Do foreign investors destabilize stock markets? The Korean experience in 1997

*Corresponding author. Tel.: #1-614-292-1970; fax: #1-614-292-2359.

E-mail address: [email protected] (R.M. Stulz)q We are grateful to Jan Jindra, Dong Lee, and Gueorgui Slavov for research assistance. We

received valuable comments from Warren Bailey, Joshua Feuerman, John Gri$n, CampbellHarvey, Andrew Karolyi, Andy Lo, Anil Makhija, N. Prabhala, G. William Schwert, MichaelRashes, Andrei Shleifer, Robert Shiller, Russ Wermers, Ingrid Werner, an anonymous referee, andseminar participants at the Federal Reserve Board, Korea University, London Business School,MIT, Seoul National University, The Ohio State University, Virginia Tech University, YaleUniversity, University of California at Berkeley, World Bank, and participants at meetings of theNBER Asset Pricing Group, Q-Group, and the Korean Finance Association.

Journal of Financial Economics 54 (1999) 227}264

Do foreign investors destabilize stock markets?The Korean experience in 1997q

Hyuk Choe!, Bong-Chan Kho!, ReneH M. Stulz",#,*!Seoul National University, College of Business Administration, Seoul, South Korea

"Department of Finance, Fisher College of Business, The Ohio State University, 1775 College Road,Columbus, OH 43210-1144, USA

#The National Bureau of Economic Research, 1050 Massachusetts Avenue, Cambridge, MA 02138, USA

Received 24 June 1998; received in revised form 21 January 1999

Abstract

This paper examines the impact of foreign investors on stock returns in Korea fromNovember 30, 1996 to the end of 1997 using order and trade data. We "nd strongevidence of positive feedback trading and herding by foreign investors before the periodof Korea's economic crisis. During the crisis period, herding falls, and positive feedbacktrading by foreign investors mostly disappears. We "nd no evidence that trades byforeign investors had a destabilizing e!ect on Korea's stock market over our sampleperiod. In particular, the market adjusted quickly and e$ciently to large sales by foreigninvestors, and these sales were not followed by negative abnormal returns. ( 1999Elsevier Science S.A. All rights reserved.

PACS: G11; G15; G28

Keywords: Foreign investors; Positive feedback trading; Herding; East Asian crisis

0304-405X/99/$ - see front matter ( 1999 Elsevier Science S.A. All rights reserved.PII: S 0 3 0 4 - 4 0 5 X ( 9 9 ) 0 0 0 3 7 - 9

1Cross-border equity #ows are dwarfed by other cross-border capital #ows [see Stulz (1999) forsome data] but no other #ows are documented at the trade level, so we could not conduct our studyon non-equity transactions by foreign investors.

2Stulz (1999) provides a review of the earlier studies that use monthly data. Bailey et al. (1998)investigate the Tequila e!ect focusing on order-imbalances on the NYSE without being able todistinguish trades by foreign and domestic investors. They "nd no signi"cant evidence of such ane!ect for foreign stocks traded on the NYSE.

1. Introduction

In the second half of 1997, it became extremely common for political leadersas well as journalists to argue that foreign investors exert a destabilizingin#uence on emerging market economies. Foreign investors were often blamedfor the dramatic di$culties of the East Asian countries and for the collapse oftheir currencies and stock markets. For instance, Stiglitz (1998) calls for greaterregulation of capital #ows, arguing that &2developing countries are morevulnerable to vacillations in international #ows than ever before'. Academicshave pointed out that foreign investors could have a destabilizing e!ect fora variety of reasons. Dornbusch and Park (1995) argue that foreign investorspursue strategies that make stock prices overreact to changes in fundamentalsand more recently Radelet and Sachs (1998) attribute the East Asian economiccrisis to "nancial panic. If foreign investors can indeed destabilize economies,the bene"ts from opening markets to investors from all countries are substan-tially weakened and perhaps reversed. It is therefore crucially important tounderstand whether this is the case.

This paper examines the impact of foreign investors on stock returns in Koreaover the period from November 30, 1996, to the end of 1997.1 In 1996, Koreawas bigger than Mexico but smaller than Canada in terms of GDP and of stockmarket capitalization. Our sample period includes Korea's dramatic economiccrisis during the last few months of 1997. A good measure of the intensity of thiscrisis is that a dollar invested in Korea's stock market index on October 1, 1997would have been worth 35 cents on the last day of trading of 1997. Informationfor each trade on the Korea Stock Exchange allows us to classify buying andselling investors into three categories: Korean individual investors, Koreaninstitutional investors, and foreign investors.

These data enable us to investigate the pattern and impact of net purchases byinstitutions, individuals, and foreign investors during the trading day as well asacross days. Since the data include the time that an order arrives at the market,we can identify all trades that are initiated by foreign investors. We can thereforeinvestigate stock returns around the time of trades initiated by foreign investors.Furthermore, a number of studies, e.g., Bohn and Tesar (1996) and Clark andBerko (1996), show a positive contemporaneous relation between equity #owsand stock returns using monthly data.2 Such a relation could hold if foreign

228 H. Choe et al. / Journal of Financial Economics 54 (1999) 227}264

3Nofsinger and Sias (1999) make a similar point concerning the assessment of herding bydomestic institutional investors using low-frequency (monthly or yearly) data. They note thatherding measured this way is consistent with positive feedback trading at high frequency as well aswith domestic institutional investors having a permanent impact on prices.

investors buy following positive returns and sell following negative returns, ifforeign investors have a permanent impact on stock prices, and/or if the foreigninvestors are good market timers and invest before prices increase.3 With ourdata, we can di!erentiate between these possible views of the relation betweenequity #ows and stock prices.

It is often argued that the trades of foreign investors are highly correlated, sothat they buy and sell as a herd. For instance, Krugman (1997) describes moneymanagers as &an extremely dangerous #ock of "nancial sheep'. This herdingcould make trades by foreign investors destabilizing because foreign investorsend up trading as a group, thereby creating disarray and possibly panic in themarkets that they exit and overheating in the markets that they enter. Toinvestigate herding, we use the measures developed in Lakonishok et al. (1992)and Wermers (1999) and "nd them to be signi"cantly positive. These herdingmeasures are large for stocks of all sizes compared to the herding measures forinstitutional investors in the U.S. During the Korean crisis period, the herdingmeasures appear to decrease when we look either at all foreign investors or onlyU.S. investors. The fact that these measures do not decrease for the larger stocksin our sample suggests that one possible reason for the decrease in herding isthat lower liquidity during the crisis may have limited the trading of foreigninvestors.

Dornbusch and Park (1995) and others contend that the trades of foreigninvestors are a!ected by past returns, so that they buy when prices haveincreased and sell when they have fallen. Such a practice is called positivefeedback trading, and it would lead to the herding we document. Theoreticalmodels have shown that investors who buy as prices increase and sell as pricesdecrease can exert a destabilizing in#uence on the stock market. DeLong et al.(1990) o!er an analysis of these potential destabilizing e!ects of positive feed-back trading. We investigate the period before the Korean crisis separately fromthe period during the crisis. Before the crisis, we "nd clear evidence of positivefeedback trading: foreign investors buy (sell) more Korean stocks on daysfollowing an increase (decrease) in the market as a whole and they buy (sell)Korean shares that outperformed (underperformed) the market over the pre-vious day. This evidence is consistent with the evidence of Froot et al. (1998).They investigate the relation between equity #ows and stock index returns withtrades of the institutions using State Street Bank & Trust as their repository, andthey conclude that past returns explain 60}85% of the quarterly covariancebetween stock index returns and equity #ows. When we turn to the crisis period,

H. Choe et al. / Journal of Financial Economics 54 (1999) 227}264 229

the evidence of positive feedback trading becomes much weaker. In particular,the sign of the previous day's market return has no information for the trades offoreign investors on the next day. We also "nd that the sign of the previous day'sstock return is not helpful in predicting the trades of foreign investors in thatstock the next day.

In theoretical models, concerns that positive feedback trading and herdinghave a destabilizing impact on prices arise because positive feedback traders canpush prices away from fundamentals when they form a herd. The argument isthat traders can make money by buying following price increases regardless ofwhether the price increases are rational or not because positive feedback traderswill push prices higher. Investors trading on fundamentals might make money inthe long run by selling short when prices are too high. Unfortunately for theseinvestors, positive feedback trading can keep prices increasing long enough toforce them to liquidate their positions before they start making pro"ts. Theargument that positive feedback trading and herding are destabilizing thereforeimplies that prices exhibit momentum. In other words, if positive feedbacktraders are selling, prices keep falling. For at least two reasons, however, positivefeedback trading and herding are not necessarily destabilizing. First, investorstrading on fundamentals may be su$ciently powerful in the markets to preventprices from moving away from fundamental values. Second, positive feedbacktraders may be trading in response to information about fundamentals, so thattheir trading does not drive prices away from fundamentals. Empirical researchis therefore required to settle the issue of whether positive feedback trading andherding are destabilizing.

To examine whether the trading practices of foreign investors are destabiliz-ing, we conduct two distinct event studies. In the "rst study, we measureabnormal returns for the 11 "ve-minute intervals centered on intervals withlarge foreign trades for the stocks in our sample. In the second study, we usedays instead of "ve-minute intervals. The hypothesis that foreign trades aredestabilizing can be rejected if additional price movements in the same directionas the price impact of the trades do not follow large foreign trades. In otherwords, if further price drops do not follow a large foreign sale, we conclude thatnet selling by foreign investors is not destabilizing. Presumably, one might beable to construct scenarios in which rational pricing implies momentum. Fromthe perspective of this study, however, the absence of momentum following largetrades by foreign investors means that we can reject the hypothesis that thesetrades are destabilizing.

Considering "rst the analysis using intraday data, we "nd that large buytrades initiated by foreign investors are associated with a stock price increase forthe "ve minutes during which the large buy trade takes place and during thenext "ve minutes. Thereafter, there are no positive signi"cant returns. Large selltrades initiated by foreign investors are associated with a stock price decline thatis partially reversed over the next 25 minutes of trading. The last three months of


4 Interestingly, the price impact around large foreign trades is not very di!erent from the evidencefor block trades in the U.S. For example, Holthausen et al. (1990) "nd that most of the e!ect ofa large block trade on the NYSE is permanent and takes place with the trade itself.

1997 do not seem to di!er from the rest of the sample with respect to theintraday returns surrounding large purchases or sales initiated by foreigninvestors.4 None of this evidence suggests that sales by foreign investors havea destabilizing impact.

When we look at daily returns, we focus on all trades by foreign investorsrather than only the trades they initiate. We "nd that the signi"cantly positivereturns on days with large net foreign buying of a stock are followed by reversalsprior to the last three months of 1997, but not during the Korean crisis period.Rather surprisingly, days with large net foreign selling of a stock have a positivemarket-adjusted return before the crisis period. Such a result is consistent withtrades by Korean individuals having a much stronger impact on returns thantrades by foreign investors. Days with large net selling by foreign investors aredays with large net buying by Korean individuals. Based on our evidence, theimpact of domestic buying on stock returns dominates the impact of foreignselling. During the last three months of 1997, days with large foreign net sellingdo not have signi"cant market-adjusted returns; even raw returns on these daysare not signi"cantly negative. There is therefore no convincing evidence thatforeign investors play a destabilizing role.

An interesting feature of the Korean stock market is the existence of dailyprice limits on individual stocks. The daily return of individual stocks cannotexceed 8% in absolute value during our sample period. Price limits make it morelikely that we will "nd e!ects consistent with a destabilizing in#uence of foreigninvestors because, on average, a stock that has fallen 8% in one day is likely tohave a negative return the next day. Hence, whenever foreign investors trade ondays when the limit is hit, one would expect a negative return the next day ratherthan a reversal. As discussed later, our conclusions are not a!ected by theexistence of price limits.

The paper proceeds as follows. In Section 2, we introduce our data andpresent information on foreign holdings of stocks. In Section 3, we test forherding. In Section 4, we investigate whether foreign investors engage in positivefeedback trading and how their trading compares to domestic trading. InSection 5, we look at the intraday and daily returns associated with large foreigntrades. We conclude in Section 6.

2. The Korea Stock Exchange and sample construction

The Korea Stock Exchange (KSE) holds two trading sessions on each week-day: a morning session and an afternoon session. The morning session operates


from 9:30 through 11:30, and the afternoon session starts at 13:00 and closes at15:00. Batch auctions are used three times a day to determine the opening pricesof each session and the daily closing prices. There are no trades during the lastten minutes of each day, when orders are collected for the closing batch auctionat 15:00. Trading prices during the rest of the trading hours are determined bycontinuous (or non-batch) auction. On Saturdays, there is only a morningsession and the closing price is determined by the batch auction. The KSE doesnot have designated market makers. Buyers and sellers meet via the AutomatedTrading System (ATS). Before November 25, 1996, only limit orders wereallowed, but since then markets orders have been allowed as well.

The database used is the intersection of two databases with raw data providedby the KSE and compiled by the Institute of Finance and Banking (IFB) atSeoul National University. The "rst database includes all transactions on theKSE for the period from 1993 to 1997. This database has each order time-stamped as of the time that it arrives at the exchange and as of the time that theorder is executed. The data provide information on the country of residence ofinvestors as well as on whether they are individuals or institutions. The seconddatabase has daily foreign ownership data from November 30, 1996, through theend of 1997 for all stocks listed on the KSE. Foreign investors in Korea mustregister with the Securities Supervisory Board (SSB) and obtain an ID numberbefore they can start trading stocks. The SSB uses this ID number to ensure thatthe foreign ownership limit for each company and each foreign investor is notexceeded by informing the KSE whether a foreign order satis"es the companyand investor limits. The ownership limit for each individual foreign investor was5% of a "rm's shares until May 2, 1997, when it increased to 6%. It thenincreased to 7% on November 3, and to 50% on December 11, 1997. Inaddition, foreign investors as a group could not own more than 20% of a "rm'sshares. This aggregate ownership limit on foreign investors increased to 23% onMay 2, 1997, to 26% on November 3, 1997, and "nally to 50% on December 11,1997. Foreign investors are not allowed to sell shares short. There is no record ofdaily foreign ownership available prior to November 30, 1996. Since we needboth databases for our study, we are therefore constrained to use the sampleperiod of the daily foreign ownership database. One limitation of the foreignownership database is that it is possible that trades we identify as foreign tradesare actually trades by Korean investors who set up a foreign nominee companyto trade on the KSE.

Over our sample period there are three Korean American DepositoryReceipts (ADRs) trading on the NYSE in addition to the Korea Fund. At theend of 1996, the shares corresponding to the ADRs represent a small fraction ofthe outstanding shares of the "rms that have issued ADRs (2.3% for Kepco,5.2% for Posco, and 1.8% for SK Telecom). Our data exclude New Yorktrading. If Korean residents trade in New York, we have no way of knowing it.This may not be important, however, because all ADRs are treated as shares


held by foreign investors that count towards the foreign ownership limit. Inaddition, New York trading does not appear to be important. NYSE ADR dailytrading represents less than 20% of the daily trading volume in Korea of thecompany's stock for the whole sample period and less than 10% during thecrisis.

We conduct our analysis on two types of trades. First, we use all trades,regardless of who initiates the trade. Since we are concerned about the impact oftrades by foreign investors on prices, we would like to identify those trades thatare most likely to a!ect prices. Thus, we consider what we call price-settingtrades. For NYSE data, it is common to identify the initiating party of a tradeusing a tick test such as the one proposed by Lee and Ready (1991). There is noneed for a tick test with our data because we know which party initiates thetrade. A buy-side (sell-side) price-setting trade for foreign investors is a trade inwhich the foreign investors' buy (sell) order comes after the sell-side (buy-side)order and hence makes the trade possible. For price-setting trades, we cantherefore only consider trades that take place during the non-batch auctionperiod. We consider split trades originated from one order as one trade irre-spective of whether we consider all trades or only price-setting trades.

The sample of stocks we use throughout the paper contains 414 of the 762stocks listed on the KSE at the end of November 1996 and is constructed asfollows. We start from the common stocks whose foreign investment ceiling was20% at the end of November 1996. Government-operated "rms and "rms withdirect foreign investment had a di!erent ceiling and were excluded. A largenumber of stocks have infrequent trading by foreign investors. We thereforerequire stocks to have more than 20 days of foreign price-setting trades fromDecember 2, 1996 to October 31, 1997. We also exclude 48 stocks that were atthe foreign ownership limit on December 2, 1996. Since foreign investors cannotbuy shares of these stocks on the exchange, we cannot use them for tests thatrequire foreign investors to be able to both buy and sell shares. These 48 stocksare interesting because they are larger "rms. The capitalization of the stocks atthe limit on December 2, 1996, is on average 521 billion Won and in total 21%of the market's capitalization, in contrast to the capitalization of the stocks inour sample which is on average 159 billion Won and in total 55% of themarket's capitalization. Since the ownership limits are higher during the crisisperiod and are no longer binding, we discuss results of an analysis of the impactof trades by foreign investors on the 48 stocks during the crisis.

Our dataset misses two types of transactions by foreign investors. First,foreign investors could buy Korean shares from other foreign investors on theover-the-counter market by o!ering a premium. Such trades enabled investorsindexing to a Korean benchmark to do so when stocks in the benchmarkwere at the foreign ownership limit. There would seem to be no reason forforeign investors to use the over-the-counter market for stocks where foreignownership is below the aggregate foreign investment ceiling. Consequently,


over-the-counter trades should not be relevant for our sample. Second, foreigninvestors could enter into over-the-counter derivatives transactions functionallyequivalent to trades in Korean shares, such as equity swaps on the KOSPIindex. Such transactions are not included in our dataset. Discussions withpractitioners suggest that such swap transactions occurred during the crisisperiod between equity portfolio managers and hedge funds.

Table 1 provides information on foreign ownership by size deciles at the endof November 1996 as well as information on foreign volume and foreignprice-setting volume for the period from December 2, 1996 to December 27,1997. As documented by Kang and Stulz (1997) for Japan, foreign ownership isstrongly positively related to size. For the smallest decile of Korean stocks,median foreign ownership is 2.09%. It increases to 13.48% for the largest decile.The average foreign ownership is 3.16% for the smallest decile and 11.89% forthe largest decile. Overall, average foreign ownership is 6.47% and medianforeign ownership is 4.52%. Although we do not reproduce these results ina table, we "nd that the median market model beta of the stocks held by foreigninvestors is 1.05 and the mean beta (against the KOSPI index) is 1.01. There istherefore no evidence that foreign investors choose stocks that have di!erentsystematic risk than the typical stock. Table 1 shows that foreign investors arenot trading disproportionately relative to their ownership, and the volume oftheir price-setting trades is not disproportionate either. These results contrastwith the results of Tesar and Werner (1995) who "nd that the turnover rates offoreign investors in Canada, the U.K., and the U.S. are higher than the turnoverrates of domestic investors in these markets.

It is useful to compare foreign ownership in our sample to foreign ownershipin the market as a whole. The average foreign ownership across the samplestocks is higher at the beginning of our sample period (6.47%) than in themarket as a whole (5.69%). In contrast, value-weighted foreign ownership,de"ned as the value of shares held by foreign investors divided by the marketcapitalization, is higher in the market as a whole (12.00%) than in our sample(9.38%). The di!erence between the equal-weighted and value-weightedmeasures arises because foreign investors hold more stocks in large companiesand the foreign ownership limit is more likely to be binding for large capitaliza-tion stocks than for other stocks.

Fig. 1 shows the time-series of the KOSPI index, the Won/USD exchangerate, and various measures of foreign ownership. In our sample, foreign inves-tors hold 4.62% of a "rm's shares on average at the end of the sample period ascompared to 6.47% at the beginning of the sample period. The same pattern,though less pronounced, holds for the market as a whole, where average foreignownership falls from 5.69% to 5.04%. Using a value-weighted measure offoreign ownership, however, there is no drop in ownership for the market aswhole. For the market as a whole, foreign investors own 12.00% of the marketcapitalization at the start of our sample period and 14.73% at the end. In


Tab

le1

Fore

ign

ow

ner

ship

,per

cent

age

ofdai

lytr

ade

volu

me

and

price

-set

ting

trad

evo

lum

efo

rth

ree

types

ofi

nve

stor

sat

the

KSE

from

Dec

.2,1

996

toD

ec.2

7,19

97.

The

sam

ple

consist

sof

414

com

mon

stock

sse

lect

edfrom

all7

62co

mm

onst

ockslis

ted

onth

eK

SEon

Nov.

30,1

996.

The

sam

ple

stoc

ksar

eth

ose

that

dono

thit

the

fore

ign

inve

stm

ent

ceili

ngof

20%

asof

Nov.

30,19

96.Sto

cks

with

ceili

ngs

oth

erth

an20

%ar

eex

clude

d(e

.g.,

those

with

fore

ign

dire

ctin

vest

men

tsh

ares

orgo

vern

men

t-op

erat

ed"rm

s).T

he

sam

ple

stock

sal

sohav

em

ore

than

20day

sof

fore

ign

pric

e-se

ttin

gtr

ades

from

Dec

.2,1

996

toO

ct.

31,19

97.T

heta

ble

show

sav

erag

esofth

efo

reig

now

ners

hip

acro

ssth

e41

4st

ocks

bysize

dec

ileas

ofN

ov.30

,19

96.A

lso

pre

sent

edar

eth

eav

erag

eper

centa

gesofd

aily

trad

evo

lum

ean

dpr

ice-

sett

ing

trad

evo

lum

eat

trib

uted

toin

stitutions,

indi

vidua

ls,a

nd

fore

igner

sfrom

Dec

.2,1

996

toD

ec.2

7,19

97.

Aprice

-set

ting

buy

(sel

l)tr

ade

isde"

ned

asa

trad

ew

here

the

buy

-sid

e(sel

l-side

)ord

eris

rece

ived

atth

eex

chan

gela

terth

anth

ese

ll-side

(buy

-sid

e)ord

er.

Firm

size

Dec

ileFore

ign

ow

ners

hip

(%)

Per

cent

age

ofda

ily

trad

evo

lum

e(%

)P

erce

nta

geofdai

lypr

ice-

sett

ing

volu

me

(%)

dof

stock

sM

ean

Med

ian

Sto

ck-d

ays

Inst

itution

Indi

vidua

lFore

igne

rSto

ck-d

ays

Inst

itution

Indi

vidua

lFore

igne

r

1(S

mal

lest

)41

3.16

2.09

12,6

816.

4191

.43

1.33

11,4

036.

2292

.04

1.10

241

4.31

2.77

12,6

228.

7387

.92

2.19

12,3

508.

0189

.04

2.03

342

5.94

3.95

12,7

889.

9986

.38

2.75

12,2

109.

3887

.43

2.53

441

5.37

3.74

12,7

0411

.27

85.3

32.

4612

,234

10.6

186

.13

2.35

542

4.95

3.50

12,9

6012

.21

83.3

93.

2612

,450

11.7

084

.13

3.12

641

5.13

3.49

12,7

1113

.08

82.2

63.

4912

,463

12.6

182

.87

3.54

742

6.08

4.11

12,9

9013

.49

81.2

83.

8012

,764

12.8

782

.09

3.80

841

8.22

6.23

12,7

4417

.10

76.7

44.

8312

,050

17.1

576

.67

5.01

942

9.64

8.85

13,0

8717

.64

73.6

27.

3412

,847

17.7

973

.91

7.22

10(L

arge

st)

4111

.89

13.4

812

,753

18.6

067

.31

12.8

612

,711

18.6

167

.87

12.4

4

All

(414

stock

s)41

46.

474.

5212

8,04

012

.87

81.5

44.

4412

3,48

212

.58

82.0

74.

37


Fig. 1. Time-series plots of the daily foreign ownership for 414 stocks by three groups (F1}F3) basedon their foreign ownership rankings as of Nov. 30, 1996, the daily value-weighted foreign ownershipcomputed as total value of foreign holdings to total market value (for 762 common stocks), theKOSPI daily index, and the Won/USD daily exchange rates for the period from Nov. 30, 1996 toDec. 27, 1997.

contrast, in our sample, foreign investors own 9.38% at the start of our sampleperiod and 9.04% at the end. It is important to note, however, that whilethe share of the market capitalization held by foreign investors increased for themarket as a whole, the dollar value of the Korean shares held by foreigninvestors fell dramatically because of the fall both in equity values and in thedollar price of the Won, which is documented in Fig. 1.

3. Do foreign investors herd?

In this section, we evaluate the extent to which foreign investors herd bothbefore and during the crisis period. There is a growing theoretical and empiricalliterature on herding. Recent papers by Nofsinger and Sias (1999) and Wermers(1999) contain overviews of the theoretical and empirical literatures on herding.In this literature, herding can be the outcome of investors using the sameinformation to trade or the product of irrational psychological factors. Our testsof herding take the narrow and simple view of herding that is prevalent in theempirical literature. We consider that foreign investors herd if they tradesimilarly over a short interval of time, namely a day. We therefore do not test


5Trading from Malaysia is unlikely to be trading by Malaysian investors. The government ofMalaysia established in 1990 an international o!shore "nancial center in the federal territory ofLabuan, which a lawyer described to us as the equivalent of the Cayman Islands in the Asia-Paci"cregion. Although we cannot identify Korean investors using the Labuan "nancial center to trade inKorea under nominee names, our results do not change even after excluding Malaysian investors asa whole.

alternate models in which herding takes place over longer periods of time in thispaper. Such tests would be useful, but the crisis period is too short to make itpossible to implement such tests. Investigating herding would not be veryinteresting if there were few foreign investors. This is not the case. At the end of1996, there were 5,294 registered foreign investors. Interestingly, this numberincreased sharply over the last two years. In June 1998, there were 7,998registered foreign investors. At that time, U.S. investors represented 37.7% ofthe total number of foreign investors.

As a starting point for our investigation, Table 2 provides statistics for tradingby foreign investors during our whole sample period for the 414 stocks in oursample. For the 16 countries with trading reported separately, we "nd that 14countries are net sellers of shares. More importantly, when we compute thedollar value of shares sold and the dollar value of shares bought for each of the16 countries using Won transaction prices and daily exchange rates, we "nd thatthe dollar value of the purchases exceeds the dollar value of the sales for onlyfour countries. The preponderance of net selling cannot be explained by chanceat reasonable signi"cance levels if the sign of the net trading is random acrossinvestors. This result is, however, consistent with herding. Most of the trading isdone by investors from the U.K., U.S., and Malaysia.5 The U.S. investors buyand sell the most shares. U.S. investors sell more shares than they buy, but thevalue of their purchases exceeds the value of their sales because they buy higherpriced shares than they sell. The U.K. investors trade less but their net selling ofshares is dramatic compared to the net selling of shares of U.S. investors. TheU.K. investors bought 64 million shares, but sold 123 million shares. Finally, theMalaysian investors are the second most active group of investors. Investorsfrom Germany and Taiwan actually bought more shares than they sold. We alsoconstructed but do not report a table like Table 2 for the 762 stocks traded onthe Korea Stock Exchange. The results for that table are similar to those shownhere, except that the dollar amount of net selling is smaller. The reason for that isthat foreign investors bought more shares of the stocks at the foreign ownershiplimit as that limit was relaxed.

To investigate whether foreign investors herd, we follow the approach ofLakonishok et al. (1992) and Wermers (1999) to estimate the importance ofherding. We compute their herding measure using a daily horizon and treat eachtrade on a day as made by a di!erent foreign investor since we do not haveidenti"cations for individual foreign investors. Speci"cally, the herding measure


Tab

le2

Tra

ding

activi

tyoffo

reig

nin

vest

ors

by

coun

try

atth

eK

SE

from

Dec

.2,

1996

toD

ec.27

,199

7.

For

each

ofth

efo

reig

nbuy

and

sell

trad

esat

the

KSE

from

Dec

.2,

1996

toD

ec.

27,

1997

,th

efo

reig

nin

vest

or's

count

ryis

iden

ti"ed

usin

gth

eco

rres

pond

ing

code

sin

the

IFB

/KSE

dat

abas

e.The

sam

ple

stock

sin

clude

414

stock

sse

lect

edfrom

all7

62co

mm

on

stock

slis

ted

onth

eK

SEon

Nov.

30,1

996.

The

table

show

sth

etr

adin

gvo

lum

eby

fore

ign

inve

stor

sag

greg

ated

acro

ssth

eirco

untr

iesan

dex

pres

sed

both

inte

rmsoft

hou

sand

sof

shar

esan

ddolla

rs.The

dolla

rvo

lum

eis

valu

edat

each

tran

sact

ion

price

inK

ore

anW

onan

dth

enco

nver

ted

into

dol

lars

using

the

dai

lyex

chan

gera

te.

Fore

ign

inve

stors'

countr

y

Shar

esboug

htShar

esso

ldN

etbo

ugh

tA

mount

bough

tA

mount

sold

Net

boug

ht

('000

shrs

)(%

)('0

00sh

rs)

(%)

('000

shrs

)('0

00U

$)(%

)('0

00U

$)(%

)('0

00U

$)

US

132,

019

(31.

1)14

8,03

1(2

5.1)

!16

,012

1934

,366

(28.

6)15

93,4

86(2

1.2)

340,

880

UK

64,4

34(1

5.2)

123,

394

(20.

9)!

58,9

6010

36,6

13(1

5.3)

1400

,242

(18.

6)!

363,

629

Irel

and

45,2

86(1

0.7)

72,6

74(1

2.3)

!27

,388

893,

825

(13.

2)11

44,7

44(1

5.2)

!25

0,91

9G

erm

any

1,04

8(0

.2)

856

(0.1

)19

212

,841

(0.2

)10

,848

(0.1

)1,

993

Fra

nce

4,95

3(1

.2)

6,94

9(1

.2)

!1,

996

76,7

96(1

.1)

83,9

51(1

.1)

!7,

155

Can

ada

14,2

54(3

.4)

16,9

16(2

.9)

!2,

662

165,

873

(2.4

)16

8,01

9(2

.2)

!2,

146

Swiss

7,45

8(1

.8)

17,4

24(3

.0)

!9,

966

124,

681

(1.8

)22

1,78

0(2

.9)

!97

,099

Net

herlan

ds

3,20

1(0

.8)

8,75

8(1

.5)

!5,

557

61,0

74(0

.9)

92,0

28(1

.2)

!30

,954

Luxe

mbou

rg12

,574

(3.0

)14

,659

(2.5

)!

2,08

520

6,82

0(3

.1)

183,

421

(2.4

)23

,399

Aust

ralia

6,74

1(1

.6)

10,0

57(1

.7)

!3,

316

95,8

19(1

.4)

124,

544

(1.7

)!

28,7

25N

ewZea

land

5,38

0(1

.3)

7,99

8(1

.4)

!2,

618

122,

664

(1.8

)12

5,55

6(1

.7)

!2,

892

Japa

n1,

758

(0.4

)3,

482

(0.6

)!

1,72

428

,460

(0.4

)32

,726

(0.4

)!

4,26

6M

alay

sia

105,

666

(24.

9)13

3,53

6(2

2.6)

!27

,870

1674

,544

(24.

7)19

82,0

99(2

6.3)

!30

7,55

5Tai

wan

1,27

8(0

.3)

595

(0.1

)68

314

,418

(0.2

)7,

793

(0.1

)6,

625

Singa

pore

2,68

5(0

.6)

3,88

4(0

.7)

!1,

199

52,4

09(0

.8)

70,0

46(0

.9)

!17

,637

Hong

Kon

g1,

049

(0.2

)2,

535

(0.4

)!

1,48

617

,607

(0.3

)26

,874

(0.4

)!

9,26

7O

ther

s(3

1co

untr

ies)

14,6

59(3

.5)

17,9

99(3

.1)

!3,

340

252,

122

(3.7

)25

7,63

6(3

.4)

!5,

514

Tota

l42

4,44

3(1

00.0

)58

9,74

7(1

00.0

)!

165,

304

6770

,932

(100

.0)

7525

,793

(100

.0)

!75

4,86

1


is computed as Dpit!E(p

it)D!EDp

it!E(p

it)D, where p

itis the proportion of

foreign investors buying stock i on day t among all foreign investors trading thatstock on that day, E(p

it) is the expected proportion of foreign investors buying

on day t relative to all foreign investors. EDpit!E(p

it)D is an adjustment factor

computed assuming that in the absence of herding the number of foreigninvestors with net purchases follows a binomial distribution that we compute asin Wermers (1999). We compute this herding measure for each stock each dayusing all 414 stocks. We then create portfolios based on size and past-weekreturns and take an average across stocks for each portfolio. The portfolios areequally weighted and rebalanced every week.

Table 3 shows estimates of the herding measure using all foreign investors. Itis immediately apparent that the herding measures were uniformly positivebefore the Korean crisis. Wermers (1999) provides herding measures for U.S.mutual funds from 1975 to 1994. He splits his sample across size quintiles andpast-return quintiles. Looking at his data across all funds, the highest herdingmeasure he "nds is 8.39%, except for the smallest stocks. In contrast, before thecrisis, no herding measure we report is below 20%. Wermers (1999) looks atherding over a quarter. In contrast, we measure herding on a daily basis. Whenmeasured over longer time intervals, herding could increase if not all institutionsthat move in the same direction do so on the same day or it could decrease ifsimilar trades by institutions lead some institutions to trade in the oppositedirection. The "rst possibility would seem to be more important than the secondone, so that we may understate herding. In contrast to Wermers (1999), it doesnot seem that herding is systematically related to prior return or size. Whilethere is strong evidence of herding prior to the Korean crisis, there is noevidence that herding is more important during the crisis period. Twenty-four of25 measures are lower during the crisis period than before and 18 are signi"-cantly lower at the 10% level using t-tests for mean di!erences. As explainedearlier, though, herding does not have to be destabilizing.

In Table 3, we treat each purchase as a purchase by a distinct foreign investor,which may overstate the degree of herding, since the same foreign investor maybuy or sell the same stock several times during the same day. One way we avoidthis problem is to investigate herding across classes of foreign investors ratherthan across all foreign investors. Using the identi"cation codes in our data forthe country of residence as well as the investor type, we attribute each trade byforeign investors to one of 658 classes based on both 47 countries and 14investor types. Foreign investors are divided into resident and non-resident. Theclasses of non-resident foreign investors are individuals, banks, insurance com-panies, securities "rms, mutual funds divided into two types (corporate types,which issue stocks, and contract types, which issue bene"ciary certi"cates withgenerally a stated maturity), other corporations, non-resident Korean investors,and pension funds. (Non-resident Korean investors account for less than 1/100of 1% of the shares sold by foreign investors.) There are also "ve classes of


Tab

le3

Lak

oni

shok

etal

.(19

92)he

rdin

gm

easu

res

(inper

centg

es)fo

rfo

reig

nin

vest

ors

bysize

and

past

-wee

kre

turn

por

tfolios

on

the

KSE

stoc

ks

To

com

put

eher

din

gm

easu

resfo

rfo

reig

nin

vest

ors

,eac

hof

thefo

reig

nbu

yan

dse

lltr

ades

on

asa

mple

oft

he41

4st

ock

sat

theK

SE

from

Dec

.2,1

996

toD

ec.2

7,19

97is

assu

med

toorigi

nat

efrom

ase

par

ate

fore

ign

inve

stor

,us

ing

the

iden

ti"ca

tion

code

sin

the

IFB

/KSE

data

bas

e.The

herd

ing

mea

sure

for

agi

ven

stock

-day

isth

enco

mpu

ted

asDp

it!E

(pit)D!

EDp

it!

E(p

it)D,w

here

p itis

the

pro

port

ion

offo

reig

nin

vest

ors

buyi

ng

stoc

kion

day

tam

ong

allfo

reig

nin

vest

ors

trad

ing

that

stoc

kon

that

day

,E(p

it)is

the

expec

ted

pro

port

ion

offo

reig

nin

vest

orsbuy

ing

onth

atda

yre

lative

toal

lfo

reig

nin

vest

ors

active

on

that

day

,and

EDp

it!E

(pit)D

isth

ead

just

men

tfa

ctorco

mput

edund

erth

ehy

pot

hes

isth

atin

the

abse

nce

ofh

erdin

gth

enum

berofp

urc

hase

sis

binom

ially

distr

ibute

d.T

hest

ock

isex

clude

dif

itis

trad

edby

less

than

two

fore

ign

inve

stor

son

day

t.The

herd

ing

mea

sure

com

pute

dab

ove

forea

chst

ock

-day

isth

enav

erag

edw

ithi

nsize

and

pas

t-w

eek

retu

rnpo

rtfo

lios(b

oth

inte

rmsoft

he

U.S

.dol

lar),w

hich

are

reba

lance

dev

ery

wee

k.The

t-st

atistics

for

the

mea

ns

are

pre

sent

edin

par

enth

eses

,an

dth

enu

mber

ofst

ock

-day

sar

epr

esen

ted

belo

wth

et-st

atistics

.Tes

tst

atistics

for

di!

eren

cein

mea

ns

(ass

umin

gun

equal

varian

ces)

and

med

ians

(using

the

Wilco

xon

rank-

sum

test

)ac

ross

two

subp

erio

ds

are

pres

ente

din

bra

ces

and

brac

kets

,res

pect

ivel

y.

Pas

t-w

eek

retu

rnpo

rtfo

lio

Bef

ore

Kore

anC

risis

(Dec

.2,1

996}

Sep

t.30

,199

7)D

urin

gK

ore

anC

risis

(Oct

.1,1

997}

Dec

.27

,199

7)D

i!er

ence

inm

eans

and

med

ians

(bef

ore

-after

)

S1 (Sm

alle

stC

ap)

S2S3

S4S5 (L

arge

stC

ap.)

S1 (Sm

alle

stC

ap.)

S2S3

S4S5 (L

arge

stC

ap.)

S1 (Sm

alle

stC

ap.)

S2S3

S4S5 (L

arge

stC

ap.)

P1

22.8

5723

.36

22.8

4723

.234

21.8

0619

.495

18.8

8417

.684

16.8

7818

.483

M1.6

8NM2

.40N

M3.1

3NM5

.65N

M3.6

5N(L

ow

est)

(28.

59)

(33.

30)

(35.

69)

(41.

22)

(50.

59)

(10.

60)

(10.

94)

(11.

64)

(17.

32)

(23.

08)

[2.9

1][3

.85]

[4.3

8][7

.25]

[4.9

9]41

355

061

688

514

8911

411

614

928

141

8

P2

22.7

4324

.899

23.8

3121

.475

22.3

8119

.861

26.2

8218

.978

19.0

4918

.693

M1.4

6NM!

0.72

NM3

.17N

M1.8

9NM3

.69N

(29.

77)

(39.

33)

(37.

73)

(40.

57)

(49.

75)

(10.

88)

(14.

51)

(13.

61)

(16.

29)

(20.

97)

[2.3

2][!

0.07

][4

.80]

[3.7

3][5

.59]

458

545

573

971

1455

9413

919

326

444

4

P3

23.1

1124

.958

24.3

8923

.222

21.8

4516

.161

22.0

1619

.481

21.1

7818

.795

M3.0

7NM1

.41N

M3.1

0NM1

.51N

M3.1

6N(2

8.20

)(3

7.93

)(3

9.43

)(4

1.12

)(4

7.45

)(7

.66)

(11.

08)

(13.

37)

(17.

26)

(22.

12)

[4.0

5][2

.01]

[4.7

2][4

.03]

[4.9

6]36

551

864

687

514

4396

110

164

288

395

P4

24.8

1422

.355

24.4

8523

.174

23.2

4116

.058

16.4

6916

.729

16.9

7219

.484

M4.4

5NM3

.31N

M5.2

9NM5

.38N

M3.6

4N(3

6.50

)(3

2.16

)(4

1.65

)(4

0.32

)(5

3.94

)(8

.70)

(10.

07)

(12.

45)

(17.

00)

(20.

80)

[5.4

8][4

.46]

[7.6

7][7

.18]

[5.6

1]46

254

664

384

514

6287

117

174

287

425

P5

23.6

1623

.391

23.3

1323

.113

22.8

7423

.193

19.1

7517

.593

21.7

9121

.27

M0.1

6NM2

.11N

M3.2

8NM0

.96N

M1.4

8N(H

ighe

st)

(31.

85)

(29.

28)

(36.

58)

(41.

95)

(53.

10)

(9.4

1)(1

0.47

)(1

0.83

)(1

7.22

)(2

1.36

)[1

.02]

[3.7

2][5

.09]

[1.7

9][3

.24]

390

461

654

883

1461

9412

614

527

239

7


resident foreign investors: individual, bank, insurance, securities "rms, and othercorporations. If the net purchase on a stock by one of the classes during a day ispositive (negative), we count it as a buy (sell) for the stock on that day by theclass of foreign investors, and then compute the herding measure as describedearlier. This approach aggregates all trades on a stock during a day by a class offoreign investors, e.g., U.S. mutual funds, in either one buy or one sell. Since weaggregate many buys and sells into one trade with this measure, it couldunderstate herding substantially by ignoring herding within classes. It is there-fore not surprising that Table 4 reports lower herding measures than Table 3.The average of the herding measures over all stock-days falls from 22.2% inTable 3 to 3.5% in Table 4. The magnitude of this average is more similar to thatof Wermers (1999), but remember that the herding measures of Table 4 representa lower bound for herding among foreign investors and that they are computedfor a day. Twenty-two out of 25 measures are signi"cantly positive at the 10%level before the crisis. Herding drops o! during the crisis, but not for the largeststocks. This is consistent with lower liquidity reducing the ability of foreigninvestors to trade in the smaller stocks.

4. Are foreign investors positive feedback traders?

Investors can be positive feedback traders for rational reasons or because ofbehavioral biases. Investors who pursue portfolio insurance strategies as well asinvestors with extrapolative expectations are positive feedback traders. Inves-tors with such strategies are often viewed as destabilizing because their sales leadthe market to fall further and their purchases increase prices further. Positivefeedback traders are often blamed for the stock market crash of 1987. In somemodels, positive feedback trading leads to bubbles, where prices depart fromfundamentals, and to crashes when bubbles burst. Besides contributing to thevolatility of stock returns, it is argued that such trading leads to destabilizingcapital #ows because equity investors rush into countries whose stock marketsare booming and #ee from countries whose stock markets are falling.

Foreign investors may act like positive feedback traders without destabilizingequity markets. One reason is that greater foreign ownership can lead to a lowerrisk premium for stocks in a country since the risks of these stocks can be bettershared internationally. Stulz (1998) provides an analysis of the relation betweenforeign ownership and the risk premium and a review of the evidence. Asa result, a period when foreign investors enter a market can also be a periodwhen the market is doing well because of these investors. Equity markets alsobecome more receptive to foreign investors as economies liberalize. Liberalizationitself leads to stock market appreciation and in this scenario this appreciation isfollowed by in#ows of foreign equity investments. Bekaert and Harvey (1998,2000) and Henry (2000) give evidence on the relation between liberalization


Tab

le4

Lak

oni

shok

etal

.(19

92)he

rdin

gm

easu

res

(inper

centa

ges)

for

fore

ign

inve

stor

clas

ses

by

size

and

past

-wee

kre

turn

port

folio

son

the

KSE

stock

s.

To

com

pute

herd

ing

mea

sure

sfo

rfo

reig

nin

vest

or

clas

ses,

each

offo

reig

nbuy

and

sell

trad

eson

asa

mple

ofth

e41

4st

ock

sat

the

KSE

from

Dec

.2,1

996

toD

ec.2

7,19

97is

attr

ibut

edto

one

ofth

efo

reig

nin

vest

or

clas

ses

(658

clas

ses

acro

ss47

count

ries

and

14in

vest

or

type

s),u

sing

the

iden

ti"ca

tion

codes

inth

eIF

B/K

SEdat

abas

e.T

heher

din

gm

easu

refo

ra

give

nst

ock

-day

isth

enco

mpu

ted

asDp

it!

E(p

it)D!EDp

it!

E(p

it)D,

whe

rep it

isth

epr

opor

tion

offo

reig

nin

vest

orcl

asse

snet

buyi

ng

stock

ion

day

tam

ong

all

fore

ign

inve

stor

clas

sestr

adin

gth

atst

ock

on

that

day

,E(p

it)i

sth

eex

pect

edpr

opor

tion

off

orei

gnin

vest

orcl

asse

sne

tbuyi

ng

on

that

day

rela

tive

toal

lfor

eign

inve

storcl

asse

sac

tive

,and

EDp

it!E(p

it)Dis

thead

just

men

tfac

torco

mput

edund

erth

enul

lofn

oher

din

gth

atth

enum

ber

ofn

etbuy

ing

clas

sesis

bin

omia

llydistr

ibute

d.T

hest

ock

isex

clud

edif

ittr

aded

by

less

than

two

fore

ign

inve

stor

clas

seson

day

t.T

he

her

din

gm

easu

reco

mpute

dab

ove

forea

chst

ock

-day

isth

enav

erag

edw

ithi

nsize

and

pas

t-w

eek

retu

rnpo

rtfo

lios

(bot

hin

term

softh

eU

.S.dol

lar),w

hic

har

ere

bal

ance

deq

ual

lyev

ery

wee

k.The

t-st

atistics

for

the

mea

nsar

epr

esen

ted

inpa

renth

eses

,an

dth

enu

mber

ofst

ock

-day

sar

epr

esen

ted

bel

ow

the

t-st

atistics

.Tes

tst

atistics

fordi!er

ence

inm

eans(a

ssum

ing

uneq

ualv

aria

nce

s)an

dm

edia

ns(u

sing

the

Wilc

oxon

rank

-sum

test

)acr

osstw

osu

bpe

riod

sar

epr

esen

ted

inbr

aces

and

bra

cket

s,re

spec

tive

ly.

Pas

t-w

eek

retu

rnpo

rtfo

lio

Bef

ore

Kore

anC

risis

(Dec

.2,

1996}Se

pt.30

,199

7)D

urin

gK

ore

anC

risis

(Oct

.1,

1997}D

ec.27

,199

7)D

i!er

ence

inm

eans

and

med

ians

(bef

ore

-dur

ing)

S1 (Sm

alle

stC

ap.)

S2S3

S4S5 (L

arge

stC

ap.)

S1 (Sm

alle

stC

ap.)

S2S3

S4S5 (L

arge

stC

ap.)

S1 (Sm

alle

stC

ap.)

S2S3

S4S5 (L

arge

stC

ap.)

P1

6.94

29.

351

3.87

02.

482

2.57

40.

869

4.73

4!

0.52

15.

472

2.44

8M1

.79N

M1.5

1NM2

.44N

M!1.

52N

M0.1

1N(L

ow

est)

(4.0

4)(5

.61)

(3.0

9)(2

.53)

(4.1

8)(0

.30)

(1.8

5)(!

0.40

)(3

.22)

(2.4

3)[1

.68]

[1.4

4][1

.83]

[!1.

54]

[0.0

3]12

614

424

738

672

529

5496

119

213

P2

4.12

96.

352

2.41

60.

224

3.88

64.

626

5.24

91.

433

3.59

63.

909

M!0.

17N

M0.4

5NM0

.47N

M!1.

95N

M!0.

02N

(2.6

2)(4

.68)

(1.9

6)(0

.28)

(6.3

9)(1

.92)

(2.5

8)(0

.85)

(2.3

5)(4

.13)

[0.0

4][0

.50]

[0.1

8][!

2.05

][!

0.01

]15

522

526

151

478

740

7889

135

264

P3

4.34

46.

427

2.34

12.

491

3.16

52.

854

8.26

10.

524

0.69

43.

901

M0.3

8NM!

0.64

NM0

.94N

M1.0

3NM!

0.59

N(2

.17)

(4.3

9)(1

.98)

(2.4

8)(4

.50)

(0.8

3)(3

.38)

(0.3

4)(0

.48)

(3.7

7)[0

.29]

[!0.

30]

[0.6

3][0

.63]

[!0.

48]

112

201

254

363

639

3044

9810

720

3

P4

8.23

65.

855

3.35

33.

907

3.65

41.

236

1.22

80.

941

1.03

56.

488

M2.4

0NM1

.74N

M1.1

9NM1

.73N

M!2.

09N

(4.4

6)(3

.99)

(2.6

5)(4

.10)

(5.6

4)(0

.55)

(0.5

5)(0

.59)

(0.7

6)(5

.45)

[2.1

2][1

.36]

[1.0

2][1

.37]

[!1.

64]

126

202

237

399

729

3755

9315

124

2

P5

2.59

20.

522

4.54

14.

294

3.50

16.

098

3.78

55.

389

2.30

74.

235

M!0.

68N

M!1.

11N

M!0.

36N

M1.1

3NM!

0.53

N(H

ighe

st)

(1.3

3)(0

.32)

(3.9

2)(4

.32)

(5.5

6)(1

.28)

(1.5

5)(2

.63)

(1.5

9)(3

.41)

[!0.

33]

[!1.

29]

[!0.

23]

[0.7

6][!

0.28

]10

515

426

438

373

322

5697

118

193


Fig. 2. Time-series plots of the daily foreign net buy-sell volume in shares (order imbalances) forthe 414 sample stocks, the KOSPI daily index, and the Won/USD daily exchange rates for theperiod of Nov. 30, 1996 to Dec. 27, 1997.

and stock market appreciation. Finally, in models that emphasize informationasymmetries between domestic and foreign investors, as in Brennan and Cao(1997), foreign investors learn more from stock returns than domestic investorsdo because stock prices impound the information that domestic investors have,so price increases reveal the domestic investors' favorable information to foreigninvestors. Since this information leads investors to have more favorable expecta-tions for stock returns, it leads them to invest more without acting irrationally.

Fig. 1 suggests that, if positive feedback trading by foreign investors hada signi"cant impact on stock price behavior in Korea, it must have been subtlerthan implied by the most aggressive critiques of the in#uence of foreign inves-tors. Fig. 2 shows the net buying of foreign investors on a daily basis. It is clearthat, starting with the end of September 1997, foreign investors are net sellersevery day for the stocks in our sample. This is consistent with a withdrawal offoreign equity investors. It has to be kept in mind, however, that when we lookat levels of foreign ownership there is simply no collapse so that foreign investorsmust have bought stocks that were at the foreign ownership limit at the end of1996 and sold stocks in our sample. At the same time, the behavior of the foreigninvestors in Fig. 2 is largely consistent with the view that these investors sell infalling markets and buy in rising markets. The periods of mostly foreign netbuying occur when stock prices are increasing, and the periods of persistent netselling by foreign investors are periods of falling stock prices.


6Note that there is no reason for the averages to sum to zero across trader types.

To investigate whether the impression of positive feedback trading given byFig. 2 is statistically signi"cant, we consider the price-setting order imbalance ofinvestors, conditioning on the sign of the market return of the prior day, of thesame day, and of the next day. A stock's price-setting order imbalance iscomputed as the price-setting buy volume minus the price-setting sell volume bya class of investors for a day normalized by the stock's average daily price-setting volume over the sample period. In Table 5, we show equally-weightedaverages of the normalized price-setting order imbalances across stocks.6 Thereis evidence that foreign investors buy following a positive market return and sellfollowing a negative market return before the Korean crisis. The same is true,however, for Korean individual investors. The contemporaneous relation be-tween order imbalance and the sign of the market return is dramatic for Koreanindividual investors. The order imbalances are much smaller in absolute valuefor foreign investors and institutional investors. Finally, there are no signi"cantdi!erences between order imbalances for positive or negative lead marketreturns for either domestic investors or foreign investors.

The results for the Korean crisis period are quite di!erent for foreign inves-tors. Irrespective of the market return, foreign investors are net sellers of sharesfollowing days with positive market returns. This is not surprising in light ofFig. 2. What is surprising, though, is that foreign investors sell more when themarket is doing well than when it is doing poorly. Foreign investors havea signi"cantly higher sell-order price-setting imbalance if the market was up theprevious day than if it was down and they have a signi"cantly higher sell-orderimbalance if the market is up the same day than if it is down. Table 5 providesevidence that during the crisis Korean individual investors acted like positivefeedback traders while foreign investors did not. Since domestic individualinvestors have large price-setting net buy order imbalances when the market wasup the previous day or is up the same day, foreign investors are selling the mostwhen domestic demand is the highest. This cannot be viewed as destabilizingbehavior.

We also investigate the relation between the sign of the market return andtotal order imbalances but do not report the results. The results for foreigninvestors similarly support the positive feedback trading hypothesis. The keydi!erence in the results has to do with Korean individual investors. The totalorder imbalances of Korean individual investors are systematically positiveboth before and during the crisis, thus absorbing the negative order imbalancesof Korean institutions and foreign investors.

Western business hours are over when the Korean stock market is open. Thisraises the question of whether returns on Korean shares observed in New Yorka!ect the trading of foreign investors in Korea. We "nd that an equally weighted


Tab

le5

Price

-set

ting

order

imbal

ance

s(]

100)

and

mar

ket

retu

rns

atth

eK

SE

from

Dec

.2,1

996

toD

ec.27

,19

97.

Theprice

-set

ting

ord

erim

bal

ance

forea

choft

he41

4st

ock

son

day

tis

com

put

edas

dai

lyprice

-set

ting

buy

volu

mele

sspr

ice-

settin

gse

llvo

lum

eat

trib

ute

dto

each

type

ofin

vest

or(in

stitut

ions,

indi

vidual

s,an

dfo

reig

ner

s),a

ndth

ennor

mal

ized

by

the

stock's

aver

age

dai

lyprice

-set

ting

volu

me

ove

rth

epe

riod

from

Dec

.2,1

996

toD

ec.2

7,19

97.T

heta

ble

show

sm

eans

ofth

eda

ily

norm

aliz

edord

erim

bala

nce

(]10

0)on

day

sof

(lagg

ed,c

urre

nt,

and

lead

)mar

ket

incr

ease

san

dde

crea

ses,

sepa

rate

ly,w

ith

the

KO

SP

Iin

dex

asth

em

arket

.A

t-te

stfo

rth

em

ean

di!

eren

ceis

repor

ted

inpa

rent

hes

es.

Mar

ket

retu

rns

(1)

Inst

itutions

ord

erim

bal

ance

(2)

Indiv

idua

lsord

erim

bal

ance

(3)

For

eign

ers

ord

erim

bal

ance

Tes

tof

(3)!

(1)

Tes

tof

(3)!

(2)

Bef

ore

the

Kor

ean

Cri

sis

(Dec

.2,

1996}

Sept

.30

,19

97)

R.t~

1'0

(108

day

s;43

,278

stock

-day

s)!

0.55

90.

471

0.19

0(3

.60)

(!1.

25)

(0

(134

day

s;52

,970

stock

-day

s)!

0.84

1!

0.35

9!

0.14

3(3

.40)

(1.0

8)t-st

atistics

for

mea

ndi!

eren

ce(1

.12)

(3.1

8)(2

.23)

R.t'

0(1

09day

s;43

,652

stoc

k-day

s)!

0.16

25.

004

0.22

1(1

.51)

(!20

.05)

(0

(133

day

s;52

,596

stoc

k-day

s)!

1.17

3!

4.12

7!

0.17

1(5

.96)

(21.

19)

t-st

atistics

for

mea

ndi!

eren

ce(3

.84)

(34.

74)

(2.6

0)R

.t`

1'0

(109

day

s;43

,040

stock

-day

s)!

0.58

0!

0.18

10.

064

(2.5

6)(1

.10)

(0

(133

day

s;53

,208

stock

-day

s)!

0.82

30.

172

!0.

039

(4.5

9)(!

1.05

)t-st

atistics

for

mea

ndi!

eren

ce(0

.91)

(!1.

35)

(0.6

9)

Dur

ing

the

Kor

ean

Cri

sis

(Oct

.1,

1997}D

ec.

27,

1997

)

R.t~

1'0

(29

days

;11

,022

stock

-day

s)!

7.81

05.

561

!2.

955

(7.9

6)(!

8.92

)(

0(4

3da

ys;15

,846

stock

-day

s)!

2.18

8!

0.38

1!

1.60

6(1

.69)

(!2.

30)

t-st

atistics

for

mea

ndi!

eren

ce(!

10.5

7)(5

.98)

(!2.

96)

R.t'

0(2

9da

ys;10

,923

stoc

k-da

ys)

!8.

005

13.5

50!

3.13

0(8

.62)

(!19

.07)

(0

(43

days

;15

,945

stoc

k-da

ys)

!2.

089

!5.

817

!1.

494

(1.5

7)(7

.32)

t-st

atistics

for

mea

ndi!

eren

ce(!

10.8

0)(1

9.87

)(!

4.05

)R

.t`

1'0

(30

days

;11

,133

stock

-day

s)!

6.51

45.

651

!2.

104

(7.2

0)(!

8.81

)(

0(4

2da

ys;15

,735

stock

-day

s)!

3.06

5!

0.48

7!

2.19

9(2

.55)

(!2.

90)

t-st

atistics

for

mea

ndi!

eren

ce(!

6.33

)(6

.37)

(0.2

2)


portfolio of the returns of Korean ADRs for the NYSE trading day has noinformation for trading by foreign investors in Korea. In contrast, however, thereturn on the Korea Fund seems to a!ect trading by foreign investors in Korea.A positive (negative) open-to-close return on the Korea Fund is associated witha positive (negative) foreign order imbalance in Korea the next day. Thedi!erence in foreign order imbalances following a positive or negative open-to-close return on the Korea Fund is signi"cant at the 10% level.

The fact that Western business hours are over when the Korean stock marketopens creates the concern that the opening batch auction may be important forforeign investors. Results that use price-setting trades do not include thatauction. In our sample of all trades by foreign investors, less than 5% of thetrades by these investors take place during that auction. We neverthelessinvestigate the trades during the opening batch auction. Though we do notreport these results, they are consistent with the results we have discussed. Themean order imbalance of foreign investors across stocks is positively related tothe previous trading day market return, but not during the crisis period.Interestingly, this mean order imbalance is not related to the overnight marketreturn. In contrast, the overnight market return has a positive e!ect on theopening trades of Korean individuals and a negative e!ect on the opening tradesof Korean institutions.

Table 5 provides evidence on positive feedback strategies that are conditionedon the sign of the market return. The rationale for this dichotomous approach isthat it is unlikely that foreigners are going to sell twice as much if the marketfalls by 6% instead of 3%. With this approach, we do not have to take a positionon the precise relation between trades and the size of the market return. We alsoinvestigate a model in which trading depends linearly on the level of the marketreturn. In results not reported, we estimate a pooled regression of daily price-setting order imbalances of individual stocks (normalized by daily averageprice-setting volume of the stock over our sample period) of the three groups oftraders on the market return of the previous day and the current day. We useinteractive dummy variables to evaluate di!erences in the impact of marketreturns on the trading of the three groups of investors. There is clear evidence ofpositive feedback trading by foreign investors in these regressions for thepre-crisis period, but it disappears for the crisis period. If we use the pooledregressions to estimate the impact of the daytime returns of the Korea Fund onthe NYSE on trading by foreign investors in Korea, there is a positive insigni"c-ant e!ect both before and during the crisis. This suggests that the previoustrading-day return on the Korean stock market has more in#uence on tradingby foreign investors than the previous trading-day return of the Korea Fund onthe NYSE for the period before the crisis.

In Table 6, we investigate the extent of positive feedback trading usingindividual stock returns rather than market returns. In Tables 6 and 7, we usethe total order imbalances to include the opening and closing sessions. The


Tab

le6

Ord

erim

bal

ance

s(]

100)

and

indi

vidu

alst

ock

s'la

gged

exce

ssre

turn

sov

erm

arke

tre

turn

sat

the

KSE

from

Dec

.2,

1996

toD

ec.27

,19

97.

The

orde

rim

bala

nce

forea

choft

he

414

stock

son

day

tis

com

pute

das

dai

lybu

yvo

lum

ele

ssse

llvo

lum

eat

trib

ute

dto

each

type

ofin

vest

or(in

stitutions,

indiv

idua

ls,a

ndfo

reig

ners

),an

dth

enno

rmal

ized

by

the

stoc

k'sav

erag

edai

lyvo

lum

eove

rth

eper

iod

from

Dec

.2,1

996

toD

ec.2

7,19

97.T

heta

ble

show

sm

eansofth

eda

ily

norm

aliz

edor

der

imba

lanc

es(]

100)

for

qui

ntile

por

tfol

ios

ofst

ock

s,P1(

low

est)

thro

ugh

P5(

hig

hes

t),f

orm

edbas

edon

lagg

edex

cess

retu

rnsov

erm

arke

tre

turn

s(K

OSPIin

dex

),w

hic

har

ere

bala

nced

daily

.The

t-st

atistics

ofth

em

eans

are

pres

ente

din

par

enth

eses

,and

t-te

stsfo

rth

em

ean

di!

eren

ceac

ross

di!er

entin

vest

or

type

sar

eal

sore

port

edin

par

enth

eses

.

Prior-

day

retu

rnpor

tfol

ios

Num

ber

ofda

ys(1

)In

stitutions

ord

erim

bal

ance

(2)In

divi

dua

lsord

erim

bal

ance

(3)Fore

igne

rsord

erim

bal

ance

Tes

tof

(3)!

(1)

Tes

tof

(3)!

(2)

Bef

ore

the

Kor

ean

cris

is(D

ec.

2,19

96}Se

pt.

30,

1997

)

Rit~

1!R

.t~

1P1

(Low

est)

241

!1.

230

(!7.

08)

2.16

7(1

0.97

)!

1.00

6(!

8.42

)(1

.06)

(!13

.75)

P2

241

!0.

469

(!3.

25)

0.62

8(3

.72)

!0.

258

(!2.

49)

(1.1

9)(!

4.47

)P3

241

!0.

036

(!0.

22)

0.00

9(0

.05)

!0.

075

(!0.

58)

(!0.

18)

(!0.

39)

P4

241

0.05

0(0

.29)

!0.

464

(!2.

83)

0.42

1(3

.12)

(1.7

0)(4

.16)

P5

(Hig

hest

)24

1!

0.41

2(!

1.88

)!

0.18

1(!

0.71

)0.

394

(2.3

1)(2

.91)

(1.8

8)P5!

P1

0.81

8(2

.93)

!2.

348

(!7.

29)

1.40

0(6

.72)

Dur

ing

the

Kor

ean

cris

is(O

ct.

1,19

97}

Dec

.27

,19

97)

Rit~

1!R

mt~

1P1

(Low

est)

72!

3.79

6(!

4.06

)5.

982

(5.4

4)!

2.58

7(!

6.25

)(1

.18)

(!7.

29)

P2

72!

2.38

8(!

2.48

)4.

650

(4.1

9)!

2.64

4(!

6.49

)(!

0.24

)(!

6.17

)P3

72!

2.75

9(!

2.86

)4.

570

(4.5

6)!

2.08

8(!

5.92

)(0

.65)

(!6.

27)

P4

72!

1.13

5(!

0.84

)2.

858

(2.0

8)!

1.97

3(!

6.93

)(!

0.61

)(!

3.44

)P5

(Hig

hest

)72

!2.

709

(!3.

66)

3.40

2(3

.83)

!1.

191

(!2.

63)

(1.7

5)(!

4.61

)P5-

P1

1.08

7(0

.91)

!2.

580

(!1.

82)

1.39

6(2

.28)


Tab

le7

Cro

ss-s

ection

alav

erag

esof

regr

essions

oford

erim

bal

ance

s(]

100)

on

indi

vidu

alst

ock

retu

rns,

mar

ket

retu

rns,

and

FX

retu

rnsat

the

KSE

from

Dec

.2,1

996

toD

ec.27

,199

7.

The

ord

erim

bala

nce

for

each

ofth

e41

4st

ocks

on

day

tis

com

put

edas

dai

lybu

yvo

lum

ele

ssse

llvo

lum

eat

trib

uted

toea

chty

pe

ofin

vest

or

(inst

itutions,

indiv

idua

ls,a

ndfo

reig

ners

),an

dth

enno

rmal

ized

by

the

stoc

k's

aver

age

dai

lyvo

lum

eov

erth

epe

riod

from

Dec

.2,1

996

toD

ec.2

7,19

97.T

heta

ble

show

sth

ecr

oss

-sec

tion

alav

erag

esofre

gres

sion

sof

the

norm

aliz

edord

erim

bal

ance

s(]

100)

ofea

chin

vest

or

type

for

each

stoc

kon

the

stoc

k's

retu

rns,

mar

ket

retu

rns

(KO

SPI

inde

x),an

dW

on/U

SD

FX

retu

rns.

Eac

hst

ock

retu

rnis

calc

ulat

edas

Rit"

ln(P

it/Pit~

1)]10

0,an

dth

eFX

retu

rns

iss t"

ln(S

t/St~

1)]10

0w

her

eSt"

(Won

/USD

) 5.The

tabl

epre

sents

cros

s-se

ctio

nal

mea

ns

ofth

ees

tim

ated

para

met

ers

acro

sssa

mple

stoc

ks.T

he

t-st

atistics

ofth

ecr

oss

-sec

tion

alm

eans

are

pres

ente

din

pare

nthes

es,a

nd

t-te

sts

for

the

mea

ndi!

eren

cein

the

estim

ated

par

amet

ers

acro

ssdi!

eren

tin

vest

orty

pes

are

also

repor

ted

inpa

rent

hes

es.

Dep

ende

ntva

riab

le:

Norm

aliz

edord

erim

bal

ance

Cro

ss-s

ection

alm

eans

ofslop

eco

e$ci

ents

(t-st

atistics

)fo

rA

vera

geA

dj.R

2

Num

berof

stoc

ks

Con

stan

tR

it~1

Rit

Rit`

1R

.5~

1R

.5

R.5`

1s t~

1s t

s t`1

Bef

ore

the

Kor

ean

cris

is(D

ec.

2,19

96}Se

pt.

30,

1997

)

(1)In

stitution's

order

imbal

ance

!0.

383

0.07

0!

0.32

4!

0.15

2!

0.15

5!

0.04

30.

328

0.42

40.

294

!0.

743

0.02

041

4(!

3.26

)(1

.93)

(!7.

09)

(!4.

38)

(!2.

53)

(!0.

71)

(5.5

8)(1

.38)

(1.1

5)(!

2.26

)

(2)In

divi

dua

l'sor

der

imbal

ance

0.39

4!

0.26

30.

055

0.26

40.

172

0.48

6!

0.50

1!

0.44

6!

0.44

00.

412

0.04

041

4(2

.66)

(!6.

04)

(0.9

3)(7

.46)

(2.4

6)(6

.15)

(!7.

98)

(!1.

22)

(!1.

59)

(1.1

5)

(3)Fore

igne

r's

ord

erim

bal

ance

!0.

083

0.19

90.

337

!0.

109

!0.

026

!0.

472

0.19

60.

136

!0.

084

0.12

20.

025

414

(!0.

72)

(6.7

0)(7

.87)

(!4.

25)

(!0.

44)

(!7.

14)

(3.7

8)(0

.45)

(!0.

39)

(0.5

5)

t-st

atistics

for

mea

ndi!

eren

ce:(

3)!

(1)

(1.8

3)(2

.77)

(10.

55)

(0.9

9)(1

.51)

(!4.

77)

(!1.

69)

(!0.

67)

(!1.

13)

(2.1

8)

t-st

atistics

for

mea

ndi!

eren

ce:(

3)!

(2)

(!2.

54)

(8.7

6)(3

.84)

(!8.

53)

(!2.

16)

(!9.

30)

(8.5

6)(1

.23)

(1.0

2)(!

0.69

)

t-st

atistics

for

mea

ndi!

eren

ce:(

2)!

(1)

(4.1

1)(!

5.89

)(5

.06)

(8.3

9)(3

.52)

(5.3

0)(!

9.64

)(!

1.83

)(!

1.95

)(2

.38)


Dur

ing

the

Kor

ean

cris

is(O

ct.

1,19

97}D

ec.

27,

1997

)

(1)In

stitution's

ord

erim

bala

nce

!2.

747

0.28

7!

0.10

0!

0.14

2!

0.92

7!

0.54

40.

035

!0.

196

!0.

167

!0.

044

0.10

941

3(!

6.52

)(3

.17)

(!0.

72)

(!1.

54)

(!7.

78)

(!3.

62)

(0.3

6)(!

2.38

)(!

1.64

)(!

0.77

)

(2)In

divi

dual's

ord

erim

bala

nce

4.80

8!

0.42

9!

0.12

10.

215

1.01

01.

090

!0.

108

0.22

7!

0.00

30.

018

0.11

641

3(9

.53)

(!4.

80)

(!0.

89)

(2.2

8)(8

.49)

(7.0

1)(!

1.04

)(1

.96)

(!0.

03)

(0.3

0)

(3)Fore

igner's

ord

erim

bala

nce

!2.

262

0.15

90.

235

!0.

077

!0.

116

!0.

511

0.10

1!

0.06

00.

066

0.02

60.

045

413

(!8.

57)

(2.6

7)(4

.05)

(!1.

98)

(!2.

11)

(!7.

27)

(1.7

9)(!

0.79

)(1

.70)

(0.5

6)

t-st

atistics

for

mea

ndi!er

ence

:(3

)!(1

)(0

.97)

(!1.

18)

(2.2

2)(0

.64)

(6.1

7)(0

.19)

(0.5

8)(1

.22)

(2.1

4)(0

.95)

t-st

atistics

for

mea

ndi!er

ence

:(3

)!(2

)(!

12.4

2)(5

.47)

(2.4

0)(!

2.87

)(!

8.59

)(!

9.38

)(1

.77)

(!2.

08)

(0.6

3)(0

.11)

t-st

atistics

for

mea

ndi!er

ence

:(2

)!(1

)(1

1.50

)(!

5.63

)(!

0.11

)(2

.71)

(11.

50)

(7.5

6)(!

1.00

)(2

.98)

(1.1

3)(0

.75)


results are similar if we use only price-setting imbalances. We proceed as follows.On a given day, we compute each stock's return in excess of the market. We rankthese excess returns and form "ve portfolios each day. For each portfolio, wethen compute the average of the normalized order imbalance of the stocks in theportfolio for the following day. The table shows that for foreign investors beforethe crisis the order imbalance increased monotonically with the stock returns onthe previous day. The di!erence in order imbalance between the two extremeportfolios is highly signi"cant. This evidence strongly supports the positivefeedback trading hypothesis. In general, the order imbalances fall for individualsand increase for institutions as one moves towards the portfolio with the highestreturn. This is consistent with individuals being contrarians and institutionsbeing positive feedback traders. However, institutions have sell-order imbalan-ces for all portfolios except the fourth portfolio. When we turn to the results forthe crisis period, the order imbalances generally increase for foreign investors asone moves from portfolio 1 to portfolio 5, but foreign investors are net sellers forall portfolios. Institutions are also net sellers, but the absolute value of theirorder imbalances "rst falls and then increases. Domestic individuals are netbuyers, but they buy more of the stocks that have performed poorly. Except forthe best-performing portfolio, the order imbalances of foreign investors andinstitutional investors are not statistically di!erent both before and during thecrisis. In contrast, the order imbalances of foreign investors are signi"cantlydi!erent from the order imbalances of Korean individual investors for allportfolios but the middle one before the crisis and for all portfolios during thecrisis.

We have now seen results relating order imbalances to the performance of themarket and to the performance of individual stocks. We try to assess the relativeimportance of the market return and individual stock returns by estimatingregressions of individual common stock order imbalances on measures of themarket return, the stock return, and the foreign exchange return. In all ourregressions, we include lagged, contemporaneous, and leading values ofour independent variables. The average regression coe$cients are reported inTable 7. Before the crisis, the average coe$cient of the lagged individual stockreturn for foreign order imbalances is positive and signi"cant, as one wouldexpect if foreign investors are positive feedback traders. The coe$cient of thelagged market return is negative and insigni"cant. This suggests that positivefeedback trading is driven by individual stock returns rather than the marketreturn. The foreign exchange returns do not seem to a!ect trading by foreigninvestors. Korean individual investors are again contrarian with respect toindividual stock returns, but seem to be positive feedback traders with respect tothe market return. Korean institutions are positive feedback traders for indi-vidual stocks and contrarian with respect to the market. During the crisis,foreign investors are still positive feedback traders at the individual stock level,but now they are signi"cantly contrarian at the market level. As a result,


7Bailey et al. (1998) "nd stronger evidence of a positive relation between currency depreciationand sales of equities using high-frequency data for Mexican ADRs and closed-end funds trading onthe NYSE. They "nd that Peso depreciation is accompanied by an increase in transactions at the bidrelative to transactions at the ask for these securities over half-hour intervals from December 21,1994 to April 30, 1995.

a stock's negative return has no impact on a stock's order imbalance if themarket has a similar negative return. The lagged foreign exchange return doesnot matter, but there is weak evidence that the contemporaneous foreignexchange return has a positive e!ect, so that a depreciation is correlated withselling by foreign investors.7 For the crisis period, Korean individual investorsare also contrarians with respect to individual stock returns, but positivefeedback traders with respect to the market return. Again, Korean institutionsbehave in the opposite way and more like foreign investors. We also estimate theregressions of Table 7 with dummy variables for stocks a!ected by price limits atthe end of the previous trading day, but our inferences do not change.

One concern with our tests of positive feedback trading is that we focus onnormalized trades. The normalization by average price-setting volume that weuse is helpful in understanding the importance of positive feedback tradingrelative to trading volume. We investigated whether the results that we reporthold if we use dependent variables that are not normalized. In particular, werepeated all our tests using the fraction of the sum of the purchases and salesattributed to each type of investor (institutions, individuals, and foreigners) asthe dependent variable. All our results hold up using this approach. They alsohold if we use the price-setting purchases and sales. One might also be concernedthat we condition on Won returns in our tests. Table 7 suggests that exchangerate returns are unlikely to a!ect our results. To make sure, we have alsochecked the results of Table 5 conditioning on dollar market returns, and wereach the same conclusions. Finally, all our tests use daily data. Tests thatexamine the existence of positive feedback trading over longer intervals wouldbe useful, but we leave those for further research. Such tests would not enable usto study the crisis period separately as we did with daily data because therewould be too few observations.

5. Do foreign investors have a destabilizing in6uence?

We have now seen that there is strong evidence of positive feedback tradingand herding on the part of foreign investors, especially before the Korean crisis.The last question we want to address is whether these investors have a de-stabilizing in#uence. With the data available, we can address this issue bylooking at whether large trading imbalances by foreign investors are followed byprice continuations and by an increase in volatility. For that purpose, we


conduct two event studies. The "rst study uses intraday returns, and the seconduses daily returns. Our approach complements the studies that focus on theimpact of capital market liberalization on stock market volatility. These studiesgenerally "nd that opening a stock market to foreign investors does not increaseits volatility [see Stulz (1999) for a review and Bekaert and Harvey (2000) formore recent evidence], but they do not directly consider whether trades byforeign investors can have a destabilizing impact. As discussed earlier, there aretax advantages for Korean investors trading out of Malaysia and Ireland. Inresults not reported here, we perform all the tests of this section excluding tradesfrom Malaysia and Ireland. Our results are unchanged. We also perform theevent studies during the crisis period for the 48 stocks that are at the ownershiplimit at the start of our sample period and "nd that trades by foreign investorshave less impact on these stocks than on our sample stocks.

5.1. Intraday event study

We divide each day into 46 "ve-minute intervals from 9:30 to 15:00, treatingthe time interval of 11:30}13:05 as a single interval containing the lunch breakand similarly the time interval of 14:45}15:00, which contains an order collectionperiod for the close. We exclude Saturdays since we have much less freedom inchoosing an event that has a su$cient number of intraday intervals prior to andafter an event. For each of the intervals for each of the 414 stocks over thesample period, we compute foreign order imbalances by subtracting foreign sellvolume from foreign buy volume during the interval. We then select the "veintervals for net buy (positive) imbalances and net sell (negative) imbalanceswith the largest foreign order imbalances in absolute value for each of the 414stocks. For each of the selected events, we examine stock returns from theprevious "fth (!5) to the subsequent "fth (#5) interval surrounding the event.To avoid crossing day boundaries when we examine !5 to #5 intervals, theevents are selected from the seventh interval (10:00}10:05) through the 41stinterval (14:20}14:25), excluding the 25th interval (11:30}13:05), which containsa batch auction period. Among the events selected above, we exclude those withforeign order imbalances less than 1,000 shares in absolute value. The abovesampling procedure is also applied to the case of the foreign price-settingvolume.

Table 8 describes the samples constructed with our procedure. We report theresults for the whole sample, the subperiod before the crisis, and the crisisperiod. Though we study the returns around the events using all the largestimbalances as well as the largest price-setting imbalances, we report only theresults for the largest price-setting imbalances. These are trades initiated byforeign investors, so we expect the impact of their imbalances to be the largest.For each class of events, we report raw returns, mean-adjusted returns, and theabsolute value of the mean-adjusted returns as a measure of volatility. We


Table 8Descriptive statistics of large foreign order imbalances (in shares) in "ve-minute intervals.

For each of the 414 stocks at the KSE from Dec. 2, 1996 to Dec. 27, 1997, the foreign orderimbalance (buy volume less sell volume) within a "ve-minute trading interval during continuousauction sessions (excluding Saturdays) is computed in two ways: one is based on all foreign tradesand the other based on foreign price-setting trades. Among these, "ve intervals with the largest orderimbalances in terms of net buy, net sell, price-setting net buy, and price-setting net sell are selectedfor each stock, but excluding those with less than 1,000 shares. The results are presented below forthe full sample period and two subperiods before and during the Korean crisis.

Descriptivestatistics

Net buy order imbalances Net sell order imbalances

Fullsample

Beforecrisis

Duringcrisis

Fullsample

Beforecrisis

Duringcrisis

Nobs 1,970 1,685 285 2,009 1,444 565Mean 19,860 20,132 18,255 20,964 16,255 33,000Maximum 2,000,000 2,000,000 292,760 1,387,500 315,840 1,387,500Q3 15,100 15,000 17,020 19,790 15,000 30,000Median 6,180 6,140 6,810 8,160 6,990 12,000Q1 3,000 3,000 3,890 4,000 3,685 5,000Minimum 1,000 1,000 1,000 1,000 1,000 1,000


Price-setting net buy order imbalances Price-setting net sell order imbalances

Fullsample

Beforecrisis

Duringcrisis

Fullsample

Beforecrisis

Duringcrisis

Nobs 1,826 1,595 231 1,915 1,386 529Mean 17,849 17,103 22,998 16,602 14,070 23,235Maximum 2,000,000 2,000,000 463,000 1,387,500 316,400 1,387,500Q3 13,070 12,250 16,970 14,230 11,140 20,000Median 5,490 5,330 6,140 5,590 5,000 9,000Q1 2,700 2,610 3,000 3,000 2,900 3,820Minimum 1,000 1,000 1,000 1,000 1,000 1,000

obtain mean-adjusted returns by subtracting the sample mean return for thestock on the same day of the week and same time of day over the whole sampleperiod as the event return to control for the well-known day-of-the-week e!ectas well as the time-of-the-day e!ect.

Panel A of Table 9 presents the returns for the "ve-minute intervals forthe "ve intervals preceding the event and the "ve intervals after it. It isimmediately apparent that the largest price-setting net buy order imbalancesoccur following positive returns for the stock, so that foreigners buy follow-ing price increases. The price increase continues for one period after the pur-chase by foreign investors, but then the returns are insigni"cantly di!erent


Tab

le9

Intr

aday

retu

rns

and

vola

tility

(%)ar

ound"ve

-min

ute

inte

rval

sof

larg

efo

reig

nprice

-set

ting

ord

erim

bala

nce

s.

The

fore

ign

pric

e-se

ttin

gor

der

imbal

ance

withi

na"ve

-min

ute

trad

ing

inte

rval

dur

ing

continuous

auct

ion

sess

ions

for

each

ofth

e41

4st

ock

sat

the

KSE

from

Dec

.2,1

996

toD

ec.2

7,19

97,e

xclu

ding

Satu

rday

s,is

com

pute

das

fore

ign

price

-set

ting

buy

volu

me

less

price

-set

ting

sell

volu

me.

The

sam

ple

sin

Pan

elsA

and

Bco

mprise

the"ve

inte

rval

sw

ith

the

larg

estprice

-set

ting

netbu

yan

dnet

sell

orde

rim

bal

ance

s,re

spec

tive

ly,f

orea

chst

ock,b

ute

xclu

ding

thos

eofl

essth

an1,

000

shar

es.T

heM

ean-

adjr

etfo

rea

chin

terv

alis

them

ean

ofth

ein

terv

alre

turn

forth

est

ock

exce

edin

gth

em

ean

obse

rved

on

thesa

meda

yoft

hew

eek

and

sam

etim

eofda

yove

rth

esa

mple

per

iod,an

dth

eDM

ean-a

djre

tDis

am

easu

reof

vola

tilit

yco

mput

edas

the

mea

nofab

solu

teva

lues

ofth

em

ean-a

dju

sted

retu

rns.

The

CA

R(0

,5)is

the

cum

ula

tive

retu

rns

from

inte

rval

s0

thro

ugh

5,an

dth

et-st

atistics

are

report

edin

par

enth

eses

.

Mea

nst

atistics

5-m

inute

inte

rval

sre

lative

toth

efo

reig

nnet

buy

ord

erim

bala

nce

(inte

rval

0)C

AR

(0,5

)

!5

!4

!3

!2

!1

01

23

45

Pan

elA

.F

ive-

min

ute

retu

rns

and

vola

tili

ty(%

)ar

ound

larg

efo

reig

npr

ice-

sett

ing

net

buy

orde

rim

bala

nces

Ful

lpe

riod

(Dec

.2,

1996

toD

ec.

27,

1997

;N"

1,82

6)

Raw

ret

0.05

90.

065

0.02

10.

072

0.15

50.

649

0.10

90.

004

!0.

005

!0.

015

!0.

015

0.72

8(3

.19)

(3.4

2)(1

.01)

(3.1

3)(6

.22)

(17.

60)

(5.2

3)(0

.15)

(!0.

21)

(!0.

62)

(!0.

66)

(14.

91)

Mea

n-ad

jre

t0.

057

0.06

60.

024

0.07

50.

157

0.64

80.

107

0.00

7!

0.00

3!

0.01

4!

0.01

20.

733

(3.1

4)(3

.51)

(1.1

5)(3

.31)

(6.3

9)(1

7.75

)(5

.20)

(0.3

1)(!

0.12

)(!

0.59

)(!

0.57

)(1

5.18

)

DMea

n-ad

jre

tD0.

401

0.39

60.

427

0.46

40.

516

0.95

90.

483

0.50

50.

449

0.48

10.

429

Bef

ore

the

Kor

ean

cris

is(D

ec.

2,19

96to

Sept

.30

,19

97;

N"

1,59

5)

Raw

ret

0.06

30.

058

0.04

80.

085

0.17

40.

574

0.12

30.

025

0.00

5!

0.01

4!

0.03

50.

679

(3.4

6)(3

.34)

(2.6

1)(4

.34)

(8.0

5)(1

7.54

)(5

.79)

(1.0

9)(0

.24)

(!0.

64)

(!1.

88)

(14.

80)

Mea

n-ad

jre

t0.

062

0.05

90.

052

0.08

80.

176

0.57

50.

122

0.02

90.

007

!0.

012

!0.

033

0.68

7(3

.39)

(3.4

1)(2

.80)

(4.5

2)(8

.26)

(17.

74)

(5.7

9)(1

.24)

(0.3

3)(!

0.58

)(!

1.79

)(1

5.14

)

DMea

n-ad

jre

tD0.

391

0.36

80.

390

0.42

90.

476

0.87

60.

474

0.47

80.

434

0.44

90.

389

Dur

ing

the

Kor

ean

cris

is(O

ct.

1,19

97to

Dec

.27

,19

97;

N"

231)

Raw

ret

0.02

60.

114

!0.

167

!0.

021

0.02

51.

166

0.01

2!

0.14

7!

0.07

0!

0.02

20.

125

1.06

4(0

.36)

(1.2

6)(!

1.60

)(!

0.17

)(0

.20)

(6.4

6)(0

.16)

(!1.

55)

(!0.

87)

(!0.

18)

(1.0

7)(4

.86)

Mea

n-ad

jre

t0.

026

0.11

8!

0.16

7!

0.01

20.

027

1.15

30.

004

!0.

141

!0.

067

!0.

025

0.12

91.

052

(0.3

7)(1

.32)

(!1.

61)

(!0.

10)

(0.2

1)(6

.46)

(0.0

5)(!

1.49

)(!

0.85

)(!

0.21

)(1

.11)

(4.8

5)

DMea

n-ad

jre

tD0.

470

0.58

90.

680

0.70

00.

795

1.53

30.

545

0.69

20.

552

0.70

40.

705

Pan

elB

.F

ive-

min

ute

retu

rns

and

vola

tili

ty(%

)ar

ound

larg

efo

reig

npr

ice-

sett

ing

net

sell

orde

rim

bala

nces

Ful

lpe

riod

(Dec

.2,

1996

toD

ec.

27,

1997

;N"

1,91

5)

Raw

ret

0.04

3!

0.00

80.

005

!0.

021

0.06

1!

0.63

90.

203

0.10

30.

053

0.04

90.

013

!0.

218

(1.7

9)(!

0.31

)(0

.25)

(!0.

91)

(2.6

1)(!

17.4

3)(6

.94)

(4.4

7)(2

.51)

(2.3

7)(0

.58)

(!4.

86)

Mea

n-a

djre

t0.

042

!0.

003

0.01

0!

0.01

50.

065

!0.

630

0.20

30.

103

0.05

30.

051

0.01

6!

0.20

4(1

.76)

(!0.

14)

(0.4

8)(!

0.64

)(2

.80)

(!17

.36)

(7.0

2)(4

.51)

(2.5

3)(2

.47)

(0.6

9)(!

4.61

)

DMea

n-ad

jre

tD0.

452

0.47

40.

444

0.47

70.

535

0.95

90.

581

0.48

30.

460

0.45

00.

463

Bef

ore

the

Kor

ean

cris

is(D

ec.

2,19

96to

Sept

.30

,19

97;

N"

1,38

6)

Raw

ret

0.05

30.

020

0.02

50.

015

0.07

8!

0.48

70.

200

0.08

30.

044

0.04

60.

023

!0.

091

(2.7

2)(0

.92)

(1.1

5)(0

.70)

(3.1

9)(!

14.2

3)(7

.52)

(3.5

9)(2

.04)

(2.1

3)(1

.02)

(!1.

98)

Mea

n-ad

jre

t0.

051

0.02

30.

028

0.02

20.

083

!0.

478

0.20

20.

085

0.04

40.

048

0.02

5!

0.07

5(2

.65)

(1.0

7)(1

.30)

(1.0

2)(3

.40)

(!14

.14)

(7.6

3)(3

.67)

(2.0

5)(2

.23)

(1.1

5)(!

1.64

)

DMea

n-ad

jre

tD0.

374

0.41

60.

407

0.42

80.

503

0.82

60.

512

0.45

80.

429

0.42

00.

407

Dur

ing

the

Kor

ean

cris

is(O

ct.

1,19

97to

Dec

.27

,19

97;

N"

529)

Raw

ret

0.01

8!

0.08

0!

0.04

7!

0.11

50.

018

!1.

037

0.20

90.

154

0.07

80.

057

!0.

011

!0.

551

(0.2

6)(!

1.22

)(!

1.00

)(!

1.94

)(0

.31)

(!10

.84)

(2.6

3)(2

.71)

(1.5

0)(1

.17)

(!0.

18)

(!5.

12)

Mea

n-a

djre

t0.

020

!0.

073

!0.

038

!0.

110

0.02

0!

1.02

70.

207

0.15

10.

077

0.05

7!

0.00

9!

0.54

4(0

.28)

(!1.

12)

(!0.

82)

(!1.

90)

(0.3

6)(!

10.8

4)(2

.63)

(2.6

8)(1

.50)

(1.1

9)(!

0.15

)(!

5.13

)

DMea

n-ad

jre

tD0.

658

0.62

40.

538

0.60

40.

618

1.30

70.

761

0.54

90.

541

0.53

00.

610

from zero. There is a large contemporaneous positive return with the event,so the large foreign net buy imbalance is associated with a large stock return.If there is positive information in the foreign net buy imbalance, it gets im-pounded in prices immediately, since "ve minutes later prices seem to haveadjusted. The absolute "ve-minute return is obviously large for the event period,but there is little evidence of a persistent sharp increase in volatility followingthe event.

The patterns for events before the Korean crisis are quite similar to those wehave just discussed for the whole sample period. The volatility "ve periods afterthe event is back to where it was "ve periods before the event. The patternsduring the crisis are di!erent, however. First, there is no longer any evidencethat foreign net buy imbalances follow positive returns. Second, the impact ofthe event is larger. Third, the event seems to have no positive e!ect on sub-sequent returns. In particular, the return for the "rst period following the eventis not signi"cantly positive.

Although we do not report these results in a table, we perform severaladditional investigations to check the robustness of our conclusions. First,we compute returns around the largest net buy imbalances instead of the largestprice-setting net buy imbalances. The only noticeable di!erence is thatthe event returns are substantially smaller. For the whole sample period,the average mean-adjusted return is 0.347%, in contrast to 0.648% inTable 9. This is not surprising, since the sample in Table 9 uses only tradesinitiated by foreign investors. Either way, it is clear that there is a signi"cantpositive return associated with the event. There is no evidence that large tradesby foreign investors are associated with positive signi"cant mean-adjustedreturns beyond the next "ve minutes, so the market adjusts quickly and e$cient-ly to the trades by these investors. Second, we estimate returns splittingthe sample between stocks that hit the price limit during or at the end of thetrading day and those that do not. Price limits do not appear to a!ect ourresults.

Panel B of Table 9 shows the analysis for the largest price-setting net sellevents. One would expect this panel to have results that are symmetric to thosein Panel A. Surprisingly, this is not the case. Looking at the subperiod before thecrisis, the returns are not negative before the event. In fact, the returns in theperiod immediately before the event are signi"cantly positive. There is a largenegative return associated with the event. In absolute value, it is slightly lessthan the return for the event in Panel A. We observe signi"cant positive returnsfor four periods after the event. As a result, the cumulative return associatedwith the event is small in absolute value and insigni"cant for mean-adjustedreturns. Although there is at best a small permanent e!ect of sales by foreigninvestors, there is a temporary e!ect that is consistent with a reward for thosewho provide the liquidity to the market. It seems, therefore, that selling byforeign investors is followed by a reversal that is quite substantial in relation to


the event return. The evidence is similar on days that end with the price limitbeing binding. On these days, the positive returns following the event arenot signi"cant, but the price impact of foreign sales is completely reversed25 minutes after the sale.

The patterns during the crisis are similar to those before the crisis, except thatthe event return is larger in absolute value. As a result, the reversals end upo!setting about half the price impact of the sale. Neither before nor during thecrisis is there evidence that large foreign sales lead to a period of signi"cantnegative mean-adjusted returns. During the crisis, the largest net sell imbalanceshave an average mean-adjusted return of !1.027%, while the largest net buyimbalances have an average mean-adjusted return of 1.153%. There is thereforeno evidence that sales by foreign investors lead to disproportionate eventreturns compared to purchases by foreign investors. Results are similar on dayswith binding price limits.

The evidence on the largest net sales by foreigners is comparable to theevidence for block sales in the U.S. Holthausen et al. (1990) investigate theadjustment of prices to block sales since block sales are initiated by the seller.They consider the 50 largest downtick transactions for 109 "rms selectedrandomly from December 1, 1982 to January 31, 1984. They "nd that the tradeimpact of the block trade is a mean-adjusted return of !1.23% and that it isfollowed by a signi"cant mean-adjusted return of 0.28%. Strikingly, we "ndsimilar stock price impacts for large sales by foreign investors in Korea, since weobserve a mean-adjusted return of !1.027% for the "ve minutes when largesales take place followed by a mean-adjusted return of 0.207% over the next "veminutes.

5.2. Interday event study

The sampling procedure for daily events is similar to the one described earlier.That is, we compute daily foreign order imbalances by subtracting foreign sellvolume from foreign buy volume for the day for each of the 414 stocks(excluding Saturdays). We then select the "ve trading days with the largestabsolute-value order imbalance for each of the 414 stocks. For each of theselected events, we examine stock returns from the previous "fth (!5) tothe subsequent "fth (#5) trading day surrounding the event day. Among theselected events, we exclude those with daily foreign order imbalances less than10,000 shares in absolute value. The above sampling procedure is also applied tothe daily foreign price-setting volume. We ignore batch auction trades for thecase of the price-setting volume, since it is impossible to de"ne price-settingvolume during batch auctions.

In the daily event study, we present the results using the "ve largest imbalan-ces rather than the "ve largest price-setting imbalances for two reasons. First, asdiscussed above, for the price-setting imbalances we cannot capture all the


Table 10Descriptive statistics of daily large foreign order imbalances (in shares)

For each of the 414 stocks at the KSE from Dec. 2, 1996 to Dec. 27, 1997, the daily foreign orderimbalance (buy volume less sell volume) is computed in two ways: one is based on all foreign trades,and the other based on foreign price-setting trades. Among these, "ve days with the largest orderimbalances in terms of net buy, net sell, price-setting net buy, and price-setting net sell are selectedfor each stock, but excluding those with less than 10,000 shares. The results are presented below forthe full sample period and two subperiods before and during the Korean crisis


Net buy order imbalances Net sell order imbalances

Fullsample

Beforecrisis

Duringcrisis

Fullsample

Beforecrisis

Duringcrisis

Nobs 1,397 1,201 196 1,588 1,079 509Mean 87,572 86,552 93,822 89,061 66,178 137,569Maximum 5,832,980 5,832,980 989,380 2,773,950 2,289,390 2,773,950Q3 59,000 54,380 109,415 71,920 56,160 120,280Median 27,000 26,250 31,315 32,835 29,500 50,500Q1 15,000 15,000 16,285 19,225 17,140 23,940Minimum 10,000 10,000 10,000 10,000 10,000 10,000


Price-setting net buy order imbalances Price-setting net sell order imbalances

Fullsample

Beforecrisis

Duringcrisis

Fullsample

Beforecrisis

Duringcrisis

Nobs 952 816 136 1,113 749 364Mean 59,306 56,398 76,751 62,526 47,473 93,500Maximum 2,016,630 2,016,630 649,500 2,230,130 1,038,260 2,230,130Q3 50,000 48,905 73,500 55,400 43,500 85,610Median 23,770 22,860 31,750 25,470 21,930 35,580Q1 14,380 14,180 15,890 15,220 14,660 17,195Minimum 10,000 10,000 10,000 10,000 10,000 10,000

imbalance in a day because we cannot include the batch auction in our samplingprocedure. Second, foreign imbalances over a day are publicized in Korea, sothat information as to whether foreign investors sold or bought is generallyavailable by the next day. This information concerns foreign imbalances asa whole rather than foreign price-setting imbalances. Table 10 summarizes thesamples. All the mean and median imbalances increase after the crisis, but theincrease is more dramatic for the net sell imbalances, where the mean more thandoubles. Table 11 presents our results, which now include market-adjustedreturns.

From Panel A, we can see that before the crisis foreign investors buystocks that have done well over recent days. Both the mean-adjusted and the


Tab

le11

Dai

lyre

turn

san

dvo

latilit

y(%

)ar

ound

day

sof

larg

efo

reig

nord

erim

bala

nces

.

Theda

ilyfo

reig

nord

erim

bala

nce

forea

chof

the41

4st

ocksat

the

KSE

from

Dec

.2,1

996

toD

ec.2

7,19

97,e

xclu

din

gSa

turd

ays,

isco

mpu

ted

asda

ilyfo

reig

nbuy

volu

mele

ssse

llvo

lum

e.T

hesa

mpl

esin

Pan

elsA

and

Bco

mpr

iseth

e"ve

days

with

thela

rges

tne

tbuy

and

net

sell

orde

rim

bal

ance

s,re

spec

tive

ly,s

elec

ted

forea

chst

ock,

excl

udi

ng

those

with

less

than

10,0

00sh

ares

.The

Mea

n-ad

jret

forea

chda

yis

them

ean

oft

hedai

lyre

turn

forth

est

ock

exce

edin

gth

em

ean

obse

rved

on

the

sam

eday

ofth

ew

eek

ove

rth

esa

mpl

epe

riod,a

ndth

eM

arket

-adjre

tis

the

mea

nofth

edai

lyre

turn

exce

edin

gth

eK

OSPI

index

retu

rn.T

heDM

ean-a

djre

tDis

am

easu

reof

vola

tilit

yco

mpu

ted

asth

em

ean

ofa

bsolu

teva

lues

oft

he

mea

n-a

djus

ted

retu

rns.

The

CA

R(0

,1)i

sth

ecu

mul

ativ

ere

turn

sfrom

day

s0

to1,

and

the

t-st

atistics

are

report

edin

par

enth

eses

.

Mea

nst

atistics

Day

sre

lative

toth

efo

reig

nne

tbuy

orde

rim

bal

ance

(day

0)C

AR

(0,1

)

!5

!4

!3

!2

!1

01

23

45

Pan

elA

.D

aily

retu

rns

and

vola

tili

ty(%

)ar

ound

larg

efo

reig

nne

tbu

yor

der

imba

lanc

es

Ful

lpe

riod

(Dec

.2,

1996

toD

ec.

27,

1997

;N"

1,39

7)

Raw

ret

0.22

70.

297

0.22

70.

640

1.01

61.

340

!0.

344

!0.

306

!0.

369

!0.

252

!0.

353

0.99

6(2

.13)

(2.7

6)(2

.17)

(5.8

5)(8

.76)

(12.

11)

(!3.

30)

(!2.

91)

(!3.

46)

(!2.

37)

(!3.

29)

(6.0

9)

Mea

n-ad

jre

t0.

561

0.61

50.

608

1.01

01.

363

1.65

9!

0.02

60.

017

0.00

40.

098

!0.

028

1.63

3(5

.18)

(5.6

6)(5

.71)

(9.1

4)(1

1.56

)(1

4.84

)(!

0.24

)(0

.16)

(0.0

3)(0

.90)

(!0.

26)

(9.8

8)

Mar

ket

-adjre

t0.

302

0.37

70.

238

0.56

90.

924

1.51

2!

0.24

4!

0.30

0!

0.21

5!

0.00

3!

0.17

01.

267

(3.2

9)(4

.04)

(2.6

0)(6

.02)

(9.1

7)(1

5.16

)(!

2.70

)(!

3.29

)(!

2.39

)(!

0.03

)(!

1.85

)(8

.77)

DMea

n-ad

jre

tD3.

062

3.10

63.

037

3.20

43.

535

3.51

03.

001

2.97

13.

016

3.06

13.

061

Bef

ore

the

Kor

ean

cris

is(D

ec.

2,19

96to

Sept

.30

,19

97;

N"

1,20

1)

Raw

ret

0.58

20.

613

0.45

40.

740

1.31

21.

718

!0.

217

!0.

382

!0.

348

!0.

123

!0.

271

1.50

2(5

.93)

(6.1

4)(4

.60)

(7.2

3)(1

2.00

)(1

6.67

)(!

2.21

)(!

3.90

)(!

3.49

)(!

1.22

)(!

2.73

)(1

0.07

)

Mea

n-ad

jre

t0.

925

0.94

20.

851

1.12

11.

661

2.04

60.

107

!0.

056

0.03

60.

237

0.05

72.

154

(9.2

6)(9

.32)

(8.5

1)(1

0.86

)(1

4.91

)(1

9.62

)(1

.08)

(!0.

56)

(0.3

5)(2

.31)

(0.5

7)(1

4.29

)

Mar

ket

-adjre

t0.

486

0.49

80.

388

0.62

61.

108

1.69

6!

0.21

9!

0.34

9!

0.24

5!

0.05

6!

0.21

51.

477

(5.2

7)(5

.33)

(4.3

4)(6

.68)

(11.

19)

(17.

73)

(!2.

38)

(!3.

86)

(!2.

73)

(!0.

62)

(!2.

35)

(10.

74)

DMea

n-ad

jre

tD2.

690

2.71

52.

676

2.80

93.

196

3.23

42.

599

2.53

62.

606

2.67

92.

609

Dur

ing

the

Kor

ean

cris

is(O

ct.

1,19

97to

Dec

.27

,19

97;

N"

196)

Raw

ret

!1.

951

!1.

639

!1.

163

0.02

9!

0.79

2!

0.97

8!

1.12

40.

157

!0.

493

!1.

042

!0.

854

!2.

102

(!4.

51)

(!3.

77)

(!2.

73)

(0.0

6)(!

1.71

)(!

2.23

)(!

2.60

)(0

.35)

(!1.

10)

(!2.

41)

(!1.

84)

(!3.

08)

Mea

n-a

djre

t!

1.67

0!

1.38

8!

0.88

10.

325

!0.

460

!0.

716

!0.

842

0.46

0!

0.19

4!

0.75

9!

0.55

2!

1.55

8(!

3.82

)(!

3.17

)(!

2.04

)(0

.70)

(!0.

98)

(!1.

63)

(!1.

92)

(1.0

1)(!

0.43

)(!

1.74

)(!

1.19

)(!

2.26

)

Mar

ket

-adjre

t!

0.82

9!

0.36

3!

0.68

30.

218

!0.

205

0.38

3!

0.39

70.

001

!0.

028

0.32

40.

108

!0.

014

(!2.

63)

(!1.

08)

(!1.

98)

(0.6

2)(!

0.55

)(0

.98)

(!1.

27)

(0.0

0)(!

0.09

)(0

.89)

(0.3

2)(!

0.02

)

DMea

n-ad

jre

tD5.

342

5.50

25.

249

5.62

75.

617

5.20

55.

460

5.63

85.

529

5.40

25.

827

Pan

elB

.D

aily

retu

rns

and

vola

tili

ty(%

)ar

ound

larg

efo

reig

nne

tse

llor

der

imba

lanc

es

Ful

lpe

riod

(Dec

.2,

1996

toD

ec.

27,

1997

;N"

1,58

8)

Raw

ret

!0.

334

!0.

325

!0.

638

!0.

639

!0.

151

0.49

70.

152

!0.

390

!0.

452

!0.

454

!0.

464

0.64

8(!

3.15

)(!

3.02

)(!

5.69

)(!

5.51

)(!

1.24

)(4

.12)

(1.3

3)(!

3.52

)(!

4.17

)(!

4.14

)(!

4.26

)(3

.66)

Mea

n-a

djre

t!

0.03

80.

009

!0.

242

!0.

290

0.16

10.

824

0.46

7!

0.04

7!

0.05

7!

0.09

1!

0.15

71.

291

(!0.

36)

(0.0

8)(!

2.13

)(!

2.47

)(1

.31)

(6.7

5)(4

.03)

(!0.

42)

(!0.

53)

(!0.

82)

(!1.

42)

(7.1

5)

Mar

ket

-adjre

t!

0.09

0!

0.09

3!

0.28

5!

0.32

3!

0.09

40.

266

0.37

30.

037

!0.

064

!0.

230

!0.

221

0.63

9(!

1.02

)(!

1.03

)(!

3.08

)(!

3.47

)(!

0.87

)(2

.39)

(3.8

6)(0

.40)

(!0.

73)

(!2.

48)

(!2.

45)

(4.0

7)

DMea

n-ad

jre

tD3.

304

3.31

03.

471

3.65

23.

927

4.06

43.

665

3.47

43.

384

3.37

73.

371

Tab

le11

(con

tinu

ed)

Mea

nst

atistics

Day

sre

lative

toth

efo

reig

nne

tbuy

orde

rim

bala

nce

(day

0)C

AR

(0,1

)

!5

!4

!3

!2

!1

01

23

45

Bef

ore

the

Kor

ean

cris

is(D

ec.

2,19

96to

Sept

.30

,19

97;

N"

1,07

9)

Raw

ret

0.38

90.

361

!0.

024

!0.

104

0.42

40.

744

0.31

8!

0.30

5!

0.23

9!

0.28

2!

0.21

21.

062

(3.7

1)(3

.42)

(!0.

22)

(!0.

94)

(3.4

5)(5

.79)

(2.7

3)(!

2.76

)(!

2.38

)(!

2.80

)(!

2.01

)(5

.88)

Mea

n-ad

jre

t0.

703

0.71

20.

403

0.26

50.

752

1.08

60.

648

0.04

20.

175

0.09

90.

110

1.73

3(6

.60)

(6.6

4)(3

.62)

(2.3

7)(6

.08)

(8.3

7)(5

.46)

(0.3

8)(1

.73)

(0.9

7)(1

.03)

(9.4

3)

Mar

ket

-adjre

t0.

293

0.26

10.

008

!0.

054

0.39

50.

657

0.44

6!

0.12

1!

0.13

7!

0.24

8!

0.19

11.

104

(3.0

5)(2

.67)

(0.0

8)(!

0.53

)(3

.36)

(5.4

6)(4

.13)

(!1.

17)

(!1.

50)

(!2.

64)

(!1.

97)

(6.6

1)

DMea

n-ad

jre

tD2.

668

2.68

72.

751

2.76

93.

180

3.53

73.

025

2.80

82.

540

2.50

52.

623

Dur

ing

the

Kor

ean

cris

is(O

ct.

1,19

97to

Dec

.27

,19

97;

N"

509)

Raw

ret

!1.

865

!1.

778

!1.

940

!1.

773

!1.

368

!0.

028

!0.

201

!0.

572

!0.

903

!0.

817

!0.

999

!0.

229

(!8.

10)

(!7.

49)

(!7.

70)

(!6.

59)

(!5.

09)

(!0.

11)

(!0.

79)

(!2.

25)

(!3.

45)

(!3.

08)

(!3.

93)

(!0.

58)

Mea

n-a

djre

t!

1.60

9!

1.48

2!

1.61

1!

1.46

6!

1.09

20.

269

0.08

3!

0.23

5!

0.54

9!

0.49

3!

0.72

20.

352

(!6.

89)

(!6.

18)

(!6.

31)

(!5.

39)

(!4.

03)

(1.0

3)(0

.32)

(!0.

92)

(!2.

08)

(!1.

85)

(!2.

81)

(0.8

7)

Mar

ket

-adjre

t!

0.90

20.

843

!0.

906

!0.

896

!1.

131

!0.

562

0.21

80.

374

0.09

0!

0.19

2!

0.28

3!

0.34

5(!

5.07

)(!

4.55

)(!

4.93

)(!

4.64

)(!

5.00

)(!

2.43

)(1

.11)

(1.9

1)(0

.47)

(!0.

92)

(!1.

47)

(!1.

03)

DMea

n-ad

jre

tD4.

651

4.63

04.

995

5.52

25.

511

5.18

15.

023

4.88

55.

175

5.22

54.

956

market-adjusted returns are signi"cantly positive every day before the event. Onthe day of the event, there is a large positive signi"cant abnormal return. Incontrast to the intraday results, however, there are reversals over the next "vedays when we use raw or market-adjusted returns. During the crisis, foreignersno longer appear to buy following positive signi"cant abnormal returns. Fur-thermore, there is no signi"cant impact on the day of the event. Using market-adjusted returns, there is no reversal following the event. The same results applyto days when the price limit is in force at the end of the day.

As with the intraday results, there is a noticeable lack of symmetry betweennet buy and net sell events. For net sell events, we have the surprising resultbefore the crisis that on net sell days market-adjusted returns are positive. Thisagain suggests that what matters when foreigners sell is that domestic investorsbuy. The positive market-adjusted return is surrounded by days with positivemarket-adjusted returns. There are, however, signi"cant negative market-ad-justed returns on days #4 and #5. When we look at days when the price limitis hit, the results are similar. When we look at the crisis period, we "nd thatforeign investors sell following signi"cant negative market-adjusted returns.There is a negative market-adjusted return on the event day, and it is followedby "ve days without a signi"cant negative market-adjusted return. The market-adjusted return on the day that foreign investors sell and on each of the next "vedays is smaller in absolute value than the market-adjusted return of any of the"ve days that precede the event day. There is no permanent signi"cant negativee!ect following large sales by foreign investors. For the sample of events inwhich the price limit is hit on the event day, the pattern is similar, but themarket-adjusted returns are insigni"cant except for a positive signi"cant mar-ket-adjusted return on day #2. Overall, there is no evidence that large foreignsales directly cause falling stock prices.

6. Conclusion

In this paper, we use a large sample of Korean stocks to explore how foreigninvestors trade and how they impact stock prices. We "nd evidence that, beforethe Korean crisis over the last months of 1997, foreign investors engage inpositive feedback trading and herd. During the crisis, the evidence of positivefeedback trading is much weaker. There is no evidence that herding is moreimportant during the crisis period, and some evidence that it is less important.Neither positive feedback trading nor herding, however, are necessarily de-stabilizing. When we investigate the impact of episodes of heavy foreign tradingon stock prices during the day, or across days, no convincing case can be madethat foreign equity investors play a destabilizing role in the equity markets.Although policymakers are often concerned about foreign equity investorsbecause they can withdraw their capital from a country rapidly, it is important


to remember that equity markets have built-in mechanisms that can makeforeign equity investors stay when creditors do not. In e$cient markets, assetprices fall to re#ect new adverse public information even in the absence of trades.After the new information is incorporated into prices, the incentive to sell is nolonger as powerful, since one sells at a fair price. With bank loans, however,this mechanism does not work. Loans still have to be paid in full even afteradverse information becomes known, so that a creditor wants to take his moneyout while he can, before the loan has to be renegotiated or the "rm goesbankrupt.

References

Bailey, W., Chan, K., Chung, Y.P., 1998. Depository receipts, country funds, and the Peso crash: theintraday evidence. Unpublished Paper. Cornell University.

Bekaert, G., Harvey, C.R., 1998. Capital #ows and the behavior of emerging market equity returns.Unpublished Working Paper. Duke University.

Bekaert, G., Harvey, C.R., 2000. Foreign speculators and emerging equity markets. Journal ofFinance, forthcoming.

Bohn, H., Tesar, L., 1996. U.S. equity investment in foreign markets: portfolio rebalancing or returnchasing? American Economic Review 86, 77}81.

Brennan, M.J., Cao, H.H., 1997. International portfolio investment #ows. Journal of Finance 52,1851}1880.

Clark, J. Berko E., 1996. Foreign investment #uctuations and emerging market stock returns: thecase of Mexico. Unpublished Working Paper. Federal Reserve Bank of New York.

DeLong, J.B., Shleifer, A., Summers, L.H., Walmann, R.J., 1990. Positive feedback investmentstrategies and destabilizing rational speculators. Journal of Finance 45, 379}395.

Dornbusch, R., Park, Y.C., 1995. Financial integration in a second-best world: are we still sure aboutour classical prejudices. In: Dornbusch, R., Park, Y.C. (Eds.), Financial Opening: Policy Lessonsfor Korea. Korea Institute of Finance, Seoul, Korea.

Froot, K., O'Connell, P.G.J., Seasholes, M.S., 1998. The portfolio #ows of international investors.Unpublished Working Paper. Harvard University.

Henry, P.B., 2000. Stock market liberalization, economic reform, and emerging market equity prices.Journal of Finance, forthcoming.

Holthausen, R.W., Leftwich, R.W., Mayers, D., 1990. Large-block transactions, the speed ofresponse, and temporary and permanent stock-price e!ects. Journal of Financial Economics 26,71}95.

Kang, J.-K., Stulz, R.M., 1997. Why is there a home bias? An analysis of foreign portfolio equityownership in Japan. Journal of Financial Economics 46, 2}28.

Krugman, P., 1997. Seven habits of highly defective investors. Fortune, December 27.Lakonishok, J., Shleifer, A., Vishny, R.W., 1992. The impact of institutional trading on stock prices.

Journal of Financial Economics 32, 23}43.Lee, C.M.C., Ready, M.A., 1991. Inferring trade direction from intraday data. Journal of Finance 46,

733}746.Nofsinger, J.R., Sias, R.W., 1999. Herding and feedback trading by institutional and individual

investors. Journal of Finance, forthcoming.Radelet, S., Sachs, J., 1998. The onset of the East Asian "nancial crisis. Unpublished Working Paper.

Harvard University.Stiglitz, J., 1998. Boats, planes and capital #ows. Financial Times, March 25.


Stulz, R.M., 1998. Globalization of equity markets and the cost of capital. Unpublished WorkingPaper. The Ohio State University.

Stulz, R.M., 1999. International portfolio #ows and security markets. In: Feldstein, M. (Ed.),International Capital Flows. University of Chicago Press, Chicago, IL.

Tesar, L.L., Werner, I.M., 1995. Home bias and high turnover. Journal of International Money andFinance 14, 467}492.

Wermers, R, 1999. Mutual fund herding and the impact on stock prices. Journal of Finance 54,581}622.


Documents

Do foreign investors destabilize stock markets? The Korean experience in 1997