- Email: [email protected]

PII:

S1042-444X(19)30049-0

DOI:

https://doi.org/10.1016/j.mulﬁn.2019.100591

Reference:

MULFIN 100591

To appear in:

Journal of Multinational Financial Management

Received Date:

2 March 2019

Revised Date:

6 September 2019

Accepted Date:

9 October 2019

Please cite this article as: Tsai L-Ju, Shu P-Gi, Chiang S-Jane, Foreign investors’ trading behavior and market conditions: Evidence from Taiwan, Journal of Multinational Financial Management (2019), doi: https://doi.org/10.1016/j.mulﬁn.2019.100591

This is a PDF ﬁle of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the deﬁnitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its ﬁnal form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. © 2019 Published by Elsevier.

Foreign investors’ trading behavior and market conditions: Evidence from Taiwan

Li-Ju Tsaia, Pei-Gi Shub, and Sue-Jane Chiangc*1

[email protected],

Department of Finance and International Business, Fu Jen Catholic

[email protected],

ro of

University, New Taipei City, Taiwan

Graduate Institute of Business Administration, Fu Jen Catholic University,

New Taipei City, Taiwan [email protected],

Department of Business Administration, Fu Jen Catholic University, New

author

re

*Corresponding

-p

Taipei City, Taiwan

1

We thank Chuan-Yang Hwang (Discussant and Session Chair) and seminar participants at the 21th

Conference on the Theories and Practices of Securities and Financial Markets. This paper is previously

lP

titled as “Market sentiment and bounded rationality of foreign investors: the evidence of Taiwan.” Corresponding author: Sue-Jane Chiang, Department of Business Administration, Fu Jen Catholic University, No. 510, ZhongZheng Rd. Xinzhuang, Dist., New Taipei City 24205, Taiwan. Tel: +886-2-

We analyze foreign investors’ trading behavior under different market conditions.

Jo

ur

Highlights

na

2905-3935, Fax: +886-2-2908-9219. E-mail addresses: [email protected] (S-J. Chiang)

Two proxies for market conditions are stock market turnover and market return.

Foreign investors facilitate price discovery when the stock market is hot.

However, foreign investors become market followers when the market is cold.

1

Abstract We investigate the trading behavior of foreign investors and how it related to stock market price under different market conditions in Taiwan. Specifically, we focus on two surrogates of market conditions: stock market turnover and return. Applying Hansen and Seo’s (2002) threshold cointegration model to avoid arbitrarily chosen

ro of

cutoff values for market conditions, we find that foreign investors facilitate the price discovery function when the market is hot (e.g., high stock market turnover and/or high

market return). In contrast, they become market followers when the market is cold (e.g.,

re

-p

low stock market turnover and/or low market return).

lP

Keywords: Market Conditions; Foreign Investors; Threshold Cointegration.

na

JEL classification: C32, G15, G23

1. Introduction

ur

Whether the introduction of foreign investors into local stock markets enhances

Jo

market efficiency has been a debatable issue. Classical finance theory assumes that the market price is determined primarily by the behavior of rational investors, in the sense that the proportion of irrational traders is offset by another proportion of arbitragers (e.g., Baker et al., 2004; Baker and Wurgler, 2006). Foreign investors with their professional ability and information advantage tend to be deemed as rational in the 2

sense that their trading predicts stock returns.1 A recent study of Choi et al. (2017) and Onishchenko and Ülkü (2019) indicates that foreign institutional investors are sophisticated and have knowledge advantages through selection and targeting research. Nevertheless, the experiences from 1998 financial crises indicate that the introduction of foreign investors does not necessarily help host economies, 2 especially if they

ro of

engage in trade herding.3 De Long et al. (1990) show that rational speculators follow positive feedback strategies as to move stock prices away from their fundamental values.

-p

Ikizlerli et al. (2019) find that the sophisticated investors act in a way consistent with

momentum trading. Still, some studies indicate that foreign investors behave in a biased

re

manner.4 If their trading behavior leads the market, foreign investors facilitate the price

lP

discovery function of local markets. In contrast, if they are market followers, their

na

trading behavior increases market volatility and destabilizes local markets.

Whether foreign investors are stabilizers or destabilizers to local markets is the

See Grinblatt and Keloharju (2000), Seasholes (2000), Froot and Ramadorai (2008), and Ferreira et al.

Jo

1

ur

focal issue to be investigated in this study. Our hypothesis is to examine the issue of

(2017).

2

Some studies show that foreign investors do not have a destabilizing effect on stock prices (e.g., Choe,

et al., 1999; Froot et al. 2001; Kim and Wei, 2002). However, Jo (2002) indicates that foreign investors cause higher volatility in the stock market than domestic investors.

3

See Brennan and Cao (1997), Kim and Singal (2000), Dvořák (2005), Bae et al. (2006), and Jeon and Moffett (2010).

4

Griffin and Tversky (1992) argue that professionals may be even more overconfident than amateurs. Bowe and Domuta (2004) find that foreigners herd more than locals, and foreign herding hastens the onset of a crisis. 3

whether the relationship between the trading behavior of foreign investors and stock prices could be nonlinear and be subject to market conditions. Moreover, prior studies fail to reach consensus of whether foreign investors’ trading leads or lags the stock market. We postulate that the relation could be subject to certain conditions and therefore nonlinear. Therefore, we explore the possible nonlinear dynamic relation

threshold effect.

ro of

between foreign investors’ trading and stock market returns, and the existence of We use two market conditions to explore the possible non-linear

-p

relationship between the trading behavior of foreign investors and local stock prices: stock market turnover and return. This is inspired by a battery of prior studies 5

re

indicating that foreign investors may behave differently when the market is stable and

lP

when it is in turmoil.

na

The first market condition is daily relative stock market turnover. This is chosen because high turnover implies the strength of the market sentiment, high risk of

For example, Choe et al. (1999) find that the relationship changes when the market is in turmoil, as it

Jo

5

ur

uncertainty, or overconfidence.6 The second proxy is market return, which is defined

was during the South Asia financial crises. Ülkü (2015) finds that foreigners engage in negativefeedback trading only when local market returns are positive but not negative. Chuang and Susmel (2011) find that foreign institutional investors are overconfident in bull markets, and Baker and Stein (2004) find that the market is less rational when investor sentiment is bullish. That is, foreign investors’ trading behavior may lead the market, and they may facilitate the price discovery function of local markets if they have insight into the market or employ advanced analytic techniques. By contrast, they may be market followers under other market conditions.

6

See Lo and Wang (2000), Baker and Stein (2004), Richards (2005), Baker and Wurgler (2006), Ogunmuyiwa (2010), and Stambaugh et al. (2012). Odean (1999) and Barber and Odean (2001) find 4

as the relative trade price. A high (low) market return implies a bull (bear) market.7 The selection of the two variables is primarily due to the fact that the two variables are the most accessible and widely accepted attributes to categorize market conditions. Furthermore, other variables have been tested but failed to yield the threshold effect 8. Using the two proxy variables of market conditions, we explore the possibility of a

return.

ro of

nonlinear relationship between foreign investors’ trading behavior and the stock market The two proxy variables are used to capture foreign investors’ bounded

re

within and beyond certain market conditions.

-p

rationality in the sense that they may behave and impact differently on the market return

By the upsurge in behavioral finance, investors’ buying and selling behavior may

lP

In other words, the net trading volume may not capture the true

na

not be symmetric.9

ur

that investors’ overconfidence results in excessive trading and poor performance. Hence, data on excessive trading may contain information about overconfidence. 7

Kim and Zumwalt (1979), Chen (1982), and Gombola and Liu (1993) use the risk-free rate as the

Jo

benchmark. If the stock index return is greater (less) than the risk-free rate during the same period, that period is considered to be a bull (bear) market. Other variables that have been tested include foreign investors’ trade volume, foreign investors’

8

turnover ratio, foreign investors’ trade ratio, overall market trade volume, and the stock price index.

The relevant literature on investor’s asymmetric behavior includes discussion of the prospect theory

9

developed by Kahneman and Tversky (1979) and Shefrin and Statman’s (1985) proposal of the disposition effect as an extension of the prospect theory. The prospect theory proposes that people make decisions based on their potential value of losses and gains, and that people evaluate these losses and gains using interesting heuristics. Shefrin and Statman (1985) propose that people tend to sell winners prematurely and hold losers at a delayed basis. 5

relationship between foreign investors’ trading behavior and the stock market index.10 For example, Chan and Lakonishok (1993) find that institutional investors’ buys and sells impact prices asymmetrically. Under this assumption of the asymmetric buying and selling behavior associated with foreign investors, and the stylized fact that stock prices and foreign investors’ buys/sells are non-stationary and determined

ro of

simultaneously in a nonlinearly dynamic relation, we use market turnover ratio and stock index return as the threshold variables and adopt the threshold cointegration

-p

model that simultaneously considers the foreign investors’ buys, sells, and stock prices. The model aims at the exploration of the possible asymmetric relation between foreign

re

investors’ buys, foreign investors’ sells, and stock market index.

lP

The choice of the Taiwan stock market as our investigated platform is briefly

na

elucidated as follows. Firstly, as the Taiwan stock market has gradually opened to foreign investors, foreign investors play a pivotal role in the market. During our sample

ur

period, foreign investors comprise more than 30.62% of total market trading. Their

Jo

daily buys and sales amount to over NT$ 10 billion11. As foreign investors become

10

The results from Panel B of Table 2 indicate that the patterns for foreign investors’ buys, sells, and

net trades are different. Foreign buys and sells increase with the passage of time, while foreign net trades fluctuate.

11

Foreign investors were firstly inducted into Taiwan market in 1983. The system of qualified foreign institutional investors (QFII) was in effect in 1991 and abolished in 2003. Right after then, the reviewing process of foreign investors was changed from a permit system to a registration system. Foreign investors comprise more and more of the total market trading as the passage of time. See Table 1 for detailed statistics. 6

more and more important as the passage of time, their trading behavior has been a noted issue. Second, the Taiwan stock market is characterized as a shallow-dish market which is comprised of a high portion of individual investors and high turnover rate. The nature of this market makes it possible to depict how foreign investors affect an emerging stock market. Finally, foreign investors are restricted to engage in short selling. The

ro of

short-restriction rule implies that foreign investors’ buys and sells may impact differently on the stock market returns. All of these make Taiwan as an ideal forum for

-p

the investigated issue.

re

Our empirical results are briefly summarized as following. First, we illustrate the existence of the threshold effect for the two market condition variables. That is, the

lP

relation between the foreign investors trading and local stock market indexes is different

na

under different market conditions. Second, the trading behavior of foreign investors has a bi-direction causal relationship with the stock index return under the condition of high

ur

market turnover and/or return. In contrast, the foreign investors become market

Jo

followers when market turnover and/or return is low: their purchases (sales) increase with an increase (a decrease) in the stock index returns. Even though foreign investors predict the stock market returns when the market is hot, their trading behavior is also affected by the stock market returns. When market turnover and/or market returns are high, foreign investors increase their purchase of shares after noticing an increase in the 7

stock market returns. This finding that foreign investors aggressively trade in a highturnover and/or a high-returns market is consistent with the finding of Chuang and Susmel (2011) that foreign institutional investors trade more aggressively following market gains in a bull market.

In general terms, the potential contributions of this study can be summarized as

ro of

follows. First, we illustrate the nonlinear and asymmetric behavior pattern of foreign

investors. The relation between foreign investors’ trading and local stock market

-p

indexes is different under different market conditions: foreign investors facilitate the

re

price discovery when the market is hot and become market followers when the market

lP

is cold.

Second, the extended threshold cointegration model of Hansen and Seo (2002) was

na

adopted to avoid arbitrarily chosen cutoff values for market conditions. The model could estimate an optimal threshold value instead of arbitrarily chosen cutoff values,

ur

and could capture non-stationary nature of financial variables and the possibly

Jo

nonlinear dynamic behavior across regimes in a cointegration model. Specifically, our estimated threshold turnover ratio differs significantly from the mean and median turnover ratios12 or an unobservable state of bull or bear market. We note that our

12

The mean and median of the turnover ratio are 0.628% and 0.534% respectively. The threshold value is 0.6850% in our empirical study. The mean and median of the stock index return are 0.014% and 0.041% respectively. The corresponding threshold value is 0.8701% in our empirical study. 8

model is superior to the two-stage Markov switching model 13 in term of the improvement on AIC of Johansen’s (1988) model. Moreover, from the practitioners’ perspective, our threshold estimate such as 0.6850% of the previous day’s turnover ratio provides a more convenient and intuitive information than the unobservable state of Markov switch.

ro of

Third, in contrast with the related study of Chiang et al. (2012), who uses the errorcorrection term as the only threshold variable, we consider both market turnover and

-p

stock index return as the threshold variables. The use of alternative threshold variables

re

is associated with the benefit of a better depiction of market conditions on one hand and a more sustainable empirical result on the other. Finally, our investigation covers a long

lP

range of sampling period from 1995 to 2018. This long coverage is especially well

na

suited to cointegration model because it implies long-term co-movements of variables. Moreover, the long coverage of sampling period is likely to detect changes in

ur

relationships between variables.

Jo

The rest of this paper is organized as follows. In Section 2 we describe our methodology. In Section 3 we describe the sources of our data and present the empirical

13

Markov switching model has been used extensively to identify the business cycle or bull and bear of financial market (see, e.g., Hamilton 2011, Nikolsko-Rzhevskyy et al., 2012 and Maheu et al., 2000). 9

results, and we discuss the relationship between foreigner investors’ behavior and the Taiwan stock market index. In Section 4 we present our conclusions.

2. Methodology The threshold regression model initially developed by Tong (1978), Tong and Lim

ro of

(1980) and Tong (1983) does not provide for a clear identification of threshold values. Tsay (1989) proposes the autoregressive threshold model as a way to estimate the

parameters of the threshold value, and the model has attracted much academic attention.

-p

Hansen (1997, 1999, 2000) develops the asymptotic distribution theory for threshold

re

estimation, which allows the threshold model to be used to determine whether the

lP

threshold effect of the nonlinear model is statistically significant. Threshold VAR models have become the focal point of many studies for which it is necessary to depict

na

nonlinear dynamic relation for multivariate time series (e.g., Potter, 1995; Shen and

ur

Chiang, 1999; Altissimo and Violante, 2001).

If the individual variables are associated with a unit root in a multivariate

Jo

threshold model, the issue of threshold cointegration needs to be further addressed. Hansen and Seo (2002) propose a threshold cointegration model in which the relationships between cointegrated variables are adjusted for different threshold values of the error-correction terms. Since many financial variables are commonly found to

10

have a unit root and to be cointegrated, the threshold cointegration model is appropriate for examining the threshold effect. Thus, the preliminary unit root test and the cointegration test are needed prior to the application of the threshold cointegration model. The approach proposed by Johansen (1988) is commonly used for testing the cointegration effect. However, Balke and Fomby (1997) argue that if the model is a

ro of

threshold model rather than a linear model, the power of the cointegration test is greater using Engle and Granger’s (1987) approach than Johansen’s (1988) approach. In this

-p

study we use both approaches as a robustness check for the existence of cointegration.

Hansen and Seo’s (2002) approach is then used to determine if there is a threshold effect

re

and if so determine the threshold value, as well as to further examine the causality for

lP

multivariate time series. Although we follow Hansen and Seo (2002) in using the

na

error-correction term as a threshold variable,14 our threshold variables, which include the stock turnover ratio and stock index return on the previous day, are stationary and

ur

have no unit root. Thus our empirical model meets Hansen’s (1996, 2000) requirement

Jo

that the threshold variable is stationary. Statistical inferences can still be made, using the bootstrapping procedures of Hansen and Seo (2002).

14

Our empirical analyses find that the threshold effect for error-correction term is not significant. Moreover, using error-correction term as the threshold variable has no important financial implications. 11

Assume that the cointegration model includes the three-component variable

xt {x1t , x2t , x3t } , where x1t represents the foreign investor’s buys, x2t represents the foreign investor’s sells, and x 3t represents the stock price index. If x1t , x2t and

x 3t follow the stochastic process with a unit root, and they are cointegrated with errorcorrection term t 1 () x t , where is the cointegration vector. The cointegration

ro of

model can be expressed as follows:

AX ( ) ut , if z t-1 xt 1 t 1 , A2 X t 1 ( ) ut , if z t-1

(1)

,

zt 1 is the threshold variable (error-

-p

where X t 1 ( ) 1, t , t 1 ( ), xt' 1 , ... , xt' k

re

correction term ωt--1 , stock turnover ratio or stock index return), and A1 and A2 denote alternative coefficient matrices. 15 From equation (1) we know that the coefficient

lP

matrix is A2 if the threshold variable zt 1 is greater than the threshold value , and

A1 if zt 1 is not greater than the threshold value τ. If A1 significantly differs from

na

A2 , the threshold effect for the cointegration model exists, which implies that that the estimate of coefficient matrix will be biased if one employs the conventional error-

ur

correction model of Johansen (1988).

Jo

First, following Hansen and Seo (2002) we estimate the cointegration vector from the cointegration model. Then, to obtain estimates of the cointegration vector ( ) and the threshold value ( ) we conduct a grid search to find the highest value of maximum

15

As shown in Table 1, foreign investors’ trade volume significantly increases with the passage of time. We therefore include time trend t in formula (1). 12

likelihood from all the possible combinations of the various cointegration vectors and threshold values.16 A further test of the existence of threshold effect is still needed, for which the null hypothesis is:

H 0 : A1 A2 .

(2)

There are different cointegration vectors ( ) and thresholds ( ) for different Lagrange

ro of

multipliers (LMs). If SupLM, a statistic introduced by Hansen and Seo (2002), is greater than its critical value, the null hypothesis A1 A2 is rejected, that is, the threshold

effect represented by equation (1) is significant.17 In contrast, the null hypothesis is not

-p

rejected if SupLM is equal to or smaller than the critical value. The critical value of

re

SupLM is obtained by bootstrapping.18

lP

3. Empirical results 3.1. Data

na

The data sample is taken from the Taiwan Economic Journal (TEJ), published by a data company in Taiwan, and covers the period from January 5, 1995 to December 28,

Jo

ur

2018. The data consist of the daily Taiwan stock price index (SP), the stock market

16

The cointegration vector is then divided into 500 possible values within six standard deviations above

and below Johansen’s estimated cointegration vector. Five hundred possible threshold values ( )

17

within the range of 5% and 95% of the values of the threshold variable are arranged in ascending order. ~ 𝑆𝑢𝑝𝐿𝑀 = 𝑠𝑢𝑝 𝐿𝑀(𝛽̃ , 𝜏), where LM ( , ) is the heteroskedasticity-robust LM-like statistic as 𝜏𝐿 ≤𝜏≤𝜏𝑈

defined by Hansen and Seo (2002), 𝛽̃ is the estimation of the cointegration vector, and(𝜏𝐿 , 𝜏𝑈 ) are the upper and lower bounds of the threshold value (τ). 18

Hansen and Seo (2002) use the “fixed regressor bootstrap” and “parametric residual bootstrap” methods to calculate the critical values. 13

turnover ratio (TR), the stock market trading volume, the volume of foreigner investors’ purchases (FP), the volume of their sales (FS), the volume of their total trades (FP + FS), the volume of their net buy (FP - FS), and their trading ratio (total trades / stock market trading volume). All variables except the turnover ratio are log transformed. The stock index return is the rate of change in the stock price index ( SP ) that is defined

ro of

as the change in the log-transformed stock price index.

3.2. Summary statistics

-p

The summary statistics in Tables 1 and 2 indicate that the average daily volume of

re

market trades is 96.1 billion and that there was an increase from NT$ 35.99 billion in 1995 to NT$ 135.71 billion in 2007. After 2007, the trading volume dropped slightly to

lP

NT$79.7 billion in 2013 and then gradually increased to NT$ 130.21 billion in 2018.

na

The averages of the stock price index, turnover ratio, and index return are 7,420, 0.628, and 0.014 respectively. None of these reveal an obvious pattern over time. In 16 of the

ur

24 sampled years the index returns are positive. The stock price index was highest in

Jo

2018 (10,620) and lowest in 2001 (4,907). The foreigner investors’ trade ratio significantly increased during the sampling period and reached the highest ratio of 60.28% in 2016. Furthermore, their buys increased from NT$0.56 billion in 1995 to NT$33.5 billion in 2018, and their sells increased from NT$0.4 billion in 1995 to

14

NT$35.0 billion in 2018. Foreign investors in Taiwan are net buyers in all the sampled years except 1997, 2008, 2011, and 2018. <

3.3. Empirical results 3.3.1 Unit-root test

ro of

We first test each variable of interest to see if it is a stationary series. Specifically, we use the KPSS test, proposed by Kwiatkowski et al. (1992), to test the null hypothesis

-p

of stationarity or trend-stationarity against an alternative series with a unit root. The

re

KPSS statistics are for stationarity and for trend-stationarity. 19 The results summarized in Panel A of Table 3 indicate that the null hypothesis of stationarity is

lP

rejected at the 5% significance level for the Taiwan stock index, foreign investors’ buys,

na

and their sells, with corresponding of 4.127, 8.956 and 8.754, and of 0.835, 0.683 and 0.514. 20 However, after taking the first difference of the data, the null

ur

hypothesis of stationarity is not rejected. The corresponding values of are 0.047,

Jo

0.034 and 0.054, and the corresponding values of s are 0.026, 0.029 and 0.046 for the first difference of the stock price index, foreign investor’s buys, and their sells,

19

Traditional unit-root tests, such as those of Dickey and Fuller (1979, 1981), lack power in the sense that the null hypothesis is hard to be rejected in the vicinity of the unit circle. Moreover, if the series is not a pure autoregressive series, the unit test of Dickey and Fuller (1979, 1981) can lead to a wrong conclusion because of the inappropriate inclusion of a lag term. This is why we adopt the stationarity test of Kwiatkowski et al. (1992).

20

The critical values for and at the 5% significance level are 0.463 and 0.146 respectively. 15

respectively. This implies that these variables are a series with a unit root and integrated order one (I(1)).

We also examine the stationarity for the threshold variables stock market turnover ratio and stock index return. The results summarized in Panel B of Table 3 indicate that the null hypothesis of stationarity is not rejected for either variable, implying that both

ro of

are stationary. <

-p

3.3.2. Cointegration test

re

After confirming that the stock market index and foreign investors’ buys and sells are I(1), we examine whether they are cointegrated. We use Engle and Granger’s (1987)

lP

test in addition to Johansen’s (1988) test as a robustness check, because Balke and

na

Fomby (1997) note that Engle and Granger’s (1987) test has greater power than Johansen’s (1988) test for assessing threshold cointegration. 21 The results for

ur

Johansen’s test, summarized in Panel A of Table 4, indicate that the stock market index

Jo

and foreign investors’ buys and sells are all cointegrated. The cointegration among these variables is further supported by Engle and Granger’s test, the results of which are summarized in Panel B of Table 4. <

21

Engle and Granger’s (1987) approach can be used for one set of cointegration relationships only. 16

3.3.3. Test of the threshold effect We use the SupLM test proposed by Hansen and Seo (2002) to investigate whether the relationship between foreign investors’ trading behavior and the stock price index differs under different market conditions. The threshold variables we use are stock market turnover ratio and stock index return. We also consider the error-correction term

ro of

as a threshold variable because Hansen and Seo (2002) and Chiang et al. (2012) use it as the threshold variable in their studies. The results summarized in Table 5 indicate

-p

that the null hypothesis of no threshold effect is significantly rejected for stock market turnover and stock index return but not for the error-correction term.

re

<

lP

3.3.4. Estimation using the threshold cointegration model

na

Since the threshold effect is saliently found for stock market turnover ratio and stock index return, we continue to use them as threshold variables to estimate the

ur

threshold cointegration model in the following empirical analysis.

Jo

3.3.4.1 Using stock market turnover ratio as the threshold variable The estimation results using stock market turnover ratio on the previous day (TRt-

1)

as the threshold variable are summarized in Table 6. The time series of stock market

turnover ratio and the estimated threshold value 0.6850% are depicted in Figure 1. The result in Table 6 confirms the existence of the threshold effect and shows that the 17

estimated coefficients are quite different depending on whether the stock market turnover ratio is larger or smaller than the threshold value. Condition I, which is defined as the turnover ratio being lower than 0.6850%, is depicted in the upper part of Table 6. It comprises 66.8% of the sample. Condition II, which is defined as the turnover ratio being higher than 0.6850%, is illustrated in the lower part of Table 6. It comprises

ro of

33.2% of the sample.

When the market is in Condition I (low turnover; upper part of Table 6), foreigner

-p

investors’ buys (∆𝐹𝑃𝑡 ) are not only negatively affected by their buys and sells on the

re

previous day, but also positively affected by the lag-one error-correction term. Foreign investors’ sells (∆𝐹𝑆𝑡 ) are also negatively affected by the error-correction term, their

lP

buys and sells on the previous day, and the stock price index on the previous day. The

na

stock price index ( ∆𝑆𝑃𝑡 ) is negatively affected by the error-correction term and positively affected by the lag-one stock price index. However, we do not find that the

ur

foreign investors’ buys and sells on the previous day affect stock price index, implying

Jo

that foreign investors’ trading behavior do not lead the stock index return when the market turnover ratio is low. In contrast, the stock price index has leading effect on foreign investors’ sells; that is, foreign investors will increase their sells after noticing a decrease in the stock price index.

18

When the market is in Condition II (high turnover; lower part of Table 6), foreign investors’ buys (∆𝐹𝑃𝑡 ) are not only negatively affected by their lag-one buys and lagone sells, but also positively affected by the error-correction term and the lag-one stock price index. Moreover, foreign investors’ sells (∆𝐹𝑆𝑡 ) are negatively affected by the error-correction term and lag-one foreign investors’ sells. The stock price index (∆𝑆𝑃𝑡 )

ro of

is negatively affected by the error-correction term and lag-one foreign investors’ sells, and positively affected by lag-one foreign investors’ buys. These results indicate that

-p

there is bi-directional causality between the foreign investors' trading behavior and the stock price index when market turnover is high. That is, they increase their buys after

re

noticing an increase in the stock price index, and their purchases (sales) on the previous

lP

day further result in an increase (a decrease) in the stock price index on the following

na

day.

<

ur

The aforementioned results illustrate that the stock price index and foreign

Jo

investors’ trading behavior can have varying lead-and-lag relationships depending on the stock market turnover ratio: foreign investors change from behaving like market followers to influencing as well as being influenced by the stock price index when the stock market turnover ratio exceeds the threshold value of 0.6850%. However, this relationship could not be uncovered using Johansen’s (1988) traditional cointegration 19

model. Our investigation using the threshold cointegration model sheds light on the prior puzzling finding that foreign investors’ trading behavior leads the Taiwan stock market index (e.g., Yu and Lai, 1999) and that the two can be mutually causally related (e.g., Chiang et al., 2003).

A follow-up question is why foreigners would change their role from being

ro of

market followers to mutually interacting with the stock price index when the market

turnover ratio is higher than the threshold. If high market turnover is an indicator of a

-p

relatively large number of irrational investors (Baker and Stein, 2004) and varying

re

beliefs among all investors (e.g., Chen et al., 2009; Verardo, 2009), our finding that foreign investors are not followers in a speculative market seems to imply that they are

lP

relatively calm and rational when other investors are in turmoil. This finding is

na

inconsistent with the finding of Richards (2005) who demonstrates that foreigner investors have a relatively large impact on low-turnover-market countries such as

ur

Indonesia and the Philippines and a relatively slight impact on high-turnover-market

Jo

countries such as Korea and Taiwan.

3.3.4.2 Using stock index return as the threshold variable In Table 7 we report results using the stock index return on the previous day(∆𝑆𝑃𝑡−1 ) as the threshold variable. The time series of stock index return and the estimated threshold value 0.8701% are depicted in Figure 2. The result in Table 7 20

confirms the existence of a threshold effect, but the estimated coefficients are quite different from the results of Table 6 using the turnover ratio on the previous day (TRt-1) as the threshold variable. Condition I (II) is defined as the stock index return being lower (higher) than 0.8701%, as depicted in the upper (lower) part of Table 7. Conditions I and II comprise 79.92% and 20.08% of the sample respectively.

ro of

When the market is in Condition I (low stock index return; upper part of Table 7),

foreign investors’ buys (∆𝐹𝑃𝑡 ) are positively affected by the error-correction term and

-p

the lag-one stock price index, but negatively affected by their lag-one buys and lag-one

re

sells. Foreign investors’ sells (∆𝐹𝑆𝑡 ) are negatively affected by the error-correction term, their lag-one buys, their lag-one sells, and the stock price index. The stock price

lP

index (∆𝑆𝑃𝑡 ) is negatively affected by the error-correction term but not significantly

na

affected by foreign investors’ lag-one sells or buys. These results indicate that foreign investors’ buys and sells do not lead the stock index return when the stock index return

ur

is low. In contrast, the stock price index has leading effect on foreign investors’ buys

Jo

and sells when the stock index return is low. That is, foreign investors’ purchase (sell) more after an increase (decrease) in the stock price index in a low-return market.

When the market is in Condition II (high stock index return; lower part of Table 7), foreign investors’ buys (∆𝐹𝑃𝑡 ) are not only negatively affected by their lag-one buys but also positively affected by the error-correction term and the lag-one stock price 21

index. Moreover, foreign investors’ sells (∆𝐹𝑆𝑡 ) are negatively affected by the errorcorrection term, their lag-one buys, and their lag-one sells, and they are positively affected by the lag-one stock price index. The stock price index (∆𝑆𝑃𝑡 ) is negatively affected by the error-correction term and foreign investors’ lag-one sells, and it is positively affected by foreign investors’ lag-one buys and the stock price index. These

ro of

results indicate that foreign investors are not market followers in a high-return market. That is, foreign investors’ trading behavior exhibits a mutually causal interaction

-p

between stock price index in a high-return market. When the stock index return (∆𝑆𝑃𝑡−1 ) is high, the stock index return will increase (decrease) as foreign investors bought (sold)

re

on the previous day. However, we find that foreign investors’ trading behavior is also

lP

significantly affected by the stock price index: they will increase their buys and sells

na

after noticing an increase in the stock price index on the previous day.

<

ur

From the aforementioned empirical results, we know that foreign investors

Jo

facilitate the price discovery function when the market is hot (e.g., high stock market turnover and/or high market return). In contrast, they become market followers when the market is cold (e.g., low stock market turnover and/or low market return). However, hot markets only comprise a small portion of our sample (high stock market turnover and/or high market return comprise 33.2% and 20.08%, respectively). Most of the cases 22

investigated in this study are cold markets (low stock market turnover and/or low market return comprise 66.8% and 79.92%, respectively). The difference in market composition could account for why our findings are different from Ülkü (2015) who finds that net foreign flows bear no significant relation to future local returns in the

ro of

period of 2001-2012 in Taiwan.

3.3.5 Discussion

We further discuss whether the contrast between bull and bear market conditions

-p

is similar to the contrast between hot and cold market conditions, with the hot (cold)

re

market being defined as high (low) stock market turnover and/or high (low) market

lP

return estimated by threshold model. Prior studies extensively use the Markov switching model to identify the business cycle or bull and bear of financial market. (see,

na

e.g., Hamilton 2011, Nikolsko-Rzhevskyy et al., 2012 and Maheu et al., 2000). We

ur

adopt a two-state Markov switching model as follows:22

Jo

𝑅𝑡 = 𝜇(𝑆𝑡 ) + 𝜌(𝑆𝑡 )𝑅𝑡−1 + 𝜎(𝑆𝑡 )𝜐𝑡 ,

𝜐𝑡 ~𝑁(0,1),

(3)

where 𝑅𝑡 is the stock return; 𝜇(𝑆𝑡 ) is the mean of stock return that depends on the state variable 𝑆𝑡 . The state variable 𝑆𝑡 is unobservable, and 𝑆𝑡 = 1 or 2.

22

𝜎(𝑆𝑡 ) is the

We also consider that the transition probability of Markov switching depends on the lags of return or turnover ratio, and the empirical results are similar. 23

variance of stock return and it also depends on the state variable 𝑆𝑡 . The transition probabilities are: Pr[𝑆𝑡 = 𝑗| 𝑆𝑡−1 = 𝑖} = 𝑝𝑖𝑗

𝑖, 𝑗 = 1, 2 .

(4)

The estimates for the Markov switching model are presented in Table 8. The estimate of 𝜇(𝑆𝑡 = 1) is smaller than the estimate of 𝜇(𝑆𝑡 = 2), and 𝜎(𝑆𝑡 = 1) is

ro of

larger than 𝜎(𝑆𝑡 = 2). We follow Maheu et al., (2000) to label the bear-market regime when 𝑆𝑡 = 1 , and the bull-market regime when 𝑆𝑡 = 2 . The smoothed regime

-p

probabilities of bear market ( 𝑆𝑡 = 1) are displayed in Figure 3. As the figure shows,

lP

<

re

the probabilities of bear market are high during the financial crisis in 2008.

We then estimate the cointegration model of the stock market index, foreign

na

investors’ buys and sells under two market regimes of bull and bear.23 The empirical results are summarized in Table 9. The stock price index (∆𝑆𝑃𝑡 ) is positively affected

ur

by foreign investors’ lag-one buys (∆𝐹𝑃𝑡−1 ) and the foreign investors’ sells (∆𝐹𝑆𝑡 ) are

Jo

negatively affected by the stock price index on the previous day (∆𝑆𝑃𝑡−1 ) under both regimes of bear and bull markets. That is, foreign investors’ trading behavior and the price index are mutually and causally related under both regimes of bull and bear

23

We identify the bear-market state when the smoothed regime probability of bear market is greater than 50% in the cointegration model. 24

markets. The results are quite different from the estimates of threshold cointegration in Table 6 and Table 7. <

We use the AIC model selection criteria proposed by Akaike (1973) to determine which cointegration model is better. The results summarized in Table 10 indicate that

ro of

according to the AIC criterion both threshold cointegration models are better than Johansen’s (1988) traditional cointegration model and the cointegration model with the

-p

bull and bear markets identified by Markov switching method . Moreover, the model

re

with market turnover ratio as the threshold variable is the best among the four models. The result from Johansen’s traditional model yields the worst AIC performance in Table This reconfirms the inappropriateness of using the traditional cointegration model

lP

10.

na

to explore the dynamic relationship between foreigner investors’ trading behavior and the stock price index. Besides, the use of Markov switching model does not help

ur

because it shows limited improvement on the AIC of Johansen’s (1988) model. Another

Jo

merit of the threshold cointegration is that the use of previous turnover in estimation is that our threshold estimate such as 0.6850% of the previous day’s turnover ratio is more convenient, accessible, and intuitive to practitioners than the unobservable state of Markov switch in identifying the condition of financial market. <

4. Conclusion Extending the threshold cointegration model developed by Hansen and Seo (2002) to avoid arbitrarily chosen cutoff values for market conditions, Whether foreign

ro of

investors enhance market efficiency has been a debatable issue. The issue merits further study of Taiwan stock market that is comprised of a high percentage of individual investors. We adopt the threshold cointegration model to explore the nonlinear

-p

threshold effect using the error-correction term, turnover ratio and stock index return as

re

threshold variables. The empirical results indicate that the threshold effect is salient for

lP

two of these threshold variables: turnover ratio and stock index return. Foreign investors’ trades (buys and sells) are shown to have a bi-direction causal relationship with the

na

stock price index when the stock market turnover ratio and/or the stock index return is high. This implies that the trading behavior of foreign investors facilitates the price

ur

discovery function for local stock markets when the overall market is hot. In contrast,

Jo

when the overall market is cold (i.e., low market turnover and/or low stock index return), foreign investors become market followers.

A finding worthy of special note is that foreign investors increase their purchase of shares after noticing an increase in the stock market return when market turnover

26

and/or market return is high. The result indicates that foreign investors are aggressive traders in a high-turnover and/or a high-return market, and this is similar to the finding of Chuang and Susmel (2011) that foreign institutional investors trade aggressively in a bull market. In a nutshell, our answer to the old puzzle of whether the involvement of foreign

ro of

investors enhances or impairs the efficiency of local stock markets is that it depends on market conditions. When the market is hot, foreign investors tend to interact with the

-p

market so as to facilitate price discovery and therefore enhance market efficiency. However, they become market followers when the market is cold.

Jo

ur

na

lP

re

`

27

References Akaike, H., 1973. Maximum Likelihood identification of Gaussian autoregressive moving average models. Biometrika 60, 255-265. Altissimo, F. and Violante, G., 2001. The nonlinear dynamics of US GNP and unemployment. Journal of Applied Econometrics 16, 461-486. Bae, K.H., Bailey, W. and Mao, C.X., 2006. Stock market liberalization and information environment. Journal of International Money and Finance 25, 404-428. Baker, M., Ruback, R.S. and Wurgler, J., 2004. Behavioral corporate finance: a survey. National Bureau of Economic Research, No. w10863.

ro of

Baker, M. and Stein, J.C., 2004. Market liquidity as a sentiment indicator. Journal of Financial Markets 7, 271-299. Baker, M. and Wurgler, J., 2006. Investor sentiment and the cross-section of stock returns. Journal of Finance 61, 1645-1680. Balke, N.S. and Fomby, T.B., 1997. Threshold cointegration.” International Economic Review 38, 627- 645.

-p

Barber, B.M. and Odean, T., 2001. Boys will be boys: gender, overconfidence, and common stock investment. Quarterly Journal of Economics 116, 261-292.

re

Bowe, M. and Domuta, D., 2004. Investor herding during financial crisis: a clinical study of the Jakarta stock exchange. Pacific-Basin Finance Journal 12, 387- 418.

lP

Brennan, M. and Cao, H., 1997. International portfolio investment flows. Journal of Finance 52, 1851-1880. Chan, L.K.C. and Lakonishok, J., 1993. Institutional trades and intraday stock price behavior,” Journal of Financial Economics 33, 173-199.

na

Chen, S.N., 1982. An examination of risk-return relationship in bull and bear markets using time-varying betas. Journal of Financial and Quantitative Analysis 17, 265-282. Chen, C.R., Lung, P.P. and Wang, F.A., 2009. Stock market mispricing: money illusion or resale option? Journal of Financial and Quantitative Analysis 44, 1125-1147.

Jo

ur

Chiang, S.J., Tsai, L.J., Shu, P.G. and Chen, S.L., 2012. The trading behavior of foreign, domestic institutional, and domestic individual investors: evidence from the Taiwan stock market. Pacific-Basin Finance Journal 20, 745-754. Chiang, S.M., Cheng, W.H. and Chiu, C.L., 2003. The dynamic relationship among foreign investments, stock market and foreign exchange market. Journal of Risk Management 15, 45-64. Choe, K., Kho, B.C. and Stulz, R.M., 1999. Do foreign investors destabilize stock markets? The Korea experience in 1997. Journal of Financial Economics 54, 227-264. Choi, N., Fedenia, M., Skiba, H., and Sokolyk, T., 2017. Portfolio concentration and performance of institutional investors worldwide. Journal of Financial Economics 123, 189-208. 28

Chuang, W.I. and Susmel, R., 2011. Who is the more overconfident trader? Individual vs. institutional investors. Journal of Banking and Finance 35, 1626-1644. De Long, J.B., Shleifer, A., Summers, L.H. and Waldmann, R.J., 1990. Positive feedback investment strategies and destabilizing rational speculation. Journal of Finance 45, 379395. Dickey, D. and Fuller, W.A., 1979. Distribution of the estimates for autoregressive time series with a unit root. Journal of the American Statistical Association 74, 427-431. Dickey, D. and Fuller, W.A., 1981. Likelihood ratio statistics for autoregressive time series with a unit root. Econometrica 49, 1057-1072. Dvořák, T., 2005. Do domestic investors have an informational advantage? Evidence from Indonesia. Journal of Finance 60, 817-839.

ro of

Engle, R.F. and Granger, C.W.J., 1987. Cointegration and error correction: representation, estimation, and testing. Econometrica 55, 251-276. Ferreira, M.A., Matos, P., Pereira, J.P., Pires, P., 2017. Do locals know better? A comparison of the performance of local and foreign institutional investors. Journal of Banking & Finance 82, 151-164.

-p

Froot, K.A., O’Connell, P.G.J. and Seasholes, M.S., 2001. The portfolio flows of international investors. Journal of Financial Economics 59, 151-193.

re

Froot, K.A. and Ramadorai, T., 2008. Institutional portfolio flows and international investments. Review of Financial Studies 21, 937–971.

lP

Gombola, M.J. and Liu, F.Y.L., 1993. Dividend yields and stock returns: evidence of time variation between bull and bear markets. Financial Review 28, 303-327. Griffin, D. and Tversky, M., 1992. The weighing of evidence and the determinants of confidence. Cognitive Psychology 24, 411-435.

na

Grinblatt, M. and Keloharju, M., 2000. The investment behavior and performance of various investor types: a study of Finland's unique data set. Journal of Financial Economics 55, 4367.

ur

Hamilton, J. D., 2011. Calling recessions in real time. International Journal of Forecasting 27, 1006-1026.

Jo

Hansen, B., 1996. Inference when a parameter is not identified under the null hypothesis. Econometrica 64, 413-430. Hansen, B., 1997. Inference in TAR models. Studies in Nonlinear Dynamics and Econometrics 2, 1-14. Hansen, B., 1999. Testing for linearity. Journal of Economic Surveys 13, 551-576. Hansen, B., 2000. Sample splitting and threshold estimation. Econometrica 68, 575-603. Hansen, B. and Seo, B., 2002. Testing for two-regime threshold cointegration in vector error correction models. Journal of Econometrics 110, 293-318.

29

Ikizlerli, D., Holmes, P., and Anderson, K., 2019. The response of different investor types to macroeconomic news. Journal of Multinational Financial Management, forthcoming. Jeon, J.Q. and Moffett, C.M., 2010. Herding by foreign investors and emerging market equity returns: evidence from Korea. International Review of Economics and Finance 19, 698710. Jo, Gab-Je., 2002. Foreign equity investment in Korea. Paper presented at the Association of Korean Economic Studies. Johansen, S., 1988, “Statistical analysis of cointegration vectors. Journal of Economic Dynamics and Control 12, 231-254. Kahneman, D. and Tversky, A., 1979. Prospect theory: an analysis of decision under risk. Econometrica 47, 263-291.

ro of

Kim, E.H. and Singal, V., 2000. Stock market openings: experience of emerging economies. Journal of Business, 73, 25-66. Kim, M.K. and Zumwalt, J.K., 1979. An analysis of risk in bull and bear markets. Journal of Financial and Quantitative Analysis 14, 1015-1025.

-p

Kim, W. and Wei, S.J, 2002. Foreign portfolio investors before and during a crisis. Journal of International Economics 56, 77-96.

re

Kwiatkowski, D., Phillips, P., Schmidt, P. and Shin, Y., 1992. Testing the null hypothesis of stationarity against the alternative of a unit root. Journal of Econometrics 54, 159-178. Lo, A.W. and Wang, J., 2000. Trading volume: definitions, data analysis, and implications of portfolio theory. Review of Financial Studies 13, 257-300.

lP

Maheu, J.M., McCurdy, T.H., 2000. Identifying bull and bear markets in stock returns. Journal of Business and Economic Statistics 18, 100–112.

na

Nikolsko-Rzhevskyy, A. and Prodan, R., 2012. Markov switching and exchange rate predictability. International Journal of Forecasting 28, 353-365. Odean, T., 1999. Do investors trade too much? ” American Economic Review 89, 1279-1298.

ur

Ogunmuyiwa, M.S., 2010. Investor’s sentiment, stock market liquidity and economic growth in Nigeria. Journal of Social Science 23, 63-67.

Jo

Onishchenko, O., and Ülkü, N., 2019.Foreign investor trading behavior has evolved. Journal of Multinational Financial Management, forthcoming. Potter, S.M., 1995. A nonlinear approach to US GNP. Journal of Applied Econometrics 10, 109-125. Richards, A., 2005. Big fish in small ponds: the trading behavior and price impact of foreign investors in Asian emerging equity markets. Journal of Financial and Quantitative Analysis 40, 1-27. Seasholes, M., 2000. Smart foreign traders in emerging markets. Working paper. Harvard Business School.

30

Shefrin, H. and Statman, M., 1985. The disposition to sell winners too early and ride losers too long: theory and evidence. Journal of Finance 40, 777-790. Shen, C.H. and Chiang, C.H., 1999. Retrieving the vanishing liquidity effect—a threshold vector autoregressive model. Journal of Economics and Business 51, 259-277. Simon, H.A., 1955. A behavioral model of rational choice. Quarterly Journal of Economics 69, 99-118. Stambaugh, R.F., Yu, J. and Yuan, Y., 2012. The short of it: investor sentiment and anomalies. Journal of Financial Economics 104, 288-302. Tong, H., 1978. On a threshold model in pattern recognition and signal processing. NATO ASI Series E: Applied Sciences No. 29, ed. C. H. Chen. The Netherlands: Sijthoff & Noordhoff, 575-586.

ro of

Tong, H., 1983. Threshold models in nonlinear time series analysis. Lecture Notes in Statistics No.21, New York: Springer-Verlag.

Tong, H. and Lim, K.S., 1980. Threshold autoregression, limit cycles and cyclical data. Journal of the Royal Statistical Society 42, 245-292.

-p

Tsay, R.S., 1989. Testing and modeling threshold autoregressive process. Journal of the American Statistical Association 84, 231-240.

re

Ülkü, N., 2015. The interaction between foreigners' trading and stock market returns in emerging Europe. Journal of Empirical Finance 33, 243-262. Verardo, M., 2009. Heterogeneous beliefs and momentum profits. Journal of Financial and Quantitative Analysis 44, 795-822.

lP

Weber, M. and Camerer, C.F., 1998. The disposition effect in securities trading: an experimental analysis. Journal of Economic Behavior & Organization 33, 167-184.

Jo

ur

na

Yu, C.H. and Lai, Y.C., 1999. On the informational leading role of foreign institutional investors. Journal of Financial Studies 7, 1-26.

31

f oo

3

Turnover Ratio (%)

pr

2.5

Threshhold Value 0.6850%

Pr

e-

2

1.5

Jo ur

0.5

na l

1

0

Figure 1. Stock market turnover ratios and the threshold value. 32

f oo

8

Stock Return (%)

Threshhold Value 0.8701%

pr

6

e-

4

Pr

2

0

-6

-8

Jo ur

-4

na l

-2

Figure 2. Stock market returns and the threshold value.

33

f oo

1.00

pr

0.90

e-

0.80 0.70

Pr

0.60 0.50

0.20 0.10 0.00

Jo ur

0.30

na l

0.40

Figure 3. Smoothed regime probabilities of bear market

34

Table 1 Summary statistics of foreign investors’ trades and the Taiwan stock market.

Panel A: Stock market—daily data Stock price index

Turnover (%)

Return (%)

Trade volume

(SP)

(TR)

( SP )

Mean

7,420

0.628

0.014

96,092

Median

7,580

0.534

0.041

90,704

Maximum

11,253

2.757

6.742

326,463

Minimum

3,446

0.097

-6.738

7,699

(in NT$ million)

Panel B: Foreign investors’ trades—daily data Sells (FS)

(in NT$ million) (in NT$ million)

Total trades

Net buys

Trading ratio

(in NT$ million)

(in NT$ million)

(%)

655

30.62

384

33.22

ro of

Buys (FP)

15,277

14,622

29,899

Median

14,763

13,618

29,017

Maximum

135,109

116,055

218,482

125,288

108.38

Minimum

26

17

74

-62,398

0.24

Jo

ur

na

lP

re

-p

Mean

35

Table 2 Data description: Breakdown of foreign investors’ trades and the Taiwan stock market. Panel A: Stock market —daily data Stock price index

Turnover (%)

Return (%)

(SP)

(TR)

( SP )

1995

5,544

0.775

-0.102

35,987

1996

6,004

0.826

0.109

45,619

1997

8,411

1.242

0.069

131,856

1998

7,738

0.949

-0.079

109,816

1999

7,427

0.887

0.116

110,868

2000

7,847

0.717

-0.190

113,713

2001

4,907

0.655

0.084

2002

5,226

0.832

-0.073

2003

5,162

0.830

0.122

2004

6,034

0.827

0.028

2005

6,092

0.520

2006

6,842

0.545

2007

8,510

0.655

2008

7,024

0.558

2009

6,460

0.754

2010

7,950

2011

8,156

2012

7,481

2013

8,093

2014 2015

2017

ro of

-p

75,453 88,456 82,258 96,711

0.077

97,570

0.043

135,714

-0.225

107,086

0.243

119,993

0.559

0.042

115,080

0.443

-0.087

109,294

0.341

0.039

83,154

0.341

0.048

79,688

8,992

0.342

0.034

92,908

8,959

0.306

-0.040

92,234

8,763

0.258

0.046

77,523

10,208

0.342

0.058

104,874

10,620

0.337

-0.032

130,212

lP

77,129

na

Jo

2018

(in NT$ million)

re

2016

Trading volume

0.029

ur

Year

(to be continued)

36

Table 2 (Continued) Data description: Breakdown of foreign investors’ trades and the Taiwan stock market.

Panel B: Foreign investors’ trades—daily data Year

Buys (FP)

Sells (FS)

Total trades

(in NT$ million)

(in NT$ million)

(in NT$ million)

Net buys (in NT$ million)

Trading ratio (%)

560

400

960

160

2.85

1996

1,044

853

1,898

191

4.42

1997

2,507

2,637

5,144

-130

4.01

1998

2,237

2,146

4,384

91

4.16

1999

3,906

2,652

6,558

1,254

6.31

2000

5,402

4,861

10,263

541

10.32

2001

5,944

4,688

10,632

1,255

16.15

2002

6,827

6,714

13,541

112

16.58

2003

9,868

7,664

17,532

2,205

22.65

2004

12,659

11,523

24,182

1,136

27.06

2005

15,240

12,328

27,568

2,913

36.31

2006

19,064

16,814

35,878

2,250

37.36

2007

26,745

26,445

53,190

300

39.74

2008

25,087

26,974

52,061

-1,888

48.96

2009

20,581

18,668

39,249

1,913

33.69

2010

21,821

20,701

42,523

1,120

37.90

2011

23,279

24,403

47,683

-1,124

43.66

2012

19,093

18,516

37,609

577

46.06

2013

20,152

19,164

39,316

988

48.89

2014

22,824

21,394

44,217

1,430

47.43

2015

26,272

26,082

52,354

189

56.14

24,263

22,950

47,213

1,312

60.28

27,504

26,873

54,378

631

51.35

34,950

68,460

-1,440

52.18

2018

-p

re

lP

na

Jo

2017

ur

2016

33,510

ro of

1995

37

Table 3 KPSS tests for existence of stationarity or trend stationary series. Panel A: Trading variables

Taiwan stock price index (SP)

Raw series (SP) First difference ( SP )

Foreign investors’ buys (FP)

Raw series (FP) First difference(ΔFP)

Foreign investors’ sells (FS)

Raw series (FS)

4.127**

0.835**

0.047

0.026

8.956**

0.683**

0.034

0.029

8.754**

0.514**

0.054

0.046

ro of

First difference(ΔFS)

Panel B: Threshold variables

Stock turnover ratio (TR)

0.349

0.090

Stock index return ( SP )

0.035

0.027

is the statistic developed by Kwiatkowski et al. (1992) for testing the null hypothesis

-p

Note:

of stationarity against an alternative series with a unit root. The critical value for the 5% significance level is 0.463.

re

is the statistic developed by Kwiatkowski et al. (1992) for testing the existence of trend

Jo

ur

na

lP

stationary series. The critical value for the 5% significance level is 0.146. ** denotes significance at the 5% level.

38

Table 4 Cointegration test for stock price index and foreign investors’ buys and sells.

Panel A: Johansen (1988) cointegration test Maximum eigenvalue Statistic

Null hypothesis :

Critical value for

Trace Statistic

Critical value for

5% significance Number of cointegration = 0 Number of cointegration = 1

120.940** 8.474

5% significance

21.132

133.345**

29.797

14.265

12.405

15.495

Panel B: Engle and Granger (1987) cointegration test Statistic

Nonexistence of the cointegration relation

-3.182**

Jo

ur

na

lP

re

-p

** denotes significance at the 5% level.

Critical value for 5% significance

ro of

Null hypothesis:

39

-2.8618

Table 5 Test of threshold effect for different threshold variables.

SupLM test of threshold effect Threshold variable: Fixed regressor bootstrap

Residual bootstrap

(p-value)

(p-value)

Statistic Error-correction term

30.4465

0.11

0.18

Market turnover ratio

84.3419

0.00**

0.00**

Stock index return

35.1410

0.01**

0.01**

Note: The p-values of the fixed regressor bootstrap and the residual bootstrap are based on Hansen and Seo (2002) and calculated from 500 repeated samplings.

Jo

ur

na

lP

re

-p

ro of

**denotes significance at the 5% level.

40

Table 6 The estimates of coefficients (standard deviations) for the threshold cointegration model with stock market turnover ratio as the threshold variable. Condition I: Stock market turnover ratio (TRt-1) < 0.6850%

FPt

FS t

-0.199 (0.022)**

0.285 (0.022)**

0.002 (0.001)**

0.340 (0.035)**

-0.432 (0.035)**

-0.004 (0.001)**

FPt 1

-0.334 (0.022)**

-0.158 (0.022)**

0.0003 (0.001)

FS t 1

-0.068 (0.021)**

-0.267 (0.021)**

0.0002 (0.001)

SPt 1

0.316 (0.646)

-3.073 (0.653)**

0.035 (0.018)**

t 1

Condition II: Stock market turnover ratio (TRt-1) > 0.6850%

FPt

(33.2% of the sample)

FS t

SPt

0.203 (0.026)**

0.001 (0.001)

t 1

0.328 (0.041)**

-0.369 (0.041)**

-0.002 (0.001)*

FPt 1

-0.182 (0.026)**

-0.0001 (0.025)

FS t 1

-0.056 (0.025)**

-0.271 (0.024)**

-0.002 (0.001)**

SPt 1

4.200 (0.761)**

-0.937 (0.757)

-0.003 (0.021)

t 1

-p

-0.245 (0.026)**

Constant

Note:

SPt

ro of

Constant

(66.8% of the sample)

0.002 (0.001)**

denotes the error-correction term; FP denotes foreign investors’ buys; FS denotes foreign

Jo

ur

na

lP

re

investors’ sells, and SP denotes the stock price index. The threshold value 0.6850% is estimated by a grid search to find the highest value of maximum likelihood following the method proposed by Hansen and Seo (2002). ** denotes significance at the 5% level.

41

Table 7 The estimates of coefficients (standard deviations) for the threshold cointegration model with stock index return as the threshold variable. Condition I: Stock index return: ( SPt 1 ) < 0.8701%

(79.92% of the sample)

FPt

FS t

SPt

Constant

-0.240 (0.019)**

0.231 (0.019)**

0.002 (0.001)**

t 1

0.389 (0.032)**

-0.356 (0.031)**

-0.003 (0.001)**

FPt 1

-0.268 (0.020)**

-0.069 (0.019)**

0.001 (0.001)

FS t 1

-0.096 (0.019)**

-0.300 (0.019)**

-0.001 (0.001)

SPt 1

2.630 (0.705)**

-1.876 (0.701)**

-0.014 (0.020)

( SPt 1 ) > 0.8701%

(20.08% of the sample)

FPt Constant

FS t

-0.202 (0.019)**

SPt

ro of

Condition II: Stock index return:

0.279 (0.019)**

0.003 (0.001)**

0.269 (0.031)**

-0.553 (0.032)**

-0.006 (0.001)**

FPt 1

-0.247 (0.019)**

-0.136 (0.020)**

0.002 (0.001)**

FS t 1

-0.013 (0.019)

-0.241 (0.019)**

-0.002 (0.001)**

SPt 1

3.170 (0.697)**

-p

t 1

2.513 (0.717)**

0.083 (0.020)**

Note: t 1 denotes the error-correction term; FP denotes foreign investors’ buys; FS denotes foreign

re

investors’ sells, and SP denotes the stock price index. The threshold value 0.8701% is estimated by a grid search to find the highest value of maximum likelihood following the method proposed by

lP

Hansen and Seo (2002).

Jo

ur

na

**denotes significance at the 5% level.

42

Table 8 The estimates for the two-state Markov switching model. parameter

parameter estimate

standard error

p-value

𝜇(𝑆𝑡 = 1)

-0.0862**

0.0408

0.035

𝜇(𝑆𝑡 = 2)

0.0766**

0.0146

0.000

𝜌(𝑆𝑡 = 1)

0.0449**

0.0209

0.032

𝜌(𝑆𝑡 = 2)

0.0516**

0.0175

0.003

𝜎(𝑆𝑡 = 1)

1.9207**

0.0183

0.000

𝜎(𝑆𝑡 = 2)

0.8251**

0.0171

0.000

𝑝11= 0.976 ,

𝑝22 = 0.985

Jo

ur

na

lP

re

-p

ro of

Note: **denotes significance at the 5% level.

43

Table 9 The estimates of coefficients (standard deviations) for the cointegration model with the regimes of bull and bear markets identified by Markov switching method.

Bear-market regime: 61.76% of the sample

FPt

FS t

SPt

Constant

-0.251 (0.025)**

t 1

0.385 (0.040)**

-0.426 (0.039)**

FPt 1

-0.305 (0.024)**

-0.103 (0.023)**

0.001 (0.0003)**

FS t 1

-0.056 (0.023)**

-0.277 (0.023)**

-0.001 (0.0004)**

SPt 1

2.657 (1.034)**

-4.160 (1.026)**

-0.031 (0.018)

Bull-market regime: 38.24% of the sample

FPt Constant

FS t

-0.211 (0.028)**

0.001 (0.0004)** -0.0008 (0.0007)

ro of

0.282 (0.025)**

SPt

0.224 (0.028)**

0.002 (0.001)**

0.305 (0.043)**

-0.331 (0.044)**

-0.005(0.002)**

FPt 1

-0.232 (0.026)**

-0.079 (0.026)**

0.002 (0.001)**

FS t 1 SPt 1

-p

t 1

-0.096 (0.025)** 1.527 (1.128)

-0.280 (0.025)**

-0.001 (0.001)

-1.781 (1.1443)**

0.010 (0.042)

re

Note: t 1 denotes the error-correction term; FP denotes foreign investors’ buys; FS denotes foreign investors’ sells, and SP denotes the stock price index. The bull and bear markets are identified by

Jo

ur

na

lP

two-state Markov switching model. **denotes significance at the 5% level.

44

Table 10 AIC for estimation models.

AIC Johansen (1988) cointegration model

-37391.3

Cointegration model with bull and bear markets identified by

-37393.7

Markov switch Threshold Cointegration Models (for different threshold variable): Threshold variable:

Stock market turnover ratio

-37457.8

Stock index return

-37421.6

ro of

Note: AIC is Akaike’s (1973) information criterion. The optimal model is the model with the

Jo

ur

na

lP

re

-p

smallest AIC value.

45