WP/18/04

Investors Behavior and Trading Strategies: Evidence from Indonesia Stock Exchange

December 2018

Abstract

This study reveals new evidence about the behavior and trading strategies of institutional and individual investors in the Indonesia Stock Exchange. Firstly, individual (institutional) investors are most likely to trade frequently (infrequently) with small (large) amounts of money and short (long) holding period. Secondly, individual (institutional) investors are consistent to perform contrarian (momentum) strategy. Lastly, past trading activities done by individual (institutional) investors are significantly affecting the current trading behavior and strategy of individual investors (both investor types). The above findings related to individual investors are robust when this study further breakdowns institutional investors into eight different investor types.

JEL Classifications: G14, G15. Keywords: Market Microstructure, Emerging Market, Institutional Investors, Individual Investors, Trading Strategies.

Corresponding author: Inka Yusgiantoro ([email protected]). The findings and interpretations expressed in this paper are entirely those of the authors and do not represent the views of Indonesia Financial Services Authority (OJK). All remaining errors and omissions rest with the authors.

INTRODUCTION

The technique of conducting research in the capital market already has been changed quite significantly in several decades. In data source perspectives, the research was mainly employing the closing daily data, and/or aggregate market trading activities, the later method starts to consider the market microstructure analysis that using intraday detail transaction data. In term of the unit of analysis, the analysis shifted from the market aggregate dynamics to the specific type of investors or trader’s behavior. These types of studies are being the common and major research methods and used in analyzing many developing economy, the microstructures research for emerging market has less being study.

Brzeszczyński, Gajdka, and Kutan (2015) stated that there are some important reasons for conducting microstructure research method in emerging market. First, emerging market economies grow significantly and more resilience over time. Stronger growth and lower corporate leverage, alongside with prospects for growth spillovers from advanced economies, has improved due to their macroeconomic outlook. Second, the biggest support pillar of emerging market economies to grow fast is significant economic reforms and major structural changes. It is proven by China’s stock market’s share which currently ranks second right after the US, surpassing Japan and the European Union. Third, institutional investor desire to trade in emerging market was increasing, proven by the growth of institutional capital flow until 2015, along with the rise of non-residents capital flow number.

Unfortunately, while the research of microstructure data in emerging markets starts gaining attention, such as a study is still rare in Indonesia. Most of the capital market research in Indonesia mainly used daily closing and aggregate data, whereas many other previous researches used the fundamental data obtained from financial statement. Below are some evidence to support this argument.

First, it is worth to mention that studies conducted by Comerton-Forde (1999) and Bonser- Neal, Linnan, and Neal (1999) are among the first study that intensively using the microstructure approach in Indonesia. Specifically, Comerton-Forde (1999) examines the impact of opening rules on stock market efficiency in Australia and Jakarta Stock Exchange (JSX). She finds that the use of a call can increases market efficiency through increased liquidity and lower volatility at the open. Meanwhile, Bonser-Neal, Linnan, and Neal (1999) undertake a research about transaction cost in Indonesia and find that JSX execution cost is surprisingly similar to those non-US developed markets. Moreover, they also find that execution costs are affected by broker identity and trades initiated by foreigners have significantly bigger execution costs.

Later on, the more advanced research was conducted by Dvořák (2005) and Agarwal, Faircloth, Liu, and Rhee (2009) to study the profitability of foreign and domestic investors in Indonesia. Specifically, Dvořák (2005) finds that domestic clients of global brokerage get

1 more profits than foreign clients of global brokerages, indicating there is an advantage from the combination of global expertise and local information. In other words, domestic investors who have better information still need the expertise of foreign firms to make use of that information into greater profits. Meanwhile, Agarwal, Faircloth, Liu, and Rhee (2009) find similar results that foreign investors underperform domestic investors. This underperformance of foreign investors is totally attributable to their non-initiated orders because they outperform domestic investors in initiated orders.

Unlike those previous researches, this study will address more on the effects of the behavior of institutional and individual investors in Indonesia, an area that has not been addressed often. To shows the dynamic behavior, this study uses the longer and more recent data period of 2013–2015. It is expected to portray the more recent of the behavior both individual and institutional capital market investors and traders in Indonesia. One of the motivations of this study seeks the answer why the individual equity ownership is significantly low (around 6- 7%) when compared to the institutional equity ownership (around 93-94%) as documented in Table 1. At the same time, it is reported by the Indonesia Central Securities Depository in 2018 that the capital market participation is less than one percent of the population.

In addition of using a longer and more recent data set, this study also benefited from the information of the actual respective type of investors or traders so that it is no need for this study to proxy the investors type like in previous studies. The data that we used are investors that classified into one general individual investor and eight different types of institutional investors, namely corporations, financial institutions, securities firms, insurance firms, mutual funds, pension funds, foundations, and other institutions. With this detailed transaction data, it is interesting and possible to research the dynamic interaction of stocks return and players trading activity of a particular type of institutional investors and individual investors. Likewise, with this information, we also can study in more detail regarding which type of investors behavior is having significant effects on the return of Indonesia stock exchange.

The main discussion in this study is focused on examines (1) the dynamics relation of trading behavior of various institutional and individual investors, (2) the underlying strategy applied by each investor type in its trading activities, i.e. contrarian and/or momentum, and (3) how the contemporaneous relationship among players trade and stocks return (herding behavior activity). All imply the trading dynamics relation amongst investors.

Particularly, this study adopts the idea of the dynamics model of analysis between institutional and individual trading studied by Griffin, Harris, and Topaloglu (2003), Ng and Wu (2007), as well Dorn, Huberman, and Sengmueller (2008). This study will also observe the dynamics of players trading based on studies conducted by Lakonishok, Shleifer, and Vishny (1992).

2

Finally, Vector Autoregressive (VAR) methodology will be used to estimate this relationship. The estimation of parameters will use maximum Likelihood Estimation while the standard error of parameters will be adjusted with heteroscedasticity and autocorrelation using Newey West (NW) covariance estimation. The implementation of NW in the VAR model follows the suggestion from Cochrane and Piazzesi (2005).

The remaining contents of this article is organized as follows. Section 2 describes the literature review. Section 3 explains the institutional background and data. Section 4 elaborates the methodology. Section 5 performs preliminary analysis for determining the optimal lag selection as well testing the autocorrelation and heteroscedasticity for all models. Section 6 reports the results of general players. Section 7 documents the results of detailed players. Finally, Section 8 concludes and provides some policy implications.

LITERATURE REVIEW

Overview of Institutional and Individual Investors

As the detailed data of stocks market transaction become available to researcher today, the research regarding the behavior of players (both institutional and individual investors) in the stocks market is gaining much more attention than ever before. In general, institution investors can be defined as investors that trade on behalf of other interest while individual investors trade on their interest. Theoretically, institutional investors are viewed as informed investors with the power to drive the market while individual investors are believed as proverbial noise trader with a tendency to perform psychological biased in trading (Kyle, 1985; Black, 1985).

Nevertheless, defining institutional and individual investors through transactional data in stocks market is not easy since in most researches there is only broker name recorded in the transaction without no detail of who is the player behind it (Khwaja and Mian, 2005; İmişiker, Özcan, and Taş, 2015; Aaron, Koesrindartoto and Takashima, 2018). Moreover, by knowing that institutional or individual investors can use more than one brokers to trade in stocks market, it is not an appropriate way to directly judge a particular broker as an institutional or individual investor. As an alternative approach, some researches like Laskonishok, Shleifer, and Vishny (1992), Barber, Odean, and Zhu (2009), as well Ng and Wu (2007) use a dollar cut-off for a transaction to classify whether the transaction initiated by a certain broker is executed by institutional or individual investors. Fortunately, as this study have a direct access to the regulator, namely Indonesian Financial Services Authority, we did not face this kind of issue, and therefore the results of this study will be free from biases caused by using a proxy.

3

Then, for the dynamic interaction between the players, the growing literature on this area gives different findings yet with a decent explanation. In short, the main focus of the research focus on examining (1) the investor trading strategy based on the relationship between stocks return and institutional and individual trading behavior, (2) how the players interact each other and (3) how the contemporaneous relationship between the change in players ownership to stocks return.

Trading Strategies

The first topic is to understand how players in the stock market buy (sell) stocks tomorrow in response to increase (decrease) of the return. This behavior is also known as momentum trading behavior (trend chasing or positive feedback trading) (Griffin, Harris, Topaloglu, 2003). Empirical literature finds different results regarding this behavior toward institutional and individual investors. Lakonishok, Shleifer, and Vishny (1992) find a weak evidence of trend chasing behavior in institutional investors in overall. As the analysis goes deep to the characteristic of the stocks (based on size), however, they find there is some evidence that institutional investors perform positive-feedback trading in small stock but not in the big stocks. On the other hand, Grinblatt, Titman and Wermers (1995) show that institutional investors are trend chasing investors that tend to follow the past price movement.

Moreover, Badrinath and Wahal (2001) explain that momentum trading behavior varies across the institution types and primarily limited to new equity position and by using detail transaction data from the Australian market, Foster, Gallagher, and Looi (2011) find that momentum trading behavior depends on the investment style of institutional investors. They further argue that growth-oriented investment manager tends to perform momentum trading while the value-oriented manager is not. In dynamics model, Griffin, Harris, Topaloglu (2003) find that there is a strong contemporaneous relation between past stock returns and institutional trading. With a similar thought, Ng and Wu (2007) conduct research in China. Using the detailed transaction record of 77.12 million trade accounts in Shanghai stocks market, they find that Chinese institutions are momentum investors.

The other perspective looks the momentum trading behavior of individual investors as a contrarian. Odean (1998) finds that individual investors tend to sell the winning stock and hold on to the past losing stock. This condition is also known as disposition effect (Dharma and Koesrindartoto, 2018). Barber and Odean (2000) explain that individual investors perform disposition in their trading because they are “anti-momentum” investors. Individual investors relatively do more buy trades than sell trades when there is an extreme positive return in the past. However, the value of sell trades that are executed is larger compared to the value of buy trades. In overall, the individual investor is a net seller in the market regarding market value following the extreme positive movement in previous days (Barber

4 and Odean, 2008). With the same market data with Barber and Odean (2008), Kaniel, Saar, and Titman (2008) also find the tendency of individual investors to buy a stock after prices decrease and sell it after the prices increase. Ng and Wu (2007) explain that the behavior of individual investors depends on their wealth. The less wealthy individual, in general, behave as contrarian investors while the wealthiest individual makes the momentum trade like Chinese institution. Based on that literature, this research believes that while institutional investors perform momentum trading strategy, individual investors perform anti-momentum or contrarian trading strategy.

Herding Behavior

The second topic explains how institutional and individual trading activity as well as the interaction between traders (herding). Lakonishok, Shleifer, and Vishny (1992) find a weak evidence of herding behavior within the pension funds manager using based on quarterly data in NYSE. Even though there is an evidence of herding in small stocks, the magnitude of herding behavior is far from huge. On the other hand, Wermers (1999) uses mutual fund holding data and find an evidence of herding behavior of mutual funds in small and growth stocks.

Another literature explains about how individual investors herd one another. In contrast, Barber, Odean, and Zhu (2009) explain that individual is correlated in their trading and tend to herd. The results also supported by the research from Dorn, Huberman, and Sengmueller (2008) that explain individual investors trade similarly based on their sample data from the large discount brokerage in German.

Then, Kaniel, Saar, and Titman (2008) see a different perspective of how individual trade toward institutional. They also find that individual investors are contrarian toward institutional investors. The tendency of contrarian of individual investors leads them to act as liquidity provider for institutional investors that require immediacy. This argument is also supported by Grinblatt and Keloharju (2000) that find similar results in Finish stocks market.

Since there is different opinion regarding which investors herd more, Lakonishok, Shilffer, and Vishny (1992) give a logical explanation of why institution herding is more important than the individual investors. First, the institution will try to infer information about the quality of investment from one and another institution. As a result, the institution will have more understanding about each other trading than individuals so that they will herd to a greater extent (Shiller and Pound, 1989; Banerjee, 1992). Second, institutional investors have an incentive to hold the same stocks as another money manager to avoid falling behind a peer group performance (Scharfstein and Stein, 1990). Third, an institution might react to the same exogenous signal, and since the signal that is received by the institution is typically the same, they tend to herd more than individual investors.

5

Besides the explanation above there is also another literature that explains why money manager (institutional) do herd. Other models explain that institution may trade with the herd because of slowly diffusing private information (Froot, Scharfstein, and Stein, 1992; Hirshleifer, Subrahmanyam, and Titman, 1994; Hong and Stein, 1999), or career concerns (Scharfstein and Stein, 1990). This research believes that both of institution and individual investors perform herding behaviour to infer same information.

Price Impact

The third topic discusses the contemporaneous relationship between changes in ownership (usually proxied by players trading imbalances) and stocks return. There is a different time frame of analysis from quarterly data (Wermers, 1999) and annual data (Nofsinger and Sias, 1999). Sias, Starks, and Titman (2001) use covariance decomposition method to find out how institutional ownership changes in quarterly data could affect the daily return of stocks. Since this research will use microstructure perspective, the literature will be more related to the research that uses daily and intra-daily data.

There is a different perspective in microstructure horizon about which player, institutional or individual, that has a significant impact toward stock price. In 1993, Barclay and Warner (1993) discovered medium trade size that between 500 to 10,000 in one transaction has a price impact toward stocks price compare to another size. Accordingly, Chakravarty (2001) explains that medium trade size can impact the stock price because it is mainly initiated by institutional trade. Additionally, Chan and Lakonishok (1995) also find that a sequence of institutional block trades can give an effect on stock prices and explain that this link can be a result of institutional trading activity that could predict future return, contemporaneous stock return, or intra-quarter trend chasing of institutional. Contrarily, Foster, Gallagher, and Looi (2011) find different results in Australia. They conclude that neither a number of funds trading nor the volume of shares that are bought or sold by institutional investors correlated with the contemporaneous return of stocks. Their findings are also supported by Lakonishok, Shilffer, and Vishny (1992) who discover that institutional investors are neither stabilizing nor destabilizing stocks price in the US market.

Nonetheless, some literature captures significant findings that individual investors trade can affect stocks price. Using unique data set from Individual Investor Express Delivery Service in NYSE, Kaniel, Saar, and Titman (2008) find that individual investors trade (proxied by net individual trading) significantly can be used to forecast return. Moreover, Barber, Odean, and Zhu (2009) support previous finding by discovering that stocks that heavily buy (sell) in a week by individual investors ears strong (poor) returns in a subsequent week.

While those researches observed the impact of players (both institutional and individual investors) toward stock return independently, recent studies apply dynamics model to observe

6 this. Griffin, Harris, and Topaloglu (2003) use VAR model with five days lag and find that there is a strong contemporaneous relation between institutional trading and stock return at daily level while there is no evidence of individual trading. Furthermore, Ng and Wu (2007) put the same idea on their research in Shanghai stock market and report that only the trading activity from Chinese institutional and wealthiest individuals can affect future stock volatility, whereas other Chinese individual investors trade, in general, have no predictive power for stock future return. Stoffman (2014) also supports above argument by documenting that, in Finland, stock price, on average, will increase (decrease) due to institutional investors buy (sell) from individual investors. Also, if price move due to individual trade among themselves, the impact will quickly revert and vanish. Accordingly, this research believes that both institutional and individual investors transaction can affect stocks return.

INSTITUTIONAL BACKGROUND AND DATA

Institutional Background

The dataset of this study is coming from the Indonesia Stock Exchange (IDX) that was originally established in 1912 by Dutch colonials under the name of the Jakarta Stock Exchange (JSE) due to it is located in the Jakarta, the capital city of Indonesia. Later on, as the consequences of merging activities in 2007 between the JSE and the Surabaya Stock Exchange (SSE), the second stock market in Indonesia that was established in 1989 in Surabaya which intended for supporting the economic development in East Indonesia, the IDX is established and becoming the sole stock market in Indonesia (Aaron, Koesrindartoto, and Takashima, 2018). We provide the current landscape of the IDX in Table 1.

Based on the illustration, it is known that there are two general players in the market, namely institutional and individual investors, where institutional investors can be further divided into eight different types, such as corporations, financial institutions, securities firms, insurance firms, mutual funds, pension funds, foundations, and other institutions. Accordingly, it is obvious that institutional investors are dominating individual investors in the IDX in terms of equity ownership and trading value even if individual investors have greater number of players. Among the institutional players, corporations are the biggest player, while financial institutions and securities firms is placed in the second and third biggest player in terms of trading value, respectively. One should also note that sometimes the proportion of equity ownership and trading value might not be strongly correlated and therefore it needs to be analyzed carefully.

Data

7

This research uses the data from the IDX from January 2013 to December 2015. We provide our data description in Table 3. Moreover, the following are the details of information that our dataset comprises of:

1. Daily closing data which consist of stock code, board code, lowest price, highest price, opening price, closing price, total volume, date, and market capitalization. 2. Transaction data which consist the data consist of the transaction number, transaction date, transaction time, transaction board, transaction price, transaction lot, transaction value, buyer and seller broker ID, buyer and seller account ID, buyer and seller investor type, and transaction order number.

According to Table 2, it is known that during these full three years period, there are 726 trading days, 582 stocks, and more than 285 million past transactions that will be observed and analyzed. With such a big data (Over 25 GB), it requires sophisticated computational procedures to clean the data from inappropriate observations, such as missing data elements and outlier that may disrupt the quality of data. To do so, this study uses SQL, a programming language that is design specifically for storing and managing data.

METHODOLOGY

Variables Measurement

Portfolio Return

The return that will be used in this research is value weighted return based on stocks market cap in each day. To construct this variable, first, calculate the daily log return of each stock. Adjusted closing price is used to adjust the stock price due to corporate ownership action such as stock split, reverse stock, and to reissue. Then, by using market capitalization data, calculate the proportion of a particular stock at period t by dividing its market capitalization with total market capitalization of portfolio. Finally, the value weighted return can be calculated by aggregated the daily return of the stocks with their weight. We formalize this equation as follows:

푁

푟푝,푡 = ∑ 푤푖. 푟푖,푝,푡 (1) 푖=1 Where:

rp,t : Portfolio return at period t

wi,t : Weight of stock i at period t based on the proportion of its market capitalization in the portfolio at period t

rp,t : Return of stock i at period t

8

Trading Imbalances

As the proxy of trading activity, trading imbalances is used in this research following the research from (Barber, Odean, and Zhu, 2009; Foster, Gallagher, and Looi, 2011; Griffin, Harris, and Topaloglu, 2003; Ng and Wu, 2007). Trading imbalances variables for each type of investor can be easily calculated by by subtracting the total value buy with total value sell of each type of investor and divide it by its total transaction value. Accordingly, the range of this variable will be between -1 and 1. We then can interpret this trading imbalances in a very straightforward way, that is a positive (negative) sign is an implication of accumulation (distribution) process and the greater trading imbalances toward the certain sign, the greater accumulation or distribution that occur by the players. The equation for this calculation is originated by Griffin, Harris, and Topaloglu (2003) and as follows:

퐵푢푦푇푉푖,푡 − 푆푒푙푙푇푉푖,푡 푇푟푎푑푖푛푔 퐼푚푏푎푙푎푛푐푒푠푖,푡 = (2) 퐵푢푦푇푉푖,푡 + 푆푒푙푙푇푉푖,푡 Where:

BuyTVi,t : Buy trading value of investor i during period t

SellTVi,t : Sell trading value of investor i during period t

Estimation Methodology

Method Selection

Research in stocks market that is using microstructure has a various statistical approach to create a model and its inferences. In general, researchers have already got a sense about how the variables interact by analyzing descriptive statistics of the data. In static point of view, most of the researches take the basic idea of linear regression under the Fama-Machbeth procedure to create the relationship model between microstructure variables. One leading research by (Kaniel, Saar, and Titman, 2008) performs Fama-Machbeth procedure regression with adjusted standard error using Newey-West correction to analyze how the individual investors trading activity could affect stocks return. The Newey-West correction is used to accommodate the heteroscedasticity in the data. Close to the Kaniel, Saar, and Titman (2008) research, Barber, Odean, and Zhu (2009) also using Fama-Machbeth regression to analyze whether individual investors can move the market. Foster, Gallagher, and Looi (2011) also do a research in Australia using similar procedure but with a different focus. They concentrate on evaluating institutional trading and stocks return relationship.

Although Fama-Machbeth procedure regression is common in analyzing the relationship between investors trading activity and stocks return, the method is not appropriate to be used

9 in dynamics model. In a static model, we can only evaluate the direct interaction between investors trading and stocks return, however, dynamics model allowed us to assume all the variables depend on one and another. This condition required particular statistical method create inference.

For the market microstructure research, the common method to analyze the dynamics relationship is Vector Autoregressive (VAR) method. VAR is commonly used instead of Vector Error Correction Model (VECM) because of the contemporaneous characteristics of the trading activity variable and stocks return (Dorn, Huberman, and Sengmueller, 2008; Griffin, Harris, and Topaloglu, 2003; Hasbrouck, 2007). There is some researches in a top journal that use VAR to analyze dynamics relation in market microstructure research. The close literature to this study, Griffin, Harris, and Topaloglu (2003) and Dorn, Huberman, and Sengmueller (2008) use the VAR method to analyze the dynamics of individual, institutional and stocks return with lag 5. Recent research by Ben-Rephael, Kandel, and Wohl (2012) using the VAR method to evaluate the dynamics relation of equity funds manager flows and market return. They use four lags in the VAR model and create the impulse response to see how one standard deviation shocks in certain variables can affect the system. On the same year, Moskowitz, Ooi, and Pedersen (2012) research to analyze times series momentum within asset classes (equity, bond, and currencies) and its impact toward speculators trade. They use monthly bivariate VAR with 24 months lags of returns and changes in net speculator position, and as a robustness check 12 months lags are used. They also create the impulse response from the VAR model using Cholesky decomposition to estimate variance- covariance matrix of the residuals.

Nevertheless, although VAR is commonly used, there is a concern that has to be addressed. Supposed that there is a bivariate VAR with k lags in equation 1. The standard estimation for this VAR model can be done by maximum likelihood (asymptotic sample) or ordinary least square (finite sample) estimation. Based on the estimation vector of β and λ can be obtained with its standard error. However, this condition can be applied under the assumption that εt,R and εt,X has no heteroscedasticity and autocorrelation (white noise) (Hasbrouck, 1991). If one of the residual vectors in the system contains heteroscedasticity and autocorrelation, the assumption is violated, and inference of the model can be biased. While the coefficient of the estimation is robust, the standard error is the cause of bias due to miscalculation. To address this problem, Cochrane and Piazzesi (2005) propose a modified model on VAR estimation for bond securities. They still use maximum likelihood to estimate the VAR model but with adjusted heteroscedasticity and autocorrelation using Generalized Method of Moments (GMM) covariance estimator with adjusted Newey West standard error calculation. With this adjustment, the inference from VAR model is expected to be more accurate:

푘 푘 (3) 푅푡 = 훼 + ∑ 훽푖푅푡−푖 + ∑ 휆푖푋푡−푖 + 휀푡,푅 푖=1 푖=1

10

푘 푘

푋푡 = 훼 + ∑ 훽푖푅푡−푖 + ∑ 휆푖푋푡−푖 + 휀푡,푋 (4) 푖=1 푖=1 Based on all the above literature, this research will conduct a preliminary test to select the lags of VAR model and find whether there are heteroscedasticity and autocorrelation in VAR residuals. If the assumption of standard VAR is violated, then the inference will be discussed after adjusting the VAR model with NW standard error.

Vector Auto Regression Methodology

Vector Autoregressive is similar to univariate autoregressive. The intuition behind most results are similar and carries over by simply replacing scalar with matrices and scalar operation with matrix operation. The VAR system that will be built in this research is 3- variate VAR for general players and 10-variate VAR for detailed players. The optimum lag selection is based on the Akaike Information Criterion (AIC) and Likelihood Ratio (LR) tests following the idea from Griffin, Harris, and Topaloglu (2003). All variables in VAR equation are portfolio return and trading imbalance for each type of investor. In general matrix model, the system can be written as below:

푘

푌푡 = 훼 + ∑ 훷푌푡−푖 + 휀푡,푟 (5) 푖=1

Where Yt is a T by K variables matrix and Φ is a vector of parameters for the VAR systems.

In this research, Yt contains variable of portfolio return and trading imbalances of each investor type. The estimation of the coefficient and standard error from the system above will use maximum likelihood procedure. Maximum likelihood is believed to be more precise than conditional maximum likelihood and ordinary least square that does not require backtest of data or errors (Sheppard, 2013): 푇 푇 1 ℒ(휃|푦) = − ln(2휋) − ln(Σ) − 푣′Σ−1푣 (6) 2 2 2 ∑ is the covariance matrix of residuals and v is a matrix of the VAR residuals. The coefficient from the VAR is obtained by maximizing the likelihood function above. For the standard error, it is achieved from the square of diagonal in the covariance matrix:

−1 Σ휃 = Η (7) The covariance matrix of the coefficient is calculated by inversed the Hessian of maximum likelihood. However, if there are heteroscedasticity and autocorrelation in residuals, this procedure to calculate covariance matrix is not relevant anymore. This research believes that there is heteroscedasticity and autocorrelation in the data due to the high frequency of the data. To accommodate those issue, the covariance matrix should be adjusted by Newey West

11

(NW) covariance matrix. In general, NW covariance matrix follows the Generalized Method of Moment (GMM) procedure. GMM covariance matrix calculated by the formula below:

1 ′ Σ = 푑−1푆푑−1 (8) 휃 푇

Where d and S:

′ 푑 ≡ 퐸(푥푡푥푡) (9) ∞ ′ 푆 = ∑ 퐸(휀푡푥푡푥푡−푗휀푡−푗) (10) 푗=−∞ The adjustment of heteroscedasticity and autocorrelation is on the S matrix or precision matrix of GMM. Adjusted precision matrix by Newey West become:

푘 푘 − |푗| 푆 = ∑ 퐸(휀 푥 푥′ 휀 ) (11) 푘 푡 푡 푡−푗 푡−푗 푗=−푘 Where k is the lag of autocorrelation in residuals and (k-|j|)/k is called weighting matrix. So, the complete adjusted covariance will be:

푘 1 푘 − |푗| ′ Σ = 퐸(푥 푥′)−1 [ ∑ 퐸(휀 푥 푥′ 휀 )] 퐸(푥 푥′)−1 (12) 휃 푇 푡 푡 푘 푡 푡 푡−푗 푡−푗 푡 푡 푗=−푘 In most research, the lag of autocorrelation in residuals in determined by a mental model of how investor or traders look historical data. Cochrane and Piazzesi (2005) use 12 months lag and 18 months lag to check the consistency of their results. This research will choose term lag of 7 days since it satisfies the general rule of thumb formula 0.75T1/3 for the S matrix since the variables in this VAR system is contemporaneous. This lag also considers the trading indicator Moving Average indicator that is usually used by a trader in a short term.

VAR has two exclusive concepts for its analysis (Sheppard, 2013). First is Granger Causality (GC). GC is the standard method in VAR to determine whether one variable is useful in predicting other and evidence of Granger Causality is a good indicator that a VAR is needed. To test the GC, Wald test is used for this specification

푦푡 = Φ0 + Φ1푌푡−1 + Φ2푌푡−2 + ⋯ + Φ푝푌푡−푝 + 휖푡 (13)

{yj,t} does not granger cause {yi,t} if (H0 = Φ푖,푗,1 = Φ푖,푗,2 = ⋯ = Φ푖,푗,푃 = 0). Accordingly, the Wald statistics are written in the equation below and follow distribution

′ ′ −1 푊 = 푇[푅휃1 − 푄] [푅Ω푅 ] [푅휃1 − 푄] (14)

Where θ1 is the vector of unrestricted parameter estimates, Ω is the asymptotic covariance matrix of θ1 and R and Q are matrices based on the restrictions. Under the null hypothesis, the

12

Wald statistic is distributed asymptotically as χ2 where the degrees of freedom equal the number of zero restrictions being tested

The second concept that exclusive to VAR is impulse response function. In univariate time series, the ACF is sufficient to understand how the shocks decay. However, the condition is not the same when analyzing vector of data. A shock to a series of data not only has an immediate impact on that series but also affect other variables in the system which, in turn, can feedback to the original variables. After a few iterations of this cycle, it can be difficult to determine how a shock propagates even in a simple VAR (1) model. To accommodate this problem, impulse response function is created to see how the shocks of one vector variables affect others. Impulse response function can be illustrated through Vector Moving Average (VMA)

푦푡 = 휇 + ϵ푡 + Ξ1휖푡−1 + Ξ2휖푡−2 + ⋯ (15)

Using this VMA, the impulse response of yi with respect to a shock in ϵj is simply

{1, Ξ1[푖푖], Ξ2[푖푖], Ξ3[푖푖], … }. Then, Ξ1 is calculated by

Ξ1 = Φ1e푗 (16) The second will be

2 Ξ2 = Φ1e푗 + Φ2e푗 (17) The third is

2 Ξ3 = Φ1e푗 + Φ1Φ2e푗 + Φ2Φ1e푗 + Φ3e푗 (18)

This procedure can be continued to compute any Ξj up to specified steps observation ahead. From the VAR estimation, Granger Causality, and Impulse Response function, the relationship between institutional trading, individual trading, and stocks return can be clearly observed.

PRELIMINARY ANALYSIS

Before discussing the results and its analysis, the preliminary analysis will be presented to discuss the proper estimation environment. First, we determine the optimal lag selection using Akaike information Criterion (AIC) and sequential modified Likelihood Ratio (LR) test statistics. Accordingly, based on Table 3, lag 3 (based on AIC) and lag 6 (based on LR) are selected as the optimal lag for the general players, while lag 1 (based on AIC) and lag 8 (based on LR) are chosen for the detailed players. It is important to notice that AIC and LR might not give similar results due to the following reason. AIC tells us whether it pays to

13 have a richer model when the goal approximating the underlying data generating process the best we can in terms of Kullback-Leibler distance, whereas LR tells us whether at a chosen confidence level we can reject the hypothesis that some restrictions on the richer model hold. Therefore, it could be implied that AIC is preferable when the goal of our model is to forecast, while LR is more suitable when the goal of our model is to significance test. Given our research objectives, thus it could be inferred that LR is preferable for this study.

After we find the optimal for each case, we then test the autocorrelation issue for each selected lag using Lagrange Multiplier test with no autocorrelation at lag order as the null hypothesis as well the heteroscedasticity issue for each selected lag using White’s heteroscedasticity test with the variances for the errors are equal or no heteroscedasticity as the null hypothesis. Accordingly, Table 3 suggests that although there is no autocorrelation issue in lag (6) for the general players and lag (8) for the detailed players, all model exhibit heteroscedasticity problem so that the estimation of VAR should be adjusted with Newey- West correction for standard errors. The details of this diagnostic tests are provided in Table 3.

RESULTS OF GENERAL PLAYERS

Firstly, we investigate the dynamic behavior and trading strategies of the two general players in the market, namely individual and institutional investors. The estimation results are presented in Table 4.

According to the above table, there are several findings that are interested to be discussed. In term of price impact, it can be seen that both institutional and individual imbalances in lags 3 and 6 that significantly affect the return of stocks. The significant value is very strong and robust at 1% after the adjustment with lag of NW in 7. Statistically, 1% increase in institutional (individual) imbalances at t-1 can decrease around 9% (5%) of portfolio return at time t. This result is consistent with the results from Lakonishok, Shleifer, and Vishny (1992) and Foster, Gallagher, and Looi (2011). However, this is contrary to the results by Griffin, Harris, Topaloglu (2003), Ng and Wu (2007), and Stoffman (2014).

Continue on the trading behavior of each types of investors, it is known that individual investors are contrarian or anti-momentum traders, while institutional investors are momentum traders. These results are very strong since they are significant at 1% level. Moreover, this finding aligns with the findings of Barber and Odean (1999) as well Kaniel, Saar, and Titman (2008) for individual investors and Lakonishok, Shleifer, and Vishny (1992) as well Grinblatt, Titman and Wermers (1995) for institutional investors. The contrarian (momentum) behavior is considered as sell (buy) the winning stocks and buy (sell) the losing stocks according to Odean (1998). Considered only the lag 1, 1% increase in stocks return will decrease (increase) the imbalances of individual (institutional) investors around

14

175% (95%). It means that individual (institutional) will reverse (strengthen) the position that they have if there is an increase in stocks price.

Different result is observed in herding behavior on the past imbalances from each investor type. Individual investors imbalances at time t-3 and t-6 significantly affect in a positive way the imbalances at time t in 1%. This is an indication that they herd with their own group as well as their counterpart. Although both imbalances are significant, an increase in institutional imbalance at t-3 or t-6 will have higher magnitude effect than an increase in individual imbalance on individual imbalance at time t. Particularly, 1% increase in institutional imbalances t-1 will increase around 40% of individual imbalances at time t, while 1% increase in individual imbalances at time t-1 will increase about 20% of individual imbalances at time t.

Conversely, institutional investors imbalances at time t-3 and t-6 significantly affect in a negative way the imbalances at time t in 1%. This is an indication that they counter herd with their own group as well as their counterpart. Although both imbalances are significant, an increase in institutional imbalance at t-3 or t-6 will have higher magnitude effect than an increase in individual imbalance on individual imbalance at time t. Particularly, 1% increase in institutional imbalances t-1 will decrease around 20% of institutional imbalances at time t, while 1% increase in individual imbalances at time t-1 will decrease about 10% of institutional imbalances at time t.

As the robustness check to the significant in the VAR system, granger causality is performed to test simultaneously whether each variable has causality effect toward one and another. The results of granger causality for the general players are presented in Table 5.

Based on the table above, there is some evidence that consistent with the results above. First, both individual and institutional imbalances have granger cause the stock return. Second, past return is granger cause the individual and individual imbalances at time t. This result confirms the contrarian (momentum) trading behavior performed by individual (institutional) investors. Then, herding behavior done by individual investors is also fully confirmed by this test, whereas the counter herding behavior done by institutional investors is partially confirmed by this test since there is no evidence of granger causality between past individual imbalances with current institutional imbalances. As the explanation, even though individual imbalances might have enough evidence to affect the institutional imbalances, it is not sufficient to reject the null hypothesis of causality.

RESULTS OF DETAILED PLAYERS

After investigating the dynamic behavior and trading strategies of two general players in the market. This study then further breakdowns the general institutional investors into eight

15 different types in order to know in more detail the characteristic of each investor type. Those specific institutional investors are, corporations, financial institutions, securities firms, other institutions, insurance firms, mutual funds, pension funds, and foundations. Using the similar methodology, we present the estimation results and granger causality of this analysis in Tables 6 and 7.

According to the above table, it could be easily seen that the findings related to the trading behavior and strategy of individual investors remain the same as the former analysis. However, there are some cases where the findings related to the trading behavior and strategy of institutional investor are different with each specific investor type.

More specifically to the detailed institutional type model, while there is understandable behavior for the corporations, financial institutions, and securities firms, less intuitive behavior is observed for insurance firms, pension funds, and foundations. It might be because of both the number of respected institutions and empirical trading data are relatively low and not significant. Therefore, more observations are needed to be done to make a good behavior interpretation and policy recommendations. This is the primary agenda for future research.

CONCLUDING REMARKS

Individual investors, although only holding assets in fractions (6%–7%) compared to institutional (93%-94%), their activities in trading cannot be ignored. Empirical evidence shows that the total value of their transactions cannot be ignored since they contribute around one-third from the total transactions. Moreover, individual investors actions have strong relations with both individual and institutions investors action in the past. It is also discovered that individual investors also granger affected by both types of investors.

To be more detail, the individual types past action has a stronger cause to the current individual actions. Institutional investors' action in the past has the relations with both individual and institutional, however, the relations are stronger on institutional actions. Interestingly, institutional investors only granger affected by market return and institutional and not by individual investor past actions.

The effects of the previous market return to the individual and institutional investor can be seen by looking at the sign of the VAR model. The sign shows that while the institutional investor is significant and has a positive sign, the individual investor is significant and has a negative sign, this implies that the institutional investor employs the momentum strategy while individual uses contrarian strategy.

For the detailed institution model, the individual investors trading behaviors are robust compared to the general model. Using more detailed institutional investors, robust observation observed for the institutions that have significant trading values such as the

16 corporations, financial institutions, and security firms. Other institutions are observed to have mixed results, therefore need further analysis.

The aggregate individual investors tend to conduct daily trading activities at which it can cause high transaction costs. At the same time, individual investor tends to employ a contrarian strategy and in term of trading, this behavior might be classified as dispositional effects (Dharma and Koesrindartoto, 2018). Both activities might hinder the individual investor to obtain the better return from the market. Related to the current policy, at which to increase the number of individual investors, the strategy should be simultaneous with conducting the increasing the capital market literacy. It is also good to mention that the individual behavior findings are robust for the general model and for the detailed institutional type model

For the detailed institutional type model, while there is understandable behavior for the corporations, financial institutions, and securities firms, less intuitive behavior is observed for insurance firms, pension funds, and foundations. It might be because of both the number of respected institutions and empirical trading data are relatively low and not significant. More observations are needed to be done to make a good behavior interpretation and policy recommendations.

ACKNOWLEDGMENTS

All views expressed herein are solely those of the authors and do not necessarily reflect those of the Indonesia Capital Market and the Indonesia Financial Services Authority. The authors thank to the Indonesia Financial Services Authority for the funding support. All errors remain our responsibility.

17

REFERENCES

Aaron, A., Koesrindartoto, D.P., Takashima, R., 2018. Micro-foundation investigation of price manipulation in Indonesian capital market. Emerging Markets Finance and Trade 10, 1–15.

Agarwal, S., Chiu, I.-M., Liu, C., Rhee, S.G., 2011. The brokerage firm effect in herding: Evidence from Indonesia. Journal of Financial Research 34 (3), 461–479.

Agarwal, S., Faircloth, S., Liu, C., Ghon Rhee, S., 2009. Why do foreign investors underperform domestic investors in trading activities? Evidence from Indonesia. Journal of Financial Markets 12 (1), 32–53.

Badrinath, S.G., Wahal, S., 2016. Momentum trading by institutions. The Journal of Finance 71 (4), 1920.

Banerjee, A.V., 1992. A simple model of herd behavior. The Quarterly Journal of Economics 107 (3), 797–817.

Barber, B.M., Odean, T., 2000. Trading is hazardous to your wealth: The common stock investment performance of individual investors. The Journal of Finance 55 (2), 773– 806.

Barber, B.M., Odean, T., 2008. All that glitters: The effect of attention and news on the buying behavior of individual and institutional investors. Review of Financial Studies 21 (2), 785–818.

Barber, B.M., Odean, T., Zhu, N., 2009. Do retail trades move markets? Review of Financial Studies 22 (1), 151–186.

Barclay, M.J., Warner, J.B., 1993. Stealth trading and volatility. Journal of Financial Economics 34, 281–305.

Ben-Rephael, A., Kandel, S., Wohl, A., 2012. Measuring investor sentiment with mutual fund flows. Journal of Financial Economics 104 (2), 363–382.

Black, F., 1985. Noise. The Journal of Finance 41 (3), 281–305.

Bonser-Neal, C., Linnan, D., Neal, R., 1999. Emerging market transaction costs: Evidence from Indonesia. Pacific-Basin Finance Journal 7 (2), 103–127.

18

Brzeszczyński, J., Gajdka, J., Kutan, A.M., 2015. Investor response to public news, sentiment and institutional trading in emerging markets: A review. International Review of Economics and Finance 40, 338–352.

Chakravarty, S., 2001. Stealth-trading: Which traders' trades move stock prices? Journal of Financial Economics 61 (2), 289–307.

Chan, L.K.C., Lakonishok, J., 1995. The behavior of stock prices around institutional trades. The Journal of Finance 50 (4), 1147–1174.

Cochrane, J.H., Piazzesi, M., 2005. Bond risk premia. American Economic Review 95 (1), 138–160.

Comerton-Forde, Carole, 1999. Do trading rules impact on market efficiency? A comparison of opening procedures on the Australian and Jakarta stock exchanges. Pacific-Basin Finance Journal 7, 495–521.

Dharma, W.A., Koesrindartoto, D.P., 2018. Reversal on disposition effect: Evidence from Indonesian stock trader behavior. International Journal of Business and Society 19 (1), 233–244.

Dorn, D., Huberman, G., Sengmueller, P., 2008. Correlated trading and returns. The Journal of Finance 63 (2), 885–920.

Douglas Foster, F., Gallagher, D.R., Looi, A., 2011. Institutional trading and share returns. Journal of Banking and Finance 35 (12), 3383–3399.

Dvorak, T., 2005. Do domestic investors have an information advantage? Evidence from Indonesia. The Journal of Finance 60 (2), 817–839.

Froot, K.A., Scharfstein, D.S., Stein, J.C., 1992. Herd on the street: Informational inefficiencies in a market with short-term speculation. The Journal of Finance 47 (4), 1461–1484.

Griffin, J.M., Harris, J.H., Topaloglu, S., 2003. The dynamics of institutional and individual trading. The Journal of Finance 58 (6), 2285–2320.

Grinblatt, M., 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, 43–67.

19

Grinblatt, M., Titman, S., Wermers, R., 1995. Momentum investment strategies, portfolio performance, and herding: A study of mutual fund behavior. American Economic Review 85 (5), 1088–1105.

Hasbrouck, J., 1991. Measuring the Information Content of Stock Trades. The Journal of Finance 46 (1), 179.

Hasbrouck, J., 2007. Empirical Market Microstructure: The institutions, economics, and econometrics of securities trading. Oxford University Press, Inc., New York.

HIrshleifer, D., Subrahmanyam, A., Titman, S., 1994. Security analysis and trading patterns when some investors receive information before others. The Journal of Finance 49 (5), 1665–1698.

Hong, H., Stein, J.C., 1999. A Unified Theory of Underreaction, Momentum Trading, and Overreaction in Asset Markets. The Journal of Finance 54 (6), 2143–2184.

İmişiker, S., Özcan, R., Taş, B.K.O., 2015. Price Manipulation by Intermediaries. Emerging Markets Finance and Trade 51 (4), 788–797.

Kaniel, R., Saar, G., Titman, S., 2008. Individual investor trading and stock returns. The Journal of Finance 63 (1), 273–310.

Khwaja, A., Mian, A., 2005. Unchecked intermediaries: Price manipulation in an emerging stock market. Journal of Financial Economics 78 (1), 203–241.

Kyle, A.S., 1985. Continuous auctions and insider trading. Econometrica 53 (6), 1315–1336.

Lakonishok, J., Shleifer, A., Vishny, R.W., 1992. The impact of institutional trading on stock prices. Journal of Financial Economics 32 (1), 23–43.

Moskowitz, T.J., Ooi, Y.H., Pedersen, L.H., 2012. Time series momentum. Journal of Financial Economics 104 (2), 228–250.

Ng, L., Wu, F., 2007. The trading behavior of institutions and individuals in Chinese equity markets. Journal of Banking and Finance 31 (9), 2695–2710.

Nofsinger, J.R., Sias, R.W., 1999. Herding and Feedback Trading by Institutional and Individual Investors. The Journal of Finance 54 (6), 2263–2295.

Odean, T., 1998. Are investors reluctant to realize their losses. The Journal of Finance 53 (5), 1775–1798.

20

Scharfstein, D.S., Stein, J.C., 1990. Herd behavior and investment. American Economic Review 80 (3), 465-479.

Sheppard, K., 2013. Financial Econometrics Notes. University of Oxford, Oxford.

Shiller, R.J., Pound, J., 1989. Survey evidence on diffusion of interest and information among investors. Journal of Economic Behavior and Organization 12 (1), 47–66.

Sias, R.W., Starks, L.T., Titman, S., 2001. The price impact of institutional trading. SSRN Electronic Journal.

Stoffman, N., 2014. Who trades with whom? Individuals, institutions, and returns. Journal of Financial Markets 21, 50–75.

Wermers, R., 1999. Mutual Fund Herding and the Impact on Stock Prices. The Journal of Finance 54 (2), 581–622.

21

APPENDIX

22

Table 1. Landscape of the Indonesia Stock Exchange based on Investor Types

The table below gives the big picture of the Indonesia Stock Exchange (IDX) based on its investor types in 2015. Generally, investor types in the IDX can be categorized into individual and institutional investors, but in more specific institutional investors can be further divided into corporations, financial institutions, securities firms, other institutions, insurance firms, mutual funds, pension funds, and foundations. The detail of each investor type, such as its equity ownership, trading value, number of players, and average trading value of a player is described in table below. Note that other than the equity ownership data that we obtained from the Statistics of Indonesian Capital Market published by the Indonesian Financial Services Authority, the remaining contents of this table are derived from our data.

Equity Ownership Trading Value Number of Players Average Trading Value Investor Type as of 30 Dec 2015 in 2015 in 2015 of a Player in 2015 in trillion Rp in % in billion Rp in % in # in % (in billion Rp)

Individual Investors 173.65 6.51% 962,808.85 34.24% 151,617 98.61% 6.35

Institutional Investors 2,494.19 93.49% 1,849,113.09 65.76% 2,142 1.39% 863.26

Corporations 833.11 31.23% 770,248.30 27.39% 1,159 0.75% 664.58

Financial Institutions 309.51 11.60% 437,566.08 15.56% 123 0.08% 3,557.45

Securities Firms 230.71 8.65% 311,826.99 11.09% 122 0.08% 2,555.96

Other Institutions 413.73 15.51% 134,279.21 4.78% 158 0.10% 849.87

Insurance Firms 101.69 3.81% 100,136.06 3.56% 85 0.06% 1,178.07

Mutual Funds 418.66 15.69% 53,075.85 1.89% 223 0.15% 238.01

Pension Funds 180.73 6.77% 35,580.83 1.27% 221 0.14% 161.00

Foundations 6.05 0.23% 6,399.77 0.23% 51 0.03% 125.49

Total 2,667.84 100.00% 2,811,921.93 100.00% 153,759 100.00% 18.29

23

Table 2. Data summary

The table below gives the summary of data. The sample is in daily level spanning from 2013-2015. In overall, our data suggest that there are 726 trading days, 582 stocks, and more than 285 million transactions happened during our sample period that involve around 8,2 billion shares and around 8,700 trillion rupiah.

Trading Trading Stocks Trading Trading Value Period Volume Days Traded Frequency (in billion Rp) (in billion) 2013 240 485 73,105,756 2,632.13 2,972,772.82 Q1 60 451 19,393,710 749.69 751,915.62 Q2 59 455 19,550,760 717.81 893,518.60 Q3 61 462 18,983,014 597.89 724,901.16 Q4 60 470 15,178,272 566.74 602,437.43 2014 242 570 103,714,922 2,712.37 2,908,436.33 Q1 60 517 25,813,196 581.73 714,970.69 Q2 59 520 24,344,006 596.07 711,822.75 Q3 60 529 25,947,892 734.29 760,149.79 Q4 63 536 27,609,828 800.28 721,493.11 2015 244 582 108,558,876 2,917.01 2,811,921.93 Q1 62 534 28,807,152 816.08 816,296.24 Q2 61 534 26,570,562 747.32 739,468.62 Q3 60 538 25,127,206 629.87 565,480.00 Q4 61 544 28,053,956 723.75 690,677.07 2013-2015 726 582 285,379,554 8,261.51 8,693,131.08

24

Table 3. Diagnostic Tests

The table below gives the results of diagnostic tests, namely optimal lag selection, autocorrelation and heteroscedasticity tests. Panel A reports the results of the optimal lag selection using Akaike information Criterion (AIC) and sequential modified Likelihood Ratio (LR) test statistics. # indicates the lag order selected by the criterion. Accordingly, lag 3 (based on AIC) and lag 6 (based on LR) are selected as the optimal lag for the general players, while lag 1 (based on AIC) and lag 8 (based on LR) are chosen for the detailed players. Panel B reports the results of autocorrelation test for each selected lag using Lagrange Multiplier test with no autocorrelation at lag order as the null hypothesis. Panel C reports the results of heteroscedasticity test for each selected lag using White’s heteroscedasticity test with the variances for the errors are equal or no heteroscedasticity as the null hypothesis. * indicates the violation of null hypothesis for both autocorrelation and heteroscedasticity tests. The p-value is reported in the parantheses.

Panel A. Optimal Lag Selection General Players Detailed Players Lag

LR AIC LR AIC

0 NA -14.24 N/A -9.74 # 1 93.28 -14.34 689.30 -10.43

2 13.09 -14.34 162.52 -10.39 # 3 25.36 -14.34 124.60 -10.29

4 3.86 -14.33 107.88 -10.17

5 4.09 -14.31 122.76 -10.08 # 6 28.58 -14.33 113.34 -9.97

7 7.05 -14.31 98.129 -9.85 8 10.79 -14.30 124.57# -9.76 Panel B. Autocorrelation Test General Players Detailed Players Lag Lag 3 Lag 6 Lag 1 Lag 8 2.753 6.845 164.7* 96.09 1 (0.973) (0.653) (0.000) (0.591) 4.095 6.393 117.8 90.65 2 (0.905) (0.700) (0.107) (0.737) 12.24 14.37 112.2 80.46 3 (0.199) (0.109) (0.190) (0.924) 3.886 7.843 120.1 77.67 4 (0.918) (0.550) (0.082) (0.952) 6.058 10.34 122.7 113.7 5 (0.734) (0.323) (0.061) (0.164) 25.46* 9.075 103.1 97.28 6 (0.002) (0.430) (0.393) (0.558) 8.502 10.54 102.9 98.59 7 (0.484) (0.307) (0.398) (0.520) 8.462 8.5499 159.0* 98.21 8 (0.488) (0.479) (0.000) (0.531) Panel C. Heteroscedasticity Test General Players Detailed Players

Lag 3 Lag 6 Lag 1 Lag 8 Chi-Squared 780.8* 1848.2* 4118.9* 9032.8* (Joint-test) (0.000) (0.000) (0.000) (0.040)

25

Table 4. VAR for General Players

The table below gives the results of Vector Autoregressive (VAR) for general players with lag three in panel A and lag six in panel B under maximum likelihood procedure with adjusted heteroscedasticity and autocorrelation in residuals using Newey West (NW) standard errors using the following equation:

푘 푘 푘

푅퐸푇푡 = 훼 + ∑ 훽1푖푅퐸푇푡−푖 + ∑ 훽2푖퐼푁푆푡−푖 + ∑ 훽3푖퐼푁퐷푡−푖 + 휀푡,푅퐸푇 푖=1 푖=1 푖=1

푘 푘 푘

퐼푁푆푡 = 훼 + ∑ 훽1푖푅퐸푇푡−푖 + ∑ 훽2푖퐼푁푆푡−푖 + ∑ 훽3푖퐼푁퐷푡−푖 + 휀푡,퐼푁푆 푖=1 푖=1 푖=1

푘 푘 푘

퐼푁퐷푡 = 훼 + ∑ 훽1푖푅퐸푇푡−푖 + ∑ 훽2푖퐼푁푆푡−푖 + ∑ 훽3푖퐼푁퐷푡−푖 + 휀푡,퐼푁퐷 푖=1 푖=1 푖=1

Where RET is the value weighted portfolio return of all stocks listed in IDX, whereas INS and IND are the trading imbalances of the institution and individual investors. The sample is in daily level spanning from 2013- 2015. Table below reports the result of VAR (k) estimation for all time sample period (T) with adjusted 7 lags in NW standard error. The truncation parameter is determined by using the formula of 0.75T1/3. The standard error of parameters is reported in the parantheses. Wald statistics test is applied for the hypothesis testing. ***, **, * indicates the significance level at 10%, 5%, and 1%, respectively.

Panel A. VAR (3) for General Players RET INS IND (1) (2) (3) RET (-1) 0.075 0.947*** -1.757*** (0.046) (0.120) (0.255) RET (-2) -0.051 0.052 -0.362 (0.046) (0.116) (0.345) RET (-3) -0.087 0.042 -0.254 (0.057) (0.178) (0.320) INS (-1) 0.011 0.102 -0.077 (0.021) (0.089) (0.183) INS (-2) -0.003 0.099* -0.136 (0.013) (0.058) (0.092) INS (-3) -0.088*** -0.218*** 0.443*** (0.017) (0.062) (0.116) IND (-1) 0.003 0.018 0.029 (0.012) (0.049) (0.106) IND (-2) -0.006 0.031 -0.061 (0.006) (0.027) (0.042) IND (-3) -0.043*** -0.099*** 0.177*** (0.008) (0.033) (0.062) CONS 0.0005 0.002 -0.001 (0.000) (0.001) (0.002) df_r 713 713 713 df_m 9 9 9 F-stat 5.006 16.92 11.76 No of Obs. 723 723 723

26

Table 4. VAR for General Players (Continued)

Panel B. VAR (6) for General Players RET INS IND (4) (5) (6) RET (-1) 0.080* 0.965*** -1.808*** (0.048) (0.125) (0.268) RET (-2) -0.051 0.041 -0.356 (0.047) (0.118) (0.343) RET (-3) -0.111* 0.015 -0.160 (0.061) (0.184) (0.330) RET (-4) -0.041 -0.018 0.111 (0.046) (0.154) (0.284) RET (-5) 0.015 -0.091 0.0443 (0.059) (0.137) (0.228) RET (-6) -0.071 0.093 0.241 (0.052) (0.122) (0.363) INS (-1) 0.007 0.104 -0.064 (0.022) (0.086) (0.172) INS (-2) -0.006 0.086 -0.120 (0.014) (0.055) (0.084) INS (-3) -0.090*** -0.226*** 0.463*** (0.018) (0.064) (0.121) INS (-4) 0.018 0.091 -0.206 (0.017) (0.092) (0.143) INS (-5) 0.0239 -0.021 -0.042 (0.015) (0.090) (0.138) INS (-6) -0.081*** -0.149** 0.308*** (0.027) (0.061) (0.098) IND (-1) 0.001 0.018 0.036 (0.012) (0.047) (0.100) IND (-2) -0.009 0.023 -0.052 (0.006) (0.026) (0.039) IND (-3) -0.049*** -0.102*** 0.191*** (0.010) (0.035) (0.065) IND (-4) 0.006 0.038 -0.101 (0.008) (0.047) (0.075) IND (-5) 0.013* 0.002 -0.030 (0.008) (0.050) (0.076) IND (-6) -0.053*** -0.096*** 0.204*** (0.014) (0.026) (0.049) CONS 0.000 0.002* -0.001 (0.000) (0.001) (0.002) df_r 701 701 701 df_m 18 18 18 F-stat 4.491 9.891 7.531 No of Obs. 720 720 720

27

Table 5. Granger Causality for General Players

The table below gives the results of Granger causality test for general players based on VAR (3) as reported in Panel A and VAR (6) in Panel B. RET is the value weighted portfolio return of all stocks listed in IDX, whereas INS and IND are the trading imbalances of the institution and individual investors. The null hypothesis of this test is that lagged values of x do not explain the variation in y, or in other words x does not granger cause y. The p-value of parameters is reported in the parantheses. ***, **, * indicates the significance level at 10%, 5%, and 1%, respectively.

Panel A. Granger Causality for General Players based on VAR (3)

Effect (t) Variables RET INS IND 2.296** 16.33*** 16.94*** RET (0.033) (0.000) (0.000) Cause 4.233*** 9.381*** 3.12** INS (t-i) (0.005) (0.000) (0.025) 4.137*** 1.817 10.44*** IND (0.006) (0.142) (0.000) Panel B. Granger Causality for General Players based on VAR (6)

Effect (t) Variables RET INS IND 2.546*** 8.750*** 9.125*** RET (0.002) (0.000) (0.000) Cause 4.044*** 5.359*** 2.559** INS (t-i) (0.000) (0.000) (0.018) 4.857*** 1.778 5.894*** IND (0.000) (0.101) (0.000)

28

Table 6. VAR for Detailed Players

The table below gives the results of Vector Autoregressive (VAR) for detailed players with lag one in panel A and lag eight in panel B under maximum likelihood procedure with adjusted heteroscedasticity and autocorrelation in residuals using Newey West (NW) standard errors using the following equation: 푘 푘 푘 푘 푘 푘 푘 푘 푘 푘

푅퐸푇푡 = 훼 + ∑ 훽1푖 푅퐸푇푡−푖 + ∑ 훽2푖 퐶푃푡−푖 + ∑ 훽3푖퐹퐷푡−푖 + ∑ 훽4푖 퐼퐵푡−푖 + ∑ 훽5푖퐼퐷푡−푖 + ∑ 훽6푖퐼푆푡−푖 + ∑ 훽7푖푀퐹푡−푖 + ∑ 훽8푖푂푇푡−푖 + ∑ 훽9푖푃퐹푡−푖 + ∑ 훽10푖 푆퐶푡−푖 + 휀푡,푅퐸푇 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푘 푘 푘 푘 푘 푘 푘 푘 푘 푘

퐶푃푡 = 훼 + ∑ 훽1푖 푅퐸푇푡−푖 + ∑ 훽2푖 퐶푃푡−푖 + ∑ 훽3푖퐹퐷푡−푖 + ∑ 훽4푖 퐼퐵푡−푖 + ∑ 훽5푖퐼퐷푡−푖 + ∑ 훽6푖퐼푆푡−푖 + ∑ 훽7푖푀퐹푡−푖 + ∑ 훽8푖푂푇푡−푖 + ∑ 훽9푖푃퐹푡−푖 + ∑ 훽10푖 푆퐶푡−푖 + 휀푡,푅퐸푇 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푘 푘 푘 푘 푘 푘 푘 푘 푘 푘

퐹퐷푡 = 훼 + ∑ 훽1푖 푅퐸푇푡−푖 + ∑ 훽2푖퐶푃푡−푖 + ∑ 훽3푖퐹퐷푡−푖 + ∑ 훽4푖 퐼퐵푡−푖 + ∑ 훽5푖퐼퐷푡−푖 + ∑ 훽6푖퐼푆푡−푖 + ∑ 훽7푖푀퐹푡−푖 + ∑ 훽8푖푂푇푡−푖 + ∑ 훽9푖푃퐹푡−푖 + ∑ 훽10푖 푆퐶푡−푖 + 휀푡,푅퐸푇 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푘 푘 푘 푘 푘 푘 푘 푘 푘 푘

퐼퐵푡 = 훼 + ∑ 훽1푖푅퐸푇푡−푖 + ∑ 훽2푖퐶푃푡−푖 + ∑ 훽3푖 퐹퐷푡−푖 + ∑ 훽4푖퐼퐵푡−푖 + ∑ 훽5푖 퐼퐷푡−푖 + ∑ 훽6푖 퐼푆푡−푖 + ∑ 훽7푖푀퐹푡−푖 + ∑ 훽8푖푂푇푡−푖 + ∑ 훽9푖푃퐹푡−푖 + ∑ 훽10푖푆퐶푡−푖 + 휀푡,푅퐸푇 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 퐾 푘 푘 푘 푘 푘 푘 푘 푘 푘

퐼퐷푡 = 훼 + ∑ 훽1푖푅퐸푇푡−푖 + ∑ 훽2푖퐶푃푡−푖 + ∑ 훽3푖퐹퐷푡−푖 + ∑ 훽4푖퐼퐵푡−푖 + ∑ 훽5푖퐼퐷푡−푖 + ∑ 훽6푖퐼푆푡−푖 + ∑ 훽7푖푀퐹푡−푖 + ∑ 훽8푖 푂푇푡−푖 + ∑ 훽9푖푃퐹푡−푖 + ∑ 훽10푖푆퐶푡−푖 + 휀푡,푅퐸푇 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푘 푘 푘 푘 푘 푘 푘 푘 푘 푘

퐼푆푡 = 훼 + ∑ 훽1푖푅퐸푇푡−푖 + ∑ 훽2푖 퐶푃푡−푖 + ∑ 훽3푖 퐹퐷푡−푖 + ∑ 훽4푖 퐼퐵푡−푖 + ∑ 훽5푖퐼퐷푡−푖 + ∑ 훽6푖퐼푆푡−푖 + ∑ 훽7푖 푀퐹푡−푖 + ∑ 훽8푖푂푇푡−푖 + ∑ 훽9푖푃퐹푡−푖 + ∑ 훽10푖 푆퐶푡−푖 + 휀푡,푅퐸푇 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푘 푘 푘 푘 푘 푘 푘 푘 푘 푘

푀퐹푡 = 훼 + ∑ 훽1푖 푅퐸푇푡−푖 + ∑ 훽2푖퐶푃푡−푖 + ∑ 훽3푖퐹퐷푡−푖 + ∑ 훽4푖퐼퐵푡−푖 + ∑ 훽5푖퐼퐷푡−푖 + ∑ 훽6푖퐼푆푡−푖 + ∑ 훽7푖푀퐹푡−푖 + ∑ 훽8푖푂푇푡−푖 + ∑ 훽9푖푃퐹푡−푖 + ∑ 훽10푖 푆퐶푡−푖 + 휀푡,푅퐸푇 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푘 푘 푘 푘 푘 푘 푘 푘 푘 푘

푂푇푡 = 훼 + ∑ 훽1푖 푅퐸푇푡−푖 + ∑ 훽2푖 퐶푃푡−푖 + ∑ 훽3푖퐹퐷푡−푖 + ∑ 훽4푖 퐼퐵푡−푖 + ∑ 훽5푖퐼퐷푡−푖 + ∑ 훽6푖퐼푆푡−푖 + ∑ 훽7푖푀퐹푡−푖 + ∑ 훽8푖푂푇푡−푖 + ∑ 훽9푖푃퐹푡−푖 + ∑ 훽10푖 푆퐶푡−푖 + 휀푡,푅퐸푇 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 퐾 푘 푘 푘 푘 푘 푘 푘 푘 푘

푃퐹푡 = 훼 + ∑ 훽1푖푅퐸푇푡−푖 + ∑ 훽2푖퐶푃푡−푖 + ∑ 훽3푖퐹퐷푡−푖 + ∑ 훽4푖퐼퐵푡−푖 + ∑ 훽5푖 퐼퐷푡−푖 + ∑ 훽6푖 퐼푆푡−푖 + ∑ 훽7푖푀퐹푡−푖 + ∑ 훽8푖 푂푇푡−푖 + ∑ 훽9푖푃퐹푡−푖 + ∑ 훽10푖푆퐶푡−푖 + 휀푡,푅퐸푇 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 퐾 푘 푘 푘 푘 푘 푘 푘 푘 푘

푆퐶푡 = 훼 + ∑ 훽1푖푅퐸푇푡−푖 + ∑ 훽2푖퐶푃푡−푖 + ∑ 훽3푖퐹퐷푡−푖 + ∑ 훽4푖퐼퐵푡−푖 + ∑ 훽5푖 퐼퐷푡−푖 + ∑ 훽6푖퐼푆푡−푖 + ∑ 훽7푖푀퐹푡−푖 + ∑ 훽8푖 푂푇푡−푖 + ∑ 훽9푖푃퐹푡−푖 + ∑ 훽10푖푆퐶푡−푖 + 휀푡,푅퐸푇 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1 푖=1

Where RET is the value weighted portfolio return of all stocks listed in IDX, whereas CP, FD, IB, ID, IS, MF, OT, PF, and SC is the trading imbalances of the corporations, foundations, financial institutions, individual investors, insurance firms, mutual funds, other institutions, pension funds, and securities firms, respectively. The sample is in

29 daily level spanning from 2013-2015. Table below reports the result of VAR (k) estimation for all time sample period (T) with adjusted 7 lags in NW standard error. The truncation parameter is determined by using the formula of 0.75T1/3. The standard error of parameters is reported in the parantheses. Wald statistics test is applied for the hypothesis testing. ***, **, * indicates the significance level at 10%, 5%, and 1%, respectively.

Table 6. VAR for Detailed Players (Continued)

Panel A. VAR (1) for Detailed Players RET CP FD IB ID IS MF OT PF SC (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) RET (-1) 0.078 0.613 -2.134 0.174 -1.458*** 0.931 0.257 2.061* -2.079 1.342* (0.057) (0.551) (1.954) (0.527) (0.310) (1.566) (1.138) (1.125) (1.368) (0.765) CP (-1) 0.001 0.158** -0.163 -0.102* -0.010 0.154 0.026 0.101 0.038 -0.060 (0.004) (0.063) (0.215) (0.058) (0.029) (0.185) (0.122) (0.136) (0.193) (0.079) FD (-1) -0.001 -0.001 -0.028 -0.031** 0.0022 0.037 -0.056* 0.054* 0.040 0.0053 (0.001) (0.012) (0.039) (0.014) (0.006) (0.040) (0.031) (0.028) (0.036) (0.019) IB (-1) 0.005 -0.049 -0.063 0.157*** -0.019 -0.142 -0.246** 0.142 -0.133 0.075 (0.003) (0.041) (0.166) (0.043) (0.023) (0.134) (0.096) (0.105) (0.135) (0.049) ID (-1) 0.008 0.099 0.196 -0.004 0.060 0.012 -0.159 0.310* -0.211 -0.045 (0.008) (0.080) (0.300) (0.088) (0.044) (0.315) (0.243) (0.181) (0.274) (0.123) IS (-1) 0.000 -0.007 0.000 -0.066*** 0.008 0.554*** -0.015 -0.030 0.037 -0.082*** (0.001) (0.016) (0.044) (0.016) (0.006) (0.051) (0.034) (0.035) (0.042) (0.021) MF (-1) 0.002 0.048** -0.035 -0.039** 0.008 0.009 0.059 -0.012 -0.044 -0.017 (0.001) (0.018) (0.047) (0.020) (0.008) (0.045) (0.044) (0.030) (0.044) (0.026) OT (-1) 0.001 -0.010 -0.040 -0.007 0.003 -0.009 -0.056 0.240*** -0.056 0.009 (0.001) (0.018) (0.063) (0.021) (0.009) (0.067) (0.046) (0.043) (0.061) (0.035) PF (-1) 0.000 0.010 0.068 -0.014 0.004 -0.077 0.093** -0.036 0.184*** -0.001 (0.001) (0.016) (0.057) (0.021) (0.008) (0.051) (0.043) (0.037) (0.051) (0.023) SC (-1) 0.002 -0.063 -0.023 0.045 -0.005 -0.177 -0.124 0.154** -0.090 0.201*** (0.002) (0.046) (0.113) (0.037) (0.018) (0.109) (0.086) (0.077) (0.091) (0.067) CONS 0.000 -0.001 0.025 0.000 -0.003 0.037** 0.058*** -0.006 -0.007 0.005 (0.000) (0.005) (0.016) (0.005) (0.002) (0.017) (0.013) (0.011) (0.014) (0.008) df_r 714 714 714 714 714 714 714 714 714 714 df_m 10 10 10 10 10 10 10 10 10 10 F-stat 1.100 5.199 2.360 10.79 8.418 33.94 4.824 10.05 9.969 14.43 No of Obs. 725 725 725 725 725 725 725 725 725 725

30

Table 6. VAR for Detailed Players (Continued)

Panel B. VAR (8) for Detailed Players – Independent Variables: Return (RETt-i) RET CP FD IB ID IS MF OT PF SC (17) (18) (19) (20) (21) (22) (23) (24) (25) (26) RET (-1) 0.063 0.067 -3.205 0.534 -1.426*** -1.550 -0.064 2.751** -3.193* 2.347*** (0.057) (0.562) (2.155) (0.593) (0.348) (1.673) (1.306) (1.311) (1.666) (0.824) RET (-2) -0.057 -0.157 4.964** -0.894 -0.327 1.555 -1.367 0.263 1.804 0.117 (0.062) (0.594) (2.361) (0.691) (0.443) (1.678) (1.293) (1.628) (1.796) (0.837) RET (-3) -0.153* 0.453 4.144* -1.974*** 0.189 2.016 0.979 0.146 0.463 -0.353 (0.079) (0.586) (2.319) (0.636) (0.311) (1.784) (1.289) (1.220) (1.875) (0.877) RET (-4) -0.036 -0.682 -3.324 -0.341 0.246 1.237 0.241 1.698 -0.734 -0.390 (0.063) (0.657) (2.393) (0.711) (0.378) (1.741) (1.301) (1.298) (1.795) (0.921) RET (-5) -0.020 -0.605 -1.450 -0.601 0.332 0.899 2.414* 0.068 0.614 -0.232 (0.075) (0.589) (2.129) (0.733) (0.259) (1.905) (1.248) (1.364) (1.978) (0.902) RET (-6) -0.006 -0.355 1.323 0.153 -0.064 -2.463 1.491 -0.622 -0.287 2.299** (0.066) (0.586) (2.271) (0.763) (0.459) (2.066) (1.549) (1.340) (1.749) (0.964) RET (-7) 0.011 -0.409 -0.148 -0.452 -0.010 2.541 -1.354 0.799 -0.681 0.116 (0.077) (0.605) (2.271) (0.637) (0.412) (1.885) (1.550) (1.495) (1.721) (0.834) RET (-8) -0.009 -1.410** 1.073 -0.067 0.933** 2.124 0.611 -0.317 0.623 0.153 (0.058) (0.619) (2.228) (0.612) (0.421) (1.992) (1.358) (1.393) (1.448) (0.778)

31

Table 6. VAR for Detailed Players (Continued)

Panel B. VAR (8) for Detailed Players – Independent Variables: Corporations (CPt-i) RET CP FD IB ID IS MF OT PF SC (17) (18) (19) (20) (21) (22) (23) (24) (25) (26) CP (-1) 0.001 0.099 -0.135 -0.034 -0.003 -0.017 -0.049 0.092 -0.058 -0.008 (0.005) (0.076) (0.220) (0.061) (0.031) (0.174) (0.134) (0.143) (0.187) (0.078) CP (-2) 0.003 0.024 0.0513 0.021 -0.063** 0.128 0.075 -0.069 0.048 0.138* (0.004) (0.063) (0.180) (0.063) (0.029) (0.216) (0.122) (0.156) (0.162) (0.080) CP (-3) -0.016*** 0.005 0.445** -0.088 0.082** -0.025 -0.056 0.126 0.518*** -0.105 (0.005) (0.065) (0.195) (0.098) (0.034) (0.161) (0.144) (0.126) (0.128) (0.092) CP (-4) -0.001 0.090** -0.016 -0.119** -0.024 0.165 0.290*** -0.130 -0.020 0.142 (0.005) (0.044) (0.203) (0.057) (0.032) (0.192) (0.109) (0.147) (0.176) (0.102) CP (-5) 0.007 0.103* 0.021 0.000 -0.024 -0.309* -0.192 0.003 -0.324* -0.061 (0.006) (0.060) (0.195) (0.069) (0.030) (0.166) (0.146) (0.129) (0.179) (0.099) CP (-6) -0.002 0.044 0.049 -0.135** 0.012 0.047 0.184 0.175 -0.024 -0.110 (0.006) (0.047) (0.216) (0.066) (0.040) (0.181) (0.134) (0.123) (0.195) (0.082) CP (-7) 0.003 -0.081 -0.133 0.042 -0.012 0.147 0.042 0.153 -0.172 0.038 (0.005) (0.060) (0.218) (0.066) (0.033) (0.172) (0.153) (0.125) (0.190) (0.073) CP (-8) -0.002 0.051 0.125 0.089 0.006 -0.131 0.062 0.007 0.229 -0.258*** (0.004) (0.061) (0.224) (0.063) (0.033) (0.170) (0.116) (0.117) (0.189) (0.082)

32

Table 6. VAR for Detailed Players (Continued)

Panel B. VAR (8) for Detailed Players – Independent Variables: Foundations (FDt-i) RET CP FD IB ID IS MF OT PF SC (17) (18) (19) (20) (21) (22) (23) (24) (25) (26) FD (-1) -0.001 -0.001 -0.055 -0.031** 0.000 0.044 -0.050 0.055* 0.051 0.005 (0.001) (0.014) (0.045) (0.015) (0.006) (0.040) (0.032) (0.028) (0.038) (0.019) FD (-2) 0.001 0.023* -0.095** -0.004 -0.004 -0.046 0.002 0.020 0.009 0.0190 (0.001) (0.013) (0.048) (0.015) (0.006) (0.041) (0.029) (0.026) (0.041) (0.019) FD (-3) -0.001 0.014 0.034 -0.010 -0.002 0.006 0.043 0.031 0.084* -0.024 (0.001) (0.012) (0.047) (0.016) (0.008) (0.040) (0.031) (0.034) (0.043) (0.018) FD (-4) 0.001 -0.001 -0.031 -0.031** 0.001 -0.027 0.028 0.025 -0.067 0.015 (0.001) (0.012) (0.042) (0.013) (0.006) (0.042) (0.031) (0.030) (0.042) (0.022) FD (-5) 0.002* -0.019 -0.045 0.033** -0.010 -0.054 0.047* 0.070** -0.091** 0.005 (0.001) (0.013) (0.041) (0.014) (0.006) (0.039) (0.028) (0.028) (0.037) (0.020) FD (-6) 0.001 -0.004 0.093** -0.017 -0.002 0.007 0.019 0.021 -0.028 0.007 (0.001) (0.012) (0.044) (0.013) (0.005) (0.042) (0.034) (0.030) (0.037) (0.017) FD (-7) 0.000 0.000 0.037 -0.010 -0.003 0.067 -0.026 -0.016 -0.046 -0.026 (0.001) (0.012) (0.045) (0.016) (0.009) (0.045) (0.031) (0.033) (0.042) (0.019) FD (-8) -0.001 -0.005 0.028 0.010 0.015* 0.016 -0.054* -0.012 -0.029 -0.041** (0.001) (0.014) (0.042) (0.015) (0.008) (0.044) (0.030) (0.024) (0.039) (0.019)

33

Table 6. VAR for Detailed Players (Continued)

Panel B. VAR (8) for Detailed Players – Independent Variables: Financial Institutions (IBt-i) RET CP FD IB ID IS MF OT PF SC (17) (18) (19) (20) (21) (22) (23) (24) (25) (26) IB (-1) 0.006* -0.045 -0.129 0.152*** -0.019 -0.043 -0.254*** 0.148 -0.158 0.036 (0.004) (0.041) (0.168) (0.044) (0.025) (0.135) (0.094) (0.111) (0.140) (0.060) IB (-2) 0.004 -0.018 0.047 0.041 -0.024 -0.013 0.075 0.103 -0.020 0.118* (0.004) (0.045) (0.143) (0.043) (0.021) (0.149) (0.090) (0.087) (0.117) (0.065) IB (-3) -0.008* 0.002 0.135 0.004 0.013 0.086 -0.020 0.017 0.167 0.0119 (0.004) (0.044) (0.148) (0.055) (0.029) (0.128) (0.109) (0.093) (0.123) (0.064) IB (-4) 0.000 -0.023 -0.034 0.003 0.016 0.072 0.002 0.058 -0.001 0.052 (0.003) (0.039) (0.136) (0.039) (0.019) (0.122) (0.093) (0.113) (0.110) (0.066) IB (-5) 0.002 0.109** 0.058 -0.010 -0.021 -0.349** -0.096 -0.022 -0.164 0.030 (0.003) (0.047) (0.140) (0.048) (0.020) (0.141) (0.106) (0.090) (0.120) (0.061) IB (-6) -0.002 0.019 -0.177 -0.083* 0.018 0.009 0.158* -0.009 -0.056 -0.092 (0.004) (0.037) (0.157) (0.048) (0.023) (0.139) (0.085) (0.086) (0.130) (0.067) IB (-7) -0.001 -0.123*** 0.034 0.039 0.014 0.227* 0.005 0.112 0.118 -0.074 (0.004) (0.041) (0.141) (0.054) (0.022) (0.123) (0.110) (0.089) (0.131) (0.063) IB (-8) 0.000 0.008 0.053 0.067 -0.006 -0.004 0.002 -0.123 0.069 -0.087 (0.004) (0.038) (0.163) (0.049) (0.022) (0.141) (0.088) (0.081) (0.127) (0.062)

34

Table 6. VAR for Detailed Players (Continued)

Panel B. VAR (8) for Detailed Players – Independent Variables: Individual Investors (IDt-i) RET CP FD IB ID IS MF OT PF SC (17) (18) (19) (20) (21) (22) (23) (24) (25) (26) ID (-1) 0.004 0.043 0.097 -0.003 0.071 -0.074 -0.396 0.305 -0.395 0.021 (0.008) (0.084) (0.344) (0.088) (0.050) (0.286) (0.250) (0.205) (0.301) (0.143) ID (-2) -0.001 -0.038 0.838*** 0.109 -0.084 0.104 0.301 0.287 -0.020 0.004 (0.007) (0.097) (0.291) (0.104) (0.051) (0.281) (0.213) (0.194) (0.245) (0.137) ID (-3) -0.024** 0.059 0.401 -0.217* 0.038 0.142 0.078 0.067 0.513* -0.062 (0.012) (0.111) (0.458) (0.117) (0.097) (0.298) (0.268) (0.247) (0.277) (0.154) ID (-4) -0.001 0.027 -0.306 -0.096 -0.002 0.087 0.410* 0.145 0.063 -0.047 (0.008) (0.075) (0.366) (0.089) (0.059) (0.311) (0.234) (0.222) (0.258) (0.119) ID (-5) 0.011 -0.011 0.094 0.082 -0.047 -0.487* -0.701*** -0.244 0.037 0.289 (0.009) (0.126) (0.324) (0.116) (0.051) (0.276) (0.231) (0.208) (0.278) (0.196) ID (-6) -0.017 0.048 0.220 -0.115 0.063 -0.185 0.596*** -0.032 0.059 -0.102 (0.013) (0.080) (0.392) (0.109) (0.068) (0.405) (0.223) (0.188) (0.374) (0.157) ID (-7) -0.002 -0.154* 0.164 0.096 0.048 0.603** -0.507** 0.096 0.151 -0.050 (0.008) (0.086) (0.407) (0.095) (0.048) (0.248) (0.226) (0.172) (0.256) (0.127) ID (-8) 0.004 -0.024 0.411 -0.014 0.040 -0.120 0.091 -0.122 0.544* -0.033 (0.007) (0.106) (0.378) (0.120) (0.050) (0.289) (0.178) (0.171) (0.307) (0.148)

35

Table 6. VAR for Detailed Players (Continued)

Panel B. VAR (8) for Detailed Players – Independent Variables: Insurance Firms (ISt-i) RET CP FD IB ID IS MF OT PF SC (17) (18) (19) (20) (21) (22) (23) (24) (25) (26) IS (-1) 0.001 -0.015 -0.004 -0.037* 0.004 0.358*** -0.038 -0.027 -0.016 -0.040* (0.001) (0.019) (0.053) (0.019) (0.007) (0.050) (0.041) (0.037) (0.045) (0.022) IS (-2) 0.000 -0.010 0.053 0.007 -0.015 0.133** 0.041 -0.003 0.016 0.007 (0.001) (0.019) (0.055) (0.018) (0.009) (0.053) (0.041) (0.040) (0.049) (0.022) IS (-3) -0.001 -0.002 0.038 -0.008 0.016* 0.037 -0.053 0.039 0.056 -0.001 (0.001) (0.017) (0.061) (0.020) (0.009) (0.047) (0.043) (0.035) (0.055) (0.024) IS (-4) -0.001 0.003 -0.020 -0.013 -0.002 0.087 0.007 0.005 0.026 -0.015 (0.001) (0.017) (0.054) (0.020) (0.009) (0.054) (0.045) (0.037) (0.047) (0.026) IS (-5) -0.001 0.000 -0.030 0.014 -0.006 -0.013 0.015 -0.010 -0.064 0.014 (0.001) (0.018) (0.060) (0.020) (0.010) (0.052) (0.040) (0.037) (0.055) (0.025) IS (-6) 0.000 0.025 -0.017 -0.028 0.001 -0.034 0.037 -0.002 -0.024 -0.003 (0.001) (0.016) (0.053) (0.019) (0.011) (0.051) (0.043) (0.039) (0.052) (0.024) IS (-7) -0.001 -0.023 0.026 0.001 -0.001 0.087* -0.086** 0.075* 0.038 -0.023 (0.001) (0.019) (0.066) (0.021) (0.008) (0.053) (0.040) (0.043) (0.057) (0.026) IS (-8) 0.000 -0.002 0.063 0.004 0.002 0.065 0.092*** -0.043 0.021 -0.035 (0.001) (0.017) (0.061) (0.017) (0.007) (0.051) (0.035) (0.037) (0.049) (0.024)

36

Table 6. VAR for Detailed Players (Continued)

Panel B. VAR (8) for Detailed Players – Independent Variables: Mutual Funds (MFt-i) RET CP FD IB ID IS MF OT PF SC (17) (18) (19) (20) (21) (22) (23) (24) (25) (26) MF (-1) 0.002 0.039** -0.034 -0.023 0.005 -0.064 0.047 0.008 -0.082* 0.009 (0.001) (0.018) (0.058) (0.022) (0.009) (0.047) (0.042) (0.033) (0.048) (0.027) MF (-2) 0.000 0.008 -0.024 0.009 -0.006 0.054 0.097** -0.032 0.0075 0.015 (0.001) (0.015) (0.056) (0.016) (0.008) (0.050) (0.041) (0.041) (0.050) (0.021) MF (-3) -0.002* -0.017 -0.028 -0.013 0.003 0.147*** -0.023 -0.001 0.110** -0.010 (0.001) (0.017) (0.055) (0.018) (0.009) (0.055) (0.038) (0.033) (0.051) (0.026) MF (-4) -0.001 0.000 0.015 0.008 0.009 0.063 0.018 -0.062* 0.0071 -0.020 (0.001) (0.015) (0.058) (0.019) (0.009) (0.050) (0.041) (0.036) (0.051) (0.021) MF (-5) 0.000 0.003 -0.042 -0.005 -0.003 -0.060 0.091** 0.011 0.045 0.011 (0.001) (0.021) (0.051) (0.021) (0.008) (0.057) (0.041) (0.041) (0.053) (0.024) MF (-6) -0.001 0.024 0.009 -0.039** -0.001 0.025 0.077* -0.001 -0.009 -0.038 (0.001) (0.016) (0.053) (0.017) (0.009) (0.055) (0.043) (0.042) (0.051) (0.025) MF (-7) 0.000 -0.025 -0.014 0.042** 0.001 0.008 -0.051 0.012 -0.004 0.009 (0.001) (0.016) (0.054) (0.018) (0.007) (0.049) (0.041) (0.037) (0.047) (0.023) MF (-8) -0.001 0.036** 0.067 -0.029 0.002 -0.063 0.113*** -0.038 -0.019 -0.053** (0.001) (0.016) (0.064) (0.022) (0.008) (0.050) (0.039) (0.034) (0.053) (0.023)

37

Table 6. VAR for Detailed Players (Continued)

Panel B. VAR (8) for Detailed Players – Independent Variables: Other Institutions (OTt-i) RET CP FD IB ID IS MF OT PF SC (17) (18) (19) (20) (21) (22) (23) (24) (25) (26) OT (-1) 0.001 -0.014 -0.042 -0.007 0.011 0.000 -0.080* 0.202*** -0.053 -0.001 (0.001) (0.021) (0.069) (0.023) (0.010) (0.064) (0.045) (0.042) (0.069) (0.037) OT (-2) -0.001 -0.023 0.098 0.042 -0.019** 0.028 0.116** 0.055 0.051 0.024 (0.001) (0.025) (0.070) (0.026) (0.009) (0.064) (0.048) (0.046) (0.057) (0.031) OT (-3) -0.004** -0.011 0.055 -0.022 0.033** 0.038 0.007 0.044 0.128** -0.050 (0.002) (0.020) (0.081) (0.027) (0.013) (0.070) (0.053) (0.045) (0.059) (0.033) OT (-4) 0.001 0.013 -0.068 -0.001 -0.004 -0.008 0.029 0.035 0.006 -0.005 (0.001) (0.020) (0.070) (0.023) (0.011) (0.070) (0.047) (0.045) (0.062) (0.032) OT (-5) 0.002 0.049** -0.079 -0.042* -0.010 -0.105 -0.107** -0.026 -0.103 0.040 (0.002) (0.021) (0.060) (0.024) (0.010) (0.066) (0.051) (0.043) (0.069) (0.033) OT (-6) 0.002 0.032 0.032 -0.015 -0.014 -0.092 0.040 0.096* -0.124* -0.015 (0.002) (0.021) (0.075) (0.021) (0.012) (0.070) (0.053) (0.052) (0.070) (0.032) OT (-7) 0.003 -0.022 -0.087 0.034 -0.017 0.063 -0.084* 0.068 -0.017 0.004 (0.002) (0.021) (0.071) (0.026) (0.012) (0.067) (0.048) (0.050) (0.068) (0.029) OT (-8) -0.001 0.006 -0.003 0.023 -0.005 -0.010 -0.007 0.016 -0.001 -0.047 (0.001) (0.022) (0.074) (0.026) (0.010) (0.062) (0.052) (0.049) (0.057) (0.029)

38

Table 6. VAR for Detailed Players (Continued)

Panel B. VAR (8) for Detailed Players – Independent Variables: Pension Funds (PFt-i) RET CP FD IB ID IS MF OT PF SC (17) (18) (19) (20) (21) (22) (23) (24) (25) (26) PF (-1) 0.000 -0.012 0.035 0.003 0.001 -0.065 0.089** -0.001 0.157*** 0.007 (0.001) (0.016) (0.06) (0.021) (0.009) (0.056) (0.044) (0.039) (0.055) (0.027) PF (-2) 0.000 0.002 0.183*** -0.031 0.001 0.041 -0.004 -0.024 0.092 -0.015 (0.001) (0.019) (0.067) (0.021) (0.009) (0.058) (0.045) (0.037) (0.062) (0.024) PF (-3) -0.003* 0.000 0.041 -0.004 0.017 0.029 -0.035 -0.038 0.016 -0.023 (0.001) (0.017) (0.071) (0.022) (0.011) (0.053) (0.045) (0.049) (0.053) (0.026) PF (-4) 0.000 -0.026 -0.001 0.010 0.000 -0.010 0.062 0.0207 -0.005 0.035 (0.001) (0.018) (0.078) (0.022) (0.010) (0.059) (0.046) (0.046) (0.055) (0.028) PF (-5) -0.001 0.035* -0.054 -0.050** 0.014 0.037 -0.024 -0.026 0.013 -0.045* (0.001) (0.019) (0.068) (0.022) (0.009) (0.065) (0.043) (0.041) (0.053) (0.026) PF (-6) 0.003** 0.018 -0.019 0.008 -0.019* -0.102* -0.001 0.030 -0.104* 0.043 (0.001) (0.019) (0.065) (0.021) (0.011) (0.061) (0.046) (0.037) (0.058) (0.030) PF (-7) 0.000 -0.020 -0.084 0.018 -0.008 0.008 0.056 -0.007 0.014 0.041 (0.001) (0.020) (0.068) (0.022) (0.009) (0.063) (0.047) (0.039) (0.057) (0.027) PF (-8) -0.001 0.000 -0.024 0.013 0.011 -0.022 0.035 -0.044 -0.082 -0.014 (0.001) (0.018) (0.067) (0.018) (0.010) (0.057) (0.044) (0.042) (0.055) (0.023)

39

Table 6. VAR for Detailed Players (Continued)

Panel B. VAR (8) for Detailed Players – Independent Variables: Securities Firms (SCt-i) RET CP FD IB ID IS MF OT PF SC (17) (18) (19) (20) (21) (22) (23) (24) (25) (26) SC (-1) 0.003 -0.060 -0.080 0.041 -0.003 -0.137 -0.120 0.140* -0.076 0.164*** (0.002) (0.047) (0.12) (0.041) (0.019) (0.103) (0.088) (0.078) (0.097) (0.058) SC (-2) 0.004 -0.031 0.096 0.011 -0.034* -0.001 0.125 0.069 -0.049 0.141*** (0.002) (0.038) (0.117) (0.036) (0.019) (0.111) (0.080) (0.081) (0.099) (0.050) SC (-3) -0.006** -0.029 0.138 0.013 0.014 0.082 -0.111 0.069 0.118 0.042 (0.003) (0.035) (0.124) (0.048) (0.020) (0.104) (0.089) (0.070) (0.092) (0.062) SC (-4) -0.001 -0.006 0.148 -0.021 0.007 -0.155 0.130 0.005 0.083 0.045 (0.003) (0.034) (0.122) (0.036) (0.019) (0.117) (0.096) (0.078) (0.101) (0.054) SC (-5) 0.003 0.061* 0.052 0.040 -0.009 -0.098 -0.163* -0.146* -0.108 0.003 (0.003) (0.032) (0.115) (0.049) (0.019) (0.106) (0.096) (0.081) (0.096) (0.056) SC (-6) -0.003 -0.029 0.074 -0.012 -0.006 -0.064 0.0461 0.088 -0.090 0.001 (0.003) (0.036) (0.137) (0.035) (0.021) (0.114) (0.093) (0.086) (0.112) (0.050) SC (-7) 0.003 -0.068** 0.043 0.077 0.009 -0.126 0.0370 0.080 -0.111 0.017 (0.003) (0.033) (0.106) (0.047) (0.018) (0.104) (0.086) (0.073) (0.113) (0.047) SC (-8) -0.001 0.054 0.124 0.019 -0.023 -0.026 0.136* -0.139** 0.064 -0.087* (0.003) (0.040) (0.118) (0.039) (0.020) (0.114) (0.078) (0.064) (0.121) (0.052) CONS 0.000 0.001 0.019 0.003 -0.003 0.012 0.035*** -0.006 -0.009 0.008 (0.000) (0.005) (0.019) (0.006) (0.002) (0.017) (0.012) (0.012) (0.016) (0.007) df_r 637 637 637 637 637 637 637 637 637 637 df_m 80 80 80 80 80 80 80 80 80 80 F-stat 2.133 4.606 1.707 5.521 4.388 16.63 4.483 4.878 4.313 6.426 No of Obs. 718 718 718 718 718 718 718 718 718 718

40

Table 7. Granger Causality for Detailed Players

The table below gives the results of Granger causality test for detailed players based on VAR (1) as reported in Panel A and VAR (6) in Panel B. RET is value weighted portfolio return. ID, CP, IB, SC, OT, IS, MF, PF, and FD is trading imbalances of the individual investors, corporations, financial institutions, securities firms, other institutions, insurance firms, mutual funds, pension funds, and foundations, respectively. The null hypothesis of this test is that lagged values of x do not explain the variation in y, or in other words x does not granger cause y. The p-value of parameters is reported in the parantheses. ***, **, * indicates the significance level at 10%, 5%, and 1%, respectively.

Panel A. Granger Causality for Detailed Players based on VAR (1)

Effect (t) Variables RET ID CP IB SC OT IS MF PF FD 0.651 23.43*** 1.153 0.070 2.619 2.887* 0.267 0.036 1.606 1.283 RET (0.753) (0.000) (0.283) (0.790) (0.106) (0.089) (0.605) (0.849) (0.205) (0.257) 0.977 6.472*** 1.074 0.002 0.108 2.304 0.001 0.486 0.587 0.383 ID (0.323) (0.000) (0.300) (0.964) (0.742) (0.129) (0.967) (0.485) (0.443) (0.536) 0.046 0.117 3.066*** 2.375 0.513 0.665 0.705 0.037 0.053 0.727 CP (0.829) (0.731) (0.001) (0.123) (0.473) (0.414) (0.401) (0.846) (0.817) (0.394) 1.823 0.758 1.371 7.013*** 1.536 2.508 1.147 6.055** 1.206 0.211 IB (0.177) (0.384) (0.242) (0.000) (0.215) (0.113) (0.284) (0.014) (0.272) (0.646) Cause 0.647 0.106 3.497* 1.370 5.673*** 4.517** 2.712* 2.356 0.848 0.045 SC (t-i) (0.421) (0.743) (0.061) (0.242) (0.000) (0.033) (0.100) (0.125) (0.357) (0.831) 0.517 0.099 0.253 0.113 0.104 3.907 0.021 1.349 0.929 0.355 OT (0.472) (0.753) (0.614) (0.735) (0.746) (0.000) (0.883) (0.245) (0.335) (0.551) 0.010 1.424 0.264 16.44*** 15.85*** 0.997 1.730* 0.219 0.824 0.000 IS (0.918) (0.233) (0.607) (0.000) (0.000) (0.318) (0.078) (0.639) (0.364) (0.990) 2.144 1.043 9.195*** 4.762** 0.592 0.130 0.034 3.520 0.952 0.447 MF (0.143) (0.307) (0.002) (0.029) (0.441) (0.718) (0.853) (0.000) (0.329) (0.503) 0.164 0.234 0.338 0.505 0.006 0.974 1.994 5.082** 1.663* 1.437 PF (0.685) (0.628) (0.561) (0.477) (0.937) (0.323) (0.158) (0.024) (0.094) (0.231)

41

0.267 0.114 0.021 4.737** 0.085 4.214** 0.916 3.656* 1.265 1.794* FD (0.605) (0.735) (0.884) (0.029) (0.770) (0.040) (0.338) (0.056) (0.260) (0.0659) Table 7. Granger Causality for Detailed Players (Continued)

Panel B. Granger Causality for Detailed Players based on VAR (8) Effect (t) Variables RET ID CP IB SC OT IS MF PF FD

77.97 31.94*** 8.720 11.80 15.87** 6.393 8.580 6.868 5.236 17.22** RET (0.295) (0.000) (0.366) (0.160) (0.044) (0.603) (0.379) (0.551) (0.732) (0.028) 14.49* 129.5*** 3.345 7.382 4.967 6.838 7.335 26.06*** 9.469 12.40 ID (0.070) (0.000) (0.911) (0.496) (0.761) (0.554) (0.501) (0.001) (0.304) (0.134) 13.05 11.62 105.2*** 134.5*** 17.32** 6.574 5.052 8.836 15.29* 6.714 CP (0.110) (0.169) (0.007) (0.000) (0.027) (0.583) (0.752) (0.356) (0.054) (0.568) 9.612 4.862 15.57** 12.07 12.56 8.423 10.67 10.72 6.788 3.612 IB (0.293) (0.772) (0.049) (0.148) (0.128) (0.393) (0.221) (0.218) (0.560) (0.890) 10.64 6.203 14.94* 7.450 114.7*** 15.75** 8.480 15.81** 6.062 8.150 SC Cause (0.222) (0.625) (0.060) (0.489) (0.001) (0.046) (0.388) (0.045) (0.640) (0.419) (t-i) 16.14** 17.40** 12.43 9.606 6.882 91.18* 6.280 16.23** 13.61* 7.270 OT (0.040) (0.026) (0.133) (0.294) (0.549) (0.063) (0.616) (0.039) (0.092) (0.508) 3.227 5.708 5.282 9.094 10.09 7.092 79.63 12.73 4.639 4.078 IS (0.919) (0.680) (0.727) (0.334) (0.258) (0.527) (0.251) (0.121) (0.795) (0.850) 8.739 2.468 17.06** 15.05* 10.9 6.341 14.86* 120.4*** 9.215 3.487 MF (0.365) (0.963) (0.029) (0.058) (0.207) (0.609) (0.062) (0.000) (0.324) (0.900) 9.979 11.97 7.866 10.26 10.65 3.953 6.063 9.847 88.71* 16.73** PF (0.266) (0.152) (0.447) (0.247) (0.222) (0.861) (0.640) (0.276) (0.088) (0.033) 9.151 8.209 7.014 19.85** 10.46 13.07 9.226 13.64* 19.21** 86.35 FD (0.330) (0.413) (0.535) (0.011) (0.234) (0.109) (0.324) (0.092) (0.014) (0.119)

42

43