How Does Globalization Affect the Synchronizationof Business Cycles?

Bv M. AYHAN KOSE, ESWAR S. PRASAD, AND MARCO E . TERRONES*

The phenomenon of globalization, which re-fers to the rising trade arid financial integrationof the world economy, has gathered steam inrecent decades. The grov/th rate of world tradehas been greater than that of world output inalmost all years since I960, and the cumulativeincrease in the volume of world trade is almostthree times larger than that of world output overthis period. A more dramatic element in theprocess of globalization has been the surge incross-border capital flows over tlie last two de-cades. Since the eariy 198O's, gross capitalflows have jumped from less than 5 percent toapproximately 20 percent of GDP for advancedcountries. For emerging markets, gross capitalflows have increased almost fourfold over thesame period and now account for roughly 5percent of GDP in these economies.'

What is the impact of globalization on thesynchronization of business cycles across coun-tries? In this paper, we attempt to address thisquestion by systematically examining the im-pact of increased trade and financial integrationon intemational business-cycle comovements.In particular, we analyzi^ the pattems of cor-relations for industrial as well as developingcountries within a unified empirical framework.We also examine the effects of different aspectsof globalization on output as well as consump-tion comovement across countries.

I. What Do We Learn from Economic Theory?

Economic theory does not provide definitiveguidance conceming the impact of increased

* International Monetary Fund, 700 19th Street, N.W.Washington, DC 20431 (e-maib;: akose@inif.org, eprasad@imf.org, and mterrones@iinf.oig, respectively). We thankCesar Calderon, Frank Diebold, Hideaki Hirata, and HenryKim for useful suggestions. The views presented are thoseof the authors and do not necessarily reflect the views of theIMF or IMF policy.

' Prasad et al. (2003) examine the increase in globaltrade and financial linkages.

trade and financial linkages on the degree ofbusiness-cycle synchronization." Intemational tradelinkages generate both demand- and supply-sidespillovers across countries. For example, on thedemand side, an investment or consumptionboom in one country can generate increased demandfor imports, boosting economies abroad. Throughthese types of spillover effects, stronger intema-tional trade linkages can result in more highlycorrelated business cycles across countries. How-ever, trade fiows could also induce increased spe-cialization of production resulting in changes inthe nature of business-cycle correlations. If stron-ger trade linkages are associated with increasedinterindustry specialization across countries, andindustry-specific shocks are important in drivingbusiness cycles, then uitemational business-cyclecomovement might be expected to decrease.

Financial linkages could result in a higherdegree of business-cycle synchronization bygenerating large demand side effects. For in-stance, if consumers from different countrieshave a significant fraction of their investmentsin a particular stock market, then a decline inthat stock market could induce a simultaneousdecline in the demand for consumption andinvestment goods in these countries. Further-more, contagion efTects that are transmittedthrough financial linkages could also resuit inheightened cross-coimtry spillovers of macro-economic fluctuations.

Intemational financial linkages could stimu-late specialization of production through thereallocation of capital in a manner consistentwith countries' comparative advantage in theproduction of different goods. Such specializa-tion of production, which could result in moreexposure to industry- or country-specific shocks,

^ See Kose and Yi (2002) for a discussion about thetiieoredcal impact of increasing trade integration on business-cycle comovement. Heathcote and Perri (2002) examine theimplications of increasing financial linkages on cross-countrybusiness-cycle correlations.

57

ss. AEA PAPERS AND PROCEEDINGS MAY 2003

would typically be expected to be accompaniedby the use of intemational financial markets todiversify consumption risk. This implies thatfinancial integration, in particular, should resultin stronger comovement of consumption acrosscountries. This effect would be expected to bestronger for developing countries that are typi-cally less diversified in terms of their endow-ment and production structures and haveintrinsically more volatile output, implying thattheir potential gains from international risk-sharing are even greater than for industrialcountries.

n. Data and Methods

Our empirical analysis is based on annualdata over the period 1960-1999 for a sample of76 countries; 21 industrial and 55 developing.^Per capita real GDP and real private consump-tion constitute the measures of national outputand consumption, respectively. We use twomeasures of trade openness; a (binary) measureof restrictions on current account transactionsand a standard openness ratio (ratio of importsand exports to GDP). To measure financial in-tegration, we use an indicator measure of re-strictions on capital-account transactions andalso a measure of accumulated gross capitalflows to GDP, where the latter is analogous tothe trade openness ratio. The restrictiveness in-dicators can be considered as measures of dejure trade and financial openness, while themeasures based on flows capture de facto open-ness. This distinction is of particular importancein understanding the effects of financial integra-tion, since many countries that have maintainedcontrols on capital-account transacfions havefound them ineffective in preventing capital out-flows. Furthermore, the de jure measure cannotfully capture differences in the degree of finan-cial integration across countries and over time.

While it would be interesting to examine theeffects of bilateral trade and financial linkageson correlations across country pairs, obtainingsuch data on financial linkages is not feasible. Inthis paper, therefore, we adopt a simpler ap-

^ For a detailed description of the data and sources seeKose et al. (2003a).

proach of examining correlations of individualcountry output and consumption growth fluctu-ations with the fluctuations of corresponding'"world" aggregates. To minimize the effects ofthe large economies on the results, we use pur-chasing power parity (PPP)-weighted aggre-gates of output and consumption in the G-7countries as measures of the relevant worldaggregates. These countries are then excludedfrom the empirical analysis that follows. Theuse of G-7 aggregates has some additional ad-vantages. Since the G-7 countries account for asub.stantial fraction of financial and trade flowsto developing countries, correlations with theG-7 aggregates are most relevant for under-standing the effects of integration on business-cycle comovements. In any case, as one wouldexpect, cyclical fiuctuations in the G-7 countriesare highly correlated with fluctuations in totalworld output.

We begin by presenting a descriptive analysisof changes in patterns of correlations of differ-ent groups of countries with the world businesscycle. For this part of the analysis, we dividedeveloping countries into two groups; more fi-nancially integrated (MFI) economies and lessfinancially integrated (LFT) economies. The formeressentially constitute the group of "emergingmarkets" and account for a substantial fractionof net capital flows from industrial to develop-ing countries in recent decades. Since outputand consumption are nonstationary series, andin order to avoid the complications with stan-dard filtering methods, we use growth rates ofthe variables in the empirical analysis.

UI. Correlations

We first examine some summary statistics onthe correlations of output growth rates in eachcountry with the growth rate of the compositemeasure of world output. Table lA shows that,on average, industrial countries have strongercorrelations with world output than do develop-ing economies. For industrial countries, thesecorrelations on average increase sharply in the197O's (the oil-shock period) and rise further inthe 199O's. For developing countries, on theother hand, these correlations are in generalmuch lower compared to industrial countriesand, if anything, decline in the 199O's. In fact.

VOL. 93 NO. 2 INTERNATIONAL TRANSMISSION OF BUSINESS CYCLES 59

TABLB 1—CORRELATIONS WITH "WORLD"MACROECONOMIC AGGREGATES

(MEDIANS FOR EACH GROUP)

Sample

A. Oiilpiit

Full sample

Industrial countries

Developing countries

MFI ecDnoniies

LFI economies

B. Consumption

Full sample

industrial eouniries

Developing countries

MR economies

LFI economies

1^^60-199')

0.17(0,06)o.4y

10,05)0.11

(0.04)0.17

10.07)0,07

(0,09)

0,10(0,04)0,45

(0,06)0.02

(0.03)0.04

(0,06)0.01

10.05)

l960's

0,12(0.05)0,05

(0,15)0,12

(0,05)0.12

(0,11)0,12

(0.12)

0.06(0.06)

-0.02(0,16)0.06

(0.09)0,14

(0,12)0,06

(0,13)

l')70\

o.ia(0.07)0.3.1

(0.08)0.14

(0.08)0.26

(0.14)0,11

(0,08)

0.09(0.07)0,34

(0.12)-0.01(0.09)

-0.04(0-11)0.01

(0.14)

l9S0"s

0,15(0,05)0.35

(0.13)0,11

(0,07)0.10

(0,09)O.ll

10.10)

0,07(0,05)0.35

(0,10)0.04

(0,04)-0,00(0.10)0.04

(0,05)

l9Ws

0.09(0.12)0.58

(0,09)-0,06(0.07)

-0.18(0,12)0.02

(0.15)

0.01(0.08)0,50

(0.07)-0.09(0,06)

-0,26(O.ll)

-0,07(0,08)

Notes: The numbers shown tn this table are medians, withineach group, of the correlations for each country with thecorresponding world aggregate. Standard errors are shownin parenth&ses.

for MFI economies, these correlations becomenegative during this period. Thus, there is littleevidence, in terms of these coarse countrygroupings, that business-cycle comovementshave on average become more synchronized ata global level during the most recent period ofglobalization.

Table IB indicates that the temporal evolu-tion of consumption correlations is quite similarto that of output. In addition,, as has been doc-umented by severai other authors, consumptioncorrelations are typically smaller than outputcorrelations. A particularly interesting result isthat, for MFI economies, the average correlationturns significantly negative in the 199O's. Thisseems at odds with the notion that financialintegration should have helped these economiesto better share consumption risk.

While cross-country correlations of outputand consumption are useful in understandingthe degree of synchronization, they only capturethe contemporaneous dimension of business-cycle comovement. To further study the extentof and the change in the degree of synchroni-zation, we estimate dynamic unobserved factor

models. This approach allows us to decomposefluctuations in each macroeconomic aggregateinto a common factor (common across all coun-tries) and a country-specific factor."̂ We exam-ine changes in the relative importance of thecommon factor by estimating the model overtwo periods: 1960-1980 and 1981-1999.^ Ifglobalization has a positive impact on the de-gree of business-cycle synchronization overtime, the contribution of the common factor tothe variation of output and consumption growthshould rise in the second period.

Table 2 presents the median (within eachgroup of countries) of the fraction of the vari-ance of output and consumption fluctuationsexplained by the common factor, for the fullsample as well as the two subperiods. There arefour results to note. First, the common factoraccounts for less than 10 percent of the variationin output and consumption fluctuations acrossthe full sample of industrial and developingeconomies. Second, the importance of the com-mon factor for output fluctuations has notchanged much across the two subperiods, sug-gesting that integration has not significantlychanged the extent of business-cycle comove-ment. Third, the common factor explains amuch laiger fraction of output and consumptionfluctuations in industrial countries than it doesin the developing countries. Moreover, for in-dustrial countries, there has been a noticeableincrease in the share of viiriance of consumptionfluctuations explained by the common factor inthe second period. Fourth, on average, theglobal common factor has played only a verysmall role in explaining the variance of outputand consumption fluctuations in the MFI andLFT economies, and the importance of the com-mon factor has not changed much over time ineither group. Overall, the results from the factormodel estimates reveal a picture similar to thatobtained from the simple correlations.

" Our estimations of dynamic factor models closely fol-low Christopher Otrok and Charies H. Whiteman (1998).Robin Lumsdaine and Prasad (2003) employ a differentmethod and estimate a common factor using the data ofOECD countries to sludy the dynamics of internationalbusiness cycles.

^ We estimated factor models using shorter sample pe-riods: however, the results are not very informative sinceshorter sample periods result in less precisely estitnatedparameters.

AEA PAPERS AND PROCEEDINGS MAY 2003

TABLE 2—-SHARE OF VARIANCE EXPLAINEDBY THE COMMON FACTOR (PERCENTAGES)

1960-1999 1%0-1980 1981-1999

A, Oulput

Full sample

TABLE 3—DETERMINANTS OF CORRELATIONS

Variable OLS OLS IV

lodustriaJ countries

Developing countries

MFI economies

LFI economies

B. Consumption

Full sample

Industrial countries

Developing countries

MFI economies

LFI economies

9.1(2.1)41.4(4.7)4,6

(1-2)5,1

(1.9)4.6

(1.5)

5.0(1.8)27.1(4,1)2,9

(0,9)1.9

(1.4)3,0

(1.1)

7,2(i.7)30.5(3.8)4.7

(1-0)3.6

(2,0)5.9

(1.0)

6.4(1.2)16.0(2.9)3.3

fO.9)3.1

(1.3)3.8

(1.3)

5.6(1.6)27.2(3,7)2.9

(1.1)3.2

(1-0)2.8

(1.7)

5.6(1.5)22.7(3.9)2,3

(1.0)2,1

(2.0)2.5

(i.l)

Notes: All data series are transformed into logarithms, firstdifferenced, and demeaned before the estimations. In eachcell, the median fraction of variance explained hy the com-mon factor in each group is reported. The sample standarderTors are shown in parentheses.

IV. Regression Analysis

In this section, we present a more formalregression analysis of the factors that influencecorrelations of individual country macroeco-nomic aggregates with the corresponding worldaggregates. We use non overlapping ten-yeai'correlations as the dependent variable. The firstcolumn of Table 3 shows the results of ordinaryleast-squares (OLS) regressions for output. Inthis table, we present coefficient estimates foronly the main variables of interest.^ Trade open-ness appears to have a weak negative effect onoutput correlations. While this result could beexplained as a consequence of more open econ-omies being more vulnerable to external shocks,it appears to run counter to other studies suggest-ing that trade linkages increase cross-country

^ We provide more detailed regression resuits and ro-bustness tests in Kose el al. (2003b).

A. Oiiipui

Current accountresirictions

Trade openness

Trade linkages wiiii G-7

Capital-acctiunl rcstrietions

Financial openness

Relative income

Terms-of-irade volatility

Volatility of fiscal impulse

Regional dummies

Number of observaiions;

B. Consitmpiion

Current-accountrestrictions

Trade openness

Trade linkages wilh G-7

Capiial-account restrictions

Financial openness

Relative income

Terms-of-trade volatility

Volatility of fiscal impulse

Regional dummies

Number of observations:

-0.0130(0.0706)

-0,00211(0.0013)0.0050*

(0.0019)-0.1859*(0.0767)

(0.0023)0.1846

(O.I2O6i-0.7680*(0.2466)

-0.1768(0.1162)

0.15235

-0.0480(0.0741)

-0,0006(0.0013)0.0047*

(0.0019)0.0639

(0.0927)0.0028

(0.0020)0.3033*

(0.1197)

- 0 . 8 7 7 1 *(0.3310)

-0.5247*(0.0521)

0,17235

0.0257(0.0715)

-0.0013(0.0014)0.0046*

(0.0021)-0.2051*'(0.0862)0.0009

(0.0024)-0.0879(0.2098)

-0.7127*(0.2513)

-0.1519

yes

(UB235

-6.0262(0.0765)0.0005

(0.0013)0.00381

(0.0019)0.0433

(0.0956)0.0021

(0.002!)-0.0153(0.2003)

-0.8679*(0.3296)

-0.4971*(0.05571

yes

0.19235

0,0496(0,0753)

-0.0006(0.0023)0.0107*

(0.0045)-0.2054*(0.1179)

-0.0013(0.0088)

-0.0754(0.2497)

-0.7394*(0.2821)

-0.1522(0.1633)

yes

229

-O.()314(0.0793)0.0001

(0.0021)0.0017

(0,0048)0.0325

(0.1184)0.0009

(0.0075)

-0.0065(0.2399)

-0.8975*(0.3527)

-0.4999*(0.0627)

yes

229

Notes: The dependent variable is the correlation for eachcountr>''s outpui or consumplion with the correspondingworid aggregate over each ten-year period. Time dummiesare included in all regressions. For the instrumental-van abies(IV) regressions, the instruments include relative income(vs. the United States) as of 1960, share.s of agriculture andmanufacturing in total outpu! in l9fiO, a weighted conflictindex, and dummies for oil-exporting countries, landlockedcountries, conntries in tropical climates, existence ofmultiple exchange rate,s. and existence of export surrenderrequirements. Robust standard errors are reported in paren-theses. Additional controls that we experimented with andthai did not affect these results include the ratio of themoney supply (1V12) to GDP and its standard deviation,manufacturing output as a share of GDP, inflation, fuelexports as a share of total exports, and an indicator of theexchange-rate regime.

t Statistically significant at the 10-percent level.* Statistically significant at the 5-percent level.

VOL 93 NO, 2 INTERNATIONAL TRANSMISSION OF BUSINESS CYCLES 61

business-cycle correlations. Our regressions alsoinclude a variable that measures direct traderelationships with the G-7 (trade with the G-7 asa share of a country's total trade). The positivecoefficient on this variable indicates that tradelinkages do indeed have a positive effect onoutput correlations. Thus, while trade opennessby itself seems to reduce domestic business-cycle correlations with world output, a measureof trade linkages does reveal evidence of cross-country business-cycle transmission via thetrade channel.

The capital-account restrictions measure en-ters with a negative coefficient. In other words,countries with restricted capital flows havelower business-cycle correlations with worldoutput. This result suggests, as expected, thatfinancial linkages are more important, in termsof business-cycle transmission, for economiesthat are more open to capital flows. However,the measure of actual gross capital flows (finan-cial openness) does not reflect this effect. Un-fortunately, data limitations prevented us frombeing able to construct a measure of financiallinkages with the G-7 in a manner analogous tothe trade linkage variable. It should be noted,however, that the G-7 countries account formore than two-thirds of all private capital flowsand, in recent years, an even greater fraction offlows to developing countries.

Among other variables that are included inthe regressions, only the volatility of the termsof trade has a statistically significant coefficient.The negative coefficient on this variable is con-sistent with the earlier result on trade opennessand indicates that economies that are subject tomore volatile terms-of-trade shocks are less cor-related with world output.

How robust are these results? We eschew theuse of fixed-effects estimators in order to avoidrestricting the empirical analysis to within-countrychanges in volatility. Most of the variation inour sample comes from the between-countrycomponent, which is of far more relevance forthe issues of interest in this analysis. Instead ofusing fixed effects, we examined the sensitivityof the results to the inclusion of regional dum-mies (column 2 in Table 3) as well as numerouscountry-specific variables (reflecting politicaland economic structures and other relevant in-stitutional features) that are potentially impor-tant for explaining cross-country differences in

correlation pattems. In general, these variablesdid not affect our main results except for thecoefficient on trade openness, which is nolonger statistically significant.

A more important concem with the baselineresults is that of possible endogeneity. Measure-ment error in the integration variables is also apotential problem. To address these concerns, weinstrumented for the trade and financial-integrationvariables. The results of these instrumental-variable(IV) regressions are shown in the last column ofTable 3. Again, the main results are preserved. Inparticular, the coefficients on the trade-linkagesvariable and the capital-account restrictiveness in-dicator remain significant.

Our findings are generally in line with theresults of other recent studies. For instance,Glenn Otto et ai. (2001) and Jean Imbs (2002)find that trade and financial linkages are im-portant in accounting for business-cycle co-movement among OECD economies.' Imbsalso finds that specialization patterns drivebusiness-cycle correlations. We introduced somebroad sectoral measures (agriculture and manu-facturing output as shares of GDP) in ourregressions to capture these effects, but thesewere not statistically significant in any ofthespecifications.

We now turn to regressions for consumptioncorrelations (Table 3B). The results for thisvariable are weaker. Of the integration vari-ables, only the trade linkages with the G-7 mat-ter, and these appear to have a positive effect oncross-country movements in consumption. Eventhis result is not robust to IV estimation. Thus,there is little evidence that globalization hasinfluenced consumption comovements acrosscountries. This is consistent with other researchshowing that imperfections in intemational cap-ital markets have thus far thwarted the use ofthese markets for effectively sharing risk acrosscountries and reducing within-country con-sumption volatility (see Kose et al., 2003a).

Terms-of-trade volatility affects consumptioncomovement in a manner similar to that ofoutput. Our findings regarding the importanceof tenms-of-trade volatility are consistent with

'̂ Cesar A. Caldercn et al. (2002) find that trade linkagesplay a more important role in explaining business-cyclecomovement in advanced countries than in developingcountries.

AEA PAPERS AND PROCEEDINGS MAY 2003

the results of several recent studies (see Kose etal., 2003b). Interestingly, the coefficient on thefiscal impulse volatility measure is negative,perhaps indicating the use of fiscal policies as acountercyclical tool that dampens the effects ofglobal shocks. An alternative interpretation, ofcourse, is that fiscal policies exacerbate country-specific fluctuations.

To examine the question of whether trade andfinancial integration have differential effects oncorrelation pattems for industrial versus devel-oping countries, we included interactions of theintegration variables with an industrial-countrydummy. While the results discussed above werenot materially affected, the effects of the tradeand financial linkages on output correlationsappeared to be stronger for industrial countriesthan for the developing countries in our sample.In addition, for these countries, there is someevidence that consumption correlations with theworid aggregate are significantly positively af-fected by both trade and financial linkages. Al-though not reported in detail here, we tested thesensitivity of our regression results to the inclu-sion of a large number of additional controls.The main results shown here appeared to bequite robust in these experiments. We also didnot find any obvious evidence of threshold ef-fects or nonlinearities in the relationships thatwe have documented in this paper, although theresults with the industrial-country dummy inter-action terms noted above suggest that such ef-fects remain a possibility.

V. Conclusions

The results in this paper provide at best lim-ited support for the conventional wisdom thatglobalization leads to an increase in the degreeof synchronization of business cycles. We foundsome evidence for the proposition that tradeand financial-market integration enhance globalspillovers of macroeconomic fluctuations. Onestriking result is that, on average, consumptioncorrelations have not increased in the 199O's,precisely when financial integration would havebeen expected to result in better risk-sharingopportunities, especially for developing countries.

While this paper has provided a number ofpreliminary results, richer data sets and morerigorous estimation methods are needed to im-prove our understanding of the effects of glob-

alization, which has important implications forthe conduct of macroeconomic policies in anincreasingly integrated global economy.

REFERENCES

Calderon, Cesar A.; Chong, Alberto E. and Stein,Ernesto H. "Trade Intensity and Business Cy-cle Synchronization: Are Developing Coun-tries Any Different?" Mimeo, Central Bankof Chile, Santiago, 2002.

Heathcote, Jonathan and Perri, Fabrizio. "Finan-cial Globalization and Real Regionalization."National Bureau of Economic Research(Cambridge, MA) Working Paper No. 9292,October 2002.

Imbs, Jean. "Trade, Finance, Specialization andSynchronization." Mimeo, London BusinessSchool, London, U.K., 2002.

Kose, M. Ayhan; Prasad, Eswar and Terrones,Marco E. "Financial Integration and Macro-economic Volatility." IME Staff Papers, In-temational Monetary Fund, Washington, DC,2003a (forthcoming).

. "Globalization and Business CycleComovement." Mimeo, Tntemational Mone-tary Fund, Washington, DC, 2003b.

Kose, M. Ayhan and Yi, Kei-Mu. "The TradeComovement Problem in Intemational Mac-roeconomics." Federal Reserve Bank of NewYork Staff Report No. 155, December 2002.

Lumsdaine, Robin and Prasad, Eswar. "Identify-ing the Common Component in IntemationalEconomic Fluctuations: A New Approach."Economic Journal, January 2003, I13(4H4),pp. 101-27.

Otrok, Christopher and Whiteman, Charles H."Bayesian Leading Indicators: Measuringand Predicting Economic Conditions inIowa." International Economic Review, No-vember 1998, 39(4), pp. 997-1014.

Otto, Glenn; Voss, Graham and Willard, Luke."Understanding OECD Output Correlations."Reserve Bank of Australia (Canberra) Re-search Paper No. 2001-05, September 2001.

Frasad, Eswar; Rogoff, Kenneth S.; Wei, Shang-Jin and Kose, M. Ayhan. "The Effects of Fi-nancial Globalization on Developing Countries—Some Empirical Evidence." IMF OccasionalPapers, Intemational Monetery Fund, Wash-ington, DC, 2003 (forthcoming).