Investor Contributions to Price Discovery and Trading Performance: Evidence from the Taiwan Stock Exchange

Abstract

This study examines the relationship between price discovery and trading performance across different investor types in Taiwan’s active order-driven market. Using five-second intraday data, we construct a stock-trader-direction information share (IS) measure and link it to trading performance. Our results reveal several key findings: institutional investors have a higher IS per order, reflecting greater contributions to price discovery, and they outperform individual investors in trading performance. While higher IS is associated with better contemporaneous outcomes, it does not predict long-term performance. Determinants of price discovery include investor type, price aggressiveness, trade size, herding behavior, firm characteristics, and macroeconomic conditions. Robustness tests, covering one-minute IS, high-volatility periods, earnings announcements, and macroeconomic influences, support these conclusions.

1. Introduction

This paper examines how trading behavior, price discovery, and trading performance interact across different investor types in an order-driven market. Unlike quote-driven markets, where designated market makers provide bids and ask quotes, an order-driven market displays all bids and offers openly. This transparency means that settlement prices in an order-driven call market are determined directly by the interaction of supply and demand, giving order submissions a stronger influence on final prices than in quote-driven systems.

Informed traders drive price discovery by quickly incorporating new information into asset values through private insights or superior analysis of public data, with their trades revealing information to the market. Uninformed investors often follow these price leaders, making information disparities and dissemination across investor types crucial in stock markets. Understanding these roles clarifies how spot prices emerge from buyer–seller interactions. Robust price discovery improves return prediction (Johnson & So, 2018), identifies volatility spillovers (Fassas & Siriopoulos, 2019; Miao et al., 2017), and informs trading strategies (Benos & Sagade, 2016; J.-C. Hung et al., 2021). For example, Benos and Sagade (2016) find that aggressive high-frequency traders are highly informed, though their contribution declines as trading aggressiveness increases.

In Taiwan’s professional investment landscape, foreign investors are often compared with domestic investment trusts. Foreign investors bring broader international experience, while investment trusts typically possess local expertise. Amateur traders frequently follow investors with superior information advantages when placing orders. As a result, informed traders can influence the behavior of market followers, affecting overall market efficiency, volatility, and liquidity.

The literature offers conflicting views on whether domestic or foreign investors hold superior information advantages. Some studies suggest that domestic investors benefit from local knowledge, earning higher trading profits than foreign investors (e.g., S. Agarwal et al., 2009; Choe et al., 2005; Kalev et al., 2008). In contrast, other studies find that foreign investors, leveraging advanced technology, substantial financial resources, and extensive international experience, outperform their domestic counterparts (Barber et al., 2009; C.-N. Chen et al., 2014; Gao & Lin, 2015; Wang & Chiao, 2008). Although information heterogeneity is well-established, the specific contributions of different investor types remain underexplored.

While many studies have examined stock returns, this study fills a gap by investigating the interplay among trading behavior, price discovery, and trading performance in the Taiwan Stock Exchange (TWSE), a highly active order-driven market. The analysis first assesses price discovery across investor categories for individual stocks and trade directions, then examines the factors influencing information shares (IS), and finally links investor trading behavior and IS to trading performance.

Price discovery is shaped by a variety of factors related to trader characteristics and behaviors (Brogaard et al., 2019; W.-K. Chen et al., 2019; Piccotti & Schreiber, 2020). Different trader groups exhibit distinct motivations and behaviors, including information-driven trading (S. Agarwal et al., 2009; Baruch et al., 2017), cognitive limitations (Kuo et al., 2015), behavior-driven factors (Choi & Skiba, 2015; Li et al., 2017), preferences for realizing gains or losses (Shefrin & Statman, 1985), momentum versus contrarian strategies (Chhimwal & Bapat, 2021), and tax-driven trading (Badrinath & Lewellen, 1991). Collectively, these factors influence investor trading behavior and, consequently, their contribution to price discovery.

When using IS to measure price discovery, Kang et al. (2016) and W.-K. Chen et al. (2019) found that foreign investors are the primary contributors in Asian derivative markets, while individual investors contribute the least. In Canadian equity markets, Brogaard et al. (2019) showed that the limit orders of high-frequency traders drive price discovery. Piccotti and Schreiber (2020) examined foreign exchange markets across inter-dealer and dealer-customer tiers, revealing that investor types differ in their influence on price discovery. These studies largely focus on derivatives or FX markets rather than equities.

Lien and Hung (2023) found that institutional herding can hinder price discovery on the TWSE, whereas foreign investors play a larger role for cross-listed firms. These findings collectively demonstrate that different investor groups play distinct roles in the price discovery process. While existing studies provide valuable insights, they exhibit certain limitations when it comes to linking price discovery with trading performance. This study provides direct evidence regarding the nexus between price discovery and trading performance from the perspective of investor types. Even when a stock’s return is the same, trading performance can vary among trader categories. We address this gap by examining the connections between trading behavior, IS, and trading performance.

Following Hasbrouck (1995), Lien and Shrestha (2009), and Lien and Hung (2023), we construct a daily stock-trader-direction IS measure within a single stock market. Most research on price discovery focuses on how new information is incorporated into efficient security pricing across markets (e.g., Hasbrouck, 1995; Lien & Shrestha, 2009). Although Lien and Hung (2023) examine the TWSE, they do not consider trading performance. In contrast, this study links investor-specific trading behavior and price discovery to trading performance, thereby providing a more comprehensive perspective on how different investor types contribute to market outcomes.

The significance of price discovery in financial markets is multifaceted. First, it concerns regulators, academics, and market participants because it exhibits characteristics of a public good (i.e., non-rivalrous and non-excludable). Second, arbitrage relationships and equilibrium considerations have broad applications in market transactions. Understanding the link between security’s efficient price and its fundamental value is central to financial markets. Traders often attempt to infer an implicit efficient price to guide derivative pricing and compare it with the actual stock price. Consequently, the process of security price formation warrants careful attention and evaluation.

While IS has been studied in interconnected markets, our study examines price discovery across investor types and its link to trading performance within a single stock market. Dominant and satellite participants require separate price series, which is challenging in quote-driven markets, but the TWSE provides an ideal setting. Investors who quickly and accurately incorporate information drive price discoveries. We employ a daily stock-trader-direction IS measure and analyze a single stock traded by four investor categories, including foreign investors, investment trusts, other institutions, and individuals, each with distinct characteristics, behaviors, and information access.

In an order-driven call market, buyers and sellers submit orders that are matched centrally by a priority rule. Buyers bid at higher prices, sellers offer at lower prices, and these opposing forces determine the settlement price. Given this, separating buy and sell orders is essential when analyzing IS. Prior studies confirm distinct patterns and asymmetric effects between buy and sell orders (Baruch et al., 2017; Chou & Wang, 2009; Lien et al., 2020b; Wang & Chiao, 2008). For example, Baruch et al. (2017) showed buy–sell asymmetry in response to information, indicating that informed traders affect price formation differently.

Our study contributes to market microstructure and the behavioral finance literature in several ways. First, we extend the findings of Hasbrouck (1995) and Lien and Shrestha (2009) by applying the IS framework to different investor types and trade directions within a single market, offering a granular measure of price discovery and identifying key drivers in the TWSE, which relies on limit orders without market makers. Second, we provide direct evidence on information leaders, informing trading strategies and the relative informativeness of foreign investors and investment trusts. Third, we link price discovery to trade direction, revealing asymmetric patterns between buying and selling. Finally, we connect trading behavior, price discovery, and trading performance, distinguishing our work from studies focusing mainly on trading behavior and returns (W.-K. Chen et al., 2019; Lien & Hung, 2023).

Our empirical findings show that investor types play distinct roles in price discovery. Institutional investors have a higher IS per order than individuals, even though individuals dominate participation and volume. Among institutions, domestic investment trusts exceed foreign investors in IS. Key factors affecting price discovery include investor type, price aggressiveness, trade size, herding, firm characteristics, and macroeconomic conditions. Aggressive buy orders by professional institutions enhance price discovery, while larger trades have a smaller marginal effect on IS for institutions than for individuals. Herding among institutions can impair price discovery. Overall, institutional investors outperform individuals, with higher IS linked to better short-term trading performance, though not predictive of long-term outcomes.

The rest of this paper is organized as follows. Section 2 provides an overview of the data sources and sample description. Section 3 explains the measurement of information share. Section 4 presents the empirical results, and Section 5 concludes the study.

2. Data Sources and Sample Description

This study relies on two primary data sources. The first source is the Taiwan Stock Exchange Corporation (TSEC), which compiles historical intraday trading data. We obtained intraday order-, trade-, and quote-level data from TSEC. The second data source is the Taiwan Economic Journal (TEJ) database, from which we extracted daily market trading data, annual financial data, and the Taiwan Capitalization Weighted Stock Index (TAIEX). For our analysis, we consider all common stocks listed on the TWSE, encompassing both newly added and delisted firms. To be included in the effective sample, the firm must have four intraday price series pertaining to four types of investors (foreign investors, investment trusts, other institutions, and individuals), along with daily trading information and yearly financial data. To preserve data integrity, we excluded after-hours trading, which mainly involves block and odd-lot trades subject to different rules than regular sessions.

Our sample period spans from 1 June 2015 to 20 March 2020, covering over four years and nine months. On 1 June 2015, TWSE raised daily price limits from ±7% to ±10% to better reflect market demand and align with international standards. The exchange transitioned from a call auction to a continuous trading system on 23 March 2020. To avoid potential effects from this policy change, we limited our sample to the period between the introduction of the ±10% price limit and the day before continuous trading began. During this period, a call auction system with fixed price limits was in operation.

To ensure consistent comparisons across the four investor types, we restricted our sample to stocks with complete intraday price series. This yields 477 firms over 1128 trading days, resulting in 56,079 firm-day observations. At the end of each sample year, these firms collectively accounted for over 77% of the total capitalization of common stocks on the TWSE.

3. Information Share Measurement

Regarding the construction of the four investor-type price series at a five-second frequency for the IS measure, this study relies exclusively on realized executed prices rather than standing order prices. Under this framework, price points are recorded only upon the finalization of a transaction. Specifically, we aggregated all execution records within each five-second interval; if a single order was partially filled across different timestamps, each resulting execution was treated as a distinct price event at the moment it occurred. Unexecuted orders or unfilled portions of limit orders were excluded as they do not represent completed market transactions.

To form the discrete five-second series, we utilized the last execution price recorded within each interval for each investor type. In cases where no trade occurred for a specific investor type within a given interval, the price was carried forward from the previous period to maintain a continuous series. By focusing on realized executions, we ensured that the price series captured actualized information exchange rather than latent intent. This approach filters out non-informative or tactical limit orders that do not result in trades, thereby providing a cleaner measure of how each investor type contributes to the common trend.

We used subscripts to denote specific parameters: subscript $t$ for the intraday trading interval; $q$ for the number of lags; and $m$ and $n$ for distinct price series ($m\ne n$) of different investor types. Following Hasbrouck (1995), we calculated IS for each investor type, measuring the proportion of innovation variance in the efficient price attributable to that group using a cointegration-based econometric approach. Cointegration occurred when a linear combination of non-stationary but related price series was stationary, indicating a shared stochastic trend. If a cointegration relationship exists, a vector error correction model (VECM) can represent the short-term dynamics within a cointegrated vector autoregression (VAR).

Most prior studies on price discovery focus on IS, which allocates information contributions across markets (e.g., Blanco et al., 2005; Gonzalo & Granger, 1995; Hasbrouck, 1995; Lien & Shrestha, 2009, 2014; Norden & Wever, 2009; Piccotti & Schreiber, 2020; Putniņš, 2013; Zhu, 2006). Building on Hasbrouck (1995) and Lien and Shrestha (2009), we constructed an investor-specific price discovery measure by aggregating the variance contributions associated with each investor type.

The Hasbrouck-style IS, a widely used price discovery metric, relies on a VAR framework and assumes non-stationary but cointegrated prices. This study follows these principles while refining the calculation on a daily basis for each stock and investor type within a single market. Our IS measure quantifies each investor group’s contribution to the variance of efficient price innovations. The group contributing the largest share is considered the dominant driver of price discovery, reflecting the presence of informed participants.

During our sample period, TWSE conducted call auctions roughly every five seconds in regular trading sessions. To construct distinct price series for the four investor types, we maintained four series per stock, for buy or sell orders, forming a (4 × 1) vector $P_t$. Each series was treated as a unit-root process with random-walk characteristics, and price changes were assumed to be covariance-stationary, allowing representation via a vector error-correction model following Engle and Granger (1987):

$${∆P}_t={\Pi P}_{t-1}+{\sum }_{q=1}^Q{A_q∆P}_{t-q}+{\epsilon }_t,$$

(1)

$$\Pi =\alpha {\beta }^T.$$

(2)

The matrix $\Pi$ captures long-run effects and can be expressed as $\alpha {\beta }^T$, where each column of $\alpha$ contains VEC coefficients and $\beta$ is the cointegration vector. The term ${\beta }^T{P}_{t-1}$ represents the stationary cointegrated series. Although individual prices are non-stationary, differences between any two prices are stationary, indicating cointegration of order 3; thus, both $\alpha$ and $\beta$ are $4\times 3$ matrices. $A_q$ is a $4\times 4$ matrix for short-term effects; ${\epsilon }_t$ is a zero-mean serially uncorrelated error vector; and its covariance matrix is $E\left({\epsilon }_t{\epsilon }_t^T\right)=\Omega$ (i.e., innovations).

Equation (1) can be rewritten in vector moving average (VMA) form (Hasbrouck, 1995; Stock & Watson, 1988). The IS calculation is based on the following VMA representation:

$${∆P}_t=\Psi \left(L\right){\epsilon }_t,$$

(3)

where $\Psi \left(L\right)$ is a matrix polynomial in the lag operator. Alternatively, the level of $P_t$ can be expressed as follows:

$$P_t=P_0+\psi \left(1\right){\sum }_{q=1}^t{\epsilon }_q+{\psi }^{*}\left(L\right){\epsilon }_t,$$

(4)

where the first term, $P_0$, is a constant vector of initial values. The second term, $\psi \left(1\right){\sum }_{q=1}^t{\epsilon }_q$, captures the long-term cumulative impact of shocks on stock prices and forms the basis for several IS metrics. The third term, ${\psi }^{\mathrm{*}}\left(L\right){\epsilon }_t~I\left(0\right)$, is a zero-mean, covariance-stationary process, with ${\psi }^{\mathrm{*}}\left(L\right)$ as a matrix polynomial in the lag operator. Cointegration among the unit-root price series (${\beta }^T{P}_t$ stationary) implies ${\beta }^T\psi \left(1\right)=0$ and $\psi \left(1\right)\alpha =0$ (Engle & Granger, 1987; Lehmann, 2002). We denote $\psi =({\psi }_1,{\psi }_2,{\psi }_{3, }{\psi }_4)$ as the common row vector.

According to Hasbrouck (1995), the cointegrating relationships are symmetric, so all price series are equal in equilibrium, with each pairwise cointegrating vector being [1, −1]. For the four unit-root price series, the transposed $\beta$ can be expressed as follows:

$$\begin{array}{c}{\beta }^T\\ 3\times 4\end{array}={[\iota }_3: {-I}_3]=\left[\begin{array}{c}1\\ 1\\ 1\end{array}\begin{array}{c}-1\\ 0\\ 0\end{array}\begin{array}{c}0\\ -1\\ 0\end{array}\begin{array}{c}0\\ 0\\ -1\end{array}\right],$$

(5)

where ${\iota }_3$ is a 3-element column vector with all elements equal to one. Moreover, $I_3$ is a $3\times 3$ identity matrix.

We calculated Hasbrouck-style IS, ${IS}_n$, to measure the contribution of investor type $n$ to price discovery, based on the proportion of that investor type’s volatility relative to the total volatility of the long-run impact:

$${IS}_n={\displaystyle \frac{{\left({\left[\psi F\right]}_n\right)}^2}{\psi \mathsf{\Omega }{\psi }^T}},n\in \{1,2,3,4\},$$

(6)

where ${\left[\psi F\right]}_n$ is the $n$th element of the row vector $\psi F$, and $F$ is the lower-triangular matrix from the Cholesky decomposition of $\mathsf{\Omega }$ ($\mathsf{\Omega }=F{F}^T$), which imposes a hierarchy with the first price series having the maximum IS and the last the minimum. To address the ordering non-uniqueness, we consider all 24 sequences of the four-price series and compute the representative Hasbrouck-style (1995) IS using the average, median, and mean of the maximum and minimum values.1

Lien and Shrestha (2009) propose a modified information share (MIS) using a factor structure based on the correlation matrix rather than the covariance matrix. Let $\Lambda$ be a diagonal matrix of the correlation matrix’s eigenvalues, with corresponding eigenvectors in $G$, and let $V$ be a diagonal matrix of innovation standard deviations. The modified IS is then defined as follows:

$$M{IS}_n={\displaystyle \frac{{\left({\psi }_n^M\right)}^2}{\psi \mathsf{\Omega }{\psi }^T}},$$

(7)

where ${\psi }^M=\psi \widehat{F}$, with ${\psi }_n^{\mathrm{M}}$ as its $n$th element, and $\widehat{F}=[G{\mathsf{\Lambda }}^{-1/2}{G}^T{V}^{-1}{]}^{-1}$ such that $\mathsf{\Omega }=\widehat{F}{\widehat{F}}^T$. The MIS is unique and unaffected by Cholesky ordering. While Lien and Shrestha (2014) propose a generalized IS for interrelated markets, our analysis of four price series within a single market allows direct application of the MIS.

4. Analytical Framework and Empirical Findings

4.1. Descriptive Statistics

Table 1. Descriptive statistics of market trading and firm characteristics.

Variable	Number of Firm-Day Obs.	Mean	Std. Dev.	Q1	Med.	Q3
Market return (%)	-	0.00	0.92	−0.40	0.06	0.50
Stock return (%)	56,079	0.11	3.01	−1.43	0.00	1.59
Time-weighted spread (%)	56,079	0.27	0.12	0.18	0.24	0.34
Intraday volatility (bp)	56,079	0.13	0.06	0.09	0.12	0.16
Price–earnings ratio (times)	56,079	30.22	124.83	12.70	17.24	24.47
Market capitalization (NT$ billion)	56,079	198.71	747.54	16.32	31.90	94.32
Book-to-market ratio	56,079	0.20	0.19	0.08	0.14	0.26
Turnover rate (%)	56,079	1.46	1.55	0.36	0.86	1.91
Institutional holdings (%)	56,079	33.20	17.66	19.71	30.61	43.53

presents descriptive statistics for daily market returns and sample firm characteristics. The daily market return averages 0.00% with a standard deviation of 0.92%, indicating relatively low market volatility. Stock returns are slightly more variable, with a mean of 0.11% and a standard deviation of 3.01%. Sample firms have an average bid–ask spread of 0.27% and intraday volatility of 0.13 basis points. The mean price-to-earnings ratio is 30.22, and the average market capitalization is NT$198.71 billion, spanning a wide range as the third quantile is nearly six times the first. Other averages include a book-to-market ratio of 0.20, a turnover rate of 1.46% per day, and top-three institutional ownership of 33.20%. Compared with all TWSE-listed firms, our sample shows lower spreads and intraday volatility, larger market capitalization, higher turnover, growth orientation, and greater institutional ownership.

To ensure the appropriateness of applying the methods of Hasbrouck (1995) and Lien and Shrestha (2009), we began by testing for a cointegration relationship among the logarithmic stock prices across different investor types. Price series are cointegrated when different investor types move in a coordinated manner throughout the trading session, indicating the existence of a daily equilibrium toward which the stock price gradually converges.

Table 2. Descriptive statistics of information shares.

Investor Type	Buy Side						Sell Side
	Number of Obs.	Mean	Std. Dev.	Q1	Med.	Q3	Number of Obs.	Mean	Std. Dev.	Q1	Med.	Q3
Panel A: Hasbrouck-style Information Shares by Average Calculation
Foreign investors	32,041	0.26	0.09	0.20	0.25	0.30	31,984	0.27	0.10	0.20	0.25	0.31
Investment trusts	32,041	0.13	0.08	0.07	0.12	0.18	31,984	0.13	0.08	0.07	0.12	0.18
Other institutions	32,041	0.23	0.09	0.17	0.23	0.28	31,984	0.23	0.10	0.16	0.22	0.28
Individuals	32,041	0.38	0.11	0.32	0.37	0.44	31,984	0.38	0.11	0.31	0.37	0.44
Panel B: Hasbrouck-style Information Shares by Median Calculation
Foreign investors	32,041	0.25	0.09	0.19	0.24	0.29	31,984	0.26	0.10	0.20	0.25	0.31
Investment trusts	32,041	0.13	0.07	0.07	0.11	0.17	31,984	0.13	0.08	0.06	0.11	0.17
Other institutions	32,041	0.22	0.09	0.17	0.22	0.27	31,984	0.22	0.10	0.16	0.21	0.27
Individuals	32,041	0.38	0.11	0.30	0.37	0.44	31,984	0.37	0.11	0.30	0.36	0.44
Panel C: Hasbrouck-style Information Shares by Extreme Means
Foreign investors	32,041	0.26	0.10	0.20	0.25	0.31	31,984	0.27	0.10	0.21	0.26	0.32
Investment trusts	32,041	0.13	0.08	0.07	0.12	0.18	31,984	0.13	0.08	0.07	0.12	0.18
Other institutions	32,041	0.24	0.09	0.17	0.23	0.29	31,984	0.23	0.10	0.16	0.22	0.28
Individuals	32,041	0.39	0.10	0.32	0.38	0.45	31,984	0.38	0.11	0.31	0.38	0.45
Panel D: Lien and Shrestha’s Modified Information Shares
Foreign investors	32,041	0.26	0.10	0.20	0.25	0.30	31,984	0.27	0.10	0.20	0.25	0.31
Investment trusts	32,041	0.13	0.08	0.07	0.12	0.18	31,984	0.13	0.08	0.07	0.12	0.18
Other institutions	32,041	0.23	0.09	0.17	0.23	0.28	31,984	0.23	0.10	0.16	0.22	0.28
Individuals	32,041	0.38	0.11	0.32	0.38	0.45	31,984	0.38	0.11	0.31	0.37	0.45

reports descriptive statistics of information shares. Hasbrouck-style (1995) IS measures were computed for all 24 possible orderings of the four investor types, with Panels A–C showing the average, median, and mean of the maximum and minimum, and Panel D presenting Lien and Shrestha’s (2009) modified MIS. Results indicate distinct IS patterns across investor types, with individuals generally having a higher IS than institutions in the TWSE. Panel D’s modified IS aligns closely with Hasbrouck-style average ISs. Compared to the Taiwan Futures Exchange, where foreign investors dominate price discovery (W.-K. Chen et al., 2019), our results show that domestic investors play a more prominent role in the stock market, highlighting market-dependent differences in investor contributions.

4.2. Scaled Information Shares and Trading Summary

Since trading information is conveyed through order submissions, it is crucial to relate price discovery to the number of orders rather than overall trading activity. We therefore adopted the scaled IS measure, following Lien and Hung (2023), as a proxy for the extent of price discovery. The scaled measure is derived by dividing the daily IS by the number of orders (measured in thousands). This serves as a more appropriate benchmark for comparative analyses across different types of investors because it accounts for differences in trade size. Although individuals, as a group, account for the largest trading volume and consequently exhibit the highest aggregated IS (as shown in Table 2), this group also includes a significant number of non-professional participants. In this context, the aggregated IS does not adequately reflect the information content per order flow.

To accentuate the transaction information associated with a specific number of orders,

Table 3. Summary statistics for Scaled information shares and trading summary.

Investor Type	Buy Side					Sell Side
	Foreign Investors	Investment Trusts	Other Institutions	Individuals	All Investors	Foreign Investors	Investment Trusts	Other Institutions	Individuals	All Investors
Panel A: Information Shares per Thousand Orders
Mean IS	0.61	2.46	0.90	0.21	0.40	0.62	2.21	0.98	0.28	0.48
Median IS	0.59	2.40	0.87	0.21	0.39	0.60	2.16	0.96	0.28	0.47
Extreme IS	0.62	2.51	0.91	0.22	0.40	0.63	2.24	1.00	0.28	0.49
Modified IS	0.61	2.46	0.90	0.21	0.40	0.62	2.20	0.98	0.28	0.48
Panel B: Trading Summary Statistics
Number of orders (million times)	13.50	1.68	8.27	57.50	80.94	13.77	1.87	7.34	43.04	66.03
% of orders	16.67	2.07	10.22	71.03	100.00	20.86	2.83	11.12	65.19	100.00
Total order value (NT$ trillion)	5.60	0.48	3.10	14.16	23.33	5.79	0.50	2.81	12.34	21.44
% of order value	23.99	2.05	13.28	60.68	100.00	27.00	2.34	13.12	57.55	100.00
Mean order value (NT$ thousand)	414.72	284.93	374.48	246.19	288.21	420.24	267.98	383.37	286.67	324.76
Mean price aggressiveness (%)	0.58	3.25	1.18	1.04	1.23	0.57	3.89	1.40	1.34	1.48
Total trade value (NT$ trillion)	5.53	0.46	3.05	13.91	22.95	5.83	0.52	2.84	12.50	21.69
% of trade value	24.09	2.02	13.28	60.61	100.00	26.88	2.39	13.10	57.64	100.00
Mean trade value (NT$ thousand)	409.74	276.36	368.29	241.98	283.57	423.28	277.03	387.11	290.49	328.55

provides additional insights by presenting the scaled IS measures alongside trading summaries. The distinction between aggregated and scaled IS is the consideration of varying numbers of order flows. By standardizing measures to the same number of orders (identical order flow), we converted the aggregated measure into the scaled IS, which provides a consistent criterion for subsequent comparative assessments.

Panels A and B of Table 3 correspond to the IS per thousand orders and the summary statistics related to orders and trades, respectively. While individuals contribute the largest share to overall price discovery (approximately 38%), their IS per thousand orders is the lowest, irrespective of the specific IS measure or trade direction considered. This highlights that the large trading volume of these individuals does not translate to high informational efficiency per order. Among institutional investors, investment trusts exhibit the highest IS per thousand orders, followed by other institutions. On average, domestic institutions make the most substantial contribution to price discovery per thousand orders, regardless of the IS measure or trade direction. These results strongly reaffirm the notion that different investor types make highly distinctive contributions to price discovery when the volume of order flow is standardized. Once again, the modified IS measures closely align with the Hasbrouck-style average IS measures. We also conducted $t$-tests to evaluate the averages of the daily scaled IS measures and reject the hypothesis that the scaled IS equals zero. For brevity, we omit the details.

Our findings indicate that institutional investors dominate information contribution in the TWSE, while individuals act as satellite participants. Investment trusts, in particular, show strong information awareness, and when their private information is timely and reliable, it enhances market efficiency, consistent with Lien and Hung (2023). Satellite participants, including astute individuals, may use the actions of dominant investors to guide their trading strategies. Although individuals contribute most to aggregated IS (Table 2), their impact on price discovery per thousand orders is relatively limited (Table 3).

Compared with W.-K. Chen et al. (2019), our results show that investor effectiveness varies across markets even within the same investor type. In the TWSE, domestic investment trusts are the main contributors to price discovery, unlike in the futures market, where foreign investors dominate. This difference reflects Taiwan’s stock market maturity and domestic institutions’ local expertise, while foreign investors excel in the futures market due to advanced derivatives skills and multinational experience. Overall, price discovery patterns differ across markets, highlighting the role of market context in informational efficiency.

Panel B of Table 3 provides an overview of the trading activity across various investor types. Individuals dominate the market, accounting for 60.68% (buy side) and 57.55% (sell side) of the order value. They also represent nearly 57–60% of the trade value. Foreign investors follow, making up 23.99% (buy side) and 27.00% (sell side) of the order value, and approximately 24–27% of the trade value. Investment trusts contribute a significantly smaller portion, with only 2.05% (buy side) and 2.34% (sell side) of the order value. Other institutions account for around 13% of the trade value. These statistics clearly highlight substantial disparities in both the order value and trade value among investor categories.

To analyze order submission behavior, we consider two key factors: price aggressiveness and trade size. Price aggressiveness is measured based on the daily average price of trades relative to the volume-weighted average price (VWAP). On the buy side, price aggressiveness is the natural logarithm of the ratio of the average price of buy trades to VWAP, expressed as a percentage. Conversely, on the sell side, price aggressiveness is the natural logarithm of the ratio of VWAP to the average price of sell trades, also expressed as a percentage. In both cases, a higher value indicates a greater sense of order immediacy (more aggressive submission).

Our empirical findings reveal that investment trusts exhibit the most aggressive order decisions among all investor types, closely followed by other institutions. Interestingly, individuals’ price aggressiveness surpasses that of foreign investors. This indicates that domestic institutions and individuals are generally more assertive than foreign investors regarding trade execution. Investment trusts adopt a particularly aggressive approach, likely driven by operational and performance demands. These demands include disclosure obligations, performance objectives, career concerns, and the necessity to streamline trading while mitigating execution risks (V. Agarwal et al., 2014; Giambona & Golec, 2010; Lien et al., 2019; Ling & Arias, 2013; Morey & O’Neal, 2006).

In terms of trade size, foreign investors display the largest average trade value per order, closely followed by other institutions. Although individuals are dominant participants in the TWSE, contributing significantly to the total trade value, their average trade value remains relatively modest. Foreign investors may fragment their substantial orders into smaller ones: a strategy aimed at concealing their private information and minimizing potential price impacts (Chakravarty, 2001; Chan & Lakonishok, 1995). Despite this fragmentation, their average trade value still surpasses that of domestic investors. Evidently, different categories of investors employ distinct trading styles, and their contributions to price discovery may vary to some extent. Institutional investors exhibit a higher scaled IS compared to individuals. Within the realm of professional institutions, domestic investment trusts outperform foreign investors in terms of scaled IS. The evidence is consistent with the premise that domestic institutions generally possess superior information advantages (e.g., Choe et al., 2005; Dvořák, 2005; Kalev et al., 2008).

Both foreign investors and domestic investment trusts are professional institutions. Although investment trusts are prominent contributors to scaled IS, they represent a mere 2% of the total trade value. Conversely, foreign investors account for approximately 25% of the total trade value, and their average scaled IS surpasses that of individuals by more than two-fold. The trading activities of both groups draw public attention, but their influence and style differ significantly. For example, foreign investors primarily prioritize corporate fundamentals and emphasize medium- and long-term investment objectives. They prefer large-cap and high-priced stocks, often using more precise pricing for trade execution. Given their substantial financial resources, their significant buying or selling activities can lead to considerable market fluctuations. In contrast, domestic investment trusts encompass multiple mutual fund members. Fund managers, unlike foreign investors, tend to focus on small- and medium-sized stocks and lower-priced equities. These managers are obliged to disclose their portfolio shareholdings information quarterly and place an emphasis on short- and medium-term performance. Consequently, they highly value corporate fundamentals and immediate news events. To pursue short-term gains, they tend to employ aggressive pricing strategies (buying at higher prices and selling at lower prices) to complete their intended trading positions swiftly. Retail investors may not always be able to discern who possesses informed information, leading to differential responses. If foreign investors dominate trading volume, retail traders may unwittingly follow them. When foreign investors sell off holdings due to a perceived market downturn, retail traders may hasten the selling process. By contrast, when the activities of investment trusts capture public attention, retail traders may unknowingly engage in herding behavior with these trusts, chasing short-term profits. This behavior can make the market more speculative and volatile. Consequently, government authorities maintain close vigilance over market trends driven by either foreign investors or investment trusts.

Frijns et al. (2018) demonstrated that enhanced market liquidity leads to a higher share of price discovery within a given market. Since trading volume is highly correlated with market liquidity, we also estimate the daily information share per million shares traded. This alternative measure captures the information content per unit of quantity and is less sensitive to variations in trading technology or order-splitting strategies. For brevity, descriptive statistics for these scaled information shares are presented in

Table A1. Descriptive statistics of alternative information shares.

Trader Type	Buy Side						Sell Side
	Number of Obs.	Mean	Std. Dev.	Q1	Med.	Q3	Number of Obs.	Mean	Std. Dev.	Q1	Med.	Q3
Panel A: Information Share per Thousand Orders
Mean IS	140,908	1.94	3.76	0.42	0.89	2.03	141,884	2.09	4.33	0.48	0.98	2.22
Median IS	140,908	1.91	3.71	0.40	0.86	1.99	141,884	2.05	4.24	0.46	0.96	2.17
Average of Max and Min IS	140,908	1.97	3.79	0.43	0.90	2.06	141,884	2.12	4.40	0.49	1.00	2.24
Modified IS	140,908	1.94	3.76	0.42	0.88	2.02	141,884	2.09	4.32	0.48	0.98	2.21
Panel B: Information Share per Million Traded Shares
Mean IS	140,908	0.78	1.85	0.11	0.28	0.75	141,884	0.87	2.54	0.12	0.31	0.84
Median IS	140,908	0.77	1.83	0.10	0.27	0.74	141,884	0.86	2.50	0.12	0.31	0.83
Average of Max and Min IS	140,908	0.79	1.87	0.11	0.28	0.76	141,884	0.88	2.56	0.12	0.32	0.85
Modified IS	140,908	0.78	1.85	0.11	0.28	0.75	141,884	0.87	2.53	0.12	0.31	0.84

of Appendix A. The information share per million shares exhibits patterns consistent with those observed for the information share per thousand orders, albeit with a smaller magnitude. These results reinforce our finding that sell-side orders and trades tend to contribute more to price discovery than their buy-side counterparts. In sum, the order-based metric provides price discovery patterns that complement those derived from trade-based metrics.

4.3. Information Shares Across Subsamples

In high-frequency market microstructure studies, price sparsity and stale risk are inherently present. Even in active stock markets, certain investor categories may exhibit several consecutive five-second intervals without transactions, particularly when the analysis is further disaggregated into buy and sell directions. To address this concern and mitigate the potential biases arising from price imputation, we conducted a robustness check by recalculating our results at a one-minute frequency. The empirical findings remain qualitatively similar across these different time resolutions, suggesting that the observed price discovery leadership is not a mechanical artifact of the data frequency or the carry-forward treatment.

This section further conducts an array of comprehensive tests to address various issues related to stock volatility, earnings reports, and macroeconomic factors. Specifically, we segment all observations into subsets based on stock volatility, PE ratios, earnings growth forecasts, foreign exchange rates, and interest rates.

Table 4. Modified information shares per thousand orders across subsamples.

Trader Type	Buy Side										Sell Side
	Foreign Investors		Investment Trusts		Other Institutions		Individuals		All Investors		Foreign Investors		Investment Trusts		Other Institutions		Individuals		All Investors
Stock Volatility
High	0.71		2.62		0.99		0.27		0.47		0.73		2.31		1.08		0.35		0.58
Low	0.55		2.38		0.84		0.19		0.36		0.56		2.20		0.93		0.25		0.44
Diff.	0.17	***	0.24	***	0.15	***	0.08	***	0.11	***	0.18	***	0.11	***	0.14	***	0.10	***	0.13	***
PE Ratio
High	0.68		2.65		1.01		0.21		0.40		0.70		2.24		1.10		0.26		0.48
Low	0.56		2.33		0.81		0.24		0.42		0.57		2.27		0.91		0.33		0.52
Diff.	0.12	***	0.32	***	0.20	***	−0.04		−0.02	***	0.13	***	−0.03	***	0.19		−0.07	***	−0.04	***
Earning Growth Forecast
Positive	0.60		2.43		0.94		0.22		0.41		0.63		2.12		0.99		0.28		0.49
Negative	0.43		2.46		0.66		0.20		0.34		0.47		1.95		0.70		0.30		0.44
Diff.	0.17	***	−0.03	***	0.28	**	0.02		0.07	***	0.17	***	0.17	***	0.29	***	−0.02		0.05	***
Foreign Exchange Rate
Appreciation	0.61		2.38		0.88		0.23		0.41		0.67		2.31		0.95		0.31		0.52
Depreciation	0.63		2.60		0.93		0.22		0.41		0.59		2.21		1.05		0.28		0.48
Diff.	−0.02	*	−0.22	***	−0.05	***	0.01	**	0.01	***	0.08		0.10		−0.10	***	0.03		0.04	***
Interest Rate
High	0.62		2.85		0.97		0.25		0.45		0.62		2.54		1.05		0.34		0.55
Low	0.63		2.25		0.86		0.20		0.38		0.63		2.06		0.95		0.26		0.46
Diff.	−0.01	***	0.61	***	0.11	***	0.05	***	0.08	***	−0.01	***	0.48	***	0.10	***	0.07	***	0.09	***

Note: ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.

reports modified information shares per thousand orders across high- and low-volatility subsamples, classified by intraday return standard deviations relative to the daily median. Results align with overall findings: investment trusts are the main contributors to price discovery, followed by other institutions and foreign investors, regardless of trade direction. All investor types show a higher IS on volatile days, with investment trusts exhibiting the largest increase, reflecting their ability to exploit market opportunities. Foreign investors’ skill in derivatives may also contribute to this effect.

Earnings reports provide key information for investors, with the PE ratio (stock price divided by EPS) used to assess expected growth. Stocks are classified into high and low PE subsamples relative to the daily median. Our findings hold across these subsamples for both buy and sell decisions. Foreign investors show higher IS in high PE stocks, while individuals exhibit lower IS. Investment trusts display buy–sell asymmetry, with higher ISs when buying and lower ISs when selling high PE stocks, reflecting information-based trading influenced by quarter-end disclosures and fund manager career concerns (V. Agarwal et al., 2014; Gallagher et al., 2009; P.-H. Hung et al., 2020).

Research analysts’ consensus earnings forecasts help investors assess stock performance. Stocks are categorized by positive or negative earnings growth forecasts based on analysts’ average sales and earnings growth predictions over the one-month period following announcements. Most institutional investors tend to show a higher IS towards stocks with positive earnings growth forecasts, while individual investors do not display similar tendencies. This pattern suggests that institutional investors possess superior stock-picking and market-timing abilities compared to individuals, enabling them to trade in promising stocks and contribute significantly to price discovery. In summary, institutional IS correlates more strongly with earnings reports than that of individuals, reflecting the superiority of professional investment decisions.

Macroeconomic factors, such as FX and interest rates, influence stock markets by affecting portfolio returns and capital costs. We classify days by NT$ appreciation or depreciation against the US$ and segment interest rates into high and low periods using bid–ask midpoints from Refinitiv Eikon Datastream. Controlling for these factors, our findings remain consistent: investment trusts dominate price discovery, followed by other institutions and foreign investors. Macroeconomic effects are stronger for domestic investors, but investment choices have a greater impact on trading behavior, confirming that specific investment targets primarily drive outcomes.

Although IS and trading summaries of buy and sell orders show similar trends, distinguishing between them remains economically meaningful. Bid and ask prices reflect demand and supply, moving in opposite directions and affecting transaction prices through different mechanisms. This asymmetry arises from transaction costs, short-sale constraints, and trade initiation effects. In the TWSE, a 0.3% securities transaction tax applies to sales but not purchases, influencing order decisions. Short-sale restrictions also make selling harder, limiting the incorporation of private information into prices (Autore et al., 2015; Ramachandran & Tayal, 2021; Anufriev & Tuinstra, 2013; Diamond & Verrecchia, 1987). Buyer- and seller-initiated crowded trades impact prices and volatility asymmetrically (Zhou & Yang, 2019; Bissoondoyal-Bheenick et al., 2019), while seasonal patterns, such as tax-loss selling in December and buyer-initiated trades in January, further contribute to asymmetry (Chordia et al., 2016). To capture these effects on trading outcomes, we analyze buy and sell orders separately, expecting some degree of buy–sell asymmetry even when trading patterns are similar.

4.4. What Are Determinants of Information Shares?

This section investigates the determinants of information shares. The regression model is estimated using heteroscedasticity- and autocorrelation-consistent standard errors following Newey and West (1987), as specified below:

$$\begin{array}{cc}M{IS}_{ijt}^d=& {\beta }_0+{\beta }_1{For}_{it}+{\beta }_2{Trust}_{it}+{\beta }_3{OthInst}_{it}+{\beta }_4{TrdAgg}_{ijt}^d+{\beta }_5{TrdSize}_{ijt}^d+{\beta }_6{Herd}_{ijt}^d+{For}_{it}\times \hfill \\ & ({\beta }_7{TrdAgg}_{ijt}^d+{\beta }_8{TrdSize}_{ijt}^d+{\beta }_9{Herd}_{ijt}^d)+{Trust}_{it}\times ({\beta }_{10}{TrdAgg}_{ijt}^d+{\beta }_{11}{TrdSize}_{ijt}^d+{\beta }_{12}{Herd}_{ijt}^d)+\hfill \\ & {OthInst}_{it}\times ({\beta }_{13}{TrdAgg}_{ijt}^d+{\beta }_{14}{TrdSize}_{ijt}^d+{\beta }_{15}{Herd}_{ijt}^d)+{\beta }_{16}{Spr}_{i,t-1}+{\beta }_{17}{Vol}_{i,t-1}+{\beta }_{18}{PE}_{i,t-1}+\hfill \\ & {\beta }_{19}{Size}_{i,t-1}+{\beta }_{20}{BM}_{i,t-1}+{\beta }_{21}{TO}_{i,t-1}+{\beta }_{22}{InstHold}_{i,t-1}+{\beta }_{23}{CroList}_{it}+{\beta }_{24}{PosRet}_{it}+{\beta }_{25}{NegRet}_{it}+\hfill \\ & {\beta }_{26}{FX}_{i,t-1}+{\beta }_{27}{Interest}_{i,t-1}+{\beta }_{28}{MIS}_{ij,t-1}^d+{\sum }_{g=1}^5{\gamma }_g{Tick}_{gi,t-1}+{\sum }_{r=1}^{19}{\delta }_r{Indus}_{rit}+{\sum }_{s=1}^3{\theta }_s{Year}_{sit}+\hfill \\ & {\sum }_{u=1}^{11}{\pi }_u{Month}_{uit}+{\sum }_{v=1}^4{\phi }_v{Weekday}_{vit}+{\epsilon }_{ijt}^d.\hfill \end{array}$$

(8)

The dependent variable, $M{IS}_{ijt}^d$, represents the modified IS per thousand orders, where the superscript $d$ indicates trade direction ($d=b$ for buy-side and $d=s$ for sell-side trading). The independent variables include trader dummies, trading behavior, firm characteristics, macroeconomic factors, and others. In particular, ${For}_{it}$, ${Trust}_{it}$, and ${OthInst}_{it}$ are dummy variables for foreign investors, investment trusts, and other institutions, respectively, with individual investors serving as the reference group.

To assess investors’ trading behavior, we consider price aggressiveness, trade size, and herding intensity. Price aggressiveness, ${TrdAgg}_{ijt}^d$, measures the immediacy of orders. For buy-side trades, this is defined as the natural logarithm of the ratio of investor $j$’s daily average buy price to the volume-weighted average price of all investors, expressed as a percentage; sell-side aggressiveness is calculated similarly in the opposite direction. Higher values indicate greater immediacy. Trade size, ${TrdSize}_{ijt}^d$, is the natural logarithm of the daily average trade value per order in NT$ thousand. Herding intensity, ${Herd}_{ijt}^d$, is constructed for each stock-trader direction following Lakonishok et al. (1992), Wermers (1999), and Lien et al. (2020a).2

Firm characteristics include time-weighted spread, intraday volatility, price–earnings ratio, firm size, book-to-market ratio, turnover rate, institutional shareholdings, and a cross-listed dummy. To reduce potential endogeneity, lagged values are used. The time-weighted spread, ${Spr}_{i,t-1}$, is calculated by weighting the quote-level percentage spread by its duration. Intraday volatility, ${Vol}_{i,t-1}$, is the standard deviation of transitory log returns in basis points. The price–earnings ratio, ${PE}_{i,t-1}$, is the stock price compared to the latest EPS, while firm size, ${Size}_{i,t-1}$, is the natural logarithm of market capitalization (NT$ million). The book-to-market ratio, ${BM}_{i,t-1}$, is last year’s book value of common shares relative to market capitalization. Turnover rate, ${TO}_{i,t-1}$, is shares traded divided by shares outstanding, as a percentage. Institutional shareholdings, ${InstHold}_{i,t-1}$, are measured as the total percentage ownership by the three main categories of institutional investors: foreign investors, investment trusts, and dealers. The cross-listed dummy, ${CroList}_{it}$, equals one if the firm issues DRs overseas, and is zero otherwise. To capture asymmetric price effects, we included current-day positive (${PosRet}_{it}$) and negative (${NegRet}_{it}$) returns.

While this paper follows Lien and Hung (2023) in adopting an extended sample period, it further incorporates macroeconomic factors into the analysis. Specifically, we consider both foreign exchange rates and interest rates. Changes in the exchange rates of the New Taiwan Dollar against the US Dollar over the preceding month are denoted as ${FX}_{i,t-1}$. The interest rate level, represented as ${Interest}_{i,t-1}$, is determined as the midpoint between the bid and ask interest rates. To account for autocorrelation, the lagged IS measure, $MI{S}_{ij,t-1}^d$, is included. Tick size is controlled using five dummy variables, $Tic{k}_{gi,t-1}$, representing the previous day’s closing price within tick groups of NT$0.01, 0.05, 0.1, 0.5, and 1.0, with NT$5 as the reference group. Here, $g$ ranges from one to five. Fixed effects for industry, year, month, and weekday are also included. To examine whether different investor behaviors affect information shares differently, interaction terms between investor type dummies and trading activity variables are incorporated.

Table 5. Determinants of information shares.

Independent Variable	Panels A: Buy Side										Panels B: Sell Side
	Foreign Investors		Investment Trusts		Other Institutions		Individuals		Pooled		Foreign Investors		Investment Trusts		Other Institutions		Individuals		Pooled
$Intercept$	11.79	***	11.21	***	19.27	***	2.07	***	7.70	***	12.54	***	11.29	***	25.32	***	2.44	***	9.72	***
${For}_{it}$									3.12	***									2.77	***
${Trust}_{it}$									2.12	***									2.03	***
${OthInst}_{it}$									5.64	***									7.74	***
${TrdAgg}_{ijt}^d$	−0.02		0.00		−0.12	**	−0.16	***	−0.37	***	−0.04	**	−0.02	**	−0.19	***	0.04	**	−0.10	***
${TrdSize}_{ijt}^d$	0.79	***	0.59	***	1.24	***	0.09	**	1.13	***	0.69	***	0.45	***	0.84	***	0.14	***	0.98	***
${Herd}_{ijt}^d$	−3.99	***	−3.21	***	−6.69	***	−1.31	***	0.82	***	−3.17	***	−3.14	***	−7.46	***	−0.89	***	1.28	***
${For}_{it}\times {TrdAgg}_{ijt}^d$									0.28	***									0.02
${For}_{it}\times {TrdSize}_{ijt}^d$									−0.40	***									−0.31	***
${For}_{it}\times {Herd}_{ijt}^d$									−4.65	***									−4.43	***
${Trust}_{it}\times {TrdAgg}_{ijt}^d$									0.41	***									0.13	***
${Trust}_{it}\times {TrdSize}_{ijt}^d$									−0.11	*									−0.05
${Trust}_{it}\times {Herd}_{ijt}^d$									−3.73	***									−4.09	***
${OthInst}_{it}\times {TrdAgg}_{ijt}^d$									0.14	***									−0.17	***
${OthInst}_{it}\times {TrdSize}_{ijt}^d$									−0.48	***									−0.69	***
${OthInst}_{it}\times {Herd}_{ijt}^d$									−7.78	***									−8.60	***
${Spr}_{i,t-1}$	1.40	***	−0.54		2.33	***	0.79	***	1.51	***	1.62	***	0.14		1.66	***	0.62	***	1.50	***
${Vol}_{i,t-1}$	1.31	**	−3.73	***	−1.52	*	−0.12		−0.53		0.63		−4.13	***	−3.26	***	0.49	**	−1.13	***
${PE}_{i,t-1}$	0.046	***	0.002	**	0.025	***	0.279	***	0.019	***	0.044	***	0.004	***	0.024	***	0.374	***	0.019
${Size}_{i,t-1}$	−1.16	***	−0.59	***	−1.81	***	−0.12	***	−1.02	***	−1.12	***	−0.47	***	−1.95	***	−0.20	***	−1.06	***
${BM}_{i,t-1}$	−2.95	***	−0.61		−2.97	***	−0.76	***	−2.13	***	−2.71	***	−0.72	**	−3.04	***	−0.76	***	−2.23	***
${TO}_{i,t-1}$	−0.52	***	−0.37	***	−0.82	***	−0.18	***	−0.51	***	−0.59	***	−0.33	***	−1.05	***	−0.24	***	−0.63	***
${InstHold}_{i,t-1}$	−0.03	***	−0.01	***	−0.02	***	0.00	***	−0.02	***	−0.03	***	−0.01	***	−0.03	***	0.00	***	−0.02	***
${CroList}_{it}$	0.24	***	−0.11		0.24	***	−0.09	***	0.12	***	0.29	***	−0.26	***	0.14		−0.04	**	0.09	***
${PosRet}_{it}$	−0.04	***	0.08	***	−0.04	***	−0.05	***	−0.03	***	0.01		0.12	***	0.04		−0.10	***	0.01
${NegRet}_{it}$	0.02	**	−0.17	***	−0.09	***	0.07	***	−0.03	***	−0.01		−0.07	***	−0.05	**	0.05	***	−0.01
${FX}_{i,t-1}$	−0.01		0.06	**	0.13	***	0.02	***	0.05	***	−0.02		−0.05	**	0.15	***	0.00		0.04	**
${Interest}_{i,t-1}$	−0.32		0.59		−0.81		−0.13		−0.36		−1.41	***	−0.14		0.06		−0.12		−0.56
${MIS}_{ij,t-1}^d$	0.05	***	0.00	**	0.03	***	0.28	***	0.02	***	0.04	***	0.00	***	0.02	***	0.37	***	0.02	***
${Tick}_{1i,t-1}\left(0.01\le \mathrm{P}<10\right)$	1.90	***	−4.56	***	3.24	***	0.52	**	1.71	***	1.50	***	−4.14	***	1.92	**	0.61	***	1.14	***
${Tick}_{2i,t-1}\left(10\le \mathrm{P}<50\right)$	−0.06		−4.35	***	0.40		0.00		0.03		−0.23		−4.71	***	−1.22	**	0.01		−0.74	***
${Tick}_{3i,t-1} (50\le \mathrm{P}<100)$	−0.23		−4.38	***	−0.23		−0.06		−0.29		−0.29	*	−4.63	***	−1.82	***	−0.05		−1.00	***
${Tick}_{4i,t-1}\left(100\le \mathrm{P}<500\right)$	−0.66	***	−4.02	***	−0.91	***	−0.16	*	−0.73	***	−0.51	***	−4.21	***	−1.79	***	−0.13		−1.14	***
${Tick}_{5i,t-1}\left(500\le \mathrm{P}<1000\right)$	−0.25		−2.70	***	0.15		−0.01		−0.12		−0.12		−3.29	***	0.65		−0.18		−0.28
Fixed effects	Yes		Yes		Yes		Yes		Yes		Yes		Yes		Yes		Yes		Yes
Adjusted $R^2$	0.33		0.10		0.20		0.43		0.22		0.34		0.16		0.17		0.51		0.21
Number of observations	30,165		13,434		29,899		30,175		103,673		30,275		14,368		29,939		30,294		104,876
Chi-Squared Statistics of Wald Tests
${\beta }_{10}-{\beta }_7$ (difference test of price aggressiveness)									0.13	**									0.11	**
${\beta }_{11}-{\beta }_8$ (difference test of trade size)									0.29	***									0.26	***
${\beta }_{12}-{\beta }_9$ (difference test of herding intensity)									0.92	***									0.34	**

Note: ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.

analyzes the factors affecting information shares, with Panels A and B presenting buy- and sell-side trades, respectively. In each panel, the first four models are estimated separately by investor type, and the final model pools all investors. In Panel A, the coefficients for $TrdAg{g}_{ijt}^d$ are mostly insignificant for professional institutions but significantly negative for other institutions and individuals, indicating that aggressive orders by non-professional investors are associated with lower ISs. The coefficients on ${TrdSize}_{ijt}^d$ are significantly positive across all investor types, suggesting that larger trades contribute more to IS, especially for institutions. Furthermore, the estimates on ${Herd}_{ijt}^d$ are significantly negative for both institutional and individual investors, indicating that higher levels of herding intensity generally have a detrimental effect on price discovery.

Turning to the pooled model in Panel A of Table 5, several findings stand out. The coefficients for the three trader dummies are significantly positive, indicating that institutional investors contribute more to price discovery than individuals on average. The coefficient on price aggressiveness, ${TrdAgg}_{ijt}^d$, is significantly negative (−0.37), but the interaction terms with institutional dummies (${For}_{it}\times {TrdAgg}_{ijt}^d$, ${Trust}_{it}\times {TrdAgg}_{ijt}^d$, and ${OthInst}_{it}\times {TrdAgg}_{ijt}^d$) are significantly positive (0.28, 0.41, and 0.14). This suggests that aggressive orders by institutions enhance price discovery despite the overall negative effect of aggressiveness. Trade size, ${TrdSize}_{ijt}^d$, is significantly positive (1.13), while the interaction terms with institutional dummies are negative (−0.40, −0.11, and −0.48), indicating that the marginal effect of institutional trade sizes is smaller than that of individuals. Nevertheless, the net impact remains positive. For instance, a 1% increase in foreign investors’ daily trade size raises IS by 0.73% (1.13–0.40). In sum, institutional investors have a positive net impact on IS due to their larger trade sizes, though their marginal effects are smaller than those of individuals.

Regarding herding intensity, all interaction terms between trader dummies and herding ($Her{d}_{ijt}^d$) are significantly negative (−4.65, −3.73, and −7.78), whereas the main coefficient on herding is significantly positive (0.82). This indicates that institutional herding exerts a stronger adverse effect on price discovery than individual herding. For instance, the coefficients on $Her{d}_{ijt}^d$ and $Trus{t}_{it}\times Her{d}_{ijt}^d$ (0.82 and −3.73) imply that a 1% increase in buy-side herding by investment trusts reduces IS by 2.91%. Likewise, a 1% increase in foreign investors’ buy-side herding decreases IS by 3.83%. As for firm characteristics, trades in stocks with wider spreads, smaller sizes, lower turnovers, and lower institutional ownership (i.e., less liquid stocks) tend to exert a larger impact on IS than trades in other firms.

In summary, the findings suggest that greater price aggressiveness by institutional investors enhances price discovery. Across different trader categories, professional institutions such as foreign investors and investment trusts contribute more to price discovery through aggressive trading than individual investors. With respect to trade size, institutional investors exhibit smaller marginal effects on price discovery, though their overall impact remains positive. This pattern likely reflects the much larger average trade sizes of institutions relative to individuals, which naturally result in smaller marginal effects. In contrast, stronger herding behavior diminishes the contributions of institutions to the price discovery process, with institutional herding exerting a more pronounced negative influence. Overall, this study demonstrates clear links between trading behaviors and price discovery across investor types, providing new insights into the dynamics of informed trading.

We continued to employ the Wald tests to assess whether the trading behaviors of domestic and foreign institutions have differential effects on IS. The test rejects the null hypothesis that ${\beta }_{10}-{\beta }_7=0$, indicating that the price aggressiveness of domestic investment trusts contributes more to price discovery than that of foreign investors. Similarly, the rejections of ${\beta }_{11}-{\beta }_8=0$ and ${\beta }_{12}-{\beta }_9=0$ indicate that the trading behaviors of investment trusts and foreign investors both significantly influence price discovery, albeit in different ways. In particular, foreign investors exert stronger adverse effects on IS through trade size and herding intensity, resulting in lower positive marginal impacts compared with investment trusts. Macroeconomic factors also matter as changes in FX rates affect domestic investors more, while interest rates have a relatively minor impact.

The sell-side results in Panel B of Table 5 exhibit patterns similar to the buy-side results in Panel A. To provide further economic justification, we performed a Chow test to assess whether the regression coefficients differ between the buy- and sell-side datasets. The results support using two separate regression models. Detailed results are available upon request.

Taken together, multiple factors, including investor types, trading behaviors (price aggressiveness, trade size, and herding intensity), firm characteristics, and macroeconomic conditions, are associated with the process of price discovery. Our findings align with the theories of short-horizon adjustment, variance decomposition, and private information. Specifically, foreign investors and investment trusts exhibit strong “price leadership,” utilizing large orders to rapidly steer prices and drive short-term adjustments. Notably, compared to foreign investors, investment trusts exhibit trading motivations that are more strongly constrained by government policy regulations (Mugerman et al., 2019). In contrast, individual investors show significantly weaker dominance, highlighting the disparity in market influence across investor classes.

Regarding variance decomposition, the persistence of lagged MIS terms and the significance of volatility and spreads indicate that informational contributions are reallocated as market frictions change. Crucially, the negative correlation with herding behavior confirms that high MIS stems from independent, private information rather than trend-following. Even after controlling for liquidity, the superior coefficients for institutional traders demonstrate that their status reflects a fundamental advantage in information-gathering and price-discovery capabilities.

4.5. Information Shares and Trading Performance

Trading performance is a critical aspect of security investments. This section assesses trading performance from the perspective of investor types rather than individual stocks. Following Kalev et al. (2008) and Barber et al. (2009), the measurement proceeds as follows:

$${Perf}_{ij(t+1,t+\tau )}={\sum }_{m=1}^{\tau }({\displaystyle \frac{{Buy}_{ijt}-{Sell}_{ijt}}{{Buy}_{ijt}+{Sell}_{ijt}}})R_{i,t+m},$$

(9)

where ${Buy}_{ijt}$ and ${Sell}_{ijt}$ represent the values of buy- and sell-side trades, respectively. The stock return, $R_{i,t+m}$, denotes the logarithmic return on the $m$th day immediately following trading day $t$, where $m$ ranges from 1 to τ. The holding periods considered are one day, three days, one week, two weeks, and one month, corresponding to τ = 1, 3, 5, 10, 21. In addition, we calculated trading performance for the current day, denoted as ${Perf}_{ij0}$.

Table 6. Trading performances across investor types.

Trader Type	${\mathit{P}\mathit{e}\mathit{r}\mathit{f}}_{\mathit{i}\mathit{j}0}$		${\mathit{P}\mathit{e}\mathit{r}\mathit{f}}_{\mathit{i}\mathit{j}1}$		${\mathit{P}\mathit{e}\mathit{r}\mathit{f}}_{\mathit{i}\mathit{j}\left(\mathrm{1,3}\right)}$		${\mathit{P}\mathit{e}\mathit{r}\mathit{f}}_{\mathit{i}\mathit{j}\left(\mathrm{1,5}\right)}$		${\mathit{P}\mathit{e}\mathit{r}\mathit{f}}_{\mathit{i}\mathit{j}\left(\mathrm{1,10}\right)}$		${\mathit{P}\mathit{e}\mathit{r}\mathit{f}}_{\mathit{i}\mathit{j}\left(\mathrm{1,21}\right)}$
Foreign investors	0.37	***	0.10	***	0.13	***	0.13	***	0.16	***	0.11	***
Investment trusts	0.64	***	0.10	***	0.16	***	0.18	***	0.25	***	0.24	***
Other institutions	0.37	***	0.01	***	0.05	***	0.07	***	0.14	***	0.23	***
Individuals	−0.34	***	−0.02	***	−0.03	***	−0.02	***	−0.04	***	−0.03	***

Note: *** denote statistical significance at the 1% levels, respectively.

reports the trading performance across investor types. On the trading day, institutional investors achieve superior performance, whereas individual investors incur losses. Over the subsequent one-month holding period, institutional investors continue to realize positive returns, while individuals persistently experience losses.

Following the theoretical frameworks of Kyle (1985) and Glosten and Milgrom (1985), we posit that leadership in price discovery is a clear manifestation of information asymmetry. In this context, informed investors leverage private information to initiate trades, thereby driving the price discovery process. This “informational lead” enables them to capture superior returns by exploiting price inefficiencies before information is fully impounded into market prices. To mitigate potential endogeneity and the risk of reverse causality, we employed a lagged regression framework where the MIS was measured prior to the evaluation of trading performance. This temporal structure ensures that the informational leadership on day $t$ is assessed before the subsequent trading performance from days $t+1$ to $t+\tau$. Establishing this timeline proves that information shares predict future performance, and not vice versa. Accordingly, we estimated the following regression model using the Newey–West method:

$$\begin{array}{cc}{Perf}_{ij\left(t+1, t+\tau \right)}=& {\beta }_0+{\beta }_1{For}_{it}+{\beta }_2{Trust}_{it}+{\beta }_3{OthInst}_{it}+{\beta }_{4}M{IS}_{ijt}^b+\left({\beta }_5{For}_{it}+{\beta }_6{Trust}_{it}+{{\beta }_{7}OthInst}_{it}\right)\times M{IS}_{ijt}^b+\hfill \\ & {\beta }_8{Agg}_{ijt}^b+{\beta }_9{TrdSize}_{ijt}^b+{\beta }_{10}{Herd}_{ijt}^b+{\beta }_{11}M{IS}_{ijt}^s+({\beta }_{12}{For}_{it}+{\beta }_{13}{Trust}_{it}+{{\beta }_{14}OthInst}_{it})\times \hfill \\ & M{IS}_{ijt}^s+{\beta }_{15}{Agg}_{ijt}^s+{\beta }_{16}{TrdSize}_{ijt}^s+{\beta }_{17}{Herd}_{ijt}^s+{\beta }_{18}{Ret}_{i(t+1\tau , t+\tau )}+Conrols+FE+{\epsilon }_{ij(t+1, t+\tau )}.\hfill \end{array}$$

(10)

Table 7. Impacts of information shares of investor types on trading performances.

Independent Variable	${\mathit{P}\mathit{e}\mathit{r}\mathit{f}}_{\mathit{i}\mathit{j}0}$		${\mathit{P}\mathit{e}\mathit{r}\mathit{f}}_{\mathit{i}\mathit{j}1}$		${\mathit{P}\mathit{e}\mathit{r}\mathit{f}}_{\mathit{i}\mathit{j}\left(\mathrm{1,3}\right)}$		${\mathit{P}\mathit{e}\mathit{r}\mathit{f}}_{\mathit{i}\mathit{j}\left(\mathrm{1,5}\right)}$		${\mathit{P}\mathit{e}\mathit{r}\mathit{f}}_{\mathit{i}\mathit{j}\left(\mathrm{1,10}\right)}$		${\mathit{P}\mathit{e}\mathit{r}\mathit{f}}_{\mathit{i}\mathit{j}\left(\mathrm{1,21}\right)}$
$Intercept$	−0.19		−0.21		−0.32		0.40		−0.02		−0.25
${For}_{it}$	0.59	***	0.07	***	0.11	***	0.04		0.07		−0.03
${Trust}_{it}$	0.49	***	0.10	***	0.05		0.04		0.07		−0.05
${OthInst}_{it}$	0.38	***	−0.01		0.01		−0.01		0.06		0.19	**
$M{IS}_{ijt}^b$	0.08	**	0.02		0.02		−0.01		−0.04		−0.16	*
${For}_{it}\times M{IS}_{ijt}^b$	−0.09	***	−0.02		−0.01		0.02		0.05		0.17	*
${Trust}_{it}\times M{IS}_{ijt}^b$	−0.07	**	−0.02		0.00		0.02		0.06		0.18	*
${OthInst}_{it}\times M{IS}_{ijt}^b$	−0.07	**	−0.03		−0.03		−0.01		0.03		0.12
${Agg}_{ijt}^b$	0.01	*	−0.01	*	−0.01		−0.01	*	−0.01		0.01
${TrdSize}_{ijt}^b$	0.01		0.01		0.03		0.01		0.05		0.11	*
${Herd}_{ijt}^b$	2.96	***	0.07		0.22	*	0.16		0.98	***	1.83	***
$M{IS}_{ijt}^s$	0.09	***	0.01		0.05		0.01		0.07		0.20	**
${For}_{it}\times M{IS}_{ijt}^s$	−0.11	***	0.01		−0.02		0.01		−0.04		−0.18	*
${Trust}_{it}\times M{IS}_{ijt}^s$	−0.08	**	−0.02		−0.06	*	−0.03		−0.08		−0.22	**
${OthInst}_{it}\times M{IS}_{ijt}^s$	−0.09	***	−0.01		−0.04		0.00		−0.05		−0.19	**
${Agg}_{ijt}^s$	0.00		0.00		0.01		0.00		0.00		−0.02
${TrdSize}_{ijt}^s$	−0.03	**	−0.01		−0.03		−0.05		−0.09	**	−0.20	***
${Herd}_{ijt}^s$	2.57	***	0.04		−0.05		−0.15		−0.25		−0.82	**
${Ret}_{i(t+1\tau , t+\tau )}$	−0.01		−0.01	**	−0.01	***	−0.01	*	0.00		−0.01
Controls & fixed effects	Yes		Yes		Yes		Yes		Yes		Yes
Adjusted $R^2$	0.2135		0.004		0.003		0.001		0.002		0.004
Number of obs.	30,299		30,299		30,299		30,299		30,299		30,299

Note: ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.

presents the empirical results. In the first model, the coefficients of the three investor dummies are significantly positive, indicating that institutional investors generally exhibit better performance than individuals. Both $M{IS}_{ijt}^b$ and $M{IS}_{ijt}^s$ show significantly positive coefficients, while the interaction terms are significantly negative. This suggests that a higher IS is associated with improved trading performance, although the incremental association for institutional ISs is smaller than that for individuals. This pattern likely reflects diminishing marginal returns at high levels of institutional informational leadership.

In the first three models, the coefficients for ${For}_{it}$ are significantly positive, indicating that foreign investors generally achieve better trading performance than individuals over the three days following their trades. Similar patterns are observed in the first two models for ${Trust}_{it}$, suggesting that investment trusts also outperform individuals shortly after executing trades. However, IS does not appear to have long-term predictive power for trading performance.

While the estimated coefficients may appear modest in isolation, their economic impact is substantial when scaled by assets under management typical of the institutional investors in our sample. For large funds managing billions in assets, even small informational advantages translate into significant annualized alpha. Our interaction terms reveal that the performance gap between institutional and retail investors is dynamic rather than static. For instance, in environments with high information shares, the “performance premium” for foreign institutions is 0.59% higher than that of domestic individuals. This highlights the practical value of the sophisticated information-processing systems employed by institutional players in the Taiwan market. Furthermore, the lead–lag relationship documented here suggests that retail investors face a systematic disadvantage when trading against the direction of informed institutional flow, providing a clear signal that market participants can utilize to anticipate future price movements.

Our empirical evidence in Table 7 reveals that institutional sell-side presence is more predictive of future price movements than buy-side presence. For example, the interaction terms for investment trusts (${Trust}_{it}\times M{IS}_{ijt}^s$) show significant negative coefficients across various horizons. This suggests that when sophisticated domestic institutions exhibit high information shares on the sell side, it serves as a potent signal for subsequent price corrections. In the Taiwan market, buy orders are often heterogeneous, driven by index tracking, momentum, or speculative retail sentiment. In contrast, sell orders from large foreign investors and investment trusts are more likely to be driven by fundamental reassessments. Our results support the “informed selling” signal, which posits that the market processes sell-side institutional flow is a high-conviction signal of private information.

Given that short-selling on the TWSE involves higher frictions and costs compared to long positions, an aggressive sell-side presence from informed players acts as a more credible signal. Investors are unlikely to overcome these structural hurdles unless they possess significant negative private information. This analysis is further supported by the distinct patterns observed in Table 5, where the determinants of information shares (such as trade size and herding behavior) exhibit different magnitudes across the buy and sell panels. The price discovery process is fundamentally non-linear across trading directions.

To address potential multiway correlation within the error terms of our panel data, we follow Cameron et al. (2011) by accounting for both firm and date dimensions. Accordingly, we re-estimate our regression models to examine the heterogeneous impact of information shares on trading performance across investor types. This estimation controls for fixed effects per industry, year, and month, with standard errors clustered two ways by both firm and date. The empirical results of this robust specification are presented in

Table A2. The heterogeneous impact of information shares on trading performance across investor types.

Independent Variable	${\mathit{P}\mathit{e}\mathit{r}\mathit{f}}_{\mathit{i}\mathit{j}0}$		${\mathit{P}\mathit{e}\mathit{r}\mathit{f}}_{\mathit{i}\mathit{j}1}$		${\mathit{P}\mathit{e}\mathit{r}\mathit{f}}_{\mathit{i}\mathit{j}\left(\mathrm{1,3}\right)}$		${\mathit{P}\mathit{e}\mathit{r}\mathit{f}}_{\mathit{i}\mathit{j}\left(\mathrm{1,5}\right)}$		${\mathit{P}\mathit{e}\mathit{r}\mathit{f}}_{\mathit{i}\mathit{j}\left(\mathrm{1,10}\right)}$		${\mathit{P}\mathit{e}\mathit{r}\mathit{f}}_{\mathit{i}\mathit{j}\left(\mathrm{1,21}\right)}$
$Intercept$	−0.19		−0.21		−0.35		0.44		0.02		−0.28
${For}_{it}$	0.59	***	0.07	***	0.11	***	0.04		0.07		−0.03
${Trust}_{it}$	0.49	***	0.10	***	0.05		0.04		0.08		−0.05
${OthInst}_{it}$	0.38	***	−0.01		0.01		−0.01		0.06		0.19	**
$M{IS}_{ijt}^b$	0.08	**	0.02		0.02		0.00		−0.04		−0.15
${For}_{it}\times M{IS}_{ijt}^b$	−0.09	***	−0.02		−0.01		0.01		0.05		0.17	*
${Trust}_{it}\times M{IS}_{ijt}^b$	−0.07	**	−0.02		0.00		0.02		0.05		0.18	*
${OthInst}_{it}\times M{IS}_{ijt}^b$	−0.07	**	−0.03		−0.03		−0.01		0.02		0.12
${Agg}_{ijt}^b$	0.01	*	−0.01	*	−0.01		−0.02	*	−0.01		0.01
${TrdSize}_{ijt}^b$	0.01		0.01		0.03		0.01		0.05		0.11	*
${Herd}_{ijt}^b$	2.96	***	0.06		0.22		0.17		0.99	***	1.83	***
$M{IS}_{ijt}^s$	0.09	***	0.02		0.05		0.01		0.06		0.20	**
${For}_{it}\times M{IS}_{ijt}^s$	−0.11	***	0.01		−0.02		0.01		−0.03		−0.18	*
${Trust}_{it}\times M{IS}_{ijt}^s$	−0.08	**	−0.02		−0.06	*	−0.02		−0.08		−0.22	**
${OthInst}_{it}\times M{IS}_{ijt}^s$	−0.09	***	−0.01		−0.04		0.01		−0.05		−0.18	**
${Agg}_{ijt}^s$	0.00		0.00		0.01		0.00		0.00		−0.02
${TrdSize}_{ijt}^s$	−0.03	*	−0.01		−0.03		−0.05		−0.09	**	−0.20	***
${Herd}_{ijt}^s$	2.57	***	0.04		−0.04		−0.16		−0.26		−0.82	**
${Ret}_{i(t+1\tau , t+\tau )}$	−0.01		−0.01		−0.01		−0.03		−0.03		−0.05
Controls & fixed effects	Yes		Yes		Yes		Yes		Yes		Yes
Clustered standard errors	Yes		Yes		Yes		Yes		Yes		Yes
Adjusted $R^2$	0.216		0.006		0.004		0.003		0.004		0.006
Number of obs.	30,299		30,299		30,299		30,299		30,299		30,299

Note: ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.

of Appendix A. As demonstrated in Table 7 and further corroborated by the robustness checks in Table A2, the findings remain highly consistent. Specifically, the coefficients for the primary independent variables maintain their original signs and statistical significance levels across both specifications. The persistent significance of the interaction terms between investor types and modified IS underscores the stability of our findings, regardless of the clustering specifications or the inclusion of additional controls. Furthermore, the nearly identical adjusted $R$-square values across corresponding models validate the reliability and explanatory power of our baseline results.

To address concerns regarding potential mechanical dependence,

Table A3. The heterogeneous impact of information shares on trading performance across investor types based on placebo tests.

Independent Variable	${\mathit{P}\mathit{e}\mathit{r}\mathit{f}}_{\mathit{i}\mathit{j}0}$		${\mathit{P}\mathit{e}\mathit{r}\mathit{f}}_{\mathit{i}\mathit{j}1}$		${\mathit{P}\mathit{e}\mathit{r}\mathit{f}}_{\mathit{i}\mathit{j}\left(\mathrm{1,3}\right)}$		${\mathit{P}\mathit{e}\mathit{r}\mathit{f}}_{\mathit{i}\mathit{j}\left(\mathrm{1,5}\right)}$		${\mathit{P}\mathit{e}\mathit{r}\mathit{f}}_{\mathit{i}\mathit{j}\left(\mathrm{1,10}\right)}$		${\mathit{P}\mathit{e}\mathit{r}\mathit{f}}_{\mathit{i}\mathit{j}\left(\mathrm{1,21}\right)}$
$Intercept$	−0.41	**	−0.19		−0.27		0.48		−0.04		−0.39
${For}_{it}$	0.53	***	0.09	***	0.12	***	0.06		0.08	*	−0.01
${Trust}_{it}$	0.45	***	0.09	***	0.06		0.06		0.07		−0.05
${OthInst}_{it}$	0.35	***	−0.02		−0.01		−0.01		0.06		0.16	**
$M{IS}_{ijt}^b$	0.11		0.04		0.00		−0.07		−0.09		−0.34	*
${For}_{it}\times M{IS}_{ijt}^b$	−0.11		−0.05		0.00		0.07		0.10		0.31
${Trust}_{it}\times M{IS}_{ijt}^b$	−0.09		−0.03		0.02		0.10		0.15		0.43	**
${OthInst}_{it}\times M{IS}_{ijt}^b$	−0.10		−0.04		−0.01		0.05		0.09		0.32
${Agg}_{ijt}^b$	0.01		−0.01	*	−0.01	*	−0.02	*	−0.01		0.01
${TrdSize}_{ijt}^b$	0.02		0.02		0.04		0.02		0.08	*	0.13	**
${Herd}_{ijt}^b$	2.94	***	0.09		0.27	*	0.24		1.10	***	1.97	***
$M{IS}_{ijt}^s$	0.13	*	0.03		0.10		0.10		0.18		0.50	**
${For}_{it}\times M{IS}_{ijt}^s$	−0.16	**	0.01		−0.06		−0.05		−0.12		−0.40	*
${Trust}_{it}\times M{IS}_{ijt}^s$	−0.11		−0.04		−0.14	*	−0.15		−0.25	*	−0.56	***
${OthInst}_{it}\times M{IS}_{ijt}^s$	−0.12	*	−0.03		−0.09		−0.07		−0.16		−0.50	**
${Agg}_{ijt}^s$	0.00		0.00		0.01		0.00		0.00		−0.01
${TrdSize}_{ijt}^s$	−0.02		−0.01		−0.04		−0.06	*	−0.12	***	−0.22	***
${Herd}_{ijt}^s$	2.57	***	0.03		−0.07		−0.21		−0.36		−0.94	**
${Ret}_{i(t+1\tau , t+\tau )}$	−0.01		−0.01		−0.01		−0.03		−0.03		−0.05
Controls & fixed effects	Yes		Yes		Yes		Yes		Yes		Yes
Clustered standard errors	Yes		Yes		Yes		Yes		Yes		Yes
Adjusted $R^2$	0.215		0.006		0.004		0.003		0.004		0.006
Number of obs.	30,299		30,299		30,299		30,299		30,299		30,299

Note: ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.

of Appendix A presents the results of a label-reshuffling placebo test. Compared to the baseline results in Table 7, the majority of key interaction terms (e.g., the interactions between investor types and information shares) either lose statistical significance or exhibit substantially reduced coefficients and inconsistent patterns across horizons. While certain coefficients remain significant, specifically in the current-day model, the overall results lack the robust and systematic alignment observed in our primary analysis. This lack of a consistent pattern indicates that our main findings are unlikely to be a mere artifact of the mathematical construction of the indicators, thereby supporting the existence of a genuine relationship between specific investor identities and their informational advantages.

We recalibrate Equation (10) across subsamples based on market conditions, Amihud illiquidity (Amihud, 2002), stock volatility, and bid–ask spread. Bull and bear markets are defined following the work of Bry and Boschan (1971) and Pagan and Sossounov (2003).3 The Amihud illiquidity measure quantifies the price impact of trading using the daily ratio of absolute stock return to dollar trading volume, expressed in return basis points per million trading value.4 Observations equal to or above the median form the high illiquidity group, while those below the median form the low group. Stock volatility is measured as the standard deviation of daily returns over the previous month, expressed as a percentage; observations above or equal to the median comprise the high volatility group, and those below the median comprise the low volatility group. The bid–ask spread is calculated as the daily time-weighted average, expressed as a percentage, with observations at or above the median forming the high spread group and those below forming the low spread group.

Table 8. Extended tests for the impacts of various investor type information shares on current-day trading performances.

Independent Variable	Market Condition						Amihud Illiquidity						Stock Volatility						Bid–Ask Spread
	Bull			Bear			High			Low			High			Low			High			Low
	Coeff.	t-Stat.		Coeff.	t-Stat.		Coeff.	t-Stat.		Coeff.	t-Stat.		Coeff.	t-Stat.		Coeff.	t-Stat.		Coeff.	t-Stat.		Coeff.	t-Stat.
$Intercept$	−0.16	−0.28		0.10	0.31		−0.03	−0.09		−0.12	−0.38		−0.38	−1.14		−0.50	−0.89		−0.64	−2.95	***	−0.12	−0.49
${For}_{it}$	0.56	20.91	***	0.66	19.43	***	0.59	18.93	***	0.61	20.92	***	0.45	14.30	***	0.75	26.29	***	0.58	19.65	***	0.62	19.67	***
${Trust}_{it}$	0.45	11.56	***	0.54	11.56	***	0.47	10.55	***	0.51	12.59	***	0.31	7.00	***	0.64	15.98	***	0.57	13.46	***	0.40	9.71	***
${OthInst}_{it}$	0.36	13.31	***	0.42	11.73	***	0.44	13.73	***	0.34	10.89	***	0.34	10.53	***	0.42	13.80	***	0.43	14.47	***	0.34	10.56	***
$M{IS}_{ijt}^b$	0.09	2.04	**	0.07	1.34		0.10	2.38	**	0.06	1.19		0.12	1.76	*	0.07	1.95	*	0.10	2.34	**	0.08	1.54
${For}_{it}\times M{IS}_{ijt}^b$	−0.11	−2.51	**	−0.08	−1.38		−0.12	−2.77	***	−0.06	−1.12		−0.12	−1.75	*	−0.11	−2.86	***	−0.10	−2.20	**	−0.14	−2.49	**
${Trust}_{it}\times M{IS}_{ijt}^b$	−0.08	−1.82	*	−0.06	−1.15		−0.09	−2.12	**	−0.05	−0.99		−0.10	−1.51		−0.06	−1.67	*	−0.09	−2.16	**	−0.07	−1.32
${OthInst}_{it}\times M{IS}_{ijt}^b$	−0.08	−1.78	*	−0.07	−1.31		−0.10	−2.34	**	−0.06	−1.04		−0.12	−1.69	*	−0.06	−1.63		−0.10	−2.22	**	−0.07	−1.37
${Agg}_{ijt}^b$	0.01	1.01		0.01	1.34		0.01	1.69	*	0.00	0.54		0.02	2.97	***	0.00	−0.90		0.00	0.20		0.01	2.26	**
${TrdSize}_{ijt}^b$	0.00	−0.22		0.02	0.85		0.02	1.06		0.00	−0.10		0.02	1.01		0.00	−0.26		0.00	0.18		0.01	0.69
${Herd}_{ijt}^b$	2.94	32.55	***	3.00	24.21	***	3.13	29.16	***	2.79	28.35	***	3.70	32.63	***	2.31	25.07	***	3.26	31.79	***	2.63	25.47	***
$M{IS}_{ijt}^s$	0.08	2.01	**	0.09	1.54		0.06	1.39		0.13	2.34	**	0.03	0.45		0.12	3.33	***	0.08	1.82	*	0.11	2.13	**
${For}_{it}\times M{IS}_{ijt}^s$	−0.11	−2.58	***	−0.12	−1.96	*	−0.08	−1.78	*	−0.18	−3.13	***	−0.07	−1.08		−0.13	−3.25	***	−0.11	−2.52	**	−0.12	−2.16	**
${Trust}_{it}\times M{IS}_{ijt}^s$	−0.07	−1.69	*	−0.09	−1.54		−0.06	−1.32		−0.12	−2.17	**	−0.03	−0.43		−0.11	−3.02	***	−0.08	−1.86	*	−0.09	−1.71	*
${OthInst}_{it}\times M{IS}_{ijt}^s$	−0.08	−2.03	**	−0.09	−1.51		−0.06	−1.39		−0.13	−2.38	**	−0.04	−0.57		−0.12	−3.19	***	−0.08	−1.80	*	−0.12	−2.17	**
${Agg}_{ijt}^s$	0.00	−0.50		0.01	1.71	*	0.01	1.51		0.00	−0.85		0.00	0.83		0.00	−0.37		0.00	0.89		0.00	0.27
${TrdSize}_{ijt}^s$	−0.03	−2.00	**	−0.03	−1.04		−0.01	−0.53		−0.04	−2.38	**	−0.03	−1.25		−0.03	−1.54		−0.01	−0.62		−0.04	−2.14	**
${Herd}_{ijt}^s$	2.44	28.93	***	2.79	23.72	***	2.74	27.67	***	2.38	24.86	***	3.05	28.98	***	2.11	23.67	***	2.68	27.14	***	2.46	25.63	***
${Ret}_{i(t+1\tau ,t+\tau )}$	−0.01	−1.10		0.00	−0.30		−0.02	−1.67	*	0.01	0.63		−0.03	−2.98	***	0.02	2.01	**	−0.01	−1.38		0.01	0.46
Controls & fixed effects	Yes			Yes			Yes			Yes			Yes			Yes			Yes			Yes
Adjusted $R^2$	0.2115			0.2171			0.2100			0.2195			0.2233			0.2093			0.2176			0.2118
Number of Obs.	18,773			11,526			16,239			14,060			16,251			14,048			16,248			14,051

Note: ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively.

reports extended tests on the effects of information shares from various investor types on current-day trading performance. Across all subsamples, the coefficients on investor-type dummies are consistently positive, indicating that institutional investors outperform retail investors on the trading day. During bull markets and among firms with low stock volatility or high bid–ask spreads, both buy- and sell-side IS measures are significantly positive. In other words, bull markets, low volatility, and high spreads create conditions that enhance trading performance. In bull markets, rising prices generate positive momentum that encourages trading and facilitates profits. Low-volatility stocks allow traders to better anticipate price movements with lower risk, while high-spread stocks, despite higher transaction costs, offer opportunities for skilled traders to achieve gains.

The interaction term estimates show a significantly negative impact during bull markets and for firms with high Amihud illiquidity or high bid–ask spreads. These results suggest that the marginal effects of institutional ISs on trading performance are smaller under these conditions compared with individual investors’ IS, even though the average IS remains positively associated with performance. Overall, the extended tests confirm that the effects of ISs from different investor types on trading performance are influenced by market conditions, Amihud illiquidity, stock volatility, and bid–ask spread.

5. Concluding Remarks

While numerous studies have examined price discovery across various markets, few have specifically explored the relationship between IS and trading performance across different investor groups within a single stock market. This study clarifies the contributions of institutional and individual investors to price discovery in a unified market and links these contributions to trading behavior and performance. Building on the methodologies of Hasbrouck (1995), Lien and Shrestha (2009), and Lien and Hung (2023), we develop a daily stock-trader-direction IS metric. Our empirical analysis reveals several notable findings across investor types and trade directions, filling important gaps in the literature.

Our findings show that different investor types contribute unevenly to price discovery, reflecting their distinct trading styles. Although individuals are the majority and account for a substantial portion of stock price changes in the TWSE, institutional investors achieve a higher IS per order. Among professional institutions, domestic investment trusts outperform foreign investors in IS per order. These results are robust across extended tests, including one-minute IS, high-volatility days, earnings announcements, and macroeconomic influences. When adjusted for order flows, investment trusts contribute most to price discovery, followed by other institutions, foreign investors, and finally individuals. Most investor groups exhibit higher IS during volatile periods, with institutional IS more strongly linked to earnings reports. Macroeconomic factors have a greater impact on domestic investors than on foreign investors.

Regarding behavioral factors, institutional investors’ price aggressiveness contributes more to price discovery than that of individuals. The marginal effect of trade size on IS is slightly smaller for institutional investors, though trade size still has a positive overall impact. Institutional herding behavior tends to have a more adverse effect on price discovery compared with that of individuals. On average, institutional investors outperform individuals, and higher IS generally corresponds to better trading performance. Notably, the link between IS and trading performance is strongest for the current day, with the latter being less predictive of longer-term performance.

While our empirical findings are primarily relevant to other Asian markets with similar trading environments, the TWSE offers unique structural advantages that enhance the internal validity of our study. First, the TWSE is characterized by a high degree of retail participation coupled with sophisticated foreign institutional presence, making it an ideal setting to observe information asymmetry and investor interactions. Second, the high-resolution, investor-specific dataset from the TWSE enables a detailed estimation of information shares that might be less discernible in Western markets characterized by a higher degree of execution fragmentation. Consequently, our findings serve as a benchmark for understanding price discovery mechanisms in order-driven markets transitioning from emerging to developed status.

Due to data constraints, several areas remain beyond the scope of the present study. First, the absence of intraday dealer-specific data precludes a detailed examination of the intermediary role of broker-dealers in the price discovery process. Second, while our results indicate that institutional investors outperform individuals across various horizons, these performance measures represent gross returns and do not explicitly account for shortfalls in implementation (Perold, 1988) or market impact costs. Consequently, future research could utilize granular order-log data to compute high-frequency post-trade markouts (e.g., 5 min or 30 min horizons) to further refine these trading performance estimates. Finally, the transition of the TWSE from call auctions to continuous trading in March 2020 represents a significant structural shift. Future investigation into how this change in market microstructure affects the speed of price discovery, or how it alters asymmetric volatility effects, would provide a valuable extension to the baseline established in this study.

Our findings suggest a potential link between institutional information shares and future price movements. These results point to the possibility that the TWSE could enhance market efficiency by increasing the granularity and frequency of institutional trading data disclosure, which might help reduce the informational disadvantage faced by retail participants. Furthermore, policy considerations could benefit from accounting for the asymmetry in trading direction and intensity identified in our study. To assist retail investors at a systematic disadvantage, educational initiatives should promote the use of public transaction data and institutional positioning to curb impulsive trading. Finally, given the evidence that the sell-side presence of investment trusts relates to subsequent price corrections, regulators such as the Financial Supervisory Commission (FSC) could consider more focused monitoring of concentrated institutional selling. This is particularly relevant as institutional dominance in price discovery is not only a function of capital size but also of the distinct regulatory environments they operate within. Consistent with Mugerman et al. (2019), institutional trading patterns often reflect policy-imposed constraints and incentives, which is a phenomenon clearly manifested in the behavior of domestic investment trusts in our study. Such oversight could be critical to mitigating the systemic risks and volatility spikes associated with institutional herding behavior during market downturns.