Herd behavior in the Bangladeshi banking sector

Abstract

This study aims to investigate the existence and nature of herding behavior in the banking sector in Bangladesh. It uses daily closing prices of 30 listed commercial banks from January 2007 to December 2018 to detect herding. This study confirms the presence of herding in the banking industry for the whole study period. Further, it finds herding in the banking sector during the domestic financial crisis and no herding in the Global financial crisis. By separating the study period into a rising market and a falling market condition, this paper finds asymmetry in herding behavior in the banking sector. This phenomenon is further confirmed when the extreme market condition of the bull and bear phase is used. Investors in the banking industry flock together ignoring any private information to reap the market return at the bull period while they avoid the crowd during the bear phase.

Keywords:

• Herding behavior
• financial crisis
• banking sector
• asymmetric behavior
• Bangladesh

Jel Classification:

• G14
• G15

1. Introduction

In asset pricing literature in finance, herding refers to an investor’s behavior to pursue the activity of other investors (Javaira & Hassan, 2015; Spyrou, 2013) which causes a regular departure from the true value and price of an asset. The magnitude and consistency of herding behavior direct the market towards unexpected upward movements and asset bubbles and abrupt falls. When conventional theories couldn’t answer the irregularities of asset pricing, the herding behavior provides the possible explanation for the asset market volatility (Tan et al., 2008). Recall Keynes (1936) in understanding the investors’ decision-making based on psychological factors mentioned as “beauty contest” and “animal spirit”. At times of fear and uncertainty, people tend to make impulsive and sentimental decisions and in most cases, these are irrational (Keynes, 1936). The greater the information asymmetry the larger the uncertainty and people take decision-based on short-run asset valuation and disregards fundamentals (Sewell, 2011) that intensify asset bubble and eventually bubble burst (Olsen, 2011). Thus, herd behavior becomes a major force of market volatility (Bekiros et al., 2017).

Earlier studies have conflict in explaining herd behavior theoretically and empirically. Demirer et al. (2010) explain three theories, i.e. Psychological herding, information-related herding, and agency herding, for herd behavior from prior studies. Psychological herding takes place due to the substantial number of non-sophisticated traders and their optimism where pessimistic investors become optimistic and feel secure by following the crowd (Devenow & Welch, 1996; Lux, 1995). Information related herding works in two ways where the informed investors take their actions based on the information that is not available to others (Avery & Zemsky, 1998; Shleifer & Summers, 1990), and the behavior of informed investors signals market movements to the individual investors (Banerjee, 1992; Shiller, 2002) which eventually saves cost and time to make an investment decision. Another explanation arises from agency theory where agents tend to follow others disregarding private information so that their market reputation sustains (Scharfstein & Stein, 1990). Thus, there exist both rational and irrational theories of herd behavior.

Javaira and Hassan (2015) and Sharma (2018) summarize existing literature on herding from two different viewpoints, one is individual stock herding and the other is market-wise herding. Particular stock-wise herding implies investor behavior focusing on individual securities without thinking of other stocks with the same characteristics. This type of herding arises because of the informational surge, reputational case, institutional investment, compensational difficulty, psychological and communication setbacks of the investors (Banerjee, 1992;Bikhchandani et al., 1992; Brennan, 1993; Conlisk, 1980; Lakonishok et al., 1992; Lux, 1995; Nofsinger & Sias, 1999; Pingle, 1995; Roll, 1992; Scharfstein & Stein, 1990; Shiller & Pound, 1986; Trueman, 1994; Wermers, 1999; Zwiebel, 1995). Unlike herding in a particular stock, market-wise herding means investors’ tendency to follow the market movements disregarding fundamentals (Chang et al., 2000; Christie & Huang, 1995). Most of the empirical researches are based on market-wise herding and finds the mixed result of positive responses (Chang et al., 2000; Chiang & Zheng, 2010; Lindhe, 2012; Prosad et al., 2012), negative responses (Demirer et al., 2010; Garg & Jindal, 2014; Lakonishok et al., 1992) of herding behavior.

Prior studies (Cakan & Balagyozyan, 2014; Chang et al., 2000; Javaira & Hassan, 2015) already reasoned the cause and implications of the higher level of herding in the emerging market compared to developed market. Emerging markets have considerable speculative investors, high information asymmetry, and recurrent regulatory intervention which weaken the market efficiency and endorse herding. Even though herding is more prevalent in the emerging market, often composite market index in an emerging market can not reveal the herding behavior because the absence of herding in one industry may offset the presence of herding in another industry. Thus industry-specific herding analysis is more implicative for investors, academicians, and regulatory bodies.

The robust and dynamic financial sector is crucial for the growth and expansion of the economy in Bangladesh (Barua & Barua, 2021; Moudud-Ul-Huq et al., 2020; Yesmine et al., 2022). Even the economy is heavily dependent on the money market; the performance of the banking system is nothing but the growth and prosperity of the economy (Mujeri & Rahman, 2009). Each year government permits new banks and bank branches towards economic development (Banerjee et al., 2020). The market capitalization of banking sectors is about 14 percent of the total market capitalization in Dhaka stock exchange and remains at the top three industries in the economy (DSE, 2021). Again, banks are the financial institutions that survive on the public image, how its stakeholders perceive banks are crucial to its operation and success. Thereby examining the price discovery process in the banking industry can play a substantial role in the decision-making process for the stakeholders. As this study attempts to detect any herding evidence that demonstrates the investors’ behavior to flock together disregarding private or public information, it becomes an empirical confirmation of whether asset prices in the banking sector represent fundamentals or just follow the impulse of the investors. Only Ahsan and Sarkar (2013) studied herding behavior in the Dhaka stock exchange for the period January 2005 to December 2011 and used techniques of Christie and Huang (1995) and Chang et al. (2000). They were unable to detect the presence of herding in the stock market in Bangladesh and reported the investors’ rational decision-making power using their own or otherwise public information. Since earlier studies claimed the inefficiency in this market (Ahmed, 2002; Mobarek et al., 2008; Arefin & Rahman, 2011; Shiblu and Ahmed, 2015;) and reported most investors as less informed (Nguyen et al., 2011; Rahman et al., 2017). Thus, the findings of Ahsan and Sarkar (2013) is quite misleading. Considering the varying herding evidence at the different industries in the same market and its ability to lead absence of herding if anyone uses a composite market index, only industry-wise investigation can lead to a better understanding of the herding behavior. As a result, this study particularly focuses on the herding behavior in banking sector for its diverse implication to the economy.

The remainder of the paper is organized as follows. Section 2 describes the data and methodology of the study and section 3 presents the results and discussion and finally, section 4 concludes the paper.

2. Data and methodology

2.1. Data

This study uses daily data of thirty listed commercial banks from January 2007 to December 2018. The sample includes 3126 trading days and the daily closing stock prices of 30 banks. DSE library is the source of the data used in this study. Stock return is measured as $R_t=log\left(\frac{R_t}{R_{t-1}}\right)$. The market return of this industry is calculated by averaging the daily return of all banks at the time.

2.2. Methodology

The study uses the widely accepted methodology of cross-sectional standard deviation (CSSD) and cross-sectional absolute deviation (CSAD) proposed by Christie and Huang (1995) and Chang et al. (2000), respectively to find the herding pattern in banks listed in Dhaka Stock Exchange, Bangladesh. The recommended CSSD method follows;

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

Where $R_t$ is the individual stock return and $R_{m,t}$ is the market return calculated by averaging return of N securities at t. N and t represent no. of securities in the banking sector and time, respectively. Now the CSSD is regressed on a constant and two dummy variables representing extreme market behavior.

(2) ${CSSD}_t=\alpha +{\beta }_1{D}_t^L+{\beta }_1{D}_t^U+{\epsilon }_t$(2)

Where,$D_t^L$, $D_t^U$ are two dummy variables for distinguishing the extreme period, one for a lower extreme period and another is for the upper extreme period. Both periods of extreme are defined at 99%, 95%, and 66% levels in the upper or lower tail of the market return distribution. $D_t^L$ = 1, when market return falls in the lower tail in the distribution otherwise 0 and $D_t^U$=1, when market return falls in the upper tail in the distribution otherwise 0. At the time of extreme period, CSSD becomes smaller. Thus when ${\beta }_1$ and ${\beta }_2$ are negative and significant, then herding exists in the market.

Due to the use of the squared term, the cross-sectional standard deviation of return is highly affected by outliers. A better measure, cross-sectional absolute deviation (CSAD), is proposed by Chang et al. (2000) which not only captures the herding but also assesses the non-linearity with the change in market conditions. Where the calculation of $R_{m,t}$ remains the same as the CSSD model The measure of CSAD is as follows;

(3) ${CSAD}_t=\frac{1}{N_t}\sum _{i=1}^{N_t}\left|R_{i,t}-R_{m,t}\right|$(3)

To detect the herding in the market the following equation is used that is just the same as the equation except for the use of CSAD replacing CSSD as a dependent variable:

(4) ${CSAD}_t=\alpha +{\beta }_1{D}_t^L+{\beta }_1{D}_t^U+{\epsilon }_t$(4)

Now according to Chang et al. (2000) the equation for detecting the non-linear pattern of herding is as follows:

(5) ${CSAD}_t=\alpha +{\delta }_1\left|R_{m,t}\right|+{\delta }_2{R}_{m,t}^2+{\epsilon }_t$(5)

Now the calculated CSAD is regressed on constant, absolute average market return and squared market return at time t. When ${\delta }_1$ is positive and significant, it expresses the direct impact of absolute market return to the variability of individual stock return and when ${\delta }_2$ is negative and significant, it implies the rise in market return lowers the variability of individual stock return. Due to the use of quadratic terms, it captures both herding and non-linearity at the same time. Chang et al. (2000). He claims a nonlinear connection between CSAD and market return if there is herding and linear relationship when there is no herding. In the case of herding, CSAD should rise or fall at less than proportional to the average market return; while in absence of herding, the dispersion grows (fall) proportionately with the increase (decrease) of the average market return.

The model is improved following Cakan and Balagyozyan (2014) by incorporating dummy variables reflecting the period of the world financial crisis in 2008 and the domestic financial crisis in 2010. In following equations, ${DV}_1$ and ${DV}_2$ represent the global financial crisis and domestic share market crash, respectively. Where the period of global financial distress constitutes from 15 September 2008 to 15 July 2009 and the domestic financial crisis covers from 6 December 2010 to 5 August 2012.

(6) ${CSAD}_t=\alpha +{\delta }_1\left|R_{m,t}\right|+{\delta }_2{R}_{m,t}^2+{\delta }_3{DV}_1+{\epsilon }_t$(6)

(7) ${CSAD}_t=\alpha +{\delta }_1\left|R_{m,t}\right|+{\delta }_2{R}_{m,t}^2+{\delta }_3{DV}_2+{\epsilon }_t$(7)

The study then approaches to identify the pattern of herding during rising (upward) and falling (downward) market conditions. As a result, the following models of CASD are used:

(8) ${CSAD}_t^{UP}=\alpha +{\delta }_1^{\hspace{0.17em}UP}\left|R_{m,t}^{UP}\right|+{\delta }_2^{UP}{R}_{m,t}^{2}UP+{\epsilon }_t$(8)

(9) ${CSAD}_t^{DOWN}=\alpha +{\delta }_1^{DOWN}\left|R_{m,t}^{DOWN}\right|+{\delta }_2^{DOWN}{R_{m,t}^2}^{DOWN}+{\epsilon }_t$(9)

Here lower than 0 market return indicates downmarket or upmarket otherwise. Negative and significant ${\delta }_2$ will indicate herding behavior.

Now the question may arise on what happens at the time of extreme up (bull) and extreme down (bear) market condition, to answer it following models are used:

(10) ${CSAD}_t^{UP}=\alpha +{\delta }_1^{UP}\left|R_{m,t}^{UP}\right|\ast D_t^U+{\delta }_2^{UP}{R}_{m,t}^{2}UP\ast D_t^U+{\epsilon }_t$(10)

(11) ${CSAD}_t^{DOWN}=\alpha +{\delta }_1^{DOWN}\left|R_{m,t}^{DOWN}\right|\ast D_t^L+{\delta }_2^{DOWN}{R_{m,t}^2}^{DOWN}\ast D_t^L+{\epsilon }_t$(11)

Where $D_t^U$ takes the value of 1 when market return falls in the upper tail of the distributions of returns or otherwise zero. Similarly, $D_t^L$ takes the value of 1 when market return falls in the lower tail of the distributions of returns or otherwise zero. Both the tails are defined at .99, .95, and .66 levels of confidence.

3. Empirical results

The descriptive statistics of cross-sectional standard deviation (CSSD) and cross-sectional absolute deviation (CSAD) and the market return (MR) are presented in Table 1. It displays that mean of CSSD, CSAD and MR are 3.44 percent, 1.14 percent, and 1.16 percent, respectively. The standard deviations of CSSD, CSAD, and MR are 3.91 percent, 0.75, and 49.59 percent respectively. All the series are positively skewed and kurtosis is greater than 3. The study employs OLS regression in CSSD and CSAD models assuming the normality tendency of a large sample size. Only the model which can detect herding will be further checked for robustness through the GARCH (1,1) method assuming nonlinearity since the null hypothesis of normality is rejected due to significant Jarque-Bera statistics in Table 1. The ADF test statistics of—28.82,—23.76, and −48.38 for CSSD, CSAD, and MR, respectively, confirm the stationary at a 1 percent level.

Table 1. Descriptive Statistics

	CSSD	CSAD	MR
Mean	0.034488	0.011441	0.011629
Median	0.027367	0.009893	−0.01388
Maximum	0.845105	0.148834	4.681158
Minimum	0.008912	0.003492	−3.61779
Std. Dev.	0.039128	0.00754	0.495942
Skewness	11.71488	7.972988	0.548762
Kurtosis	188.668	111.5907	11.20622
Jarque-Bera	4,561,559	1,569,021	8928.192
Probability	0	0	0
ADF statistics	−28.8211	−23.7671	−48.3843
Sum	107.8101	35.7638	36.35196
Sum Sq. Dev.	4.784391	0.177662	768.6216
Observations	3126	3126	3126
Test ADF critical values: 1% level			−3.96098
5% level			−3.41124
10% level			−3.12746

Firstly the study approaches with a comparative investigation of cross-sectional standard deviation (CSSD) and cross-sectional absolute deviation (CSAD) model to detect herding in the banking sector in Bangladesh. Table 2 presents the results of the OLS regression of the CSSD and CSAD model at .99, .95, and .66 confidence levels of upper and lower tails of the market. Even though the CSSD and CSAD models have significant ${\beta }_1$ and ${\beta }_2$, they cannot detect the herding due to positive coefficients. Since positive values of ${\beta }_1$ and ${\beta }_2$ imply stock return dispersion becomes larger with the rise in market return, and vice-versa; and their relation is linear, it reflects the CAPM assumptions of rational pricing theory. This result is consistent with Ahsan and Sarkar (2013), Sharma (2018), Huang et al. (2015).

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

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

Table 2. Regression results for CSSD and CSAD model

	CSSD			CSAD
	.99	.95	.66	.99	.95	.66
$\alpha$	0.0258*	0.0252*	0.0252*	0.008617*	0.008562*	0.008618*
${\beta }_1$	0.0072*	0.0076*	0.0075**	0.002158*	0.002143*	0.002068*
${\beta }_2$	0.0119*	0.0123*	0.0120*	0.00413*	0.004127*	0.003944*
F statistic	11.639	10.247	7.916	43.54	38.704	31.403
LM Test	1682.68*	1689.31*	1698.77*	1795.96*	1813.472*	1833.438*
ARCH	581.69*	582.01*	582.89*	539.0618*	539.696*	541.672*

*, **, and *** denote the null hypothesis is rejected at one percent, five percent, and ten percent level, respectively.

Now the study approaches towards the basic model of CSAD of Chang et al. (2000), which assumes stock return dispersion to nonlinearly increase or decrease with the change in market return and is regarded as more appropriate than the method of Christie and Huang (1995). Again this study incorporates dummies to represent global financial crisis (GFC) and domestic financial crisis (DFC). The models with GFC and DFC run just after the basic model of CSAD. In table 3, the study presents the results of three models of CSAD, one is the basic model of Chang et al. (20,000), other models have GFC and DFC dummy respectively. All these model studies the whole sample period and report the significant negative value of ${\delta }_2$, which indicates investors follow the market, flock together and herding exists. The insignificant dummy for GFC expresses the insusceptibility of investors’ behavior in the banking sector in Bangladesh to the international market. On the other hand, a dummy for DFC reports the existence of herding during a domestic financial crisis.

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

Table 3. Regression results for CSAD

	Basic	GFC	DFC
$\alpha$	0.00786* (41.644)	0.0078* (41.193)	0.0079* (42.032)
${\delta }_1$	0.3541* (19.348)	0.3501* (19.109)	0.3700* (19.954)
${\delta }_2$	−1.2308* (−4.741)	−1.2082* (−4.658)	−1.2633* (−4.881)
DV1		0.00152* (3.096)
DV2			−0.0017* (−4.75038)
F-statistic	437.993	295.995	301.533
Adjusted R-squared	0.2185	0.2206	0.2239
LM Test	1331.317*	1328.152*	1272.954*
ARCH	583.55*	569.4171*	581.7628*

*, **, and *** denote the null hypothesis is rejected at one percent, five percent, and ten percent level, respectively.

The existence of herding for the whole period is further examined through the GARCH (1, 1) model and the estimated results are presented in Table 4. It is noticeable that ${\delta }_2$ is significant and negative. Thus, herding persists in the GARCH (1, 1) model which confirms the robustness of the findings in the OLS regression.

${CSAD}_t=\alpha +{\delta }_1\left|R_{m,t}\right|+{\delta }_2{R}_{m,t}^2+{\delta }_{3}D{V}_1+{\epsilon }_t$

${CSAD}_t=\alpha +{\delta }_1\left|R_{m,t}\right|+{\delta }_2{R}_{m,t}^2+{\delta }_{3}D{V}_1+{\epsilon }_t$

Table 4. GARCH (1, 1) results

	Basic
$\alpha$	0.0073* (149.52)
${\delta }_1$	0.2857* (49.751)
${\delta }_2$	−0.9203* (−8.869)
${\gamma }_1$	0.8734* (401.373)
${\phi }_1$	0.1265* (58.173)
Adjusted R-squared	0.18664

*, **, and *** denote the null hypothesis is rejected at one percent, five percent, and ten percent level, respectively.

After detecting the herding for the entire study period, this study divides the study period in terms of upward market and downward market. This division is initiated to detect any asymmetry in the herding behavior that persists if the market condition changes. Table 5 presents the summary statistics of the upward and downward markets. This table shows division of market return into two parts based on its sign that is positive market return means upward market movement and the negative market return is downward market movement. Out of 3126 observations, 1632 represent upmarket, and the rest 1494 indicate the downmarket. The mean of CSAD at the upward market and downward market are 1.22 percent and 1.07 percent, respectively. Both CSAD series is positively skewed and kurtosis is greater than 3. The mean market return of the upmarket is 1.2 percent whereas the mean return of the downmarket is −1.02 percent. However, their standard deviations lie between 1.10 percent to 1.37 percent. The Jarque-Bera statistics and ADF test statistics infers the same conclusion of non-normality and stationary time series, respectively.

Table 5. Descriptive statistics of upmarket and downmarket condition

	CSAD^UP	CSAD^DOWN	Market Return^UP	Market Return^DOWN
Mean	0.012245	0.010705	0.012006	−0.01022
Median	0.010596	0.009395	0.007844	−0.00711
Maximum	0.118572	0.148834	0.156039	−7.71E-07
Minimum	0.0035	0.003492	2.43E-07	−0.12059
Std. Dev.	0.007414	0.007581	0.013743	0.011029
Skewness	5.38757	10.43076	2.841268	−3.20535
Kurtosis	57.46576	163.1287	17.18818	20.44302
Jarque-Bera	191,893.3	1,773,195	14,541.33	23,484.22
Probability	0	0	0	0
ADF statistics	−15.541	−5.618	−10.7144	−6.10019
Sum	18.29373	17.47007	17.93682	−16.6771
Sum Sq. Dev.	0.082074	0.093739	0.281971	0.198408
Observations	1494	1632	1494	1632
Test ADF critical values: 1% level			−3.9642
5% level			−3.4128
10% level			−3.1283

Table 6 presents the herding behavior at the upward market and downward market conditions. The upmarket confirms herding behavior through the significant negative value of ${\delta }_2$. While the downmarket is unable to pronounce any herding behavior. These findings are in line with the papers like Cakan and Balagyozyan (2014), Huang et al. (2015), Medhioub and Chaffai (2019).

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

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

Table 6. Regression results for Upmarket and Downmarket

	Upmarket	Downmarket
$\alpha$	0.008207* (32.357)	0.00779* (27.667)
${\delta }_1$	0.3888* (17.158)	0.28235* (9.3806)
${\delta }_2$	−1.8961* (−6.208)	0.10865 (0.2346)
F-statistic	273.130	174.079
Adjusted R-squared	0.2671	0.1750
LM Test	690.239*	320.269*
ARCH	350.951*	87.512*

*, **, and *** denote the null hypothesis is rejected at one percent, five percent, and ten percent level, respectively.

As herding behavior is mostly observed in the rising market in the banking sector in Bangladesh, this paper further investigates herding behavior incorporating both models of Christie and Huang (1995) and Chang et al. (2000) so that herding behavior can be evaluated in the context of extreme upmarket (bull market) and extreme downmarket (bear market). Table 7 reports the results for an extreme upmarket and an extreme downmarket. In different criteria, the bull market persistently reports an inverse connection between return deviation and market return by significant negative values of ${\delta }_2$. On the other hand, the bear market presents a significant positive value of ${\delta }_2$ that implies return deviation moves with the changes of the market return. This positive relation indicates reverse herding or no herding in the bear market. Investors tend to escape from their flock as it appears to be a falling market. As result, this study confirms the asymmetry in herding behavior which is consistent with the findings of earlier literature like Chaffai and Medhioub (2018), Shrotryia and Kalra (2019).

${CSAD}_t^{UP}=\alpha +{\delta }_1^{UP}\left|R_{m,t}^{UP}\right|\ast D_t^U+{\delta }_2^{UP}{R}_{m,t}^{2}UP\ast D_t^U+{\epsilon }_t$

${CSAD}_t^{DOWN}=\alpha +{\delta }_1^{DOWN}\left|R_{m,t}^{DOWN}\right|\ast D_t^L+{\delta }_2^{DOWN}{R_{m,t}^2}^{DOWN}\ast D_t^L+{\epsilon }_t$

Table 7. Regression results for extreme up and extreme down market

	Bull market			Bear market
	.99	.95	.66	.99	.95	.66
$\alpha$	0.00961 (56.488)	0.010109 (57.6732)	0.010008 (59.139)	0.01034* (41.245)	0.010338 (41.01594)	0.010316 (40.548)
${\delta }_1$	0.480376 (25.235)	0.307199 (28.542)	0.304688 (28.341)	−0.14481** (−2.418)	−0.1386** (−2.350)	−0.12762** (−2.221)
${\delta }_2$	−2.86247* (−10.406)	−1.2001* (−2.0399)	−0.93675** (−1.628)	3.661183* (14.660)	3.662799* (14.653)	3.668147* (14.6414)
F-statistic	510.386	418.306	412.537	127.298	127.017	126.617
Adjusted R-squared	0.40559	0.3585	0.3537	0.1341	0.1338	0.1334
LM Test	503.352*	467.254*	478.367*	342.897*	343.512*	343.692*
ARCH	311.468*	361.696*	363.393*	92.953*	92.939*	92.919*

*, **, and *** denote the null hypothesis is rejected at one percent, five percent, and ten percent level, respectively.

4. Conclusion

Primarily the study focuses on the detection of herding in the banking sector in Bangladesh; it extends its analysis to explain herd behavior in market asymmetries. By utilizing the daily data from January 2007 to December 2018, the study reveals the existence of herding in the banking sector and confirms the herding nature of investors during the domestic financial crisis but an absence of herding during the Global financial crisis. By dividing the study period into upmarket and downmarket contexts, this paper finds herding in the banking sector is common in a rising market but blurred in a falling market. This phenomenon is further confirmed when the extreme market condition of the bull and bear phase is used. At the bear phase investors expose anti-herd behavior to avoid the market catastrophe whereas they tend to mimic others overlooking any private information to gain from the market momentum.

This study affirms the nonlinearity in herding behavior as a quadratic term in the model of Chang et al. (2000) is negative and significant. This indicates the incremental changes in the level of herding are less than proportional to the average market return. Eventhough herding persists in the banking sector of Bangladesh and there is connection between level of herding and average market return, but market return is inconclusive to define the exact level of herding because the same amount of market return may yield varying levels of herding. This study may turn out a significant insight to the regulatory agency in explaining investors’ behavior during domestic and international stock market crises. The asymmetry and nonlinearity in herding behavior in the banking sector have both social and financial implications that investors and regulatory bodies should perform with caution as the stocks are mispriced and market participants have varying behavior thereby banking sector would pronounce surprise now and then.

The study could have been further improved considering macroeconomic variables and fundamentals, volatility, and sentiments along with more industries and expanded periods in the domestic and foreign stock market.

Acknowledgements

The author is very thankful for the comments of the Editor and the anonymous reviewers which significantly improved this work.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This research did not receive any specific grant from funding agencies.