Investor Psychology in the Bangladesh Equity Market: An Examination of Herding Behavior Across Diverse Market States

Abstract

The results reveal significant evidence of herding in the overall, bearish, and extended crisis market phases during extreme downturns, while the magnitude of market returns in the tail distribution is considered. Asymmetric herding behavior is more pronounced and prevalent, conditioned by market dimensions like return direction, trading volume, and volatility, with CSSD proving more effective than CSAD in detecting asymmetric patterns. Notably, herding strongly appears in the COVID-19 market during periods of abnormally high market volatility, reflecting heightened market sentiment. Applying Dow Theory to delineate bull and bear market phases significantly improved the methodological complexity and analytical depth related to herding behavior. These findings suggest policy implications for regulators and market participants in minimizing herding effects to create an efficient market environment through enhanced market surveillance, improved investor education, and the use of advanced technologies.

1. Introduction

Financial markets have become increasingly volatile and play a critical role in market stability. The rapid development of technology-driven trading activities has reshaped the behavioral patterns of investors, which remain unexplored in the context of equity markets. Investor psychology, particularly behavioral biases, may significantly impact equity market dynamics. Herding behavior is one of the fundamental biases in behavioral finance, reflecting the psychological tendency of market participants to follow others in their investment decisions. This irrational behavior is particularly pertinent in frontier and emerging markets, where lower market efficiency, weaker regulatory frameworks, and information asymmetry may intensify such tendencies.

The Bangladesh equity market, classified as a frontier market by MSCI (Morgan Stanley Capital International), is characterized by small market capitalization, low liquidity, high risk, the limited participation of foreign investors, non-transparent trading practices, weak corporate governance, and increased vulnerability to political tensions. It is predominately influenced by retail investors’ participation, signifying institutional investors’ stabilizing role in the market. These inexperienced and less educated retail investors tend to initiate investment decisions based on psychological factors like market rumors, emotions, and speculation, rather than on investment fundamentals. They do not adopt sophisticated risk management investment strategies. In addition, the equity market has experienced two extreme market shocks: the endogenous 2010–2011 market crash and the exogenous global COVID-19 pandemic. These structural inefficiencies may make markets more prone to behavioral biases such as herding, which stems from investors’ cognitive shortcuts and social pressures. Prior empirical studies (Alam et al. 2017; Mobarek et al. 2008; Shiblu and Ahmed 2015) confirm that the Dhaka Stock Exchange (DSE) represents inefficiency that could provide an environment conducive to herding behavior.

Despite extensive research in the existing literature, very limited studies have investigated how herding behavior manifests under different market states in Bangladesh. The lack of a structured framework to quantify herding across these market states exacerbates market stability and regulatory intervention concerns. Addressing this gap, the study focuses on systematically investigating market-wide herding behavior in a frontier equity market like Bangladesh across diverse market states, including bearish, bullish, crisis, extended crisis, and COVID-19 markets. Comprehending the nature of herding behavior under these market conditions is crucial for policymakers, investors and other market participants to reduce the risks associated with irrational investment choices.

The study employs two commonly applied dispersion metrics—cross-sectional standard deviation (CSSD) and cross-sectional absolute deviation (CSAD)—to analyze whether equity return dispersions decrease as market returns increase, supporting herding behavior. These models are estimated in multiple market conditions considering returns, trading volume, and volatility, even at the tail of the distribution, to capture how investors behave in highly stressed conditions. Dow Theory is used to classify the market into bearish and bullish periods, while crisis and extended crisis markets are identified based on the massive market crash of 2010–2011. This classification provides a more robust and structured framework for examining the study’s market states.

The study’s novelty lies in analyzing herding behavior by identifying bullish and bearish market phases based on the principles of Dow Theory. Additionally, our study provides valuable insights into the asymmetric herding patterns of market participants across diverse market states in a frontier equity market.

The rest of the study is organized as follows: Section 2 presents relevant herding literature, Section 3 provides the landscape of the Bangladesh equity market, Section 4 describes data characteristics and the methodology employed in the study, Section 5 analyzes the empirical results, Section 6 presents a discussion of the results, and finally, Section 7 offers a conclusion and research implications.

2. Literature Review

Herding behavior can be categorized into two dimensions: rational and irrational. Rational herding can be described as a situation wherein the intention of the equity market participants is to execute the investment behaviors of others, overlooking their own information and believing that others possess better information than themselves. Irrational herding assures that investors blindly follow the market flow without assessing the fundamentals of investments, which is driven by the psychological traits of investors. Theoretical models of rational herding include information cascades (Banerjee 1992; Bikhchandani et al. 1992), compensation structure models (Scharfstein and Stein 1990), and reputational motivation herding (Trueman 1994). In the behavioral finance literature, empirical research on herding behavior can be categorized into two broad types, as follows: Herding towards certain stocks using microdata and data on stock-level characteristics. Professional institutional investors do this. The second dimension is market-wide herding, which focuses on measuring the herding tendency of investors using aggregate market data.

Herding behavior in the equity markets represents a violation of rational utility-maximizing finance theory (Lao and Singh 2011), and may produce increased equity market volatility and distort security prices (Venezia et al. 2011; Demirer and Kutan 2006). Herding can be described as the emotional tendency of equity market investors to copy the decisions of others, overlooking their independent judgment (Vieira and Pereira 2013; Li et al. 2018). Empirical studies show that herding behavior may have crucial effects on securities’ fair prices, and can significantly impact the risk–return characteristics of securities (Tan et al. 2008). The tendency of market participants to herd may induce and destabilize the equity market, reflecting the possibility of bringing about, or to some extent contributing to, market crashes or bubbles (Bikhchandani and Sharma 2000; Bekiros et al. 2017; Scharfstein and Stein 1990; Shiller 2003; Kabir and Shakur 2018; Thoma 2013).

2.1. Chronological Overview of Previous Research

2.1.1. Origins and Early Developments (1992–2010)

(Lakonishok et al. 1992) pioneered the study of institutional herding in the US market but failed to document any evidence. The study of (Christie and Huang 1995) has been treated as the originator in the empirical literature, examining market-wide herding behavior in the US equity market. They used CSSD to analyze herding behavior in the US equity market during extreme market movements, but found no evidence of herding activity. Using the CSAD model, (Chang et al. 2000) examined five international equity markets. They found no herding behavior in the US, Hong Kong, and Japan, but documented its presence in Korea and Taiwan. (Caparrelli et al. 2004) confirmed that herding is only present in extreme situations in the Italian market. Tan et al. (2008) and (Chiang and Zheng 2010) showed herding in the Shanghai and Shenzhen A share markets, while (Yao and Tangjitprom 2019) found a strong presence of herding in the Chinese B stock market. Chiang and Zheng (2010) reported herding behaviors in developed and Asian markets, but could not find evidence in Latin American and US markets.

2.1.2. Developments in Herding Behavior Studies (2011–2022)

(Khan et al. 2011) showed that herding does exist in France, Germ:any, Italy, or England. Investors in Abu Dhabi, Dubai, Qatar, Kuwait, and Saudi Arabian equity markets display herding behaviors under different market regimes (Balcilar et al. 2013). Herding is more prevalent in a downward market and during times with high trading volumes in the Chinese market (Lao and Singh 2011). (Prosad et al. 2012) signified the existence of herding behavior in the Indian equity market during the rising market state, but it was not present in the falling market phase. A study by (Poshakwale and Mandal 2014) also confirmed the herding behavior of investors in the Indian equity market. (Mobarek et al. 2014) investigated 11 European countries from 2001 to 2012 and reported that herding behavior prevailed for most countries during the global financial crisis under scenarios of asymmetric market conditions. (Choi and Skiba 2015) tested institutional herding behavior in 41 countries and suggested a greater tendency to herd in markets with a high level of information asymmetry. (Rahman et al. 2015) examined herding behavior in the Saudi stock market and reported pervasive herding among retail investors, irrespective of up and down market conditions. (Lam and Qiao 2015) confirmed the herding behavior in the Hong Kong market during periods of high trading volume, high and low volatility, and a bullish market. (Vo and Phan Dang 2016) found the presence of herding behavior in the Vietnam stock market. (Arjoon and Bhatnagar 2017) tested herding behavior in the Singapore stock market and discovered the apparent existence of herding at both the portfolio and aggregate market levels.

(Adem 2020) displayed that investors in the Istanbul equity market show stronger herding behaviors in the declined market than in the increased market. Herding behavior was observed during different crisis periods in the US equity market (Clements et al. 2017). (Economou 2020) examined the four frontier markets, and the results indicate that Romania offers the most intensive evidence of herding compared to other countries across different estimations. (Espinosa-Méndez and Arias 2021) obtained substantial empirical evidence suggesting that the COVID-19 pandemic has intensified herding in the European stock markets. Herding behavior was more intensive after COVID-19 than before the Pandemic in Russia, Poland, the Czech Republic, Hungary, and Slovenia (Fang et al. 2021). (Jiang et al. 2022) demonstrated the clear presence of herding during COVID-19 in six Asian stock markets (Japan, South Korea, China, Hong Kong, Singapore, and Taiwan). (Maquieira and Espinosa Méndez 2022) showed that herding behavior was more pronounced after the COVID-19 period in the Chinese equity market. (Bogdan et al. 2022) examined 15 European countries and observed that herding was most prominent in emerging markets, subsequently trailed by frontier markets and developed markets. (Bangalore Nagendra 2022) found no evidence supporting herding during periods of extreme market conditions or under normal conditions in the Irish stock market. Herding coefficients were insignificant under normal and asymmetric market conditions in the BRICS equity market, except for China and South Africa (Shrotryia and Kalra 2022).

2.1.3. Recent Herding Developments (2023-Present)

(Ah Mand et al. 2023) stated that investors in Shariah-compliant stocks exhibit herding behavior during the upward and downward markets in Malaysia.

(Ampofo et al. 2023) revealed herding behavior was not present before the COVID-19 period in both the USA and UK bullish markets, but it was discovered in both countries’ bullish markets during the COVID-19 period. (Dam et al. 2023) investigated the Vietnam stock market and reported evidence of herding in the period of the global COVID-19 crisis, but not in the pre-COVID-19 period. (Hasan et al. 2023) examined the herding behaviors of 33 global stock markets and found strong evidence of herding driven by non-fundamental information under negative market conditions for all countries. They also found that herding tendency increases with the associated increase in systematic risk. (Hakmaoui and Jebari 2023) confirmed the existence of herding in the US and Morocco, but not in the French and Tunisian stock markets. (Khan and Imam 2023) investigated herding behavior in the Bangladesh equity market, covering a data set from 3 January 2007 to 29 December 2011. The results suggest that herding behavior is more pronounced when market trading volume is high and during a highly upward market state.

(Dhuri and Patkar 2024) reported the prevalence of herding behavior in the Indian equity market only during structural breakpoints and times of high trading volume. (Nguyen and Vo 2024) investigated herding behavior in the Vietnam stock market from 2018 to 30 June 2022, considering before, during, and after COVID-19. The study confirmed the presence of herding activity for the during- and post-COVID-19 periods. The study further shows robust herding behavior conditioned by market liquidity and investors’ information demand during the various COVID-19 periods.

Ahn et al. (2024) examined the relationship between business cycles and herding behavior, and the results confirm that herding is more strongly present in recession than in boom phases, influenced by increased economic uncertainty. Metawa et al. (2024) investigated herding in Egyptian mutual funds and found evidence of herding behavior only in down markets. However, they found that no herding is present in a rising market, as managers follow their own judgment. Political and social issues do not have any positive impact on herding behavior. Yahya et al. (2024) studied factors affecting herding behavior in the Jakarta Islamic Index and found that exchange rate, market volatility, market sentiment, and firm size significantly impact herding behavior. Asymmetric information intensifies the influence of herding on investment decisions, highlighting its role in investors’ irrational behavior. Gavrilakis and Floros (2025) focused on herding behavior and how it interacts with risk aversion in selected sustainable indices. The finding does not reveal any direct herding effects, but implies herding on risk aversion factors, identifying long-term impacts of the contagion in most countries except emerging equity markets.

Xing et al. (2025) used a nonparametric approach to investigate time-varying herding behavior between the US stock market and the A share Hong Kong market. The results show persistent adverse herding in the US and Hong Kong markets and a weakening herding behavior in the Hong Kong market before the global financial crisis, periods without herding, and alternating adverse herding from 2008 to 2021. Medhioub (2025) utilized a quantile regression model from 2011 to 2023 to examine herding behavior in MENA countries against global geopolitical risks. They discovered that herding behavior exists in all markets except Lebanon at the lower 5% quantile level in a down market. Except in Morocco and Saudi Arabia, geopolitical risk significantly influences equity return dispersion. In Jordan and Tunisia, geopolitical risk affects herding behavior, while in Lebanon, it increases return dispersion.

2.2. Hypothesis Development

A number of hypotheses developed in the study have been extensively explored in previous herding behavior research, with different authors investigating its effects on various equity market aspects such as normal markets, up and down markets, high and low market volatility, high and low trading volume, as well as market crises phenomena. These hypotheses are aggregated from several herding research findings, thus integrating ideas that have been analyzed independently in different equity market environments. However, no comprehensive herding study has examined these hypotheses across a structured classification of market states, particularly in frontier equity markets like Bangladesh. The following hypotheses are developed based on existing herding literature, with citations supporting each hypothesis.

H1. 

Herding behavior is more pronounced in extreme market conditions, with a greater tendency during periods of significant market movements and high volatility (Arjoon and Bhatnagar 2017; Bhaduri and Mahapatra 2013; Bangalore Nagendra 2022; Christie and Huang 1995; Indārs et al. 2019; Jirasakuldech and Emekter 2021; Khan and Imam 2023; Dhuri and Patkar 2024; Sadewo and Cahyaningdyah 2022).

During times of extreme market conditions characterized by high price movements and uncertainty, investors tend to follow aggregate market behavior rather than rely on their own analysis. In times of market exuberance and stress, investors may display herd-like behavior due to fear of missing out or panic selling (Chiang and Zheng 2010). In the context of the high uncertainty caused by asymmetric information, a cascade flow of information leads investors to make decisions sequentially based upon the previous decisions being made public information, and hence, information cascade-based herding tends to occur (Bikhchandani and Sharma 2000). During extreme conditions, investors’ rational decision behavior is overshadowed by behavioral biases, producing herding behavior.

H2. 

Herding behavior is more prevalent in a bearish market relative to a bullish market, with a greater tendency to follow market consensus due to risk and loss aversion.

Investors’ psychology may differ depending on the market phases. A loss aversion theory proposes that investors tend to react more strongly in bearish markets as potential losses outweigh equivalent gains, making them more prone to herding dynamics (Kahneman and Tversky 2013). Herding intensity tends to be stronger in bearish markets because investors attempt to minimize perceived losses by following the crowd (Chang et al. 2000).

H3. 

Investors tend to exhibit herding during crisis periods caused by endogenous shocks to the equity market since heightened uncertainty may increase reliance on collective market behavior (Chiang and Zheng 2010; Clements et al. 2017; Economou 2020; Ferreruela and Mallor 2021; Hasan et al. 2023; Jirasakuldech and Emekter 2021; Bhaduri and Mahapatra 2013; Vo and Phan Dang 2016).

Endogenous shock, mainly incurred by massive market crashes, may trigger market-wide distress, leading to panic-driven investor behavior. In this situation, information asymmetry increases, making it difficult for investors to determine true security value and tempting them to imitate the actions of others (Chiang and Zheng 2010). During crisis periods, numerous studies have evidenced significant herding behavior, as investors sacrifice their investment strategies and follow the collective market actions to reduce uncertainty (Bogdan et al. 2022; Chiang and Zheng 2010; Mobarek et al. 2014).

H4. 

Investors tend to exhibit herding during crisis periods caused by exogenous shocks such as the COVID-19 pandemic, due to panic-driven market behavior and heightened uncertainty (Ampofo et al. 2023; Bogdan et al. 2022; Bouri et al. 2019; Dam et al. 2023; Espinosa-Méndez and Arias 2021; Fang et al. 2021; Ferreruela and Mallor 2021; Jiang et al. 2022; Maquieira and Espinosa Méndez 2022; Vidya et al. 2023; Nguyen and Vo 2024).

An exogenous shock originating from an external event like COVID-19 provides a market framework in which uncertainty regarding overall market stability and performance prompts widespread herding behavior. Several studies have documented the significant presence of herding activity during this pandemic, as market participants collectively reacted to various government restrictions and other panic-driven news (Jirasakuldech and Emekter 2021; Nguyen and Vo 2024).

H5. 

Herding behavior may occur in both up and down market conditions, with a greater tendency in the down market due to more investor fear and collective risk aversion attitudes (Adem 2020; Ah Mand et al. 2023; Chiang and Zheng 2010; Lao and Singh 2011; Bhaduri and Mahapatra 2013; Chaffai and Medhioub 2018; Dhuri and Patkar 2024; Prosad et al. 2012; Rahman et al. 2015; Tan et al. 2008).

Herding behavior can be observed in both up and down market movements; investors are more likely to herd during down market trends because of increased anxiety and market pessimism, suggesting that negative sentiment drives stronger collective behavior (Demirer and Kutan 2006). In an up market, herding is mainly driven by positive momentum and speculative bubbles in anticipation of future positive movements. However, during market declines, panic and uncertainty may prompt investors to enact a stronger herding behavior, as they tend to sell off securities in a collective effort to mitigate perceived losses.

H6. 

Herding behavior may prevail during the periods of high or low trading volume, as speculative trading behavior or limited information availability may induce investors’ herding tendency (Arjoon and Bhatnagar 2017; Chang et al. 2000; Choi and Yoon 2020; Galariotis et al. 2015; Jirasakuldech and Emekter 2021; Khan and Imam 2023; Nguyen and Vo 2024; Dhuri and Patkar 2024; Yao and Tangjitprom 2019).

Trading volume represents market activity and liquidity; herding can be present under conditions of high and low volume. When there is a high trading volume, speculative and momentum-driven trading strategies may encourage buying and selling behavior, and encourage following the crowd. When there is a low volume, a lack of public information increases uncertainty, prompting investors to follow the aggregate market rather than considering their information strongly. In a frontier equity market, where information asymmetry is common, and market efficiency is relatively low, herding tendency is more likely to be prevalent under high or low trading volume conditions.

H7. 

Herding behavior may prevail during the periods of high or low market volatility, with stronger herding expected in high volatility conditions in which increased market uncertainty amplifies collective decision-making (Arjoon and Bhatnagar 2017; Batmunkh et al. 2020; Choi and Yoon 2020; Maquieira and Espinosa Méndez 2022; Yao and Tangjitprom 2019; Vo and Phan Dang 2016).

Herding behavior may be strongly influenced by market volatility since abrupt equity market movements induce uncertainty, prompting investors to imitate prevailing market trends (Asgharian et al. 2012). In a high-volatile market, risk perception increases, making investors more dependent on reflecting the crowd’s behavior. Herding may also be apparent in a low-volatility environment if investors regard stable market trends as an indicator of future market direction.

This study contributes to the existing literature in the following ways: Firstly, our study makes a remarkable contribution to the herding literature by introducing an innovative approach to determining bullish and bearish market phases based on the principles of the Dow Theory. No previous study has employed the Dow Theory to distinguish between bullish and bearish markets. Consequently, this distinct market classification extends a fresh methodological development to understand the dynamics of investors’ behavior during different market periods. Secondly, our study makes a significant contribution by delineating asymmetric herding patterns within each market state classification developed in the research, such as bullish, bearish, crisis, extended crisis, and the COVID-19 market, which may provide valuable insights into the differential dynamics of investment behavior shown by market participants across market states.

3. Landscape of Bangladesh Equity Market

Bangladesh’s equity market is a frontier market categorized by MSCI that operates through two stock exchanges: the Dhaka Stock Exchange (DSE) and the Chittagong Stock Exchange (CSE). The DSE is the country’s prime bourse, with a market capitalization of BDT 6,761,820 million as of 25 March 2025.

The Bangladesh Securities and Exchange Commission (BSEC) is the principal regulator responsible for supervising the equity market, implementing compliance with market laws, monitoring trading mechanisms to protect against market manipulation, and approving initial public offerings (IPOs) in order to promote market stability. The Central Depository Bangladesh Limited (CDBL), which oversees the electronic trading and settlement procedures, facilitates the trading activities of both stock exchanges. To ensure market transparency and investors’ interest, BSEC tends to implement various intervention strategies such as trading halts, circuit breakers, and corporate governance policies. Special investment guidelines monitor foreign investment to make sure that domestic financial policies are followed. Despite these regulations prevailing, challenges still remain in our market, like inside trading, manipulation by institutional investors, market gambling, and enforcement issues, which require further regulatory improvement.

The DSE and CSE are actively involved in offering both primary and secondary market services to ensure the smooth flow of funds in the economy. The market also has an over-the-counter (OTC) market platform where non-listed and delisted securities are traded. Both stock exchanges provide a modern electronic trading infrastructure for trading with T+2 settlement, confirming that transactions are finalized within two business days. The DSE has three market indices: DSEX, DS30 and DSES. The DSEX is a broad benchmark market index representing almost 97% of the total equity market capitalization. DS30, a blue-chip index, is composed of the top 30 leading companies based on financial viability and liquidity criteria. DSES is a Shariah-compliant index constructed to align with investment strategies based on Islamic finance principles. Both DSEX and DS30 indices methodologies were designed and developed by Standard & Poor Dow Jones Indices, and implemented on 28 January 2013. The DSE represented a market capitalization of 9.32% of GDP compared to the 60.84% for BSE India Limited at the end of 2023. Our equity market is primarily equity-centric, while the bond market still remains underdeveloped. Although Bangladesh government treasury bonds dominate the DSE in terms of issued capital 77% (as of 2023), these bonds are rarely traded in the secondary market.

Bangladesh’s equity market is heavily composed of retail investors, who make up almost 80% of the total equity market participation. Retail investors in our market tend to exhibit rumors- and speculation-driven trading, rely on collective market behavior, and sacrifice their independent fundamental analyses. This may have consequences in terms of frequent market volatility and securities price distortions, particularly during periods of rapid market movements. Institutional investors, such as bank and non-bank financial institutions, insurance companies, mutual funds, and asset management companies, play a crucial role in stabilizing the market under volatile conditions, although their participation is negligible relative to other emerging and developed economies. Foreign investors have limited participation in our equity market mainly because of governance, regulatory limitations, and liquidity constraints.

To understand investor demographic attributes more, we provide findings from recent research on the Dhaka Stock Exchange. The findings of Khanam (2017) indicate that 61.7% of investors were aged between 31 and 40 years, 52.3% were service holders, and 24% were business people; 60% of them had graduated, while 18% had finished their higher secondary education. Regarding yearly income, 38% had income levels ranging between BDT 300,000 and BDT 600,000, and 19% ranged between BDT 600,000 and BDT 900,000. These results suggest that most investors in our market are retail small equity investors. Regarding experience, 57.3% of investors had a trading experience of 5 years, and 30.7% had 6 to 10 years of experience.

4. Data and Methodology

4.1. Data

This section describes the dataset, including its key characteristics, followed by a methodological framework employed to examine the herding behavior in the Bangladesh equity market across various market states. The data used for the study include the daily closing prices and trading volumes of all companies listed on the Dhaka Stock Exchange (DSE) between 3 January 2010 and 30 December 2021. The data were sourced from the DSE library, which analyzes the annual trade information files. The study considers two variables: equity market returns (determined as the weighted average return of all listed companies) and market trading volume (determined by summing up the daily trade volume of all listed firms). We chose 290 companies, excluding mutual funds and bonds, for the analysis after addressing data issues like non-synchronized trading days and missing values.

The selected sample period (2010–2021) encompasses a crucial phase in the evolution of the Bangladesh equity market, incorporating two significant market crises. The first is an endogenous massive market crash that occurred in 2010–2011, and the second comprises exogenous shocks stemming from the unprecedented global COVID-19 pandemic. The sample also evidences the post-crash regulatory reforms implemented by the Bangladesh Securities and Exchange Commission and the Bangladesh Bank to revive investor confidence and market stability. This provides a unique opportunity to explore whether herding dynamics persisted, declined or evolved as a result of major market changes or policy interventions in a frontier equity market setting like Bangladesh.

The equity return realized for each individual company is determined as follows:

$$R_{i,t}=In\left[{\displaystyle \frac{P_t}{P_{t-1}}}\right]100$$

(1)

where $R_{i,t}$ is the realized equity return of company I at time t. $P_t and$ $P_{t-1}$ are the adjusted closing price of each company at time t and t − 1.

To investigate herding behavior, we classified the whole sample into bearish, bullish, crisis, extended crisis, and COVID-19 markets to provide valuable insights into how the herding tendency of investors may vary under distinct market dynamics. The novel contribution of this research lies in how we classify bearish and bullish markets. The existing literature defines a bearish market as when the market return on a particular day increases, and a bullish market as when the market return decreases on a particular day. In this study, we segregated bearish and bullish markets according to the principles of the Dow Theory.

Table 1. Identification of bull/bear trends according to Dow Theory.

Intermediate Up/Down	Bull/Bear	Start Date	End Date
Up	Bull	9-Sep-2009	17-Feb-2010
Down	Bull	17-Feb-2010	10-May-2010
Up	Bull	10-May-2010	5-Dec-2010
Down	Bear	5-Dec-2010	25-May-2011
Up	Bear	25-May-2011	24-Jul-2011
Down	Bear	24-Jul-2011	6-Feb-2012
Up	Bear	6-Feb-2012	17-Apr-2012
Down	Bear	17-Apr-2012	10-Jul-2012
Up	Bear	10-Jul-2012	23-Sep-2012
Down	Bear	23-Sep-2012	10-Dec-2012
Up	Bear	10-Dec-2012	14-Feb-2013
Down	Bear	14-Feb-2013	29-Apr-2013
Up	Bull	30-Apr-2013	10-Jul-2013
Down	Bull	10-Jul-2013	21-Oct-2013
Up	Bull	21-Oct-2013	6-Feb-2014
Down	Bull	6-Feb-2014	22-Jun-2014
Up	Bull	22-Jun-2014	12-Oct-2014
Down	Bear	12-Oct-2014	4-May-2015
Up	Bear	4-May-2015	5-Aug-2015
Down	Bear	5-Aug-2015	2-May-2016
Up	Bull	2-May-2016	24-Jan-2017
Down	Bull	24-Jan-2017	19-May-2017
Up	Bull	19-May-2017	26-Nov-2017
Down	Bull	26-Nov-2017	28-Mar-2018
Down	Bear	29-May-2018	14-Mar-2021
Up	Bull	15-Mar-2021	28-July-2021
Down	Bear	29-Jul-2021	9-Nov-2021
Up	Bull	10-Nov-2021	30-Dec-2021

segregates the bull and bear periods by applying the Dow Theory.

Applying the Dow Theory, the study classifies the equity market into bearish and bullish phases. This theory, rooted in technical analysis, tends to categorize market trends into primary, secondary, and minor trends by analyzing historical price movements. A primary trend indicates the overall direction of the equity market over a long period, usually lasting from several months to years. Market prices can trend up (bull) or down (bear) based on sustained movements. A secondary trend, which lasts for a few weeks to a few months, is a correction that happens within the primary trend. It is a temporary price decline in a bull market, whereas it is characterized by a temporary increase in prices (rally) in a bear market. A minor trend can be defined as the daily or weekly price changes incurred by news events, market noise, or any speculative trading. Such a trend does not define the market direction. Instead, it contributes to short-term fluctuations.

In Figure 1, we present the evolution of return development along with the bullish and bearish periods that Dow Theory classifies.

According to Dow Theory, the equity market is classified as bullish, where each successive price change exhibits higher highs and higher lows over a sustained period. Conversely, a market is categorized as bearish when successive price movements exhibit a sustained pattern of lower highs and lower lows. An uptrend market is ensured when each successive high (peak) and low (trough) is higher than the prior one. At the same time, a downtrend market is defined when each of the subsequent peaks and troughs are lower than the previous ones. Market trends are only confirmed when they are consistently observed, avoiding short-term fluctuations that might give misleading results. This approach enables a systematic and objective equity market classification without depending on modern technical indicators or oscillators, which are not a component of conventional Dow Theory.

An intermediate trend is the equity market price movements identified within a broader primary trend. The intermediate trend takes place within both bullish and bearish periods. Any part of an intermediate up and down trend that persists throughout a high up period is considered bullish, whereas any part that persists throughout a low down period is considered bearish. A secondary trend is defined as the short-term price fluctuations within an intermediate trend. Therefore, this theory as a method of identifying bearish and bullish markets improves the accuracy and theoretical rigor of market phase classification, contributing a new perspective to the herding literature on how investors’ collective behavior may vary in different phases of market trends.

4.2. Reformulation of Hypotheses

The hypotheses tested in the study are derived from existing herding literature, but this paper extends their application by examining multiple market states, such as overall, bearish, bullish, crisis, extended crisis, and COVID-19 markets. This innovative approach enables a greater understanding of how investors’ herding tendencies may vary across different market states, which has not previously been revealed in a single study. As a result, hypotheses are formulated that emphasize applicability to each market state.

H1. 

Herding behavior may prevail during times of extreme market conditions across all market states.

H2. 

Herding behavior is more prevalent in a bearish market compared to a bullish market, as classified by Dow Theory.

H3. 

Investors may exhibit herding behaviors during a market crisis caused by endogenous equity market shocks.

H4. 

Investors may exhibit herding behaviors during a market crisis caused by exogenous equity market shocks.

H5. 

Herding behaviors may prevail in up-and-down markets, during periods of high and low volatility, and with high and low trade volume across all market states.

Hypothesis 1 is refined in this study by analyzing different market states characterized by abnormal market returns, volatility, and trading volume levels. The respective distributions’ top 5% or 1% tails determine these extreme market conditions. By examining herding dynamics within these extreme market scenarios, the study provides a deeper understanding of investor psychology under stressful and unstable market conditions. Reformulating these hypotheses aligning with the study’s methodological approach will provide a comprehensive investigation of the herding dynamic in the Bangladesh equity market, which has not been attempted in any previous herding literature globally.

4.3. Methodology

In this study, we have used the measures of cross-sectional standard deviation of returns (CSSD) proposed by Christie and Huang (1995) and cross-sectional absolute deviation of returns (CSAD) by Chang et al. (2000) to capture herding behavior in all linear and non-linear estimations. Both dispersion measures are calculated as follows:

$${CSSD}_t=\sqrt{{\sum }_{i=1}^N{\displaystyle \frac{{\left(R_{i,t}-R_{m,t}\right)}^2}{N-1}}}$$

(2)

$$CSA{D}_t={\displaystyle \frac{1}{N}}{\displaystyle \sum _{i=1}^N\left|R_{i,t}-R_{m,t}\right|}$$

(3)

$R_{i,t}$ is the returns of security i at time t. $R_{m,t}$ is an equally weighted realized return of all available securities on day t included in the market portfolio.

4.3.1. Herding Behavior During Extreme Market Movements

$${CSSD}_t=\alpha +{\gamma }_1{D}_t^U+{\gamma }_2{D}_t^D+{\epsilon }_t$$

(4)

$${CSAD}_t=\alpha +{\gamma }_1{D}_t^U+{\gamma }_2{D}_t^D+{\epsilon }_t$$

(5)

where $D_t^U$ takes a value of 1 if the market returns on day t are in the extreme upper tail of the return distributions and zero otherwise, and $D_t^D$ takes a value of 0 if the returns of the market on day t lie in the extreme lower tail of the return distributions.

We here modify the Christie and Huang’s model specified in Equation (4) using the absolute magnitude of market return and squared market return with dummy variables at its tail distributions in a non-linear form, as follows:

$${CSSD}_t=\alpha +{\gamma }_1{D}_t^U\left|R_{m,t}\right|+{\gamma }_2{D}_t^D\left|R_{m,t}\right|+{\gamma }_3{D_t^{U}R}_{m,t}^2+{\gamma }_4{D}_t^D{R}_{m,t}^2+{\epsilon }_t$$

(6)

where $D_t^U$ takes a value of 1 if the market returns on day t are in the extreme upper tail of the return distributions and zero otherwise, and $D_t^D$ takes a value of 0 if the returns of the market on day t lies in the extreme lower tail of the distributions. $\left|R_{m,t}\right|$ is the absolute value of equally weighted market return, while $R_{m,t}^2$ is the market return square. To detect herding behavior, the coefficients ${\gamma }_3$ and ${\gamma }_4$ should be negatively significant.

4.3.2. Herding Behavior Under Non-Linear Normal Market Conditions

Chang et al. (2000) employed the following models to detect herding behavior under non-linear market conditions and capture asymmetric herding in up and down markets:

$${CSAD}_t=\alpha +{\gamma }_1\left|R_{m,t}\right|+{\gamma }_2{R}_{m,t}^2+{\epsilon }_t$$

(7)

The asymmetric herding behavior conditions in up and down markets are as follows:

$${CSAD}_t^{Up}=\alpha +{\gamma }_1\left|R_{m,t}^{Up}\right|+{\gamma }_2{R_{m,t}^{Up}}^2+{\epsilon }_t$$

(8)

$${CSAD}_t^{Down}=\alpha +{\gamma }_1\left|R_{m,t}^{Down}\right|+{\gamma }_2{R_{m,t}^{Down}}^2+{\epsilon }_t$$

(9)

where $\left|R_{m,t}^{Up}\right|$ and $\left|R_{m,t}^{Down}\right|$ are the absolute values of an equally weighted realized return of all available securities on day t when the market is up (down). ${R_{m,t}^{Up}}^2$ and ${R_{m,t}^{Down}}^2$ are the squared market value return. The presence of herding behavior will be detected by the statistically significant negative value of the coefficient ${\gamma }_2$.

4.3.3. Asymmetric Herding Behavior Conditions on High and Low Trading Volume

$${CSAD}_t^{V\text{-}High}=\alpha +{\gamma }_1^{V\text{-}High}\left|R_{m,t}^{V\text{-}High}\right|+{\gamma }_2^{V\text{-}High}{\left(R_{m,t}^{V\text{-}High}\right)}^2+{\epsilon }_t$$

(10)

$${CSAD}_t^{V\text{-}Low}=\alpha +{\gamma }_1^{V\text{-}Low}\left|R_{m,t}^{V\text{-}Low}\right|+{\gamma }_2^{V\text{-}Low}{\left(R_{m,t}^{V\text{-}Low}\right)}^2+{\epsilon }_t$$

(11)

where $\left|R_{m,t}^{V\text{-}High}\right| \mathrm{a}\mathrm{n}\mathrm{d} \left|R_{m,t}^{V\text{-}Low}\right|$ are the equally weighted realized returns of all available securities on day t when trading volumes are high and low, respectively. ${\left(R_{m,t}^{V\text{-}High}\right)}^2 \mathrm{a}\mathrm{n}\mathrm{d}$ ${\left(R_{m,t}^{V\text{-}Low}\right)}^2$ are the squared market returns of this term. A statistically significant negative coefficient of ${\gamma }_2$ implies the presence of herding behavior under the consideration of market trading volume.

4.3.4. Asymmetric Herding Behavior Conditions on High and Low Market Volatility

$${CSAD}_t^{Vol\text{-}High}=\alpha +{\gamma }_1^{Vol\text{-}High}\left|R_{m,t}^{Vol\text{-}High}\right|+{\gamma }_2^{vol\text{-}High}{\left(R_{m,t}^{vol\text{-}High}\right)}^2+{\epsilon }_t$$

(12)

$${CSAD}_t^{vol\text{-}Low}=\alpha +{\gamma }_1^{Vol\text{-}Low}\left|R_{m,t}^{Vol\text{-}Low}\right|+{\gamma }_2^{Vol\text{-}Low}{\left(R_{m,t}^{Vol\text{-}Low}\right)}^2+{\epsilon }_t$$

(13)

where $\left|R_{m,t}^{Vol\text{-}High}\right|$ and $\left|R_{m,t}^{Vol\text{-}Low}\right|$ are the equally weighted realized returns of all available securities on day t when market volatility is high and low, respectively. ${\left(R_{m,t}^{vol\text{-}High}\right)}^2$ and ${\left(R_{m,t}^{Vol\text{-}Low}\right)}^2$ are the squared market returns of this term. A statistically significant negative coefficient of ${\gamma }_2$ implies the presence of herding behavior.

4.3.5. Herding Behavior Under Extreme Trading Volume

$${CSAD}_t=\alpha +{\gamma }_1\left|R_{m,t}\right|+{{\gamma }_2{R}_{m,t}^2+\gamma }_3{D}^{V\text{-}Abn High}{R}_{m,t}^2+{\gamma }_4{D}^{V\text{-}Abn Low}{R}_{m,t}^2+{\epsilon }_t$$

(14)

where ${ D}^{V\text{-}Abn High}$ and $D^{V\text{-}Abn Low}$ are the dummy variables that take a value of 1 if market trading volume on day t lies in the extreme upper tail (lower tail) of the trading volume distributions, and zero otherwise.

4.3.6. Herding Behavior Under Extreme Market Volatility

$${CSAD}_t=\alpha +{\gamma }_1\left|R_{m,t}\right|+{{\gamma }_2{R}_{m,t}^2+\gamma }_3{D}^{Vol\text{-}Abn High}{R}_{m,t}^2+{\gamma }_4{D}^{Vol\text{-}Abn Low}{R}_{m,t}^2+{\epsilon }_t$$

(15)

where $D^{Vol\text{-}Abn High} \mathrm{a}\mathrm{n}\mathrm{d}$ $D^{Vol\text{-}Abn Low}$ are the dummy variables that take a value 1 if market volatility on day t lies in the extreme upper tail (lower tail) of the volatility distributions and zero otherwise.

4.4. Estimation of the Models with AR (1) Term

Time-series data are to be auto-correlated and may result in inaccurate estimations of regression coefficients. To overcome this autocorrelated problem, we estimate each model with the AR(1) term.

5. Empirical Results

Before we estimate the regression model, the cross-sectional standard deviation (CSSD), cross-sectional absolute deviation (CSAD) and market return are tested for unit roots by applying the Augmented Dickey and Fuller (ADF) and Phillip–Peron (PP) tests. The results of the unit roots test confirm that all variables are free of unit roots, as the null hypothesis of the unit root has been rejected at the 1% level in both the Augmented Dickey and Fuller (ADF) and Phillip–Peron tests. Thus, all series of CSSD, CSAD, and Rm are stationary at the level form and can be used for model estimation purposes. We do not report the results here due to space constraints.

5.1. CSSD Regression Results During Extreme Market Movements

In this research, we apply Equations (4) and (5) at 1% and 5% criteria for the lower and upper tails of the market return distributions defined as extreme market movements. The results confirm the statistically positive sign of both dummy variables under the CSSD model, indicating that equity return dispersion tends to increase at an increasing rate, evidencing an absence of herding behavior. This finding is contradictory to Christie and Huang’s (1995) original assumption that investors tend to be aligned with the average collective market behavior during periods of market turmoil. Our result is consistent with the findings of (Ahsan and Sarkar 2013); (Javaira and Hassan 2015; Sharma 2018). We do not report the results here, and they can be available upon request.

We also examine the herding effect under extreme market conditions by using the absolute value of market return with its tail distributions in the non-linear fashion specified in Equation (6).

The results in

Table 2. Regression results of CSSD under extreme market movements using the modified dummy variable model.

This table reports the estimated parameters using the following regressions with AR(1) models: $${CSSD}_t=\alpha +{\gamma }_1{D}_t^U\left|R_{m,t}\right|+{\gamma }_2{D}_t^D\left|R_{m,t}\right|+{\gamma }_3{D_t^{U}R}_{m,t}^2+{\gamma }_4{D}_t^D{R}_{m,t}^2+{\epsilon }_t$$ where $D_t^U$ takes a value of 1 if the market returns on day t are in the extreme upper tail of the return distributions and zero otherwise, and $D_t^D$ takes a value of 0 if the market returns on day t are in the extreme lower tail of the distributions and zero otherwise. The 5% criterion is used as the percentage of the observations in the upper and lower tails of the distribution defined as extreme market conditions. $\left|R_{m,t}\right|$ is the absolute value of an equally weighted realized return of all available securities on day t included in the market portfolio, while $R_{m,t}^2$ is the square of market return. To detect herding behavior, the coefficients ${\gamma }_3$ and ${\gamma }_4$ should be negatively significant. The estimation is carried out for the overall, bearish, bullish, crisis (04/07/2010–30/06/2011), extended crisis (04/01/2010–29/12/2011), and COVID-19 markets (08/03/2020–31/12/2021). Dow Theory is applied to segregate the overall market into bullish and bearish. The crisis market occurred because a massive crash happened in January 2011. The extended market prolongs the crisis period by six months before and after the event. The COVID-19 periods starts on the date when the first patient was identified in Bangladesh.
Regression results at 5% extreme market movements
	α	${\gamma }_1$	${\gamma }_2$	${\gamma }_3$	${\gamma }_4$	Adj. R2
Overall Market	3.60 ***	0.024	1.12 ***	0.030	−0.022 ***	0.224
	(18.7)	(0.02)	(12.43)	(0.06)	(5.77)
Bearish Market	3.35 ***	0.08	1.08 ***	0.015	−0.021 ***	0.244
	(16.0)	(0.07)	(14.02)	(0.03)	(7.09)
Bullish Market	3.58 ***	−0.332	0.214	0.427	0.869	0.211
	(11.7)	(−0.15)	(0.13)	(0.30)	(1.18)
Crisis Market	4.21 ***	0.252	1.31	−0.007	−0.188	0.064
	(5.33)	(0.17)	(1.68)	(−0.01)	(−0.76)
Extended Crisis Market	3.90 ***	−0.174	1.14 ***	0.091	−0.022 ***	0.292
	(7.34)	(−0.13)	(7.63)	(0.13)	(3.42)
COVID-19 Market	5.29 ***	−3.03	0.436	2.43	−0.340	0.214
	(6.24)	(−0.11)	(0.00)	(0.09)	(−0.00)

Note: *** The coefficient is significant at the 1% level. The t-statistics are reported in parentheses. CSSD is the cross-sectional standard deviation. Equation (6) is estimated by the HAC (Newey–West) method to account for heteroscedasticity and autocorrelation. Source: Author’s calculation using EViews software 13.

clearly demonstrate that the coefficient ${\gamma }_4$ is statistically significant and negative for the overall market, bearish, and extended crisis periods at the 5% criterion of extreme market conditions. It confirms that investors in the Bangladesh equity market avoid their own private analysis and tend to follow the aggregate market movements during extreme down-market movements. As similar results are generated in both the 5% and 1% criteria of extreme market conditions, we report only the results from the 5% criterion. It is interesting to observe that herding behavior is exhibited during extreme down market conditions, and this supports hypothesis 1. It suggests that the magnitude of market returns may motivate inexperienced investors to behave irrationally to replicate others’ decisions and be blinded by greed (Luo and Schinckus 2015).

5.2. Examining the Nexus Between the Cross-Sectional Absolute Deviation (CSAD), and Cross-Sectional Standard Deviation (CSSD), and Non-Linear Market Return

Table 3. Panel A: Regression results of the CSAD (CSSD) using a non-linear market return model.

Panel A: Regression results of the CSAD model
This table reports the estimated parameters using the following regressions with AR(1) models: $${CSAD}_t=\alpha +{\gamma }_1\left|R_{m,t}\right|+{\gamma }_2{R}_{m,t}^2+{\epsilon }_t$$ where $\left|R_{m,t}\right|$ is the absolute value of an equally weighted realized return of all accessible securities on day t included in the market portfolio, and $R_{m,t}^2$ is the market return square. A negative and statistically significant value of the coefficient ${\gamma }_2$ would ensure the presence of herding behavior. The estimation is carried out for the overall, bearish, bullish, crisis (04/07/2010–30/06/2011), extended crisis (04/01/2010–29/12/2011), and COVID-19 markets (08/03/2020–31/12/2021). Dow Theory is applied to segregate the overall market into bullish and bearish. The crisis market is identified by a massive crash that happened in January 2011. The extended market prolongs the crisis period by six months before and after the event. The COVID-19 periods starts on the date when the first patient was identified in Bangladesh.
	α	${\gamma }_1$	${\gamma }_2$	Adj. R2	F-Stat
Overall Market	1.723 ***	0.378 ***	0.013 ***	0.889	(570) ***
	(35.04)	(64.98)	(85.27)
Bearish Market	1.752 ***	0.331 ***	0.012 ***	0.933	(425) ***
	(23.47)	(38.08)	(74.48)
Bullish Market	1.641 ***	0.175 ***	0.189 ***	0.663	(432) ***
	(22.93)	(3.05)	(7.07)
Crisis Market	1.808 ***	0.520 ***	−0.059 **	0.725	(126) ***
	(8.44)	(5.05)	(−2.20)
Extended Crisis Market	1.660 ***	0.411 ***	0.010 ***	0.941	(169) ***
	(14.03)	(19.82)	(26.04)
COVID-19 Market	2.461 ***	0.078	0.148	0.556	(100) ***
	(22.98)	(0.493)	(1.153)
Panel B: Regression results of the CSSD model
Overall Market	3.333 ***	0.828 ***	−0.015 ***	0.222	(200) ***
	(16.43)	(9.19)	(−4.70)
Bearish Market	3.095 ***	0.697 ***	−0.020 ***	0.240	(96) ***
	(13.66)	(3.29)	(−6.95)
Bullish Market	3.539 ***	−0.217	0.796 ***	0.213	(60) ***
	(9.88)	(−0.39)	(2.90)
Crisis Market	3.557 ***	1.416	−0.236	0.081	(5.21) ***
	(3.94)	(1.51)	(−0.914)
Extended Crisis Market	3.587 ***	0.716 ***	−0.025 ***	0.196	(27.2) ***
	(6.26)	(2.52)	(−3.40)
COVID-19 Market	5.541 ***	−1.739	1.334	0.217	(23.5) ***
	(6.41)	(−0.61)	(0.34)

Note: *** The coefficient is significant at 1% level. ** The coefficient is significant at the 5% level. The t-statistics are reported in parentheses. CSAD is the cross-sectional absolute deviation. Equation (7) is estimated using the HAC (Newey–West) method to account for heteroscedasticity and autocorrelation. Source: Author’s calculation using EViews software 13.

provides a non-linear specification of the CSAD and CSSD regression results to measure the tendency of herd behavior among investors in the Bangladesh equity market.

Panel A of Table 3 shows the statistically significant negative coefficient of the non-linear herding component only in the crisis market under the CSSD model. In times of crisis, investors imitate the actions carried out by others in the market to a greater extent by sacrificing their information and knowledge. The findings support those from the study by Khan and Imam (2023) on herding behavior in the Bangladesh equity market. The presence of herding during the crisis period confirms hypothesis 3 and is also supported by the studies of (Clements et al. 2017; Economou 2020; Li et al. 2018; Litimi 2017; Ouarda et al. 2013; Shah et al. 2017). Panel B shows that herding did exist in the overall market, bearish, and extended crisis periods under the CSSD model, whereas it is absent in the bullish market. During periods of uncertainty, individual investors suppress their own private information and follow the market consensus.

5.3. Examining the Asymmetric Effect of Herding Behavior Under Different Market States: Up and Down Markets

The behavior of investors in the equity market may represent an asymmetric pattern depending on the different market conditions. The direction of market return (up or down) may affect investors’ herding tendencies differently (Chang et al. 2000; Demirer and Kutan 2006). Table 3 provides the results estimated using Equations (8) and (9) for both CSAD and CSSD measures.

Panel A in

Table 4. Regression results of the CSAD (CSSD) model under up- and down-market conditions.

Panel A: Regression results of CSAD
This table reports the parameters estimated using the following regressions in AR(1) models: $${CSAD}_t^{Up}=\alpha +{\gamma }_1\left|R_{m,t}^{Up}\right|+{\gamma }_2{R_{m,t}^{Up}}^2+{\epsilon }_t$$ $${CSAD}_t^{Down}=\alpha +{\gamma }_1\left|R_{m,t}^{Down}\right|+{\gamma }_2{R_{m,t}^{Down}}^2+{\epsilon }_t$$ where $\left|R_{m,t}^{Up}\right|$ and $\left|R_{m,t}^{Down}\right|$ are the absolute values of an equally weighted realized return of all accessible securities on day t when the market is up (down). ${R_{m,t}^{Up}}^2$ and ${R_{m,t}^{Down}}^2$ are the squared market value returns. The prevalence of herding behavior can be detected by the statistically significant negative value of the coefficient ${\gamma }_2$. The estimation is carried out for the overall, bearish, bullish, crisis (04/07/2010–30/06/2011), extended crisis (04/01/2010–29/12/2011), and COVID-19 markets (08/03/2020–31/12/2021). Dow Theory is applied to segregate the overall market into bullish and bearish. The crisis market is identified by a massive crash that happened in January 2011. The extended market prolongs the crisis period by six months before and after the event. The COVID-19 periods starts on the date when the first patient was identified in Bangladesh.
	Regression results from up-market return			Regression results from down-market return
	α	${\gamma }_1$	${\gamma }_2$	α	${\gamma }_1$	${\gamma }_2$
Overall Market	1.74 ***	0.515 ***	−0.067	1.61 ***	0.504 ***	0.010 ***
	(25.98)	(11.4)	(−3.6) ***	(29.55)	(56.52)	(44.66)
Bearish Market	1.72 ***	0.577 ***	−0.089	1.65 ***	0.483 ***	0.011 ***
	(18.57)	(8.88)	(−3.59) ***	(20.31)	(52.23)	(45.50)
Bullish Market	1.66 ***	0.347 ***	0.085	1.55 ***	0.122	0.224
	(16.11)	(3.86)	(1.70)	(18.92)	(1.26)	(5.56) ***
Crisis Market	1.63 ***	0.766 ***	−0.132 *	1.67 ***	0.598 ***	−0.014
	(3.65)	(3.36)	(−1.83)	(8.58)	(3.52)	(−0.35)
Extended Crisis Market	1.55 ***	0.605 ***	−0.088 **	1.49 ***	0.632 ***	0.008 ***
	(7.19)	(4.99)	(−1.94)	(13.53)	(28.52)	(12.98)
COVID-19 Market	2.502 ***	−0.208	0.836 ***	2.451 ***	0.120	−0.063
	(30.34)	(−0.87)	(4.88)	(12.13)	(0.43)	(−0.29)
Panel B: Regression results of CSSD measure
Overall Market	3.789 ***	−0.140	0.100	3.095 ***	1.171 ***	−0.022 ***
	(12.66)	(−0.23)	(0.27)	(11.58)	(10.60)	(−4.97)
Bearish Market	3.514 ***	−0.152	0.144	2.859 ***	1.167 ***	−0.021 ***
	(8.45)	(−0.16)	(0.41)	(10.92)	(11.78)	(−5.2)
Bullish Market	3.601 ***	0.011	0.237	3.146 ***	0.638	0.766 ***
	(8.59)	(0.01)	(0.40)	(5.40)	(0.74)	(20.3)
Crisis Market	3.845 ***	0.454	−0.022	3.253 ***	2.353	−0.402
	(3.51)	(0.29)	(−0.04)	(20.6)	(31.54)	(−1.01)
Extended Crisis Market	4.350 ***	−1.004	0.318	3.284 ***	1.365 ***	−0.027 ***
	(5.71)	(−1.09)	(0.81)	(3.48)	(6.08)	(−2.8)
COVID-19 Market	5.774 ***	−1.932	1.787	5.773 ***	−3.888	2.371
	(5.91)	(−0.45)	(0.33)	(3.68)	(−0.75)	(0.36)

Note: *** The coefficient is significant at the 1% level. ** The coefficient is significant at the 5% level. * The coefficient is significant at the 10% level. The t-statistics are reported in parentheses. CSSD is the cross-sectional standard deviation and CSAD is the cross-sectional absolute deviation. Equations (8) and (9) are estimated using the HAC (Newey–West) method to account for heteroscedasticity and autocorrelation. Source: Author’s calculation using EViews software 13.

reveals the presence of herding in the overall, bearish, crisis, and extended crisis markets under the CSAD model under up-market conditions, while Panel B provides the results derived using the CSSD model, showing the existence of herding in the overall, bearish, and extended crisis markets during down-market periods. Both results are supported by the negatively significant coefficient of the non-linear market term.

5.4. Examining Asymmetric Herding Effect Under Different Market States: High and Low Volume

The trading volume incorporates the quality and precision of market information as well as information about price movements (Blume et al. 1994).

Table 5. Regression results of the CSAD (CSSD) model in high- and low-volume states.

Panel A: Regression results of the CSAD measure
This table reports the parameters estimated using the following regressions with AR(1) models: $${CSAD}_t^{V\text{-}High}=\alpha +{\gamma }_1^{V\text{-}High}\left|R_{m,t}^{V\text{-}High}\right|+{\gamma }_2^{V\text{-}High}{\left(R_{m,t}^{V\text{-}High}\right)}^2+{\epsilon }_t$$ $${CSAD}_t^{V\text{-}Low}=\alpha +{\gamma }_1^{V\text{-}Low}\left|R_{m,t}^{V\text{-}Low}\right|+{\gamma }_2^{V\text{-}Low}{\left(R_{m,t}^{V\text{-}Low}\right)}^2+{\epsilon }_t$$ where $\left|R_{m,t}^{V\text{-}High}\right|$ $\mathrm{a}\mathrm{n}\mathrm{d} \left|R_{m,t}^{V\text{-}Low}\right|$ are the equally weighted realized returns of all accessible securities on day t when trading volume is high and low, respectively. $\left(R_{m,t}^{V\text{-}High}\right) \mathrm{and}$ ${\left(R_{m,t}^{V\text{-}Low}\right)}^2$ are the squared market returns of this term for high and low trading volumes, respectively. A statistically significant negative coefficient of ${\gamma }_2$ implies the presence of herding behavior. The estimation is carried out for the overall, bearish, bullish, crisis (04/07/2010–30/06/2011), extended crisis (04/01/2010–29/12/2011), and COVID-19 markets (08/03/2020–31/12/2021). Dow Theory is applied to segregate the overall market into bullish and bearish. The crisis market is identified by a massive crash that happened in January 2011. The extended market prolongs the crisis period by six months before and after the event. The COVID-19 periods starts on the date when the first patient was identified in Bangladesh.
	Regression results under a state of high trading volume			Regression results under a state of low trading volume
	α	${\gamma }_1$	${\gamma }_2$	α	${\gamma }_1$	${\gamma }_2$
Overall Market	1.617 ***	0.561 ***	0.010 ***	1.744 ***	0.169 ***	0.052 ***
	(33.33)	(60.36)	(44.44)	(22.76)	(8.24)	(15.07)
Bearish Market	1.834 ***	0.135 ***	0.071 ***	1.647 ***	0.487 ***	0.011 ***
	(26.70)	(4.11)	(18.64)	(18.17)	(72.61)	(58.90)
Bullish Market	1.706 ***	0.080	0.349 ***	1.589 ***	0.275 ***	0.063
	(19.32)	(1.04)	(9.99)	(14.75)	(2.77)	(0.95)
Crisis Market	1.689 ***	0.399 ***	0.006	2.052 ***	0.422 ***	−0.032
	(9.95)	(3.87)	(0.16)	(5.39)	(2.85)	(−0.88)
Extended Crisis Market	1.515 ***	0.555 ***	0.015 ***	1.811 ***	0.322 ***	0.013 ***
	(14.13)	(19.77)	(13.11)	(10.29)	(9.44)	(16.83)
COVID−19 Market	2.359 ***	0.293 ***	0.771 ***	2.773 ***	0.376	0.100
	(18.10)	(1.78)	(3.72)	(9.37)	(1.10)	(0.38)
Panel B: Regression results of the CSSD measure
Overall Market	3.005 ***	1.192 ***	−0.021 ***	3.454 ***	0.132 *	0.150
	(11.0)	(10.61)	(−4.86)	(13.5)	(1.57)	(1.26)
Bearish Market	3.611 ***	−0.201 *	0.123 ***	2.848 ***	0.975 ***	−0.016 ***
	(8.08)	(−1.58)	(2.76)	(11.2)	(14.7)	(−7.79)
Bullish Market	3.684 ***	−1.533 ***	2.330 ***	3.448 ***	0.575	−0.381
	(8.36)	(−2.37)	(8.03)	(6.33)	(0.38)	(−0.20)
Crisis Market	4.086 ***	−0.349	0.173	3.426 **	2.461 **	−0.558
	(4.36)	(−0.26)	(0.23)	(2.07)	(1.98)	(−1.61)
Extended Crisis Market	3.661 ***	1.153 ***	−0.014	3.280 ***	0.987 ***	−0.016
	(3.00)	(2.89)	(−0.37)	(4.87)	(3.17)	(−0.55)
COVID−19 Market	4.240 ***	−0.580	1.660	6.733 ***	−1.499	1.187
	(6.98)	(−0.29)	(1.24)	(3.15)	(−0.28)	(0.17)

Note: *** The coefficient is significant at the 1% level. ** The coefficient is significant at the 5% level. * The coefficient is significant at the 10% level. The t-statistics are reported in parentheses. CSSD is the cross-sectional standard deviation, and CSAD is the cross-sectional absolute deviation. Equations (10) and (11) are estimated using the HAC (Newey–West) method to account for heteroscedasticity and autocorrelation. Source: Author’s calculation using EViews software 13.

reports the results estimated using Equations (10) and (11) for high and low trading volume states under both CSAD and CSSD measures.

The findings affirm that the coefficient of the non-linear term remains positively significant for most of the market conditions, particularly in times of high and low trading volumes under the CSAD model. These findings contradict the assumption that cross-sectional dispersion should be inversely associated with trading volume in a situation where investors tend to herd with correlated market behavior. As a result, the positive significance indicates that trading volume may not be a major driver of herding in the Bangladesh equity market. In the case of the CSSD measure, the coefficient ${\gamma }_2$ becomes significantly negative only for the overall market during days of high trading volume. In a low-volume state, there is evidence of asymmetric herding for the bearish market, attributed to increased risk aversion.

5.5. Examining Asymmetric Herding Effect Under Different Market States: High and Low Volatility

Market volatility conditions may influence herding behavior. Investors might find themselves more comfortable suppressing their prior beliefs in favor of market consensus during periods of high volatility and oscillation (Gleason et al. 2004; Tan et al. 2008; Chiang and Zheng 2010; Ferreruela and Mallor 2021).

Panel A in

Table 6. Regression results of the CSAD (CSSD) model under high and low volatility.

Panel A: Regression results of the CSAD measure
This table reports the parameters estimated using the following regressions in AR(1) models: $${CSAD}_t^{Vol\text{-}High}=\alpha +{\gamma }_1^{Vol\text{-}High}\left|R_{m,t}^{Vol\text{-}High}\right|+{\gamma }_2^{vol\text{-}High}{\left(R_{m,t}^{vol\text{-}High}\right)}^2+{\epsilon }_t$$ $${CSAD}_t^{vol\text{-}Low}=\alpha +{\gamma }_1^{Vol\text{-}Low}\left|R_{m,t}^{Vol\text{-}Low}\right|+{\gamma }_2^{Vol\text{-}Low}{\left(R_{m,t}^{Vol\text{-}Low}\right)}^2+{\epsilon }_t$$ where $\left|R_{m,t}^{Vol\text{-}High}\right|$ and $\left|R_{m,t}^{Vol\text{-}Low}\right|$ are the equally weighted realized return of all accessible securities on day t when market volatility is high and low. ${\left(R_{m,t}^{vol\text{-}High}\right)}^2$ and ${\left(R_{m,t}^{Vol\text{-}Low}\right)}^2$ are the squared market returns of this term. A statistically significant negative coefficient of ${\gamma }_2$ implies the presence of herding behavior. The estimation is carried out for the overall, bearish, bullish, crisis (04/07/2010–30/06/2011), extended crisis (04/01/2010–29/12/2011), and COVID-19 markets (08/03/2020–31/12/2021). Dow Theory is applied to segregate the overall market into bullish and bearish. The crisis market is identified by a massive crash that happened in January 2011. The extended market prolongs the crisis period by six months before and after the event. The COVID-19 periods starts on the date when the first patient was identified in Bangladesh.
	Estimation for high-volatility state			Estimation for low-volatility state
	α	${\gamma }_1$	${\gamma }_2$	α	${\gamma }_1$	${\gamma }_2$
Overall Market	1.553 ***	0.611 ***	0.008 ***	1.792 ***	0.065 ***	0.067 ***
	(25.61)	(52.55)	(28.89)	(32.75)	(2.05)	(10.90)
Bearish Market	1.545 ***	0.601 ***	0.009 ***	1.807 ***	0.097 ***	0.065 ***
	(19.37)	(45.84)	(27.05)	(21.10)	(2.38)	(13.56)
Bullish Market	1.483 ***	0.393 ***	0.219 ***	1.692 ***	−0.079	0.491 *
	(10.48)	(2.26)	(3.53)	(18.69)	(−0.39)	(1.81)
Crisis Market	1.892 ***	0.504 ***	−0.053 **	1.811 ***	0.271 *	0.099
	(6.92)	(4.64)	(−1.89)	(8.37)	(1.77)	(0.04)
Extended Crisis Market	1.080 ***	0.853 ***	0.003 ***	1.799 ***	0.041 *	0.071 ***
	(9.35)	(21.30)	(2.48)	(12.02)	(1.65)	(2.96)
COVID-19 Market	2.519 ***	0.132 *	0.214	2.398 ***	0.201	0.703
	(9.41)	(1.69)	(1.07)	(27.13)	(0.33)	(0.59)
Panel B: Regression results of the CSSD measure
Overall Market	2.746 ***	1.290 ***	−0.024 ***	3.777 ***	0.547	0.229
	(8.38)	(10.00)	(−4.18)	(14.32)	(−1.80) **	(0.64)
Bearish Market	2.414 ***	1.163 ***	−0.019 ***	3.524 ***	0.411*	0.215
	(8.41)	(13.08)	(−6.16)	(11.47)	(−1.77)	(0.82)
Bullish Market	3.161 ***	−0.481	1.512 ***	3.889 ***	3.048 **	4.929 **
	(3.75)	(−0.315)	(2.64)	(8.92)	(−1.86)	(1.84)
Crisis Market	3.100 ***	1.980 ***	−0.345	3.873 ***	8.425	−16.87
	(2.65)	(1.83)	(−1.29)	(2.21)	(0.57)	(−0.71)
Extended Crisis Market	2.896 ***	1.347 ***	−0.025 ***	4.390 ***	−0.901**	0.267 ***
	(4.01)	(3.88)	(−2.59)	(11.26)	(−1.92)	(3.24)
COVID-19 Market	3.790	2.935	−0.883	6.009	−5.613	8.410
	(1.30)	(0.37)	(−0.18)	(7.38) *	(−0.86)	(0.67)

Note: *** The coefficient is significant at the 1% level. ** The coefficient is significant at the 5% level. * The coefficient is significant at the 10% level. The t-statistics are reported in parentheses. CSSD is the cross-sectional standard deviation, and CSAD is the cross-sectional absolute deviation. Equations (12) and (13) are estimated using the HAC (Newey–West) method to account for heteroscedasticity and autocorrelation. Source: Author’s calculation using EViews software 13.

shows that herding behavior prevails only in crisis markets during periods of high volatility under the CSAD model. This suggests that high uncertainty during a crisis may prompt investors to ignore their own analysis and adhere to market consensus. Panel B presents a broader presence of herding in the entire, bearish and extended crisis markets during periods of high market volatility under the CSSD measure, as evidenced by the negatively significant coefficient of ${\gamma }_2$.

5.6. Herding Behavior Under Extreme Trading Volume and Market Volatility

We also examine whether investors tend to herd under conditions of extreme trading volume and market volatility by using the magnitude of market return at its tail distributions in a non-linear fashion, as specified in Equations (14) and (15).

Table 7. Regression results of the CSAD under conditions of extreme trading volume and extreme market volatility using the modified dummy variable model.

Regression Results of the CSAD Measure at the 5% Extreme Trading Volume and Market Volatility
This table reports the parameters estimated using the following regressions in AR(1) models: $${CSAD}_t=\alpha +{\gamma }_1\left|R_{m,t}\right|+{{\gamma }_2{R}_{m,t}^2+\gamma }_3{D}^{V\text{-}Abn High}{R}_{m,t}^2+{\gamma }_4{D}^{V\text{-}Abn Low}{R}_{m,t}^2+{\epsilon }_t$$ $${ CSAD}_t=\alpha +{\gamma }_1\left|R_{m,t}\right|+{{\gamma }_2{R}_{m,t}^2+\gamma }_3{D}^{Vol\text{-}Abn High}{R}_{m,t}^2+{\gamma }_4{D}^{Vol\text{-}Abn Low}{R}_{m,t}^2+{\epsilon }_t$$ where ${ D}^{V\text{-}Abn High}$ ${(D}^{Vol\text{-}Abn High)}$ is the dummy variable that takes a value 1 if market trading volume (market volatility) on day t lies in the extreme upper tail of the trading volume (volatility) distributions and zero otherwise. $D^{V\text{-}Abn Low}$ ${(D}^{Vol\text{-}Abn Low})$ is the dummy variable that takes a value 1 if market trading volume (market volatility) on day t falls into the extreme lower tail of the trading volume (volatility) distributions and zero otherwise. $\left|R_{m,t}\right|$ is the absolute value of an equally weighted realized return of all accessible securities on day t included in the market portfolio, and $R_{m,t}^2$ is the market return square. The 5% criterion is used as the percentage of observations in the upper and lower tails of the distribution to define the extreme market conditions. To detect herding behavior, the coefficients ${\gamma }_3$ and ${\gamma }_4$ should be significantly negative. The estimation is carried out for the overall, bearish, bullish, crisis (04/07/2010–30/06/2011), extended crisis (04/01/2010–29/12/2011), and COVID-19 markets (08/03/2020–31/12/2021). Dow Theory is applied to segregate the overall market into bullish and bearish. The crisis market is identified by a massive crash that happened in January 2011. The extended market prolongs the crisis period by six months before and after the event. The COVID-19 periods starts on the date when the first patient was identified in Bangladesh.
	${\gamma }_1$	${\gamma }_2$	${\gamma }_3$	${\gamma }_4$	${\gamma }_1$	${\gamma }_2$	${\gamma }_3$	${\gamma }_4$
Under extreme trading volume					Under extreme market volatility
Overall Market	0.386 ***	0.013 ***	−0.140 ***	−0.001	0.418 ***	0.012 ***	−0.066 ***	0.064 ***
	(0.000)	(0.000)	(0.000)	(0.978)	(0.000)	(0.000)	(0.001)	(0.000)
Bearish Market	0.364 ***	0.013 ***	−0.204 **	−0.065	0.413 ***	0.012 ***	−0.104 ***	0.058
	(0.000)	(0.000)	(0.026)	(0.212)	(0.000)	(0.000)	(0.000)	(0.031)
Bullish Market	0.183 ***	0.187 ***	0.177 ***	−0.043	0.242 ***	0.172 ***	−0.043	0.054
	(0.000)	(0.000)	(0.001)	(0.416)	(0.000)	(0.000)	(0.384)	(0.100)
Crisis Market	0.504 ***	−0.043	−0.004	−0.632	0.562 ***	0.064 *	−0.016	−0.015
	(0.000)	(0.126)	(0.968)	(0.289)	(0.003)	(0.098)	(0.915)	(0.940)
Extended Crisis Market	0.459 ***	0.011 ***	−0.136	0.157	0.532 ***	0.010 ***	−0.154 ***	0.166
	(0.000)	(0.000)	(0.198)	(0.598)	(0.000)	(0.000)	(0.004)	(0.10)
COVID-19 Market	0.097	0.128	0.895 **	0.270 ***	0.078	0.384	−0.470 ***	0.027
	(0.637)	(0.346)	(0.050)	(0.007)	(0.720)	(0.036)	(0.004)	(0.649)

Note: *** The coefficient is significant at the 1% level. ** The coefficient is significant at the 5% level. * The coefficient is significant at the 10% level. The p-values are reported in parentheses. CSAD is the cross-sectional absolute deviation. The Equations (14) and (15) are estimated by use of the HAC (Newey–West) method to account for heteroscedasticity and autocorrelation. Source: Author’s calculation using EViews software 13.

reports the CSAD estimations of Equations (14) and (15). The results delineate that the coefficient ${\gamma }_3$ is statistically significant and negative for the entire and bearish markets at the 5% upper tail distribution of trading volume, while it is significantly negative for the entire, bearish, extended crisis, and COVID-19 markets at the 5% upper tail of the distribution of market volatility. These findings further show that investors in the Bangladesh equity market devalue their personal analysis and follow popular market behaviors, revealing a significant herding effect under extreme market conditions, in particular when trading volume and market volatility reach their highest levels.

5.7. Herding Behavior Under Structural Break Analysis

The two equity market crises are significant events experienced by our equity market during the sample period; they are already reflected in the market state classification used in the study. Despite these, we incorporate several structural break analyses to examine whether herding behaviors vary before or after certain market events or policy reforms in the Bangladesh equity market. The reform structural break captures how the regulatory authorities implemented a series of reforms after the massive 2010–2011 market crash to restore investor confidence and stabilize the overall equity market. We include a structural break corresponding to the post-market crash reform period to examine whether these reforms impact herding effects. We develop short-term, medium-term, and long-term reform windows to analyze immediate and delayed behavioral shifts arising from policy changes. The COVID-19 structural break represents the exogenous shock caused by the unprecedented global pandemic that significantly disrupted investor behavior and equity market dynamics. Both structural breaks will help us to evaluate whether investor herding tendencies significantly changes during or after this disruption, providing a more robust examination of investor psychology.

Table 8. Regression results of CSAD under several structural breaks using the dummy variable approach.

This table presents the parameters estimated for structural breaks using the following regressions in the AR(1) model: $${CSAD}_t=\alpha +{\gamma }_1\left|R_{m,t}\right|+{{\gamma }_2{R}_{m,t}^2+\gamma }_3{D}^{COVID}+{\gamma }_4{D}^{COVID}{R}_{m,t}^2+{\epsilon }_t$$ $${CSAD}_t=\alpha +{\gamma }_1\left|R_{m,t}\right|+{{\gamma }_2{R}_{m,t}^2+\gamma }_3{D}^{Reform}{R}_{m,t}^2+{\gamma }_4{D}^{Reform}{R}_{m,t}^2+{\epsilon }_t$$ where $D^{COVID}$ is the dummy variable that takes a value of 1 during the COVID-19 period and zero otherwise, and $D^{Reform}$ is the dummy variable that takes a value of 1 during regulatory reform periods and zero otherwise. $\left|R_{m,t}\right|$ is the absolute value of an equally weighted realized return of all accessible securities on day t included in the market portfolio, and $R_{m,t}^2$ is the market return square. The COVID period covers from 08/03/2020 to 31/12/2021. The short-reform period is from 01/04/2012 to 31/03/2013, the medium-reform window spans from 01/04/2012 to 31/03/2013, and the long-reform window is from 01/04/2012 to 31/12/2021.
	COVID-19 Structural Break	Short-Reform Structural Break	Medium-Reform Structural Break	Long-Reform Structural Break
$\alpha$	1.599 ***	1.801	1.738 ***	1.682 ***
${\gamma }_1$	0.402 ***	0.386	0.383 ***	0.406 ***
${\gamma }_2$	0.0132 ***	0.014	0.013 ***	0.013 ***
${\gamma }_3$	0.811 ***	−0.317	−0.183	0.054
${\gamma }_4$	−0.105 ***	−0.031	−0.088 ***	−0.088 ***
AR(1)	0.761 ***	0.789	−0.029 ***	0.813 ***
Adj. R-Square	0.892	0.889	0.888	0.889

Note: *** The coefficient is significant at the 1% level. CSAD is the cross-sectional absolute deviation. Both equations are estimated using the HAC (Newey–West) method to account for heteroscedasticity and autocorrelation. Source: Author’s calculation using EViews software 13.

reports the results of incorporating several structural breaks into the CSAD model using a dummy variable approach. The findings related to the COVID-19 structural break show strong evidence of herding effects among investors in the Bangladesh equity market. Though the overall return dispersion increased due to heightened uncertainty, the negatively significant coefficient of ${\gamma }_4$ suggests that investors moved collectively in the same direction despite market volatility. This shift in investor behavior confirms psychological convergence under crisis market conditions, consistent with herding behavior.

The short-reform structural break analysis reveals that there is no statistically significant evidence of a shift in herding deviation, as both the dummy variable and the interactive term coefficients are negatively insignificant. However, the positive and significant ${\gamma }_2$ coefficient shows the absence of herding in the pre-reform period. These findings imply that the regulatory reforms introduced immediately after the market crash do not significantly impact investors’ tendency to imitate one another in the short term. The results of the medium-reform structural shifts indicate a decrease in overall equity return dispersion, signaling improved market stability. Though reforms helped reduce noise and market uncertainty, they did not significantly influence collective investor behavior during market movements, as evidenced by the statistically insignificant change in herding behavior. The long-term reform window gives clear evidence of strong herding behavior, as indicated by the negatively significant ${\gamma }_4$ coefficient. While overall equity return dispersion did not change significantly, investors exhibited more collective behavior during market movements. This implies that, even in the presence of ongoing equity market volatility, regulatory reforms may require time to impact investor psychology, with long-term changes gradually fostering behavioral convergence.

6. Discussion of the Results

The Bangladesh equity market shows strong and consistent evidence of herding behavior, with its effects varying across equity market conditions and market states. Our examination of herding behavior during extreme market movements initially yielded no significant evidence. However, by modifying the dummy variables model to incorporate the magnitude of the market returns in its tail distribution, we found herding in the overall, bearish, and extended crisis markets during extreme downward movements. This modification explores the model’s sensitivity to the intensity of market movements, potentially capturing herding behavior more effectively during extreme market downturns.

Examining the non-linear relationship under the CSAD model indicates that herding is observed only in the crisis market, suggesting that investors imitate others’ behaviors, ignoring their own decisions. This finding supports those of previous studies (Clements et al. 2017; Economou 2020; Li et al. 2018; Litimi 2017; Ouarda et al. 2013; Shah et al. 2017) and confirms hypothesis 3 regarding herding during the crisis period. However, under the CSSD model, herding is present in the overall, bearish, and extended crisis markets. This suggests that investors are more inclined toward collective behavior in contexts of heightened risk aversion or negative market sentiment, aligning with behavioral finance theory. This finding develops a key implication that the CSSD model appears to capture herding better than the CSAD model. During times of market stress, CSSD’s sensitivity to extreme return dispersion may be more suitable for identifying imitation-based trading behaviors. The more substantial herding effects under crisis and bearish conditions confirm hypothesis 2.

We also investigated the potential asymmetric herding behavior of investors conditioned on different market dimensions like positive (negative) returns, high (low) trading volume, and high (low) market volatility. In upward market movements, we find strong evidence of herding activity for the overall, bearish, crisis, and extended crisis markets under the CSAD model. This indicates that investing behavior may involve positive feedback trading, where investors follow past price movements by buying when prices increase and selling when prices decrease (Nofsinger and Sias 1999). Investor sentiment and momentum-based trading strategies may drive such behavior in the up-market phase, as investors collectively reinforce the price trends. Using the CSSD model, herding prevails in the overall, bearish, and extended crisis markets during down-market periods. This aligns with the loss aversion theory, stating that investors react more strongly to losses than gains (Tversky and Kahneman 1986). During down markets, investors may suppress their judgment and conform to collective market behavior, exacerbating herding tendencies. These findings are supported by previous studies (Economou 2016; Ouarda et al. 2013; Dang and Lin 2016; Choi and Yoon 2020), confirming hypothesis 5.

Asymmetric herding analysis under the CSAD model affirms that trading volume is not an important driver of herding in the Bangladesh equity market. One possible explanation is that the relatively low market liquidity may limit investors’ ability to engage in collective market behavior. Transaction restrictions may make it difficult for investors to execute trades simultaneously, weakening the expected negative relationship between return dispersion and trading volume. However, under the CSSD model, herding is evident in the overall market during periods of high trading volume. This finding is supported by some studies (Tan et al. 2008; Lao and Singh 2011; Jirasakuldech and Emekter 2021; Khan and Imam 2023; Dhuri and Patkar 2024). On the other hand, under the CSSD model, it is present only in a bearish market, implying that investors respond more strongly to negative market signals and behave in line with market trends to reduce their perceived risks. This validates hypothesis 6, specifically with reference to the findings of (Economou et al. 2011).

Our examination of volatility further delineates the capabilities of two dispersion measures. The CSAD model shows that asymmetric herding was observed under high-volatility conditions in the crisis market. In contrast, the CSSD model captures broader herding effects, with herding being present in the overall, bearish, and extended crisis markets. This finding is consistent with behavioral finance theories, contending that investors lack confidence in their own knowledge and become more prone to mimicking the actions of others during periods of extreme market turbulence. The stronger evidence of herding detection under the CSSD model reinforces the idea that this approach incorporates magnified price movements and rapid shifts in investor sentiment, which are more prominent under volatile market conditions. The existence of herding during periods of high volatility is supported by (Javaira and Hassan 2015; Ferreruela and Mallor 2021; Maquieira and Espinosa Méndez 2022). Interestingly, the analysis of the low-volatility market reveals that neither the CSSD nor the CSAD detects any herding dynamics. This suggests that low market volatility is represented by greater market stability, lower uncertainty, and reduced risk perception among investors. In these environments, investors feel more confident in their own knowledge and strategies, making independent decisions rather than imitating the behaviors of others.

Finally, the more substantial evidence of herding during times of abnormally high trading volumes and volatility reveal two aspects that provide systematic justification. First, this coefficient is significantly higher relative to those observed under other market conditions, ensuring that investors’ herding behavior is intensified when market turbulence reaches extreme levels. This may be attributed to increased risk aversion, amplified uncertainty, and more substantial psychological biases among investors. The drastic surges in market trading activity during these extreme conditions ensure the self-reinforcing feedback loop, whereby investors perceive strong market movements as signals of informed trading and feel forced to imitate others in order to avoid being left behind. Second, the emergence of herding behavior in the COVID-19 period with extreme market volatility and its absence under other market conditions reveals that herding was not a constant feature of equity market dynamics during the pandemic. Instead, it only surfaced during times of extreme stress and uncertainty. During times of extreme market volatility, investors most likely panicked, abandoned their individual analysis, and excessively replicated aggregate market behaviors, resulting in an exceptionally high herding coefficient. This further suggests that extreme volatility leads to more emotional trading and heightened risk aversion among investors, which made them more susceptible to crowd behavior during the COVID-19 pandemic. The findings confirm hypothesis 4 and are also supported by previous studies (Ampofo et al. 2023; Bogdan et al. 2022; Bouri et al. 2019; Dam et al. 2023; Vidya et al. 2023; Nguyen and Vo 2024). In addition to the 5% criterion, the study also employed a 1% tail criterion to ensure the robustness of the herding analysis under conditions of both extreme trading volume and market volatility. The finding remained consistent, evidencing the more substantial herding effects at the 1% criterion. However, we do not present the 1% criterion results here due to space constraints.

Another unique aspect of the study is the superior efficacy of the CSSD compared to the CSAD measure in uncovering latent herding tendencies among investors in a frontier equity market. We also highlight that using Dow Theory to identify bullish and bearish periods contributed significantly to the methodological rigor of our herding analysis.

The structural breaks analysis provides valuable insights into how major market events and regulatory policy reforms shaped investor herding behavior in the Bangladesh equity market. The unprecedented COVID-19 pandemic was accompanied by strong evidence of herding, signifying heightened investor conformity under conditions of extreme market uncertainty. In contrast, the immediate post-crash reform analysis revealed no significant behavioral shifts, confirming that a very short period may be insufficient to influence herding effects. The medium-term reform widow led to a reduction in overall equity return dispersion but not an apparent shift in herding dynamics, suggesting that reform policies initially stabilized the equity market without changing collective behavior. However, herding behavior became statistically significant during the long-term structural reform window, highlighting that regulatory market reforms may require a sufficient period to affect investor psychology. Overall, these findings emphasize that while exogenous shocks might trigger immediate herding effects, regulatory reforms’ effects on investor behavior develop gradually, highlighting the importance of sustained reform policies in shaping collective decision-making and equity market discipline.

7. Conclusions

This study attempts to comprehensively evaluate the empirical validity of the herding behavior among market participants in the Bangladesh equity market by focusing on diverse market states, such as the bullish, bearish, crisis, extended crisis, and COVID-19 markets. In the study, we employ both the CSAD and CSSD dispersion measures to capture herding behavior. The critical findings convey intricate herding patterns that challenge conventional understandings and highlight the distinct dynamics of a frontier market.

We document strong evidence of asymmetric herding in the Bangladesh equity market under different market conditions. Herding is more pronounced in overall, bearish, crisis, and extended crisis markets during up and down periods. It also persists during high-volatility periods. While herding effects are not largely determined by high trading volume, they appear in bearish markets under low-volume conditions due to greater risk aversion, reinforcing the assumption that extreme market volatility intensifies investor imitation-driven trading activity.

One novel finding we document here is that investors’ tendencies to herd were strongly pronounced and more prevalent in the overall and bearish markets during periods of extremely high trading volume. In addition, herding was evident in the overall, bearish, extended crisis and COVID-19 markets during periods of extremely high market volatility. Another remarkable discovery made in the Bangladeshi equity market is that herding only occurred during COVID-19 in periods of extremely high market volatility. An extremely high trading volume may reflect strong market sentiment, driving herding as investors conform to the market majority. On the other hand, extremely high volatility may signal market turbulence, prompting market participants to seek safety by collectively reacting to market trends.

Our structural break findings reveal that endogenous policy reforms and exogenous COVID-19 shocks may affect investor behavior in the Bangladesh market. However, the magnitude and timing of these effects may differ.

Limitations and Policy Implications

This study investigates herding behavior based on aggregate market-level information, overlooking individual stock-level dynamics that could reveal the behavioral trading patterns of institutional investors. Future research could address this by exploring whether the heterogeneity of herding effects among market participants varies across distinct market variations.

Our comprehensive empirical findings on the presence of herding behavior may have significant implications for equity market policymakers and various market participants. They signal potential price distortions, allowing investors to recognize these anomalies and make informed decisions. Investors can also initiate strategies to hedge against downturns or adopt a contrarian strategy of acting against market trends. By incorporating behavioral insights, investors can adhere to long-term investment objectives rather than resorting to short-term market opportunities.

Regulators can improve market transparency by enforcing strict reporting standards and disclosure requirements for listed companies, enabling investors to make informed decisions rather than relying on psychological biases. They can also reduce information asymmetry by offering customized training through workshop or certificate-based courses, ensuring easy access to critical market information. Furthermore, by introducing technology-based trading platforms, regulators can monitor market behavior and regulate speculative trading. Implementing circuit breakers and other trade-related actions can help reduce market volatility. These efforts may promote equity market stability and investor confidence, fostering a more efficient and resilient market environment.