The dynamics of emerging market equity ﬂows · The dynamics of emerging market equity ﬂows ... macro-economic stabilization programs ... (2002) 295–350 297

Journal of International Money and Finance21 (2002) 295–350

www.elsevier.com/locate/econbase

The dynamics of emerging market equity flows

G. Bekaerta,d, C.R. Harveyb,d,∗, R.L. Lumsdainec,d

a Columbia University, New York, 10027 USAb Duke University, Durham, NC 27708, USA

c Deutsche Bank, London, EC2N 2DB UKd National Bureau of Economic Research, Cambridge, MA 02138, USA

Abstract

We study the interrelationship between capital flows, returns, dividend yields and worldinterest rates in 20 emerging markets. We estimate a vector autoregression with these variablesto measure the degree to which lower interest rates contribute to increased capital flows andshocks in flows affect the cost of capital among other dynamic relations. We precede the VARanalysis by a detailed examination of endogenous break points in capital flows and the othervariables. These structural breaks are traced to the liberalization of emerging equity markets.Our evidence of structural breaks calls into question past research which estimates VAR mod-els over the full sample. After a liberalization, we find that equity flows increase by 1.4% ofmarket capitalization. We also show that shocks in equity flows initially increase returns whichis consistent with a price pressure hypothesis. While the effect is diminished over time, therealso appears to be a permanent impact. This is consistent with our finding that our proxy forthe cost of capital, the dividend yield, decreases. Finally, our analysis of the transition dyna-mics from pre-liberalization to post-liberalization suggests that when capital leaves, it leavesfaster than it came in. These results may help us understand the dynamics of the recent crisesin Latin America and East Asia. 2002 Published by Elsevier Science Ltd.

JEL classification: F3; F4; G1; E4

∗ Corresponding author. Tel.:+1-919-660-7768; fax:+1-919-660-8030.E-mail address: [email protected] (C.R. Harvey).

0261-5606/02/$ - see front matter 2002 Published by Elsevier Science Ltd.PII: S0261 -5606(02 )00001-3

Journal of International Money and FinancePublished in:

COLUMBIA BUSINESS SCHOOL 1

296 G. Bekaert et al. / Journal of International Money and Finance 21 (2002) 295–350

1. Introduction

With a number of recent crises in emerging markets, the role of foreign capitalin developing countries is under intense scrutiny once again. One country, Malaysia,imposed capital controls on October 1, 1998, in an effort to thwart the perceiveddestabilizing actions of foreign speculators. After a decade of capital market liberaliz-ations and increased portfolio flows into developing countries, this process may nowbe stalled or even reversed. The goal of this paper is to explore the dynamics, causesand consequences of capital flows in 20 emerging markets over the last 20 years.Importantly, we explicitly investigate the role of the recent financial liberalizationprocess in these dynamics.

Our work is related to two literatures. First, there is a growing body of researchthat studies the joint dynamics of capital flows and equity returns [see for example,Warther, 1995; Choe et al., 1999, Froot et al., 2001; Clark and Berko, 1997; Edelenand Warner, 1999; Stulz, 1999]. The first hypothesis of interest is whether foreigninvestors are “return chasers”, in the terms of Bohn and Tesar (1996), that is, areflows caused by changes in expected returns? A related hypothesis is that inter-national investors are momentum investors, leading to a positive relation betweenpast returns and flows. The second set of hypotheses focuses on the effect of flowson returns. Both Froot et al. (focusing on 28 emerging markets) and Clark and Berko(focusing on Mexico) find that increases in capital flows raise stock market prices,but the studies disagree on whether the effect is temporary or permanent. If theincrease in prices is temporary, it may be just a reflection of “price pressure”, whichhas also been documented for mutual fund flows and stock indices [Warther, 1995;Shleifer, 1986]. If the price increase is permanent, it may reflect a long-lastingdecrease in the cost of equity capital associated with the risk sharing benefits ofcapital market openings in emerging markets.

Our work is also related to a second literature on capital market liberalizationsand the integration process in emerging markets [see Bartolini and Drazen, 1997;Bekaert and Harvey, 1995, 2000a,b; Henry, 2000a,b; Kim and Singal, 2000]. Duringthe sample period, many emerging markets removed capital controls, which oftenwent hand in hand with other reforms in the domestic financial system, trade lib-eralization, macro-economic stabilization programs (especially in Latin-America)and large scale privatizations [see Bekaert and Harvey, 2000b for detailed time lineson important structural changes in emerging countries]. These structural changescomplicate any empirical analysis of emerging markets during this period, since theycould cause permanent or at least long-lasting changes in the data-generating pro-cesses. In Bekaert et al. (2002), we use the structural break methodology of Bai,Lumsdaine, and Stock (hereafter BLS) (1998) to “date” when market integrationoccurred and document structural changes in a number of financial and economictime-series.1

1 Kawakatsu and Morey (1999) also use endogenous break point techniques to date stock marketopenings and examine stock market efficiency before and after the opening.


297G. Bekaert et al. / Journal of International Money and Finance 21 (2002) 295–350

The main tool of analysis in this paper is a vector-autoregressive (VAR) frame-work as in Froot et al. (2001), but with a number of differences. First, we add twovariables to the bivariate set-up of returns and equity flows in Froot et al.: the worldinterest rate and local dividend yields. The low level of US interest rates has oftenbeen cited as one of the major reasons for increased capital flows to emerging mar-kets in 1993 [see World Bank, 1997; as well as Calvo et al. (1993, 1994) and Fernan-dez-Arias, 1996] and our framework will allow us to trace out the effects of anunexpected reduction in world interest rates on capital flows to emerging markets.We are ultimately interested in the effects of structural reforms in emerging marketson local returns and particularly, on the local cost of capital. The inclusion of theworld interest rate helps in that endeavor in that it removes the effect of exogenousglobal determinants of capital flows. We add dividend yields to the VAR as our costof capital measure, since they capture potential permanent price effects induced byincreased foreign capital after liberalizations better than average returns [see Bekaertand Harvey, 2000a] and they also serve as an indicator of expected returns allowingdifferentiation of the ‘return chasing’ and ‘momentum investing’ hypotheses.

Second, we precede our VAR analysis with a detailed endogenous break pointanalysis of our three main time series (net equity flows as a proportion of localmarket capitalization, log returns and the log dividend yield) using the novel tech-niques in BLS (1998) and Bai and Perron (1998a,b). This analysis helps pin downa relevant time-period over which to conduct the VAR analysis but is also interestingin its own right. For example, we study the transition dynamics of some of ourvariables around the break points. Such analysis is particularly important given thatrecent events in South-East Asia indicate that the integration process may now behalted and reversed. Studying capital flow dynamics and their impact on the localmarket may therefore yield predictions for the likely effects of the recent re-impo-sition of capital controls in some countries. Also, if capital market liberalizationsinduce one-time portfolio rebalancing on the part of global investors, one may expectnet flows to increase substantially after a liberalization and then to decrease again[see Bachetta and van Wincoop (2000) for a formal model generating such dynam-ics]. The Bai-Perron statistics look for multiple breaks in a time series and mayuncover such dynamics. We also test this prediction directly.

Third, although equity flows are our main focus, we also investigate how theyrelate to bond flows. For example, increased equity flows may substitute fordecreased bond flows or both may increase simultaneously as a result of a generalfinancial liberalization package.

We find that equity capital flows increase after liberalization but level out threeyears after their liberalization. This provides evidence that foreign investors rebalancetheir portfolios towards the newly available emerging market assets. Our analysis ofthe transition dynamics suggests that the movement of equity capital is much fasterwhen it leaves than when it enters. In general, we find sharply different results ifour models are estimated over the entire sample—which ignores a fundamental non-stationarity in the data—versus a post-break (liberalization) sample. One of our mainfindings is that unexpected equity flows are indeed associated with strong short-livedincreases in returns. However, we also find that they lead to permanent reductions



in dividend yields. As we explain later, the dividend yield change may reflect achange in the cost of capital. Hence, the reduction in the dividend yield suggeststhat additional flows reduce the cost of capital and that the actual return effect isnot a pure price pressure effect.

The paper is organized as follows. The second section presents the vector autore-gressive empirical model that we use to interpret the relation between expectedreturns, dividend yields and capital flows. The third section explores the methodsthat we use to establish structural breaks in the time-series of interest. The fourthsection describes the data. The results are presented in the fifth and sixth sections.The final section offers some concluding remarks.

2. The econometric framework

2.1. Variables and main hypotheses

Since there is currently no well-accepted model of transition dynamics around aliberalization, we conduct our empirical analysis in the context of vector autore-gressions (VARs). For part of our analysis, these VARs are reduced-form represen-tations of an unspecified structural model, but some of the empirical hypotheses wewant to test require a structural interpretation of the VARs. In our primary empiricalsystem, we generalize the set-up of Froot et al. (2001), who run a bivariate VARon the ratio of net capital flows to market capitalization and market returns. Theirmain identifying structural restriction is that the shock to flows may affect returnsbut not vice versa. In addition, past returns only affect current returns through theireffects on flows. Two interesting questions can be investigated in this framework:

1. What is the effect of an unexpected shock to capital flows on current returns andwhat are its dynamics (that is, does it die out or is it permanent)? The dynamiceffects of the shock serve to distinguish the price pressure hypothesis from thepermanent change in the cost of capital hypothesis.

2. Do past returns affect current capital flows? In particular, Bohn and Tesar (1996)argue that capital flows are motivated by capital “chasing” high expected returns,rather than portfolio rebalancing motives. One issue here is that high past returnsneed not signal high future returns, unless momentum is an important determinantof expected return (see Bohn and Tesar (1997). In our framework, we will beable to distinguish the expected return-updating hypothesis from the momentumhypothesis.

In contrast to previous work, our primary VAR will contain four variables. LetYt � [it,nft,dyt,rt]�, whereit is the world interest rate,nft is the net equity capital flowdivided by market capitalization,dyt is the log-dividend yield andrt is the loggedequity return.

The presence of the world interest rate allows a more subtle testing of the hypoth-eses in (1). It has often been argued that the emerging markets received a lot of US



capital in 1993 because investors were chasing higher yielding assets with interestrates in the US reaching historical lows. It should be mentioned that there may begood reasons for an inverse link between US interest rates and capital flows to emerg-ing markets. For example, the low US interest rates may have increased the Amer-icans’ wealth and therefore increased their risk tolerance, leading them to rebalancetowards riskier emerging market securities. Whatever the reason, our framework willallow a direct test of the magnitude and dynamics of a decrease in the world interestrate on capital flows to emerging markets.

With the world interest rate in the system, we can now divide the effect of highercapital flows on returns into two components. Capital flow increases induced bylower world interest rates may, for example, be less likely to lead to permanent priceincreases than capital flow increases that are not caused by world factors, but alsomay reflect portfolio rebalancing after a capital market liberalization.

The addition of the log-dividend yield is motivated by the work of Bekaert andHarvey (2000a). They argue that the extreme volatility in emerging market returnsimplies that changes in the cost of capital can be better assessed by investigatingchanges in dividend yields. In any rational pricing model, dividend yields will bedecreasing in the growth rate of dividends (cash flows) and increasing in the discountrate. Because of their low variability, dividend yields will capture permanent priceeffects induced by cost of capital changes more accurately than average returns.Bekaert and Harvey document that liberalizations tend to lead to small drops individend yields. One problem is that a lower dividend yield may also reflect animprovement in growth opportunities. Nevertheless, our set-up will allow us to testthe effect of a change in the world interest rate, and/or capital flows on the dividendyield in emerging markets and contrast that with the effect on returns. In addition,given that the dividend yield is a good proxy for expected returns, which is alsoborne out in the predictability tests for emerging markets by Harvey (1995), itsinclusion allows for a proper test of the Bohn-Tesar hypotheses mentioned above.

To perform tests of the main hypotheses, we use impulse response analysis basedon a structural interpretation of the VAR. Consider, without loss of generality, afirst-order VAR, suppressing the constant:

Yt � AYt�1 � et, (1)

with all eigenvalues ofA having moduli less than one so that the VAR is stationary.We compute impulse responses,IR(i,j,k) � ∂e�iYt�k /∂e∗t,j, wheree∗t,j are the “struc-

tural” shocks andei is an indicator variable selecting theith variable, e.g.e1 =[1,0,0,0]. We look at one standard deviation shocks. The structural shocks,e∗

t,j, aredetermined by the ordering in the VAR, that is,et � P�e∗t , whereP is an upper-triangular matrix ande∗t are uncorrelated structural shocks, such that� �E[et e�t] � P�P.

The ordering of the variables is as defined above, with the world interest ratecoming first. Hence, the world interest residual is implicitly assumed not to be affec-ted by the other shocks in the system. As in Froot et al. (2001), we order flowsbefore returns, but we insert the dividend yield in the middle. Dividend yields andreturns are contemporaneously negatively correlated, but a shock to the dividend



yield may reflect a near-permanent price change due to the liberalization process ora change in expected returns, and it is therefore natural to order dividend yieldsbefore returns. We are interested in:

1. the effects of the world interest rate, it, on the ratio of net capital flows to marketcapitalization, returns and dividend yields,

2. the impact of flows on returns and dividend yields to test the price pressure versuspermanent impact hypotheses, and

3. the effect of past returns and dividend yields on flows to test ‘return chasing’ and‘momentum investing’.

For this last response, our setup removes the contemporaneous correlation betweenreturns or dividend yields and flows, ascribed potentially to price pressure effects,because flows are ordered before these two variables. For example, a shock to thedividend yield not contemporaneously correlated with capital flows may reflect achange in growth opportunities or a change in expected returns which may affectfuture foreign capital inflows.

We also report impulse responses of shocks to the world interest rate, currentcapital flows, returns, and dividend yields on two cumulative variables: the averagelong-run return and the change in US holdings.

First, we definert�k,k � 1k�k�1

i � 0 rt�1�i as the per period cumulative long-horizon

return. Next, consider the definition of net capital flows to capitalization:

nft �ft

mcapt

,

where ft is the net capital flow andmcapt is the equity market capitalization. Thecumulative capital flow to market capitalization is:

cft,t+k � �k

i � 1

nf t+i

�ft+1

mcapt+1

� % �ft+k

mcapt+k

�1

mcapt+k�ft+1 ×

mcapt+k

mcapt+1

� % � ft+k�.

So,cft,t+k accumulates the flows occurring betweent andt+k and allows each flowto change value as a result of the market return.

Now consider the definition of the cumulative holdings in the local market:



ht � ��t

i�0

fimcapt

mcapi� 1

mcapt

� �t

i�0

fimcapi

� �t

i�0

nfi.

So holdings accumulate the flows from the beginning of time (notice the counterbegins ati = 0) allowing for the market return and express them as a proportion ofcurrent market capitalization. From this analysis, it is immediate that

cft,t+k � ht+k�ht,

is the change in holdings. Whereas impulse responses on the VAR variables die outin any stationary VAR, these cumulative effects represent the permanent effect onreturns and capital flows when we letk go to infinity.

2.2. VAR dynamics

Before we test the main hypotheses, we document the information in the VARsregarding the dynamic relations between our four variables using two different stat-istics.

First, we investigate dynamic regression coefficients between the various variables,for example, what is the correlation between world interest rates today and capitalflows or returns in the future? Formally, we investigate:

bi,j,k �Cov [e�iYt+k, e�jYt]

Var[e�jYt], (2)

whereei are indicator vectors, e.g.e1 = [1,0,0,0].Given that the VAR is stationary, Var[Yt] = C(0) can be computed as

Vec[C(0)] � [I�A�A]�1vec(�), (3)

where � � E[et e�t] and I is the identity matrix. The VAR fully summarizes theshort-run and long-run dynamics ofYt ; e.g.

E[YtYt�k] � C(k) � AkC(0). (4)

With this information, projection coefficients at all horizons can be computed.With the long-run variables introduced above, changes in holdings and long-run

returns, we can investigate the relation between these variables on the one hand andworld interest rates and capital flows on the other hand. For example, we compute:



gk �Cov [rt+k,k, nft]

Var[nft]

�

1ke�4[I � A � % � Ak�1]C(0)e�2

e�2C(0)e2,

(5)

wheree2 (e4) are indicator variables selecting flows (returns). Also, [I+A+%+Ak�1]= [I�Ak][I�A]�1. We can also letk go to infinity here. Analogously, we can investi-gate the long-run beta of changes in holdings with respect to current interest rates,returns and dividend yields.

Second, whereas the regression coefficients provide useful summary information,they are univariate relations that may hide intricate dynamic patterns. For example,there may be a positive relation between current capital flows and future returns, butpart of this correlation may come indirectly through the effect of world interest rateson capital flows. It would be interesting to see whether there is still a relation betweencapital flows and future returns, controlling for the world interest rate effect. Simi-larly, are capital flows mostly predictable by external variables like world interestrates, or by internal variables (returns and dividend yields) that may proxy forexpected returns for example? This question has been addressed before [see Calvoet al., 1993, 1994; and Fernandez-Arias, 1996], but in the context of our VAR it isparticularly simple to implement. We conduct a series of Granger-causality tests,testing whether world interest rates Granger-cause the other variables, and whetherreturns and dividend yields Granger-cause capital flows or vice versa. If liberalizationis a gradual process and is pre-announced, dividend yields may decrease and returnstemporarily increase before flows pick up. In this case, returns and dividend yieldsmay Granger-cause flows. Hence, it may be hard to distinguish this effect frommomentum investing (positive feedback trading), since this would also imply thatpositive returns predict higher flows.

2.3. The dynamics of capital flows and breaks

It does not make much sense to conduct this analysis over the full sample of data,given that many of the markets that we study may have undergone an integrationprocess somewhere in the middle of the sample. If the market truly went from seg-mented to integrated, the dynamics of all the variables in our VAR except the worldinterest rate would be affected. Consider Fig. 1 which sketches what a standardmodel of risk sharing [see the description in Bekaert et al., 2002] would predict,namely “a permanent change in prices leading to a new regime of lower expectedreturns”. Interestingly, this process is associated with short-term increases in returns(the return to integration) and long-run decreases in expected returns, plus a perma-nent decrease in dividend yields. This pattern would more generally result from“investor base broadening”, the spreading of risks among more investors (see Stulz,1999, for a summary of the scarce evidence to date on this). If flows drive pricestemporarily away from fundamentals, the price effect ought to be temporary (“theprice pressure hypothesis”). This behavior can be examined in the context of our



Fig. 1. Asset prices and market integration.

VAR by contrasting the effects of capital flows onrt, rt+k,k and dyt. We expect apositive (negative) contemporaneous relation between returns (dividend yields) andunexpected flows that is permanent under the first but transitory and reversed underthe second hypothesis.

Fig. 1 also suggests a potential scenario for capital flows. Flows are of coursevirtually zero before the liberalization. After the liberalization, which may be gradual,large inflows occur as foreign investors include the emerging market into their port-folios. However, once the rebalancing is accomplished, net flows need no longer bepositive. Bachetta and van Wincoop (2000) model the dynamics of capital flows byallowing for a gradual decrease in a tax on investments into an emerging marketand show how capital flows can over-shoot. Of course, this is a simple story andthere are many competing models for the behavior of emerging market capital flows



including models that predict a foreign lending boom followed by the inevitablecrash! [see, for example, Calvo and Mendoza, 1998.]

To deal with this problem, we apply the most recent methodology on break dateinference. We first apply the methodology developed in BLS to all of our univariateseries, and also use it to determine a break date for the joint system. This breakpoint analysis then determines the post-break period to which we will apply ourVAR analysis described above. The break point analysis also reveals a period oftransitions and the transition dynamics are of interest in their own right.

A disadvantage of the BLS methodology is that it only allows for one break. Thereare a number of reasons why there may be more than one break especially in thenet flows series. As indicated above, flows may be temporarily high to effect aportfolio rebalancing after capital market integration. There may be a second breakat the end of this process. Recently, we have seen a reversal of capital flows witha number of well-publicized crises in Mexico 1994–1995, and in South-East Asiain 1997–98. Even much before this, the debt crisis may have caused some LatinAmerican markets to become effectively segmented from the rest of the world,although capital flows before then were small. Therefore we also apply the techniquesof Bai and Perron (1998a,b) which allow for multiple breaks, but only apply tounivariate series. Additional details are presented in the econometrics section.

We also investigate the dynamics of capital flows around a potential liberalizationbreak using a very simple regression procedure. If the capital flow story of Fig. 1is accurate, mean equity flows should be higher after the break than before, butdecrease again after portfolio re-balancing is completed. In other words, letD1t bea dummy that comes on after the break andD2t the dummy that comes on threeyears after the break, then in the regression

nft � a � bD1t � cD2t � et, (6)

we would findb to be positive, andc to be negative. We choose three years becauseof the time it takes from announcement to effective implementation of a marketliberalization. Bekaert and Harvey (2000a) provide evidence that liberalizations areoften gradual.

Finally, to examine the transition dynamics, we compute a statistic we call the“transition half-life statistic” (THL). The THL statistic is measured as follows. Con-sider the point in time at which the break occurs and imagine the current realizationof the variable is at the unconditional meanbefore the break. Now consider fore-castingk periods in the future using the new dynamics. If the VAR is stationary,eventually the forecasts will reach the new unconditional mean. How fast they willget there depends on the persistence of the system and how far away the post-breakmean is from the starting point (pre-break mean). Our THL statistic records the timeit takes (in months) to reach half the distance between old and new mean. We canalso reverse the computation. That is, we compute the THL statistic starting fromthe new mean going to the old mean using the pre-break dynamics. Comparing thetwo statistics is informative about the different dynamics before and after the break.We will compute this statistic for capital flows and dividend yields. For capital flows,a bold interpretation of the pre-break THL statistic is that it reveals something about



how capital flows will react when the integration process is reversed, as is recentlyhappening in a few countries. We do not compute the THL statistic for returns,because we conjecture that the measurement of mean returns is too noisy to makethe computation valuable.

3. Data

Our data consists of capital flows, interest rates, dividend yields, and returns. Oursource of monthly data on capital flows is the US Treasury International Capital(TIC) reporting system.2 For the 20 emerging markets we study, we are able tocalculate the net US flows for stocks and bonds for 17 countries. The Treasury doesnot track data on Jordan, Nigeria, and Zimbabwe. We use the International FinanceCorporation’s Emerging Markets Database as a source of the US dollar returns andthe dividend yields. The world interest rate is constructed as a GDP weighted averageof the short-term government Treasury bills in the G-7 countries.3 The interest rateand GDP data are from Datastream.

With so many countries and relatively small sample periods, a country by countryanalysis may both lack power and prevent us from presenting the results in an intelli-gible way. Therefore we present most of our results using country groupings. Weuse three different types of groupings. The first set is aimed primarily at noisereduction. We aggregate results over all countries with three weighting schemes:equally weighted, value weighted (using the market capitalization of the equity mar-ket from the IFC) and volatility weighted. The volatility weighting constructs weightsusing the inverse of the sum of squared residuals of all the regressions in the VARfor a particular country relative to the inverse sum of the squared residuals overall countries. Hence, the “noisiest” VARs are down-weighted. Before applying thisprocedure, we re-scaled the residuals, so that at a global, all-country level, capitalflows accounted for 40% of the total variance, returns and dividend yields each 25%,and interest rates, 10%. If the coefficients (like impulse responses) are independentacross countries, this aggregation would lead to a reduction of the typical country-specific standard error by a factor of over four, since there are 17 countries in oursample. Hence, even if the country-specific standard errors are double the size ofthe coefficients, we would obtain significance.

The second grouping is geographical. We contrast the results for six Latin Amer-ican countries (Argentina, Brazil, Chile, Colombia, Mexico and Venezuela) and sixAsian countries (Indonesia, Korea, Malaysia, the Philippines, Taiwan and Thailand).Although these results may be sensitive to country-specific outliers, the contrastsbetween the development models of Latin-America and South-East Asia and therecent crises in both areas make this grouping meaningful.

2 See Tesar and Werner (1994, 1995) for a description and analysis of these data.3 See the data appendix for additional details on the construction of the dividend yields and the world

interest rate.



Our last groupings focus on the characteristics of capital flows. Table 1 summar-izes some of the characteristics we use in the selection process. First, we investigatethe magnitude of equity and bond flows. We calculate the return-adjusted cumulativeequity flows divided by market capitalization. This is a measure of US ownershipin the country. We present average ownership over the full sample as well as the1990s. The largest average ownership is found for Mexico (since 1990) followed byBrazil and Argentina. The country with the largest average ownership in Asia isThailand in the 1990s. To allow comparison with bond flows, where return adjust-ments and market capitalizations are not available, we also calculate the cumulativeequity flows to GDP without return adjustments. In this analysis, Mexico, Chile andMalaysia have the largest cumulative U.S. equity flows to GDP. Bond flows to GDPare presented in the next column. The highest averages of the cumulative bond flowsto GDP are found in Mexico, Venezuela and Argentina. Clearly, cumulative net bondflows are much smaller than equity flows. By adding cumulative equity and bondflows, we obtain a measure of total external financing through portfolio capital. Wealso report sample averages and averages in the 1990s. There are two countries withnet negative external financing, Chile and Taiwan, both, not surprisingly, countrieswith stringent capital controls. Mexico, Venezuela and Malaysia have the highestexternal financing to GDP. Generally, both equity and bond flows are larger in thenineties than for the full sample, and total external financing is larger in the ninetiesfor 15 out of 17 countries.

We now consider different groups of countries based on this information for the1990s sample. First, we rank the countries according to the importance of externalfinance. In addition to the three countries previously mentioned (Mexico, Venezuelaand Malaysia), we add Argentina, Brazil and Korea to the top group. The bottomsix countries are, in addition to Chile and Taiwan, India, Greece, the Philippines andPakistan, all with less than 0.35% of GDP in external financing through US bondand equity flows.

Finally, we want to select countries that primarily rely on equity capital and coun-tries that primarily rely on fixed income. This is not trivial to do, since for somecountries the absolute flows may be very small and, in particular, for bonds, theymay be negative. Our approach was to rank countries based on the difference betweenthe average cumulative post-1989 equity and bond flows to GDP. The bottom sixcountries are deemed bond-reliant, the top six equity reliant. We exclude countrieswhen they do not place in the top 10 ranked according to cumulative bond flows orcumulative equity flows to GDP, respectively. The countries relying primarily onequity are Chile, the Philippines, Malaysia, Portugal, Thailand and Korea. Weexcluded Taiwan because of its insignificant absolute equity flows. Although Mexico,Brazil and Argentina are top countries in terms of equity flows to GDP, they appearin the fixed income group because of their substantial positive fixed income flows.The other three countries are Venezuela, Pakistan, and Indonesia. Because of therelative non-importance of bond flows in general, it is possible that countries groupedin the “rely on bonds” category receive more equity than bond capital.

At the bottom of the table, we present country groupings. We find that LatinAmerican countries tend to have high US equity ownership. Asian countries have



Tab

le1

Sum

mar

yan

alys

isof

net

equi

tyflo

ws

and

net

bond

flow

s

Cou

ntry

Ave

rage

ofcu

mul

ativ

eA

vera

geof

cum

ulat

ive

Ave

rage

ofcu

mul

ativ

eA

vera

geof

tota

lE

quity

flow

H0:

Equ

ityflo

ws

H 0:

Bon

dflo

ws

equi

tyflo

ws/

equi

tyeq

uity

flow

s/G

DP

bond

flow

s/G

DP

cum

ulat

ive

exte

rnal

and

bond

flow

dono

tG

rang

erdo

not

Gra

nger

capi

taliz

atio

nfin

anci

ngto

GD

Pco

rrel

atio

nca

use

bond

flow

sca

use

equi

tyflo

ws

Ful

lP

ost-

1990

Ful

lP

ost-

1990

Ful

lP

ost-

1990

Ful

lP

ost-

1990

P-v

alue

P-v

alue

Arg

entin

a0.

0093

0.08

370.

0035

0.00

900.

0040

0.01

040.

0074

0.01

940.

055

0.55

80.

114

Bra

zil

0.05

190.

1271

0.00

370.

0093

0.00

300.

0077

0.00

670.

0170

0.01

30.

000

0.16

5C

hile

0.02

330.

0578

0.00

900.

0226

�0.

0326

�0.

0642

�0.

0236

�0.

0416

�0.

049

0.03

60.

897

Col

ombi

a0.

0137

0.03

230.

0017

0.00

44�

0.00

010.

0022

0.00

150.

0065

0.18

90.

000

0.00

0G

reec

e0.

0012

0.01

730.

0008

0.00

23�

0.00

09�

0.00

09�

0.00

010.

0014

0.04

60.

262

0.01

9In

dia

0.00

480.

0112

0.00

080.

0019

�0.

0010

�0.

0009

�0.

0002

0.00

110.

195

0.00

00.

000

Indo

nesi

a0.

0302

0.07

640.

0027

0.00

690.

0023

0.00

580.

0050

0.01

27�0.

007

0.83

40.

181

Jord

ann.

a.n.

a.n.

a.n.

a.n.

a.n.

a.n.

a.n.

a.n.

a.n.

a.n.

a.K

orea

0.04

170.

0781

0.00

370.

0088

0.00

160.

0056

0.00

530.

0144

0.07

10.

000

0.00

0M

alay

sia

0.01

150.

0263

0.00

880.

0210

�0.

0058

0.00

040.

0030

0.02

14�

0.13

50.

996

0.76

3M

exic

o0.

1186

0.24

110.

0104

0.02

530.

0196

0.04

660.

0300

0.07

190.

103

0.97

30.

000

Nig

eria

n.a.

n.a.

n.a.

n.a.

n.a.

n.a.

n.a.

n.a.

n.a.

n.a.

n.a.

Pak

ista

n0.

0058

0.01

480.

0007

0.00

180.

0006

0.00

140.

0012

0.00

320.

110

0.00

00.

100

Phi

lippi

nes

0.01

760.

0444

0.00

460.

0115

�0.

0052

�0.

0101

�0.

0006

0.00

140.

035

0.83

20.

998

Por

tuga

l0.

0284

0.07

010.

0028

0.00

70�

0.00

16�

0.00

330.

0012

0.00

370.

043

0.00

00.

002

Tai

wan

0.00

100.

0020

0.00

010.

0004

�0.

0110

�0.

0278

�0.

0109

�0.

0274

0.02

20.

973

0.95

3T

haila

nd0.

0320

0.06

670.

0026

0.00

59�

0.00

05�

0.00

010.

0021

0.00

58�

0.06

30.

013

0.00

1T

urke

y0.

0188

0.04

770.

0010

0.00

280.

0003

0.00

100.

0014

0.00

38�0.

169

0.61

10.

441

Ven

ezue

la�

0.00

49�

0.00

910.

0004

0.00

110.

0139

0.03

270.

0144

0.03

380.

0070

0.99

00.

892

Zim

babw

en.

a.n.

a.n.

a.n.

a.n.

a.n.

a.n.

a.n.

a.n.

a.n.

a.n.

a.(c

onti

nued

onne

xtpa

ge)



Tab

le1

(cont

inue

d)

Cou

ntry

Ave

rage

ofcu

mul

ativ

eA

vera

geof

cum

ulat

ive

Ave

rage

ofcu

mul

ativ

eA

vera

geof

tota

lE

quity

flow

H0:

Equ

ityflo

ws

H 0:

Bon

dflo

ws

equi

tyflo

ws/

equi

tyeq

uity

flow

s/G

DP

bond

flow

s/G

DP

cum

ulat

ive

exte

rnal

and

bond

flow

dono

tG

rang

erdo

not

Gra

nger

capi

taliz

atio

nfin

anci

ngto

GD

Pco

rrel

atio

nca

use

bond

flow

sca

use

equi

tyflo

ws

Ful

lP

ost-

1990

Ful

lP

ost-

1990

Ful

lP

ost-

1990

Ful

lP

ost-

1990

P-v

alue

P-v

alue

Ful

l0.

0238

0.05

810.

0034

0.00

83�

0.00

080.

0004

0.00

260.

0087

0.02

7A

sia

0.02

230.

0490

0.00

370.

0091

�0.

0031

�0.

0044

0.00

060.

0047

�0.

013

Latin

0.03

530.

0888

0.00

480.

0119

0.00

130.

0059

0.00

610.

0178

0.05

3R

ely

onst

ocks

0.02

570.

0572

0.00

520.

0128

�0.

0073

�0.

0119

�0.

0021

0.00

09�

0.01

6R

ely

onbo

nds

0.03

520.

0890

0.00

360.

0089

0.00

720.

0174

0.01

080.

0263

0.04

7T

opex

tern

al0.

0380

0.09

120.

0051

0.01

240.

0060

0.01

720.

0111

0.02

960.

019

Bot

tom

exte

rnal

0.00

890.

0246

0.00

270.

0067

�0.

0083

�0.

0171

�0.

0057

�0.

0103

0.06

0

Ful

lre

pres

ents

aneq

ually

-wei

ghte

dav

erag

eac

ross

all

coun

trie

s.A

sia

isan

equa

lly-w

eigh

ted

aver

age

acro

sssi

xA

sian

coun

trie

s(I

ndon

esia

,K

ore

a,M

alay

sia,

the

Phi

lippi

nes,

Tai

wan

and

Tha

iland

),an

dLa

tinis

equa

lly-w

eigh

ted

aver

age

acro

sssi

xLa

tinA

mer

ican

coun

trie

s(A

rgen

tina,

Bra

zil,

Chi

le,

Col

ombi

a,M

exic

oan

dV

enez

uela

).W

eal

sore

port

equa

lly-w

eigh

ted

acro

ssth

esi

xco

untr

ies

that

rely

mos

ton

equi

tyfin

anci

ng(C

hile

,th

eP

hilip

pine

s,M

alay

sia,

Por

tuga

l,T

haila

ndan

dK

orea

),eq

ually

-wei

ghte

dac

ross

the

six

coun

trie

sth

atre

lym

ost

onbo

ndfin

anci

ng(V

enez

uela

,M

exic

o,A

rgen

tina,

Pak

ista

n,In

done

sia

and

Bra

zil),

equa

lly-w

eigh

ted

acro

ssth

esi

xco

untr

ies

that

have

the

high

est

prop

ort

ion

ofex

tern

alfin

anci

ng,

that

is,

the

bond

plus

equi

tyca

pita

lflow

sto

GD

P(M

exic

o,M

alay

sia,

Ven

ezue

la,

Arg

entin

a,B

razi

land

Kor

ea)

and

the

six

coun

trie

sw

ithth

elo

wes

tpr

opor

tion

ofex

tern

alfin

anci

ngto

GD

P(C

hile

,T

aiw

an,

Pak

ista

n,th

eP

hilip

pine

s,G

reec

ean

dIn

dia)

.T

heco

rrel

atio

nsar

ees

timat

edov

erth

efu

llsa

mpl

e.T

hebr

eak

date

sfo

rea

chco

untr

yar

epr

esen

ted

inT

able

2.T

heG

rang

er-c

ausa

lity

test

sar

efr

oma

biva

riate

VA

Rof

equi

tyre

turn

san

dbo

ndre

turn

ses

timat

edov

erth

efu

llsa

mpl

e.



had sharply negative bond outflows in the 1990s perhaps due in part to the Asiancrisis. Latin American countries tend to rely more on equity financing than Asiancountries. The countries grouped in the “relying on bonds” category are generallycountries that heavily rely on external portfolio capital, including equity capital. Asa proportion of market capitalization, these countries have higher U.S. holdings thanthe “stock reliant” countries.

Table 1 presents some additional information on the relation between bond andstock flows. Given that our analysis focuses on equity flows, it is important to knowwhether bond flows are substitutes for equity flows. We present the correlation ofthe net capital flows which is, in general, small. The highest correlation is found forIndia and Colombia. The largest negative correlations are found for Turkey andMalaysia. In the majority of the countries, the correlation is positive, and of the fivenegative correlations, three are in South-East Asian countries.

We also present Granger-causality tests based on a bivariate system of stock flowsto market capitalization and bond flows to GDP. We can reject the hypothesis at the5% level that equity flows do not Granger-cause bond flows in Brazil, Chile, Colom-bia, India, Korea, Pakistan, Portugal and Thailand. We can reject the hypothesis atthe 5% level that bond flows do not Granger-cause equity flows in Colombia, Greece,Indonesia, Korea, Mexico, Portugal and Thailand. Hence, there are five countries forwhich there are significant predictive relations in both directions, potentially suggest-ing the effect of a third variable on general capital flows or persistence in capitalflows after a capital market liberalization. There is no support for a consistent patternwhere bonds always lead equity (or vice versa) or where bonds and equity areclear substitutes.

Fig. 2 presents the results from the impulse response analysis based on the bivari-ate VAR. We examine value- and equally-weighted impulse responses as well asthe regional groupings. There are two interesting observations. First, positive shocksto stock (bond) flows are followed by positive responses in bond (stock) flows.Again, this is potentially consistent with gradual portfolio rebalancing towardsemerging markets in general (both equities and bonds, after a capital market liberaliz-ation or induced by changes in world interest rates, for example). Second, notice thedistinction between Latin America and Asia. A shock to equity flows in Latin Amer-ica has a much larger short-term impact on bond flows than it does in Asia. Moreover,there is an initial slightly negative effect of bond flows on stock flows in Latin

Fig. 2. The relation between net bond and net equity flows.



America whereas the effect in Asia is positive. From the second period onwards,the effects are positive and very similar in magnitude across the two regions. Fromboth graphs, it is clear that the effects die out within six months. We also conducteda break point analysis of bond and equity flows. In the countries that experiencedsignificant breaks in equities and bond capital flows, the break in bonds precededthe break in equities in four countries (Argentina, Korea, Malaysia and Thailand).The break in equities preceded the break in bonds in five countries (Brazil, Colombia,Greece, India and the Philippines). With the exception of one country, the Philip-pines, the equity and bond breaks are clustered closely in time for all of thesecountries.4

Taken together, our results suggest that bond and equity flows are not substitutesbut both increased in the nineties with the liberalization process.

4. Breaks and initial pre and post-break analysis

4.1. Break analysis

To select an appropriate break date for use in subsequent analysis, we rely onseveral pieces of information. First, we use the results of the univariate break analysisfor three series: returns, dividend yields and net equity flows. We compute themedian break date, the 90% confidence interval for the date in months as well as astatistic that provides a test of the null hypothesis that no break occurred. Detailedresults on all break tests are reported in Appendix Tables 1–3.

Second, we conduct a multivariate analysis of breaks using the BLS framework.We investigate three specifications. The first is a trivariate system with equity capitalflows, dividend yields and returns. The second specification is a quadravariate systemthat includes an equation for the world interest rate. However, the coefficients in theworld interest rate equation are not allowed to break in the estimation but the depen-dence of the other variables in the system on the world interest rate may break. Thethird specification adds the bond capital flows to the system of equations to see ifthe breaks we find are affected by this addition. For 13 of the 17 countries, themultivariate breaks fall within the range of the univariate breaks for either bond orequity flows. For four countries, (Greece, Mexico, Portugal and Taiwan), the breakdates correspond to the break dates in dividend yields or returns, but the confidenceintervals for the breaks are always tighter in the multivariate estimation consistentwith what the theory would predict. Finally, the break dates from the three multivari-ate systems are very close to each other.

Third, we conduct the Bai-Perron tests that allow for multiple breaks. In a numberof countries more than one break occurs. For example, in the dividend yield esti-mation for Mexico there are three significant breaks: January 1983, July 1986 andMarch 1991. The first date closely follows the onset of the Latin American debt

4 See Appendix Table 1.



crisis. The second date closely follows the abolition of the official exchange rateand coincides with major debt restructuring. The final date closely follows the privat-ization of Telmex and the beginning of the NAFTA negotiations. For the flows esti-mation, we find little evidence in favor of multiple breaks, although this may be dueto low power. Argentina shows two breaks in equity flows, with the second onebeing detected in the BLS tests. There are three breaks for equity flows in Korea,partially reflecting the gradual lifting of foreign ownership. Finally, there are twobreak dates for bond flows in Taiwan, one preceding and one following the liberaliz-ation plan of January 1991. Hence, there appears to be a pattern in the multiplebreak analysis that is linked to important events tied to either deliberalization orliberalization of capital markets.

Finally, we link the statistical breaks to the economic events detailed in Bekaertand Harvey (2000b). The end-result is presented in Table 2, where we also brieflyindicate the motivation for each break date. All the dates we choose are dates foundby one of the break analyses we conducted, but the final choice uses exogenousinformation on regulatory reforms to make sure we use a relevant break date. Werely heavily on the break dates that arise from the quadravariate system as well asthe Bai-Perron breaks for dividend yields and equity flows. For example, the Bai-Perron break for equity capital flows in Thailand is August 1988. This closely corre-sponds to the opening of the market to foreigners by the creation of the Alien Boardfor trading in late 1987. The quadravariate and quintravariate date for Taiwan isApril 1988. This closely corresponds to the lifting of exchange controls. Pakistan isparticularly interesting. Most consider the official liberalization to be February 1991.The Bai-Perron break in dividend yields is earlier, in December 1990. However,an examination of the chronology shows that the liberalization was announced inNovember 1990.5

4.2. The impact of breaks on unconditional means

Table 3 presents an analysis of both the means and standard deviations of the fourcountry-specific variables that we study in the VARs: dividend yields, log returns,net bond flows to GDP and net equity flows to equity market capitalization. Individ-ual country results and results for our country groupings are presented for the fullsample, as well as the pre- and post-break periods using the dates in Table 2.

There are a number of interesting differences between the pre-break and post-break periods. The dividend yields are sharply lower on average (2.5 in post-breakand 4.7 in pre-break). The yields decrease in 13 of 17 countries. This is consistentwith the idea that expected returns decrease after a liberalization. Although returnsare much noisier than dividend yields, we also observe that the average return

5 We also closely examined the results of estimations with a smaller trimming factor, which suggestedbreaks near the end of the sample that are associated with the Asian crisis. Since these dates are verymuch near the end of our sample, we did not exclude post-break observations.


312 G. Bekaert et al. / Journal of International Money and Finance 21 (2002) 295–350T

able

2S

elec

tion

ofbr

eak

date

s

Cou

ntry

Dat

eS

erie

sR

easo

n

Arg

entin

aM

ay92

dy-B

PA

lso

dy-U

.C

lose

toE

q.flo

ws-

BP

(Dec

92).

Fol

low

sin

trod

uctio

nof

Arg

entin

eF

und

(Oct

91)

whi

chis

the

first

time

US

inve

stor

sco

uld

acce

ssth

ism

arke

tin

mod

ern

times

.

Bra

zil

Mar

90Q

uad

Clo

seto

dy-U

(Nov

90)

and

Eq.

Flo

ws-

U(J

an90

).O

ffici

alis

May

91.

Exa

ctly

coin

cide

sw

ithin

trod

uctio

nof

Col

lor

Pla

n(M

ar90

).

Chi

leN

ov90

dy-B

PA

bout

one

year

from

Offi

cial

(Jan

92).

Qua

did

entifi

esth

ede

btcr

isis

.S

hort

lyfo

llow

sin

trod

uctio

nof

refo

rmpa

ckag

e(A

pr90

),m

ajor

AD

R(J

uly

90)

and

rene

wal

ofA

ndea

nP

act

(Nov

90).

Col

ombi

aO

ct91

dy-B

P1

Clo

seto

Offi

cial

(Feb

91).

Eq.

Flo

ws-

Ufo

llow

s(J

uly

93).

Exa

ctm

onth

that

Pes

ode

regu

late

d(O

ct91

).In

addi

tion,

exac

tm

onth

that

fore

ign

coun

try

fund

sar

eal

low

edto

have

upto

10%

ofow

ners

hip

and

fore

ign

firm

sar

epe

rmitt

edto

rem

it10

0%of

profi

ts.

Gre

ece

Jul

88Q

uad

Als

ody

-BP

date

.O

ffici

al(D

ec87

).C

lose

lyco

inci

des

with

date

offir

stA

DR

(Aug

88).

Indi

aA

pr92

Qua

dB

etw

een

dy-U

(Jan

92)

and

Eq.

Flo

ws-

U(M

ay93

).O

ffici

alis

Nov

92.

Clo

sely

follo

ws

esta

blis

hmen

tof

Sec

uriti

esE

xcha

nge

Boa

rd(J

an92

).In

addi

tion,

the

first

AD

Ris

Feb

92.

Indo

nesi

aM

ay92

dy-B

PO

ffici

alS

ep89

.E

quity

and

bond

flow

sbr

eak

muc

hla

ter.

Firs

tA

DR

laun

ched

Feb

92.

For

eign

Boa

rdfo

rtr

adin

gby

fore

igne

rses

tabl

ishe

din

July

92.

Jord

anA

pr92

dy-B

PO

ffici

alis

Dec

95.

Clo

sely

prec

edes

the

liftin

gof

cont

rols

onou

tbou

ndan

din

boun

ddi

rect

inve

stm

ents

;al

low

ance

ofpr

ivat

eho

ldin

gof

fore

ign

exch

ange

and

othe

rfin

anci

alas

sets

,an

dpr

ovis

ion

ofm

arke

tac

cess

tofo

reig

nfin

anci

alin

stitu

tions

in19

93.

Kor

eaO

ct92

Eq.

Flo

ws-

BP

3C

lose

tody

-BP

(Feb

91).

Offi

cial

isJa

n92

.F

orei

gns

can

own

upto

10%

ofst

ocks

inse

lect

edin

dust

ries

(Jan

92).

Mor

ein

dust

ries

incl

uded

inM

ay92

.

Mal

aysi

aF

eb91

dy-B

PE

xact

lyco

inci

des

with

the

Priv

atiz

atio

nM

aste

rP

lan

(Feb

91).

Pre

cede

sth

eO

utlin

eP

ersp

ectiv

eP

lan

inJu

n91

whi

chen

cour

aged

fore

ign

inve

stm

ent

and

priv

atis

atio

n.O

ffici

alda

teis

earli

erD

ec88

.(c

onti

nued

onne

xtpa

ge)


313G. Bekaert et al. / Journal of International Money and Finance 21 (2002) 295–350T

able

2(co

ntin

ued)

Cou

ntry

Dat

eS

erie

sR

easo

n

Mex

ico

Mar

91dy

-BP

3Le

ssth

ana

year

from

Eq.

Flo

ws-

U(J

ul90

).C

lose

lyfo

llow

sT

elm

expr

ivat

izat

ion

inD

ec90

and

the

initi

atio

nof

NA

FT

Ata

lks

inF

eb90

and

bond

flow

sin

Mar

90.

Offi

cial

isea

rly,

May

89.

Nig

eria

Jun

95dy

-UO

ffici

alis

Aug

95bu

tea

rlier

in19

95th

eB

udge

tca

lled

for

the

open

ing

ofm

arke

tsto

fore

ign

port

folio

inve

stor

s.H

ence

,th

em

arke

tm

ight

have

antic

ipat

edth

eof

ficia

lin

trod

uctio

nof

the

Nig

eria

nIn

vest

men

tD

ecre

ein

Aug

95.

Pak

ista

nD

ec90

dy-B

P1

Ver

ycl

ose

toO

ffici

alin

Feb

91.

Inde

ed,

the

anno

unce

men

tof

the

liber

aliz

atio

nsth

atw

ere

impl

emen

ted

inF

eb91

was

mad

ein

Nov

90.

Phi

lippi

nes

Jul

90E

q.F

low

s-U

Clo

seto

dy-U

(Oct

91).

Offi

cial

isJu

n91

.F

ollo

ws

maj

orba

nkre

stru

ctur

ing

agre

emen

tin

Feb

90.

Firs

tA

DR

isM

ar91

.

Por

tuga

lF

eb89

Qua

dC

lose

tody

-U.

Offi

cial

isJu

ly86

.P

ortu

gal

Fun

dla

unch

edin

Aug

87.

Pre

cede

spr

ivat

izat

ion

law

inM

ar90

.

Tai

wan

Apr

88Q

uad

Als

ody

-Uda

te.

Exc

hang

eco

ntro

lslif

ted

inJu

l87

.O

ffici

alis

late

r,in

Jan

91.

Tha

iland

Aug

88E

q.F

low

s-B

PO

ffici

alS

ep87

whe

nA

lien

Boa

rdin

trod

uced

.dy

-Uis

Jan

90w

hich

seem

sla

te

Tur

key

Oct

89dy

-BP

Jul

89co

mm

uniq

uest

atin

gth

atm

arke

tsw

illbe

open

tofo

reig

ners

.O

ffici

alin

Aug

89.

Tur

kish

Inve

stm

ent

Fun

din

Dec

89.

Eq.

Flo

w-B

Pis

late

r,in

Dec

91an

dB

ond

Flo

w-B

Pis

Jun

92.

Ven

ezue

laM

ar92

dy-U

Offi

cial

Jun

90bu

tfir

stA

DR

isla

ter,

inA

ug91

.M

ajor

priv

atiz

atio

nsin

Sep

91an

dN

ov91

.A

llot

her

estim

atio

nsst

rong

lyin

fluen

ced

by19

97-9

8.

Zim

babw

eJu

n93

dy-U

Offi

cial

issa

me

date

whe

nne

win

vest

men

tgu

idel

ines

took

effe

ct.

How

ever

,br

eak

isno

tsi

gnifi

cant

.

The

datin

gin

volv

esa

num

ber

ofdi

ffere

ntes

timat

ion

stra

tegi

es:

(i)a

join

tes

timat

ion

ofqu

adra

varia

te(Q

uad)

syst

emw

ithw

orld

inte

rest

rate

s,ne

teq

uity

flow

s(E

q.F

low

s),

divi

dend

yiel

ds(d

y),

and

retu

rns,

(ii)

univ

aria

te(U

)es

timat

ion

allo

win

gfo

ra

sing

lebr

eak,

and

(iii)

Bai

-Per

ron

(BP

)te

sts

whi

ch

allo

wfo

rup

toth

ree

brea

ks(B

P1,

BP

2,B

P3)

and

are

also

univ

aria

te.

The

offic

ial

liber

aliz

atio

nda

tes

(Offi

cial

)ar

efr

omB

ekae

rtan

dH

arve

y(2

000a

).W

eal

soco

nsul

ted

with

the

chro

nolo

gyof

impo

rtan

tfin

anci

al,

econ

omic

and

polit

ical

even

tsde

velo

ped

inB

ekae

rtan

dH

arve

y(2

000b

)w

hich

isav

aila

ble

at

http

://w

ww

.duk

e.ed

u/~

char

vey.



able

3P

re-p

ost

anal

ysis

ofth

eva

riabl

esus

edin

vect

orau

tore

gres

sion

s

Cou

ntry

Div

iden

dyi

eld

Net

bond

flow

s/G

DP

Net

equi

tyflo

ws/

mca

pLo

gre

turn

s

Mea

nS

td.

dev.

Mea

nS

td.

dev.

Mea

nS

td.

dev.

Mea

nS

td.

dev.

Arg

entin

aF

ull

1.55

91.

372

0.00

018

0.00

088

0.00

075

0.00

547

0.01

160.

2138

Pre

1.12

51.

318

�0.

0000

20.

0000

8�

0.00

023

0.00

452

0.01

590.

2458

Pos

t2.

639

0.78

30.

0006

60.

0015

40.

0032

00.

0067

40.

0011

0.09

63

Bra

zil

Ful

l5.

494

4.40

30.

0000

70.

0006

40.

0008

50.

0020

50.

0083

0.16

02P

re7.

029

4.07

80.

0000

00.

0000

40.

0002

40.

0011

90.

0056

0.15

66P

ost

3.06

93.

775

0.00

019

0.00

103

0.00

180

0.00

268

0.01

260.

1664

Chi

leF

ull

4.94

62.

022

�0.

0002

70.

0019

10.

0003

40.

0013

90.

0177

0.09

66P

re5.

779

2.02

2�

0.00

026

0.00

088

0.00

021

0.00

138

0.01

880.

1067

Pos

t3.

441

0.76

0�

0.00

030

0.00

299

0.00

057

0.00

139

0.01

580.

0756

Col

ombi

aF

ull

4.65

53.

119

0.00

011

0.00

087

0.00

057

0.00

206

0.02

060.

0797

Pre

7.37

72.

503

�0.

0000

30.

0001

1�

0.00

005

0.00

080

0.02

240.

0580

Pos

t2.

269

0.73

70.

0004

40.

0015

00.

0012

00.

0026

80.

0187

0.09

71

Gre

ece

Ful

l5.

593

3.29

90.

0000

20.

0002

70.

0002

00.

0018

50.

0054

0.09

48P

re6.

173

3.91

5�

0.00

001

0.00

005

�0.

0001

00.

0020

0�

0.00

380.

0865

Pos

t4.

926

2.24

20.

0000

50.

0003

90.

0005

50.

0016

20.

0161

0.10

29

Indi

aF

ull

2.61

31.

350

�0.

0006

50.

0024

00.

0001

40.

0004

10.

0088

0.07

93P

re3.

172

1.18

6�

0.00

102

0.00

268

0.00

001

0.00

008

0.01

670.

0725

Pos

t1.

247

0.45

60.

0002

50.

0010

30.

0004

50.

0006

6�

0.01

030.

0916

Indo

nesi

aF

ull

1.36

80.

695

0.00

007

0.00

033

0.00

147

0.00

259

�0.

0191

0.13

00P

re0.

276

0.15

90.

0000

00.

0000

40.

0010

10.

0014

2�

0.01

500.

1045

Pos

t1.

633

0.48

10.

0002

30.

0005

90.

0016

40.

0029

2�

0.02

070.

1391

Kor

eaF

ull

3.81

92.

846

0.00

011

0.00

068

0.00

069

0.00

188

0.00

350.

0998

Pre

4.63

21.

593

0.00

000

0.00

027

0.00

024

0.00

090

0.01

020.

0885

Pos

t1.

593

0.54

90.

0004

00.

0012

00.

0019

30.

0030

0�

0.01

480.

1247

(con

tinu

edon

next

page

)



Tab

le3

(cont

inue

d)

Cou

ntry

Div

iden

dyi

eld

Net

bond

flow

s/G

DP

Net

equi

tyflo

ws/

mca

pLo

gre

turn

s

Mea

nS

td.

dev.

Mea

nS

td.

dev.

Mea

nS

td.

dev.

Mea

nS

td.

dev.

Mal

aysi

aF

ull

1.99

00.

624

0.00

015

0.00

195

0.00

020

0.00

065

�0.

0002

0.09

60P

re2.

160

0.39

8�

0.00

007

0.00

118

0.00

022

0.00

044

0.00

610.

0860

Pos

t1.

870

0.72

20.

0005

60.

0028

60.

0001

80.

0007

8�

0.00

550.

1037

Mex

ico

Ful

l4.

298

4.17

20.

0003

20.

0017

80.

0011

10.

0040

40.

0132

0.13

37P

re5.

743

4.49

60.

0001

50.

0019

50.

0007

70.

0025

40.

0155

0.14

57P

ost

1.50

60.

422

0.00

064

0.00

132

0.00

175

0.00

592

0.00

890.

1076

Pak

ista

nF

ull

4.49

62.

619

0.00

002

0.00

024

0.00

039

0.00

229

0.00

420.

0828

Pre

7.70

90.

789

0.00

000

0.00

000

0.00

000

0.00

020

0.01

010.

0299

Pos

t2.

765

1.88

20.

0000

60.

0004

00.

0006

90.

0030

3�

0.00

050.

1073

Phi

lippi

nes

Ful

l1.

540

1.47

60.

0001

20.

0010

50.

0005

40.

0023

90.

0175

0.10

39P

re2.

263

1.90

6�

0.00

008

0.00

028

0.00

005

0.00

313

0.04

430.

1042

Pos

t0.

913

0.53

40.

0004

70.

0016

20.

0008

70.

0016

2�

0.00

090.

1001

Por

tuga

lF

ull

2.45

61.

128

�0.

0000

20.

0001

60.

0010

10.

0039

70.

0206

0.10

45P

re1.

006

0.54

40.

0000

00.

0000

20.

0001

50.

0006

90.

0515

0.17

96P

ost

2.80

20.

939

�0.

0000

30.

0002

30.

0012

90.

0045

20.

0108

0.06

28

Tai

wan

Ful

l0.

994

0.54

1�

0.00

017

0.00

043

0.00

003

0.00

028

0.01

310.

1331

Pre

1.67

70.

701

�0.

0000

30.

0000

70.

0000

20.

0002

60.

0400

0.14

69P

ost

0.83

20.

334

�0.

0003

30.

0005

80.

0000

40.

0002

90.

0046

0.12

79

Tha

iland

Ful

l5.

523

3.31

10.

0000

40.

0004

30.

0003

00.

0009

80.

0056

0.09

62P

re7.

680

2.86

0�

0.00

001

0.00

013

0.00

035

0.00

100

0.01

650.

0736

Pos

t3.

005

1.55

70.

0001

10.

0006

10.

0002

40.

0009

6�

0.00

700.

1162

(con

tinu

edon

next

page

)



Tab

le3

(cont

inue

d)

Cou

ntry

Div

iden

dyi

eld

Net

bond

flow

s/G

DP

Net

equi

tyflo

ws/

mca

pLo

gre

turn

s

Mea

nS

td.

dev.

Mea

nS

td.

dev.

Mea

nS

td.

dev.

Mea

nS

td.

dev.

Tur

key

Ful

l4.

349

2.20

50.

0000

10.

0001

30.

0004

90.

0024

00.

0177

0.18

03P

re7.

275

2.91

70.

0000

00.

0000

2�

0.00

006

0.00

052

0.05

350.

2191

Pos

t3.

708

1.35

20.

0000

20.

0002

00.

0006

60.

0027

20.

0064

0.16

59

Ven

ezue

laF

ull

1.64

80.

914

0.00

015

0.00

287

0.00

001

0.00

396

0.00

830.

1409

Pre

1.39

40.

914

0.00

021

0.00

329

�0.

0001

00.

0015

80.

0266

0.13

83P

ost

1.89

00.

847

0.00

001

0.00

146

0.00

015

0.00

557

�0.

0123

0.14

18

All

coun

trie

sF

ull

3.37

32.

123

0.00

002

0.00

100

0.00

053

0.00

227

0.00

920.

1192

Pre

4.26

31.

900

�0.

0000

70.

0006

50.

0001

60.

0013

30.

0197

0.12

01P

ost

2.35

91.

081

0.00

020

0.00

115

0.00

101

0.00

277

0.00

140.

1134

Latin

Am

eric

aF

ull

3.76

72.

667

0.00

009

0.00

149

0.00

061

0.00

316

0.01

330.

137

Pre

4.74

12.

555

0.00

001

0.00

106

0.00

014

0.00

200

0.01

750.

142

Pos

t2.

469

1.22

10.

0002

70.

0016

40.

0014

50.

0041

60.

0074

0.11

4

Asi

aF

ull

2.53

91.

582

0.00

005

0.00

081

0.00

054

0.00

146

0.00

340.

110

Pre

3.11

51.

270

�0.

0000

30.

0003

30.

0003

20.

0011

90.

0170

0.10

1P

ost

1.64

10.

696

0.00

024

0.00

124

0.00

082

0.00

159

�0.

0074

0.11

9

Rel

yon

stoc

ksF

ull

3.37

91.

901

0.00

002

0.00

103

0.00

051

0.00

188

0.01

080.

0995

Pre

3.92

01.

554

�0.

0000

70.

0004

60.

0002

00.

0012

60.

0246

0.10

64P

ost

2.27

10.

844

0.00

020

0.00

159

0.00

085

0.00

205

�0.

0003

0.09

72

Rel

yon

bond

sF

ull

3.14

42.

363

0.00

013

0.00

112

0.00

076

0.00

340

0.00

440.

1436

Pre

3.87

91.

959

0.00

006

0.00

090

0.00

028

0.00

191

0.00

980.

1368

Pos

t2.

250

1.36

50.

0003

00.

0010

50.

0015

40.

0044

8�

0.00

180.

1264

(con

tinu

edon

next

page

)



Tab

le3

(cont

inue

d)

Cou

ntry

Div

iden

dyi

eld

Net

bond

flow

s/G

DP

Net

equi

tyflo

ws/

mca

pLo

gre

turn

s

Mea

nS

td.

dev.

Mea

nS

td.

dev.

Mea

nS

td.

dev.

Mea

nS

td.

dev.

Top

exte

rnal

Ful

l3.

135

2.38

90.

0001

60.

0014

70.

0006

00.

0030

10.

0075

0.14

07P

re3.

681

2.13

30.

0000

50.

0011

30.

0001

90.

0018

60.

0133

0.14

35P

ost

2.09

51.

183

0.00

041

0.00

157

0.00

150

0.00

411

�0.

0017

0.12

34

Bot

tom

exte

rnal

Ful

l3.

364

1.88

5�

0.00

016

0.00

105

0.00

027

0.00

144

0.01

110.

0984

Pre

4.46

21.

753

�0.

0002

40.

0006

60.

0000

30.

0011

70.

0210

0.09

11P

ost

2.35

41.

035

0.00

004

0.00

117

0.00

053

0.00

143

0.00

410.

1009

All

coun

trie

sre

pres

ents

aneq

ually

-wei

ghte

dav

erag

eac

ross

all

coun

trie

s.A

sia

iseq

ually

-wei

ghte

dac

ross

six

Asi

anco

untr

ies

(Ind

ones

ia,

Kor

ea,

Mal

aysi

a,th

eP

hilip

pine

s,T

aiw

anan

dT

haila

nd),

and

Latin

iseq

ually

-wei

ghte

dac

ross

six

Latin

Am

eric

anco

untr

ies

(Arg

entin

a,B

razi

l,C

hile

,C

olom

bia,

Mex

ico

and

Ven

ezue

la).

We

also

repo

rteq

ually

-wei

ghte

dac

ross

the

six

coun

trie

sth

atre

lym

ost

oneq

uity

finan

cing

(Chi

le,

the

Phi

lippi

nes,

Mal

aysi

a,P

ortu

gal

,T

haila

ndan

dK

orea

),eq

ually

-wei

ghte

dac

ross

the

six

coun

trie

sth

atre

lym

ost

onbo

ndfin

anci

ng(V

enez

uela

,M

exic

o,A

rgen

tina,

Pak

ista

n,In

done

sia

and

Bra

zil),

equa

lly-w

eigh

ted

acro

ssth

esi

xco

untr

ies

that

have

the

high

est

prop

ortio

nof

exte

rnal

finan

cing

,th

atis

,th

ebo

ndpl

useq

uity

capi

tal

flow

sto

GD

P(M

exi

co,

Mal

aysi

a,V

enez

uela

,A

rgen

tina,

Bra

zil

and

Kor

ea)

and

the

six

coun

trie

sw

ithth

elo

wes

tpr

opor

tion

ofex

tern

alfin

anci

ngto

GD

P(C

hile

,T

aiw

an,

Pak

ista

n,th

eP

hilip

pine

s,G

reec

ean

dIn

dia)

.T

hem

eans

and

stan

dard

devi

atio

nsar

ees

timat

edov

erth

efu

llsa

mpl

e,th

epr

e-br

eak

sam

ple

and

the

post

-bre

aksa

mpl

e.T

hebr

eak

date

sfo

rea

chco

untr

yar

epr

esen

ted

inT

able

2.



decreases after the breaks. Note that return volatility actually declines from 41.6%to 39.3% on an annualized basis.

The capital flows also exhibit differences. The equity capital flows to capitalizationratio increases five-fold after the break. Increases occur in 15 of 17 countries. Simi-larly, the bond flows to GDP ratio increases from a negative number in the pre-break period to a positive number in the post-break period. Increases occur in 13 of17 countries. It is also interesting to note that while the volatility of equity returnsand dividend yields decreases after our breaks, the volatility of both the equity andbond capital flows increases. Indeed, the volatility of the bond and equity flow ratiosdoubles from pre to post-break.

4.3. Mean dynamics of capital flows

We estimated the regression specified in eq. (6) for each individual country andpooled across all countries for the various country groups introduced in Section 3.The pooled regression allows for fixed effects, and corrects for temporal heterosked-asticity. The results are in Table 4. We report the regressions both for equity flowsand bond flows. The equity results are consistent with the liberalization inducing alarge portfolio rebalancing that induces large capital flows just after the break, butless after the transition period (which we fixed at three years). On an annualizedbasis, equity flows increase by 1.4% of local market capitalization, but then drop by0.55% after three years. Bond flows continue to increase throughout.

One might expect this phenomenon to be artificially driven by the Asian crisis.During late 1997 and 1998, large amounts of foreign capital left Asia and this wasnot driven by a liberalization-induced portfolio rebalancing. However, the overallphenomenon we find does not occur for Asia, where equity flows on average continueto increase three years after the liberalization. The positive/negative pattern howeveris very strong in Latin-America. Not surprisingly the result also strongly appears forthe 6 countries relying most on foreign capital, whereas it does not show for theleast foreign portfolio capital reliant countries. The fact that the result shows up forthe countries that rely more heavily on bonds and not for those that rely most heavilyon stocks is probably due to the geographical composition of these groups, withLatin-American countries dominating the former, but not the latter.

4.4. Transition dynamics

The results of our THL statistic using the quadravariate VAR are summarized inFig. 3 for capital flows and dividend yields. We can interpret these results as indica-tive of how flows and dividends react when the markets become integrated and howthey might react when the integration process is reversed. The pre-mean/new dynam-ics plot illustrates that for almost half the countries, capital comes in very fast, sothat adjustment happens very quickly, usually within one month. This may provideindirect evidence of portfolio rebalancing. But there are also some notable excep-tions. In Chile, Colombia, Greece, India, Korea, the Philippines, Portugal, Thailand,and Venezuela the transition lasts at least five months. More striking is that in 15



able

4C

hang

esin

capi

tal

flow

saf

ter

brea

ks

Cou

ntry

Net

equi

tyflo

ws

Net

bond

flow

s

Inte

rcep

tB

reak

3-yr

spo

stIn

terc

ept

Bre

ak3-

yrs

post

Arg

entin

a�

0.00

023

0.00

613

�0.

0052

6�

0.00

002

0.00

053

0.00

029

�0.

694.

04�

3.13

�2.

342.

920.

87

Bra

zil

0.00

030

0.00

135

0.00

019

�0.

0000

10.

0001

20.

0001

42.

132.

060.

27�

0.98

1.54

0.89

Chi

le0.

0002

10.

0001

50.

0003

5�

0.00

026

�0.

0004

20.

0006

31.

710.

520.

86�

3.09

�1.

161.

09

Col

ombi

a�

0.00

005

0.00

124

0.00

007

�0.

0000

40.

0002

00.

0005

2�

0.53

2.75

0.11

�3.

661.

591.

92

Gre

ece

�0.

0001

00.

0002

40.

0005

9�

0.00

001

�0.

0000

10.

0001

1�

0.57

1.30

1.92

�2.

52�

1.04

2.23

Indi

a0.

0000

10.

0002

70.

0003

2�

0.00

102

0.00

102

0.00

047

1.49

3.24

2.02

�3.

153.

162.

19

Indo

nesi

a0.

0008

00.

0015

9�

0.00

132

0.00

000

0.00

017

0.00

011

3.03

2.65

�1.

970.

662.

040.

75

Kor

ea0.

0002

40.

0010

20.

0014

10.

0002

40.

0010

20.

0014

12.

244.

191.

702.

244.

191.

70

Mal

aysi

a0.

0002

20.

0002

5�

0.00

050

�0.

0000

7�

0.00

004

0.00

112

3.19

1.19

�2.

27�

0.63

�0.

152.

35

Mex

ico

0.00

077

0.00

350

�0.

0042

70.

0001

50.

0006

3�

0.00

022

3.25

2.70

�3.

171.

002.

00�

0.65

Pak

ista

n0.

0000

00.

0002

60.

0007

10.

0000

00.

0000

10.

0000

8�

0.04

2.10

1.20

�0.

201.

111.

14(c

onti

nued

onne

xtpa

ge)



Tab

le4

(cont

inue

d)

Cou

ntry

Net

equi

tyflo

ws

Net

bond

flow

s

Inte

rcep

tB

reak

3-yr

spo

stIn

terc

ept

Bre

ak3-

yrs

post

Phi

lippi

nes

0.00

005

0.00

089

�0.

0001

0�

0.00

008

�0.

0001

00.

0010

50.

101.

40�

0.24

�2.

68�

1.09

3.42

Por

tuga

l0.

0001

50.

0004

20.

0010

60.

0000

0�

0.00

007

0.00

006

1.47

2.23

1.72

�3.

00�

3.40

1.99

Tai

wan

0.00

002

�0.

0000

20.

0000

6�

0.00

003

�0.

0001

1�

0.00

027

0.39

�0.

451.

15�

4.53

�1.

69�

1.82

Tha

iland

0.00

035

�0.

0000

3�

0.00

012

�0.

0000

1�

0.00

005

0.00

025

2.20

�0.

15�

0.63

�1.

19�

0.76

2.54

Tur

key

�0.

0000

60.

0008

9�

0.00

026

0.00

000

0.00

005

�0.

0000

5�

0.77

3.41

�0.

58�

1.17

1.29

�1.

14

Ven

ezue

la�

0.00

010

0.00

066

�0.

0007

90.

0002

1�

0.00

022

0.00

004

�0.

601.

13�

0.61

0.89

�0.

820.

13

All

coun

trie

s0.

0011

6�

0.00

046

0.00

012

0.00

022

7.56

�2.

752.

823.

51

Asi

a0.

0004

60.

0000

30.

0001

80.

0004

04.

720.

232.

553.

51



Tab

le4

(cont

inue

d)

Cou

ntry

Net

equi

tyflo

ws

Net

bond

flow

s

Inte

rcep

tB

reak

3-yr

spo

stIn

terc

ept

Bre

ak3-

yrs

post

Latin

0.00

219

�0.

0015

40.

0001

40.

0002

25.

97�

3.72

1.43

1.50

Rel

yon

stoc

ks0.

0004

30.

0002

4�

0.00

007

0.00

054

3.82

1.39

�0.

823.

86

Rel

yon

bond

s0.

0023

6�

0.00

168

0.00

021

0.00

006

6.06

�3.

972.

750.

68

Top

exte

rnal

0.00

224

�0.

0015

40.

0002

10.

0002

96.

18�

3.69

2.33

2.08

Bot

tom

exte

rnal

0.00

032

0.00

031

0.00

008

0.00

026

3.31

2.55

1.05

2.27

Bas

edon

are

gres

sion

ofne

teq

uity

flow

san

dne

tbo

ndflo

ws

ontw

oin

dica

tor

varia

bles

.T

hene

teq

uity

(bon

d)flo

ws

repr

esen

tth

ene

teq

uity

(bon

d)flo

ws

betw

een

the

Uni

ted

Sta

tes

and

each

emer

ging

mar

ket.

The

flow

data

are

mon

thly

from

the

Dep

artm

ent

ofth

eT

reas

ury.

The

first

indi

cato

rta

kes

the

valu

eof

one

afte

ra

brea

k.T

hese

cond

indi

cato

rta

kes

onth

eva

lue

ofon

eth

ree

year

saf

ter

the

brea

k.T

here

are

seve

npo

oled

estim

atio

ns:

all

coun

trie

s,si

xA

sian

coun

trie

s(I

ndon

esia

,K

orea

,M

alay

sia,

the

Phi

lippi

nes,

Tai

wan

and

Tha

iland

),si

xLa

tinA

mer

ican

coun

trie

s(A

rgen

tina,

Bra

zil,

Chi

le,

Col

ombi

a,M

exic

oan

dV

enez

uela

),th

esi

xco

untr

ies

that

rely

mos

ton

equi

tyfin

anci

ng(C

hile

,th

eP

hilip

pine

s,M

alay

sia,

Por

tuga

l,T

haila

ndan

dK

orea

),th

esi

xco

untr

ies

that

rely

mos

ton

bond

finan

cing

(Ven

ezue

la,M

exic

o,A

rgen

tina,

Pak

ista

n,In

done

sia

and

Bra

zil),

the

six

coun

trie

sth

atha

veth

ehi

ghes

tpro

port

ion

ofex

tern

alfi

nanc

ing,

that

is,

the

bond

plus

equi

tyca

pita

lflo

ws

toG

DP

(Mex

ico,

Mal

aysi

a,V

enez

uela

,A

rgen

tina,

Bra

zil,

and

Kor

ea)

and

the

six

coun

trie

sw

ithth

elo

wes

tpr

opor

tion

ofex

tern

alfin

anci

ngto

GD

P(C

hile

,T

aiw

an,P

akis

tan,

the

Phi

lippi

nes,

Gre

ece

and

Indi

a).W

edo

not

repo

rtth

ein

terc

epts

inth

egr

oup

esti

mat

ions

.T

hebr

eak

date

sfo

rea

chco

untr

yar

epr

esen

ted

inT

able

2.A

llre

gres

sion

sha

vehe

tero

sked

astic

ity-c

onsi

sten

tst

anda

rder

rors

.T

hepo

oled

regr

essi

ons

allo

wfo

rfix

edef

fect

s.T

heva

lues

belo

wea

chco

effic

ient

repr

esen

tth

et-

stat

istic

.



Fig. 3. Transition dynamics.

of the 17 countries, the transition implied by the post-mean/old dynamics is faster(with Brazil moving from 1 to 2 months), suggesting that when capital leaves, itleaves faster than it came! This is an interesting finding in light of the recent criseswhich resulted in capital flight from many emerging markets. Indeed, our empiricalfindings use data from long before these crises.

In 13 of the 17 countries, a dividend yield decrease followed the liberalization(structural change), indicating a decrease in the cost of capital. In contrast to theevidence on capital flows, this decrease in the cost of capital seems to take sometime (more than 1 month in all countries besides Korea, and a median of 4 months),but in the reverse experiment, this process would take even longer (median of 10months). One interpretation of this finding is that liberalizations have made dividendyields less persistent.

5. VAR dynamics

5.1. Dynamic regression coefficients

Table 5 reports the results of our analysis of dynamic regression coefficients. Wereport the coefficients at three horizons,k = 12 (one year),k = 36 (three years), andk = 60 (five years). We also report the cumulative responses over these horizons, andthe infinite cumulative response. To give an idea of the cross-sectional distribution ofthe coefficients across countries, we order the coefficients from low to high andreport the coefficients for the third and 15th country (there are 17 countries). Wealso report the equally-weighted average over the six Latin American, over the sixAsian and over all countries. The table only reports the post-break analysis. Asreported in a previous version of the paper [Bekaert et al., 1999], many of the resultsare not robust across time periods demonstrating once again that capital liberaliza-tions have caused breaks in the dynamic relations between our variables that cannotbe ignored.

The first three panels investigate slope coefficients with the world interest rate asthe regressor. These reflect the long-run correlations between capital flows, dividendyields or returns at timet+k and the world interest rate at timet. The relation betweenfuture capital flows and current world interest rates varies significantly across coun-tries. The equally-weighted response is positive, but the relation is negative for both



Tab

le5

Dyn

amic

anal

ysis

Res

pons

eS

ampl

ek=12

k=36

k=60

cum

k=12

cum

k=36

cum

k=60

Infin

ity

Fut

ure

capi

tal

flow

sP

ost

3rd

coun

try

−0.0

194

−0.0

155

−0.0

047

−0.4

688

−0.7

078

−1.1

395

−1.5

049

tow

orld

inte

rest

15th

coun

try

0.03

270.

0057

0.00

190.

8146

1.05

141.

0640

1.06

47ra

tes

Equ

alw

eigh

t0.

0061

0.03

350.

0974

0.04

350.

4887

1.97

611.

7526

Latin

Am

eric

a−0

.009

9−0

.006

0−0

.001

9−0

.017

6−0

.235

6−0

.319

7−0

.363

5A

sia

−0.0

051

−0.0

007

−0.0

003

−0.1

356

−0.1

784

−0.1

883

−0.1

542

Fut

ure

divi

dend

Pos

t3r

dco

untr

y−1

7.37

−8.6

5−7

.04

−254

.19

−639

.35

−878

.84

−192

1.20

yiel

dsto

wor

ld15

thco

untr

y12

.90

4.94

3.12

143.

1133

7.66

387.

3843

8.63

inte

rest

rate

sE

qual

wei

ght

9.75

29.0

880

.52

93.4

053

2.26

1774

.50

−432

.55

Latin

Am

eric

a−5

.49

−5.1

5−3

.77

−60.

82−1

91.6

3−2

98.7

0−4

84.8

2A

sia

−2.0

6−1

.46

−1.2

6−2

0.08

−58.

44−9

1.34

−205

.83

Fut

ure

retu

rns

toP

ost

3rd

coun

try

−1.6

042

−0.5

033

−0.3

056

−2.4

314

−1.2

862

−0.8

134

−70.

7920

wor

ldin

tere

stra

tes

15th

coun

try

1.29

790.

5596

0.46

891.

4222

1.27

630.

8150

96.3

130

Equ

alw

eigh

t−0

.272

5−1

.122

1−3

.203

1−0

.096

9−0

.480

3−1

.104

1−1

2.21

10La

tinA

mer

ica

0.04

330.

1353

0.09

93−0

.036

40.

0625

0.08

529.

4885

Asi

a0.

9303

0.39

310.

2868

1.35

510.

8231

0.62

5911

5.82

00

Fut

ure

divi

dend

Pos

t3r

dco

untr

y−1

6.95

−1.7

5−0

.58

−445

.25

−563

.99

−567

.52

−565

.10

yiel

dsto

capi

tal

15th

coun

try

15.5

215

.45

5.90

176.

8354

6.87

919.

7319

37.0

0flo

ws

Equ

alw

eigh

t0.

88−6

.57

−28.

4816

.76

−34.

97−4

24.6

2−7

7.44

Latin

Am

eric

a−4

.46

−0.2

80.

30−1

16.6

5−1

53.6

5−1

50.9

9−1

23.1

3A

sia

8.20

12.9

513

.52

102.

0437

0.81

692.

8843

58.0

0(c

onti

nued

onne

xtpa

ge)



Tab

le5

(cont

inue

d)

Res

pons

eS

ampl

ek=12

k=36

k=60

cum

k=12

cum

k=36

cum

k=60

Infin

ity

Fut

ure

retu

rns

toP

ost

3rd

coun

try

−0.1

451

−0.0

212

−0.0

041

−0.1

238

−0.0

674

−0.0

446

−12.

6570

capi

tal

flow

s15

thco

untr

y3.

3869

1.79

171.

6850

5.05

914.

2570

2.64

4714

1.28

00E

qual

wei

ght

1.42

841.

4957

2.29

632.

4706

1.75

631.

7887

−133

2.80

00La

tinA

mer

ica

−0.0

225

0.02

650.

0080

0.10

330.

0559

0.03

972.

4257

Asi

a0.

8740

0.46

170.

3981

3.31

171.

4720

1.05

4218

5.17

00

Fut

ure

capi

tal

flow

sP

ost

3rd

coun

try

−0.0

0015

−0.0

0001

0.00

000

−0.0

0588

−0.0

0462

−0.0

0543

−0.0

2422

todi

vide

ndyi

elds

15th

coun

try

0.00

044

0.00

015

0.00

009

0.00

801

0.01

374

0.01

609

0.01

565

Equ

alw

eigh

t0.

0002

30.

0004

60.

0011

20.

0023

00.

0102

50.

0282

6−0

.015

13La

tinA

mer

ica

0.00

006

0.00

006

0.00

003

0.00

019

0.00

183

0.00

285

0.00

370

Asi

a−0

.000

010.

0000

20.

0000

2−0

.000

69−0

.000

320.

0001

00.

0033

4

Fut

ure

capi

tal

flow

sP

ost

3rd

coun

try

−0.0

0030

−0.0

0011

−0.0

0006

−0.0

0125

−0.0

0437

−0.0

0408

−0.0

0627

tore

turn

s15

thco

untr

y0.

0004

30.

0000

70.

0000

70.

0096

70.

0137

30.

0171

30.

0141

8E

qual

wei

ght

0.00

009

0.00

012

0.00

016

0.00

484

0.00

727

0.01

055

−0.2

2599

Latin

Am

eric

a−0

.000

07−0

.000

04−0

.000

020.

0065

00.

0050

70.

0043

40.

0035

6A

sia

0.00

004

−0.0

0001

0.00

000

0.00

280

0.00

276

0.00

264

0.00

665

For

our

anal

ysis

ofdy

nam

icre

gres

sion

coef

ficie

nts,

we

repo

rtth

eco

effic

ient

sat

thre

eho

rizon

s,k

=12

(1ye

ar),

k=36

(thr

eeye

ars)

,an

dk=6

0(fi

veye

ars)

.W

eal

sore

port

the

cum

ulat

ive

resp

onse

sov

erth

ese

horiz

ons,

and

the

infin

itecu

mul

ativ

ere

spon

se.

To

inve

stig

ate

the

cros

s-se

ctio

nal

dist

ribut

ion

ofth

eco

effic

ient

sac

ross

coun

trie

s,w

eor

der

the

coef

ficie

nts

from

low

tohi

ghan

dre

port

the

coef

ficie

nts

for

the

third

and

15th

coun

try

(the

rear

e17

coun

tri

es).

We

also

repo

rtth

eeq

ually

-wei

ghte

dav

erag

eov

erth

esi

xLa

tinA

mer

ican

,ov

erth

esi

xA

sian

and

over

allc

ount

ries.

The

brea

kda

tes

are

prov

ided

inT

abl

e2.



the Latin American and Asian countries. For example, a 1% decrease in the worldinterest rate is associated with a cumulative 36-month increase of US holdings ofthe local equity market equal to 0.24% in Latin America and 0.18% in South-EastAsia in the post-break period.

The relation between the world interest rate and future dividend yields equally-weighted across the countries is at first positive but the cumulative infinite responseis negative. The responses in Latin America and Asia are always negative, indicatingthat lower interest rates are associated with future increases in dividend yields. Thisis surprising under the “push” hypothesis where lower interest rates drive developedmarket capital into emerging markets and drive up prices there, hence lowering divi-dend yields. Of course, we report long-run effects, and the “push” effect may bevery short-lived. The effects we document die out very slowly, which is largely dueto the large persistence of the dividend yield in most countries. Note that the coef-ficients seem large primarily because of the log-transformation of dividend yields.To get an approximate regression coefficient for the level of the dividend yield, oneshould multiply the reported coefficients by the average dividend yield level (0.035).

For the implicit regression of future returns on the world interest rate, we dividethe cumulative effects byk, to obtain a per period return effect (except, of course,for k = �). Not surprisingly, given the noisiness in returns, the coefficients showalso a wide range across countries. Nevertheless, they are consistently positive forboth Latin America and Asia. Hence, decreases in world interest rates are associatedwith long-run decreases in returns, not increases as we might expect. The samecaveats we mentioned above apply. In addition, we should add that a true test ofthe “push” hypothesis should focus on the effects of an unexpected shock to worldinterest rates, especially since world interest rates are quite predictable. Such ananalysis follows later when we consider impulse responses.

The following panels consider the long-run relation between capital flows andfuture dividend yields and returns. The relation between capital flows and futuredividend yields also shows a lot of cross-sectional dispersion, with positive coef-ficients for Asia and negative ones for Latin America. Hence, the slope coefficientsdo not clearly reveal a permanent cost of capital effect from increased capital flows.The correlations between capital flows and returns are largely positive; higher flowsare correlated with higher future returns but the coefficients are quite small.

The next panel considers an implicit regression of future capital flows on currentlog-dividend yields. The cumulative responses can now be interpreted as the changein total holdings over this period. The coefficients are generally small for the samereason previous coefficients involving dividend yields were large — the log-trans-formation. For an average dividend yield level of 0.035, one should multiply thecoefficients by 28.6. If higher dividend yields proxy for higher expected returns, andthe return chasing hypothesis is true, one would expect a positive contemporaneousrelation, which may persist because of positive autocorrelation in dividend yieldsand flows. This is indeed the case for both Latin-America and Asia at the five-yearhorizon, but the effects are very small. It remains true when the country-by-countryresults are averaged at the five-year horizon but not when the infinite cumulativeresponse is computed.



The last panel investigates the correlation between current returns and future capi-tal flows. At the horizons that we consider, we would not expect to see much of aneffect, even if foreigners are momentum traders. Indeed, we find very small effects,that are positive when cumulated for Asia and Latin-America, but negative whenaveraged over all countries.

In general, the slope coefficients do not lead to strong conclusions about univariatecorrelation patterns between the various variables. This motivates investigating morecomplex patterns, either partial regression coefficients as in the Granger-causalityanalysis that follows, or impulse responses which control for the predictability ofcausal factors.

5.2. Granger-causality tests

Granger-causality tests are reported in Table 6. The first three columns investigatethe predictive power of a much discussed external factor (a “push” factor), the worldinterest rate, on equity flows, dividend yields and returns. The statistical results areweak. This is not surprising before the break if markets were truly segmented pre-venting free capital flows. The only significant result is that in Korea we can rejectthat the world interest rate fails to Granger-cause returns. But even in the POSTperiod, significant results are rare. Except for single cases of near or below 5%rejections of the no Granger-causality hypotheses in Korea, the Philippines, and Tur-key, the only country where the world interest rate played a significant role predictingcapital flows, dividend yields and returns simultaneously is Brazil.

These results may indicate that the world interest rate is not an important predictorof capital flows and returns and that the previous literature (e.g. Froot et al., 2001)justifiably ignored it, but it may also reflect the short sample periods we have avail-able. When looking at impulse responses in the next sub-section, our various countrygroupings may yet reveal the interest rate to be an important external determinantof capital flows.

The price pressure hypothesis would suggest that increased capital flows tempor-arily induce high returns which are reversed afterwards in which case we wouldexpect capital flows to Granger-cause returns. In the portfolio rebalancing story, onthe other hand, even before the full liberalization is implemented, anticipation shouldhave already led to permanent price increases, making it less obvious that we willsee significant results in the post period. As Fig. 1 indicates, returns show the mostinteresting dynamics during the transition period. If there is an effect on the dividendyield, the effect may represent a more permanent change in the cost of capital. Thestatistical evidence on Granger-causality is not overwhelming. We reject the hypoth-esis of no predictability of returns by capital flows at the 5% level for only sixcountries: Brazil (pre-break), Greece (pre-break), Korea (post-break), Malaysia (fullperiod), Pakistan (post-break), and Thailand (pre-break).

In three of the six countries where capital flows predict returns, they also predictdividend yields significantly. In three other countries (Indonesia, the Philippines andVenezuela), we also record a significant relation between capital flows and divi-dend yields.



able

6G

rang

erca

usal

ityte

sts

Pro

babi

lity

valu

esfr

omth

enu

llhy

poth

esis

that

:

Cou

ntry

Wor

ldin

tere

stW

orld

inte

rest

Wor

ldin

tere

stN

eteq

uity

Net

equi

tyR

etur

nsdo

not

Div

iden

dyi

elds

rate

sdo

not

rate

sdo

not

rate

sdo

not

flow

sdo

not

flow

sdo

not

Gra

nger

caus

edo

not

Gra

nger

Gra

nger

caus

eG

rang

erca

use

Gra

nger

caus

eG

rang

erca

use

Gra

nger

caus

ene

teq

uity

caus

ene

tne

teq

uity

divi

dend

yiel

dsre

turn

sdi

vide

ndyi

elds

retu

rns

flow

seq

uity

flow

sflo

ws

Arg

entin

aF

ull

0.95

10.

652

0.28

30.

927

0.43

80.

819

0.95

5P

re0.

981

0.94

00.

183

0.97

80.

395

0.68

70.

903

Pos

t0.

316

0.62

40.

172

0.80

40.

064

0.83

20.

557

Bra

zil

Ful

l0.

936

0.91

10.

921

0.00

20.

575

0.62

10.

093

Pre

0.94

00.

962

0.88

50.

000

0.00

80.

024

0.15

7P

ost

0.09

00.

028

0.07

10.

029

0.45

40.

305

0.73

3

Chi

leF

ull

0.69

10.

162

0.34

80.

595

0.95

60.

526

0.92

4P

re0.

806

0.26

80.

121

0.20

80.

991

0.58

60.

607

Pos

t0.

406

0.98

60.

340

0.08

20.

938

0.69

70.

234

Col

ombi

aF

ull

0.14

60.

298

0.05

10.

581

0.31

00.

913

0.00

0P

re0.

299

0.79

70.

757

0.34

90.

299

0.98

20.

127

Pos

t0.

191

0.66

40.

134

0.80

20.

415

0.95

90.

015

Gre

ece

Ful

l0.

535

0.09

30.

765

0.49

60.

129

0.48

10.

947

Pre

0.90

90.

327

0.73

40.

720

0.02

40.

188

0.79

2P

ost

0.63

50.

826

0.45

30.

898

0.57

50.

994

0.99

4

Indi

aF

ull

0.90

10.

588

0.97

80.

879

0.05

10.

507

0.77

4P

re0.

312

0.66

50.

999

0.99

60.

900

0.73

00.

121

Pos

t0.

570

0.24

00.

141

0.92

70.

090

0.81

20.

842

Indo

nesi

aF

ull

0.99

50.

308

0.29

00.

187

0.51

10.

307

0.66

9P

re0.

099

0.19

50.

372

0.04

00.

201

0.49

50.

833

Pos

t0.

999

0.92

00.

513

0.16

30.

754

0.38

40.

566

(con

tinu

edon

next

page

)



able

6(co

ntin

ued)

Pro

babi

lity

valu

esfr

omth

enu

llhy

poth

esis

that

:

Cou

ntry

Wor

ldin

tere

stW

orld

inte

rest

Wor

ldin

tere

stN

eteq

uity

Net

equi

tyR

etur

nsdo

not

Div

iden

dyi

elds

rate

sdo

not

rate

sdo

not

rate

sdo

not

flow

sdo

not

flow

sdo

not

Gra

nger

caus

edo

not

Gra

nger

Gra

nger

caus

eG

rang

erca

use

Gra

nger

caus

eG

rang

erca

use

Gra

nger

caus

ene

teq

uity

caus

ene

tne

teq

uity

divi

dend

yiel

dsre

turn

sdi

vide

ndyi

elds

retu

rns

flow

seq

uity

flow

sflo

ws

Kor

eaF

ull

0.74

10.

953

0.04

90.

934

0.01

00.

000

0.98

1P

re0.

144

0.99

80.

030

1.00

00.

614

0.51

80.

982

Pos

t0.

999

0.07

00.

558

0.28

70.

010

0.00

00.

890

Mal

aysi

aF

ull

0.94

80.

876

0.34

00.

413

0.02

30.

092

0.12

5P

re0.

759

0.44

20.

462

0.28

00.

194

0.11

40.

423

Pos

t0.

901

0.78

80.

592

0.46

80.

095

0.21

30.

033

Mex

ico

Ful

l0.

998

0.05

10.

617

0.68

20.

884

0.96

00.

734

Pre

0.87

10.

111

0.18

90.

326

0.30

20.

996

0.45

5P

ost

0.57

50.

214

0.72

60.

623

0.64

70.

994

0.51

0

Pak

ista

nF

ull

0.34

60.

426

0.14

20.

000

0.00

00.

466

0.04

8P

re0.

545

0.09

00.

458

0.94

30.

702

0.49

50.

862

Pos

t0.

532

0.27

70.

205

0.00

00.

015

0.84

40.

179

Phi

lippi

nes

Ful

l0.

015

0.44

40.

597

0.41

50.

831

0.56

30.

499

Pre

0.38

30.

249

0.68

50.

008

0.32

00.

353

0.00

1P

ost

0.05

20.

240

0.45

60.

833

0.97

60.

423

0.18

0

Por

tuga

lF

ull

0.69

20.

438

0.57

00.

958

1.00

00.

988

0.96

4P

re0.

527

0.88

90.

903

0.96

30.

795

0.73

80.

977

Pos

t0.

750

0.13

30.

281

0.82

00.

983

0.98

50.

907

Tai

wan

Ful

l0.

598

0.69

70.

819

0.95

60.

651

0.75

90.

408

Pre

0.65

00.

303

0.42

10.

491

0.84

70.

703

0.59

9P

ost

0.69

20.

740

0.80

30.

723

0.30

10.

782

0.47

2



Tab

le6

(cont

inue

d)

Pro

babi

lity

valu

esfr

omth

enu

llhy

poth

esis

that

:

Cou

ntry

Wor

ldin

tere

stW

orld

inte

rest

Wor

ldin

tere

stN

eteq

uity

Net

equi

tyR

etur

nsdo

not

Div

iden

dyi

elds

rate

sdo

not

rate

sdo

not

rate

sdo

not

flow

sdo

not

flow

sdo

not

Gra

nger

caus

edo

not

Gra

nger

Gra

nger

caus

eG

rang

erca

use

Gra

nger

caus

eG

rang

erca

use

Gra

nger

caus

ene

teq

uity

caus

ene

tne

teq

uity

divi

dend

yiel

dsre

turn

sdi

vide

ndyi

elds

retu

rns

flow

seq

uity

flow

sflo

ws

Tha

iland

Ful

l0.

992

0.26

50.

585

0.00

50.

251

0.00

60.

383

Pre

0.99

60.

159

0.34

90.

326

0.00

10.

069

0.13

7P

ost

0.56

20.

490

0.58

90.

013

0.20

20.

187

0.53

8

Tur

key

Ful

l0.

951

0.92

80.

040

0.99

60.

331

0.30

30.

490

Pre

0.42

70.

991

0.03

70.

054

0.14

80.

705

0.52

4P

ost

0.95

30.

651

0.00

90.

984

0.23

00.

427

0.09

0

Ven

ezue

laF

ull

0.99

00.

167

0.29

60.

288

0.52

60.

938

0.54

6P

re0.

823

0.58

00.

323

0.01

90.

054

0.89

30.

759

Pos

t0.

977

0.96

70.

830

0.30

30.

386

0.86

90.

398

Gra

nger

caus

ality

anal

ysis

isba

sed

ona

quad

rava

riate

syst

emof

wor

ldin

tere

stra

tes,

net

equi

tyca

pita

lflow

s,di

vide

ndyi

elds

and

equi

tyre

turn

s.T

heV

AR

sar

ees

timat

edov

erth

efu

llsa

mpl

e,th

epr

e-br

eak

sam

ple

and

the

post

-bre

aksa

mpl

e.T

hebr

eak

date

sfo

rea

chco

untr

yar

epr

esen

ted

inT

able

2.



Do endogenous factors play a large role in determining capital flows? If that isthe case, returns and/or dividend yields ought to have predictive power for capitalflows. Interestingly, there is little predictive power of dividend yields (only Colom-bia, Malaysia, Pakistan, and the Philippines), and returns predict capital flows onlyin Brazil (pre-break), Korea and Thailand.

The lack of significant predictive relations between our variables, revealed bycountry-specific Granger-causality tests, stresses again the need to focus on cross-sectionally averaged results. To analyze some of the predictive relations further, wereport three partial regression coefficients in Table 7. The first is the coefficient inthe flow equation on past flows. Froot et al. (2001) note that the predictive powerof flows for future returns may in fact be due to both a strong contemporaneousrelation between flows and returns (see next section) and the persistence in flows.Our estimates suggest a monthly persistence coefficient of around 0.135, which ismuch higher than the monthly persistence implied by Froot et al.’s daily model.

The second coefficient is the coefficient on returns in the flows equation. Positivecoefficients suggest feedback trading or may be the result of returns anticipatingpositive news about future market reforms that will bring in foreign capital. Whereasthe country by country results showed little significance (and are not reported), thecoefficients are predominantly positive as they are in Froot et al. (2001). One countryfor which a gradual liberalization story would have much appeal, Korea, records anegative coefficient.

Finally, we report the flow coefficient in the return equation. As in Froot et al.(2001), we find predominantly positive and large coefficients. A 1% increase inforeign holdings leads on average to a 4.74% increase in returns next month (seeTable 7). Froot et al. argue that this is not inconsistent with the price pressure hypoth-esis (which would require a positive contemporaneous effect and negative long-runeffect) given that flows are persistent. The response could also reflect a return tointegration, as we discussed before, although one would expect much of the increasein foreign holdings to be anticipated and, hence, not induce large ex post pricechanges. In the impulse response analysis in the next section, we look at the dynamiceffects of unexpected shocks, including the contemporaneous relations between vari-ables.

6. Tests of the main hypotheses

The impulse response analysis of the quadravariate VAR is presented in Figs. 4–6. For each country, we estimate the VAR on three samples: full sample, pre-breakand post-break. We then aggregate the impulse responses across various countrygroups using equal weights. To help interpret the evidence, in light of the hypothesesformulated in Section 2, we provide two additional tables. Table 8 presents an analy-sis of impulse responses on changes in equity holdings and long-run returns, whereasTable 9 reports some important contemporaneous betas. When rescaled, these betasconstitute the impulse responses at time 0 and they also appear in Figs. 4–6. Bothtables focus exclusively on the post-break period.



Tab

le7

Ana

lysi

sof

VA

Rco

effic

ient

s(p

ost-

brea

kes

timat

ion)

VA

Req

uatio

nE

qual

Val

ueA

sian

Latin

Rel

yon

Rel

yon

Top

Bot

tom

Res

idua

l3r

d15

thw

eigh

ted

wei

ghte

deq

uity

bond

sex

tern

alex

tern

alw

eigh

tsco

effic

ient

coef

ficie

ntfin

anci

ngfin

anci

ng

Equ

ityflo

ws

on0.

1320

0.13

600.

1133

0.12

490.

0781

0.15

710.

1280

0.20

840.

1351

�0.

0245

0.29

02la

gged

flow

s

Equ

ityflo

ws

on0.

0019

0.00

230.

0009

0.00

350.

0012

0.00

360.

0034

0.00

040.

0016

�0.

0005

0.00

49la

gged

retu

rns

Ret

urns

on4.

7378

6.67

469.

1445

1.92

158.

2919

1.28

165.

1814

4.51

666.

5033

�0.

9226

10.6

367

lagg

edeq

uity

flow

s

We

repo

rtth

eV

AR

coef

ficie

ntde

note

din

the

first

colu

mn

from

coun

try

spec

ific

estim

atio

nsba

sed

inth

epo

st-b

reak

sam

ple.

Equ

alw

eigh

ted

isan

equa

lw

eigh

ting

ofal

l117

coun

trie

s.V

alue

wei

ghte

dre

pres

ents

17co

untr

ies

wei

ghte

dby

thei

rm

arke

tca

pita

lizat

ion

inD

ecem

ber

1991

.A

sia

repr

esen

tssi

xA

sian

coun

trie

s.La

tinre

pres

ents

six

Latin

Am

eric

anco

untr

ies.

Rel

yon

stoc

ksar

eth

esi

xco

untr

ies

that

tend

tore

lyon

stoc

ksm

ore

than

bond

sfo

rex

tern

alfin

anci

ngsi

nce

1989

.R

ely

onbo

nds

are

the

six

coun

trie

sw

hich

tend

tore

lym

ore

onbo

nds

for

exte

rnal

finan

cing

.T

opex

tern

alar

eth

esi

xco

untr

ies

whi

chha

veth

ela

rges

tpr

opor

tion

ofcu

mul

ativ

ene

tbo

ndpl

useq

uity

flow

sto

GD

P.

Bot

tom

exte

rnal

are

the

six

coun

trie

sw

ithth

esm

alle

stpr

opor

tion

ofcu

mul

ativ

ene

tbo

ndpl

useq

uity

flow

sto

GD

P.

For

the

3rd

and

15th

coef

ficie

nt,

we

rank

the

coef

ficie

nts

acro

ssco

untr

ies,

elim

inat

e“o

utlie

rs”,

that

isth

esm

alle

stan

dhi

ghes

ttw

oan

dre

port

the

rang

efr

omth

ere

mai

nder

(min

and

max

),(t

he3r

dan

d15

thco

effic

ient

give

nth

ere

are

17co

untr

ies)

.



Table 8Cumulative impulse response analysis on post break sample

Country group k=12 k=36 k=60

A. Interest rate shock on cumulative flowsEqually weighted 0.00020 0.00069 0.00158Value weighted 0.00011 0.00010 0.00026Asia 0.00031 0.00040 0.00041Latin 0.00003 0.00033 0.00044Rely on stocks 0.00033 0.00048 0.00054Rely on bonds 0.00006 0.00108 0.00340Top external 0.00008 0.00024 0.00028Bottom external 0.00036 0.00143 0.003883rd coefficient �0.00022 �0.00045 �0.0006315th coefficient 0.00101 0.00145 0.00172

B. Dividend yield shock on cumulative flowsEqually weighted 0.00004 0.00077 0.00253Value weighted 0.00042 0.00074 0.00112Asia �0.00012 �0.00003 0.00003Latin �0.00037 �0.00032 �0.00025Rely on stocks �0.00014 �0.00001 0.00008Rely on bonds 0.00014 0.00189 0.00667Top external �0.00019 �0.00021 �0.00017Bottom external 0.00013 0.00197 0.006803rd coefficient �0.00077 �0.00133 �0.0015515th coefficient 0.00100 0.00184 0.00216

C. Return shock on cumulative flowsEqually weighted 0.00012 0.00002 �0.00010Value weighted 0.00012 0.00002 �0.00001Asia 0.00011 0.00011 0.00011Latin 0.00022 0.00004 0.00000Rely on stocks 0.00011 0.00011 0.00010Rely on bonds 0.00020 �0.00004 �0.00031Top external 0.00014 �0.00002 �0.00004Bottom external 0.00009 �0.00001 �0.000293rd coefficient �0.00015 �0.00021 �0.0002815th coefficient 0.00044 0.00045 0.00046

D. Interest rate shock on average long-run returnsEqually weighted 0.00054 �0.00018 �0.00061Value weighted 0.00071 0.00007 �0.00008Asia 0.00079 0.00016 0.00002Latin 0.00125 0.00035 0.00015Rely on stocks �0.00017 �0.00046 �0.00042Rely on bonds 0.00137 �0.00031 �0.00147Top external 0.00156 0.00049 0.00027Bottom external 0.00009 �0.00084 �0.001833rd coefficient �0.00220 �0.00130 �0.0010215th coefficient 0.00284 0.00130 0.00084

(continued on next page)



Table 8 (continued)

Country group k=12 k=36 k=60

E. Net equity flows shock on average long-run returnsEqually weighted 0.00097 0.00040 0.00030Value weighted 0.00127 0.00049 0.00033Asia 0.00080 0.00038 0.00028Latin 0.00122 0.00041 0.00025Rely on stocks 0.00144 0.00061 0.00043Rely on bonds 0.00086 0.00036 0.00034Top external 0.00182 0.00072 0.00047Bottom external 0.00053 0.00027 0.000303rd coefficient �0.00079 �0.00032 �0.0001915th coefficient 0.00189 0.00091 0.00071

Notes to the table: We analyze cumulative impulse response functions for two variables: average longrun returns and change in equity holdings. The change in equity holdings is simply the cumulative sumof net equity capital flows. For the average long-run returns we examine shocks in capital flows andworld interest rates. For the change in equity holdings we measure the effect of a shock in world interestrates, returns and dividend yields. The effects are calculated by summing the impulse response until k(in months) not including the contemporaneous effect. Equal weighting is an equal weighting of all 17countries. Value weighting represents 17 countries weighted by their market capitalization in December1991. Asia represents six Asian countries. Latin represents six Latin American countries. Rely on stocksare the six countries that tend to rely on stocks more than bonds for external financing since 1989. Relyon bonds are the six countries which tend to rely more on bonds for external financing. Top external arethe six countries which have the largest proportion of cumulative net bond plus equity flows to GDP.Bottom external are the six countries with the smallest proportion of cumulative net bond plus equityflows to GDP. For the 3rd and 15th coefficient, we rank the coefficients across countries, eliminate“outliers”, that is the smallest two and highest two and report the range from the remainder (min andmax), (the 3rd and 15th coefficient given there are 17 countries). The VARs are estimated on the post-break sample. The break dates are detailed in Table 2.

6.1. The effects of world interest rate shocks

The effects of a negative world interest rate shock differ greatly between the pre-break and the post-break periods. Given that we should only expect any effect post-liberalization, we only consider the post-break period. The contemporaneous effectsof a shock in world interest rates on capital flows are mixed with positive covariancesdominating (see Table 9) but after one period a negative shock generally leads tosmall increases in net equity flows (Fig. 4 panel (a)). This is definitely the case forAsia but less so for Latin America. The countries that benefit the most are thosewith the highest degree of external financing. In all cases, the effects die out quickly.From Table 8, we see that a negative interest rate shock is associated with increasedholdings at five year horizons for all groupings. The impact is greatest for countriesthat rely on bonds rather than stocks. While there is a large positive increase inholdings for Asian countries up to one year out compared to Latin American coun-tries, by five years, the impact is very similar across the regions. A 0.3% decreasein world interest rates leads to an 0.04% increase in the proportion of the equitymarket held by US investors.



Tab

le9

Impo

rtan

tco

ntem

pora

neou

sef

fect

s

Coe

ffici

ent

Equ

alV

alue

Asi

anLa

tinR

ely

onR

ely

onT

opex

tern

alB

otto

mR

esid

ual

wei

ghte

dw

eigh

ted

equi

tybo

nds

finan

cing

exte

rnal

wei

ghts

finan

cing

nfon

i0.

0623

�0.

0571

0.05

950.

0322

0.04

530.

0793

�0.

0216

0.13

940.

0761

dyon

i12

.182

814

.451

218

.558

013

.368

214

.180

214

.605

418

.708

84.

2519

11.1

440

ron

i�

8.66

53�

12.9

565

�10

.712

8�

11.9

839

�8.

2843

�14

.377

3�

15.5

770

�3.

4362

�7.

6166

dyon

nf�

4.53

55�

5.59

96�

10.0

817

�1.

6151

�11

.646

8�

0.95

37�

7.21

08�

3.56

36�

5.88

19r

onnf

3.71

905.

3209

6.54

922.

3367

8.47

07�

0.36

187.

7257

3.88

895.

6144

ron

dy�

0.29

45�

0.27

24�

0.29

88�

0.29

85�

0.30

70�

0.29

28�

0.30

91�

0.27

53�

0.32

25

The

coef

ficie

nts

repo

rted

are

base

don

the

cont

empo

rane

ous

beta

sim

plie

dby

the

caus

alor

derin

gof

our

VA

Rin

the

post

-bre

akes

timat

ion.

Tha

tis

,th

eya

reta

ken

from

the

mat

rixB

,w

here�

=BH

B�,

whe

reB

islo

wer

tria

ngle

and

Hha

sth

eva

rianc

esof

the

stru

ctur

alsh

ocks

alon

gits

diag

onal

.E

qual

wei

ghte

dis

aneq

ual

wei

ghtin

gof

all1

17co

untr

ies.

Val

uew

eigh

ted

repr

esen

ts17

coun

trie

sw

eigh

ted

byth

eir

mar

ket

capi

taliz

atio

nin

Dec

embe

r19

91.

Asi

are

pre

sent

ssi

xA

sian

coun

trie

s.La

tinre

pres

ents

six

Latin

Am

eric

anco

untr

ies.

Rel

yon

stoc

ksar

eth

esi

xco

untr

ies

that

tend

tore

lyon

stoc

ksm

ore

than

bond

sfo

rex

tern

alfin

anci

ngsi

nce

1989

.R

ely

onbo

nds

are

the

six

coun

trie

sw

hich

tend

tore

lym

ore

onbo

nds

for

exte

rnal

finan

cing

.T

opex

tern

alar

eth

esi

xco

untr

ies

whi

chha

veth

ela

rges

tpr

opor

tion

ofcu

mul

ativ

ene

tbo

ndpl

useq

uity

flow

sto

GD

P.

Bot

tom

exte

rnal

are

the

six

coun

trie

sw

ithth

esm

alle

stpr

opor

tion

ofcu

mul

ativ

ene

tbo

ndpl

useq

uity

flow

sto

GD

P.

The

varia

bles

are:

the

wor

ldin

tere

stra

te(i)

,ne

teq

uity

flow

s(n

f),

divi

dend

yiel

ds(d

y)an

dre

turn

s(r

).



Fig. 4. The impact of world interest rate shocks on equity flows, dividend yields and returns.

The analysis of dividend yields in Fig. 4 panel (b) reveals that a negative shock inthe interest rate is associated with lower dividend yields in Asian and Latin Americancountries. The only incidence of higher dividend yields is for countries that use littleexternal financing. In terms of magnitude, Table 9 is informative. The contempor-aneous beta is over 10.0, meaning that a 1% decrease in the world interest rate leadsto a drop in the level of the dividend yield of about 10 times the average dividendyield, that is 35 bp. The effect persists for quite a long time, especially in Asia.

The analysis of returns in Fig. 4 panel (c) suggests a positive effect only contem-poraneously in all countries except those that do not rely on external financing. Thisis consistent with a portfolio effect (higher capital flows to emerging markets) beinginduced by a low world interest rate. It may reflect a pure short-term price pressureeffect or the return to integration (see above) if market liberalizations happen tocoincide with periods of lower world interest rates.

Our analysis of long-run returns in Table 8 suggests that negative interest rateshocks increase long-run returns over a one-year horizon but the effect is often elim-inated by three years. Generally, both the dividend yield effects and the expectedreturn effects do not show clear enough results to distinguish between long-termbeneficial effects (lower cost of capital) or short-term price pressure effects thatmight reverse themselves.



6.2. The effects of equity flow shocks

We examine the impact of positive equity flows shocks on dividend yields andreturns in Fig. 5. We use this figure to illustrate the dramatic differences betweenthe pre- and post-break analysis. Perhaps the most powerful graph that we presentis the impulse response of a positive shock in equity flows on dividend yields. Inthe pre-break period, it is hard to see any consistent effect. However, in the postbreak period, there is a sharply lower dividend yield. Contemporaneously (see Table9), the effect is on the order of about 20 basis points per 1% increase in foreignholdings (recall that because of the log-transformation, the coefficients must be multi-plied by 0.035, the mean dividend yield, to obtain the effect on the level). In addition,the effect is very persistent. Dividend yields drop the most for Asian countries butalso drop for Latin American countries. The drop in dividend yields is very strongfor those countries that rely on external financing. Interestingly, the countries thatactively use bond financing have a larger drop in dividend yields than those countriesthat rely more on equity financing. Following Bekaert and Harvey’s (2000a) argu-ment that changes in dividend yields closely follow changes in the cost of equitycapital, this analysis suggests that increased capital flows decrease the cost of equitycapital. It is important to realize that the shock to capital flows is net of the resultof world interest rate changes.

The impulse response analysis of capital flows on returns in Fig. 5 is consistentwith the portfolio rebalancing hypothesis. In the post-break period, returns increaseonly in the very short term. These results are consistent with the dividend yieldresults. A shock in equity capital flows increases the price level which leads to higherreturns in the short-term and permanently lower dividend yields. The contempor-aneous beta is around 6. This is of the same order of magnitude as the estimate forMexico in Clark and Berko (1997), but much smaller than the estimate in the dailyflow/return model in Froot et al. (2001). None of these studies separate the interestrate effect from the capital flows effect. The cumulative effect (see Table 8) remainspositive, but becomes significantly smaller at the 60-month horizon suggesting some,but incomplete, reversal of prices. The largest impact is on countries that have arelatively large reliance on external financing.

6.3. The effects of return and dividend yield shocks

We now turn to the return chasing hypothesis. Bohn and Tesar (1996) distinguishtwo hypotheses. The “return updating” hypothesis links expected returns to capitalflows. The “momentum” hypothesis predicts positive capital flows after positivereturns. We revisit this last hypothesis by looking at realized returns and measuringthe impact of a positive shock in returns on the capital flows (see Fig. 6 panelA). In our framework, the VAR ordering implies that the return shock omits thecontemporaneous correlation with both capital flows and dividend yields and hencereflects an unusual unexpected high return, not contemporaneously correlated withforeign capital shocks or permanent cost of capital changes. We find a positiveresponse of flows to returns in all regions but it is most dramatic for Latin American



Fig. 5. The impact of equity flow shocks on dividend yields and returns in three samples.



Fig. 6. The impact of returns and dividend yield shocks on equity flows.

countries. We find little or no impact on capital flows for those countries that havesmall external financing. The impulse response analysis supports the hypothesis thatcapital flows are, in part, momentum driven. That this is an entirely short-run effectis confirmed by the cumulative responses reported in Table 8, Panel C. The positivereturn shocks lead to higher holdings in the short-term but not in the long term.

Fig. 6 Panel (b) reports the effects of dividend yields on capital flows, where thedividend yield shock is net of a contemporaneous capital flow effect. A positiveshock in the dividend yield is associated with higher flows after a few periods for thevolatility-weighted, equally-weighted and the value-weighted set of countries after aninitial negative response. For the Latin American and the Asian set, the impulseresponses basically become zero. The former results may be consistent with theshort-run momentum results and a longer term “expected return”-chasing effect. Inthe very short term, a higher dividend yield may simply reflect a negative unexpectedreturn, which leads to reduced short-term capital flows. However, higher dividendyields may be associated with higher expected returns in the future. Positive effectson capital flows are only visible after two periods and then only for a subset of ourgroupings. The countries not relying on external finance display a consistently posi-tive effect. Table 8, Panel B cumulates these effects to measure the total change inholdings. The cumulative effects, for example, show that the positive long-run effectson capital flows in South-East Asia suffice to overturn the initial negative effect. ForLatin American and stock reliant countries, the cumulative effect is slightly negative.



7. Conclusions

The goal of this paper is to develop a better understanding of the relation betweencapital flows and asset prices. We apply the latest structural break econometrics toidentify liberalizations in 20 emerging markets and then examine a VAR in pre- andpost-break periods on four variables: the world interest rate, net equity capital flows,ex post returns and a proxy for expected returns, the dividend yield.

Our results can be grouped into two main categories. Our first set of findingsconcern the break dynamics. We find that after a liberalization equity capital flowsincrease by 1.4% of market capitalization on an annual basis. However, three yearspost-liberalization, the equity flows are reduced. This is consistent with a modelwhereby investors rebalance their portfolios towards emerging markets. Also, com-paring the pre with post-break dynamics, we document that in almost all the countrieswe examine, when capital leaves it leaves much faster than it came. These intriguingresults may shed light on the recent crises in Latin America and Asia and the roleof capital flight.

Our second set of results revisits a number of important hypotheses in a structuralVAR framework. The fundamental nonstationarity in the data — structural breaksinduced by liberalizations — ignored by previous research, is not without conse-quences. We illuminate significant differences between the pre-break analysis andthe post-break analysis.

Our analysis yields three main findings. First, the “push effect” from world interestrates to capital flows appears consistently when we cumulate impulse responseswhereas contemporaneously interest rates and capital flows show no consistent corre-lation pattern. A 0.3% decrease in interest rates eventually increases foreign holdingsby about 0.04% of market capitalization, a small effect. Interest rate decreases dogenerate strong but very short-lived increases in returns.

Second, unexpected shocks to equity flows have a strongly positive contempor-aneous effect on returns, in line with the findings of Clark and Berko (1997) andFroot et al. (2001). The effect immediately dies out but there is only incompletereversal suggesting some of the effect is permanent. This is consistent with ourfinding that positive shocks in net equity capital flows lead to lower dividendyields — our proxy for expected returns. Following Bekaert and Harvey’s (2000a)argument that dividend yield changes reveal information about the cost of equitycapital, the equity capital flow shocks lead to a lower cost of capital in many coun-tries. We find that this relation is dramatically strengthened if we estimate our VARson the post-break sample. Although part of the initial effect may be due to “pricepressure”, our results suggest part of the response is near permanent and beneficial.

Third, we revisit the Bohn and Tesar (1996) argument that capital flows can bedriven by expected ‘return chasing’ and by momentum investing. We find evidencethat positive returns shocks are followed by increased short-term equity capital flows,indicating a momentum effect. Using the dividend yield as a proxy for long-runexpected returns, we find only weak evidence for the expected return chasing hypoth-esis.

There are, of course, many caveats to our analysis. Ideally, we need an economic



theory that captures the evolution from segmented to integrated financial markets.In the absence of such a theory, we rely on vector autoregressions to characterizethe behavior of important financial aggregates. In addition, the break methodologyimplicitly assumes that every break is permanent and hence theoretically ignores thatthe next break may be rationally anticipated by market participants. If this is thecase, a regime switching model, such as the one presented in Bekaert and Harvey(1995) may be a superior modeling approach. Although we have not studied thisissue in detail, we believe that structural break tests can probably be used to detectpersistent changes in regime and so are less incompatible with a regime-switchingmodel than theory may lead one to believe.

Acknowledgements

We appreciate the comments of Josh Coval, Luc Laeven, Michael Pagano, MarkSeasholes, Linda Tesar, Frank Warnock, Jeff Wurgler, as well as the participants atthe 1998 Conference on Globalization, Capital Markets Crises and Economic Reformat Duke University, the 2000 Western Finance Association meetings in Sun Valley,the University of Virginia and Erasmus University. A preliminary version of thispaper circulated under the title, “Structural breaks in emerging market capital flows”.We owe significant thanks to our RA, Chris Lundblad for his help. Lumsdaineacknowledges the support of the National Fellows Program at the Hoover Institution,Stanford University. Bekaert gratefully acknowledges financial support from an NSFgrant. The opinions expressed are the authors’ and do not represent those of DeutscheBank or its affiliates.

Technical appendix

BLS (Testing for a single break with multiple time-series that break at the sametime)

The techniques in Bai et al. (1998) (BLS) enable us to investigate structural breaksin the relationship between capital flows, returns, and dividends and to constructconfidence intervals around an estimated break date. For this part of the analysis,we assume that the data are generated by a stationary vector autoregression and thereis at most one structural break. One of the key results in BLS is that the precisionwith which a potential break date is estimated (as given by the width of the associatedconfidence interval) is a function not of the number of observations but of the numberof series in a multivariate framework that experience the same break date. In addition,they show that including series that have no break in the VAR, such as the worldinterest rate in our application, while reducing the power of the tests to detect acommon break, will not increase the width of the confidence intervals. In an earlierpaper (Bekaert et al., 2002, hereafter BHL), we provide empirical examples of howthese techniques can be used to draw inference.



App

endi

xT

able

1U

niva

riate

brea

kte

sts

Cou

ntry

Sta

tistic

5th

%ile

Med

ian

95th

%ile

#of

lags

Sta

tistic

5th

%ile

Med

ian

95th

%ile

#of

lags

Log

retu

rns

Log

divi

dend

yiel

dA

rgen

tina

8.00

Jun

84Ju

l89

Aug

941

8.84

Oct

90M

ay92

Dec

931

Bra

zil

8.17

Sep

82A

pr87

Nov

911

9.25

May

88N

ov90

May

933

Chi

le24

.75∗∗

∗M

ay81

Feb

83N

ov84

220

.48∗∗

∗Ja

n81

Sep

82M

ay84

3C

olom

bia

14.1

7∗∗A

pr90

Feb

92D

ec93

114

.98∗∗

Jun

92F

eb94

Oct

952

Gre

ece

8.38∗

Feb

81N

ov85

Aug

900

15.2

9∗∗D

ec83

Jun

86D

ec88

1In

dia

6.23

Oct

86A

pr92

Oct

970

12.6

7Ju

l89

Jan

92Ju

l94

3In

done

sia

21.7

5∗∗∗

Dec

96A

pr97

Aug

972

20.6

8∗∗∗

Apr

92M

ay92

Jun

922

Jord

an5.

94F

eb79

Apr

82Ju

n85

030

.18

∗∗∗

Nov

91A

pr92

Sep

921

Kor

ea10

.62∗∗

Sep

92N

ov94

Jan

970

92.4

9∗∗

∗D

ec90

Feb

91A

pr91

1M

alay

sia

12.5

3∗∗Ju

n95

Jul

96A

ug97

010

.32∗

May

90F

eb91

Nov

911

Mex

ico

22.6

9∗∗∗

Jan

86O

ct87

Jul

893

11.9

2∗∗

Oct

84M

ay86

Dec

871

Nig

eria

2.56

Jun

87A

ug97

35.

55F

eb94

Jun

95O

ct96

1P

akis

tan

13.0

1∗Ja

n92

Dec

93N

ov95

229

.77

∗∗∗

Jan

96M

ar96

May

962

Phi

lippi

nes

13.1

0∗∗N

ov86

Aug

87M

ay88

110

.49

Aug

90O

ct91

Dec

923

Por

tuga

l17

.19∗

Feb

87M

ay88

Aug

894

19.4

9∗∗∗

Jan

89A

pr89

Jul

892

Tai

wan

7.94

∗F

eb87

Feb

90F

eb93

018

.26

Jul

87A

pr88

Jan

894

Tha

iland

21.3

8O

ct93

Nov

94D

ec95

415

.33

∗∗M

ar89

Jan

90N

ov90

1T

urke

y2.

70A

ug90

Jan

970

20.2

3A

ug89

Oct

89D

ec89

1V

enez

uela

8.96

Apr

89F

eb92

Dec

941

12.0

4Ju

l90

Mar

92N

ov93

2Z

imba

bwe

7.17

May

91M

ar95

36.

59F

eb92

Jun

93O

ct94

1(c

onti

nued

onne

xtpa

ge)


342 G. Bekaert et al. / Journal of International Money and Finance 21 (2002) 295–350A

ppen

dix

Tab

le1

(cont

inue

d)

Cou

ntry

Sta

tistic

5th

%ile

Med

ian

95th

%ile

#of

lags

Sta

tistic

5th

%ile

Med

ian

95th

%ile

#of

lags

Net

equi

tyflo

ws

toeq

uity

capi

taliz

atio

nN

etbo

ndflo

ws

toG

DP

Arg

entin

a37

.50∗∗

∗Ja

n95

Jul

95Ja

n96

244

.00

∗∗∗

Jul

92A

pr93

Jan

940

Bra

zil

30.7

4∗∗∗

Dec

88Ja

n90

Feb

911

9.44

May

89N

ov92

May

960

Chi

le11

.31∗

Feb

85Ja

n88

Dec

901

2.08

May

840

Col

ombi

a24

.12∗∗

∗N

ov92

Jul

93M

ar94

055

.08∗∗

∗S

ep93

Feb

94Ju

l94

3G

reec

e18

.89∗∗

∗F

eb90

Mar

92A

pr94

115

.29∗∗

Apr

93N

ov94

Jun

960

Indi

a82

.20∗∗

∗F

eb93

May

93A

ug93

023

.48∗∗

∗A

ug94

May

95F

eb96

2In

done

sia

5.40

May

93D

ec95

029

.62

∗∗∗

Jul

94Ju

n95

May

962

Jord

anK

orea

34.8

9∗∗∗

Jun

95Ja

n96

Aug

962

24.4

2∗∗

∗M

ay89

Dec

90Ju

l92

0M

alay

sia

18.6

8∗∗F

eb93

Feb

94F

eb95

315

.03

∗∗Ju

l91

Jul

93Ju

l95

0M

exic

o7.

77N

ov87

Jul

90M

ar93

113

.59

∗∗M

ar87

Mar

90M

ar93

0N

iger

iaP

akis

tan

6.10

May

95D

ec96

1n.

a.P

hilip

pine

s26

.62∗∗

∗S

ep87

Jul

90M

ar93

054

.72

∗∗∗

Sep

95N

ov95

Jan

964

Por

tuga

l4.

67S

ep91

Jun

950

31.1

3∗∗

∗S

ep94

May

95Ja

n96

4T

aiw

an5.

31F

eb92

Jun

951

21.9

3∗∗

Jun

91N

ov91

Apr

924

Tha

iland

34.9

1∗∗∗

Jan

96Ju

l96

Jan

974

29.0

7∗∗

∗A

ug94

May

95F

eb96

0T

urke

y5.

05S

ep94

Nov

960

39.7

0∗∗

∗M

ay92

Jul

92S

ep92

4V

enez

uela

3.34

Dec

92A

ug96

01.

62D

ec90

0Z

imba

bwe

Uni

varia

tebr

eak

test

sba

sed

onth

em

etho

dsof

Bai

etal

.(1

998)

.T

est

isco

mpu

ted

from

anau

tore

gres

sion

,w

here

the

num

ber

ofla

gsis

chos

enby

the

BIC

,an

dis

the

max

imum

over

all

poss

ible

brea

kda

tes

k∗ �1�

k�T

�k∗

ofan

F-s

tatis

ticte

stin

gth

ehy

poth

esis

that

all

coef

ficie

nts

inth

eau

tore

gres

sion

brea

kat

date

k.k∗

isth

etr

imm

ing

valu

e,ta

ken

tobe

0.15T,

whe

reT

isth

esa

mpl

esi

ze.

Crit

ical

valu

esfo

rth

est

atis

tics

are

give

nin

BH

L(1

999)

.T

hees

timat

edbr

eak

date

isgi

ven

byth

eco

lum

nla

bele

d“M

edia

n”,

with

corr

espo

ndin

g90

%co

nfide

nce

inte

rval

give

nby

the

colu

mns

labe

led

“5th

%ile

”an

d“9

5th

%ile

”.B

lank

sin

thes

ela

tter

two

colu

mns

indi

cate

that

the

estim

ated

confi

denc

ein

terv

alex

ceed

edth

esa

mpl

esi

ze.

Sig

nific

ance

ofth

ebr

eak

test

sat

the

10,

5an

d1%

leve

lsar

ede

note

dby∗ ,

∗∗,

and

∗∗∗ ,

resp

ectiv

ely.


343G. Bekaert et al. / Journal of International Money and Finance 21 (2002) 295–350A

ppen

dix

Tab

le2

Mul

tivar

iate

brea

kte

sts

Triv

aria

tete

sts:

Qua

drav

aria

tete

sts:

Qui

ntra

varia

tete

sts:

Equ

ityflo

ws,

divi

dend

yiel

ds,

retu

rns

Wor

ldin

tere

stra

tes,

equi

tyflo

ws,

divi

dend

Wor

ldin

tere

stra

tes,

bond

flow

s,eq

uity

flow

s,yi

elds

,re

turn

sdi

v.yi

elds

,re

turn

s

Cou

ntry

Sta

tistic

5th

%ile

Med

ian

95th

#of

Sta

tistic

5th

%ile

Med

ian

95th

#of

Sta

tistic

5th

%ile

Med

ian

95th

#of

%ile

lags

%ile

lags

%ile

lags

Arg

entin

a38

.25∗∗

∗M

ar95

Apr

95M

ay95

140

.42∗∗

Aug

94S

ep94

Oct

941

126.

84∗∗M

ay94

Jun

94Ju

l94

3B

razi

l17

5.7∗∗

∗F

eb90

Mar

90A

pr90

319

8.68∗∗

∗F

eb90

Mar

90A

pr90

325

7.1∗∗

∗F

eb90

Mar

90A

pr90

3C

hile

84.3

5∗∗∗

Mar

85A

pr85

May

853

91.1

5∗∗∗

Feb

85M

ar85

Apr

853

149.

24∗∗∗

Dec

85Ja

n86

Feb

863

Col

ombi

a79

.43∗∗

∗A

pr94

May

94Ju

n94

290

.77∗∗

∗A

pr94

May

94Ju

n94

212

8.07∗∗

∗A

pr94

May

94Ju

n94

2G

reec

e50

.35∗∗

∗Ju

n88

Jul

88A

ug88

270

.8∗∗Ju

n88

Jul

88A

ug88

399

.86∗∗

Jun

88Ju

l88

Aug

883

Indi

a78

.6∗∗∗

Feb

93M

ar93

Apr

933

73.4

4∗∗M

ar92

Apr

92M

ay92

311

7.68∗∗

Mar

93A

pr93

May

932

Indo

nesi

a14

0.72∗∗

∗A

pr97

May

97Ju

n97

219

1.93∗∗

∗A

pr97

May

97Ju

n97

320

3.82∗∗

∗A

pr97

May

97Ju

n97

3Jo

rdan

Kor

ea96

.6∗∗∗

Dec

95Ja

n96

Feb

963

122.

89∗∗∗

Aug

95S

ep95

Oct

953

591.

9∗∗

∗M

ar96

Apr

96M

ay96

3M

alay

sia

70.2

5∗∗∗

May

96Ju

n96

Jul

963

81.9

7∗∗∗

Jun

96Ju

l96

Aug

963

136.

14∗∗∗

Jul

96A

ug96

Sep

963

Mex

ico

67.5

3∗∗∗

Aug

87S

ep87

Oct

873

106.

13∗∗∗

Aug

87S

ep87

Oct

873

140.

33∗∗∗

Sep

87O

ct87

Nov

873

Nig

eria

Pak

ista

n31

.52∗∗

Nov

96D

ec96

Jan

971

53.7

7∗∗

∗N

ov96

Dec

96Ja

n97

113

5.36∗∗

∗Ju

l97

Aug

97S

ep97

1P

hilip

pine

s52

.46

Jul

88A

ug88

Sep

883

99.4

∗∗∗

Feb

88M

ar88

Apr

883

163.

4∗∗∗

Nov

88D

ec88

Jan

893

Por

tuga

l89

.84∗∗

∗A

pr89

May

89Ju

n89

310

6.07∗∗

∗Ja

n89

Feb

89M

ar89

317

2.18∗∗

∗M

ay96

Jun

96Ju

l96

3T

aiw

an59

.13∗∗

Mar

88A

pr88

May

883

109.

69∗∗∗

Mar

88A

pr88

May

883

178.

77∗∗∗

Mar

88A

pr88

May

883

Tha

iland

117.

18∗∗∗

May

96Ju

n96

Jul

963

132.

3∗∗∗

May

96Ju

n96

Jul

963

180.

52∗∗∗

May

96Ju

n96

Jul

963

Tur

key

144.

53∗∗

∗D

ec96

Jan

97F

eb97

318

2.85∗∗

∗D

ec96

Jan

97F

eb97

322

9.4

∗∗∗

Dec

96Ja

n97

Feb

973

Ven

ezue

la12

8.14∗∗

∗Ju

l96

Aug

96S

ep96

315

2.34∗∗

∗Ju

l96

Aug

96S

ep96

318

3.38∗∗

∗A

ug96

Sep

96O

ct96

3Z

imba

bwe

Mul

tivar

iate

brea

kte

sts

base

don

the

met

hods

ofB

aiet

al.

(199

8),

test

ing

the

null

hypo

thes

isof

nost

ruct

ural

chan

geag

ains

tth

eal

tern

ativ

eof

asi

ngl

ebr

eak.

The

test

isco

mpu

ted

from

aV

AR

,w

here

the

num

ber

ofla

gsis

chos

enby

the

BIC

,an

dis

anal

ogou

sto

the

univ

aria

tete

sts

inT

able

2.In

the

quad

rava

riate

and

quin

trav

aria

tete

sts,

only

the

retu

rns,

divi

dend

yiel

ds,

and

flow

sar

eal

low

edto

brea

k,th

atis

,th

ete

stdo

esno

tle

tth

eva

riabl

esin

the

wor

ldin

ter

est

rate

equa

tion

brea

k.S

eeal

sono

tes

toT

able

2.



App

endi

xT

able

3T

ests

that

allo

wfo

rm

ultip

lebr

eaks

Log

retu

rns

Log

divi

dend

yiel

dN

eteq

uity

flow

sto

equi

tyN

etbo

ndflo

ws

toG

DP

capi

taliz

atio

n

Cou

ntry

5th

Med

ian

95th

Sig

nif

5th

Med

ian

95th

Sig

nif

5th

Med

ian

95th

Sig

nif

5th

Med

ian

95th

Sig

nif

%ile

%ile

%ile

%ile

%ile

%ile

%ile

%ile

Arg

entin

a1

NB

not

Mar

92M

ay92

Nov

931.

0%M

ar92

Dec

92F

eb93

2.5%

Aug

80M

ay81

Sep

8110

.0%

Arg

entin

a2

Oct

94A

ug95

Dec

952.

5%F

eb93

Mar

93A

pr93

10.0

%B

razi

l1

Feb

84M

ay87

Jun

9010

.0%

Jan

91no

tN

Bno

tN

Bno

tB

razi

l2

Feb

95no

tN

Bno

tN

Bno

tC

hile

1O

ct81

Mar

83Ja

n85

1.0%

Nov

90no

tN

Bno

tN

Bno

tC

hile

2N

ov94

not

NB

not

NB

not

Col

umbi

a1

Feb

90F

eb92

Jun

932.

5%Ja

n91

Oct

91D

ec91

5.0%

NB

not

NB

not

Col

ombi

a2

Oct

93F

eb94

Jun

952.

5%N

Bno

tN

Bno

tG

reec

eN

Bno

tD

ec86

Jul

88S

ep90

1.0%

Feb

90M

ar92

May

9210

.0%

NB

not

Indi

aN

Bno

tN

Bno

tA

pr93

May

93Ju

n93

1.0%

NB

not

Indo

nesi

aN

Bno

tF

eb92

Apr

92Ju

n92

1.0%

NB

not

Mar

94F

eb96

Oct

9610

.0%

Jord

an1

not

Oct

91A

pr92

Jul

921.

0%N

oda

taN

oda

taJo

rdan

2no

tA

ug94

Feb

95N

ov95

1.0%

No

data

No

data

Kor

ea1

NB

not

Sep

89F

eb91

Apr

911.

0%S

ep95

Jul

96O

ct96

1.0%

Oct

90N

ov90

Dec

901.

0%K

orea

2A

ug86

Dec

86A

ug87

10.0

%N

B10

.0%

Kor

ea3

Aug

91O

ct92

Dec

9210

.0%

NB

10.0

%M

alay

sia

NB

not

Jul

90Ja

n91

Feb

935.

0%N

Bno

tN

Bno

tM

exic

o1

Apr

84M

ar86

Oct

875.

0%Ja

n82

Jan

83Ju

l83

5.0%

NB

not

NB

not

Mex

ico

2S

ep94

May

95Ju

n96

5.0%

Apr

86Ju

l86

Dec

865.

0%N

Bno

tN

Bno

tM

exic

o3

Aug

90M

ar91

Feb

925.

0%N

Bno

tN

Bno

tN

iger

iaM

ar95

May

95O

ct96

1.0%

NB

not

No

data

No

data

Pak

ista

n1

NB

not

Oct

90D

ec90

Apr

915.

0%N

Bno

tN

Bno

tP

akis

tan

2A

pr95

Mar

96Ju

n96

5.0%

NB

not

NB

not

(con

tinu

edon

next

page

)



App

endi

xT

able

3(co

ntin

ued)

Log

retu

rns

Log

divi

dend

yiel

dN

eteq

uity

flow

sto

equi

tyN

etbo

ndflo

ws

toG

DP

capi

taliz

atio

n

Cou

ntry

5th

Med

ian

95th

Sig

nif

5th

Med

ian

95th

Sig

nif

5th

Med

ian

95th

Sig

nif

5th

Med

ian

95th

Sig

nif

%ile

%ile

%ile

%ile

%ile

%ile

%ile

%ile

Phi

lippi

nes

1M

ay87

Aug

87Ju

n88

10.0

%N

Bno

tN

Bno

tN

Bno

tP

hilip

pine

s2

Jan

88Ju

l89

Jan

9010

.0%

NB

not

NB

not

NB

not

Por

tuga

lN

Bno

tN

Bno

tN

Bno

tS

ep90

Jan

92A

ug92

1.0%

Tai

wan

1A

pr86

Oct

87O

ct88

10.0

%N

Bno

tN

Bno

tM

ar90

Apr

90M

ay90

2.5%

Tai

wan

2N

ov90

Oct

91M

ay93

2.5%

Tha

iland

1N

Bno

tF

eb89

Jan

90M

ar90

2.5%

Jul

87A

ug88

Mar

891.

0%N

Bno

tT

haila

nd2

Sep

92A

pr93

Sep

9410

.0%

Tur

key

Feb

93M

ay94

May

9610

.0%

Jun

89S

ep89

Sep

9010

.0%

Dec

91no

tM

ar92

Jun

92Ju

l96

not

Ven

ezue

laN

Bno

tN

Bno

tN

Bno

tN

Bno

tZ

imba

bwe

NB

not

NB

not

No

data

not

noda

ta

Mul

tiple

brea

kte

sts

use

the

repa

rtiti

onm

etho

dsof

Bai

and

Per

ron

(199

8a,b

)w

hich

allo

wfo

ra

max

imum

of5

brea

ksw

ith15

%tr

imm

ing.

All

test

sar

epe

rfo

rmed

assu

min

g2

lags

inth

eau

tore

gres

sion

.E

stim

ated

brea

kda

tes

are

give

nin

the

colu

mn

labe

lled

“med

ian”

,al

ong

with

corr

espo

ndin

gco

nfide

nce

inte

rval

sin

the

colu

mns

labe

lle

d“5

th%

ile”

and

“95t

h%

ile”.

The

sign

ifica

nce

leve

lre

port

sth

elo

wes

tsi

gnifi

canc

ele

vel

for

whi

chth

ere

part

ition

proc

edur

efo

und

each

spec

ific

date

tobe

sign

ifica

nt.

NB

=N

oB

reak

.



For this part of the analysis, it is assumed that there is at most one break. It isalso assumed that the errors in the VAR have 4+� moments for some� 0. Thegeneral form of the regression is (equation 2.2 from BLS):

Yt � (G�t�In)q � dt(k)(G�t�In)S�Sd � et, (A1)

where Yt is n×1 (defined earlier),G�t is a row vector containing a constant andplags ofYt, In is ann×n identity matrix,dt(k) = 0 for t k anddt(k) = 1 for t�k. qand d are parameter vectors with dimensionr = n(np+1). S is a selection matrixcontaining zeros and ones and having column dimensionr and row dimension equalto the number of coefficients which are allowed to change (�r; i.e., S is full rowrank).

The procedure for determining when a potential break occurred involves estimat-ing (A1) for all possible break datesk∗+1�k�T�k∗, wherek∗ represents a trimmingvalue, often taken to be 15% of the sample size,T. At each possible break date, anF-statistic is computed, testing the significance ofSd, and is denotedF(k). Then thestatistic testing for structural change is equal to

maxk∗

�1�k�T�k∗F(k).

BLS show that this statistic converges via the functional central limit theorem tomax F∗, a (known) function of Brownian motion. More details and critical valuesare provided in BHL and BLS.

A confidence interval with asymptotic coverage of at least 95% is given by (eq.2.19 in BLS):

CI � (k̂�[�k]�1,k̂ � [�k] � 1),

where k̂ is the estimated break date, [·] denotes the “greatest least integer”, and

�k � c[(Sd̂)�S(Q̂1��̂�1k̂ )S�(Sd̂)]�1,

where c=7.63, �̂k̂ is the estimator of the variance-covariance matrix of the OLSresiduals under the alternative, givenk̂, and Q̂1 � 1

T�T

t�1GtG�t.

BP (Testing for multiple breaks in a single series)

The assumption of a single structural change seems less palatable in light of theAsian crisis. In addition, as we indicated above, some theoretical models of thedynamics of capital flows [see, e.g. Bachetta and van Wincoop (2000)] may alsolead to multiple breaks in the net flows as a percentage of market capitalization.Therefore, we investigate whether returns, dividend yields, and capital flows experi-enced multiple structural breaks, using techniques recently developed by Bai andPerron (1998a,b). Because multiple break analogs to the VAR framework of BLShave not yet been developed, we consider each series separately.

Rather than assuming the number of breaks is known a priori, Bai and Perronprovide econometric tests to determine the number of breaks. The necessary assump-



tions in Bai and Perron are not particularly restrictive and admit a wide variety oflinear specifications to identify the break dates, and construct confidence intervalsaround the estimated break dates. We use one of their set-ups that has serially uncor-related errors but allows for lagged dependent variables. As in Lumsdaine and Papell(1997), it is also assumed that the breaks are asymptotically distinct (intuitively, ifa large downward spike is immediately followed by a large upward one, this wouldbe considered one break, rather than two). Bai and Perron also show that the esti-mation of a single break when the underlying series has two breaks in its data-generating process results in consistent estimation of the break fraction for one ofthe breaks. In fact, the procedure consistently estimates the break of the larger magni-tude. Hence, our work assuming single breaks may still detect useful dates.

To implement the procedure for investigating multiple breaks, we follow the rec-ommendations in Bai and Perron (1998b).6 First, we use their double maximum teststo test the null hypothesis that there is no break versus the alternative that there isat least one break. The statistic is given by

max1�m�M

F∗T,

whereF∗T is a modified F-statistic given by equation (A2) in the technical appendix

below andM is the upper bound on the number of breaks. Critical values are givenin Bai and Perron (1998a,b), for various values ofM and k∗, the trimming value.7

Second, if there is evidence of at least one break, we implement their repartitionprocedure, which is based on comparing the sum of squared residuals from estimationof the ‘best’ (in a minimum sum of squared residuals sense)l-break model to thebestl+l-break model, beginning withl = 1. The number of breaksm is the first valueof l for which the test fails to reject the null hypothesis ofl breaks in favor of thealternative ofl+1 breaks. Finally, confidence intervals are then computed aroundthe break dates estimated using them-break model (formulas are given in Bai andPerron (1998b)).

We consider the full structural change model of Bai and Perron (that is, whereall coefficients are allowed to change). Using notation analogous to (A1), the modelcan be written as

Yt � �m � 1

i � 1

dt(ki)G�tdi � et,

wherem is the number of breaks and, as in model (A1),Gt consists of a constant andlags ofYt, anddt(ki) is an indicator variable equal to 0 whent ki and 1 otherwise.

A maximal F-test is used to test the hypothesis of no structural change versus thealternative ofm breaks. The null hypothesis is thus

6 We are grateful to Pierre Perron for supplying us with the Bai-Perron programs.7 The trimming value refers to the number of observations at either end of the sample where it is

assumed that no break has occurred; in earlier literature this was most often chosen to be 0.15T or 0.01TwhereT is the sample size.



H0: d1 � % � dm+1

and can be expressed asRd = 0 where

R � �1 �1 0 % 0 0 0

0 1 �1 % 0 0 0

� � � � � � �

0 0 0 % 1 �1 0

0 0 0 % 0 1 �1�.

Then theF-statistic is (this corresponds to equation (12) in Bai and Perron (1998a))

sup(l

1,%,l

m)��

F∗T(l1,%,lm;r) �

1T�T�(m � 1)r

mr �d̂�R�(RV̂(d̂)R�)�1Rd̂, (A2)

whered̂ is an estimate ofd, andd � (d�1,%,d�m � 1)�, V̂ is the estimate of its covari-ance matrix,li is the fraction of the sample at which breaki occurs,� is the spaceof all possible m-partitions andr is the number of columns inG.

One drawback of this approach is that the choice of how many lags to includemust now be determined exogenously, rather than by using an information criterion.However, the BP procedures do allow for robust (heteroskedasticity and autocorre-lation consistent) covariance estimation and for different variances of the errors ineach of the break periods, something the BLS test theory permits but is not allowedin current implementation. There is a cost associated with this flexibility, however;in general the BP tests appear to have less power to detect breaks than the BLStests. Thus it is important to consider both sets of results together when identifyingpossible break dates.8

Data appendix

Dividend yields

There are a few instances of zeros in the 12-month moving sum of dividends thatthe IFC reports. We investigated these zeros by looking at each individual firm’sdividends. There seems to be an issue of when the IFC recorded dividends. Forexample, in Korea, there are dividends paid in January 1988 and no dividends appearin the individual company files until March 1989. After that, there is no dividend

8 When we restrict the BP test to at most use one break, the number of lags of the BLS tests, and donot allow for robust covariance estimation or different variances, we replicate the break dates, confidenceintervals, and significance levels of the BLS tests, except for Indonesia and Portugal, two countries withvery short samples, where the break dates differ slightly.



paid by an individual company until April 1990. It is not surprising that we findzero entries in January and February 1989 and in March 1990. In order to avoidzero entries which appear to be a result of the timing of the recording of dividends,rather than a canceling of dividends, we carried forward some past dividends toreplace these holes in the data.

For Korea, we calculate the dividend on the index in December 1988 and use thatvalue as the numerator for the dividend yield calculation for January and February1989. The values (in percent) for these two months are 0.5511 and 0.5273, respect-ively. In February 1990, we calculate the dividend on the index and use that valuein the numerator for March 1990. The value is 1.0919.

For Indonesia, we calculate the value of the dividend in June 1991. That value isused in the numerator for July 1991 through February 1992. The dividend yieldsare: 0.1200, 0.1380, 0.1745, 0.1936, 0.1799, 0.1718, 0.1496 and 0.1311.

For Taiwan, we calculate the value of the dividend in February 1990 and use thatin the numerator for March 1990 through April 1991. The dividend yields for thisperiod are: 0.2250, 0.2600, 0.3340, 0.4545, 0.4186, 0.6355, 0.8307, 0.6825, 0.5314,0.5037, 0.5831, 0.4703, 0.4677 and 0.4048.

World interest rates

The nominal interest rate for the G-7 countries is calculated by aggregating indi-vidual countries’ short-term interest rates weighted by using countries’ previous quar-ter’s share in G-7 GDP. The following interest rates are employed: Canada 90-dayTreasury bill (IFS 60C), France 90-day bill (IFS 60C), Germany 90-day bill (IFS60C), Italy 180-day bill (IFS 60B), Japan commercial paper from 1975–1976 (IFS60B) and the Gensaki rate from 1977–1997 (IFS GBD3M), United Kingdom 90-daybill (IFS 60C), and the United States 90-day bill (IFS 60C).

References

Bachetta, P., van Wincoop, E., 2000. Capital flows to emerging markets Liberalization, overshooting andvolatility. In: Edwards, S. (Ed.), Capital Inflows and the Emerging Economies. University of ChicagoPress and NBER, Cambridge, MA, pp. 61–98.

Bai, J., Perron, P., 1998a. Computation and analysis of multiple structural change models, Unpublishedworking paper, Boston University, Boston, MA.

Bai, J., Perron, P., 1998b. Estimating and testing linear models with multiple structural changes. Econo-metrica 66, 47–78.

Bai, J., Lumsdaine, R.L., Stock, J.H., 1998. Testing for and dating breaks in stationary and nonstationarymultivariate time series. Review of Economic Studies 65 (3), 395–432.

Bartolini, L., Drazen, A., 1997. Capital account liberalization as a signal. American Economic Review87, 138–154.

Bekaert, G., Harvey, C.R., 1995. Time-varying world market integration. Journal of Finance 50, 403–444.Bekaert, G., Harvey, C.R., 2000a. Foreign speculators and emerging equity markets. Journal of Finance

55, 565–613.Bekaert, G., Harvey, C.R., 2000b. Capital flows and the behavior of emerging market equity returns. In:

Edwards, S. (Ed.), Capital Inflows and the Emerging Economies. University of Chicago Press andNBER, Cambridge, MA, pp. 159–194.



Bekaert, G., Harvey, C.R., Lumsdaine, R.L., 1999. The dynamics of emerging market equity flows,Unpublished working paper 7219, National Bureau of Economic Research, Cambridge MA.

Bekaert, G., Harvey, C.R., Lumsdaine, R.L., 2002. Dating the integration of world capital markets. Journalof Financial Economics, forthcoming.

Bohn, H., Tesar, L.L., 1996. US equity investment in foreign markets: Portfolio rebalancing or returnchasing? American Economic Review 86 (2), 77–81.

Bohn, H., Tesar, L.L., 1997. The US international investment portfolio and mean-variance optimization,Unpublished working paper, University of California at Santa Barbara, Santa Barbara, CA.

Calvo, G.A., Leiderman, L., Reinhart, C.M., 1993. Capital inflows and real exchange rate appreciationin Latin America: The role of external factors. IMF Staff Papers, Washington, DC.

Calvo, G.A., Leiderman, L., Reinhart, C.M., 1994. The capital inflows problem: Concepts and issues.Contemporary Economic Policy 12 (July), 54–66.

Calvo, G.A., Mendoza, E., 1998. Contagion, globalization and volatility of capital flows. In: Edwards, S.(Ed.), Capital Inflows and the Emerging Economies. University of Chicago Press and NBER, Chicago,IL, pp. 15–42.

Choe, H., Kho, B.-C., Stulz, R., 1999. Do foreign investors destabilize stock markets? The Korean experi-ence in 1997. Journal of Financial Economics 54, 227–264.

Clark, J., Berko, E., 1997. Foreign investment fluctuations and emerging market stock returns: The caseof Mexico, Staff report 24, Federal Reserve Bank of New York, New York, NY.

Edelen, R. M., Warner, J. B., 1999. Why are mutual fund flow and market returns related? Evidencefrom high-frequency data. Unpublished working paper, University of Rochester, Rochester, NY.

Fernandez-Arias, E., 1996. The new wave of private capital flows: Push or pull? Journal of DevelopmentEconomics 48, 389–418.

Froot, K.A., O’Connell, P.G.J., Seasholes, M.S., 2001. The portfolio flows of international investors.Journal of Financial Economics 59, 151–194.

Harvey, C.R., 1995. Predictable risk and returns in emerging markets. Review of Financial Studies 8,773–816.

Henry, P.B., 2000a. Equity prices, stock market liberalization, and investment. Journal of Financial Eco-nomics 58, 301–334.

Henry, P.B., 2000b. Stock market liberalization, economic reform, and emerging market equity prices.Journal of Finance 55, 529–564.

Kawakatsu, H., Morey, M. R., 1999. An empirical examination of financial liberalization and the efficiencyof emerging stock market prices. Journal of Financial Research, 385-411.

Kim, E.H., Singal, V., 2000. Opening up of stock markets: Lessons from emerging economies. Journalof Business 73, 25–66.

Lumsdaine, R.L., Papell, D.H., 1997. Multiple trend breaks and the unit root hypothesis. Review ofEconomics and Statistics 79 (2), 212–218.

Shleifer, A., 1986. Do demand curves for stocks slope down? Journal of Finance 41, 579–590.Stulz, R. M., 1999. International portfolio flows and security markets. In Feldstein, M., (Ed.), International

Capital Flows, National Bureau of Economic Research, forthcoming.Tesar, L., Werner, I., 1994. International equity transactions and U.S. portfolio choice. In: Frankel, J. (Ed.),

The Internationalization of Equity Markets. University of Chicago Press, Chicago, IL, pp. 185–215.Tesar, L., Werner, I., 1995. U.S. equity investment in emerging stock markets. World Bank Economic

Review 9 (1), 109–130.Warther, V.A., 1995. Aggregate mutual fund flows and security returns. Journal of Financial Economics

39, 209–235.World Bank, 1997. Private capital flows to developing countries: the road to financial integration. Oxford

University Press.


Documents

The dynamics of emerging market equity ﬂows · The dynamics of emerging market equity ﬂows ... macro-economic stabilization programs ... (2002) 295–350 297