You are on page 1of 57

Online investors trading behaviour and performance:

Evidence from the Korean equity market


NATALIE Y OH*
University of New South Wales and SIRCA Limited
JERRY T PARWADA*
University of New South Wales and SIRCA Limited
TERRY S WALTER*
University of New South Wales and Capital Markets CRC Limited
Abstract
This paper investigates the trading behaviour and performance of online equity investors
in comparison to other investors on the Korean stock market. We find that online traders
are noise traders who provide liquidity to other investors. We also show that the
aggregate trading activity of all investor types largely fails to explain market returns.
However, returns on the index have a significant positive impact in changing online
investment flows, compared to the negative feedback trading we find for other domestic
investor types. Volatility is apparently perceived as an opportunity, and like all other
investors, online traders increase their trading during volatile periods. In periods when
there is uncertainty about the future direction of the market online investors reduce
purchases and sales indiscriminately. Although some market timing ability characterizes
online buy trades, the long run performance of online investors trading decisions is
below that of other investor types.
JEL classification: G10; G20; O33
Key words: Online trading, Performance, Investor Behaviour

*We thank the Securities Industry Research Centre of Asia Pacific for financial support and acknowledge
Mr Youngjin Kim from KSDA for valuable comments. Mr Kilhyun Ahn from the Korean Stock Exchange
gets special thanks for providing data.

Corresponding author.
Address for correspondence: SIRCA Research, PO Box H58 Australia Square, NSW 1215, Australia.
Telephone: +61 2 9236 9118, Fax: +61 2 9231 5988
E-mail. natalie@sirca.org.au (Oh); jerry@sirca.org.au (Parwada); t.walter@unsw.edu.au (Walter)

1. Introduction
Recent developments in internet-based transaction technologies have allowed online
investing to become an important, if not controversial, feature of financial markets.1
Online trading has the potential to lower transaction costs and facilitate entry, resulting in
increased trading volumes (DAvolio, Gildor, and Schleifer, 2002). Despite evidence
that internet-based stock trading now accounts for a large proportion of securities trading,
it is surprising that very few academic studies have been conducted of this rapidly
expanding form of trading. Choi, Laibson, and Metrick (2002) and Barber and Odean
(2002) are the exceptions. Choi et. al compare online traders with phone-based traders
using a sample of 100,000 members of two large US pension funds, and find that the
availability of internet trading increases transactions by 50 percent. Barber and Odean
document increased trading activity and a higher propensity to speculate amongst
investors who take up online trading. Importantly, these studies show that trading profits
quickly deteriorate (Barber and Odean) or are non-existent (Choi, Liabson and Metrick)
in the period after online trading is adopted.

Online investing research is of interest from at least two perspectives. First, by


demonstrating that online trading increases volume, these studies open up the possibility
that online investing may bring price pressure to bear in markets. Second, by analyzing
post trading performance, it can be deduced whether the phenomenal growth in volume is
driven by information or cognitive biases on the part of the investors.

See (Barber and Odean, 2002) for US data and examples of adverse press coverage of online trading.

In this paper we examine the behaviour and performance of online investors, and
compare this to other investor categories in the Korean equity market. We choose this
market because the level of online investing in Korea may be characterized as being
phenomenal. For example, online trading accounted for 65.3 percent of all stock trading
on Korean Stock Exchange (KSE) in April 2003. We examine the behaviour of
aggregated online trading to identify any systematic patterns that emerge relative to other
investors. To the best of our knowledge this paper represents the first substantive study
that directly compares online investors behaviour and performance to all other
participants in the equity market. The study closest to ours is Jackson (2003). He uses
individual client accounts at a sample of 56 Australian retail stockbroking firms,
including nine internet brokers, to investigate whether discernible flow-return patterns
exist for his dataset. The average trade size is $7,570 for internet brokerage clients, and
$10,080 for the full-service brokers, suggesting that Jacksons study does not capture
much of the institutional trading on ASX.

We utilize a detailed database of daily trading volumes and values provided by the KSE.
Importantly this database links volumes and values to each of the categories of investors
originating the trades, including online investors, individual investors, foreign investors,
local institutional investors, securities houses and unclassified institutional investors.
Such data are not readily available in other markets. This makes our study a potentially
useful supplement to the studies that have used proprietary firm level data.2 Moreover,

Barber and Odean (2002) use a dataset consisting of clients of a discount brokerage firm; Choi, Laibson

and Metrick (2002) utilise data on two US pension funds.

the dataset splits trades into purchases and sales allowing us to augment our approach
with analysis based on trade imbalances, as aggregate trading activity could mask
interesting trading traits.

This paper is organized as follows. The next section summarizes the background to
online trading and the related literature and Section 3 describes the data and market
setting. Trading behaviour and performance are analyzed in Sections 4 and 5,
respectively. Section 6 concludes.

2. Background to perceptions of online investors


This section reviews the role of internet investing and the research literature relevant to
this paper.

The benefits for online trading have been documented in various industries. For
example, Brown and Goolsbee (2002) suggest that the internet may significantly reduce
search costs by enabling price comparisons online in the insurance market. However, it
is not clear that informational advantages translate into superior return performance in
equity markets. Barber and Odean (2002) investigate the performance of investors who
switched from phone-based trading to internet trading. While those traders who opted for
internet trading initially beat the market by about 2% prior to going online, their
performance decreased afterwards, resulting in performance 3% below the market. Choi,
Laibson and Metrick (2002) also report evidence of underperformance in the market

timing of online traders in a 401(k) plan.3 Thus, access to wider information sources on
the internet does not seem to imply higher return performance.

Whilst online investing facilities may have reduced the costs of trading, there is a
downside. First, the detrimental effects of high portfolio turnover have been shown to
reduce performance (Barber and Odean, 2000, 2002 and Choi Laibson and Metrick,
2002).4 In contrast, additional trading increases liquidity and this may induce even
higher volumes, possibly creating a winners curse. Second, trading volume bubbles
in online trading may result in another detrimental feature - low information revelation.
Third, online trading may also increase noise as information sources such as discussion
groups (dominated by unsophisticated investors) become an avenue for spreading
inaccurate information (Madhavan, 2000). According to DAvolio, Gildor and Shleifer
(2001), A well functioning securities market relies on the availability of accurate
information, a broad base of investors who can process this information, legal protection
of these investors rights, and a liquid secondary market unencumbered by excessive
transaction costs and constraints.

The reliance of securities markets on a broad base of investors who can process
information has been considered questionable in the case of online investors. Are online
3

401(k) plans are the primary vehicle for retirement savings in the United States. According to Choi,

Laibson and Metrick (2002), in 1999, 401(k) plans held $1.6 trillion in assets, 72% of which represent
equity holdings.
4

See also Carhart (1997) on the negative relation between portfolio turnover and net returns in mutual

funds.

investors able to process information correctly and thus contribute to a well-functioning


market? Recently, online trading has been criticized in the financial press as being a
conduit for introducing unsophisticated individuals into the market, thus making the
market too speculative in ways synonymous with a gambling arena.5 The justification
behind this accusation is that it is believed the so-called inexperienced online traders have
access to an overwhelming amount of information through the internet, which creates an
illusion of knowledge and brings about overconfidence in trading (Barber and Odean,
2002). Further, lower transaction costs with online trading have allowed online investors
to engage in speculative trading more cheaply, thus creating speculative bubbles,
inducing higher volatility, asset mispricing and poor trading performance.6

It is surprising, given the contentious perceptions of online investors behaviour, that


there has not been any systematic analysis of the aggregate behaviour and trading returns
for online investors. We provide empirical evidence on these issues and, perhaps more
importantly, compare online investors behaviour with other investors to identify whether
internet-based traders indeed behave in a manner that warrants such criticism.

See for example, Digital Manipulation The Economist, 10th February 2001 (on the Korean market);

Online Trading May Trigger New Dimension in Volatility New Straits Times Press, 19th October 2000
(Malaysia); and Technology Boom Keeps Volatility Ticking; Australian Financial Review, 28th August
2000 (Australia).
6

See Barber and Odean (2001) for a review of the experimental economics literature on asset bubbles.

3. Data and market background


3.1. Data
This paper utilizes daily data on online trading volumes obtained from the KSE. The data
identify trading volume (number of shares traded) and value (Korean won) for both
purchases and sales over the 2001-2003 period. The sample period falls after the
worldwide tech boom which started in 1999 and the subsequent tech bust which
diminished in late 2000.

The dataset provides information on aggregate online trading activity that is not readily
available in other equity markets. The dataset also contains trading volumes and values
for other market participants. These include foreign investors (hereafter denoted
Foreigners), individual investors (Individuals), local institutional investors (Institutions),
securities houses (Securities) and unclassified institutions (Others).7 Institutions
comprise insurance, investment trust companies, commercial banks, merchant banks and
pension funds. Securities are the local securities companies that trade on their own
behalf (principal trading). Others in the dataset are mainly government agencies. We
exclude Others from this study because they trade infrequently. Others have also been
7

The foreign investor category includes both institutional and individual investors, but most trades are

from institutions. It is possible that trades we identify as foreign trades are actually trades by Korean
investors who set up a foreign nominee company to trade on the KSE. This limitation has also been
identified by Choe, Kho and Stultz (1999). Prior to May 1998, foreigners had limited access to ownership
of Korean stocks. The restrictions that remain for foreigners apply to the ownership of public corporations
that are deemed to be of national strategic importance, including electricity, telecommunications, and
airline companies.

excluded in the prior research due to insignificant trading (e.g. Choe, Kho and Stulz,
2004). In line with the notation adopted for identifying different investor types, we also
refer to online investors simply as Online.

The daily market data utilized in this study are also sourced from the KSE. The data
include the dividend yield, the 90-day commercial paper rate, the foreign exchange rate
(US dollar/Korean won), the 3-year government bond and the 3-year corporate bond rate.

3.2. Market structure and descriptive statistics


The KSE is an order driven market with trading facilitated by the Automated Trading
System (ATS). There are no designated market makers or specialists. Stock trade orders
are placed through stockbrokers. Whilst a few dedicated discount online brokers exist,
almost all the full service brokers also operate systems that facilitate online trading8.

Following the authorization of online trading in late 1997, internet-based trading was not
immediately popular and the onset of the Asian economic crisis further reduced
investors willingness to adopt the new technology. It was not until late 1998, when the
economy stabilized and the tech boom had swept the Korean market, that investors truly
embraced online trading, primarily due to heavily discounted online trading commissions
and the fierce competition between providers of the service that pushed the costs even
lower. The average commission rate of online trading in 1997 was 0.5 percent, the same

To date no foreign brokerage firms provide online trading services. All online trading is provided by

local brokerages.

level as for traditional methods, but in 1999 it dropped to 0.14 percent, further dropping
to 0.07 percent by 2001 (Byun, 2002). The new technology also brought speedy
information dissemination that made trading more accessible for existing and new
investors. A combination of lower transaction costs and easy access to information not
only encouraged new investors to enter the market but apparently also increased trading
frequency and the participation of day traders. According to Korean Securities and
Derivatives Association (KSDA) figures, in 2001 day trading was responsible for 46.6
percent of the total stock trading, an increase from 38.7 percent in 2000 (see KSDA,
2003).

The KSE is amongst the most actively traded exchanges in the Asia-Pacific region. The
total value of share trading on the KSE stood at US$597 billion in 2002, a considerable
amount when compared with the largest neighboring exchanges (e.g. Tokyo, US$1,66
trillion; Taiwan, US$541 billion; Australia, US$244 billion, and Shanghai, US$291
billion).9 As reported in Table 1, share trading in Korea is dominated by individual
investors. The trading frequency (volume) of individual investors is phenomenal in
comparison to most other equity markets, in which institutions are the dominant investor
category. Individual trading on average is above 90 percent for both purchases and sales
of KSE volumes. When total trading value is considered, individuals have a lower
presence but remain the dominant group, standing at 70 percent, whilst value of trading
for other investor types such as foreigners and institutions, are about 10 percent. The
correlation between individual investors and online investors is almost perfect, although
the percentage of trading done by individuals in volume and value terms outweighs
9

Data obtained from World Federation of Exchanges Annual Statistics 2002.


8

online investors (see Table 2). Because of this almost perfect correlation between
individuals and online traders, we exclude a comparison of individual investors and
online investors. As reported in Panel A of Table 1 mean and median online trading
volumes are 80 percent of total trading volume. Reports in the popular press suggest that
these volumes are all made up of small parcels with little information content, creating
bubbles or noise in the market place.

Figure 1 shows the extent of the increase in online trading since January 1998. Clearly,
the value of non-online investing has decreased. Total on-line trading (including stocks,
futures and options) increased 146 times from 22.5 trillion won in 1998 to 3,293 trillion
won in 2002.10 This signifies that many investors have switched from traditional
methods of transacting to online trading.

4. Stock market effects of the trading behaviour of online traders


This section investigates whether online investor flows move stock prices. If so, is this in
ways that are different from the impact of other investors trades? Studies beginning with
Kraus and Stoll (1972a, b) document temporary price pressures from trades. Dealing
with a homogeneous and dominant group of investors such as online investors may give
rise to expectations that aggregate flows from this section of the market can exert
pressure on market prices. We also consider whether shifts in returns earned by the
different investors are information driven, or a consequence of price pressure. If trades

10

On-line stock trading has been declining since 2002, but on-line derivatives trading shows a consistent

increase.

are information driven then they could be reasonably expected to result in high
performance; if trading is due to cognitive biases, poor performance would more likely
follow. To facilitate comparison between online and other investors, we investigate, for
each investor type, at an aggregate level, whether the demand curve for each investor
type is horizontal or downward sloping to understand the nature of online investors
demand. This is done by examining the relationship between flow and return - whether
flow Granger causes return or vice versa. Further, if flow contains information above
market fundamentals, then there is price pressure being exerted by that investor type,
hence rejecting the traditional assumption that the demand curve is horizontal. The latter
issue is tested by adding market fundamentals to the estimated regression equations
following Cha and Lee (2001).

The relationship between flow and market return also enables the study to conclude
whether investors are in aggregate positive or negative feedback traders. Herding and
feedback trading have been extensively studied for many markets and for different
investor types. According to Grinblatt, Titman and Wermers (1995), a trade imbalance
by an investor type that is correlated with past returns can be considered feedback
trading. In many studies, large trading imbalances are interpreted to indicate investor
herding.11 Hence we investigate investor behaviour by calculating each days trade
imbalance for that investor group. We compute trade imbalance or Net Investment Flow
(NIF) as follows:

11

Herding is defined as a group of investors buying or selling during the same time interval (Nofsinger and

Sias, 1999).

10

NIFit =

Purchasing Value it Selling Value it


.
Purchasing Value it + Selling Value it

(1)

NIFit is a proxy for ownership data which enables us to identify net purchases by investor
type i at time t. This net measure is sometimes considered to be indicative of when the
market is under or over-valued, hence reflecting the market timing ability of different
investor types.

We also analyze the behaviour of each investor type by considering stock purchases and
sales separately. We normalize purchases and sales flow measures to counteract the
upward trend in market trading volumes and valuations. In line with common practice,
we normalize flow by the 90-day moving average of market capitalization for value flow
(see Warther, 1995, Goetzmann and Massa, 2003).

4.1 Simple measures of relatedness


To provide a preliminary view of trading behaviour, Table 3 reports correlations between
NIF and return and correlations among investor types. Panel A reports a negative
correlation between Online and contemporaneous return, suggesting that when Online are
net buyers the market goes down. Based on results for one-day lagged returns, the flowreturn correlation patterns are suggestive of all domestic investors engaging in negative
feedback trading whilst Foreigners are positive feedback traders. Foreigners, Institutions
and Securities exhibit a positive correlation with contemporaneous return raising the
possibility that, in terms of market timing, Online are poor performers whereas
Foreigners, Institutions and Securities are better market timers.

11

Panel B reports the relationship between investor types. Foreigners and all other investor
flows are negatively related. Combining this result with those of the flow return
correlations suggesting Foreigners are good market timers raises the specter that
domestic market participants, especially online investors, are liquidity providers to
foreigners. However, we regard these results as preliminary, and we conduct further tests
in which we control for other possible determinants in the sections below.

4.2 Detecting feedback trading


Feedback trading analysis based on correlations alone may give erroneous conclusions
due to possible multicollinearity. We use a bivariate vector autoregressive regression
(VAR) model to analyze whether online investors engage in feedback trading, following
Seasholes (2000) and Froot, OConnell and Seasholes (2001). Econometrically, a VAR
model and its transformed representation constitute a useful tool which allows us to
effectively measure the impact of one variable on fluctuations of other variables in the
model.

The VAR model specification is:


p

Zt = C + j Zt j + t

(2)

j =1

Rt
R
C =
F
Ft

where Zt =

1,1, p
p =
1, 2, p

1, 2 , p
R ,t
=

.
2, 2 , p t F ,t

Zt is a 21 matrix of return, Rt, and flow, Ft (Purchases, Sales or NIF) for day lag j, C is
the constant and t is the 21 error matrix.

12

The VAR analysis constitutes estimates of reduced form equations with uniform sets of
lagged dependent variables from all equations as regressors. Using the Schwartz and
Alkaike criteria, we find that two lags of each variable are enough to capture the linear
interdependencies in the system. However, a five day lag is chosen to capture, and report
on, trading patterns over the preceding week.

Table 4 reports results from a bivariate VAR (5) model for all investors. Panel A, Part 1
reports the results for Online with Return as the dependent variable. All three flow types
(Purchases, Sales and NIF) and past returns fail to show any significant impact on
returns. This indicates that aggregate online flows do not move the market index. The
absence of a significant relationship between past returns and contemporaneous return
demonstrates that KSE is at least weak-form efficient. With Flow as a dependent variable
(Part 2), both market returns and past Online flows exhibit some significant relationship
to Online flows. For both Purchases and Sales flow measures, the market return impact
lasts up to four lags. It is interesting to observe that Purchases and Sales move in the
same direction. Yesterdays return has a positive correlation with todays flow for both
Purchases and Sales. Two day lagged returns are negatively correlated with both
Purchases and Sales. Hence, it is not surprising that we do not observe a significant
impact of market returns on NIF because it seems that Online as a group, do not share a
consensus on market movements and are likely to be just noise trading. It is therefore
hard to predict whether online investors are positive feedback traders (momentum
traders) or negative feedback traders (contrarians) at an aggregate level since there is no
distinct pattern of behaviour. The R2 value for Purchases and Sales is relatively high,

13

consistent with periods of boom and bust in trading levels, but low for NIF, consistent
with there being little information in the net opinion of online traders. Online exhibit
strong positive serial correlations in all three flows in that Online buys and sells tend to
lead other Online investors to buy and sell, demonstrating herding behaviour. This also
suggests that lagged online flow is a good indicator of contemporaneous flow.

Our results on Online flows relations with past returns differ from those of Jackson
(2003) who finds evidence of negative feedback trading by internet brokers driving the
overall observed relationships for a sample including full service brokers. Our results
also contrast with the findings on Finnish and US individual investors net flows by
Grinblatt and Keloharju (2001) and Odean (1998), respectively.

Similar to Online, results reported in Table 4 Panel B indicate that Foreigners past flows
and returns alike fail to explain changes in market return. However, past market returns
and flows do have some significant impacts for the three flows (Part 2). Unlike Online,
Foreigners do have a distinct pattern in their behaviour. Yesterdays return has a
significant positive correlation on todays Purchases but no significant impact for Sales,
a position confirmed by the significant positive impact on NIF. Similarly, returns posted
two days ago have no impact on todays Purchases but are significantly positively
correlated to Sales, resulting in a significant negative relationship with NIF. This
illustrates that Foreigners as a group have more consensus (than Online) on the direction
of market movements. Part 2 also reports a significant sign reversal between one day
lagged returns and two day lagged returns for NIF flows. This may suggest that

14

Foreigners are profit seekers or bargain hunters who buy in up markets and sell the
following day, as it were to take profits. However, for a market that has been shown to
be at least weak form efficient, a more plausible interpretation is perhaps that foreigners,
as net sellers after market rises, are contrarians, and that gains posted in periods of
temporary price pressure reverse shortly afterwards due to the actions of other investors.

A significant positive coefficient is observed for Foreigners one day lagged flow and
current flow, which suggests that herding behaviour amongst Foreigners is also evident.

Table 4 Panel C (Part 1) reports results for Institutions with Return as the dependent
variable, and indicates that the coefficients on past returns are not significantly different
from zero. Unlike any other investors, Institutions two day lagged flows (NIF and Sales)
display a significant impact on returns suggesting the positive NIF impact on Return is
driven by negative Sales flow. In terms of results on Flow in Panel C (Part 2),
Institutions exhibit negative feedback trading. Yesterdays return induces significantly
higher Sales and no clear impact on Purchases, thus leaving a negative NIF. The result
that Korean institutions are negative feedback traders contradicts findings for US
institutions which suggest positive feedback trading (Grinblatt, Titman and Wermers,
1995; Wermers, 1999; and Nofsinger and Sias, 1999) but portrays similarities to Japanese
institutions (Kim and Nofsinger, 2002). Since Institutions are net sellers and Foreigners
net buyers, it is possible that these two investor types interact. Positive serial correlation
is detected for up to a five day lag for Purchases and Sales but only up to a three day lag

15

for NIF. This result indicates possible herding behaviour by Institutions - yesterdays
trading activity spurs other institutional investors to trade today.

Finally, Table 4 Panel D reports results for Securities. Analogous to Online, Foreigners,
and Institutions, past flows and returns for Securities display an insignificant impact on
Returns. However, both past returns and flows show some significant results for all three
flows. Securities exhibit significant net selling activity for one day lagged returns and net
buying activity for two day lagged returns which is the exact opposite of Foreigners
conduct, but in line with the behaviour of Institutions. This suggests that Securities, like
Institutions, interact with Foreigners. Positive serial correlation is evident for flows for
up to five days.

In summary, the R2 statistics reported for all investors in Table 4 indicate that past market
returns are an important variable in explaining flows whilst the explanatory power of past
flow measures on market returns is low. Past market returns do not have any significant
impact on current market returns. As well, significant positive serial correlations are
noted for Online, as with all other investor types, which suggest that there is significant
herding behaviour evident in the Korean market. All in all, online investors, like other
investor types, do not exhibit trading patterns that suggest particular skill in trading.

4.3 Is there information in flows?


Our analysis of the flow-return relationship is augmented in this section by further tests
of whether, beyond simply reacting to market returns, online investors flows contain any

16

information about market returns. The study by Cha and Lee (2001) conjectures that
flows may affect returns through either price pressure, or their effect on aggregate market
revisions of fundamentals about stock prices. Cha and Lee devise a method of
disentangling the price pressure and information effects of investor flows. Whilst it is
difficult to pinpoint the appropriate market fundamentals for purposes of our study, there
is reason to expect that, at the market level, dividends are important drivers of prices, as
too would be the risk premium. At the macroeconomic level interest rates and, for a
fairly open market like Korea, foreign exchange rates are further possible candidates.

We conduct the following regression analyses, loosely based on Cha and Lee (2001):
3

i =1

i =1

Rett = + Rt i + Flowt i + Ddiv + Dint + Aspr + Axrate ,

(3)

and
3

i =1

i =1

Flowt = + Rt i + Flowt i + Ddiv + Dint + Aspr + Axrate

(4)

In the specifications, Flowt is measured by Purchases, Sales or NIF, Rt is market return


calculated by the natural logarithm of the difference between the level of the market
index on day t and day t-1; the market fundamentals are Ddiv, the dividend yield; Axrate,
the change in the foreign currency exchange rate (US dollar / Korean won); Aspr, the
spread, proxied by the difference between the 3-year government bond and 3-year
corporate bond rate; and Dint, the 90-day commercial paper rate. Since the frequency of
the data is daily and the market fundamentals do not fluctuate much on a daily basis, with
the exception of the foreign exchange rate, the variables are averaged over the past three

17

days. Short-term interest rates and dividend yields are differenced to meet stationarity
conditions.

Table 5 presents results of tests of Granger causality between flows and returns in the
presence of market fundamentals. The most interesting result in Panel A for Online is that
our specifications reject the null of no causality between net flows and returns whilst
failing to reject the null of causality in the opposite direction. The lack of causality from
flows to returns suggests the absence of any information in Online flows concerning
market returns. Strikingly, the significant impact (negative in sign) of two day lagged
returns on both Online Purchases and Sales followed by a positive reaction to returns the
following day, as reported in Panel A (Part 2), is consistent with the directionless nature
of online investor trades shown in Table 4. Online flows clearly contain information
about the following days flows, particularly in the case of disaggregated flows; however
only a low level of statistical significance applies to net flows lagged one day (again
emphasizing the absence of a clear trading strategy by online investors).

We contrast the flow-returns relations observed for Online with those for other investor
types. Generally net flows from Foreigners, Institutions and Securities (see Panels B
(Part 2), C (Part 2) and D (Part 2)) do not contain information about market returns in
the presence of market fundamentals. The exception is for Institutions flows lagged by
two days where we reject the null of causality from these flows to returns. This result
suggests domestic institutional investors are privy to some private information about

18

returns.12 This may be the signal used by Foreigners to purchase prudently, along with
Institutions themselves, the following day as their one day lagged purchases seem to
contain information about market returns even after incorporating market fundamentals
(possibly with Online and Securities being the providers of the necessary liquidity). The
non-online traders all appear to be influenced by past returns in their investment
strategies. The signs on the coefficients for one-day lagged returns suggest positive
feedback trading behaviour for Foreigners, and negative feedback trading by Institutions
and Securities. (See Panels B (Part 1), C (Part 1) and D (Part 1), respectively). The
difference in the behaviour of Foreigners compared to other non-online investors raises
questions about whether they act in a manner that exploits the trends observed for local
investors. However, this question is beyond the scope or our paper.

In possible vindication of online investors, it is interesting to note that all other investors
flows are also subject to positive serial correlation. The levels of severity differ though
for instance, for Institutions all lags of the various flow measures reveal strong positive
serial correlation; for Securities this phenomenon applies more significantly in the case of
disaggregated flows; and for Foreigners it is generally restricted to one lag of daily
flows. These results in part deflect the criticism of online investors to the extent that their
trading behaviour seems to be no more affected by past trading trends instead of market
fundamentals than other investor types.

12

Although we do not investigate this issue we note that the Korean corporate market is dominated by

cross-holding structures, Chaebols, of which the Institutions in our study often are part.

19

In summary, since for most investor types all three flows do not contain additional
information about returns over and above market fundamentals, it is likely that the price
pressure hypothesis does not hold. Flows from online investors, like those of other
investor types, simply respond to changes in market returns. In this regard, these findings
are consistent with results for US institutions (Cha and Lee, 2001), and imply that a
horizontal market demand curve for equities holds at an aggregate level. The presence of
strong positive serial correlation for all investors demonstrates that previous trading
activity by the same investor class has a significant impact on current flows. That this
phenomenon is shown to exist in the presence of market fundamentals emphasizes the
possible influence of herding on trading patterns amongst investors in the Korean stock
market at the expense of fundamental information.

4.4 Risk, uncertainty and market fundamentals as determinants of flows


According to finance theory investors decisions are based not only on expectations about
return but also on the uncertainty of those returns. What bearing does risk have on the
determinants of flow for online investors and how does this compare with other investor
types? In a classical CAPM model it is known that risk has negative utility effects on
investors and therefore a negative relationship with demand is expected. But if online
investors are unsophisticated and increase their comparative presence during periods of
market uncertainty, allegations about the detrimental effects of online trading are likely
justified. This is because it is usually speculators who enter the market and increase their
presence during periods of uncertainty, hoping to reap profits from volatility, driving
prices away from fundamentals (Goetzmann and Massa 2003). In the previous section

20

we investigated the relationship between flow and returns. In this section we examine
investors risk perceptions and behaviour and the role of risk in determining the volume
or demand for each investor type. We investigate the role of uncertainty in determining
flows using the following regression equation:

Ft = + Unct + InfVt + Ft 1 + t

(5)

where Ft denotes flow measurements as being Purchases, Sales, or NIF. InfVt is a vector
of information variables (or market fundamentals as explained in section 4.2), t is the
error term, and Unct represents the measure of uncertainty under consideration. The
specification is similar to that use by Goetzmann and Massa (2003).

Two proxies for uncertainty are considered. First, volatility is measured as the square of
the natural logarithm of return.13 Since anecdotal evidence suggests Korean investors are
mostly day-traders, intra-day volatility based on the Garman and Klass (1980) measure is
also used for robustness.14 Second, as an estimate of the dispersion of investor beliefs,
KOSPI 200 futures open interest, standardized by dividing daily open interest of KOSPI

13

Bae, Chan and Ng (2004) use this measure to capture the volatility in their study for all the emerging

markets which includes Korea.


14

Garman and Klass (1980) investigate the relative efficiency of various measures of volatility and identify

a volatility measure with the highest efficiency.

In the so-called GK measure, written as

t = VAR(GK ) = 0.5[ LN ( High) LN ( Low)]2 [2 LN (2) 1][ LN (Open) LN (Close)]2 , VAR(GK) is the variance using the

Garman-Klass (1980) method, LN denotes the natural logarithm, and High, Low, Open, Close are the high,
low, open, and closing prices of the day to determine the volatility.

21

200 futures by the trading volume on KOSPI 200 futures contract on the same day, is
utilized.

The results, reported in Table 6 Panel A show that a strong positive relationship exists
between Unc (volatility) and disaggregated Online flows but not net flows (NIF). This
shows that volatility is an important contributor to Online flow even though the weak
relationship between volatility and NIF may demonstrate that Online increase their
presence in the market indiscriminately during volatile periods. Increasing both
Purchases and Sales during volatile periods may further increase volatility.15 Hence,
Online trading could be seen as speculation in the market during volatile periods,
possibly creating bubbles and a winners curse.16 Similar to our results for earlier tests, a
strong positive serial correlation persists for up to three days which may intensify the
creation of bubbles. One market fundamental, Ddiv, is significantly related to all three
flow measures. Remarkably, the positive relationship for NIF reported for Online in
Table 6 defies the trend of negative relations for all other investor types net flows. This
may indicate that online investors do take into consideration some fundamental
information, particularly easily accessible variables such as dividend yield, in

15

Perhaps this is the reason why online investors have been blamed for increasing volatility and

destabilizing the market. This is because during periods of higher volatility it is expected that speculators
would enter the market attracted by the possibility of higher gains, pushing prices away from the
fundamentals.
16

Speculative bubbles can also be manifestations of the winners curse. The winners curse is more likely

when there are more bidders and when the dispersion of opinions about the value of whatever is being
auctioned is greater.

22

determining their trade decisions, but they are still unable to process the information in a
manner that results in astute trading strategies. This supports the perception of online
investors as being naive market participants.

The other investor types, Foreigners, Institutions, and Securities all display a significant
positive relationship between volatility and Purchases and Sales in the presence of
market fundamentals but fail to show a significant relationship between volatility and
NIF. Clearly Online are not the only class of investors to increase Sales and Purchases
volume during volatile periods. This suggests that the market as a whole perceives
volatility as an opportunity to gain as suggested by Goetzmann and Massa (2003). For
non-online investors, the Dint market variable bears a significant positive influence on
disaggregated flows not borne out in net flow terms.

On the alternative volatility measure we adopt, the above results are qualitatively similar
for regressions incorporating the Garman and Klass intraday volatility computation.17

Another risk measure adopted in the model is dispersion of beliefs. Dispersion of beliefs
amongst investors is proxied for by the open interest of derivative contracts, the Korean
futures contracts. The relationship between level of dispersion and demand (Purchases
and Sales) is significantly negative for Online (see Table 6 Panel B) but, once again, the
relationship is insignificant for NIF. This indicates that during periods when there is

17

Results available from the authors on request.

23

disagreement about the future Online investors withdraw from the market
indiscriminately. The same results are evident for Institutions and Securities.

Foreigners, however, show a significant negative relationship for all three flows,
including NIF. The results indicate that Foreigners are more informed about the level of
dispersion of beliefs and withdraw from the market more discerningly. This is in line
with past research documenting that increases in the level of disagreement about the
future course of the market by sophisticated investors (those who use derivatives) induces
investors to be cautious (Goetzman and Massa, 2003). For Foreigners, clarity about the
future is an important determinant of the level of investor trading activity. The
interpretation of the rest of the explanatory variables remains unchanged for regressions
incorporating the dispersion of beliefs as a risk measure.

5. Trading performance
This section investigates whether each investor types trading is based on information
rather than cognitive biases. This is examined by analyzing post trading performance.
Specifically, more rigorous tests of market timing ability are carried out to examine the
extent to which trades affect prices in subsequent periods. Online investors have been
alleged to be poor performers or market timers and the evidence we have presented so far
appears to point in this direction. Hence this section shows, in comparison to other
investor types, whether online investors are indeed the losers in the market. Due to the
absence of portfolio holdings data, a direct estimate of investor group performance
cannot be implemented. However, following the work of Kamesaka, Nofsinger and

24

Kawakita (2003), this study utilizes daily purchase and sale flows to characterize the
market timing ability of these investor groups, which serves the purpose of proxying for
ownership or portfolio holdings when examining market returns after each trading day.
We firstly examine market performance after those days when investors conducted
particularly heavy buying or selling. We then estimate the cumulative return due to the
daily changes in investment flow and the subsequent market return for each investor
group.

5.1 Market timing ability


To examine whether the trading observed is motivated by information or cognitively
biased behaviour, the performance of investor groups after heavy buying and selling days
is measured. We conjecture that there is a stronger motivation behind heavy buying and
selling and it is on these days we expect to capture clearer evidence of information driven
or cognitively biased trading. After sorting each investor groups data by level of NIF
into five equal sets, we class the highest positive NIF quintile as the buying days, and the
largest negative NIF the selling days. Daily, weekly and monthly market returns
following heavy buying and selling trading days are calculated. These give an indication
of market timing ability.

The results for performance after heavy buying and selling days in terms of value are
reported in Table 7. The imbalance of the buy and sell days is low for Online. Online,
on average, have NIFs of -0.037 and 0.054 during heavy sell and buy days, respectively.
The market rises after a day of heavy buying and selling by Online in a statistically

25

significant sense. But the market falls after a week and a month of heavy buying and
selling. This indicates that Online are relatively good at market timing for buys on a
daily basis and good market timing for sells on weekly and monthly basis. A comparison
of average trading imbalance between Online and the rest of traders on the KSE, reveals
that Online generally do not take extreme buy or sell positions. This may confirm that
Online are not well informed about the market and therefore do not have confidence to
take extreme trading positions.

Foreigners show relatively high buy and sell NIF imbalances. The market rises for up to
a week after heavy buying from Foreigners which shows Foreigners have good market
timing ability for buys. But after heavy sells from the Foreigners the market fails to
show significant declines, in fact there are significant increases. We conclude from this
asymmetry that Foreigners are therefore good at timing their buys but not their sells.
This result is similar to Online.

After a day of heavy buying and selling by Institutions, the market rises and declines,
albeit our results in this regard are without statistical significance. The t-statistics fail to
show any significance for post-trading performance for Institutions trades.

Securities on the other hand have the highest trading imbalance. The NIFs for heavy buy
and sell weeks are -0.395 and 0.330 respectively. The post-trading performance of
Securities is poor. The market rises after heavy sell days and the market declines after
heavy buy days. This poor performance persists for up to a month.

26

5.2 Cumulative performance


To investigate who are the ultimate winners and losers on the Korean Stock Exchange the
relative market timing ability of the investor groups based on the entire sample period,
rather than just heavy buying and selling days as in the previous section, is examined.
The performance measure used in this section was originally developed by Grinblatt and
Titman (1993) in their study based on the change in the portfolio holdings and later
modified by Karolyi (2002) and Kamesaka, Nofsinger and Kawakita (2003) by utilizing
net investment flows in the place of portfolio holdings. For a full description of the
derivation of the performance measure refer to Appendix 1.

The following empirical specification estimates the cumulative return due to the daily
changes in investment flow and following market returns:
T

Cumulative return = (
t =1

Purchaset -1 Salest 1
) Rt
Purchaset -1 + Salest 1

(6)

where Purchases and Sales are raw values and Rt is the market return. Equation (6) is
estimated for each investor group in analyzing the performance over the entire sample
period.

Daily cumulative performance for the duration of sample period is graphed in Figure 2.
Consistent with performance based on heavy buy and sell days, cumulative performance
shows that Foreigners and Institutions attain the best returns. Online and Securities are
the losers but the extent of Securities losses is quite extreme compared to Online. Since
Securities have extreme positions (imbalances) which is a sign of herding (Nofsinger and
Sias, 1999), the inability to time the market would have devastating effects on their

27

performance at an aggregate level. On the other hand, because Online investors net
position is more or less balanced, their losses are probably mitigated.

5.3 Robustness tests


The results for Securities are surprising given that they are commonly presumed to be
more informed than other domestic traders. As such, we conduct a robustness check by
looking at weekly rather than daily performance. The weekly aggregation follows
convention as laid out in Lo and Wang (2000). The weekly cumulative performance is
graphed in Figure 3. The results are consistent for weekly as well as daily flows for
Foreigners, Institutions and Online but the opposite result to the daily returns is recorded
for Securities. Based on weekly returns Securities have positive returns. Foreigners are
a clear winner on the Korean stock exchange. Consistent with daily return results, Online
at an aggregate level are losers.
As a further robustness test, past, contemporaneous and lead return is regressed against
NIF for each investor type to assess whether they are good market timers and also to
investigate whether their flows affect future returns at daily and weekly frequencies. The
functional specification is:
5

NIF = + i Ret t +i + ,

(7)

where NIF is defined as

Purchases Sales
for each investor type and lead/lag market
Purchases + Sales

returns Ret is the return of KOSPI index.

28

Table 8 shows the relation between NIF for each investor type and lead/lag market
returns on daily and weekly basis, respectively. From observing daily past returns Online
investors are negative feedback traders whereas Foreigners are positive feedback traders
up to three day lag returns. Institutions and Securities also show negative feedback
trading at daily level. No significant results are present for weekly past returns.

The results from contemporaneous daily return show that Foreigners, Institutions and

Securities are good market timers, whereas Online display signs of naivety, increasing
their NIF on days when the market return is negative. In the results for weekly
contemporaneous return Foreigners continue to be winners, and Online remain as losers.
No significant results are evident for Institutions and Securities at weekly frequency.

To summarize our results on investor performance, Online show some superior market
timing ability during extreme buying positions up to a week. But looking at longer
horizons and taking into consideration all rather than just extreme cases, Online perform
poorly. This suggests that Online at an aggregate level are purely liquidity providers who
lose out to superior investors, although there exists some marginal online investors who
display superior performance as shown for extreme buying positions. Foreign investors,
in both general and extreme cases, record superior performance which supports the view
that foreigner trading is information driven (Seasholes, 2000, Froot, OConnell and
Seasholes, 2001, Kamesaka, Nofsinger and Kawakita, 2003) and confirms that they are
the winners on the Korean stock exchange. However this result is contrary to Choe, Kho

29

and Stulzs (2004) results based on data sampled during the Asian crisis for the Korean
market.

6. Summary and conclusions


In this study we utilize daily data from the highly active Korean stock market to
investigate the trading behaviour and performance of online investors. We compare our
online results with other investor types. Online investors do not exhibit distinct trading
patterns at an aggregate level and show signs of noise trading whereas foreigners,
institutions and securities companies exhibit some levels of intra-group consensus in the
direction of their trades in response to market movements. In the main, aggregated flows
from investors, with the exception of lagged local institutions trades, fail to explain
changes in the market index but the market index has a significant impact on investment
movements.

Positive serial correlation is persistent for all investors which raises concerns over the
results of investors actions. Serial correlation, when interpreted as herding, could push
prices away from fundamental values, destabilising the market (Lakonishok, Shleifer and
Vishny, 1992 and Wermers, 1999). This is more of a concern for online investors
relative to other investors since they constitute the vast majority of trading activity, in
both shares and value) on the KSE.

Results on the risk-volume relationship are similar for all groups online investors join
other investor types in increasing their trading volumes indiscriminately during volatile

30

periods, suggesting that volatility is perceived as an opportunity. Our results based on the
dispersion of investor beliefs, proxied by open interest on the derivatives market, suggest
that only foreign investors discriminate between reductions in purchases and sales in
response to rising uncertainty.

Based on performance measures following periods of extreme imbalances in trading,


online investors, together foreign investors, record superior performance only at the
highest buying levels, but in aggregate their gains are likely neutralized by poor selling
positions. In the longer run, foreigners are the clear winners with online investors the
worst performers.

Our study may also have important policy implications. We find that online investors
trading decisions are related to one variable, the dividend yield, amongst the proxies for
fundamental information that we adopt. Since this information is easily available even to
fairly nave investors, one is driven to conjecture whether improvements in the way
online investors access quality information could instill more discipline in their trading
behaviour. As DAvolio, Gildor and Shleifer (2001) have outlined, accurate information
must be available to be of use to investors. Because the trading community on the
Korean stock market mainly comprises online investors, the so-called unsophisticated
investors, there may be an incentive for information providers, who post better long term
gains, to discriminate against this source of cheap liquidity. Solutions need to be
carefully thought out, but the beginning point could involve greater efforts at investor
education.

31

With regards to the implications of this study for future research, it is important to note
that online investors dominate not only the equities market, but also the futures market.
As such, the behaviour of online traders and their performance in the derivatives market
is a potentially fruitful area for further research.

32

Appendix 1: Cumulative Performance Measure


The performance measure developed by Grinblatt and Titman (1993), based on changes
in portfolio holdings, captures the covariance between return on an asset and the change
in proportional holding of an asset as follows:
T

COV = ( w j ,t w j ,t 1 ) R j ,t / T ,
t =1 j =1

where Rj is the return, wj is the weight of holding asset, N is the number of the assets and
T is the estimating sample period.

In this paper, we estimate this portfolio holding change measure for each investor i, using
modified conditions adopted by Karolyi (2002) and Kamesaka, Nofsinger and Kawakita
(2003).

The modified condition assumes that there are only two assets, one is the stock market
index and the other is the risk-free rate. The proxy for the daily or weekly market return
is calculated using the market index, the KOSPI index, and the daily risk free rate is
assumed to be zero. The modified measure is:
T

COV = ( wt wt 1 ) Rt / T
t =1

where Rt is the return on the market index during period t. Similar to Karolyi (2002) and
Kamesaka, Nofsinger and Kawakita (2003) we replace the change in portfolio weight
with net investment flows defined as

Purchases Sales
.
Purchases + Sales

33

References

Bae, K., Chan, K., Ng, A., 2004. Investibility and return volatility. Journal of Financial
Economics 71, 239-263.
Barber, B., Odean, T., 2000. Trading is hazardous to your wealth: The common stock
investment performance of individual investors. The Journal of Finance 55, 773-806.
Barber, B., Odean, T., 2001. The internet and the investor. Journal of Economic
Perspectives 15, 41-54.
Barber, B., Odean, T., 2002. Online investors: Do the slow die first? Review of Financial
Studies 15, 455-487.
Brown, J., Goolsbee, A., 2002. Does the internet make markets more competitive?
Evidence from the life insurance industry. Journal of Political Economy 110, 481-507.
Byun, J., 2002. Online securities trading in Korea: Developments and trends. Working
Paper, APEC Working Group on Electronic Financial Transactions Systems.
Carhart, M. 1997. On persistence in mutual fund performance. The Journal of Finance 52,
57-82.
Cha, H., Lee, B., 2001. The market demand curve for common stocks: Evidence from
equity mutual fund flows. Journal of Financial and Quantitative Analysis 36, 195-220.
Choe, H., Kho, B., Stulz, R., 1999. Do foreign investors destabilize stock markets? The
Korean experience in 1997. Journal of Financial Economics 40, 31-62.
Choe, H., Kho, B., Stulz, R., 2004. Do domestic investors have an edge? The trading
experience of foreign investors in Korea. Working Paper , Ohio State University.
Choi, J. J., Laibson, D., Metrick, A., 2002. How does the internet affect trading?
Evidence from investor behavior in 401(k) plans. Journal of Financial Economics 64,
397-421.
DAvolio, G., Gildor, E., Schleifer, A., 2002. Technology, information production, and
market efficiency. Economic Policy for the Information Economy, A Symposium
Sponsored by the Federal Reserve Bank of Kansas City.
Froot, K., OConnell, P., Seasholes, M., 2001. The portfolio flows of international
investors. Journal of Financial Economics 59, 151-193.
Garman, M., Klass, M., 1980. On the estimation of security price volatilities from
historical data. Journal of Business 53, 67-78.

34

Goetzmann, W., Massa, M., 2003. Index funds and stock market growth. Journal of
Business 76, 1-28.
Grinblatt, M., Keloharju, M., 2000. The investment behavior and performance of various
investor-types: A study of Finlands unique data set. Journal of Financial Economics 55,
43-67.
Grinblatt, M., Titman, S., 1993. Performance measurement without benchmarks: An
examination of mutual fund returns. Journal of Business 66, 47-68.
Grinblatt, M., Titman, S., Wermers, R., 1995. Momentum investment strategies,
portfolio performance, and herding: A study of mutual fund behavior. American
Economic Review 85, 1088-1105.
Jackson, A., 2003. The aggregate behaviour of individual investors. Working Paper,
London School of Economics.
Kamesaka, A., Nofsinger, J., Kawakita, H., 2003. Investment patterns and performance
of investor groups in Japan. Pacific-Basin Finance Journal 11, 1-22.
Karolyi, A., 2002. Did the Asian financial crisis scare foreign investors out of Japan?
Pacific-Basin Finance Journal 10, 411-442.
Kim, K., Nofsinger, J., 2002. Institutional herding, business groups, and economic
regimes: Evidence from Japan. Working Paper, Washington State University.
Kraus, A., Stoll, H., 1972a. Price impacts of block trading on the New York Stock
Exchange. The Journal of Finance 27, 569-588.
Kraus, A., Stoll, H., 1972b. Parallel trading by institutional investors. Journal of Financial
and Quantitative Analysis 7, 2107-2138
Korea Securities Dealers Association., 2003. Securities Market in Korea 2003.
Lakonishok, J., Shleifer, A., Vishny, R., 1992. The impact of institutional trading on
stock prices. Journal of Financial Economics 32, 23-43.
Lo, A., Wang, J., 2000. Trading volume: Definitions, data analysis, and implications of
portfolio theory. Review of Financial Studies 13, 257-300.
Madhavan, A., 2000. In search of liquidity in the internet era. Paper presented to the 9th
annual financial markets conference of the Federal Reserve Bank of Atlanta.
Nofsinger, J., Sias, R., 1999. Herding and feedback trading by institutional and individual
investors. The Journal of Finance 54, 2263-2295.

35

Seasholes, M., 2000. Smart foreign traders in emerging markets. Working Paper,
Harvard Business School.
Warther, V., 1995. Aggregate mutual fund flows and security returns. Journal of
Financial Economics 39, 209-235.
Wermers, R., 1999. Mutual fund herding and the impact on stock prices. The Journal of
Finance 54, 581-622.

36

Table 1: Sample Descriptive Statistics


This table presents summary statistics for daily equity sales and purchases for six investor types on the Korean Stock Exchange. Panel A presents volume
statistics (in lots of 10,000 shares), and Panel B, values (1 million Korean won). Raw reports the actual volume and value whereas Ratio represents raw
figures for that investor type divided by the total volume or value. The sample period is January 2001 to February 2003.
Purchases

Sales

Panel A : Volume

Raw

Online

Foreigners

Individuals

Institutions

Securities

Online

Foreigners

Individuals

Institutions

Securities

Mean

538,237

12,882

621,573

13,799

4,306

525,993

13,352

615,771

17,313

4,621

median

481,978

11,291

556,194

13,202

3,506

463,274

11,898

551,274

14,606

3,854

std dev

306,923

7,485

327,395

5,050

3,243

302,428

7,193

327,452

13,813

3,387

Min

138,697

873

92,674

1,764

121

133,298

2,600

90,580

1,738

327

Max

2,173,568

75,906

2,341,094

34,660

34,255

2,145,120

57,871

2,342,724

222,297

40,038

Mean

0.80

0.02

0.94

0.02

0.01

0.78

0.02

0.93

0.03

0.01

median

0.80

0.02

0.94

0.02

0.01

0.78

0.02

0.93

0.03

0.01

std dev

0.06

0.01

0.05

0.01

0.01

0.06

0.01

0.05

0.02

Min

0.29

0.24

0.28

0.22

Max

0.92

0.07

0.99

0.1

0.07

0.91

0.11

0.99

0.2

0.04

Mean

1,429,624

282,158

1,749,835

260,423

79,334

1,414,701

274,172

1,754,300

262,967

81,824

median

1,271,775

259,484

1,552,106

239,803

67,698

1,273,767

250,602

1,576,952

234,458

73,272

std dev

615,412

136,156

766,741

123,260

49,310

614,422

133,303

778,433

125,894

47,943

min

485,852

11,022

226,597

16,560

1,826

514,907

34,233

194,352

15,781

2,933

max

3,813,982

923,406

4,767,107

846,906

361,240

3,546,988

806,597

4,442,588

821,838

454,606

mean

0.58

0.12

0.71

0.11

0.03

0.57

0.12

0.71

0.11

0.03

median

0.58

0.11

0.72

0.11

0.03

0.58

0.11

0.72

0.11

0.03

std dev

0.06

0.04

0.08

0.03

0.02

0.06

0.05

0.08

0.03

0.02

min

0.24

0.16

0.01

0.22

0.02

0.14

0.01

max

0.74

0.26

0.87

0.30

0.14

0.72

0.29

0.88

0.28

0.12

Ratio

Panel B: Value

Raw

Ratio

37

Table 2: Correlation Between Investor types: Volumes and Values


This table reports Pearson correlation statistics between investor types for daily Purchases and Sales.
Panel A reports correlations between the investor types for volume. Panel B reports the correlation for the
value. P-values are in parentheses. The sample period is January 2001 February 2003.

Purchases

Sales

Panel A : Volume
(000)

Foreigners

Individuals

Institutions

Securities

Online

0.390

1.000

0.070

0.230

(<.0001)

(<.0001)

(0.090)

(<.0001)

0.400

0.340

0.320

<.0001

(<.0001)

(<.0001)

Foreigners
Individuals

Online

Individuals

Institutions

Securities

0.290

1.000

0.100

0.160

(<.0001)

(<.0001)

(0.020)

(<.0001)

0.290

0.050

0.130

(<.0001)

(0.230)

(0.001)

Foreigners

0.120

0.250

(0.010)

(<.0001)

Individuals

0.560

Institutions

Foreigners

0.100

0.180

(0.020)

(<.0001)
0.180

Institutions

(<.0001)
Securities

(<.0001)
Securities

Panel B : Value
Online
Foreigners
Individuals
Institutions

Foreigners

Individuals

Institutions

Securities

0.570

0.990

0.710

0.550

(<.0001)

(<.0001)

(<.0001)

(<.0001)

0.580

0.570

0.580

(<.0001)

(<.0001)

(<.0001)

Online
Foreigners

0.730

0.560

(<.0001)

(<.0001)

Individuals

0.710

Institutions

(<.0001)
Securities

Individuals

Institutions

0.520

0.990

0.690

Securities

0.490

(<.0001)

(<.0001)

(<.0001)

(<.0001)

0.500

0.520

0.430

(<.0001)

(<.0001)

(<.0001)

0.700

0.500

(<.0001)

(<.0001)
0.650
(<.0001)

Securities

38

Foreigners

Table 3: Correlation Between Investor types: Net Investment Flows


Pearson correlation coefficients are reported between each investor groups NIF and market returns and
also between investor types. Return is the daily return of the KOSPI index for the day of the investment
flow (t=0) and the preceding five days to capture and report on the whole trading week. NIF is defined
as

Purchase - Sales
Purchase + Sales

. The sample represents 531 days of investment flows.

Online

Foreigners

Institutions

Securities

Panel A: Flow-return correlations


Return (t=0)

-0.616

***

0.374

***

0.127

***

0.193

***

Return (t=-1)

-0.144

***

0.408

***

-0.140

***

-0.117

***

Return (t=-2)

-0.094

**

0.059

0.037

0.169

***

Return (t=-3)

-0.035

-0.011

-0.010

0.095

**

Return (t=-4)

-0.017

0.030

-0.026

0.009

Return (t=-5)

-0.009

0.054

-0.039

-0.032

Panel B: Investor type correlations


Online
Foreigners

-0.472

***

Institutions

-0.331

***

-0.372

***

Securities

-0.463

***

-0.152

***

0.289

*, **, *** denotes significance at 10, 5 and 1 percent levels.

39

***

Table 4: Vector Autoregressive Regression Analysis of Flows and Returns


This table reports results from Vector Autoregressive Regression (VAR) of flows and returns for four
different investor types. The three flows are NIF, Purchases and Sales based on daily trading value. NIF is
defined as

Purchases Sales
Purchases + Sales

, and Purchases and Sales are normalized by dividing daily trading value by the

90-day moving average of the KOSPI indexs market capitalization. Return is the daily return on the
KOSPI index. Five lags are chosen to capture up to one week effects. Daily number of observations for
the sample period is 531. ***, **, * denotes significance at 1, 5, and 10 percent levels. The t-statistics are
in italics.
Panel A: Online
Part1

Returns

Part2

Flows

NIF

Purchase

Sales

NIF

Purchase

0.000

-0.002

-0.002

0.006

0.108

-0.790

-0.881

3.814

-0.045

-0.013

-0.030

-0.083

0.014

-0.797

-0.280

-0.596

-0.982

5.958

Sales

0.000
***

3.514

0.000
***

3.113

***

Return
(-1)

(-2)

(-3)

(-4)

(-5)

-0.063

-0.059

-0.077

-0.059

-0.005

-1.115

-1.228

-1.523

-0.696

-2.036

0.028

-0.013

-0.006

-0.028

0.005

0.500

-0.271

-0.110

-0.333

1.934

0.004

-0.046

-0.037

-0.049

0.006

0.081

-0.976

-0.738

-0.579

2.636

0.009
***

3.253

***

-0.007
**

-2.678

***

0.003
*

1.235
0.005

***

1.973

-0.010

-0.001

0.000

0.009

-0.001

-0.003

-0.176

-0.026

-0.008

0.114

-0.433

-1.081

-0.043

0.469

0.925

0.121

-1.147

0.501

0.990

2.153

-0.018

0.308

0.468

0.061

0.180

-0.485

0.288

0.435

1.085

3.407

Flow
(-1)

(-2)

(-3)

(-4)

(-5)

0.576
**

12.492

0.041

-0.057

-0.760

-0.002

0.123

1.095

-0.053

-0.706

-0.039

2.333

0.603
***

12.135
0.180

***

3.144
0.068

***

1.193

0.034

0.489

0.187

-0.039

0.002

0.031

0.915

0.468

0.178

-0.699

0.047

0.549

-0.035

-0.863

-0.425

0.030

0.038

0.040

-0.948

-0.970

-0.479

0.536

0.875

0.848

0.011

0.006

0.008

0.039

0.831

0.805

40

***

***

Panel B: Foreigners
Part1

Returns

Part2

Flows

NIF

Purchase

Sales

NIF

Purchase

0.000

-0.002

-0.002

0.005

0.000

-0.069

-0.650

-0.637

0.621

6.197

Sales
0.000
***

4.754

***

-1.297

***

Return
(-1)

(-2)

(-3)

(-4)

(-5)

-0.029

-0.022

-0.006

2.787

-0.596

-0.473

-0.135

7.032

-0.052

-0.045

-0.043

-0.819

-1.019

-0.936

-0.957

-1.953

0.005
***

**

5.464

-0.001

0.000

0.002

0.024

2.418

0.015

0.016

0.002

-0.551

0.001

0.001

0.290

0.337

0.041

-1.314

0.824

1.878

-0.022

-0.023

-0.028

-0.021

0.001

0.001

-0.427

-0.498

-0.626

-0.050

0.987

1.024

0.012

-0.004

0.014

0.211

0.000

0.002

0.237

-0.091

0.314

0.510

0.171

2.032

0.008

2.895

-0.411

0.271

1.371

1.367

-0.156

5.610

**

**

Flow
(-1)

(-2)

(-3)

(-4)

(-5)

-0.006

-2.683

0.353

0.094

-1.025

-1.227

0.123

1.899

0.295

0.432

***

6.456

***

9.814

0.023

0.076

0.498

1.587

0.001

-0.147

0.698

0.013

0.036

0.035

0.209

-0.068

0.244

0.268

0.775

0.725

-0.002

-0.708

-0.470

0.058

0.081

0.095

-0.385

-0.327

-0.165

1.165

1.742

1.993

0.001

2.521

1.569

0.014

0.164

0.272

1.269

0.606

0.322

3.823

0.008

0.012

0.004

0.244

0.284

41

***

0.138
***

3.190
0.412

***

Panel C: Institutions
Part1

Returns

Part2

Flows

NIF

Purchase

Sales

NIF

Purchase

0.000

-0.003

-0.001

-0.003

0.000

0.040

-1.252

-0.474

-0.473

4.170

-0.010

-0.023

-0.005

-1.253

-0.215

-0.500

-0.123

-4.203

Sales
0.000
***

5.418

***

Return
(-1)

(-2)

(-3)

(-4)

(-5)

***

0.001

0.005

1.408

5.402

-0.047

-0.046

-0.037

0.257

0.001

0.000

-1.034

-1.002

-0.799

0.848

1.530

0.399

0.023

0.002

0.031

-0.077

0.000

0.001

0.523

0.034

0.679

-0.257

-0.568

0.976

-0.039

-0.024

-0.029

-0.053

0.001

0.002

-0.869

-0.540

-0.653

-0.179

0.927

1.982

0.010

-0.008

0.003

-0.357

0.000

0.001

0.226

-0.168

0.063

-1.200

0.054

1.602

0.006

3.640

2.238

0.123

0.301

0.345

0.906

1.416

0.952

2.771

***

Flow
(-1)

(-2)

0.020

-0.703

-7.078

-0.265

-2.852

-0.008

-2.124

2.552

0.097

-1.191

-0.800

1.026

2.152

3.058
(-3)

(-4)

(-5)

***

***

6.680

***

2.585

0.153
***

3.425

7.849

***

1.700

0.120

1.765

-0.004

-1.209

-0.243

-0.053

0.123

-0.460

-0.098

-1.184

2.679

0.004

3.752

3.694

0.010

0.172

0.529

1.488

1.614

0.219

3.898

0.026

0.014

0.022

0.081

0.428

***

0.079

0.082
**

-0.664

42

***

0.120
*

2.586

**

0.784

***

0.036

0.153
***

3.571
0.422

***

Panel D: Securities
Part1

Returns

Part2

Flows

NIF

Purchase

Sales

NIF

0.000

-0.001

-0.004

-0.027

-0.381

-0.426

-1.614

-2.385

0.005

-0.014

-0.010

-1.561

0.109

-0.297

-0.229

-2.871

-0.053

-0.036

-0.051

1.719

-1.168

-0.788

-1.151

3.142

Purchase

Sales

0.000
**

6.069

0.000
***

5.886

***

Return
(-1)

(-2)

(-3)

(-4)

(-5)

***

0.000

0.001

0.529

3.694

0.001
***

3.905

0.000
***

0.001

-0.789

0.019

0.015

-0.002

1.052

0.425

0.316

-0.048

1.906

-0.019

-0.033

-0.039

0.113

0.001

0.001

-0.416

-0.720

-0.873

0.206

1.710

2.513

1.717

***

0.000
*

-0.586

0.012

0.009

-0.006

-0.574

0.000

0.001

0.266

0.202

-0.135

-1.046

-0.822

2.049

-0.005

3.595

5.186

0.020

0.289

0.194

-1.269

0.643

0.968

0.456

6.353

***

**

Flow
(-1)

(-2)

(-3)

(-4)

(-5)

0.000

-2.170

-3.482

0.108

0.088

-0.374

-0.643

2.400

***

0.094
**

1.985

4.467
0.114

2.586

-0.005

-8.337

3.841

-0.010

0.072

0.090

-1.389

-1.441

0.710

-0.222

1.534

2.052

-0.015

0.070

-0.318

1.595

-0.005

4.588

6.316

-0.094

-1.301

0.803

1.170

-2.136

**

0.002

5.505

0.987

-0.032

0.170

0.529

1.013

0.188

-0.739

3.849

0.014

0.010

0.012

0.068

0.268

43

***

***

**

0.144
***

3.391
0.223

***

Table 5: Granger Causality Tests in the Presence of Instrumental Variables


This table reports the results of Granger causality tests of three different types of equity flows and market
return incorporating fundamental variables including dividends, interest rate, risk premia and exchange
rate.
3

i =1

i =1

Flowt = + Rt i + Flowt i + Ddiv + D int + Aspr + Axrate (1)


3

i =1

i =1

Ret t = + Rt i + Flowt i + Ddiv + D int + Aspr + Axrate (2)


The three flows are NIF, Purchases and Sales based on daily trading value. NIF is defined
as

Purchases Sales
Purchases + Sales

, and Purchases and Sales are normalized by dividing daily trading value by the 90-day

moving average of the KOSPI indexs market capitalization. Ret is the daily return of the KOSPI index.
Ddiv is the average of the preceding three days dividend yield, differenced to meet stationarity
requirements. Dint is the average of the preceding three days 90-day commercial paper rate, differenced to
meet stationarity requirement. Aspr is a proxy for the risk premia and calculated by differencing 3 year
government bond and 3 year corporate bond rate and then taking the average of the result for the past three
days. Axrate is the average change in exchange rate (USD/Won) for the preceding three days. The results
are reported in panels (A-D) by investor type. For each panel, the null hypothesis for Part (1) is that stock
market returns do not Granger-cause flows in the presence of market fundamentals and the null hypothesis
for Part (2) is flows do not Granger-cause stock market returns in the presence of market fundamentals.
The daily number of observations is 531. ***,**,* denotes significance level at 1. 5, 10 percent levels. The
figures are Newey-West heteroskedasticity and autocorrelation adjusted.
Panel A: Online
NIF
Purchases
Variable
Coefficient
t-Statistic
Coefficient
t-Statistic
Panel A (Part 1): H0 - stock market returns do not Granger cause flows
0.016
3.224
***
0.000
1.843
Intercept
0.106
1.827
*
0.584
10.415
Flowt-1
0.056
1.004
0.168
3.183
Flowt-2
-0.013
-0.251
0.168
3.831
Flowt-3
-0.084
-1.172
0.014
3.054
Rett-1
Rett-2

-0.062

-0.792

-0.005

-1.742

Rett-3
Ddiv
Dint
Aspr
Axrate

-0.024
0.003
-0.005
-0.010
0.148

-0.317
0.066
-0.097
-2.451
0.335

0.004
0.001
0.002
0.000
-0.019

1.401
0.277
0.845
1.085
-1.039

**

2
0.053
0.830
R
Panel A (Part 2): H0 - flows do not Granger cause market returns
-0.004
-1.455
-0.006
Intercept
-0.037
-0.943
0.515
Flowt-1
-0.015
-0.391
0.309
Flowt-2
0.041
1.259
-0.523
Flowt-3
0.013
0.179
0.041
Rett-1

*
***
***
***
***
*

Sales
Coefficient

t-Statistic

0.000
0.612
0.174
0.132
0.009

1.585
9.847
2.771
2.580
2.110

***
***
**
**

-0.007

-2.277

**

0.002
0.001
0.002
0.000
-0.012

0.698
0.414
0.811
1.336
-0.614

0.805
-1.770
0.456
0.267
-0.546
0.725

-0.006
0.910
0.553
-1.131
0.027

-1.823
0.842
0.509
-1.238
0.431

Rett-2

0.004

0.042

0.009

0.127

-0.009

-0.117

Rett-3
Ddiv
Dint
Aspr
Axrate

0.093
0.105
0.000
0.004
0.468

1.197
1.254
0.002
1.601
1.348

0.063
0.111
-0.007
0.004
0.510

0.934
1.391
-0.159
1.519
1.481

0.072
0.113
-0.007
0.004
0.499

1.085
1.415
-0.165
1.501
1.453

0.019

0.017

0.020

44

Panel B: Foreigners
NIF
Purchases
Variable
Coefficient
t-Statistic
Coefficient
t-Statistic
Panel B (Part 1) : H0 - stock market returns do not Granger cause flows
-0.060
-2.669
***
0.001
5.927
Intercept
0.261
6.326
***
0.336
6.007
Flowt-1
0.083
1.704
*
0.038
0.732
Flowt-2
0.018
0.455
0.117
2.563
Flowt-3
2.925
5.082
***
0.004
2.405
Rett-1

***
***
***
**

Sales
Coefficient

t-Statistic

0.000
0.461
0.095
0.102
-0.001

6.850
9.190
1.951
2.192
-1.192

Rett-2

-0.676

-1.431

-0.002

-1.104

0.001

1.340

Rett-3
Ddiv
Dint
Aspr
Axrate

-0.463
0.293
0.109
0.064
-2.619

-1.045
0.624
0.273
2.968
-1.028

-0.001
-0.002
0.001
0.000
0.004

-0.511
-0.701
1.159
-0.491
0.626

0.001
-0.001
0.000
0.000
0.001

1.027
-1.021
0.295
-2.835
0.181

***

2
0.254
0.249
R
Panel B (Part 2) : H0 - flows do not Granger cause market returns
-0.004
-1.576
-0.006
Intercept
0.007
1.286
3.760
Flowt-1
-0.007
-1.171
-1.794
Flowt-2
0.000
-0.046
-0.120
Flowt-3
0.030
0.483
0.030
Rett-1

-1.642
2.118
-0.791
-0.062
0.542

**

-0.008
0.570
1.073
1.339
0.053

-1.888
0.295
0.383
0.537
0.911

0.018

0.227

0.009

0.138

0.026

0.356

Rett-3
Ddiv
Dint
Aspr
Axrate

0.090
0.110
0.005
0.004
0.435

1.217
1.342
0.123
1.616
1.241

0.077
0.117
-0.002
0.004
0.461

1.151
1.439
-0.038
1.730
1.356

0.074
0.114
-0.006
0.005
0.461

1.026
1.355
-0.132
2.040
1.314

0.019

0.021

0.017

45

***

0.392

Rett-2

***
****
*
**

**

Panel C: Institutions
NIF
Purchases
Variable
Coefficient
t-Statistic
Coefficient
t-Statistic
Panel C (Part 1) : H0 - stock market returns do not Granger cause flows
0.013
0.623
0.000
3.985
Intercept
0.112
2.294
**
0.365
5.234
Flowt-1
0.142
3.035
***
0.174
4.327
Flowt-2
0.088
2.115
**
0.180
4.436
Flowt-3
-1.274
-3.304
***
0.000
0.004
Rett-1

***
***
***
***

Sales
Coefficient

t-Statistic

0.000
0.377
0.107
0.192
0.004

4.848
6.476
2.379
4.612
2.114

***
***
**
***
**

Rett-2

0.148

0.454

0.000

-0.122

-0.001

-0.638

Rett-3
Ddiv
Dint
Aspr
Axrate

-0.161
-0.171
-0.494
-0.018
-0.368

-0.443
-0.508
-1.630
-0.935
-0.155

-0.002
-0.001
0.001
0.000
-0.002

-1.535
-0.975
1.221
0.603
-0.340

-0.001
-0.001
0.002
0.000
0.000

-0.473
-0.690
1.729
0.923
0.015

-0.004
2.974
-6.044
3.058
0.050

-1.276
1.447
-2.136
1.214
0.875

**

2
0.081
0.393
R
Panel C (Part 2) : H0 - flows do not Granger cause market returns
-0.005
-1.826
*
-0.007
Intercept
0.006
0.937
4.603
Flowt-1
0.021
2.938
***
-0.024
Flowt-2
-0.006
-0.914
-1.593
Flowt-3
0.048
0.851
0.031
Rett-1

0.402
-2.066
2.164
-0.011
-0.807
0.564

**
**

Rett-2

0.015

0.215

0.013

0.198

0.023

0.325

Rett-3
Ddiv
Dint
Aspr
Axrate

0.095
0.118
0.015
0.005
0.425

1.357
1.437
0.382
1.926
1.240

0.066
0.117
-0.011
0.004
0.479

0.996
1.422
-0.239
1.575
1.403

0.083
0.109
0.004
0.004
0.458

1.247
1.304
0.090
1.601
1.335

0.037

0.022

0.027

46

Panel D: Securities
NIF
Purchases
Variable
Coefficient
t-Statistic
Coefficient
t-Statistic
Panel D (Part 1) : H0 - stock market returns do not Granger cause flows
-0.004
-0.115
0.000
5.978
Intercept
0.023
0.461
0.296
4.694
Flowt-1
0.093
2.227
**
0.124
2.751
Flowt-2
-0.026
-0.629
0.108
1.984
Flowt-3
-1.605
-2.161
**
0.000
-0.247
Rett-1
Rett-2

1.692

2.277

Rett-3
Ddiv
Dint
Aspr
Axrate

1.068
-0.347
-0.112
-0.020
0.123

1.384
-0.421
-0.202
-0.635
0.027

**

0.001

1.675

0.000
-0.001
0.001
0.000
0.000

0.300
-0.839
1.060
-0.125
-0.044

Sales
Coefficient

t-Statistic

***
***
***
**

0.000
0.227
0.133
0.150
0.001

4.969
6.549
3.295
3.531
1.461

-0.001

-1.611

-0.001
-0.001
0.001
0.000
0.002

-0.993
-1.097
1.425
0.214
0.679

2
0.061
0.249
R
Panel D (Part 2) : H0 - flows do not Granger cause market returns
-0.004
-1.562
-0.004
-1.346
Intercept
-0.005
-1.171
5.211
0.745
Flowt-1
0.000
-0.064
-0.227
-0.042
Flowt-2
-0.005
-1.287
-4.885
-0.865
Flowt-3
0.058
0.953
0.040
0.672
Rett-1

0.183
-0.008
7.312
-1.513
5.294
0.048

-2.019
1.306
-0.311
1.087
0.868

Rett-2

0.007

0.098

0.020

0.271

0.010

0.145

Rett-3
Ddiv
Dint
Aspr
Axrate

0.087
0.104
-0.001
0.004
0.536

1.277
1.235
-0.029
1.529
1.593

0.072
0.114
0.002
0.004
0.459

1.074
1.333
0.054
1.660
1.330

0.069
0.113
-0.015
0.004
0.544

1.047
1.410
-0.325
1.673
1.620

0.022

0.017

0.022

47

***
***
***
***

Table 6: Determinants of Flow


This table presents the regression analysis of the determinants of flows including the uncertainty variable as
an independent variable. The functional specification is Ft = + Unct + InfVt + Ft 1 + t . Ft represents NIF,
Purchases and Sales where Purchases and Sales are normalized flows and NIF is defined as
Purchase - Sales
Purchase + Sales

. Unct is an uncertainty variable proxied by the log of return squared [ln(Ret2)] - volatility in

Panel A. Ret is the market return. For Panel B, standardized open interest for the KOSPI 200 index futures
is used as a proxy for dispersion of beliefs. Ddiv is the dividend yield; Dint is the 90-day commercial paper
rate; Dspr is a proxy for the risk premium calculated by differencing the 3 year-government bond and
corporate bond rate; Xrate is the change in daily exchange rate (USD/Won). Ddiv, Dint and Dspr are
differenced to meet stationarity requirements. The figures are Newey-West heteroskedasticity and
autocorrelation adjusted. Observations are daily in frequency and the sample period is 1991-1993. *, **,
*** denotes significance at 10, 5 and 1 percent levels. All the coefficients are multiplied by a factor of 100
for ease of presentation.
Panel A: Uncertainty Variable - Volatility
NIF

Purchases

Coefficient

t-Statistic

Intercept

1.114

2.353

Unc

0.061

1.320

Xrate

-7.568

-0.240

Ddiv

34.595

4.681

Dint

-1.357

Dspr

3.347

Flowt-1

16.158

3.931

Flowt-2

5.862

Flowt-3

2.152

Sales

Coefficient

t-Statistic

Coefficient

t-Statistic

0.151

7.114

0.012

6.065

0.154

7.324

0.011

6.027

-1.734

-1.584

-0.627

-5.185

-1.169

-0.963

-1.010

-6.527

-0.448

0.139

0.905

-0.132

0.755

0.165

0.998

-1.054

-0.197

65.143

11.547

-1.546

67.560

13.225

1.375

11.590

2.038

8.264

1.547

0.601

15.635

3.756

15.255

4.032

***

0.053

7.634

0.002

3.300

***
***

Online

**

***

***

***
***
***

***
**
***

0.837

0.280

***
***
***

***

0.837

Foreigners
Intercept

2.342

0.763

0.078

7.641

Unc

0.203

0.636

0.003

3.066

Xrate

9.693

0.050

-0.534

-1.050

Ddiv

-114.185

-4.374

-0.258

-4.136

Dint

-4.540

-0.129

0.205

3.060

Dspr

-35.185

-1.430

Flowt-1

38.156

9.448

Flowt-2

1.605

Flowt-3

1.132

0.229

***

-0.091

-1.512

39.891

7.549

0.287

2.635

0.494

0.277

8.348

1.952

***

0.288

***
***
***
***
***
*

-0.312

-1.164

0.081

2.099

0.163

2.360

-0.055

-1.048

49.394

10.158

6.816

1.404

11.427

2.685

0.393

48

**
**
***
***

Panel A (Contd)
Institutions
Intercept

0.137

0.052

0.063

8.203

Unc

0.055

0.209

0.003

6.269

Xrate

176.697

1.459

-0.012

-0.036

Ddiv

-27.685

-1.474

-0.172

-2.956

Dint

-15.998

-0.743

0.147

2.400

Dspr

-5.277

-0.258

Flowt-1

8.574

1.800

Flowt-2

14.200

3.005

Flowt-3

7.568

1.787

*
***
*

-0.056

-0.969

37.031

5.305

18.758

5.053

14.685

4.295

***
***
***
**
***
***
***

0.448

0.056

0.068

9.105

0.004

5.596

-0.439

-1.249

-0.118

-2.571

0.188

2.017

-0.073

-1.110

43.494

7.420

9.188

2.012

15.778

3.582

***
***
**
**
***
***
***

0.415

Securities
Intercept

0.256

0.057

0.025

6.457

Unc

0.290

0.629

0.001

3.770

Xrate

456.260

1.967

0.004

0.027

Ddiv

-98.914

-2.066

-0.090

-2.183

Dint

-32.455

-0.672

0.056

1.636

Dspr

-49.543

-1.630

-0.056

-3.074

Flowt-1

0.205

0.046

30.688

4.929

Flowt-2

12.424

3.006

17.775

4.377

Flowt-3

-0.711

-0.177

8.458

1.464

0.056

**
**

***

0.300

***
***
**
***
***
***

0.023

7.052

0.001

2.855

-0.387

-2.442

-0.020

-1.296

0.064

2.197

-0.011

-0.475

23.239

7.040

12.638

3.173

14.347

3.436

0.176

49

***
***
**
**
***
***
***

Panel B: Uncertainty Variable KOSPI200 futures Open Interest


NIF
Coefficient

Purchases
t-Statistic

Coefficient

Sales
t-Statistic

Coefficient

t-Statistic

Online
Intercept

0.534

1.145

0.122

4.799

***

0.129

5.299

***

Unc

0.035

0.032

-0.187

-4.104

***

-0.191

-4.403

***

Xrate

-7.095

-0.228

-1.611

-1.432

-1.076

-0.869

Ddiv

34.648

4.692

-0.612

-5.121

Dint

-1.258

-0.417

0.189

1.072

Dspr

3.676

0.998

-0.073

-0.563

Flowt-1

15.755

3.829

61.369

11.188

***

65.062

12.671

Flowt-2

6.465

1.540

12.411

2.103

**

7.031

1.271

Flowt-3

1.874

0.523

17.811

3.907

***

18.305

4.356

***

7.246

***

-0.052

-3.841

***

-0.272

-1.013

***

***

0.278

***

0.830

-0.999

-6.315

0.216

1.409

-0.134

-0.984

***

***

0.832

Foreigners
Intercept

5.946

2.000

**

Unc
Xrate

-13.698

-1.898

**

9.025

0.047

Ddiv

-114.027

-4.412

Dint

-3.417

-0.097

Dspr

-34.428

-1.390

Flowt-1

37.558

9.162

36.886

7.015

47.485

9.710

Flowt-2

1.817

0.328

3.624

0.689

7.384

1.546

Flowt-3

0.487

0.121

6.705

1.586

11.412

2.650

**

***

***

0.234

0.095

7.363

***
***

0.058

-0.091

-4.376

-0.542

-1.068

-0.257

-4.114

***

0.083

2.073

**

0.222

3.736

***

0.173

2.601

***

-0.081

-1.402

-0.045

-0.886

***

0.304

***

0.396

Institutions
Intercept

-2.286

-1.045

0.060

5.814

***

0.067

6.776

***

Unc

4.752

0.901

-0.070

-5.043

***

-0.076

-4.314

***

Xrate

175.942

1.437

0.021

0.063

-0.440

-1.164

Ddiv

-27.697

-1.472

-0.170

-2.733

***

-0.118

-2.420

**

Dint

-16.197

-0.760

0.164

2.851

***

0.210

2.348

**

Dspr

-5.007

-0.247

-0.038

-0.656

-0.057

-0.897

Flowt-1

8.480

1.802

34.437

5.377

***

38.190

6.560

***

Flowt-2

13.941

2.960

***

18.023

4.915

***

12.717

2.847

***

Flowt-3

7.095

1.677

16.497

4.728

***

14.125

3.231

***

0.057

0.440

0.408

Securities
Intercept

-2.641

-0.595

0.028

8.023

***

0.037

7.822

***

Unc

0.483

0.045

-0.034

-4.839

***

-0.046

-4.913

***

Xrate

458.778

1.975

**

0.015

0.100

-0.369

-2.421

**

Ddiv

-98.596

-2.042

**

-0.089

-2.074

-0.020

-1.392

Dint

-31.835

-0.662

0.064

2.068

**

0.075

2.603

Dspr

-48.086

-1.611

-0.051

-2.892

***

-0.010

-0.465

Flowt-1

0.392

0.088

Flowt-2

12.322

2.990

Flowt-3

-0.725

-0.179

0.055

**

**

***

27.559

4.447

***

19.474

5.979

***

17.078

4.222

***

11.598

2.887

***

8.727

1.520

11.121

2.948

***

0.308

0.223

50

Table 7: Market Performance after Heavy Buying and Selling


This table reports the performance of different investor types at extreme ends of NIF. NIF is defined as

Purchases Sales
Purchases + Sales

for each investor-group is sorted into

five equal sets. Periods of the highest positive NIF are designated the buying weeks, and those with the largest negative NIF the selling weeks. Each quintile of
buy or sell weeks has 100 observations. One day, one week and one month returns are computed following the heavy buy or sell trading day. The mean NIF for
heavy buying and selling days are reported in the table. The t-statistic tests the differences in mean between current and post trading returns and are reported in
italic.

NIF
Online

1 day

Sell

Buy

Sell

Buy

Sell

-0.037

0.054

0.004

0.001

-0.003

-0.001

-0.013

-0.260

0.306

Securities

-0.214
-0.395

0.192
0.330

***

7.644

***

3.610

0.000
3.872

Institutions

1 month

Buy

5.939
Foreigners

1 week

Sell

***

4.989

***

3.915

0.003

***

3.805

***

2.618

0.007

***

3.608

***

1.985

0.000

Buy
-0.009
***

1.051

0.185

0.007

0.010

-0.002

0.002

-0.004

0.007

-0.005

-0.004

0.033

0.229

0.452

0.996

0.304

0.724

0.001
2.680

-0.003
***

*, **, *** denotes significance at 10, 5 and 1 per cent level

3.119

0.009
***

2.909

-0.009
***

3.314

0.011
***

1.791

0.002
*

0.578

Table 8: Market Timing by Different Investors


This table shows daily and weekly relation between NIF (defined as

Purchases Sales
Purchases + Sales

) for each investor type and lead/lag market returns. The functional

specification is NIF = + i Ret t + i + .. Ret is the daily return of KOSPI index for panel A and weekly returns for panel B. Daily and weekly number of
5

observations are 520 and 114 respectively. The figures are Newey-West heteroskedasticity and autocorrelation adjusted.

Panel A: Daily
Online
Coefficient

t-Statistic

Intercept
Rett-5
Rett-4

0.007
-0.005
-0.062

5.663
-0.105
-1.251

Rett-3

-0.052

-0.871

Rett-2

-0.176

-3.113

Rett-1

-0.197

-4.691

Ret
Rett+1
Rett+2

-0.923
-0.059
-0.018

-8.418
-1.130
-0.354

Rett+3

0.046

Rett+4
Rett+5
2

Foreigners

Institutions
Coefficient

t-Statistic

Intercept
Rett-5
Rett-4

-0.005
-0.349
-0.135

-0.594
-1.157
-0.469

Rett-3

-0.153

-0.513

**

Rett-2

0.306

1.135

***

Rett-1

-1.039

-3.032

***

***

Ret
Rett+1
Rett+2

0.917
0.378
0.925

2.200
1.250
2.998

**

0.081

Rett+3

-0.133

-0.451

-0.105

-0.297

Rett+4

-0.045

-0.136

0.113

0.380

Rett+5

0.188

0.631

Coefficient

t-Statistic

Intercept
Rett-5
Rett-4

0.010
0.626
0.395

1.044
1.696
1.165

Rett-3

0.045

0.148

***

Rett-2

0.648

2.153

***

Rett-1

3.792

8.403

***

Ret
Rett+1
Rett+2

3.380
0.333
-0.246

9.767
0.983
-0.739

0.980

Rett+3

0.027

0.041

0.870

Rett+4

-0.038

-0.799

Rett+5

0.399

***

0.300

52

0.065

Securities

***

Coefficient

t-Statistic

Intercept
Rett-5
Rett-4

-0.027
-0.479
0.261

-2.330
-0.848
0.599

**

Rett-3

1.003

1.758

Rett-2

2.058

3.753

***

Rett-1

-1.521

-2.344

**

Ret
Rett+1
Rett+2

2.395
-0.645
-0.096

2.271
-1.128
-0.196

**

Rett+3

-0.816

-1.547

Rett+4

-0.748

-1.472

Rett+5

0.175

0.344

0.097

Panel B: Weekly
Online
Coefficient
C
0.006
Rett-5
-0.011
Rett-4
0.066

t-Statistic
4.397
-0.342
1.915

Rett-3

0.014

0.582

Rett-3

-0.036

Rett-2

-0.007

-0.293

Rett-2

0.078

Rett-1

-0.023

-0.767

Rett-1

0.415

1.659

Ret
Rett+1
Rett+2

-0.308
-0.005
0.037

-6.698
-0.216
1.724

Ret
Rett+1
Rett+2

1.628
-0.002
-0.146

5.524
-0.008
-0.753

Rett+3

-0.022

-0.813

Rett+3

0.216

Rett+4

-0.009

-0.293

Rett+4

Rett+5

-0.016

-0.467

Rett+5

0.541

***

***

Foreigners
Coefficient
C
0.018
Rett-5
-0.044
Rett-4
-0.382

Institutions
C
Rett-5
Rett-4

Coefficient
-0.007
0.100
0.208

t-Statistic
-0.698
0.450
0.983

Securities
Coefficient
C
-0.020
Rett-5
0.247
Rett-4
0.153

t-Statistic
-1.612
0.589
0.502

-0.158

Rett-3

0.163

0.801

Rett-3

0.198

0.585

0.311

Rett-2

-0.047

-0.260

Rett-2

0.025

0.109

Rett-1

-0.105

-0.465

Rett-1

-0.110

-0.334

Ret
Rett+1
Rett+2

0.248
0.171
-0.056

1.232
0.789
-0.360

Ret
Rett+1
Rett+2

0.596
-0.174
-0.465

1.260
-0.661
-1.716

1.074

Rett+3

0.098

0.497

Rett+3

0.497

1.670

0.001

0.003

Rett+4

-0.227

-1.050

Rett+4

0.629

2.131

0.426

1.885

Rett+5

0.015

0.056

Rett+5

-0.425

-1.323

t-Statistic
1.306
-0.173
-1.598

***

0.363

53

0.054

0.176

*
**

Fig 1: Comparative Size of the Online Trading Value Compared to Total Value Traded in both Korean Stock Exchange and KOSDAQ.
(Source: KSDA)

Total Stock Value (KSE and KOSDAQ)


400000000

350000000

250000000
Total Value
Non-online

200000000

150000000

100000000

50000000

Date

54

2002 Nov

2002 Sep

2002 Jul

2002 May

2002 Mar

2002 Jan

2001 Nov

2001 Sep

2001 Jul

2001 May

2001 Mar

2001 Jan

2000 Nov

2000 Sep

2000 Jul

2000 May

2000 Mar

2000 Jan

1999 Nov

1999 Sep

1999 Jul

1999 May

1999 Mar

1999 Jan

1998 Nov

1998 Sep

1998 Jul

1998 May

1998 Mar

0
1998 Jan

Trading Value (mil. Won)

300000000

cumulative Performance

-0.05

-0.15

-0.2

-0.25
Dates

55

-0.1

Securities

20030214

20030127

20030108

20021216

20021127

Institutions
20021108

20021022

20021002

20020912

20020826

20020806

20020718

20020627

20020607

20020520

20020430

20020411

20020322

20020305

20020208

20020122

20020103

20011212

20011123

20011106

20011018

20010926

20010907

20010821

0.15

20010801

20010712

20010625

20010605

20010517

20010427

20010410

20010321

20010302

20010212

20010119

20010102

Fig 2 : Daily Cumulative Performance


0.2

Foreigners

0.1

0.05
Online

-0.05

-0.1
Dates

56
Online

02-Feb-03

04-Jan-03

01-Jan-03

02-Dec-02

03-Nov-02

05-Oct-02

02-Oct-02

03-Sep-02

04-Aug-02

01-Aug-02

03-Jul-02

04-Jun-02

01-Jun-02

03-May-02

04-Apr-02

01-Apr-02

02-Mar-02

03-Feb-02

05-Jan-02

02-Jan-02

03-Dec-01

04-Nov-01

01-Nov-01

03-Oct-01

04-Sep-01

01-Sep-01

03-Aug-01

04-Jul-01

01-Jul-01

02-Jun-01

03-May-01

04-Apr-01

01-Apr-01

03-mar-01

05-Feb-01

02-Feb-01

04-Jan-01

01-Jan-01

Cumulative Performance

Fig 3: Weekly Cumulative Performance


0.4

0.35

0.3

0.25
Foreigners

0.2

0.15
Securities

0.1

0.05
Institutions