Feldstein Horioka Review

8/3/2019 Feldstein Horioka Review

1/20

International Journal of Finance and EconomicsInt. J. Fin. Econ. 3: 169188 (1998)

SURVEY ARTICLE

The FeldsteinHorioka Puzzle andCapital Mobility: A Review

Jerry Coakleya,*, Farida Kulasib and Ron Smithca London Guildhall University, London, UKb Primark Decision Economics, London, UKc Birkbeck College, London, UK

This paper reviews how economists responded to the FeldsteinHorioka(FH) view that a high saving-investment association across OECD countriesimplied low capital mobility. This posed an uncomfortable puzzle since the

conventional wisdom in most exchange rate and open-economy macroeco-nomic models was that capital mobility was high. In the face of a variety ofreplications, the FH result of a high cross-section association between sav-ing and investment rates in OECD countries has remained remarkably ro-bust. The debate over whether saving-investment comovements areinformative about capital mobility is still unresolved although the scepticsappear to be in the ascendancy. 1998 John Wiley & Sons, Ltd.

KEY WORDS: saving-investment association; Feldstein Horioka puzzle; capital mobil-ity; current account

SUMMARY

This paper reviews the extensive literature on howeconomists responded to the Feldstein and Horioka(1980) (FH) claim that capital was relatively immo-

bile. They based this claim on the results of crosssection regressions of investment on saving (bothexpressed as shares of GDP) across 16 OECDcountries for the 196074 period. FH reasoned thatsaving and investment would be perfectly corre-lated in a closed economy but should be unrelatedin an open economy since saving could seek out thehighest global returns. The FH view of low capitalmobility posed an uncomfortable puzzle since theconventional wisdom embodied in most exchangerate and open-economy macroeconomic models

since the 1970s was that capital mobility was high.

The issue of capital mobility is significant for di-verse policy issues such as the single currencydebate within the EU or for questions of taxes oncapital and saving.

In this paper we have organized our review ofcontributions to the FH puzzle in relation to thegeneric comments suggested by Stigler (1977). Themajority of the models and explanations reviewedoppose the FH view of low capital mobility byattempting to construct models which reconcile ahigh saving-investment association with high phys-ical and financial capital mobility and/or by provid-

ing plausible econometric and other data-basedrefutations of the FH view. These include generalequilibrium, real business cycle models and, nota-

bly, intertemporal models of the current account.Against a background of ongoing debate we

draw some interim conclusions. Firstly, the FH

* Correspondence to: Department of Economics, London Guild-hall University, 84 Moorgate, London EC2M 6SQ, UK. E-mail:

[email protected]

CCC 10769307/98/020169-20$17.50 1998 John Wiley & Sons, Ltd.


2/20

J. Coakley et al.170

result of a high saving-investment association hasremained remarkably robust in OECD cross-sec-tions although the coefficient on saving has shownsome tendency to decline over recent years. Theresult persists in panels and average time-seriesand has been remarkably robust to the addition of

other variables and different estimation methodsin the OECD. However, there is less evidence for aclose relationship between saving and investmentin non-OECD samples, particularly in LDCs. Sec-ondly, using the FH regression for policy purposesis questionable because both saving and invest-ment are endogenous and the FH regression can-not distinguish exogenous shifts in saving fromendogenous shifts reflecting factors which mayalso impact on investment.

Thirdly, the FH result may not be informativeabout capital mobility since a range of theoreticalmodels can generate high saving-investment corre-lations even under perfect capital mobility. Thisnow seems to be the emerging consensus in theliterature. Nonetheless the FH puzzle is by nomeans resolved and recent contributions by Bay-oumi et al. (1996) and Sarno and Taylor (1996)establish high capital mobility within FH-relatedframeworks. Finally, the debate surrounding theFH puzzle has shown that the notion of capitalmobility itself is not analytically straightforward.Obstacles to the movement of financial, physicaland human capital may have quite different impli-cations. Assuming perfect capital mobility may

produce plausible conclusions in one set of modelsand not in another.

INTRODUCTION

Economists often use the term puzzle to refer toawkward empirical facts that refuse to complywith their established theoretical frameworks. Theequity premium puzzle of Mehra and Prescott(1985) is a well known example. In this survey wereview the responses to the puzzle posed byFeldstein and Horioka (1980). FH argued that,

under perfect capital mobility, there is no neces-sary association between national saving and in-vestment since saving can globally seek out thehighest returns. The implication is that an exoge-nous increase in investment in any country can befinanced by a perfectly elastic supply of global

funds. By contrast, zero capital mobility implies aone-to-one relationship between saving and in-vestment, since saving has to be invested domesti-cally. In this case we have a world of segmentedcapital markets in which each countrys interestrate is determined domestically and domestic

monetary and fiscal policies are relatively effec-tive.

In cross-section regressions for 16 OECD coun-tries FH failed to reject the null hypothesis of aone-to-one association between saving and invest-ment. They interpreted this as implying zero capi-tal mobility. This conclusion was unpalatable both

because most theoretical, open economy modelshad assumed perfect capital mobility and becauseit appeared that financial integration in the indus-trialized world was high and increasing, particu-larly since the introduction of floating exchange

rates in the early 1970s. The question of capitalmobility in contemporary economies is not anarcane issue since it is of critical import for policyissues such as saving and investment subsidies(Summers, 1988; Feldstein, 1994, 1995; Devereux,1996; Peeters, 1996) or the EU single currencydebate (Bayoumi et al., 1996).

The high saving-investment association is un-usual among stylized facts insofar as it is based onan econometric result. Regression coefficients arerarely sufficiently robust to become established asstylized facts. However the attempts to show thatthe FH result was fragile did not succeed as un-

derlined by the recent assessment of Baxter andCrucini (1993):

In the field of international macroeconomics, tempo-rally robust stylized facts are few and far between.One of the most stable regularities observed in thedata is the fact that national saving rates are highlycorrelated with national investment rates, both intime-series analyses of individual countries and incross sections in which each country is treated as asingle data point. High saving-investment correla-tions arise in small economies as well as largeeconomies (Ibid. p. 416).

While the high saving-investment association is

accepted, debate about its interpretation remainspolarized around one major issue: is the high FHcoefficient informative about capital mobility asFH argue? Although subsequent work has demon-strated that the FH result is consistent with perfectcapital mobility in a variety of theoretical models,

1998 John Wiley & Sons, Ltd. Int. J. Fin. Econ. 3: 169188 (1998)


3/20

The FeldsteinHorioka and Capital Mobility 171

a number of both theoretical and empirical contributions continues to offer support for the FHapproach. The resolution of this issue is not helped

by the fact that empirical econometric work issubject to a considerable degree of variability andis often interpreted within competing theoretical

frameworks. For instance, within the FH frame-work the consensus is that capital mobility hasincreased since the 1980s but is higher in particularsamples such as the EU. Such inferences are anath-ema to the many sceptics of the FH approach tocapital mobility.

The FH controversy puzzle is illuminating aboutthe various ways in which economists respond touncomfortable results. The classic compendium ofgeneric responses is given by Stigler (1977) whoprovided a numbered list of typical commentselicited by (conference) papers. It is striking howmany of the responses to FH fall within his cate-gories of criticism, and we have used his com-ments to organize the responses. We needed tomake only one addition, comment 33, Have youtested for unit roots and cointegration? We intro-duce sections with his relevant comment(s). In sodoing we do not intend to be postmodernist, ironicor relativistic1, since a less relativistic economistthan George Stigler is difficult to imagine. Ratherwe would emphasize that economists tend to useparticular lines of attack to confront difficult prob-lems and that Stigler identifies these lines withhumour.

The literature on the FH puzzle is enormousand, in mid-1996, shows no sign of abating. Theoriginal FH article was quoted some 142 times ineconomics and related journals between 1988 and1995.2 This article differs from related reviews inthat it is more narrowly focused on the FH debatethan the reviews of Frankel (1992) or Obstfeld(1995). It is organized as follows. The first sectionoutlines the FH framework in the context of com-peting approaches to capital mobility. The nextsection summarizes a variety of extensions, updat-ings, and tests of the FH result of a high saving-in-vestment association. The section after addresses

interpretational problems impinging on the valid-ity of the FH approach. The final section gives ourconclusions. Two caveats should be borne in mind.First, since the debate is ongoing, this review prob-ably already is or quickly will become incomplete.Second, many of the responses have depended on

subtle, technical points and in summarizing themwe may not have accorded them full justice.

FH REGRESSIONS AND CAPITAL

MOBILITY

S2. Unfortunately, there is an identification problemthat is not dealt with adequately in the paper.

Background and Definitions

The origins3 of the FH puzzle lie in the 1980Economic Journal article in which they estimatedcross section regressions of the form:

(I/Y)i=h+i(S/Y)i+ui i=1, 2, 3,N, (1)

where, I is national investment (public and pri-vate) by country i, S is national saving and Y isnational income. We call such relationships be-tween national investment and saving shares ofGDP (hereafter saving and investment unless oth-erwise specified), including variants such as firstdifferences, FH regressions and call i the FH coeffi-cient or the saving-investment association. The iden-tification problem relates to what exactly the FHcoefficient measures. FH interpreted the coefficientas an index of capital mobility. Their results for 16OECD countries for the 196074 period indicateda very high saving-investment association andthey could not reject the null hypothesis of i=1

or zero capital mobility. FH concluded:[a]lthough there may be perfect arbitrage of short-term yields and substantial flows of long-term directand portfolio investment, there appear to be sufficientrigidities and locational preferences to keep most ofany incremental saving invested in the country oforigin (FH p. 323).

Of itself, the FH finding is not controversial andhas been replicated with only minor modificationsfor OECD countries many times. Despite the ap-parent robustness of the FH result of a high sav-ing-investment association, the FH view orinterpretation that the FH coefficient (i) can be

identified as a measure of international capitalmobility has been widely challenged since it is notobvious what structural parameters this equationmeasures.

The FH equation can also be parameterized interms of the current account, B=SI:



4/20


(B/Y)i=h+(1i)(S/Y)i+ui. (2)

In this framework, the FH interpretation is that,given zero capital mobility, changes in saving arereflected in changes in investment and have noeffect on the current account. This runs counter to

the tradition of exchange rate and open economymodels. For example, all contemporary exchangerate models, with the exception of portfolio bal-ance and some Mundell-Fleming models, assumeperfect mobility (Taylor, 1995). Dornbusch (1991)sums up how the open economy parameterizationabove links with the original FH regression:

Feldsteins discovery of the tight link between na-tional saving and investment rates continues to bafflethe profession. Ample research over the past fewyears has failed to reject the basic finding TheFeldstein finding runs counter to the spirit of theopen economy literature in which, under of condi-tions of perfect capital mobility, changes in national

saving rates are primarily reflected in the currentaccount, not in investment. (p. 220).

The FH puzzle is further exacerbated by the highdegree of capital mobility implied by some interestparity studies4 and casual empiricism. The 1980ssaw widespread deregulation of financial marketsinvolving the removal of impediments to cross-

border trading of financial instruments, and infor-mation and communication technology (ICT)advances which facilitated the international trans-fer of capital. Even where formal exchange con-trols survive, agents seem able to avoid them as is

evidenced by the flight of capital from manydeveloping countries.5

Approaches to Capital Mobility

The FH puzzle has emphasized that defining andmeasuring capital mobility is not straightforward.Capital is not homogenous and the obstacles to themobility of financial, physical and human capitalmay be quite different. Here we briefly place theFH approach in context by outlining three compet-ing approaches to capital mobility.

FH Quantity Approach

The attractions of the FH approach are its intuitivesimplicity and data availability. However its sim-plicity proves deceptive since the FH regression isneither a structural model of investment nor a

reduced form equation which raises an identifica-tion problem. Although the original FH (1980)paper did not provide an explicit theoreticalframework, we use a linear version of the partialequilibrium model of Feldstein (1983) to examinethe identification problem. This is a classical model

in which national saving and investment and for-eign investment (the balance of payments) arefunctions of the real interest rate (r) and stochasticshocks. An open economy equilibrium conditiondetermines the real interest rate. The S, I and Bvariables are implicitly expressed as shares of GDPand, together with r, are measured as deviationsfrom their means. For a sample of countries, j=1,2,,N, the model is as follows:

Ij=rj+1j, (3)

Sj=rj+2j, (4)

Bj=prj+3j. (5)Equilibrium is given by equilibrium on the

world balance of payments:

SjIjBj0, (6)

implying that:

rj=1j+3j2j

++p. (7)

The FH coefficient is estimated as:

i.=

%N

j=1

IjSj

%

N

j=1Sj2

, (8)

which is a function of all six elements of thevariance covariance matrix and the three struc-tural parameters. Feldstein (1983) gives the de-tailed formula.

In the case of perfect capital mobility, there is nodifference between the country interest rate andthe world average (rj=0) and real interest parityholds. The implication is that testing for capitalmobility using the FH approach involves testing a

joint hypothesis of a high saving-investment asso-ciation and real interest parity. Frankel (1991) ar-

gues that the failure of real interest parity is themost likely explanation for the positive covariance

between saving and investment. The FH coeffi-cient measures the effect of an exogenous shock tosaving on investment but will be close to zero onlyif the covariance between investment and saving



5/20


shocks is close to zero. This will not hold if com-mon factors (such as technology shocks, popula-tion growth and the like) move both saving andinvestment. FH readily admitted that the highsaving-investment association may be explained

by common factors but put the burden of discover-

ing statistically significant common factors on theircritics who have largely failed in this respect.

Under imperfect capital mobility, interest ratesdiffer from country to country and, using (3) and(7), investment is given by:

Ij=1j+3j2j

++p+1j . (9)

If the covariances between 2, 1 and 3 are zero,the effect of a shock to saving (2) on investment isgiven by:

i=

++p. (10)

This shows that the estimate of i will not be unityeven under low capital mobility (p:0) unlesssaving is interest inelastic, =0. In the latter in-stance, the FH coefficient reduces to /(+p) butis less than unity for p\0.6 Thus, even withinFHs own framework, it is only under strong iden-tifying assumptions that the FH coefficient has therequisite properties for the FH interpretation. Arelated difficulty is the absence of a benchmarkvalue for i corresponding to perfect capital mobil-ity. Either perfect capital mobiity or infinitely in-

terest elastic saving (or both) imply i tending tozero and we cannot distinguish these cases.

Interest ParityNo ArbitrageRelations

These relations can be classified into three separatecomponents: covered interest parity (CIP), uncov-ered interest parity (UIP), and real interest parity(RIP). We focus on RIP since it is implicitly as-sumed in the FH approach. We follow Frankels(1991) decomposition of RIP into two components(note that the variables below such as r, i and Pnow denote levels):

rr*=(i i*fd)+(fdDPe+DPe*), (11)

where r is the real and i the nominal interest rate,DPe expected inflation and fd is the forward dis-count (starred variables refer to their foreign coun-terparts). The first bracketed term on the right is

the country premium or covered interest paritywhich captures all barriers (such as transactioncosts, information costs, capital controls, and vari-ous taxes) to integration of financial marketsacross national boundaries. Obstfeld (1995) showsthat, although the offshoreonshore money mar-

ket links (equivalent to CIP) increased in a sampleof leading economies in the 1980s, the risk ofgovernment intervention, particularly in times ofexchange rate crisis, remains significant. Anothercomplicating factor is that financial markets areless well integrated at long term maturities than at3 month maturities which are typically used to testcovered interest parity (Dooley et al., 1987; Frankel,1991, 1992).

The second bracketed term on the right of (11) isthe currency premium which consists of the ex-change risk premium (fdDSe) and expected realdepreciation (DSeDPe+DPe*) where DSe is ex-pected depreciation. Frankel argues that a cur-rency premium exists due to real and nominalexchange rate variability. Even with equalizationof covered interest rates (i i*fd=0), large dif-ferentials in real interest rates may persist due tovolatility in both components of the currency pre-mium. The failure of RIP is also highlighted byLemmen and Eijffinger (1995). In brief, RIP re-quires not only perfect capital mobility but alsothe integration of goods markets and efficiency ofexchange markets.

Consumption Smoothing Approach

The basis of this approach is that, in a world ofintegrated capital markets, consumption risks can

be traded to improve welfare. One implication isthat consumption paths should be correlatedacross countries as agents smooth consumption inthe face of shocks. However consumption smooth-ing is possible only if there is no credit rationing infinancial markets (Bayoumi, 1990; Artis and Bay-oumi, 1992).7 Bayoumi and MacDonald (1995) ar-gue that their approach has the benefit of allowingtests of RIP without assuming ex ante purchasing

power parity. Their conclusion that Japan is theonly advanced economy for which capital marketintegration is not rejected typifies the negativefindings on capital mobility found using this ap-proach. Since the consumption-smoothing ap-proach is surveyed elsewhere (Obstfeld, 1995) we



6/20


focus here on a recent variant found in Ghosh(1995) and Ghosh and Ostry (1995).

Ghosh (1995) starts from Sachs (1981) observa-tion that the current account acts as a buffer orthat countries borrow and lend abroad in order tosmooth consumption in the face of temporary

shocks. Since saving and investment seem to benon-stationary processes and the current accountmight be expected to be a stationary process, hefocuses on current account volatility rather thansaving and investment correlations to test for per-fect capital mobility. Ghosh constructs a bench-mark time series of the optimal consumptionsmoothing, current account given perfect capitalmobility and shocks to the economy. A low vari-ance of the actual to the optimal current accountimplies that the degree of capital mobility is in-sufficient to allow the actual current account toabsorb shocks to smooth consumption.

Ghoshs null hypothesis is a joint hypothesis ofperfect capital mobility and the consumption-smoothing motive. The optimal current accountpath is defined as:

CA*t=Et %

i=1

(1+r)1 D(Qt+iIt+ iGt+i),

(12)

where Q is national output or GDP, I is invest-ment, and G is government expenditure. Thisstates that the optimal current account path isequal to the expected present value of changes in

future national cash flows (Qt+iIt+ iGt+ i). Hecompares the variance of the optimal and theactual current accounts of five major OECD indus-trialized countries from 196088. If the two vari-ances are equal, then the null hypothesis isaccepted and we have perfect capital mobility andconsumption smoothing. For four out of five coun-tries the volatility of the actual current accountexceeds the optimal current account by statisticallysignificant margins. He attributed this excessvolatility to speculative capital flows. Ghosh andOstry (1995) apply similar methods to a sample ofLDCs for various periods, ranging from 195091.

The results of Ghosh and Ghosh and Ostry standout in suggesting excess capital mobility in bothadvanced and developing countries. These find-ings run counter not only to those of FH but alsothose of the consumption-smoothing approach.The Ghosh approach has two difficulties. The first

is that that Ghosh does not provide either anempirical or theoretical reconciliation of his con-trarian results with alternative approaches whichuse the saving-investment association. The secondis that one may question the manner in whichGhosh constructs his optimal consumption series,upon which his results hinge.

MEASURES OF THE

SAVING-INVESTMENT ASSOCIATION

S3. The residuals are non -normal and the specifica -tion of the model is incorrect.S8. Have you tried two stage least squares?S22. What happens when you extend the anlaysis tothe later (earlier) period?S33. Have you tested for unit roots and cointegra -

tion?8

Econometric results are notoriously fragile sinceestimates are sensitive to such issues as specifica-tion, estimation method, sample and the particularmeasures used to proxy the theoretical variables.The initial response to FH was to target economet-ric errors as the explanation for their finding (To-

bin, 1983; Westphal, 1983; Murphy, 1984; Frankel,1986). The problems of the FH regression includeidentification discussed in the previous section (itis not clear what theoretical parameter the FHcoefficient measures), misspecification (relevant

common factors may have been omitted), simul-taneity bias (saving may be endogenous), perma-nent and transitory effects (was the FH coefficientmeasuring short or long-run impacts discussed inthe previous section), sample sensitivity, and non-stationarity. In general the FH results have beenrobust to these empirical critiques in OECD sam-ples.

Replicating the FH Regressions

The main evidence provided by FH and the bulk

of the subsequent work has been based on cross-section regressions. FH conducted time-series anal-ysis but pointed out that: [s]imultaneousequations bias makes these time-series estimatestoo unreliable to warrant serious attention (fn. 1,p. 327). Other researchers have mainly followed



7/20


the FH lead in focusing on cross sections but somehave also produced time series and panel esti-mates of the saving-investment association. Notethat as long as the deviations of the time-series FHcoefficients from their mean are independent ofthe saving rate, the cross-section estimate will

provide an unbiased estimate of the mean of thetime-series coefficients for each country (Zellner,1969).

Data

FH used gross rather than net national saving andinvestment flows for two reasons. Firstly it is grossnot net saving flows which respond to world wideyield differentials. Secondly, the measurement ofdepreciation is inaccurate, in particular in the pres-ence of high inflation rates. Removing depreciationfrom gross saving and investment may cause aspurious correlation between net saving and in-vestment, biasing the FH coefficient upwards. Theactual FH estimates of the coefficient on saving inEquation (1) using annual data for 16 OECD coun-tries for the 196074 period9 were 0.89 and 0.94 forgross and net measures, respectively illustratingthat the difference between the two estimates issmall when the FH coefficient (i) is close to one.In more recent studies in which the FH regressionis estimated using gross and net measures, the netmeasures invariably produce a larger coefficienton saving (see Feldstein, 1983; Feldstein and Ba-

chetta, 1991; Tesar, 1991, 1993).The OECD publishes net rather than gross mea-sures for public and private saving and investmentin the National Accounts for its member countries.10

Net saving is derived by deducting governmentand private final consumption from national dis-posable income. Net investment is defined as grossfixed capital formation less stock building anddepreciation. Gross investment and saving are cal-culated as the net measures plus depreciation.Obstfeld (1986), Baxter and Crucini (1993),amongst others, argue that the OECD measure ofsaving and investment does not reflect true saving

and investment. The difference arises when for-eign ownership of firms is extensive. Obstfeld(1986) argues that foreign ownership in openeconomies may artificially increase the correla-tion between measured national saving and do-mestic investment (ibid. p. 84). Bayoumi et al.

(1996) use this insight to construct a new measureof capital mobility based on GNP/GDP ratios.

Cross-Section Results

The FH result based on cross-section data has

proved remarkably robust. Table 1 reports on aselection of cross-section estimates of the saving-investment association (i) for different samples ofOECD countries using a variety of time periods.Luxembourg has been identified as an outlier inmost recent studies and the FH result collapses ifit is included. The first part of Table 1 considersthe evidence in favour of the FH result. The origi-nal FH result was updated by Feldstein (1983) whoextended the FH sample period from 196074 to196079. He confirmed that the saving-investmentassociation estimates had not declined by extend-ing the sample period. More recently Feldstein andBachetta (1991) re-estimated the FH equation forthe 196086 period using a sample of 2311 OECDcountries. They found that the saving-investmentassociation had only marginally declined over thelonger sample period.

The rest of Table 1 presents the other replica-tions of the FH result in approximately chronolog-ical order. The empirical work in the 1980sappeared to confirm the original FH result andindicated that estimates of the saving-investmentassociation did not decline when the estimationperiod was extended to 1980 and beyond (Mur-

phy, 1984; Penati and Dooley, 1984; Obstfeld, 1986;Dooley et al., 1987; Golub, 1990; Artis and Bay-oumi, 1992; Sinn, 1992; Coakley et al., 1994, 1995a;Obstfeld, 1995). The only studies which show anestimated value of b less than 0.7 for OECD sam-ples are those where saving and investment areaveraged over periods of less than one decade (e.g.Artis and Bayoumi for 197480 period).12 Coakleyet al. (1994) and Coakley et al. (1995a) use the mostrecent data which yield an estimated i of 0.75 forthe 196092 period and 0.63 the more recent 198092 subperiod.

Time Series

Table 2 presents selected time series estimates ofthe FH coefficient and, where applicable, of thecorrelation coefficient between saving and invest-ment. Only a relatively small number of time



8/20


Table 1. Cross section estimates of FH regression

Period i (S.E.) R2 OECDa sampleAuthor

1960 74 0.91 16b0.887 (0.074)Feldstein and Horioka (1980)

0.796 (0.112) 0.75Feldstein (1983) 1960 79 FH 16 plus Fra

0.833 (0.094) 231960-86Feldstein and Bachetta (1991)0.848 (0.063)1960690.671 (0.121)197079

1980 86 0.868 (0.126)

196080 0.90 (0.09) 0.85 FH 16 plus FraMurphy (1984)

0.710.88 (6.12) 19197181Penati and Dooley (1984)

0.858 (0.806) 0.07Obstfeld (1986) 1970 79 FH 16 plus Fra0.41 FH 161.422 (0.456)

0.746 (0.104) 0.79Dooley et al. (1987) FH 16 minus 21960730.571974 80 0.736 (0.173)

0.53 (0.18)Artis and Bayoumi (1992) 1974 80 25 plus Yugoslavia0.79 (0.12)198188

196088 0.76 (0.12)

0.75 FH 16Golub (1990) 1960 86 0.74 (0.12)0.76 (0.12)196088

0.84 (0.13) 0.73Tesar (1991)c 1960 86 FH 160.850.89 (0.10)1960740.58 23197586 0.81 (0.18)

0.84 (0.10) 0.741960860.860.87 (0.07)196074

1975 86 0.85 (0.15) 0.59

0.715 (0.131) 0.60Obstfeld (1995) 22 minus Lux, Tur1974900.867 (0.170) 0.56197480

198190 0.636 (0.108) 0.64

0.800.752 (0.079) 23196092Coakley et al. (1994, 1995a)0.883 (0.063) 0.89196074

0.649 (0.104) 0.63197592 0.690.628 (0.090)198092

a The full list of OECD countries (with abbreviations in parentheses) comprises: Australia (Aus), Austria (Aut), Belgium(Bel), Canada (Can), Denmark (Dnk), Finland (Fin), France (Fra), Germany (Ger), Greece (Grc), Iceland (Isl), Ireland (Ire),Italy (It), Japan (Jap), Luxembourg (Lux), Netherlands (Nld), Norway (Nor), New Zealand (Nzl), Portugal (Prt), Spain(Esp), Sweden (Swe), Switzerland (Che), Turkey (Tur), UK, and the US.b The FH sample excludes Fra, Isl, Lux, Nor, Prt, Esp, Che and Tur. We refer to these as the FH 16.c Tesars results are based on net saving and investment rates.

series as compared with cross section replicationsof the FH result has been undertaken. Nonethelesstime series estimates are important especially as aguide for pooling time series observations fromdifferent countries as Obstfeld (1986) has argued.The distinguishing feature of Table 2 is the consid-erable degree of heterogeneity in estimates forindividual countries coefficients. In Obstfeld(1986) the correlation coefficients vary from 0.13 to0.92 while those of Tesar (1993) vary over a very

similar range. Over the entire 196092 period, thetime series estimates of i by Coakley et al. (1994)ranged from 0.025 for Luxembourg to 1.182 forSwitzerland.

If the interest sensitivity of saving is not zeroand the covariance between the error terms is notzero, then saving will be endogenous and theestimates of the FH coefficient will suffer simul-taneity bias. FH were aware of the potential simul-taneity problem and employed instrumental



9/20


Table 2. Selected time series estimates of the FH regression

i (S.E.) Correlation coefficientTime period Country

0.194 (0.106)Obstfeld (1986) Aus1960I84IV0.132 (0.195) Aut1970II84I0.550 (0.125)1959I 84I Can

0.649 (0.133)1960III 84II Ger0.846 (0.140) Jap1959I83IV

1959I 84II 0.604 (0.166) UKUS0.908 (0.143)1959I84II

1946I 87III US0.571Miller (1988)

na 0.848 (0.102)Tesar (1993) Can1960880.929 (0.071) Fra0.886 (0.091) Ger0.063 (0.196) It0.592 (0.154) UK0.752 (0.124) US

1947 89 0.335 (2.2) CanAfxentiou and Serlitis (1993)

0.796 (10.76) US1955I73VAlexakis and Apergis (1992)

1974I90V 0.657 (1.94)AusCoakley et al. (1994, 1995a) 1960 92 0.580 (0.087)

0.910 (0.086) Aut0.678 (0.073) Bel0.710 (0.086) Can

Dnk0.745 (0.112)0.857 (0.102) Fin0.843 (0.045) Fra

Ger1.019 (0.092)Grc0.770 (0.050)

0.741 (0.145) Isl0.160 (0.319) Ire0.753 (0.078) It

Jap1.043 (0.106)0.025 (0.054) Lux

0.308 (0.318) Nld0.469 (0.193) NorNzl0.293 (0.121)Prt0.293 (0.121)Esp0.758 (0.133)

0.740 (0.055) Swe1.182 (0.242) Che0.717 (0.89) Tur0.401 (0.147) UK0.586 (0.096) US

variable estimators. They reported that this didnot alter the main thrust of their originalfindings. This finding has been supported by oth-ers who have used instrumental variable estima-tors (Dooley et al., 1987; Bayoumi, 1990; Tesar,1991).

Panel EstimatesA few researchers have employed pooled andother panel data estimators (such as the Swamy(1971) random coefficients estimator) to obtain al-ternative measures of the saving-investment asso-ciation (Feldstein, 1983; Amirkhalkhali and Dar,



10/20


1993; Coakley et al., 1994, 1995b). However, sincethe cross-section variance dominates the totalvariation and the time series estimates of i arevery heterogenous, the pooled estimates arelargely similar to the cross section estimates forthe OECD sample, 196092. The conclusion from

the limited number of studies undertaken seemsto be that FH cross section, panel, and averagetime series estimates are close to unity for OECD

but not LDC samples. The cross section estimatematches the average time series levels estimate(Coakley et al., 1995a). Pesaran and Smith (1995)discuss the relationship between cross section,panel, and time series estimates.

Sample Sensitivity

Although the finding of a high saving-investmentassociation seems relatively robust for the OECDgroup of countries as a whole, some researchershave found anomalies when the sample is disag-gregated or changed in various ways. Below wefocus on the main anomalies.

Large Country Effect

Murphy (1984) argues that the FH regression teststwo hypotheses, perfect capital mobility andsmall open economies. Small countries are unableto influence world interest rates and prices butlarge countries are and so can bias the saving-in-vestment correlation towards one even underconditions of perfect capital mobility. FH did rec-ognize the large country effect by arguing, thatif a country is large, it will behave like a closedeconomy:

while the link between domestic saving and invest-ment may vary among countries, we found no evi-dence that it varied in relation to either the size of theeconomy or the importance to international trade.(Ibid. p. 323).

Murphy divided his sample of 17 OECD coun-tries into ten small and seven large countries. He

found that his group of large countries had aresponse coefficient on saving of 0.98 while thesmall country group had a coefficient of only0.59. He argues that these results are consistentwith the expected effect of country size whencapital is mobile between countries.

Harberger (1980) argues that the FH resultreflects large country bias rather than low capitalmobility. He considers whether capital flows(which he estimates as the current account as aratio of gross investment) are less variable andsmaller in absolute size for large countries thanfor small or poor countries. A large country willfinance most investment projects from domesticsaving and thus will have less need to borrowfrom abroad. He finds that gross capital flows areindeed more variable and greater in absolutevalue for smaller countries. The views of Murphyand Harberger on the large country effect areechoed elsewhere in the literature (Obstfeld,198613; Tobin, 1983; Baxter and Crucini, 1993).

EU Effect

Feldstein and Bachetta (1991), Artis and Bayoumi(1992) and Bayoumi et al. (1996) argue that, dueto informational and institutional links, financialflows between EU countries should be greaterthan among OECD countries and thus thesecountries should experience a lower saving-in-vestment correlation. Feldstein and Bachetta splitthe 23 OECD country sample into the then EU (9)and non-EU country samples for the 196086 pe-riod. Using net figures they show that the EUcountries experienced a small decline in the sav-ing-investment association from the 1960s to the1970s, followed by a sharp decline in the 1980s.

By comparison, the coefficient for the 14 non-EUcountries declined much more slowly. These re-sults suggest that capital mobility has been en-hanced over the years amongst EU countries,particularly in the 1980s. Using gross measures,the coefficient on saving for the EU countries issubstantial throughout the period even though itfalls somewhat in the 1980s. However the declinein the corresponding coefficient for non-EU coun-tries is much sharper, suggesting, accordingto the FH interpretation, that capital mobilityhas increased more in the non-EU countries.

These conflicting findings indicate that theFeldstein Bachetta results must be interpretedwith caution.

Artis and Bayoumi (1992) also claim to havediscovered an EU-type effect by focusing on thesix core members of the exchange rate mechanism,



11/20


the ERM-6. The coefficient on saving for the origi-nal ERM-6 countries decreased throughout thesample period 1960 88 and was insignificantlydifferent from zero (0.58, S.E.=0.33) for the 198188 period. For the same sample period, Artis andBayoumi found the entire OECD group produced

a coefficient on saving of 0.78 (S.E.=0.12). Theyconcluded that within the FH framework, theERM-6 countries are substantially integrated. Fi-nally Bayoumi et al. (1996) confirm a higher sav-ing-investment association for EU economies thanfor either Europe as a whole (excluding Luxem-

bourg) or for the OECD sample.

Developing Countries

Dooley et al. (1987) examined 62 countries, 48 ofwhich were developing countries and the remain-ing 14 OECD countries. They split the sample intotwo sub-periods, the fixed exchange rate period,1960 73, and the floating exchange rate period,1974 84, when many countries removed capitalcontrols. They find that the saving coefficient isgreater for the OECD countries than for the devel-oping countries and that the coefficient is greaterin the floating rate period for both groups. Theydivided the heterogeneous developing countriesinto 21 market borrowers and 14 countries whichdepend solely on official financing. The saving-in-vestment correlation was positive and significantfor both groups combined and the relationship

was stronger in the second sub-period and for themarket borrowers than for the official borrowers.They attribute the lower coefficient on saving forthe developing countries to the country size fac-tor, since the sample of developing countries com-prises small countries which cannot influence theworld interest rates and therefore their savings-in-vestment correlation is not biased upwards.

Mamingi (1994) investigated the FH regressionusing time series estimation for 58 developingcountries. Overall he found that the saving-invest-ment coefficient was much weaker for developingcountries than the corresponding coefficient in

studies of OECD countries. He reached a similarconclusion to Dooley et al. and argued that devel-oping countries are essentially small openeconomies where fiscal policy used for demandmanagement purposes will be unable to crowd outprivate sector investment.

Other

A number of other disaggregations of the FHequation has appeared in the literature. FH them-selves divided national saving and investment intothree separate components for nine OECD coun-

tries: the government, household and corporatesectors. Their main result is that (gross) corporateinvestment appears responsive to corporate ratherthan other forms of saving. Although they claimthat this is consistent with their general results, ithas to be treated with caution in view of thelimited degrees of freedom of their regressionequation.

A number of studies has used the FH approachto investigate the question of regional or intra-na-tional capital mobility. For instance, Bayoumi andRose (1993) used regional UK data over the 197185 period and found regional saving and invest-

ment rates were uncorrelated. They concluded thattheir results were consistent with the hypothesis ofperfect regional capital mobility but this could also

be interpreted as lack of default risk since there isno solvency constraint. Similar results have beenreported for other regional studies: Dekle (1996)for Japanese prefectures and Bayoumi and Sterne(1993) for Canadian regions. Finally Bayoumi et al.(1996) use a novel approach to confirm that re-gional capital mobility is higher than internationalcapital mobility by comparing GNP/GDP ratiosacross EU countries with equivalent ratios acrossUK regions.

Non-stationarity

In recent years it has been recognized that nationalsaving and investment rates are very persistentseries since the null of a unit root cannot berejected (Miller, 1988; Ballabriga et al., 1991; Leach-man, 1991; Gulley, 1992; Gundlach and Sinn, 1992;Afxentiou and Serlitis, 1993; Alexakis and Apergis,1992; Argimon and Roldan, 1994; de Haan andSiermann, 1994; Ghosh, 1995; Lemmen and Ei-

jffinger, 1995; Coakley et al. 1996a,b; Coakley

and Kulasi, 1997). Coakley et al. (1996a) notethat although saving and investment shares ofGDP are I(1) variables, the difference betweenthem the current account balance should bestationary or I(0) as a result of solvency con-straints. In the latter case, saving and invest-



12/20


ment will cointegrate with a unit coefficient andthe time series FH regression in levels will yieldsuper-consistent estimates of i.

Note that evidence in favour of a stationarycurrent account or cointegration of saving andinvestment can be interpreted in two diametrically

opposing manners. On one hand it can be inter-preted as confirmation of the FH result; on theother it can be interpreted as evidence of opencapital markets imposing a solvency constraint oncountries. The empirical evidence on the stationar-ity of the current account employing conventionalcointegration techniques is mixed (Miller, 1988;Gulley, 1992; Gundlach and Sinn, 1992; Argimonand Roldan, 1994; Ghosh, 1995). Some recent ap-proaches have employed panel unit root tests totest for the stationarity of the current account(Coakley et al., 1996a,b; Krol, 1996; Coakley andKulasi, 1997). The panel tests of Im et al. (1995)have higher power than individual, pairwise testsand the general conclusion is that the currentaccount is a stationary series in both in developingand OECD countries.

ALTERNATIVE INTERPRETATIONS

General Equilibrium Approaches

S29. The problem cannot be dealt with by partialequilibrium methods: it requires a general equi-

librium framework.

The implicit FH theoretical framework previ-ously set out, was a partial equilibrium frame-work. Several authors have constructed generalequilibrium models which simultaneously incor-porate a high saving-investment correlation andhigh or perfect capital mobility (Obstfeld, 1986;Engel and Kletzer, 1989; Finn, 1990; Cardia, 1991;Mendoza, 1991; Backus et al., 1992; Baxter andCrucini, 1993; Tesar, 1993; Stockman and Tesar,1995). One drawback of the general equilibriummodels is that they are mainly theoretical in nature

and generally lack supporting econometric evi-dence.14

Obstfeld (1986) constructs a simple intertempo-ral model of a small open economy in whichtemporary shocks to the productivity of domesticcapital and labour cause short run comovements

between saving and investment despite perfectcapital mobility. Because labour is assumed to beimmobile across national boundaries, productivityshocks to labour in the domestic economy cannot

be absorbed by foreign labour markets. In order tomaintain the equality between the marginalproduct of capital and the rate of return on capital,namely the world rate of interest, firms must in-crease their capital stock. Obstfeld (1986) arguesthe essential reason for this co-movement [be-tween savings and investment] is again the factthat labour is not mobile across national

boundaries (ibid. p. 74). If labour was perfectlymobile across national boundaries, the impact ofproductivity shocks to labour in the domesticeconomy would be eliminated since excess domes-tic workers would be absorbed abroad. Savingmay increase but domestic investment need not

increase since the marginal product of capital re-mains unchanged.

Tesar (1993) links the FH result to the lowcross-country consumption correlation and thedominance of domestic assets in national portfo-lios. She reports cross-country consumption corre-lations for five OECD countries, ranging from0.35 between Canada and France to 0.58 be-tween France and Germany, with an average cor-relation of only 0.09. French and Poterba (1991)find that for the five largest national stock markets,domestic ownership range from 79% in Germanyto 95.7% in Japan. Tesar argues that these empiri-

cal regularities imply a lack of consumptionsmoothing, even under conditions of perfect capi-tal mobility. She develops a simple two country,exchange model with investment and productivityshocks to non-traded goods. The model demon-strates that these empirical regularities can be ex-plained by productivity shocks to non-tradedgoods and agents preferences particularly vis-a-vis the composition of traded and non-tradedgoods in consumption smoothing. In such cases,domestic investors will bias their portfolios to-wards assets in domestic non-traded goods.

Tesars theoretical model is illustrated by simula-tion results where productivity shocks are gener-ated from real data from five OECD countries.

Baxter and Crucini (1993) (BC) set out to resolvethe conflicting results reported by Sachs (1981) andFeldstein and Horioka (1980) and to offer an expla-



13/20


nation for the large country effect put forward byMurphy (1984). Their model shows high time se-ries correlations between saving and investmentunder perfect capital mobility, higher correlationsfor larger countries, and a negative relationship

between current account deficits and investmentas found in Sachs. BC base their conclusions on atwo country, one sector growth model driven byexogenous shocks to productivity where the effectsof government fiscal policy are assumed neutral.BC conclude:

Sachss empirical findings have traditionally beeninterpreted as evidence in favour of internationalcapital mobility, while high values of saving andinvestment correlations have been interpreted as evi-dence against capital mobility. Our model starts fromthe assumption of highly mobile capital and simulta-neously accounts for both of these phenomena. (p.428).

Note however that Penati and Dooley (1984) foundthat the relationship between investment and thecurrent account was statistically insignificant.

Current Account Explanations

S21. The central argument is not only a tautology, itis false.

The tautology is the identity that saving minusinvestment equals the balance of payments oncurrent account. A number of authors has sug-

gested that, if, in response to balance of paymentsdisequilibrium, public or private decision-makersreact in such a way as to restore equilibrium, thiswould induce an association between saving andinvestment even under high capital mobility. Sev-eral variants of the current account explanation arefound in the literature (Bayoumi, 1990; Ballabrigaet al., 1991; Artis and Bayoumi, 1992; Argimon andRoldan, 1994; Ghosh, 1995; Glick and Rogoff, 1995;Coakley et al., 1996a,b).

Sachs (1981, 1982) argued that changes in invest-ment opportunities rather than oil price changes

were the main determinant of current accountimbalances in the 1970s. Sachs (1981) regressed theaverage current account (as a proportion of GDP)for 15 industrial countries on their average invest-ment rates and found that the slope coefficient wasnegative and significantly different from zero. Sim-

ilar results are reported in Sachs (1982), Table 5.2,for slightly different samples of countries andwhere the data were averaged over different sam-ple periods. Sachs finds that changes in the bal-ance on the current account are much more closelycorrelated with changes in investment than

changes in saving rates. He concludes that theresponsiveness of world capital to changes in in-vestment in the domestic economy implies thatworld capital markets are integrated to a greatextent15.

Sachs results sharply contradict the FH result.Penati and Dooley (1984) (PD) hypothesize that ifSachs results are valid, then Sachs equationshould strengthen over time as international finan-cial markets become more integrated. They repli-cated both the FH and Sachs equations using 19industrialized countries. They showed that, whilethe FH equation remained robust over time, Sachsrelationship breaks down when the sample periodis extended to include 1980 and the number ofcountries is increased to 19. The results led PD toreject capital mobility. In fact Penati and Dooleyfound that Sachs results depended heavily onoutlier countries, and once these were removedfrom the sample, an estimate of the slope coeffi-cient not significantly different from zero was ob-tained.

Another strand of this literature is that the sav-ing-investment correlation is high because govern-ments target the current account using appropriate

policy instruments (Bayoumi, 1990; Artis and Bay-oumi, 1992 (AB)). For instance, AB argue thatgovernments target the current account usingmonetary rather than fiscal policy. To examine thisthey estimated the following reaction function:

Dr=h+i Dy+k Dp+lCA/y+u, (13)

where r is a discount rate administered by mone-tary authorities, Dy is growth, Dp is inflation, CA/yis the current account balance as a ratio to GDP orGNP, and u is a random error term. If govern-

ments targeted the current account, l should benegative. Since there was a decline in the saving-investment association in the 1980s, AB hypothe-sized that current account targeting bygovernments had declined in the 1980s. They con-clude that their results:



14/20


appear to confirm that the current account was asignificant policy target for monetary policy in the1970s, but that its importance diminished somewhatin the 1980s. This behaviour appears to correspond toa reduction in the correlation between saving andinvestment among OECD countries... it seems likelythat government targeting of the current account

helps account for the high correlation. (Ibid. p. 303).

Coakley et al. (1996a) propose a resolution of theFH puzzle based on the current account whichdiffers from the previous explanations in severalways. First it stresses capital market constraintsrather than government targets for the currentaccount. Unlike the Ballabriga et al. (1991) model,the Coakley et al. model does not rely on endoge-nous government policy responding to private sec-tor behaviour or to shocks. Instead it relates savingand investment behaviour to the current accountvia a market determined risk premium on borrow-ing. Secondly, their approach does not rely oncapital controls. They argue that, even under per-fect capital mobility, the interest rate faced by acountry will contain two components, the worldrate and a risk premium/discount which averagesto zero across countries. The risk premium/dis-count will respond to the balance of paymentsthus ensuring long-run solvency and making thecurrent account as a share of GDP a stationaryprocess. Since saving and investment rates areintegrated processes, a stationary current accountimplies that they cointegrate with a unit coeffi-

cient. It is this solvency constraint that the FHcoefficient measures. They adduce empirical evi-dence for panels of both OECD countries andLDCs to support their theoretical model (Coakleyet al., 1996a,b).

Human Capital

S16. Of course if you allow for investment in humancapital the entire picture changes.

The FH puzzle about capital mobility has also

been investigated within the framework ofneoclassical growth models (Mankiw et al., 1992;Barro et al., 1995). Human capital plays a centralrole in many of these models. Barro et al. (1995)use a modified neoclassical growth model to in-vestigate the implications of various assumptions

about capital mobility in growth models. Perfectcapital mobility carries the implausible implicationof immediate convergence to steady state levels ofper capita output, physical capital and humancapital. To avoid this problem, they consider acase of partial capital mobility where only physicaland not human capital can be used as collateral oninternational financial markets. On this basis theyassert:

(a)lthough the credit-constrained open economy con-verges faster than the closed economy, the speed ofconvergence is now finite for the open economy andthe difference from the closed economy is not largefor plausible parameter values (p. 112).

Barro et al. develop a modified Ramsey, infinite-horizon optimizing model. They analyse the opti-mal economic growth path using a one sectormodel with Cobb-Douglas technology, in which

output is produced with three inputs, physicalcapital (k), human capital (h) and non-repro-ducible labour. They demonstrate that the conver-gence implications of the partial capital mobilitymodel are similar to those of the closed economy.Both models demonstrate that output is an in-creasing function of capital stock under conditionsof diminishing returns. They argue that the worldrate of interest, which is the same as under theclosed economy, implies that perfect capital mobil-ity does not affect the steady state values of k andh which are the same as for a closed economy.

However, capital mobility does affect the speed ofconvergence. The Barro et al. model predicts a highcorrelation between saving and investment irre-spective of the degree of physical or financialcapital mobility. Thus in this model the highFH coefficient sheds no light on physical or finan-cial capital mobility, only on human capital mobil-ity.

Transitory versus Permanent Influences

S12. The analysis is marred by a failure to distin - guish between transitory and permanent compo-

nents.

FH largely relied on cross-section estimates. Ini-tially time-series estimates seemed less supportiveof the puzzle, though Coakley et al. (1994) showthat the average across countries of the long-run



15/20


time-series estimate of the FH coefficient is almostexactly the same as the cross-section estimate.However, the time-series estimates, particularly infirst differences, are more likely to pick up theshort-run transitory responses of investment tosavings while cross-section estimates are more

likely to capture the long-run permanent response.It is a fairly general result that the short-run esti-mates of the FH coefficient tend to be muchsmaller than the long-run ones (Coakley et al.,1994). At first sight this seems inconsistent withthe interpretation of the FH coefficient as a mea-sure of capital mobility. Suppose there is a transi-tory increase in saving. Given the frictionsinvolved in international transactions, it may not

be worth incurring the costs of finding out aboutforeign investment opportunities or evading ex-change controls. Thus transitory increases in sav-ing would tend to be invested nationally.

However, if there were a permanent increase insavings, then it may be worth incurring those costsand one would expect part of a permanent in-crease to flow abroad. The finding that transitorychanges to saving tend to go abroad and perma-nent ones stay at home, an implication of theshort-run estimate of the FH coefficient beingsmaller than the long-run estimate, suggests thatthe explanation is not frictional obstacles to capitalmobility. To deal with this issue Feldstein (1983),p. 147, first disparages the time-series estimates,commenting that Coefficient estimates based on

annual variations in savings and investment aresubject to potentially severe simultaneity bias.Instead he provides a theoretical explanation

based on a portfolio balance model of why theshort-run effect of saving on investment should besmaller than the long-run one. His explanation isthat, when short-run portfolio adjustments arecomplete, capital flows revert to a lower level.

The Feldstein explanation has not found favourin the literature but there has been little work onthe difference between long run and short run FHcoefficients. One notable exception is Sarno andTaylor (1996). Rather than using the usual short

and long-run estimates of the FH coefficient theyemploy the Blanchard and Quah (1989) decompo-sition to distinguish between temporary and per-manent shocks to saving and investment shares ofGDP. Using quarterly UK data, they find that theshort run saving-investment correlation is signifi-

cantly higher than the long run correlation aspredicted by the FH approach. Thus transitoryincreases in saving are more likely to remain in theUK while permanent changes tend to flow abroad.They conclude that their results indicate a highdegree of capital mobility in the UK especially

following the removal of exchange controls in1979.

Public and Private Saving-Investment Gaps

Some authors have disaggregated national savingand investment into their public and private com-ponents to focus on the role of public and privatesaving-investment gaps. The general conclusion inthis approach is that capital is relatively immobile,mainly due to the operation of capital controls.Ballabriga et al. (1991) argue that correlations be-tween private and public sectoral gaps have directimplications for the FH view on the degree ofinternational capital mobility. If the sectoral gapshave a unit root, this implies a high correlation

between public and private gaps and thus lowcapital mobility since public and private intertem-poral budget constraints impose a long run exter-nal solvency condition on countries. They find thatthe null hypothesis of a unit root for sectoral gapsis not rejected, thereby violating the government

budget constraint in a large number of EC coun-tries. Since public and private sector gaps tend tooffset one another, the external solvency of the

country is preserved by the use of capital controls.Ballabriga et al. demonstrate that, although thelong run external solvency condition has not beenviolated, domestic intertemporal budget con-straints have been, implying a high correlation

between saving and investment. They argue thatthe associated low capital mobility is the result ofgovernments imposing capital controls to targetthe current account.

Argimon and Roldan (1994) base their explana-tion of the FH result on governments targeting thecurrent account and saving-investment gaps. Forfive of the nine EU countries in their sample, they

found that private and public sector saving-invest-ment gaps were cointegrated and that four of thesecountries imposed capital controls in the periodunder investigation. Since the capital account isthe mirror image of the current account, any con-straint on the former, such as exchange controls,



16/20


will necessarily impact on the latter also. Thisimplies that capital is immobile and so saving andinvestment turn out to be highly correlated. vanWincoop and Marrinan (1993) (VM) develop apartial equilibrium, perfect capital mobility, real

business cycle model which is driven by technical,

fiscal and interest rate shocks which are correlatedacross countries. Under conditions of zero capitalmobility, the correlation between total saving andinvestment should equal one and the correlation

between public and private saving-investmentgaps should be exactly minus one. VM find that,when they control for income movements, thecorrelations between total saving and investmentand public and private saving-investment gaps are0.78 and 0.81 respectively. They assert that wecannot understand the significant correlation be-tween total savings and investment by referring totechnology shocks in the context of a model withperfect capital mobility (ibid. p. 22). They con-clude that it is the immobility of capital acrosscountries which is the key factor to the high corre-lations found in the data.

In a different but related vein, Peeters (1996)carries out estimation and simulations on theGlobal Econometric Model (National Institute Lon-don) using a general equilibrium modelling ap-proach. Her results for the US, Japan, Germanyand the UK indicate a lower estimate (comparedwith partial equilibrium models) of the saving-in-vestment association and would therefore suggest

higher capital mobility in both the short and longrun. More importantly she finds that private andpublic saving gaps largely offset one another inthe short run. She concludes that increasing pri-vate saving would have little effect on the USstwin deficits since it would be almost fullyreflected in an increase in the government deficitand thus its impact on the current accountnegated.

Revised Feldstein Approach

Feldstein (1994) has provided an informal but

more sophisticated justification of the FH view. Hesupports the original FH view by linking it toanother puzzle, the home country bias puzzle(Lewis, 1995; Tesar and Werner, 1995) The highsaving-investment association is consistent withthe documented propensity of fund managers to

hold a disproportionately high share of domesticsecurities or not to diversify their portfolios inter-nationally. Feldstein offers two rationales for such

behaviour. One is the political and currency risk ofoverseas securities holdings. While he concedesthat political risk may be negligible for OECD

countries, currency risk may not be16 and this isconsistent with the Coakley et al. solvency argu-ment. Feldsteins second rationale is that hedging

behaviour creates offsetting capital flows so thatnet capital flows may be zero or negligible. Thislinks in with a more general weakness of the FHapproach which is that it focuses on net ratherthan gross international capital flows.

However Feldsteins hedging example of invest-ment in German government bonds is rather spe-cial in that (coupon) income flows in foreigncurrency can be perfectly anticipated whereas in-come from overseas equities cannot be similarlypredicted. Feldstein also overlooks a potentiallymore significant source of under recording of cap-ital flows. Increasingly fund managers use futuresand other derivatives for tactical (short term) inter-national asset allocation. Such flows are not fullyrecorded in balance of payments data. For exam-ple, the only aspect of investment by an UK fundmanager in overseas stock index futures contractsrecorded in the balance of payments statistics arethe variation margin flows which typically are buta small fraction of the total investment. Therecording of derivatives flows in the balance of

payments statistics is currently under consider-ation by central banks and the IMF.

CONCLUSIONS

In this paper we have reviewed the wayeconomists have responded to the FH puzzle, us-ing Stiglers generic comments to introduce theirlines of attack. In the process we have examined alarge number of competing interpretations of theFH coefficient. It will be apparent that the generaltone of contributions is largely but by no means

unanimously negative toward the FH interpreta-tion of low capital mobility. The approaches whichoffer support for the FH view of capital mobilityinclude sectoral gap models, the Barro et al. (1995)modified neoclassical growth model in which hu-man capital is immobile, and Sarno and Taylors



17/20


(1996) support for the FH short and long runfindings. However the majority of the models andexplanations oppose the FH interpretation by at-tempting to construct models which reconcile ahigh saving-investment association with highphysical and financial capital mobility and/or by

providing plausible econometric and other data- based refutations of the FH view. These includegeneral equilibrium, real business cycle modelsand intertemporal models of the current account.

There is no doubt that the FH debate has beenvery productive, provoking a wide range of theo-retical analysis and empirical work and has in-creased our understanding of the issues. Althoughthe debate continues, we can draw some interimconclusions. Firstly, the FH result of a high saving-investment association has remained remarkablyrobust in OECD cross-sections although the coeffi-cient on saving has shown some tendency to de-cline over recent years. The result persists inpanels and average time-series for OECD coun-tries, but not for individual countries where thereis a wide dispersion of estimates. The main resulthas also been remarkably robust to the addition ofother variables and different estimation methodsin the OECD. However, there is less evidence for aclose relationship between saving and investmentin non-OECD samples, particularly in LDCs.

Secondly, using the FH regression for policypurposes is questionable because both saving andinvestment are endogenous and the FH regression

cannot distinguish exogenous shifts in saving fromendogenous shifts reflecting factors which also im-pact on investment. In any event, governmentmeasures that promote saving are likely to be partof packages that may also promote investment.Thirdly, the FH result may not be informativeabout capital mobility since a range of theoreticalmodels can generate high saving-investment corre-lations even under perfect capital mobility. Thisnow seems to be the emerging consensus in theliterature. Baxter (1995), p. 1842) says: These re-sults show that the time-series savings investmentcorrelation is not an informative statistic concern-

ing the degree of international financial integra-tion. Similarly Obstfeld and Rogoff (1995), p.1779, observe:

Taken together, however, and combined with otherevidence indicating substantial international mobilityof capital, the arguments below suggest that the

FeldsteinHorioka finding provides no basis at all fordismissing the basic premises of the intertemporalapproach.

Nonetheless the FH puzzle is by no means re-solved and recent contributions by Bayoumi et al.(1996) and Sarno and Taylor (1996) claim to estab-lish high capital mobility within FH-related frame-works.

Fourthly, the debate surrounding the FH puzzlehas shown that the notion of capital mobility itselfis not analytically straightforward. Obstacles to themovement of financial, physical and human capi-tal may have quite different implications. Assum-ing perfect capital mobility may produce plausibleconclusions in one set of models and not in an-other. Thus while it may be sensible in open-econ-omy macroeconomics to assume perfect capitalmobility, it is certainly not in neoclassical growth

models where it would result in all countriesjumping immediately to their steady state. In suchcircumstances it seems sensible to treat perfectcapital mobility as Friedman (1953), p.36, treatedperfect competition:

Everything depends on the problem; there is noinconsistency in regarding the same firm as if it werea perfect competitor for one problem, and a monopo-list for another, just as there is none in regarding thesame chalk mark as a Euclidean line for one problem,a Euclidean surface for a second, and a Euclideansolid for a third.

ACKNOWLEDGEMENTS

We are grateful to two anonymous referees forhelpful comments on an earlier version. We acceptresponsibility for any remaining errors.

NOTES

1. This was how an earlier version (Coakley et al.,1995a) was often received. We are grateful to ananonymous referee who responded in kind by

pointing out that Stigler also provided apt responsesto this paper in his introductory remarks a) and c).2. Source: Social Sciences Citation Index.3. For a non-technical overview of the FH puzzle see

Feldstein (1994).4. For an excellent survey of this literature see Obstfeld

(1995).



18/20


5. Dooley and Kletzer (1994) argue that the increasedflow of funds to emerging country markets since thelate 1980s may be explained by reductions in thestock of capital flight rather than capital exports bythe industrial countries.

6. Feldstein (1983) gives identifying restrictions in thecontext of his real flows (quantity) model (pp. 141

45). Murphy (1984) modifies the Feldstein model toincorporate a country effect and discusses appropri-ate identifying restrictions (pp. 33031).

7. We are grateful to an anonymous referee for thisinsight.

8. The quotation marks denote that this is our sug-gested addition to Stiglers original list.

9. The countries excluded were Iceland, Portugal,Turkey, and (former) Yugoslavia due to lack of data,and France, Luxembourg, Norway, Spain, andSwitzerland due to a change in methods of nationalaccounting over the sample period.

10. Note that the terms national saving and investmentare employed in the literature to include both publicand private saving and investment.

11. Their sample excluded Luxembourg (an outlier) andthe former Yugoslavia.12. Sinn (1992) and Coakley et al. (1994) show that the

FH coefficient is much lower and more volatilewhen cross sections based on annual data are esti-mated.

13. Note this effect is not evident in the annual dataestimates of Obstfeld (1989).

14. Many of the general equilibrium models are cali- brated to match some typical empirical results butthese cannot be afforded the same status as econo-metric studies.

15. The coefficient in the Sachs regression would mea-sure (1b)/i in a determininstic model. So i closeto unity in the FH model corresponds to the Sachsregression coefficient being close to zero, ignoringthe different exogenity assumptions.

16. Note that this argument overlaps with the currencypremium component of real interest parity inFrankels (1991) exposition.

REFERENCES

Afxentiou, P.C. and Serlitis, A., International CapitalMobility and the Long Run Investment and Saving inCanada, Economia-Internazionale, 46 (1993), 14767.

Alexakis, A. and Apergis, S., The Feldstein HoriokaPuzzle and Exchange Rate Regimes: Evidence FromCointegration Tests, Journal of Policy Modelling, 16

(1992), 459 72.Argimon, I. and Roldan, J.M., Saving, Investment and

International Capital Mobility in EC Countries, Eu-ropean Economic Review, 38 (1994), 5967.

Amirkhalkhali, S. and Dar, A.A., Testing for CapitalMobility: A Random Coefficients Approach, EmpiricalEconomics, 18 (1993), 523541.

Artis, M.J. and Bayoumi, T., Global Capital MarketIntegration and the Current Account, in M.P. Taylor(Ed.) Money and Financial Markets, Cambridge, MS andOxford: Blackwell, 1992, 297307.

Backus, D.K., Kehoe, P.J. and Kydland, F.E., Interna-tional Business Cycles: Theory and Evidence, Journalof Political Economy, 100 (1992), 745 75.

Ballabriga, F.C., Dolado, J.J and Vinals, J., InvestigatingPrivate and Public Saving-Investment Gaps, CEPRWorking Paper, Number 607, 1991.

Barro, R.J., Mankiw, N.G. and Sala-i-Martin, X., CapitalMobility in Neoclassical Models of Growth, AmericanEconomic Review, 85 (1995), 103 15.

Baxter, M., International Trade and Business Cycles inG.M..Grossman and K. Rogoff (Eds.), Handbook of In-ternational Economics, vol. 3, Amsterdam: Elsevier,1995, 180164.

Baxter, M. and Crucini, M.J., Explaining Saving-Invest-ment Correlations, American Economic Review, 83(1993), 416 36.

Bayoumi, T., Saving-Investment Correlations: ImmobileCapital, Government Policy, or Endogenous Be-haviour?, IMF Staff Papers, 37 (1990), 36087.

Bayoumi, T. and Rose, A.K., Domestic Saving and Intra-national Capital Flows, European Economic Review, 37(1993), 1197202.

Bayoumi, T. and Sterne, G., Regional Trading Blocs,Mobile Capital and Exchange Rate Coordination,Mimeo, London: Bank of England, 1993.

Bayoumi, T. and MacDonald, R., Consumption, Incomeand International Capital Market Integration, IMFStaff Papers, 42 (1995), 55276.

Bayoumi, T., Sarno, L. and Taylor, M.P., Saving-Invest-ment Correlations, European Capital Flows and Re-gional Risk Discusssion Paper, Department ofEconomics and Accounting, Liverpool: University ofLiverpool, 1996.

Blanchard, O. and Quah, D., The Dynamic Effects ofDemand and Suppply Disturbances, American Eco-nomic Review, 79 (1989), 65573.

Cardia, E., The Dynamics of Savings and Investment inResponse to Monetary, Fiscal Productivity ShocksJournal of Monetary Economics, 28 (1991) 411 34.

Coakley, J. and Kulasi, F., Cointegration of Long RunSaving and Investment, Economics Letters, 54 (1997),16.

Coakley, J., Kulasi, F. and Smith, R., The Saving-Invest-ment Association Birkbeck College, Working PaperNumber 10/94, Department of Economics, Gresse St.,London, 1994.

Coakley, J., Kulasi, F. and Smith, R., Current Account

Solvency and the Feldstein Horioka Puzzle, WorkingPaper Number 8/95R, Birkbeck College, London 1995a(Revised February 1996).

Coakley, J., Kulasi, F. and Smith, R., The Relationshipbetween Cross Section, Time Series and Other Estima-tors of the FH Regression, Mimeo, Birkbeck College,London, 1995b.



19/20


Coakley, J., Kulasi, F. and Smith, R., Curent AccountSolvency and the Feldstein Horioka Puzzle, EconomicJournal, 106 (1996a), 620 27.

Coakley, J., Kulasi, F. and Smith, R., Saving, Investmentand Capital Mobility in LDCs, Working Paper Number10/96, Birkbeck College, London, 1996b. Review of Inter-national Economics, forthcoming.

Dekle, R., Saving-Investment Associations and CapitalMobility: On the Evidence from Japanese RegionalData, Journal of International Economics, 41 (1996), 5372.

de Haan, J. and Siermann, C.L.J., Saving, Investment, andCapital Mobility: A Comment on Leachman, OpenEconomies Review, 5 (1994), 517.

Devereux, M.P., Investment, Saving, and Taxation in anOpen Economy, Oxford Review of Economic Policy, 12(1996), 90 108.

Dooley, M., Frankel, J. and Mathieson, D.J., InternationalCapital Mobility: What Do Saving-Investment Correla-tions Tell US?, IMF Staff Papers, 34 (1987), 50329.

Dooley, M. and Kletzer, K.M., Capital Flight, ExternalDebt and Domestic Policies, NBER Working Paper

Number 4793, 1994.Dornbusch, R., National Saving and International Invest-

ment: Comment, in B. Bernheim andJ.B. Shoven (Eds.),National Savings and Economic Performance, NBER Re-port, Chicago: University of Chicago Press, 1991, 22026.

Engel, C.M. and Kletzer, K.M., Saving and Investment ina Open Economy with Non-Traded Goods, Interna-tional Economic Review, 4 (1989), 73552.

Feldstein, M., Domestic Saving and International CapitalMovements in the Long Run and the Short Run,European Economic Review, 21 (1983), 129 51.

Feldstein, M., Tax Policy and International CapitalFlows, Weltwirtschaftliches Archiv, 130 (1994), 67597.

Feldstein, M., The Effects of Outbound Foreign Direct

Investment on the Domestic Capital Stock in M.Feldstein et al. (Eds.), The Effects of International Taxationon Multinational Corporations,Chicago: University ofChicago Press, 1995.

Feldstein, M. and Horioka, C., Domestic Saving andInternational Capital Flows, Economic Journal, 90(1980), 314329.

Feldstein, M. and Bachetta, P., National Saving andInternational Investment in D. Bernheim and J. Shoven(Eds.), National Saving and Economic Performance,Chicago: University of Chicago Press, 1991, 20126.

Finn, M.G., On Saving and Investment Dynamics in aSmall Open Economy, Journal of International Econom-ics, 29 (1990), 122.

Frankel, J.A., International Capital Mobility and Crowd-

ing-Out in the US Economy: Imperfect Integration ofFinancial Markets or of Goods Markets?, in R.W. Hafer(Ed.), How Open Is the US Economy?, Lexington, 1986,3367.

Frankel, J.A., Quantifying International Capital Mobilityin the 1980s, in Bernheim and Shoven (Eds.), op. cit.,1991, 22770.

Frankel, J.A., Measuring International Capital Mobil-ityA Review, American Economic Review, 82 (1992),197202.

French, K. and Poterba, J., Investor Diversification andInternational Equity Markets, American Economic Re-view, 81 (1991), 222 26.

Friedman, Milton, The Methodology of Positive Econom-

ics, in Essays in Positive Economics, Chicago: Universityof Chicago Press, 1953, 343.

Ghosh, A.R., International Capital Mobility Amongst theMajor Industrialised Countries: Too Little or TooMuch, The Economic Journal, 105 (1995), 107128.

Ghosh, A.R. and Ostry, J.D., The Current Account inDeveloping Countries: A Perspective from the Con-sumption Smoothing Approach, World Bank EconomicReview, 9 (1995), 30533.

Glick, R. and Rogoff, K., Global versus Country-SpecificProductivity Shocks and the Curent Account, Journalof Monetary Economics, 35 (1995), 15992.

Golub, S.S., International Capital Mobility: Net VersusGross Stocks and Flows, Journal of International Moneyand Finance, 9 (1990), 42439.

Gulley, D.O., Are Saving and Investment Cointegrated?Another Look At The Data, Economics Letters, 39 (1992),5558.

Gundlach, E. and Sinn, S., Unit Root Tests of the CurrentAccount Balance: Implications for International CapitalMobility, Applied Economics, 26 (1992), 617 25.

Harberger, A., Vignettes on the World Capital Market, American Economic Review, 70 (1980), 331 37.

Im, K.S., Pesaran, M.H. and Shin, Y., Testing for UnitRoots in Dynamic Heterogeneous Panels, WP 9526DAE, University of Cambridge, 1995.

Krol, H., International Capital Mobility: Evidence fromPanel Data Journal of International Money and Finance,15 (1996), 467 74.

Leachman, L., Saving, Investment and Capital MobilityAmong OECD Countries, Open Economies Review, 2

(1991), 137 63.Lemmen, J.J.G. and Eijffinger, S.C.W., The Quantity

Approach to Financial Integration: The FeldsteinHo-rioka Criterion Revisited, Open Economies Review, 6(1995), 145 65.

Lewis, K.L., Puzzles in International Financial Marketsin G.M. Grossman and K. Rogoff (Eds.), op.cit., 1995,191371.

Mamingi, N., Savings-Investment Correlations and Cap-ital Mobility in Developing Countries, The World BankPolicy Research Working Paper Number 1211, 1994.

Mankiw, N.G., Romer, D. andWeil, D.N., A Contributionto the Empirics of Economic Growth, Quarterly Journalof Economics, 107 (1992), 40737.

Mehra, R. and Prescott, E.C., The Equity Premium: A

Puzzle, Journal of Monetary Economics, 15 (1985), 14561.

Mendoza, E.G., Real Business Cycles in a Small OpenEconomy, American Economic Review, 81 (1991), 797818.

Miller, S.M., Are Saving and Investment Cointegrated?,Economic Letters, 27 (1988), 3134.



20/20


Murphy, R.G., Capital Mobility and the Relationshipbetween Saving and Investment Rates, Journal of Inter-national Money and Finance, 3 (1984), 327 42.

Obstfeld, M., Capital Mobility in the World Economy:Theory and Measurement, Carnegie-Rochester Confer-ence Series on Public Policy, 24 (1986), 55 104.

Obstfeld, M., How Integrated Are World Capital Mar-

kets? Some New Tests, in G. Calvo et al. (Eds.), DebtStabilisation and Development, Basil Blackwell, 1989,13455.

Obstfeld, M., International Capital Mobility in the 1990s,in P.B. Kenen (Ed.), Understanding Interdependence:TheMacroeconomics of the Open Economy, Princeton: Prince-ton University Press, 1995.

Obstfeld, M. and Rogoff, K., The Intertemporal Ap-proach to the Current Account, in G.M. Grossman andK. Rogoff (Eds.), op.cit., 1995, 173199.

Peeters, M., The Public-Private Savings Mirror andCausality Relations among Private Savings, Investmentand (Twin) Deficits: A Full Modelling Approach,Discussion Paper No. 42, ESRC Macroeconmic Mod-elling Bureau, University of Warwick, 1996.

Penati, A. and Dooley, M., Current Account Imbalancesand Capital Formation in the Industrialised Countries,IMF Staff Papers, 31 (1984), 124.

Pesaran, M.H. and Smith, R.P., Estimating Long RunRelationships from Dynamic Heterogeneous Panels,Journal of Econometrics, 68 (1995), 79 113.

Sachs, J.D., The Current Account and MacroeconomicAdjustment in the 1970s, Brookings Papers in EconomicActivity, 12 (1981), 1089 92.

Sachs, J.D., The Current Account in the MacroeconomicAdjustment Process, Scandanavian Journal of Economics,84 (1982), 147159.

Sarno, L. and Taylor, M.P., Saving-Investment Correla-tions: Transitory versus Permanent, Mimeo, Liver-pool: Department of Economics and Accounting, Uni-versity of Liverpool, 1996.

Sinn, S., Saving-Investment Correlations and CapitalMobility: On the Evidence From Annual Data, Eco-nomic Journal, 102 (1992), 116270.

Stigler, G., The Conference Handbook, Journal of PoliticalEconomy, 85 (1977), 44143.

Stockman, A.C. and Tesar, L.L. Tastes and Technologyin a Two-Country Model of the Business Cycle: Ex-plaining International Comovements, American Eco-nomic Review, 85 (1995), 16885.

Summers, L., Tax Policy and Competitiveness, in J.

Frankel (Ed.), International Aspects of Fiscal Policy,Chicago, University of Chicago Press, 1988, 34975.

Swamy, P.A.V.B., Statistical Inference in Random CoefficientRegression Models, Lecture Notes in Operations Re-search and Mathematical Systems 55, Berlin: Springer,1971.

Taylor, M.P., The Economics of Exchange Rates, Journalof Economic Literature, 33 (1995), 13 47.

Tesar, L.L. Saving, Investment and International CapitalFlows, Journal of International Economics, 31 (1991),5578.

Tesar, L.L., International Risk-Sharing and Non-TradedGoods, Journal of International Economics, 35 (1993),6989.

Tesar, L.L. and Werner, I.M., Home Bias and High

Turnover, Journal of International Money and Finance, 14(1995), 467 93.

Tobin, J., Comment on Domestic Saving and Interna-tional Capital Movements in the Long Run and theShort Run, European Economic Review, 21 (1983), 15356.

van Wincoop, E. and Marrinan, J., Public and PrivateSaving and Investment, Seminar Paper No. 546, Insti-tute for International Economic Studies, 1993.

Westphal, U., Comment onDomestic Saving and Inter-national Capital Movements, in The Long Run andthe Short Run, European Economic Review, 21 (1983),15759.

Zellner, A., On the Aggregation Problem: A New Ap-proach to a Troublesome Problem, in K.A. Fox et al.

(Eds.), Economic Models, Estimation and Risk Program-ming: Essays in Honour of Gerhard Tintner, Berlin:Springer, 1969, 36578.


Feldstein Horioka Review

Documents

Transcript of Feldstein Horioka Review