaQ = T(T + 2)_mk=1_k2T 1048576 k_ _2where T = Sample size and m = Maximum lag lengthThe lag length selection can be based on di_erent Information Criteria likeAkaikes Information criteria (AIC) Schwarzs Bayesian information criteria(SBIC) Hannan-Quinn criterion (HQIC) Mathematically di_erent criteria arerepresented asAIC = ln(_2) + 2kTSBIC = ln(_2) + kT lnTHQIC = ln(_2) + 2kT ln(ln(T))

For a better test for stationarity we use augmented Dickey fuller Unit roottest on each time series separately Augmented Dickey Fuller test is test ofnull hypothesis that the time series contains a unit roots against a alternativehypothesis that the series is stationary421 Mathematical representation of Stationary series and unit roottestAssume a variable Y whose structure can be given by AR process with no driftequationyt = _1yt10485761 + _2yt10485762 + _3yt10485763 + + _nyt1048576n + ut (2)where ut is the residual at time t Using a Lag operator L we can write eq(1)asyt = _1L1yt + _2L2yt + _3L3yt + + _nLnyt + ut (3)EDHEC Business School 164 METHODOLOGYRearranging eqn (2) we getyt 1048576 _1L1yt 1048576 _2L2yt 1048576 _3L3yt + 1048576 _nLnyt = ut (4)yt(1 1048576 _1L1 1048576 _2L2 1048576 _3L3 + 1048576 _nLn) = ut (5)or_(L)yt = ut (6)The time series is stationary if we can write eqn(5) in formyt = _(L)10485761ut (7)with _(L)10485761 converging to zero It means the autocorrelation function woulddecline as lag length is increased If eqn (6) is expanded to a MA(1) processthe coe_cients of residuals should decrease such that the the residuals that thee_ect of residuals decrease with increase in lags SO if the process is stationarythe coe_cients of residuals will converge to zero and for non-stationary seriesthey will and converge to zero and will have long term e_ect The condition fortesting of unit root for an AR process is that the roots of eqn(6) or Charac-teristic equation should lie outside unit circle422 Augmented Dickey Fuller Unit Root TestConsider an AR(1) process of variable Yyt = _yt10485761 + ut (8)Subtracting yt10485761 from both sides of eqn(7) we get_y = (_ 1048576 1)yt10485761 + ut (9)Eqn(8) is the test equation for Dickey Fuller test For Dickey-Fuller Unit roottestNull Hypothesis The value of _ is equal to 1 or value of _10485761 is equal to 0 vsAlternate Hypothesis The value of _ is less than one or value of _ 1048576 1 is lessthan zero Augmented Dickey-Fuller test is similar to normal Dickey-Fuller testsexcept it takes the lag structure of more than one into account_y = yt10485761 +Xpi=1_i_yt1048576i + ut (10)If the series has one or more unit root it is said to be integrated of order nwhere n is the number of unit roots of the characteristic equation To makethese time series stationary they needs to be di_erenced Mathematically ifyt _ I (n) (11)then_(d) yt _ I (0) (12)To make our time-series stationary we will use the natural log returns of theseseries in the analysisEDHEC Business School 174 METHODOLOGY43 Testing Long Term RelationshipsEngle and Granger (1987) in their seminal paper described cointegration whichforms the basis for testing for long term relationship between variables Accord-ing to Engle and Granger two variables are cointegrated if they are integratedprocess in their natural form (of the same order) but a weighted combination

of the variables can be found such that the combined new variable is integratedof order less than the order of individual time series Mathematically assumeyt to be a k X 1 vector of variables then the components are cointegrated orintegrated of order (db) if1 All components of yt are I(d)2 There is at least one vector of coe_cients _ such that_0

yt _ I (d 1048576 b) (13)As most of the _nancial time series are integrated of order one we will restrictourselves to case d=b=1 Two or more variables are said to be cointegrated ifthere exist a linear combination of these variables that is stationary Many ofthe series are non-stationary but move together over time which implies twoseries are bound by some common force or factor in long run We will test forcointegration by a residual-based approach and Johansens VAR methodResidual Based approach Consider a modelyt = _1 + _2x2t + _3x3t + + ut (14)where yt x2t x3t are all integrated of order N Now if the residual of this re-gression ut is stationary then we can say that the variables are cointegrated elsethere exist no long term relationship between the variables To test the resid-ual for stationarity we will run Augmented Dickey-Fuller tests on the residualsUnder the Null hypothesis the residual are integrated of order one or more andunder alternate hypothesis the residuals are I(0)431 Johansen test for CointegrationJohansen test for cointegration presents a better model for testing multiplecointegration among multiple variables The Residual based approach can only_nd atmost one cointegration and can be tested for a model with two variablesEven if more than two variables are present in the equation that are cointegratedthe Residual based approach will give only one cointegration SO we will useJhoansen VAR based cointegration for testing more than one cointegrationSuppose that a set of g variables are under consideration that are I(1) andwhich are thought to be cointegrated A VAR with k lags containing thesevariables could be set upyt = _1yt10485761 + _2yt10485762 + _ _ _ + _kyt1048576k + ut (15)g _ 1 g _ g g _ 1 g _ g g _ 1 g _ g g _ 1 g _ 1EDHEC Business School 184 METHODOLOGYIn order to use the Johansen test the VAR above should be turned into avector error correction model of form_yt = _yt1048576k + 1_yt10485761 + 2_yt10485762 + _ _ _ + k10485761_yt1048576(k10485761) + ut (16)where _ = (_ki=1_i) 1048576 Ig and i = (_ij=1_j) 1048576 IgThe Johansens test centers around testing the _ matrix which is the matrixthat represents the long term cointegration between the variables The test fornumber of cointegration is calculated by looking at the rank of the _ matrixthrough its eigenvalues The rank of the matrix is equal to number of roots(eigenvalues) _i of the matrix that are di_erent from zero The roots should beless than 1 in absolute value and positive If the variables are not cointegratedthe rank of the matrix will not be signi_cantly di_erent from zero ie _i _ 0There are two test statistics for Johansen test _tracer and _max_trace (r) = 1048576TPgi=r+1 ln(1 1048576 _ _i)and_max(r r + 1) = 1048576Tln(1 1048576 _r_+1)_trace is a test statistic for joint test where the null hypothesis is that thenumber of cointegration vector is less than or equal to r against an alternativethat there are more than r_max conducts another separate test on eigenvalues and has null hypothesis that

the number of cointegrating vector is r against r+144 Impulse ResponseOnce we have determined whether the variables have long term relationship ornot we can form a multivariate VAR model for the variables A multivariateVAR model between g variables is a model where the current value of a variabledepend on di_ernt combinations of the previous k values of all the variables anderror terms A general representation of the model can beyBSEt = _ + _BSEyBSE + _IP yIP + CPIyCPI + _M1yM1 + _SP500ySP500 + u1t(17)where all the coe_cients except _ are g _ k matrices and all variables y are k_ 1 matricesOnce we have formed a model like this we can use the model for Impulse re-sponse A VAR(p) model can be written as a linear fuction of the past innova-tions that isrt = _ + at + 1at10485761 + 2at10485762 + (18)where _ = [_(1)]10485761_0 provided that the inverse exists and the coe_cient ma-trices i can be obtained by equating the coe_cients of Bi in the equation(I 1048576 _1B 1048576 1048576 _PBP )(I + 1B + 2B2 + ) = I (19)EDHEC Business School 194 METHODOLOGYwhere I is the Identity martix This is a moving average representation of rtwith the coe_cient matrix i being the impact of the past innovation at1048576i onrt Equivalently i is the e_ect of at on the future observation rt+i Therefore i is often referred to as the Impulse Response Function of rt For our impulseresponse we will use equation of variables in _rst di_ernce form like_BSEt = _t +Xki=0_11(i)_BSEt1048576i +Xkj=1_12(j)_MIt1048576j + _BSEt (20)_MIt = _t +Xki=0_21(i)_MIt1048576i +Xkj=1_22(j)_BSEt1048576j + _MIt (21)Grangers causality and Blocks F test of a VAR model will suggest which ofthe variables have statistically signi_cant impacts on the future values of othervariables in the system But F-test results cannot explain the sign of the re-lationship nor how long these e_ects require to take place Such informationwill however be given by an examination of the VARs impulse responses andvariance decompositions Impulse response is a technique that trace out theresponsiveness of the dependent variable in the VAR to shocks of each of theother variables So for each variable from each equation separately we will applya unit shock to the error and trace the e_ects upon the VAR system over timeBy using the impulse response technique we can determine how responsive isthe BSE stock index to Indian macro indicators and SP500 This will help usdetermine whether the BSE index is more reactive to domestic news or globalnewsEDHEC Business School 205 RESULTS5 ResultsBefore we use the time series for VAR analysis or cointegration tests we need todetermine whether the series are Stationary or not If the series are stationaryin levels we can use them directly else we need to use the di_erenced time seriesOne way to look for autocorrelation or integrated process is to see the graphsof the various time series used Section 71 shows the graphs of variables we

use for our analysis As we can see from the graphs all of the time series havea trend in long run which points to an integrated process As a second stepwe plot the graphs of di_erenced time series in Section 52 We can see thatthe di_erenced graphs in Section 72 dont show a long term trend and crossthe X-axis frequently This is usually a property of I(1) processes So we checkthe series for autocorrelations at di_erent lag lengths Section 73 shows cor-relograms graph autocorrelation coe_cient partial autocorrelation coe_cientQ-Stat and p-value for various time series up to 36 lags As can be seen in thetables the Q-stat for all lags is zero and we can reject the joint null hypothesisthat all the autocorrelations up to 36 lags are zero Table 741 shows that ifwe conduct a Unit root test on levels of the series we _nd that all the 7 seriesare integrated as we cannot reject the t-stat for unit root at 1 level But ifwe conduct the same test on di_erenced values of the series we _nd that we canreject the null hypothesis of unit root at 1 signi_cance level for all the seriesexcept CPI This tells us that all the series are I(1) as there _rst di_erence seriesare I(0)As our series are I(1) we will work with index levels of time series to determineif there exist one or more cointegrating relationships between the series Tablesin subsection 743 are based on residual approach where we run a regression ofBSE and various macroeconomic indicators and test the residuals for unit rootusing Augmented Dickey-Fuller test As we assume the two series are cointe-grated we conduct the test with no trend and intercept If the two series arecointegrated then the errors should not have any trend or intercept We see thatwe can reject the null hypothesis of unit root at 1 signi_cance for CPIIP M1We can reject the null of unit root for PPI at 5 and for SP500 and USDINRwe cant reject the null hypothesis of unit root at even 5 level This pointsto the fact that BSE has a strong long term relationship with IP M1 moneysupply CPI at 1 level with IP M1 CPI PPI at 5 signi_cance level AlsoBSE has no long term relationship with SP500 and USD INR exchange rateTo test for multiple cointegrating relationship we now employ a Johansen VARbased cointegration test The results of the test are displayed in subsection744 The _rst panel of the test results displays the value of _trace and_maxof Johansen test with di_erent assumptions about intercept and trend We cansee from this panel that when we consider a functional form of intercept and noTrend we have atleast and atmost three cointegrating relationships The secondpanel of the results display the value of information criteria for lag lengths Formost of the models we see that Akalike criteria points to a lag of three andSchwarz criteria points to a lag of one To estimate the cointegrating model wechoose the model with intercept and no trend and run a cointegration testTestresults are shown in Table 2 of subsection 744 At 5 signi_cance level wecan reject the null of atmost two cointegrating factors for _trace and same for_max Now to test which all variables have a long tern relationship we perform aRestricted cointegration with vector error correction model As we had alreadyEDHEC Business School 215 RESULTSseen in our residual based test of cointegration that BSE has no cointegratingrelationship with SP500 and USDINR we create a restricted cointegration modelwhere we set coe_cients of SP500 and USDINR as zero The test results aredisplayed in Table 3 of subsection 744 In this case as there are two restrictionsthe test statistic follow _2 with two degrees of freedom We can see that thep-value for the test is 1333 which tells us that the restrictions are supportedby data at 10 level of signi_cance So we can conclude that the BSE has along term relationship with CPIIPPPIM1 money supply but has no long termrelationship with SP500 and USDINR exchange rate One interpretation of thisresult can be that the Indian stock market represented here by BSE Sensexmoves more in accordance with domestic factors like Industrial production M1money supply Consumer price index and Producer Price index than with globalfactors or in other words as BSE is representation of largest market cap Indiancompanies we can say that the biggest companies in India are ones that are

more dependent on domestic demand rather than exports This result presentsan opportunity for international investors to diversify their portfolio by invest-ing in BSE Sensex as it is decoupled with global markets and macroeconomicfactorsWe use A bivariate Vector Autoregression (BVAR) technique to analyze thedynamic interaction between real asset prices and macro economy VAR ispreferred method to study Macroeconomy and asset prices where variables en-dogenously e_ect each otherWe begin with a bivariate VAR with no restriction Asset prices and instru-ments are allowed to respond to each other freely For paired variables withcointegration relationship VAR is performed at levels whilst for those that arenot cointegrated VAR is performed at _rst di_erence Constant term is ignoredwith loss of generality We use the Bivariate Autoregression analysis for bothimpulse response and Grangers causality testsImpulse response results are displayed in subsection 745 From _rst graph ofimpulse response of BSE to USDINR we can see that USDINR has a negativeimpact on BSE As impulse response is response of BSE to shocks given to US-DINR we can see that a positive shock or unexpected appreciation INR valuewrt USD will have a negative e_ect on BSE for few lags and will disappearafter few lags If we look at the constituents of BSE Index over time we seethat most of the time some of its constituent are companies that thrive on ex-ports Some of the biggest Market-Cap in India are companies in service sectorlike Infosys TCS etc that are hugely dependent on services provided to clientsfrom Europe and US So an appreciation of INR compared to USD makesthese _rms costlier for the global clients and in turn reduces the income of thesecompanies As the _rms revenue pro_t decreases the value of the stock alsodecreases that in turn a_ects the returns of BSE SensexSecond graph (betwen BSE and SP500) shows that increase in SP500 has a pos-itive e_ect on BSE as higher returns of SP500 indicate strong global economywhich in turn results in higher trade between countries The positive responseof BSE to one unit shock to SP500 indicates a spillover e_ect of global factorson Indian economy but the response is weak as can be seen from the graphMoving forward response of BSE to shocks in M1 money supply CPI PPImake economic sense As for M1 money supply one unit shock means increasein M1 money supply This increase in money supply allows companies to bor-row more money from banks at lower rates which they can use for investingEDHEC Business School 225 RESULTSin pro_table projects and generating larger cash ows For Ination indicatorsone unit shock means increase in ination This increase in ination results inhigher costs for the companies that in turn reduces their pro_t margins and asa result value of stocksBy looking at the graphs we can also see that shocks to Indian macroeconomicindicators creates stronger response by BSE as compared to global factors likeSP500 or USDINR This indicates that BSE Index is driven by companies thatdepend hugely on domestic demand rather than exports Response of BSE toshocks to Industrial Production are contradictory to theory In theory an in-crease in industrial production should result in positive response from BSE butour analysis shows the other way A possible reason for this response could bethat industrial production time series is seasonal as can be seen from the graphSo there is a possibility of a leadlag relationship between the two variablesTo test for possibility of leadlag relationship we run a Grangers causality testbetween BSE and IP The result in section 646 shows that at a lag lengthof 4 we can reject the Null hypothesis of BSE does not Granger cause IP at1 signi_cance level This proves that BSE is a leading indicator of industrialproduction and there exist a leadlag relationship between the two indicatorsEDHEC Business School 236 CONCLUSIONS6 Conclusions

In this paper I tested the relations between Indian stock market represented byBSE and domestic and global macro economic factors The research concludesthat the India stock markets are mainly driven by domestic demand and theinuence of global macro factors on the stock market is weak I also tested forGranger causality between BSE and IP and found that BSE is a leading indicatorof Industrial production and can help in predicting the industrial climate inIndiaThe research is insightful for investors and professionals who are looking forinvestment opportunities to diversify their risks As Indian stock markets aremore dependent on domestic factors one can invest in Indian indices and stocksto diversify their risks gained through investing in US and European stocksThe paper opens new doors for research in this _eld One can use variancedecomposition technique to see how much variance of BSE can be explained myvarious domestic and global macro factors Also one can use di_erent globalfactors like sovereign CDS spreads T-Bill rates a composite indicator of globaleconomy for further research on interaction between Indian stock market andglobal economyOne can also research on how various global macroeconomicnews a_ects India stock markets and for how long the e_ects persistsEDHEC Business School 247 GRAPHS AND TABLES7 Graphs and Tables71 Graphs of Time seriesEDHEC Business School 257 GRAPHS AND TABLESEDHEC Business School 267 GRAPHS AND TABLESEDHEC Business School 277 GRAPHS AND TABLESEDHEC Business School 287 GRAPHS AND TABLES72 Graphs of Time Series - Di_erencedEDHEC Business School 297 GRAPHS AND TABLESEDHEC Business School 307 GRAPHS AND TABLESEDHEC Business School 317 GRAPHS AND TABLESEDHEC Business School 327 GRAPHS AND TABLES73 Correlograms of Time seriesBSEEDHEC Business School 337 GRAPHS AND TABLESIPEDHEC Business School 347 GRAPHS AND TABLESSP500EDHEC Business School 357 GRAPHS AND TABLESUSDINREDHEC Business School 367 GRAPHS AND TABLESCPIEDHEC Business School 377 GRAPHS AND TABLESPPIEDHEC Business School 387 GRAPHS AND TABLES

M1EDHEC Business School 397 GRAPHS AND TABLES74 Tables741 Table for Unit root test of Time seriesVariables T-Stat p-valueBSE -2671 2495 SP500 -1315 8818 CPI -1909 6466 IP -1669 899 M1 -2420 3679 PPI -3353 601 USDINR -2955 1469 742 Tables for Unit root test of Di_erenced time seriesVariables T-Stat p-valueBSE -13848 000 SP500 -14832 000 CPI -3344 140 IP -3865 027 M1 -3867 026 PPI -9656 000 USDINR -13701 000 743 Tables for Residual based test of cointegrationTable 1BSE - CPIt-Statistic ProbADF test statistic -2622676 087Test critical values 1 level -25738185 level -19420410 level -1615891Table 2BSE - IPt-Statistic ProbADF test statistic -3738802 002Test critical values 1 level -25745135 level -194213610 level -1615828EDHEC Business School 407 GRAPHS AND TABLESTable 3BSE - M1t-Statistic ProbADF test statistic -2875518 041Test critical values 1 level -25737845 level -194203510 level -1615894Table 4BSE - PPIt-Statistic ProbADF test statistic -2399055 162Test critical values 1 level -25737845 level -194203510 level -1615894Table 5BSE - SP500t-Statistic ProbADF test statistic -1427184 1430Test critical values 1 level -25737845 level -1942035

10 level -1615894EDHEC Business School 417 GRAPHS AND TABLESTable 6BSE - USDINRt-Statistic ProbADF test statistic -1659522 917Test critical values 1 level -25738185 level -19420410 level -1615891EDHEC Business School 427 GRAPHS AND TABLES744 Johansen cointegration testEDHEC Business School 437 GRAPHS AND TABLESTable 2EDHEC Business School 447 GRAPHS AND TABLESTable 3EDHEC Business School 457 GRAPHS AND TABLES745 Impulse response testsEDHEC Business School 467 GRAPHS AND TABLESEDHEC Business School 477 GRAPHS AND TABLESEDHEC Business School 487 GRAPHS AND TABLES746 Granger causality test between IP and BSEEDHEC Business School 498 BIBLIOGRAPHY8 BibliographyEugene F Fama Ination Output and Money Journal of Business 1982Eugene F Fama Stock Returns Real activity and Money The American Eco-nomic Review 1981Eugene F Fama Stock Returns Expected Returns and Real activity Journal ofFinance 1990Pal and Mittal Impact of macroeconomic indicators in Indian capital marketsJournal of Risk Finance 2011Shahid Ahmed Aggregate Economic Variables and Stock Markets in India In-ternational Research Journal of Finance and Economics 2008Sahu and Dhiman Correlation and Causality between Stock Market and MacroEconomic Variables in India An Empirical Study 2010 International Confer-ence on E-Business and Economics 2011Mohammad Bayezid Ali Impact of Micro Variables on Emerging Stock MarketReturn A case on Dhaka Stock Exchange (DSE) Interdisciplinary Journal ofResearch in Business 2011Napphon Tangjitprom Macroeconomic Factors of Emerging Stock Market Theevidence from Thailand International Journal of Finance and Research 2012Sayed Mehdi Hosseini The Role of Macroeconomic Variables on Stock MarketIndex in China and India International Journal of Economics and Finance2011John Y Campbell Pitfalls and Opportunities What Macroeconomists shouldknow about Unit Roots NBER Working Papers 1991Hacker and Hatemi The properties of Procedures Dealing with Uncertainityabout Intercept and Deterministic Trend in Unit Root Testing CESIS Elec-tronic Working Papers 2010Elder and Kennedy Testing for Unit Roots What should Students be TaughtNasseh and Strauss Stock Prices and domestic and international macroeco-

nomic activity a cointegration approach The Quarterly Review of Economicsand Finance 2000Engle and Granger Co-Integration and Error Correction Representation Es-timation and Testing Econometrica 1987Eugene F Fama Stock Returns Real Activity Ination and Money 1981American Economic AssociationNaliniprave Tripathy Causal Relationship between Macro-Economic Indicatorsand Stock Market in India Asian Journal of Finance and Accounting 2011Rogalski and Vinso Stock Returns Money Supply and the Direction of Causal-ity The Journal of Finance 1977James et al A VARMA Analysis of the Causal Relations Among Stock Re-turns Real output and Nominal Interest Rates 1985 The Journal of FinanceBailey and Chung Risk and return in the Philippine Equity market A multi-factor exploration Paci_c-Basin Finance Journal 1996Nai-Fu Chen Financial Investment opportunities and the Macroeconomy TheJournal of Finance 1991GB Wickremasinghe Macroeconomic forces and stock prices Some empiricalevidence from an emerging stock markets University of Wollongong 2006EDHEC Business School 508 BIBLIOGRAPHYYao Juo and Loh On Chinas Monetary Policy and Asset Prices University ofNottingham- China policy Institute 2011Bilson et al Selecting macroeconomic variables as explanatory factors of emerg-ing stock market returns Paci_c-Basin Finance Journal 2001CHen Roll and Ross Economic forces and the Stock Markets The Journal ofBusiness 1986William H Greene Econometric Analysis 6th Edition Pearson InternationalEditionRuey Tsay Analysis of Financial Time seriesChris Brooks Introductory Econometrics for Finance Cambridge PublicationsEDHEC Business School 51

For a better test for stationarity we use augmented Dickey fuller Unit roottest on each time series separately Augmented Dickey Fuller test is test ofnull hypothesis that the time series contains a unit roots against a alternativehypothesis that the series is stationary421 Mathematical representation of Stationary series and unit roottestAssume a variable Y whose structure can be given by AR process with no driftequationyt = _1yt10485761 + _2yt10485762 + _3yt10485763 + + _nyt1048576n + ut (2)where ut is the residual at time t Using a Lag operator L we can write eq(1)asyt = _1L1yt + _2L2yt + _3L3yt + + _nLnyt + ut (3)EDHEC Business School 164 METHODOLOGYRearranging eqn (2) we getyt 1048576 _1L1yt 1048576 _2L2yt 1048576 _3L3yt + 1048576 _nLnyt = ut (4)yt(1 1048576 _1L1 1048576 _2L2 1048576 _3L3 + 1048576 _nLn) = ut (5)or_(L)yt = ut (6)The time series is stationary if we can write eqn(5) in formyt = _(L)10485761ut (7)with _(L)10485761 converging to zero It means the autocorrelation function woulddecline as lag length is increased If eqn (6) is expanded to a MA(1) processthe coe_cients of residuals should decrease such that the the residuals that thee_ect of residuals decrease with increase in lags SO if the process is stationarythe coe_cients of residuals will converge to zero and for non-stationary seriesthey will and converge to zero and will have long term e_ect The condition fortesting of unit root for an AR process is that the roots of eqn(6) or Charac-teristic equation should lie outside unit circle422 Augmented Dickey Fuller Unit Root TestConsider an AR(1) process of variable Yyt = _yt10485761 + ut (8)Subtracting yt10485761 from both sides of eqn(7) we get_y = (_ 1048576 1)yt10485761 + ut (9)Eqn(8) is the test equation for Dickey Fuller test For Dickey-Fuller Unit roottestNull Hypothesis The value of _ is equal to 1 or value of _10485761 is equal to 0 vsAlternate Hypothesis The value of _ is less than one or value of _ 1048576 1 is lessthan zero Augmented Dickey-Fuller test is similar to normal Dickey-Fuller testsexcept it takes the lag structure of more than one into account_y = yt10485761 +Xpi=1_i_yt1048576i + ut (10)If the series has one or more unit root it is said to be integrated of order nwhere n is the number of unit roots of the characteristic equation To makethese time series stationary they needs to be di_erenced Mathematically ifyt _ I (n) (11)then_(d) yt _ I (0) (12)To make our time-series stationary we will use the natural log returns of theseseries in the analysisEDHEC Business School 174 METHODOLOGY43 Testing Long Term RelationshipsEngle and Granger (1987) in their seminal paper described cointegration whichforms the basis for testing for long term relationship between variables Accord-ing to Engle and Granger two variables are cointegrated if they are integratedprocess in their natural form (of the same order) but a weighted combination

of the variables can be found such that the combined new variable is integratedof order less than the order of individual time series Mathematically assumeyt to be a k X 1 vector of variables then the components are cointegrated orintegrated of order (db) if1 All components of yt are I(d)2 There is at least one vector of coe_cients _ such that_0

yt _ I (d 1048576 b) (13)As most of the _nancial time series are integrated of order one we will restrictourselves to case d=b=1 Two or more variables are said to be cointegrated ifthere exist a linear combination of these variables that is stationary Many ofthe series are non-stationary but move together over time which implies twoseries are bound by some common force or factor in long run We will test forcointegration by a residual-based approach and Johansens VAR methodResidual Based approach Consider a modelyt = _1 + _2x2t + _3x3t + + ut (14)where yt x2t x3t are all integrated of order N Now if the residual of this re-gression ut is stationary then we can say that the variables are cointegrated elsethere exist no long term relationship between the variables To test the resid-ual for stationarity we will run Augmented Dickey-Fuller tests on the residualsUnder the Null hypothesis the residual are integrated of order one or more andunder alternate hypothesis the residuals are I(0)431 Johansen test for CointegrationJohansen test for cointegration presents a better model for testing multiplecointegration among multiple variables The Residual based approach can only_nd atmost one cointegration and can be tested for a model with two variablesEven if more than two variables are present in the equation that are cointegratedthe Residual based approach will give only one cointegration SO we will useJhoansen VAR based cointegration for testing more than one cointegrationSuppose that a set of g variables are under consideration that are I(1) andwhich are thought to be cointegrated A VAR with k lags containing thesevariables could be set upyt = _1yt10485761 + _2yt10485762 + _ _ _ + _kyt1048576k + ut (15)g _ 1 g _ g g _ 1 g _ g g _ 1 g _ g g _ 1 g _ 1EDHEC Business School 184 METHODOLOGYIn order to use the Johansen test the VAR above should be turned into avector error correction model of form_yt = _yt1048576k + 1_yt10485761 + 2_yt10485762 + _ _ _ + k10485761_yt1048576(k10485761) + ut (16)where _ = (_ki=1_i) 1048576 Ig and i = (_ij=1_j) 1048576 IgThe Johansens test centers around testing the _ matrix which is the matrixthat represents the long term cointegration between the variables The test fornumber of cointegration is calculated by looking at the rank of the _ matrixthrough its eigenvalues The rank of the matrix is equal to number of roots(eigenvalues) _i of the matrix that are di_erent from zero The roots should beless than 1 in absolute value and positive If the variables are not cointegratedthe rank of the matrix will not be signi_cantly di_erent from zero ie _i _ 0There are two test statistics for Johansen test _tracer and _max_trace (r) = 1048576TPgi=r+1 ln(1 1048576 _ _i)and_max(r r + 1) = 1048576Tln(1 1048576 _r_+1)_trace is a test statistic for joint test where the null hypothesis is that thenumber of cointegration vector is less than or equal to r against an alternativethat there are more than r_max conducts another separate test on eigenvalues and has null hypothesis that

the number of cointegrating vector is r against r+144 Impulse ResponseOnce we have determined whether the variables have long term relationship ornot we can form a multivariate VAR model for the variables A multivariateVAR model between g variables is a model where the current value of a variabledepend on di_ernt combinations of the previous k values of all the variables anderror terms A general representation of the model can beyBSEt = _ + _BSEyBSE + _IP yIP + CPIyCPI + _M1yM1 + _SP500ySP500 + u1t(17)where all the coe_cients except _ are g _ k matrices and all variables y are k_ 1 matricesOnce we have formed a model like this we can use the model for Impulse re-sponse A VAR(p) model can be written as a linear fuction of the past innova-tions that isrt = _ + at + 1at10485761 + 2at10485762 + (18)where _ = [_(1)]10485761_0 provided that the inverse exists and the coe_cient ma-trices i can be obtained by equating the coe_cients of Bi in the equation(I 1048576 _1B 1048576 1048576 _PBP )(I + 1B + 2B2 + ) = I (19)EDHEC Business School 194 METHODOLOGYwhere I is the Identity martix This is a moving average representation of rtwith the coe_cient matrix i being the impact of the past innovation at1048576i onrt Equivalently i is the e_ect of at on the future observation rt+i Therefore i is often referred to as the Impulse Response Function of rt For our impulseresponse we will use equation of variables in _rst di_ernce form like_BSEt = _t +Xki=0_11(i)_BSEt1048576i +Xkj=1_12(j)_MIt1048576j + _BSEt (20)_MIt = _t +Xki=0_21(i)_MIt1048576i +Xkj=1_22(j)_BSEt1048576j + _MIt (21)Grangers causality and Blocks F test of a VAR model will suggest which ofthe variables have statistically signi_cant impacts on the future values of othervariables in the system But F-test results cannot explain the sign of the re-lationship nor how long these e_ects require to take place Such informationwill however be given by an examination of the VARs impulse responses andvariance decompositions Impulse response is a technique that trace out theresponsiveness of the dependent variable in the VAR to shocks of each of theother variables So for each variable from each equation separately we will applya unit shock to the error and trace the e_ects upon the VAR system over timeBy using the impulse response technique we can determine how responsive isthe BSE stock index to Indian macro indicators and SP500 This will help usdetermine whether the BSE index is more reactive to domestic news or globalnewsEDHEC Business School 205 RESULTS5 ResultsBefore we use the time series for VAR analysis or cointegration tests we need todetermine whether the series are Stationary or not If the series are stationaryin levels we can use them directly else we need to use the di_erenced time seriesOne way to look for autocorrelation or integrated process is to see the graphsof the various time series used Section 71 shows the graphs of variables we

use for our analysis As we can see from the graphs all of the time series havea trend in long run which points to an integrated process As a second stepwe plot the graphs of di_erenced time series in Section 52 We can see thatthe di_erenced graphs in Section 72 dont show a long term trend and crossthe X-axis frequently This is usually a property of I(1) processes So we checkthe series for autocorrelations at di_erent lag lengths Section 73 shows cor-relograms graph autocorrelation coe_cient partial autocorrelation coe_cientQ-Stat and p-value for various time series up to 36 lags As can be seen in thetables the Q-stat for all lags is zero and we can reject the joint null hypothesisthat all the autocorrelations up to 36 lags are zero Table 741 shows that ifwe conduct a Unit root test on levels of the series we _nd that all the 7 seriesare integrated as we cannot reject the t-stat for unit root at 1 level But ifwe conduct the same test on di_erenced values of the series we _nd that we canreject the null hypothesis of unit root at 1 signi_cance level for all the seriesexcept CPI This tells us that all the series are I(1) as there _rst di_erence seriesare I(0)As our series are I(1) we will work with index levels of time series to determineif there exist one or more cointegrating relationships between the series Tablesin subsection 743 are based on residual approach where we run a regression ofBSE and various macroeconomic indicators and test the residuals for unit rootusing Augmented Dickey-Fuller test As we assume the two series are cointe-grated we conduct the test with no trend and intercept If the two series arecointegrated then the errors should not have any trend or intercept We see thatwe can reject the null hypothesis of unit root at 1 signi_cance for CPIIP M1We can reject the null of unit root for PPI at 5 and for SP500 and USDINRwe cant reject the null hypothesis of unit root at even 5 level This pointsto the fact that BSE has a strong long term relationship with IP M1 moneysupply CPI at 1 level with IP M1 CPI PPI at 5 signi_cance level AlsoBSE has no long term relationship with SP500 and USD INR exchange rateTo test for multiple cointegrating relationship we now employ a Johansen VARbased cointegration test The results of the test are displayed in subsection744 The _rst panel of the test results displays the value of _trace and_maxof Johansen test with di_erent assumptions about intercept and trend We cansee from this panel that when we consider a functional form of intercept and noTrend we have atleast and atmost three cointegrating relationships The secondpanel of the results display the value of information criteria for lag lengths Formost of the models we see that Akalike criteria points to a lag of three andSchwarz criteria points to a lag of one To estimate the cointegrating model wechoose the model with intercept and no trend and run a cointegration testTestresults are shown in Table 2 of subsection 744 At 5 signi_cance level wecan reject the null of atmost two cointegrating factors for _trace and same for_max Now to test which all variables have a long tern relationship we perform aRestricted cointegration with vector error correction model As we had alreadyEDHEC Business School 215 RESULTSseen in our residual based test of cointegration that BSE has no cointegratingrelationship with SP500 and USDINR we create a restricted cointegration modelwhere we set coe_cients of SP500 and USDINR as zero The test results aredisplayed in Table 3 of subsection 744 In this case as there are two restrictionsthe test statistic follow _2 with two degrees of freedom We can see that thep-value for the test is 1333 which tells us that the restrictions are supportedby data at 10 level of signi_cance So we can conclude that the BSE has along term relationship with CPIIPPPIM1 money supply but has no long termrelationship with SP500 and USDINR exchange rate One interpretation of thisresult can be that the Indian stock market represented here by BSE Sensexmoves more in accordance with domestic factors like Industrial production M1money supply Consumer price index and Producer Price index than with globalfactors or in other words as BSE is representation of largest market cap Indiancompanies we can say that the biggest companies in India are ones that are

more dependent on domestic demand rather than exports This result presentsan opportunity for international investors to diversify their portfolio by invest-ing in BSE Sensex as it is decoupled with global markets and macroeconomicfactorsWe use A bivariate Vector Autoregression (BVAR) technique to analyze thedynamic interaction between real asset prices and macro economy VAR ispreferred method to study Macroeconomy and asset prices where variables en-dogenously e_ect each otherWe begin with a bivariate VAR with no restriction Asset prices and instru-ments are allowed to respond to each other freely For paired variables withcointegration relationship VAR is performed at levels whilst for those that arenot cointegrated VAR is performed at _rst di_erence Constant term is ignoredwith loss of generality We use the Bivariate Autoregression analysis for bothimpulse response and Grangers causality testsImpulse response results are displayed in subsection 745 From _rst graph ofimpulse response of BSE to USDINR we can see that USDINR has a negativeimpact on BSE As impulse response is response of BSE to shocks given to US-DINR we can see that a positive shock or unexpected appreciation INR valuewrt USD will have a negative e_ect on BSE for few lags and will disappearafter few lags If we look at the constituents of BSE Index over time we seethat most of the time some of its constituent are companies that thrive on ex-ports Some of the biggest Market-Cap in India are companies in service sectorlike Infosys TCS etc that are hugely dependent on services provided to clientsfrom Europe and US So an appreciation of INR compared to USD makesthese _rms costlier for the global clients and in turn reduces the income of thesecompanies As the _rms revenue pro_t decreases the value of the stock alsodecreases that in turn a_ects the returns of BSE SensexSecond graph (betwen BSE and SP500) shows that increase in SP500 has a pos-itive e_ect on BSE as higher returns of SP500 indicate strong global economywhich in turn results in higher trade between countries The positive responseof BSE to one unit shock to SP500 indicates a spillover e_ect of global factorson Indian economy but the response is weak as can be seen from the graphMoving forward response of BSE to shocks in M1 money supply CPI PPImake economic sense As for M1 money supply one unit shock means increasein M1 money supply This increase in money supply allows companies to bor-row more money from banks at lower rates which they can use for investingEDHEC Business School 225 RESULTSin pro_table projects and generating larger cash ows For Ination indicatorsone unit shock means increase in ination This increase in ination results inhigher costs for the companies that in turn reduces their pro_t margins and asa result value of stocksBy looking at the graphs we can also see that shocks to Indian macroeconomicindicators creates stronger response by BSE as compared to global factors likeSP500 or USDINR This indicates that BSE Index is driven by companies thatdepend hugely on domestic demand rather than exports Response of BSE toshocks to Industrial Production are contradictory to theory In theory an in-crease in industrial production should result in positive response from BSE butour analysis shows the other way A possible reason for this response could bethat industrial production time series is seasonal as can be seen from the graphSo there is a possibility of a leadlag relationship between the two variablesTo test for possibility of leadlag relationship we run a Grangers causality testbetween BSE and IP The result in section 646 shows that at a lag lengthof 4 we can reject the Null hypothesis of BSE does not Granger cause IP at1 signi_cance level This proves that BSE is a leading indicator of industrialproduction and there exist a leadlag relationship between the two indicatorsEDHEC Business School 236 CONCLUSIONS6 Conclusions

In this paper I tested the relations between Indian stock market represented byBSE and domestic and global macro economic factors The research concludesthat the India stock markets are mainly driven by domestic demand and theinuence of global macro factors on the stock market is weak I also tested forGranger causality between BSE and IP and found that BSE is a leading indicatorof Industrial production and can help in predicting the industrial climate inIndiaThe research is insightful for investors and professionals who are looking forinvestment opportunities to diversify their risks As Indian stock markets aremore dependent on domestic factors one can invest in Indian indices and stocksto diversify their risks gained through investing in US and European stocksThe paper opens new doors for research in this _eld One can use variancedecomposition technique to see how much variance of BSE can be explained myvarious domestic and global macro factors Also one can use di_erent globalfactors like sovereign CDS spreads T-Bill rates a composite indicator of globaleconomy for further research on interaction between Indian stock market andglobal economyOne can also research on how various global macroeconomicnews a_ects India stock markets and for how long the e_ects persistsEDHEC Business School 247 GRAPHS AND TABLES7 Graphs and Tables71 Graphs of Time seriesEDHEC Business School 257 GRAPHS AND TABLESEDHEC Business School 267 GRAPHS AND TABLESEDHEC Business School 277 GRAPHS AND TABLESEDHEC Business School 287 GRAPHS AND TABLES72 Graphs of Time Series - Di_erencedEDHEC Business School 297 GRAPHS AND TABLESEDHEC Business School 307 GRAPHS AND TABLESEDHEC Business School 317 GRAPHS AND TABLESEDHEC Business School 327 GRAPHS AND TABLES73 Correlograms of Time seriesBSEEDHEC Business School 337 GRAPHS AND TABLESIPEDHEC Business School 347 GRAPHS AND TABLESSP500EDHEC Business School 357 GRAPHS AND TABLESUSDINREDHEC Business School 367 GRAPHS AND TABLESCPIEDHEC Business School 377 GRAPHS AND TABLESPPIEDHEC Business School 387 GRAPHS AND TABLES

M1EDHEC Business School 397 GRAPHS AND TABLES74 Tables741 Table for Unit root test of Time seriesVariables T-Stat p-valueBSE -2671 2495 SP500 -1315 8818 CPI -1909 6466 IP -1669 899 M1 -2420 3679 PPI -3353 601 USDINR -2955 1469 742 Tables for Unit root test of Di_erenced time seriesVariables T-Stat p-valueBSE -13848 000 SP500 -14832 000 CPI -3344 140 IP -3865 027 M1 -3867 026 PPI -9656 000 USDINR -13701 000 743 Tables for Residual based test of cointegrationTable 1BSE - CPIt-Statistic ProbADF test statistic -2622676 087Test critical values 1 level -25738185 level -19420410 level -1615891Table 2BSE - IPt-Statistic ProbADF test statistic -3738802 002Test critical values 1 level -25745135 level -194213610 level -1615828EDHEC Business School 407 GRAPHS AND TABLESTable 3BSE - M1t-Statistic ProbADF test statistic -2875518 041Test critical values 1 level -25737845 level -194203510 level -1615894Table 4BSE - PPIt-Statistic ProbADF test statistic -2399055 162Test critical values 1 level -25737845 level -194203510 level -1615894Table 5BSE - SP500t-Statistic ProbADF test statistic -1427184 1430Test critical values 1 level -25737845 level -1942035

10 level -1615894EDHEC Business School 417 GRAPHS AND TABLESTable 6BSE - USDINRt-Statistic ProbADF test statistic -1659522 917Test critical values 1 level -25738185 level -19420410 level -1615891EDHEC Business School 427 GRAPHS AND TABLES744 Johansen cointegration testEDHEC Business School 437 GRAPHS AND TABLESTable 2EDHEC Business School 447 GRAPHS AND TABLESTable 3EDHEC Business School 457 GRAPHS AND TABLES745 Impulse response testsEDHEC Business School 467 GRAPHS AND TABLESEDHEC Business School 477 GRAPHS AND TABLESEDHEC Business School 487 GRAPHS AND TABLES746 Granger causality test between IP and BSEEDHEC Business School 498 BIBLIOGRAPHY8 BibliographyEugene F Fama Ination Output and Money Journal of Business 1982Eugene F Fama Stock Returns Real activity and Money The American Eco-nomic Review 1981Eugene F Fama Stock Returns Expected Returns and Real activity Journal ofFinance 1990Pal and Mittal Impact of macroeconomic indicators in Indian capital marketsJournal of Risk Finance 2011Shahid Ahmed Aggregate Economic Variables and Stock Markets in India In-ternational Research Journal of Finance and Economics 2008Sahu and Dhiman Correlation and Causality between Stock Market and MacroEconomic Variables in India An Empirical Study 2010 International Confer-ence on E-Business and Economics 2011Mohammad Bayezid Ali Impact of Micro Variables on Emerging Stock MarketReturn A case on Dhaka Stock Exchange (DSE) Interdisciplinary Journal ofResearch in Business 2011Napphon Tangjitprom Macroeconomic Factors of Emerging Stock Market Theevidence from Thailand International Journal of Finance and Research 2012Sayed Mehdi Hosseini The Role of Macroeconomic Variables on Stock MarketIndex in China and India International Journal of Economics and Finance2011John Y Campbell Pitfalls and Opportunities What Macroeconomists shouldknow about Unit Roots NBER Working Papers 1991Hacker and Hatemi The properties of Procedures Dealing with Uncertainityabout Intercept and Deterministic Trend in Unit Root Testing CESIS Elec-tronic Working Papers 2010Elder and Kennedy Testing for Unit Roots What should Students be TaughtNasseh and Strauss Stock Prices and domestic and international macroeco-

nomic activity a cointegration approach The Quarterly Review of Economicsand Finance 2000Engle and Granger Co-Integration and Error Correction Representation Es-timation and Testing Econometrica 1987Eugene F Fama Stock Returns Real Activity Ination and Money 1981American Economic AssociationNaliniprave Tripathy Causal Relationship between Macro-Economic Indicatorsand Stock Market in India Asian Journal of Finance and Accounting 2011Rogalski and Vinso Stock Returns Money Supply and the Direction of Causal-ity The Journal of Finance 1977James et al A VARMA Analysis of the Causal Relations Among Stock Re-turns Real output and Nominal Interest Rates 1985 The Journal of FinanceBailey and Chung Risk and return in the Philippine Equity market A multi-factor exploration Paci_c-Basin Finance Journal 1996Nai-Fu Chen Financial Investment opportunities and the Macroeconomy TheJournal of Finance 1991GB Wickremasinghe Macroeconomic forces and stock prices Some empiricalevidence from an emerging stock markets University of Wollongong 2006EDHEC Business School 508 BIBLIOGRAPHYYao Juo and Loh On Chinas Monetary Policy and Asset Prices University ofNottingham- China policy Institute 2011Bilson et al Selecting macroeconomic variables as explanatory factors of emerg-ing stock market returns Paci_c-Basin Finance Journal 2001CHen Roll and Ross Economic forces and the Stock Markets The Journal ofBusiness 1986William H Greene Econometric Analysis 6th Edition Pearson InternationalEditionRuey Tsay Analysis of Financial Time seriesChris Brooks Introductory Econometrics for Finance Cambridge PublicationsEDHEC Business School 51

of the variables can be found such that the combined new variable is integratedof order less than the order of individual time series Mathematically assumeyt to be a k X 1 vector of variables then the components are cointegrated orintegrated of order (db) if1 All components of yt are I(d)2 There is at least one vector of coe_cients _ such that_0

yt _ I (d 1048576 b) (13)As most of the _nancial time series are integrated of order one we will restrictourselves to case d=b=1 Two or more variables are said to be cointegrated ifthere exist a linear combination of these variables that is stationary Many ofthe series are non-stationary but move together over time which implies twoseries are bound by some common force or factor in long run We will test forcointegration by a residual-based approach and Johansens VAR methodResidual Based approach Consider a modelyt = _1 + _2x2t + _3x3t + + ut (14)where yt x2t x3t are all integrated of order N Now if the residual of this re-gression ut is stationary then we can say that the variables are cointegrated elsethere exist no long term relationship between the variables To test the resid-ual for stationarity we will run Augmented Dickey-Fuller tests on the residualsUnder the Null hypothesis the residual are integrated of order one or more andunder alternate hypothesis the residuals are I(0)431 Johansen test for CointegrationJohansen test for cointegration presents a better model for testing multiplecointegration among multiple variables The Residual based approach can only_nd atmost one cointegration and can be tested for a model with two variablesEven if more than two variables are present in the equation that are cointegratedthe Residual based approach will give only one cointegration SO we will useJhoansen VAR based cointegration for testing more than one cointegrationSuppose that a set of g variables are under consideration that are I(1) andwhich are thought to be cointegrated A VAR with k lags containing thesevariables could be set upyt = _1yt10485761 + _2yt10485762 + _ _ _ + _kyt1048576k + ut (15)g _ 1 g _ g g _ 1 g _ g g _ 1 g _ g g _ 1 g _ 1EDHEC Business School 184 METHODOLOGYIn order to use the Johansen test the VAR above should be turned into avector error correction model of form_yt = _yt1048576k + 1_yt10485761 + 2_yt10485762 + _ _ _ + k10485761_yt1048576(k10485761) + ut (16)where _ = (_ki=1_i) 1048576 Ig and i = (_ij=1_j) 1048576 IgThe Johansens test centers around testing the _ matrix which is the matrixthat represents the long term cointegration between the variables The test fornumber of cointegration is calculated by looking at the rank of the _ matrixthrough its eigenvalues The rank of the matrix is equal to number of roots(eigenvalues) _i of the matrix that are di_erent from zero The roots should beless than 1 in absolute value and positive If the variables are not cointegratedthe rank of the matrix will not be signi_cantly di_erent from zero ie _i _ 0There are two test statistics for Johansen test _tracer and _max_trace (r) = 1048576TPgi=r+1 ln(1 1048576 _ _i)and_max(r r + 1) = 1048576Tln(1 1048576 _r_+1)_trace is a test statistic for joint test where the null hypothesis is that thenumber of cointegration vector is less than or equal to r against an alternativethat there are more than r_max conducts another separate test on eigenvalues and has null hypothesis that

the number of cointegrating vector is r against r+144 Impulse ResponseOnce we have determined whether the variables have long term relationship ornot we can form a multivariate VAR model for the variables A multivariateVAR model between g variables is a model where the current value of a variabledepend on di_ernt combinations of the previous k values of all the variables anderror terms A general representation of the model can beyBSEt = _ + _BSEyBSE + _IP yIP + CPIyCPI + _M1yM1 + _SP500ySP500 + u1t(17)where all the coe_cients except _ are g _ k matrices and all variables y are k_ 1 matricesOnce we have formed a model like this we can use the model for Impulse re-sponse A VAR(p) model can be written as a linear fuction of the past innova-tions that isrt = _ + at + 1at10485761 + 2at10485762 + (18)where _ = [_(1)]10485761_0 provided that the inverse exists and the coe_cient ma-trices i can be obtained by equating the coe_cients of Bi in the equation(I 1048576 _1B 1048576 1048576 _PBP )(I + 1B + 2B2 + ) = I (19)EDHEC Business School 194 METHODOLOGYwhere I is the Identity martix This is a moving average representation of rtwith the coe_cient matrix i being the impact of the past innovation at1048576i onrt Equivalently i is the e_ect of at on the future observation rt+i Therefore i is often referred to as the Impulse Response Function of rt For our impulseresponse we will use equation of variables in _rst di_ernce form like_BSEt = _t +Xki=0_11(i)_BSEt1048576i +Xkj=1_12(j)_MIt1048576j + _BSEt (20)_MIt = _t +Xki=0_21(i)_MIt1048576i +Xkj=1_22(j)_BSEt1048576j + _MIt (21)Grangers causality and Blocks F test of a VAR model will suggest which ofthe variables have statistically signi_cant impacts on the future values of othervariables in the system But F-test results cannot explain the sign of the re-lationship nor how long these e_ects require to take place Such informationwill however be given by an examination of the VARs impulse responses andvariance decompositions Impulse response is a technique that trace out theresponsiveness of the dependent variable in the VAR to shocks of each of theother variables So for each variable from each equation separately we will applya unit shock to the error and trace the e_ects upon the VAR system over timeBy using the impulse response technique we can determine how responsive isthe BSE stock index to Indian macro indicators and SP500 This will help usdetermine whether the BSE index is more reactive to domestic news or globalnewsEDHEC Business School 205 RESULTS5 ResultsBefore we use the time series for VAR analysis or cointegration tests we need todetermine whether the series are Stationary or not If the series are stationaryin levels we can use them directly else we need to use the di_erenced time seriesOne way to look for autocorrelation or integrated process is to see the graphsof the various time series used Section 71 shows the graphs of variables we

use for our analysis As we can see from the graphs all of the time series havea trend in long run which points to an integrated process As a second stepwe plot the graphs of di_erenced time series in Section 52 We can see thatthe di_erenced graphs in Section 72 dont show a long term trend and crossthe X-axis frequently This is usually a property of I(1) processes So we checkthe series for autocorrelations at di_erent lag lengths Section 73 shows cor-relograms graph autocorrelation coe_cient partial autocorrelation coe_cientQ-Stat and p-value for various time series up to 36 lags As can be seen in thetables the Q-stat for all lags is zero and we can reject the joint null hypothesisthat all the autocorrelations up to 36 lags are zero Table 741 shows that ifwe conduct a Unit root test on levels of the series we _nd that all the 7 seriesare integrated as we cannot reject the t-stat for unit root at 1 level But ifwe conduct the same test on di_erenced values of the series we _nd that we canreject the null hypothesis of unit root at 1 signi_cance level for all the seriesexcept CPI This tells us that all the series are I(1) as there _rst di_erence seriesare I(0)As our series are I(1) we will work with index levels of time series to determineif there exist one or more cointegrating relationships between the series Tablesin subsection 743 are based on residual approach where we run a regression ofBSE and various macroeconomic indicators and test the residuals for unit rootusing Augmented Dickey-Fuller test As we assume the two series are cointe-grated we conduct the test with no trend and intercept If the two series arecointegrated then the errors should not have any trend or intercept We see thatwe can reject the null hypothesis of unit root at 1 signi_cance for CPIIP M1We can reject the null of unit root for PPI at 5 and for SP500 and USDINRwe cant reject the null hypothesis of unit root at even 5 level This pointsto the fact that BSE has a strong long term relationship with IP M1 moneysupply CPI at 1 level with IP M1 CPI PPI at 5 signi_cance level AlsoBSE has no long term relationship with SP500 and USD INR exchange rateTo test for multiple cointegrating relationship we now employ a Johansen VARbased cointegration test The results of the test are displayed in subsection744 The _rst panel of the test results displays the value of _trace and_maxof Johansen test with di_erent assumptions about intercept and trend We cansee from this panel that when we consider a functional form of intercept and noTrend we have atleast and atmost three cointegrating relationships The secondpanel of the results display the value of information criteria for lag lengths Formost of the models we see that Akalike criteria points to a lag of three andSchwarz criteria points to a lag of one To estimate the cointegrating model wechoose the model with intercept and no trend and run a cointegration testTestresults are shown in Table 2 of subsection 744 At 5 signi_cance level wecan reject the null of atmost two cointegrating factors for _trace and same for_max Now to test which all variables have a long tern relationship we perform aRestricted cointegration with vector error correction model As we had alreadyEDHEC Business School 215 RESULTSseen in our residual based test of cointegration that BSE has no cointegratingrelationship with SP500 and USDINR we create a restricted cointegration modelwhere we set coe_cients of SP500 and USDINR as zero The test results aredisplayed in Table 3 of subsection 744 In this case as there are two restrictionsthe test statistic follow _2 with two degrees of freedom We can see that thep-value for the test is 1333 which tells us that the restrictions are supportedby data at 10 level of signi_cance So we can conclude that the BSE has along term relationship with CPIIPPPIM1 money supply but has no long termrelationship with SP500 and USDINR exchange rate One interpretation of thisresult can be that the Indian stock market represented here by BSE Sensexmoves more in accordance with domestic factors like Industrial production M1money supply Consumer price index and Producer Price index than with globalfactors or in other words as BSE is representation of largest market cap Indiancompanies we can say that the biggest companies in India are ones that are

more dependent on domestic demand rather than exports This result presentsan opportunity for international investors to diversify their portfolio by invest-ing in BSE Sensex as it is decoupled with global markets and macroeconomicfactorsWe use A bivariate Vector Autoregression (BVAR) technique to analyze thedynamic interaction between real asset prices and macro economy VAR ispreferred method to study Macroeconomy and asset prices where variables en-dogenously e_ect each otherWe begin with a bivariate VAR with no restriction Asset prices and instru-ments are allowed to respond to each other freely For paired variables withcointegration relationship VAR is performed at levels whilst for those that arenot cointegrated VAR is performed at _rst di_erence Constant term is ignoredwith loss of generality We use the Bivariate Autoregression analysis for bothimpulse response and Grangers causality testsImpulse response results are displayed in subsection 745 From _rst graph ofimpulse response of BSE to USDINR we can see that USDINR has a negativeimpact on BSE As impulse response is response of BSE to shocks given to US-DINR we can see that a positive shock or unexpected appreciation INR valuewrt USD will have a negative e_ect on BSE for few lags and will disappearafter few lags If we look at the constituents of BSE Index over time we seethat most of the time some of its constituent are companies that thrive on ex-ports Some of the biggest Market-Cap in India are companies in service sectorlike Infosys TCS etc that are hugely dependent on services provided to clientsfrom Europe and US So an appreciation of INR compared to USD makesthese _rms costlier for the global clients and in turn reduces the income of thesecompanies As the _rms revenue pro_t decreases the value of the stock alsodecreases that in turn a_ects the returns of BSE SensexSecond graph (betwen BSE and SP500) shows that increase in SP500 has a pos-itive e_ect on BSE as higher returns of SP500 indicate strong global economywhich in turn results in higher trade between countries The positive responseof BSE to one unit shock to SP500 indicates a spillover e_ect of global factorson Indian economy but the response is weak as can be seen from the graphMoving forward response of BSE to shocks in M1 money supply CPI PPImake economic sense As for M1 money supply one unit shock means increasein M1 money supply This increase in money supply allows companies to bor-row more money from banks at lower rates which they can use for investingEDHEC Business School 225 RESULTSin pro_table projects and generating larger cash ows For Ination indicatorsone unit shock means increase in ination This increase in ination results inhigher costs for the companies that in turn reduces their pro_t margins and asa result value of stocksBy looking at the graphs we can also see that shocks to Indian macroeconomicindicators creates stronger response by BSE as compared to global factors likeSP500 or USDINR This indicates that BSE Index is driven by companies thatdepend hugely on domestic demand rather than exports Response of BSE toshocks to Industrial Production are contradictory to theory In theory an in-crease in industrial production should result in positive response from BSE butour analysis shows the other way A possible reason for this response could bethat industrial production time series is seasonal as can be seen from the graphSo there is a possibility of a leadlag relationship between the two variablesTo test for possibility of leadlag relationship we run a Grangers causality testbetween BSE and IP The result in section 646 shows that at a lag lengthof 4 we can reject the Null hypothesis of BSE does not Granger cause IP at1 signi_cance level This proves that BSE is a leading indicator of industrialproduction and there exist a leadlag relationship between the two indicatorsEDHEC Business School 236 CONCLUSIONS6 Conclusions

In this paper I tested the relations between Indian stock market represented byBSE and domestic and global macro economic factors The research concludesthat the India stock markets are mainly driven by domestic demand and theinuence of global macro factors on the stock market is weak I also tested forGranger causality between BSE and IP and found that BSE is a leading indicatorof Industrial production and can help in predicting the industrial climate inIndiaThe research is insightful for investors and professionals who are looking forinvestment opportunities to diversify their risks As Indian stock markets aremore dependent on domestic factors one can invest in Indian indices and stocksto diversify their risks gained through investing in US and European stocksThe paper opens new doors for research in this _eld One can use variancedecomposition technique to see how much variance of BSE can be explained myvarious domestic and global macro factors Also one can use di_erent globalfactors like sovereign CDS spreads T-Bill rates a composite indicator of globaleconomy for further research on interaction between Indian stock market andglobal economyOne can also research on how various global macroeconomicnews a_ects India stock markets and for how long the e_ects persistsEDHEC Business School 247 GRAPHS AND TABLES7 Graphs and Tables71 Graphs of Time seriesEDHEC Business School 257 GRAPHS AND TABLESEDHEC Business School 267 GRAPHS AND TABLESEDHEC Business School 277 GRAPHS AND TABLESEDHEC Business School 287 GRAPHS AND TABLES72 Graphs of Time Series - Di_erencedEDHEC Business School 297 GRAPHS AND TABLESEDHEC Business School 307 GRAPHS AND TABLESEDHEC Business School 317 GRAPHS AND TABLESEDHEC Business School 327 GRAPHS AND TABLES73 Correlograms of Time seriesBSEEDHEC Business School 337 GRAPHS AND TABLESIPEDHEC Business School 347 GRAPHS AND TABLESSP500EDHEC Business School 357 GRAPHS AND TABLESUSDINREDHEC Business School 367 GRAPHS AND TABLESCPIEDHEC Business School 377 GRAPHS AND TABLESPPIEDHEC Business School 387 GRAPHS AND TABLES

M1EDHEC Business School 397 GRAPHS AND TABLES74 Tables741 Table for Unit root test of Time seriesVariables T-Stat p-valueBSE -2671 2495 SP500 -1315 8818 CPI -1909 6466 IP -1669 899 M1 -2420 3679 PPI -3353 601 USDINR -2955 1469 742 Tables for Unit root test of Di_erenced time seriesVariables T-Stat p-valueBSE -13848 000 SP500 -14832 000 CPI -3344 140 IP -3865 027 M1 -3867 026 PPI -9656 000 USDINR -13701 000 743 Tables for Residual based test of cointegrationTable 1BSE - CPIt-Statistic ProbADF test statistic -2622676 087Test critical values 1 level -25738185 level -19420410 level -1615891Table 2BSE - IPt-Statistic ProbADF test statistic -3738802 002Test critical values 1 level -25745135 level -194213610 level -1615828EDHEC Business School 407 GRAPHS AND TABLESTable 3BSE - M1t-Statistic ProbADF test statistic -2875518 041Test critical values 1 level -25737845 level -194203510 level -1615894Table 4BSE - PPIt-Statistic ProbADF test statistic -2399055 162Test critical values 1 level -25737845 level -194203510 level -1615894Table 5BSE - SP500t-Statistic ProbADF test statistic -1427184 1430Test critical values 1 level -25737845 level -1942035

10 level -1615894EDHEC Business School 417 GRAPHS AND TABLESTable 6BSE - USDINRt-Statistic ProbADF test statistic -1659522 917Test critical values 1 level -25738185 level -19420410 level -1615891EDHEC Business School 427 GRAPHS AND TABLES744 Johansen cointegration testEDHEC Business School 437 GRAPHS AND TABLESTable 2EDHEC Business School 447 GRAPHS AND TABLESTable 3EDHEC Business School 457 GRAPHS AND TABLES745 Impulse response testsEDHEC Business School 467 GRAPHS AND TABLESEDHEC Business School 477 GRAPHS AND TABLESEDHEC Business School 487 GRAPHS AND TABLES746 Granger causality test between IP and BSEEDHEC Business School 498 BIBLIOGRAPHY8 BibliographyEugene F Fama Ination Output and Money Journal of Business 1982Eugene F Fama Stock Returns Real activity and Money The American Eco-nomic Review 1981Eugene F Fama Stock Returns Expected Returns and Real activity Journal ofFinance 1990Pal and Mittal Impact of macroeconomic indicators in Indian capital marketsJournal of Risk Finance 2011Shahid Ahmed Aggregate Economic Variables and Stock Markets in India In-ternational Research Journal of Finance and Economics 2008Sahu and Dhiman Correlation and Causality between Stock Market and MacroEconomic Variables in India An Empirical Study 2010 International Confer-ence on E-Business and Economics 2011Mohammad Bayezid Ali Impact of Micro Variables on Emerging Stock MarketReturn A case on Dhaka Stock Exchange (DSE) Interdisciplinary Journal ofResearch in Business 2011Napphon Tangjitprom Macroeconomic Factors of Emerging Stock Market Theevidence from Thailand International Journal of Finance and Research 2012Sayed Mehdi Hosseini The Role of Macroeconomic Variables on Stock MarketIndex in China and India International Journal of Economics and Finance2011John Y Campbell Pitfalls and Opportunities What Macroeconomists shouldknow about Unit Roots NBER Working Papers 1991Hacker and Hatemi The properties of Procedures Dealing with Uncertainityabout Intercept and Deterministic Trend in Unit Root Testing CESIS Elec-tronic Working Papers 2010Elder and Kennedy Testing for Unit Roots What should Students be TaughtNasseh and Strauss Stock Prices and domestic and international macroeco-

nomic activity a cointegration approach The Quarterly Review of Economicsand Finance 2000Engle and Granger Co-Integration and Error Correction Representation Es-timation and Testing Econometrica 1987Eugene F Fama Stock Returns Real Activity Ination and Money 1981American Economic AssociationNaliniprave Tripathy Causal Relationship between Macro-Economic Indicatorsand Stock Market in India Asian Journal of Finance and Accounting 2011Rogalski and Vinso Stock Returns Money Supply and the Direction of Causal-ity The Journal of Finance 1977James et al A VARMA Analysis of the Causal Relations Among Stock Re-turns Real output and Nominal Interest Rates 1985 The Journal of FinanceBailey and Chung Risk and return in the Philippine Equity market A multi-factor exploration Paci_c-Basin Finance Journal 1996Nai-Fu Chen Financial Investment opportunities and the Macroeconomy TheJournal of Finance 1991GB Wickremasinghe Macroeconomic forces and stock prices Some empiricalevidence from an emerging stock markets University of Wollongong 2006EDHEC Business School 508 BIBLIOGRAPHYYao Juo and Loh On Chinas Monetary Policy and Asset Prices University ofNottingham- China policy Institute 2011Bilson et al Selecting macroeconomic variables as explanatory factors of emerg-ing stock market returns Paci_c-Basin Finance Journal 2001CHen Roll and Ross Economic forces and the Stock Markets The Journal ofBusiness 1986William H Greene Econometric Analysis 6th Edition Pearson InternationalEditionRuey Tsay Analysis of Financial Time seriesChris Brooks Introductory Econometrics for Finance Cambridge PublicationsEDHEC Business School 51

the number of cointegrating vector is r against r+144 Impulse ResponseOnce we have determined whether the variables have long term relationship ornot we can form a multivariate VAR model for the variables A multivariateVAR model between g variables is a model where the current value of a variabledepend on di_ernt combinations of the previous k values of all the variables anderror terms A general representation of the model can beyBSEt = _ + _BSEyBSE + _IP yIP + CPIyCPI + _M1yM1 + _SP500ySP500 + u1t(17)where all the coe_cients except _ are g _ k matrices and all variables y are k_ 1 matricesOnce we have formed a model like this we can use the model for Impulse re-sponse A VAR(p) model can be written as a linear fuction of the past innova-tions that isrt = _ + at + 1at10485761 + 2at10485762 + (18)where _ = [_(1)]10485761_0 provided that the inverse exists and the coe_cient ma-trices i can be obtained by equating the coe_cients of Bi in the equation(I 1048576 _1B 1048576 1048576 _PBP )(I + 1B + 2B2 + ) = I (19)EDHEC Business School 194 METHODOLOGYwhere I is the Identity martix This is a moving average representation of rtwith the coe_cient matrix i being the impact of the past innovation at1048576i onrt Equivalently i is the e_ect of at on the future observation rt+i Therefore i is often referred to as the Impulse Response Function of rt For our impulseresponse we will use equation of variables in _rst di_ernce form like_BSEt = _t +Xki=0_11(i)_BSEt1048576i +Xkj=1_12(j)_MIt1048576j + _BSEt (20)_MIt = _t +Xki=0_21(i)_MIt1048576i +Xkj=1_22(j)_BSEt1048576j + _MIt (21)Grangers causality and Blocks F test of a VAR model will suggest which ofthe variables have statistically signi_cant impacts on the future values of othervariables in the system But F-test results cannot explain the sign of the re-lationship nor how long these e_ects require to take place Such informationwill however be given by an examination of the VARs impulse responses andvariance decompositions Impulse response is a technique that trace out theresponsiveness of the dependent variable in the VAR to shocks of each of theother variables So for each variable from each equation separately we will applya unit shock to the error and trace the e_ects upon the VAR system over timeBy using the impulse response technique we can determine how responsive isthe BSE stock index to Indian macro indicators and SP500 This will help usdetermine whether the BSE index is more reactive to domestic news or globalnewsEDHEC Business School 205 RESULTS5 ResultsBefore we use the time series for VAR analysis or cointegration tests we need todetermine whether the series are Stationary or not If the series are stationaryin levels we can use them directly else we need to use the di_erenced time seriesOne way to look for autocorrelation or integrated process is to see the graphsof the various time series used Section 71 shows the graphs of variables we

use for our analysis As we can see from the graphs all of the time series havea trend in long run which points to an integrated process As a second stepwe plot the graphs of di_erenced time series in Section 52 We can see thatthe di_erenced graphs in Section 72 dont show a long term trend and crossthe X-axis frequently This is usually a property of I(1) processes So we checkthe series for autocorrelations at di_erent lag lengths Section 73 shows cor-relograms graph autocorrelation coe_cient partial autocorrelation coe_cientQ-Stat and p-value for various time series up to 36 lags As can be seen in thetables the Q-stat for all lags is zero and we can reject the joint null hypothesisthat all the autocorrelations up to 36 lags are zero Table 741 shows that ifwe conduct a Unit root test on levels of the series we _nd that all the 7 seriesare integrated as we cannot reject the t-stat for unit root at 1 level But ifwe conduct the same test on di_erenced values of the series we _nd that we canreject the null hypothesis of unit root at 1 signi_cance level for all the seriesexcept CPI This tells us that all the series are I(1) as there _rst di_erence seriesare I(0)As our series are I(1) we will work with index levels of time series to determineif there exist one or more cointegrating relationships between the series Tablesin subsection 743 are based on residual approach where we run a regression ofBSE and various macroeconomic indicators and test the residuals for unit rootusing Augmented Dickey-Fuller test As we assume the two series are cointe-grated we conduct the test with no trend and intercept If the two series arecointegrated then the errors should not have any trend or intercept We see thatwe can reject the null hypothesis of unit root at 1 signi_cance for CPIIP M1We can reject the null of unit root for PPI at 5 and for SP500 and USDINRwe cant reject the null hypothesis of unit root at even 5 level This pointsto the fact that BSE has a strong long term relationship with IP M1 moneysupply CPI at 1 level with IP M1 CPI PPI at 5 signi_cance level AlsoBSE has no long term relationship with SP500 and USD INR exchange rateTo test for multiple cointegrating relationship we now employ a Johansen VARbased cointegration test The results of the test are displayed in subsection744 The _rst panel of the test results displays the value of _trace and_maxof Johansen test with di_erent assumptions about intercept and trend We cansee from this panel that when we consider a functional form of intercept and noTrend we have atleast and atmost three cointegrating relationships The secondpanel of the results display the value of information criteria for lag lengths Formost of the models we see that Akalike criteria points to a lag of three andSchwarz criteria points to a lag of one To estimate the cointegrating model wechoose the model with intercept and no trend and run a cointegration testTestresults are shown in Table 2 of subsection 744 At 5 signi_cance level wecan reject the null of atmost two cointegrating factors for _trace and same for_max Now to test which all variables have a long tern relationship we perform aRestricted cointegration with vector error correction model As we had alreadyEDHEC Business School 215 RESULTSseen in our residual based test of cointegration that BSE has no cointegratingrelationship with SP500 and USDINR we create a restricted cointegration modelwhere we set coe_cients of SP500 and USDINR as zero The test results aredisplayed in Table 3 of subsection 744 In this case as there are two restrictionsthe test statistic follow _2 with two degrees of freedom We can see that thep-value for the test is 1333 which tells us that the restrictions are supportedby data at 10 level of signi_cance So we can conclude that the BSE has along term relationship with CPIIPPPIM1 money supply but has no long termrelationship with SP500 and USDINR exchange rate One interpretation of thisresult can be that the Indian stock market represented here by BSE Sensexmoves more in accordance with domestic factors like Industrial production M1money supply Consumer price index and Producer Price index than with globalfactors or in other words as BSE is representation of largest market cap Indiancompanies we can say that the biggest companies in India are ones that are

more dependent on domestic demand rather than exports This result presentsan opportunity for international investors to diversify their portfolio by invest-ing in BSE Sensex as it is decoupled with global markets and macroeconomicfactorsWe use A bivariate Vector Autoregression (BVAR) technique to analyze thedynamic interaction between real asset prices and macro economy VAR ispreferred method to study Macroeconomy and asset prices where variables en-dogenously e_ect each otherWe begin with a bivariate VAR with no restriction Asset prices and instru-ments are allowed to respond to each other freely For paired variables withcointegration relationship VAR is performed at levels whilst for those that arenot cointegrated VAR is performed at _rst di_erence Constant term is ignoredwith loss of generality We use the Bivariate Autoregression analysis for bothimpulse response and Grangers causality testsImpulse response results are displayed in subsection 745 From _rst graph ofimpulse response of BSE to USDINR we can see that USDINR has a negativeimpact on BSE As impulse response is response of BSE to shocks given to US-DINR we can see that a positive shock or unexpected appreciation INR valuewrt USD will have a negative e_ect on BSE for few lags and will disappearafter few lags If we look at the constituents of BSE Index over time we seethat most of the time some of its constituent are companies that thrive on ex-ports Some of the biggest Market-Cap in India are companies in service sectorlike Infosys TCS etc that are hugely dependent on services provided to clientsfrom Europe and US So an appreciation of INR compared to USD makesthese _rms costlier for the global clients and in turn reduces the income of thesecompanies As the _rms revenue pro_t decreases the value of the stock alsodecreases that in turn a_ects the returns of BSE SensexSecond graph (betwen BSE and SP500) shows that increase in SP500 has a pos-itive e_ect on BSE as higher returns of SP500 indicate strong global economywhich in turn results in higher trade between countries The positive responseof BSE to one unit shock to SP500 indicates a spillover e_ect of global factorson Indian economy but the response is weak as can be seen from the graphMoving forward response of BSE to shocks in M1 money supply CPI PPImake economic sense As for M1 money supply one unit shock means increasein M1 money supply This increase in money supply allows companies to bor-row more money from banks at lower rates which they can use for investingEDHEC Business School 225 RESULTSin pro_table projects and generating larger cash ows For Ination indicatorsone unit shock means increase in ination This increase in ination results inhigher costs for the companies that in turn reduces their pro_t margins and asa result value of stocksBy looking at the graphs we can also see that shocks to Indian macroeconomicindicators creates stronger response by BSE as compared to global factors likeSP500 or USDINR This indicates that BSE Index is driven by companies thatdepend hugely on domestic demand rather than exports Response of BSE toshocks to Industrial Production are contradictory to theory In theory an in-crease in industrial production should result in positive response from BSE butour analysis shows the other way A possible reason for this response could bethat industrial production time series is seasonal as can be seen from the graphSo there is a possibility of a leadlag relationship between the two variablesTo test for possibility of leadlag relationship we run a Grangers causality testbetween BSE and IP The result in section 646 shows that at a lag lengthof 4 we can reject the Null hypothesis of BSE does not Granger cause IP at1 signi_cance level This proves that BSE is a leading indicator of industrialproduction and there exist a leadlag relationship between the two indicatorsEDHEC Business School 236 CONCLUSIONS6 Conclusions

In this paper I tested the relations between Indian stock market represented byBSE and domestic and global macro economic factors The research concludesthat the India stock markets are mainly driven by domestic demand and theinuence of global macro factors on the stock market is weak I also tested forGranger causality between BSE and IP and found that BSE is a leading indicatorof Industrial production and can help in predicting the industrial climate inIndiaThe research is insightful for investors and professionals who are looking forinvestment opportunities to diversify their risks As Indian stock markets aremore dependent on domestic factors one can invest in Indian indices and stocksto diversify their risks gained through investing in US and European stocksThe paper opens new doors for research in this _eld One can use variancedecomposition technique to see how much variance of BSE can be explained myvarious domestic and global macro factors Also one can use di_erent globalfactors like sovereign CDS spreads T-Bill rates a composite indicator of globaleconomy for further research on interaction between Indian stock market andglobal economyOne can also research on how various global macroeconomicnews a_ects India stock markets and for how long the e_ects persistsEDHEC Business School 247 GRAPHS AND TABLES7 Graphs and Tables71 Graphs of Time seriesEDHEC Business School 257 GRAPHS AND TABLESEDHEC Business School 267 GRAPHS AND TABLESEDHEC Business School 277 GRAPHS AND TABLESEDHEC Business School 287 GRAPHS AND TABLES72 Graphs of Time Series - Di_erencedEDHEC Business School 297 GRAPHS AND TABLESEDHEC Business School 307 GRAPHS AND TABLESEDHEC Business School 317 GRAPHS AND TABLESEDHEC Business School 327 GRAPHS AND TABLES73 Correlograms of Time seriesBSEEDHEC Business School 337 GRAPHS AND TABLESIPEDHEC Business School 347 GRAPHS AND TABLESSP500EDHEC Business School 357 GRAPHS AND TABLESUSDINREDHEC Business School 367 GRAPHS AND TABLESCPIEDHEC Business School 377 GRAPHS AND TABLESPPIEDHEC Business School 387 GRAPHS AND TABLES

M1EDHEC Business School 397 GRAPHS AND TABLES74 Tables741 Table for Unit root test of Time seriesVariables T-Stat p-valueBSE -2671 2495 SP500 -1315 8818 CPI -1909 6466 IP -1669 899 M1 -2420 3679 PPI -3353 601 USDINR -2955 1469 742 Tables for Unit root test of Di_erenced time seriesVariables T-Stat p-valueBSE -13848 000 SP500 -14832 000 CPI -3344 140 IP -3865 027 M1 -3867 026 PPI -9656 000 USDINR -13701 000 743 Tables for Residual based test of cointegrationTable 1BSE - CPIt-Statistic ProbADF test statistic -2622676 087Test critical values 1 level -25738185 level -19420410 level -1615891Table 2BSE - IPt-Statistic ProbADF test statistic -3738802 002Test critical values 1 level -25745135 level -194213610 level -1615828EDHEC Business School 407 GRAPHS AND TABLESTable 3BSE - M1t-Statistic ProbADF test statistic -2875518 041Test critical values 1 level -25737845 level -194203510 level -1615894Table 4BSE - PPIt-Statistic ProbADF test statistic -2399055 162Test critical values 1 level -25737845 level -194203510 level -1615894Table 5BSE - SP500t-Statistic ProbADF test statistic -1427184 1430Test critical values 1 level -25737845 level -1942035

10 level -1615894EDHEC Business School 417 GRAPHS AND TABLESTable 6BSE - USDINRt-Statistic ProbADF test statistic -1659522 917Test critical values 1 level -25738185 level -19420410 level -1615891EDHEC Business School 427 GRAPHS AND TABLES744 Johansen cointegration testEDHEC Business School 437 GRAPHS AND TABLESTable 2EDHEC Business School 447 GRAPHS AND TABLESTable 3EDHEC Business School 457 GRAPHS AND TABLES745 Impulse response testsEDHEC Business School 467 GRAPHS AND TABLESEDHEC Business School 477 GRAPHS AND TABLESEDHEC Business School 487 GRAPHS AND TABLES746 Granger causality test between IP and BSEEDHEC Business School 498 BIBLIOGRAPHY8 BibliographyEugene F Fama Ination Output and Money Journal of Business 1982Eugene F Fama Stock Returns Real activity and Money The American Eco-nomic Review 1981Eugene F Fama Stock Returns Expected Returns and Real activity Journal ofFinance 1990Pal and Mittal Impact of macroeconomic indicators in Indian capital marketsJournal of Risk Finance 2011Shahid Ahmed Aggregate Economic Variables and Stock Markets in India In-ternational Research Journal of Finance and Economics 2008Sahu and Dhiman Correlation and Causality between Stock Market and MacroEconomic Variables in India An Empirical Study 2010 International Confer-ence on E-Business and Economics 2011Mohammad Bayezid Ali Impact of Micro Variables on Emerging Stock MarketReturn A case on Dhaka Stock Exchange (DSE) Interdisciplinary Journal ofResearch in Business 2011Napphon Tangjitprom Macroeconomic Factors of Emerging Stock Market Theevidence from Thailand International Journal of Finance and Research 2012Sayed Mehdi Hosseini The Role of Macroeconomic Variables on Stock MarketIndex in China and India International Journal of Economics and Finance2011John Y Campbell Pitfalls and Opportunities What Macroeconomists shouldknow about Unit Roots NBER Working Papers 1991Hacker and Hatemi The properties of Procedures Dealing with Uncertainityabout Intercept and Deterministic Trend in Unit Root Testing CESIS Elec-tronic Working Papers 2010Elder and Kennedy Testing for Unit Roots What should Students be TaughtNasseh and Strauss Stock Prices and domestic and international macroeco-

nomic activity a cointegration approach The Quarterly Review of Economicsand Finance 2000Engle and Granger Co-Integration and Error Correction Representation Es-timation and Testing Econometrica 1987Eugene F Fama Stock Returns Real Activity Ination and Money 1981American Economic AssociationNaliniprave Tripathy Causal Relationship between Macro-Economic Indicatorsand Stock Market in India Asian Journal of Finance and Accounting 2011Rogalski and Vinso Stock Returns Money Supply and the Direction of Causal-ity The Journal of Finance 1977James et al A VARMA Analysis of the Causal Relations Among Stock Re-turns Real output and Nominal Interest Rates 1985 The Journal of FinanceBailey and Chung Risk and return in the Philippine Equity market A multi-factor exploration Paci_c-Basin Finance Journal 1996Nai-Fu Chen Financial Investment opportunities and the Macroeconomy TheJournal of Finance 1991GB Wickremasinghe Macroeconomic forces and stock prices Some empiricalevidence from an emerging stock markets University of Wollongong 2006EDHEC Business School 508 BIBLIOGRAPHYYao Juo and Loh On Chinas Monetary Policy and Asset Prices University ofNottingham- China policy Institute 2011Bilson et al Selecting macroeconomic variables as explanatory factors of emerg-ing stock market returns Paci_c-Basin Finance Journal 2001CHen Roll and Ross Economic forces and the Stock Markets The Journal ofBusiness 1986William H Greene Econometric Analysis 6th Edition Pearson InternationalEditionRuey Tsay Analysis of Financial Time seriesChris Brooks Introductory Econometrics for Finance Cambridge PublicationsEDHEC Business School 51

use for our analysis As we can see from the graphs all of the time series havea trend in long run which points to an integrated process As a second stepwe plot the graphs of di_erenced time series in Section 52 We can see thatthe di_erenced graphs in Section 72 dont show a long term trend and crossthe X-axis frequently This is usually a property of I(1) processes So we checkthe series for autocorrelations at di_erent lag lengths Section 73 shows cor-relograms graph autocorrelation coe_cient partial autocorrelation coe_cientQ-Stat and p-value for various time series up to 36 lags As can be seen in thetables the Q-stat for all lags is zero and we can reject the joint null hypothesisthat all the autocorrelations up to 36 lags are zero Table 741 shows that ifwe conduct a Unit root test on levels of the series we _nd that all the 7 seriesare integrated as we cannot reject the t-stat for unit root at 1 level But ifwe conduct the same test on di_erenced values of the series we _nd that we canreject the null hypothesis of unit root at 1 signi_cance level for all the seriesexcept CPI This tells us that all the series are I(1) as there _rst di_erence seriesare I(0)As our series are I(1) we will work with index levels of time series to determineif there exist one or more cointegrating relationships between the series Tablesin subsection 743 are based on residual approach where we run a regression ofBSE and various macroeconomic indicators and test the residuals for unit rootusing Augmented Dickey-Fuller test As we assume the two series are cointe-grated we conduct the test with no trend and intercept If the two series arecointegrated then the errors should not have any trend or intercept We see thatwe can reject the null hypothesis of unit root at 1 signi_cance for CPIIP M1We can reject the null of unit root for PPI at 5 and for SP500 and USDINRwe cant reject the null hypothesis of unit root at even 5 level This pointsto the fact that BSE has a strong long term relationship with IP M1 moneysupply CPI at 1 level with IP M1 CPI PPI at 5 signi_cance level AlsoBSE has no long term relationship with SP500 and USD INR exchange rateTo test for multiple cointegrating relationship we now employ a Johansen VARbased cointegration test The results of the test are displayed in subsection744 The _rst panel of the test results displays the value of _trace and_maxof Johansen test with di_erent assumptions about intercept and trend We cansee from this panel that when we consider a functional form of intercept and noTrend we have atleast and atmost three cointegrating relationships The secondpanel of the results display the value of information criteria for lag lengths Formost of the models we see that Akalike criteria points to a lag of three andSchwarz criteria points to a lag of one To estimate the cointegrating model wechoose the model with intercept and no trend and run a cointegration testTestresults are shown in Table 2 of subsection 744 At 5 signi_cance level wecan reject the null of atmost two cointegrating factors for _trace and same for_max Now to test which all variables have a long tern relationship we perform aRestricted cointegration with vector error correction model As we had alreadyEDHEC Business School 215 RESULTSseen in our residual based test of cointegration that BSE has no cointegratingrelationship with SP500 and USDINR we create a restricted cointegration modelwhere we set coe_cients of SP500 and USDINR as zero The test results aredisplayed in Table 3 of subsection 744 In this case as there are two restrictionsthe test statistic follow _2 with two degrees of freedom We can see that thep-value for the test is 1333 which tells us that the restrictions are supportedby data at 10 level of signi_cance So we can conclude that the BSE has along term relationship with CPIIPPPIM1 money supply but has no long termrelationship with SP500 and USDINR exchange rate One interpretation of thisresult can be that the Indian stock market represented here by BSE Sensexmoves more in accordance with domestic factors like Industrial production M1money supply Consumer price index and Producer Price index than with globalfactors or in other words as BSE is representation of largest market cap Indiancompanies we can say that the biggest companies in India are ones that are

more dependent on domestic demand rather than exports This result presentsan opportunity for international investors to diversify their portfolio by invest-ing in BSE Sensex as it is decoupled with global markets and macroeconomicfactorsWe use A bivariate Vector Autoregression (BVAR) technique to analyze thedynamic interaction between real asset prices and macro economy VAR ispreferred method to study Macroeconomy and asset prices where variables en-dogenously e_ect each otherWe begin with a bivariate VAR with no restriction Asset prices and instru-ments are allowed to respond to each other freely For paired variables withcointegration relationship VAR is performed at levels whilst for those that arenot cointegrated VAR is performed at _rst di_erence Constant term is ignoredwith loss of generality We use the Bivariate Autoregression analysis for bothimpulse response and Grangers causality testsImpulse response results are displayed in subsection 745 From _rst graph ofimpulse response of BSE to USDINR we can see that USDINR has a negativeimpact on BSE As impulse response is response of BSE to shocks given to US-DINR we can see that a positive shock or unexpected appreciation INR valuewrt USD will have a negative e_ect on BSE for few lags and will disappearafter few lags If we look at the constituents of BSE Index over time we seethat most of the time some of its constituent are companies that thrive on ex-ports Some of the biggest Market-Cap in India are companies in service sectorlike Infosys TCS etc that are hugely dependent on services provided to clientsfrom Europe and US So an appreciation of INR compared to USD makesthese _rms costlier for the global clients and in turn reduces the income of thesecompanies As the _rms revenue pro_t decreases the value of the stock alsodecreases that in turn a_ects the returns of BSE SensexSecond graph (betwen BSE and SP500) shows that increase in SP500 has a pos-itive e_ect on BSE as higher returns of SP500 indicate strong global economywhich in turn results in higher trade between countries The positive responseof BSE to one unit shock to SP500 indicates a spillover e_ect of global factorson Indian economy but the response is weak as can be seen from the graphMoving forward response of BSE to shocks in M1 money supply CPI PPImake economic sense As for M1 money supply one unit shock means increasein M1 money supply This increase in money supply allows companies to bor-row more money from banks at lower rates which they can use for investingEDHEC Business School 225 RESULTSin pro_table projects and generating larger cash ows For Ination indicatorsone unit shock means increase in ination This increase in ination results inhigher costs for the companies that in turn reduces their pro_t margins and asa result value of stocksBy looking at the graphs we can also see that shocks to Indian macroeconomicindicators creates stronger response by BSE as compared to global factors likeSP500 or USDINR This indicates that BSE Index is driven by companies thatdepend hugely on domestic demand rather than exports Response of BSE toshocks to Industrial Production are contradictory to theory In theory an in-crease in industrial production should result in positive response from BSE butour analysis shows the other way A possible reason for this response could bethat industrial production time series is seasonal as can be seen from the graphSo there is a possibility of a leadlag relationship between the two variablesTo test for possibility of leadlag relationship we run a Grangers causality testbetween BSE and IP The result in section 646 shows that at a lag lengthof 4 we can reject the Null hypothesis of BSE does not Granger cause IP at1 signi_cance level This proves that BSE is a leading indicator of industrialproduction and there exist a leadlag relationship between the two indicatorsEDHEC Business School 236 CONCLUSIONS6 Conclusions

In this paper I tested the relations between Indian stock market represented byBSE and domestic and global macro economic factors The research concludesthat the India stock markets are mainly driven by domestic demand and theinuence of global macro factors on the stock market is weak I also tested forGranger causality between BSE and IP and found that BSE is a leading indicatorof Industrial production and can help in predicting the industrial climate inIndiaThe research is insightful for investors and professionals who are looking forinvestment opportunities to diversify their risks As Indian stock markets aremore dependent on domestic factors one can invest in Indian indices and stocksto diversify their risks gained through investing in US and European stocksThe paper opens new doors for research in this _eld One can use variancedecomposition technique to see how much variance of BSE can be explained myvarious domestic and global macro factors Also one can use di_erent globalfactors like sovereign CDS spreads T-Bill rates a composite indicator of globaleconomy for further research on interaction between Indian stock market andglobal economyOne can also research on how various global macroeconomicnews a_ects India stock markets and for how long the e_ects persistsEDHEC Business School 247 GRAPHS AND TABLES7 Graphs and Tables71 Graphs of Time seriesEDHEC Business School 257 GRAPHS AND TABLESEDHEC Business School 267 GRAPHS AND TABLESEDHEC Business School 277 GRAPHS AND TABLESEDHEC Business School 287 GRAPHS AND TABLES72 Graphs of Time Series - Di_erencedEDHEC Business School 297 GRAPHS AND TABLESEDHEC Business School 307 GRAPHS AND TABLESEDHEC Business School 317 GRAPHS AND TABLESEDHEC Business School 327 GRAPHS AND TABLES73 Correlograms of Time seriesBSEEDHEC Business School 337 GRAPHS AND TABLESIPEDHEC Business School 347 GRAPHS AND TABLESSP500EDHEC Business School 357 GRAPHS AND TABLESUSDINREDHEC Business School 367 GRAPHS AND TABLESCPIEDHEC Business School 377 GRAPHS AND TABLESPPIEDHEC Business School 387 GRAPHS AND TABLES

M1EDHEC Business School 397 GRAPHS AND TABLES74 Tables741 Table for Unit root test of Time seriesVariables T-Stat p-valueBSE -2671 2495 SP500 -1315 8818 CPI -1909 6466 IP -1669 899 M1 -2420 3679 PPI -3353 601 USDINR -2955 1469 742 Tables for Unit root test of Di_erenced time seriesVariables T-Stat p-valueBSE -13848 000 SP500 -14832 000 CPI -3344 140 IP -3865 027 M1 -3867 026 PPI -9656 000 USDINR -13701 000 743 Tables for Residual based test of cointegrationTable 1BSE - CPIt-Statistic ProbADF test statistic -2622676 087Test critical values 1 level -25738185 level -19420410 level -1615891Table 2BSE - IPt-Statistic ProbADF test statistic -3738802 002Test critical values 1 level -25745135 level -194213610 level -1615828EDHEC Business School 407 GRAPHS AND TABLESTable 3BSE - M1t-Statistic ProbADF test statistic -2875518 041Test critical values 1 level -25737845 level -194203510 level -1615894Table 4BSE - PPIt-Statistic ProbADF test statistic -2399055 162Test critical values 1 level -25737845 level -194203510 level -1615894Table 5BSE - SP500t-Statistic ProbADF test statistic -1427184 1430Test critical values 1 level -25737845 level -1942035

10 level -1615894EDHEC Business School 417 GRAPHS AND TABLESTable 6BSE - USDINRt-Statistic ProbADF test statistic -1659522 917Test critical values 1 level -25738185 level -19420410 level -1615891EDHEC Business School 427 GRAPHS AND TABLES744 Johansen cointegration testEDHEC Business School 437 GRAPHS AND TABLESTable 2EDHEC Business School 447 GRAPHS AND TABLESTable 3EDHEC Business School 457 GRAPHS AND TABLES745 Impulse response testsEDHEC Business School 467 GRAPHS AND TABLESEDHEC Business School 477 GRAPHS AND TABLESEDHEC Business School 487 GRAPHS AND TABLES746 Granger causality test between IP and BSEEDHEC Business School 498 BIBLIOGRAPHY8 BibliographyEugene F Fama Ination Output and Money Journal of Business 1982Eugene F Fama Stock Returns Real activity and Money The American Eco-nomic Review 1981Eugene F Fama Stock Returns Expected Returns and Real activity Journal ofFinance 1990Pal and Mittal Impact of macroeconomic indicators in Indian capital marketsJournal of Risk Finance 2011Shahid Ahmed Aggregate Economic Variables and Stock Markets in India In-ternational Research Journal of Finance and Economics 2008Sahu and Dhiman Correlation and Causality between Stock Market and MacroEconomic Variables in India An Empirical Study 2010 International Confer-ence on E-Business and Economics 2011Mohammad Bayezid Ali Impact of Micro Variables on Emerging Stock MarketReturn A case on Dhaka Stock Exchange (DSE) Interdisciplinary Journal ofResearch in Business 2011Napphon Tangjitprom Macroeconomic Factors of Emerging Stock Market Theevidence from Thailand International Journal of Finance and Research 2012Sayed Mehdi Hosseini The Role of Macroeconomic Variables on Stock MarketIndex in China and India International Journal of Economics and Finance2011John Y Campbell Pitfalls and Opportunities What Macroeconomists shouldknow about Unit Roots NBER Working Papers 1991Hacker and Hatemi The properties of Procedures Dealing with Uncertainityabout Intercept and Deterministic Trend in Unit Root Testing CESIS Elec-tronic Working Papers 2010Elder and Kennedy Testing for Unit Roots What should Students be TaughtNasseh and Strauss Stock Prices and domestic and international macroeco-

nomic activity a cointegration approach The Quarterly Review of Economicsand Finance 2000Engle and Granger Co-Integration and Error Correction Representation Es-timation and Testing Econometrica 1987Eugene F Fama Stock Returns Real Activity Ination and Money 1981American Economic AssociationNaliniprave Tripathy Causal Relationship between Macro-Economic Indicatorsand Stock Market in India Asian Journal of Finance and Accounting 2011Rogalski and Vinso Stock Returns Money Supply and the Direction of Causal-ity The Journal of Finance 1977James et al A VARMA Analysis of the Causal Relations Among Stock Re-turns Real output and Nominal Interest Rates 1985 The Journal of FinanceBailey and Chung Risk and return in the Philippine Equity market A multi-factor exploration Paci_c-Basin Finance Journal 1996Nai-Fu Chen Financial Investment opportunities and the Macroeconomy TheJournal of Finance 1991GB Wickremasinghe Macroeconomic forces and stock prices Some empiricalevidence from an emerging stock markets University of Wollongong 2006EDHEC Business School 508 BIBLIOGRAPHYYao Juo and Loh On Chinas Monetary Policy and Asset Prices University ofNottingham- China policy Institute 2011Bilson et al Selecting macroeconomic variables as explanatory factors of emerg-ing stock market returns Paci_c-Basin Finance Journal 2001CHen Roll and Ross Economic forces and the Stock Markets The Journal ofBusiness 1986William H Greene Econometric Analysis 6th Edition Pearson InternationalEditionRuey Tsay Analysis of Financial Time seriesChris Brooks Introductory Econometrics for Finance Cambridge PublicationsEDHEC Business School 51

more dependent on domestic demand rather than exports This result presentsan opportunity for international investors to diversify their portfolio by invest-ing in BSE Sensex as it is decoupled with global markets and macroeconomicfactorsWe use A bivariate Vector Autoregression (BVAR) technique to analyze thedynamic interaction between real asset prices and macro economy VAR ispreferred method to study Macroeconomy and asset prices where variables en-dogenously e_ect each otherWe begin with a bivariate VAR with no restriction Asset prices and instru-ments are allowed to respond to each other freely For paired variables withcointegration relationship VAR is performed at levels whilst for those that arenot cointegrated VAR is performed at _rst di_erence Constant term is ignoredwith loss of generality We use the Bivariate Autoregression analysis for bothimpulse response and Grangers causality testsImpulse response results are displayed in subsection 745 From _rst graph ofimpulse response of BSE to USDINR we can see that USDINR has a negativeimpact on BSE As impulse response is response of BSE to shocks given to US-DINR we can see that a positive shock or unexpected appreciation INR valuewrt USD will have a negative e_ect on BSE for few lags and will disappearafter few lags If we look at the constituents of BSE Index over time we seethat most of the time some of its constituent are companies that thrive on ex-ports Some of the biggest Market-Cap in India are companies in service sectorlike Infosys TCS etc that are hugely dependent on services provided to clientsfrom Europe and US So an appreciation of INR compared to USD makesthese _rms costlier for the global clients and in turn reduces the income of thesecompanies As the _rms revenue pro_t decreases the value of the stock alsodecreases that in turn a_ects the returns of BSE SensexSecond graph (betwen BSE and SP500) shows that increase in SP500 has a pos-itive e_ect on BSE as higher returns of SP500 indicate strong global economywhich in turn results in higher trade between countries The positive responseof BSE to one unit shock to SP500 indicates a spillover e_ect of global factorson Indian economy but the response is weak as can be seen from the graphMoving forward response of BSE to shocks in M1 money supply CPI PPImake economic sense As for M1 money supply one unit shock means increasein M1 money supply This increase in money supply allows companies to bor-row more money from banks at lower rates which they can use for investingEDHEC Business School 225 RESULTSin pro_table projects and generating larger cash ows For Ination indicatorsone unit shock means increase in ination This increase in ination results inhigher costs for the companies that in turn reduces their pro_t margins and asa result value of stocksBy looking at the graphs we can also see that shocks to Indian macroeconomicindicators creates stronger response by BSE as compared to global factors likeSP500 or USDINR This indicates that BSE Index is driven by companies thatdepend hugely on domestic demand rather than exports Response of BSE toshocks to Industrial Production are contradictory to theory In theory an in-crease in industrial production should result in positive response from BSE butour analysis shows the other way A possible reason for this response could bethat industrial production time series is seasonal as can be seen from the graphSo there is a possibility of a leadlag relationship between the two variablesTo test for possibility of leadlag relationship we run a Grangers causality testbetween BSE and IP The result in section 646 shows that at a lag lengthof 4 we can reject the Null hypothesis of BSE does not Granger cause IP at1 signi_cance level This proves that BSE is a leading indicator of industrialproduction and there exist a leadlag relationship between the two indicatorsEDHEC Business School 236 CONCLUSIONS6 Conclusions

In this paper I tested the relations between Indian stock market represented byBSE and domestic and global macro economic factors The research concludesthat the India stock markets are mainly driven by domestic demand and theinuence of global macro factors on the stock market is weak I also tested forGranger causality between BSE and IP and found that BSE is a leading indicatorof Industrial production and can help in predicting the industrial climate inIndiaThe research is insightful for investors and professionals who are looking forinvestment opportunities to diversify their risks As Indian stock markets aremore dependent on domestic factors one can invest in Indian indices and stocksto diversify their risks gained through investing in US and European stocksThe paper opens new doors for research in this _eld One can use variancedecomposition technique to see how much variance of BSE can be explained myvarious domestic and global macro factors Also one can use di_erent globalfactors like sovereign CDS spreads T-Bill rates a composite indicator of globaleconomy for further research on interaction between Indian stock market andglobal economyOne can also research on how various global macroeconomicnews a_ects India stock markets and for how long the e_ects persistsEDHEC Business School 247 GRAPHS AND TABLES7 Graphs and Tables71 Graphs of Time seriesEDHEC Business School 257 GRAPHS AND TABLESEDHEC Business School 267 GRAPHS AND TABLESEDHEC Business School 277 GRAPHS AND TABLESEDHEC Business School 287 GRAPHS AND TABLES72 Graphs of Time Series - Di_erencedEDHEC Business School 297 GRAPHS AND TABLESEDHEC Business School 307 GRAPHS AND TABLESEDHEC Business School 317 GRAPHS AND TABLESEDHEC Business School 327 GRAPHS AND TABLES73 Correlograms of Time seriesBSEEDHEC Business School 337 GRAPHS AND TABLESIPEDHEC Business School 347 GRAPHS AND TABLESSP500EDHEC Business School 357 GRAPHS AND TABLESUSDINREDHEC Business School 367 GRAPHS AND TABLESCPIEDHEC Business School 377 GRAPHS AND TABLESPPIEDHEC Business School 387 GRAPHS AND TABLES

M1EDHEC Business School 397 GRAPHS AND TABLES74 Tables741 Table for Unit root test of Time seriesVariables T-Stat p-valueBSE -2671 2495 SP500 -1315 8818 CPI -1909 6466 IP -1669 899 M1 -2420 3679 PPI -3353 601 USDINR -2955 1469 742 Tables for Unit root test of Di_erenced time seriesVariables T-Stat p-valueBSE -13848 000 SP500 -14832 000 CPI -3344 140 IP -3865 027 M1 -3867 026 PPI -9656 000 USDINR -13701 000 743 Tables for Residual based test of cointegrationTable 1BSE - CPIt-Statistic ProbADF test statistic -2622676 087Test critical values 1 level -25738185 level -19420410 level -1615891Table 2BSE - IPt-Statistic ProbADF test statistic -3738802 002Test critical values 1 level -25745135 level -194213610 level -1615828EDHEC Business School 407 GRAPHS AND TABLESTable 3BSE - M1t-Statistic ProbADF test statistic -2875518 041Test critical values 1 level -25737845 level -194203510 level -1615894Table 4BSE - PPIt-Statistic ProbADF test statistic -2399055 162Test critical values 1 level -25737845 level -194203510 level -1615894Table 5BSE - SP500t-Statistic ProbADF test statistic -1427184 1430Test critical values 1 level -25737845 level -1942035

10 level -1615894EDHEC Business School 417 GRAPHS AND TABLESTable 6BSE - USDINRt-Statistic ProbADF test statistic -1659522 917Test critical values 1 level -25738185 level -19420410 level -1615891EDHEC Business School 427 GRAPHS AND TABLES744 Johansen cointegration testEDHEC Business School 437 GRAPHS AND TABLESTable 2EDHEC Business School 447 GRAPHS AND TABLESTable 3EDHEC Business School 457 GRAPHS AND TABLES745 Impulse response testsEDHEC Business School 467 GRAPHS AND TABLESEDHEC Business School 477 GRAPHS AND TABLESEDHEC Business School 487 GRAPHS AND TABLES746 Granger causality test between IP and BSEEDHEC Business School 498 BIBLIOGRAPHY8 BibliographyEugene F Fama Ination Output and Money Journal of Business 1982Eugene F Fama Stock Returns Real activity and Money The American Eco-nomic Review 1981Eugene F Fama Stock Returns Expected Returns and Real activity Journal ofFinance 1990Pal and Mittal Impact of macroeconomic indicators in Indian capital marketsJournal of Risk Finance 2011Shahid Ahmed Aggregate Economic Variables and Stock Markets in India In-ternational Research Journal of Finance and Economics 2008Sahu and Dhiman Correlation and Causality between Stock Market and MacroEconomic Variables in India An Empirical Study 2010 International Confer-ence on E-Business and Economics 2011Mohammad Bayezid Ali Impact of Micro Variables on Emerging Stock MarketReturn A case on Dhaka Stock Exchange (DSE) Interdisciplinary Journal ofResearch in Business 2011Napphon Tangjitprom Macroeconomic Factors of Emerging Stock Market Theevidence from Thailand International Journal of Finance and Research 2012Sayed Mehdi Hosseini The Role of Macroeconomic Variables on Stock MarketIndex in China and India International Journal of Economics and Finance2011John Y Campbell Pitfalls and Opportunities What Macroeconomists shouldknow about Unit Roots NBER Working Papers 1991Hacker and Hatemi The properties of Procedures Dealing with Uncertainityabout Intercept and Deterministic Trend in Unit Root Testing CESIS Elec-tronic Working Papers 2010Elder and Kennedy Testing for Unit Roots What should Students be TaughtNasseh and Strauss Stock Prices and domestic and international macroeco-

nomic activity a cointegration approach The Quarterly Review of Economicsand Finance 2000Engle and Granger Co-Integration and Error Correction Representation Es-timation and Testing Econometrica 1987Eugene F Fama Stock Returns Real Activity Ination and Money 1981American Economic AssociationNaliniprave Tripathy Causal Relationship between Macro-Economic Indicatorsand Stock Market in India Asian Journal of Finance and Accounting 2011Rogalski and Vinso Stock Returns Money Supply and the Direction of Causal-ity The Journal of Finance 1977James et al A VARMA Analysis of the Causal Relations Among Stock Re-turns Real output and Nominal Interest Rates 1985 The Journal of FinanceBailey and Chung Risk and return in the Philippine Equity market A multi-factor exploration Paci_c-Basin Finance Journal 1996Nai-Fu Chen Financial Investment opportunities and the Macroeconomy TheJournal of Finance 1991GB Wickremasinghe Macroeconomic forces and stock prices Some empiricalevidence from an emerging stock markets University of Wollongong 2006EDHEC Business School 508 BIBLIOGRAPHYYao Juo and Loh On Chinas Monetary Policy and Asset Prices University ofNottingham- China policy Institute 2011Bilson et al Selecting macroeconomic variables as explanatory factors of emerg-ing stock market returns Paci_c-Basin Finance Journal 2001CHen Roll and Ross Economic forces and the Stock Markets The Journal ofBusiness 1986William H Greene Econometric Analysis 6th Edition Pearson InternationalEditionRuey Tsay Analysis of Financial Time seriesChris Brooks Introductory Econometrics for Finance Cambridge PublicationsEDHEC Business School 51

In this paper I tested the relations between Indian stock market represented byBSE and domestic and global macro economic factors The research concludesthat the India stock markets are mainly driven by domestic demand and theinuence of global macro factors on the stock market is weak I also tested forGranger causality between BSE and IP and found that BSE is a leading indicatorof Industrial production and can help in predicting the industrial climate inIndiaThe research is insightful for investors and professionals who are looking forinvestment opportunities to diversify their risks As Indian stock markets aremore dependent on domestic factors one can invest in Indian indices and stocksto diversify their risks gained through investing in US and European stocksThe paper opens new doors for research in this _eld One can use variancedecomposition technique to see how much variance of BSE can be explained myvarious domestic and global macro factors Also one can use di_erent globalfactors like sovereign CDS spreads T-Bill rates a composite indicator of globaleconomy for further research on interaction between Indian stock market andglobal economyOne can also research on how various global macroeconomicnews a_ects India stock markets and for how long the e_ects persistsEDHEC Business School 247 GRAPHS AND TABLES7 Graphs and Tables71 Graphs of Time seriesEDHEC Business School 257 GRAPHS AND TABLESEDHEC Business School 267 GRAPHS AND TABLESEDHEC Business School 277 GRAPHS AND TABLESEDHEC Business School 287 GRAPHS AND TABLES72 Graphs of Time Series - Di_erencedEDHEC Business School 297 GRAPHS AND TABLESEDHEC Business School 307 GRAPHS AND TABLESEDHEC Business School 317 GRAPHS AND TABLESEDHEC Business School 327 GRAPHS AND TABLES73 Correlograms of Time seriesBSEEDHEC Business School 337 GRAPHS AND TABLESIPEDHEC Business School 347 GRAPHS AND TABLESSP500EDHEC Business School 357 GRAPHS AND TABLESUSDINREDHEC Business School 367 GRAPHS AND TABLESCPIEDHEC Business School 377 GRAPHS AND TABLESPPIEDHEC Business School 387 GRAPHS AND TABLES

M1EDHEC Business School 397 GRAPHS AND TABLES74 Tables741 Table for Unit root test of Time seriesVariables T-Stat p-valueBSE -2671 2495 SP500 -1315 8818 CPI -1909 6466 IP -1669 899 M1 -2420 3679 PPI -3353 601 USDINR -2955 1469 742 Tables for Unit root test of Di_erenced time seriesVariables T-Stat p-valueBSE -13848 000 SP500 -14832 000 CPI -3344 140 IP -3865 027 M1 -3867 026 PPI -9656 000 USDINR -13701 000 743 Tables for Residual based test of cointegrationTable 1BSE - CPIt-Statistic ProbADF test statistic -2622676 087Test critical values 1 level -25738185 level -19420410 level -1615891Table 2BSE - IPt-Statistic ProbADF test statistic -3738802 002Test critical values 1 level -25745135 level -194213610 level -1615828EDHEC Business School 407 GRAPHS AND TABLESTable 3BSE - M1t-Statistic ProbADF test statistic -2875518 041Test critical values 1 level -25737845 level -194203510 level -1615894Table 4BSE - PPIt-Statistic ProbADF test statistic -2399055 162Test critical values 1 level -25737845 level -194203510 level -1615894Table 5BSE - SP500t-Statistic ProbADF test statistic -1427184 1430Test critical values 1 level -25737845 level -1942035

10 level -1615894EDHEC Business School 417 GRAPHS AND TABLESTable 6BSE - USDINRt-Statistic ProbADF test statistic -1659522 917Test critical values 1 level -25738185 level -19420410 level -1615891EDHEC Business School 427 GRAPHS AND TABLES744 Johansen cointegration testEDHEC Business School 437 GRAPHS AND TABLESTable 2EDHEC Business School 447 GRAPHS AND TABLESTable 3EDHEC Business School 457 GRAPHS AND TABLES745 Impulse response testsEDHEC Business School 467 GRAPHS AND TABLESEDHEC Business School 477 GRAPHS AND TABLESEDHEC Business School 487 GRAPHS AND TABLES746 Granger causality test between IP and BSEEDHEC Business School 498 BIBLIOGRAPHY8 BibliographyEugene F Fama Ination Output and Money Journal of Business 1982Eugene F Fama Stock Returns Real activity and Money The American Eco-nomic Review 1981Eugene F Fama Stock Returns Expected Returns and Real activity Journal ofFinance 1990Pal and Mittal Impact of macroeconomic indicators in Indian capital marketsJournal of Risk Finance 2011Shahid Ahmed Aggregate Economic Variables and Stock Markets in India In-ternational Research Journal of Finance and Economics 2008Sahu and Dhiman Correlation and Causality between Stock Market and MacroEconomic Variables in India An Empirical Study 2010 International Confer-ence on E-Business and Economics 2011Mohammad Bayezid Ali Impact of Micro Variables on Emerging Stock MarketReturn A case on Dhaka Stock Exchange (DSE) Interdisciplinary Journal ofResearch in Business 2011Napphon Tangjitprom Macroeconomic Factors of Emerging Stock Market Theevidence from Thailand International Journal of Finance and Research 2012Sayed Mehdi Hosseini The Role of Macroeconomic Variables on Stock MarketIndex in China and India International Journal of Economics and Finance2011John Y Campbell Pitfalls and Opportunities What Macroeconomists shouldknow about Unit Roots NBER Working Papers 1991Hacker and Hatemi The properties of Procedures Dealing with Uncertainityabout Intercept and Deterministic Trend in Unit Root Testing CESIS Elec-tronic Working Papers 2010Elder and Kennedy Testing for Unit Roots What should Students be TaughtNasseh and Strauss Stock Prices and domestic and international macroeco-

nomic activity a cointegration approach The Quarterly Review of Economicsand Finance 2000Engle and Granger Co-Integration and Error Correction Representation Es-timation and Testing Econometrica 1987Eugene F Fama Stock Returns Real Activity Ination and Money 1981American Economic AssociationNaliniprave Tripathy Causal Relationship between Macro-Economic Indicatorsand Stock Market in India Asian Journal of Finance and Accounting 2011Rogalski and Vinso Stock Returns Money Supply and the Direction of Causal-ity The Journal of Finance 1977James et al A VARMA Analysis of the Causal Relations Among Stock Re-turns Real output and Nominal Interest Rates 1985 The Journal of FinanceBailey and Chung Risk and return in the Philippine Equity market A multi-factor exploration Paci_c-Basin Finance Journal 1996Nai-Fu Chen Financial Investment opportunities and the Macroeconomy TheJournal of Finance 1991GB Wickremasinghe Macroeconomic forces and stock prices Some empiricalevidence from an emerging stock markets University of Wollongong 2006EDHEC Business School 508 BIBLIOGRAPHYYao Juo and Loh On Chinas Monetary Policy and Asset Prices University ofNottingham- China policy Institute 2011Bilson et al Selecting macroeconomic variables as explanatory factors of emerg-ing stock market returns Paci_c-Basin Finance Journal 2001CHen Roll and Ross Economic forces and the Stock Markets The Journal ofBusiness 1986William H Greene Econometric Analysis 6th Edition Pearson InternationalEditionRuey Tsay Analysis of Financial Time seriesChris Brooks Introductory Econometrics for Finance Cambridge PublicationsEDHEC Business School 51

M1EDHEC Business School 397 GRAPHS AND TABLES74 Tables741 Table for Unit root test of Time seriesVariables T-Stat p-valueBSE -2671 2495 SP500 -1315 8818 CPI -1909 6466 IP -1669 899 M1 -2420 3679 PPI -3353 601 USDINR -2955 1469 742 Tables for Unit root test of Di_erenced time seriesVariables T-Stat p-valueBSE -13848 000 SP500 -14832 000 CPI -3344 140 IP -3865 027 M1 -3867 026 PPI -9656 000 USDINR -13701 000 743 Tables for Residual based test of cointegrationTable 1BSE - CPIt-Statistic ProbADF test statistic -2622676 087Test critical values 1 level -25738185 level -19420410 level -1615891Table 2BSE - IPt-Statistic ProbADF test statistic -3738802 002Test critical values 1 level -25745135 level -194213610 level -1615828EDHEC Business School 407 GRAPHS AND TABLESTable 3BSE - M1t-Statistic ProbADF test statistic -2875518 041Test critical values 1 level -25737845 level -194203510 level -1615894Table 4BSE - PPIt-Statistic ProbADF test statistic -2399055 162Test critical values 1 level -25737845 level -194203510 level -1615894Table 5BSE - SP500t-Statistic ProbADF test statistic -1427184 1430Test critical values 1 level -25737845 level -1942035

10 level -1615894EDHEC Business School 417 GRAPHS AND TABLESTable 6BSE - USDINRt-Statistic ProbADF test statistic -1659522 917Test critical values 1 level -25738185 level -19420410 level -1615891EDHEC Business School 427 GRAPHS AND TABLES744 Johansen cointegration testEDHEC Business School 437 GRAPHS AND TABLESTable 2EDHEC Business School 447 GRAPHS AND TABLESTable 3EDHEC Business School 457 GRAPHS AND TABLES745 Impulse response testsEDHEC Business School 467 GRAPHS AND TABLESEDHEC Business School 477 GRAPHS AND TABLESEDHEC Business School 487 GRAPHS AND TABLES746 Granger causality test between IP and BSEEDHEC Business School 498 BIBLIOGRAPHY8 BibliographyEugene F Fama Ination Output and Money Journal of Business 1982Eugene F Fama Stock Returns Real activity and Money The American Eco-nomic Review 1981Eugene F Fama Stock Returns Expected Returns and Real activity Journal ofFinance 1990Pal and Mittal Impact of macroeconomic indicators in Indian capital marketsJournal of Risk Finance 2011Shahid Ahmed Aggregate Economic Variables and Stock Markets in India In-ternational Research Journal of Finance and Economics 2008Sahu and Dhiman Correlation and Causality between Stock Market and MacroEconomic Variables in India An Empirical Study 2010 International Confer-ence on E-Business and Economics 2011Mohammad Bayezid Ali Impact of Micro Variables on Emerging Stock MarketReturn A case on Dhaka Stock Exchange (DSE) Interdisciplinary Journal ofResearch in Business 2011Napphon Tangjitprom Macroeconomic Factors of Emerging Stock Market Theevidence from Thailand International Journal of Finance and Research 2012Sayed Mehdi Hosseini The Role of Macroeconomic Variables on Stock MarketIndex in China and India International Journal of Economics and Finance2011John Y Campbell Pitfalls and Opportunities What Macroeconomists shouldknow about Unit Roots NBER Working Papers 1991Hacker and Hatemi The properties of Procedures Dealing with Uncertainityabout Intercept and Deterministic Trend in Unit Root Testing CESIS Elec-tronic Working Papers 2010Elder and Kennedy Testing for Unit Roots What should Students be TaughtNasseh and Strauss Stock Prices and domestic and international macroeco-

nomic activity a cointegration approach The Quarterly Review of Economicsand Finance 2000Engle and Granger Co-Integration and Error Correction Representation Es-timation and Testing Econometrica 1987Eugene F Fama Stock Returns Real Activity Ination and Money 1981American Economic AssociationNaliniprave Tripathy Causal Relationship between Macro-Economic Indicatorsand Stock Market in India Asian Journal of Finance and Accounting 2011Rogalski and Vinso Stock Returns Money Supply and the Direction of Causal-ity The Journal of Finance 1977James et al A VARMA Analysis of the Causal Relations Among Stock Re-turns Real output and Nominal Interest Rates 1985 The Journal of FinanceBailey and Chung Risk and return in the Philippine Equity market A multi-factor exploration Paci_c-Basin Finance Journal 1996Nai-Fu Chen Financial Investment opportunities and the Macroeconomy TheJournal of Finance 1991GB Wickremasinghe Macroeconomic forces and stock prices Some empiricalevidence from an emerging stock markets University of Wollongong 2006EDHEC Business School 508 BIBLIOGRAPHYYao Juo and Loh On Chinas Monetary Policy and Asset Prices University ofNottingham- China policy Institute 2011Bilson et al Selecting macroeconomic variables as explanatory factors of emerg-ing stock market returns Paci_c-Basin Finance Journal 2001CHen Roll and Ross Economic forces and the Stock Markets The Journal ofBusiness 1986William H Greene Econometric Analysis 6th Edition Pearson InternationalEditionRuey Tsay Analysis of Financial Time seriesChris Brooks Introductory Econometrics for Finance Cambridge PublicationsEDHEC Business School 51

10 level -1615894EDHEC Business School 417 GRAPHS AND TABLESTable 6BSE - USDINRt-Statistic ProbADF test statistic -1659522 917Test critical values 1 level -25738185 level -19420410 level -1615891EDHEC Business School 427 GRAPHS AND TABLES744 Johansen cointegration testEDHEC Business School 437 GRAPHS AND TABLESTable 2EDHEC Business School 447 GRAPHS AND TABLESTable 3EDHEC Business School 457 GRAPHS AND TABLES745 Impulse response testsEDHEC Business School 467 GRAPHS AND TABLESEDHEC Business School 477 GRAPHS AND TABLESEDHEC Business School 487 GRAPHS AND TABLES746 Granger causality test between IP and BSEEDHEC Business School 498 BIBLIOGRAPHY8 BibliographyEugene F Fama Ination Output and Money Journal of Business 1982Eugene F Fama Stock Returns Real activity and Money The American Eco-nomic Review 1981Eugene F Fama Stock Returns Expected Returns and Real activity Journal ofFinance 1990Pal and Mittal Impact of macroeconomic indicators in Indian capital marketsJournal of Risk Finance 2011Shahid Ahmed Aggregate Economic Variables and Stock Markets in India In-ternational Research Journal of Finance and Economics 2008Sahu and Dhiman Correlation and Causality between Stock Market and MacroEconomic Variables in India An Empirical Study 2010 International Confer-ence on E-Business and Economics 2011Mohammad Bayezid Ali Impact of Micro Variables on Emerging Stock MarketReturn A case on Dhaka Stock Exchange (DSE) Interdisciplinary Journal ofResearch in Business 2011Napphon Tangjitprom Macroeconomic Factors of Emerging Stock Market Theevidence from Thailand International Journal of Finance and Research 2012Sayed Mehdi Hosseini The Role of Macroeconomic Variables on Stock MarketIndex in China and India International Journal of Economics and Finance2011John Y Campbell Pitfalls and Opportunities What Macroeconomists shouldknow about Unit Roots NBER Working Papers 1991Hacker and Hatemi The properties of Procedures Dealing with Uncertainityabout Intercept and Deterministic Trend in Unit Root Testing CESIS Elec-tronic Working Papers 2010Elder and Kennedy Testing for Unit Roots What should Students be TaughtNasseh and Strauss Stock Prices and domestic and international macroeco-

nomic activity a cointegration approach The Quarterly Review of Economicsand Finance 2000Engle and Granger Co-Integration and Error Correction Representation Es-timation and Testing Econometrica 1987Eugene F Fama Stock Returns Real Activity Ination and Money 1981American Economic AssociationNaliniprave Tripathy Causal Relationship between Macro-Economic Indicatorsand Stock Market in India Asian Journal of Finance and Accounting 2011Rogalski and Vinso Stock Returns Money Supply and the Direction of Causal-ity The Journal of Finance 1977James et al A VARMA Analysis of the Causal Relations Among Stock Re-turns Real output and Nominal Interest Rates 1985 The Journal of FinanceBailey and Chung Risk and return in the Philippine Equity market A multi-factor exploration Paci_c-Basin Finance Journal 1996Nai-Fu Chen Financial Investment opportunities and the Macroeconomy TheJournal of Finance 1991GB Wickremasinghe Macroeconomic forces and stock prices Some empiricalevidence from an emerging stock markets University of Wollongong 2006EDHEC Business School 508 BIBLIOGRAPHYYao Juo and Loh On Chinas Monetary Policy and Asset Prices University ofNottingham- China policy Institute 2011Bilson et al Selecting macroeconomic variables as explanatory factors of emerg-ing stock market returns Paci_c-Basin Finance Journal 2001CHen Roll and Ross Economic forces and the Stock Markets The Journal ofBusiness 1986William H Greene Econometric Analysis 6th Edition Pearson InternationalEditionRuey Tsay Analysis of Financial Time seriesChris Brooks Introductory Econometrics for Finance Cambridge PublicationsEDHEC Business School 51

nomic activity a cointegration approach The Quarterly Review of Economicsand Finance 2000Engle and Granger Co-Integration and Error Correction Representation Es-timation and Testing Econometrica 1987Eugene F Fama Stock Returns Real Activity Ination and Money 1981American Economic AssociationNaliniprave Tripathy Causal Relationship between Macro-Economic Indicatorsand Stock Market in India Asian Journal of Finance and Accounting 2011Rogalski and Vinso Stock Returns Money Supply and the Direction of Causal-ity The Journal of Finance 1977James et al A VARMA Analysis of the Causal Relations Among Stock Re-turns Real output and Nominal Interest Rates 1985 The Journal of FinanceBailey and Chung Risk and return in the Philippine Equity market A multi-factor exploration Paci_c-Basin Finance Journal 1996Nai-Fu Chen Financial Investment opportunities and the Macroeconomy TheJournal of Finance 1991GB Wickremasinghe Macroeconomic forces and stock prices Some empiricalevidence from an emerging stock markets University of Wollongong 2006EDHEC Business School 508 BIBLIOGRAPHYYao Juo and Loh On Chinas Monetary Policy and Asset Prices University ofNottingham- China policy Institute 2011Bilson et al Selecting macroeconomic variables as explanatory factors of emerg-ing stock market returns Paci_c-Basin Finance Journal 2001CHen Roll and Ross Economic forces and the Stock Markets The Journal ofBusiness 1986William H Greene Econometric Analysis 6th Edition Pearson InternationalEditionRuey Tsay Analysis of Financial Time seriesChris Brooks Introductory Econometrics for Finance Cambridge PublicationsEDHEC Business School 51