You are on page 1of 48

Emerging Markets Finance and Trade

ISSN: 1540-496X (Print) 1558-0938 (Online) Journal homepage: http://www.tandfonline.com/loi/mree20

Investors’ Fear and Herding Behavior: Evidence


from the Taiwan Stock Market

Teng-Ching Huang & Kuei-Yuan Wang

To cite this article: Teng-Ching Huang & Kuei-Yuan Wang (2017): Investors’ Fear and Herding
Behavior: Evidence from the Taiwan Stock Market, Emerging Markets Finance and Trade, DOI:
10.1080/1540496X.2016.1258357

To link to this article: http://dx.doi.org/10.1080/1540496X.2016.1258357

Accepted author version posted online: 08


Mar 2017.

Submit your article to this journal

Article views: 10

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at


http://www.tandfonline.com/action/journalInformation?journalCode=mree20

Download by: [University of Newcastle, Australia] Date: 12 March 2017, At: 20:36
Investors’ Fear and Herding Behavior: Evidence from the Taiwan

Stock Market

t
Teng-Ching Huang*

ip
cr
us
Kuei-Yuan Wang**

Abstract
an
This paper analyzes the influence of investors’ fear on their investment behavior in the Taiwan
M
stock market. This study used the volatility index (VIX) as a barometer of investors’ fear. Our

results show that herding behavior increases with the VIX; that is, herding behavior is
ed

encouraged by an increase in investors’ fear. Moreover, Our results demonstrate that investors

react more quickly to bad news than to good news when their fear increases, supporting the
pt

hypothesis of the presence of an asymmetric reaction to news. However, investors react more
ce

*
Teng-Ching Huang (tengshow10@gmail.com) is an Assistant Professor in the Department of Finance at Asia

University. Kuei-Yuan Wang (gueei5217@gmail.com) is an Associate Professor in the Department of Finance at


Ac

Asia University. He is also a consult of Department of Medical Research at China Medical University Hospital at

China Medical University. This study was supported by Asia University under Grant No. 103-asia-02. We

acknowledge the helpful comments and suggestions from the anonymous reviewers.

**
Corresponding Author: Kuei-Yuan Wang

Affiliation: Department of Finance, Asia University

Address: 500, Lioufeng Rd., Wufeng, Taichung 41354, Taiwan

E-mail: gueei5217@gmail.com

1
quickly to good news when their fear decreases, indicating an inverse asymmetric reaction. In

addition, our empirical results reveal that herding behavior tends to exist on days with a large

trading volume.

Keywords: herding behavior; VIX; investor’s fear; return dispersion; financial crisis

t
ip
1. Introduction

cr
Academicians and practitioners have been interested in understanding the investment

behavior of market participants for the purposes of portfolio management and financial asset and

us
risk management. In perfectly efficient markets, obtaining new information does not incur costs.
an
The new information should be quickly conveyed into asset prices, and investors should

accurately evaluate financial assets on the basis of rational expectations. However, in the real
M
world, acquiring information is costly, and some investment decisions can be made under the

framework of investors’ irrational expectations. Some investors even discard the information
ed

they hold and adopt other investors’ decisions. This phenomenon is called a type of herding

behavior.
pt

Herding behavior can cause the prices of financial assets to deviate from their efficient
ce

prices. Recently, an increasing body of literature1 studies this relevant field to understand

investor behavior. Christie and Huang (1995) proposed a cross-sectional standard deviation
Ac

(CSSD) to be a measure of return dispersion. They use the CSSD in their study of herding

behavior in the US stock market and argue that herding behavior exists in equity markets when

1
Lakonishok, Shleifer, and Vishny (1992), Christie and Huang (1995), Nofsinger and Sias (1999), Chang, Cheng,

and Khorana (2000), Sias (2004), Hwang and Salmon (2004), Gleason, Mathur, and Peterson (2004), Chang and

Dong (2006), Blasco and Ferreruela (2008), Choi and Sias (2009), Dasgupta, Prat, and Verardo (2011a, 2011b),

Chiang et al. (2013), Huang, Lin, and Yang (2015).

2
return dispersion decreases during the period of market stress. Chang, Cheng, and Khorana (2000)

presented cross-sectional absolute deviation of return (CSAD), an alternative method based on

the CSSD. They considered the entire distribution of market returns, and they examined the

relationship between return dispersion and market return in three developed markets and two

emerging markets by using a nonlinear method. Although Christie and Huang (1995) and Chang,

Cheng, and Khorana (2000) also reported an association of their methods with firm

t
ip
characteristics, such as industry and firm size, they do not take into account the factor of risk in

their empirical evidence.

cr
According to Chang, Cheng, and Khorana (2000), the investment behavior of market

us
participants can be linked to the degree of underlying market volatility. Subsequently, some

researchers investigated investors’ behavior based on this point of view. On the basis of the
an
CSSD of Christie and Huang (1995), Hwang and Salmon (2004) utilized the cross-sectional

dispersion of beta and observe that investors herd toward the market index. Furthermore, they
M

applied this model to US and Korean equity markets and observed this herding behavior in both
ed

markets. Chiang et al. (2013) examined the herding behavior of investors in the Pacific-Basin

equity market. They reported that the herding behavior of investors exists in the up and down
pt

markets and that the degree of herding behavior is negatively correlated with market volatility.

Huang, Lin, and Yang (2015) analyzed the impact of idiosyncratic volatility on the investment
ce

behavior of market participants and reported that investors exhibited more active herding
Ac

behavior when their investment portfolios have higher idiosyncratic volatility.

In this study, we contribute to the literature by investigating the relationship between

investors’ risk perception and their investment behavior. In highly volatile financial markets,

investors’ risk perception would drastically affect their investment decisions, which are further

reflected in asset prices. The Chicago Board Options Exchange (CBOE) volatility index (VIX),

constructed in 1993, is a measure of implied volatility estimated from S&P 100 option prices.

3
Whaley (2000) defined the VIX as a gauge of investors’ fear2. Empirical evidence derived from

measuring market risk perception reveals that the VIX is arguably the most favorable volatility

measure.

On the basis of the information asymmetry hypothesis and the assumption that investors are

risk averse, a remarkable increase in the VIX might represent a substantial increase in the risk.

This anomaly might result in risk-averse investors with less information becoming nervous,

t
ip
consequently prompting them to exhibit irrational behavior. Because of less information, they

might follow others to make investment decisions. In such cases, herding behavior would be

cr
more prominent. However, some studies (e.g., Avramov, Chordia, and Goyal, 2006; Venezia,

us
Nashikkar, and Shapira, 2011; Natividad, Pilar, and Sandra, 2012) demonstrate that herding

behavior is positively correlated with market volatility. Venezia, Nashikkar, and Shapira (2011)
an
examined herding behavior in professional and amateur investors in a large brokerage house in

Israel. They reported that herding behavior is persistent in both groups and that it is positively
M

and significantly related to stock market volatility.


ed

In this study, we used the VIX as a barometer of investors’ fear, and on the basis of the

CSSD (Christie and Huang 1995) and CSAD (Chang, Cheng, and Khorana 2000), we
pt

investigated the impact of changes in the VIX on the investment behavior of market participants

in the Taiwan stock market. The evidence of herding behavior is stronger in emerging markets
ce

than in mature markets because more speculators having relatively short investment horizons are
Ac

present in emerging markets (Chang, Cheng, and Khorana, 2000). Among the emerging markets,

the Taiwan stock market is very important, and a majority of its market participants are

2
An increase in the VIX represents investors’ anticipation that volatility in the future might be more dramatic. This

anomaly might induce fear in the investors. By contrast, a decrease in the VIX represents investors’ anticipation that

volatility in the future might be more stable, and they might be in the exuberance state. Therefore, the VIX reflects

the investors’ viewpoint and state of mind and can often be used as an indicator of investors’ fear.

4
individual traders. Individual investors are commonly considered to possess speculative

characteristic with relatively short investment horizons3; this thus provides an opportunity to

examine herding behavior. In addition, compared with other emerging markets4, the Taiwan

stock exchange (TSE) regulates a larger daily price limit in listing stocks, affecting the speed at

which stock prices are reflected in news. This may result in informational inefficiency, which can

impact the investment strategies of market participants.

t
ip
The rest of this paper is organized as follows. Section 2 describes the data, and Section 3

introduces the methodology. Section 4 reports the descriptive statistics and empirical results. The

cr
concluding remarks are provided in Section 5.

2. Data Description
us
an
We collected stocks traded on the Taiwan Stock Exchange (TSE) from January 1, 2007, to
M
December 31, 2014, a period of 1989 trading days. The data included individual stocks and index

levels from the Taiwan Economic Journal (TEJ) database, which provides daily stock prices,
ed

daily index values, and daily VIX.

The VIX is estimated by a weighted average of volatilities implied from four call and put
pt

options on the S & P 100, which are the closest to be at-the-money. This provides an estimate of
ce

3
The average monthly turnover rate of the Taiwan Stock Exchange (TSE) from 2007 to 2014 is 15.1%, and it

ranged from a low of 2012 (9.5%) to a high of 2007 (22.1%). These turnover rates are substantially higher than
Ac

those among the emerging markets, providing indirect evidence on the existence of relatively short investment

horizons of market participants in the TSE. In addition, Bekaert and Harvey (1997) reported that during their sample

period, Taiwan ranked second in terms of the monthly turnover rate among 20 emerging equity markets.

4
The daily price limit in listing stocks is 7% for the Taiwan stock market, which is a larger constraint in stock price

fluctuation than that in other emerging markets. For example, the price limits in listing stocks are 15%, 10%, and

30% for Korea, Thailand, and Malaysia, respectively.

5
expected stock market volatility for the subsequent 30 calendar days (approximately 22 trading

days). In 2003, the CBOE proposed a modified methodology for calculating the VIX from the

prices of S & P 500 index options in a wide range of strike prices; this methodology is

independent of any option pricing model. Because the new VIX measure contains the entire

strike price range of implied volatilities, it is commonly considered a more favorable metric of

market expectations. The TEJ provides, starting in December 2006, the new VIX calculated

t
ip
using the new VIX measure of the CBOE. Thus, we used the new VIX in our empirical evidence.

The subprime mortgage crisis that occurred in the United States in July 2007 subsequently

cr
caused a large turmoil in global financial markets. This financial turmoil has a large influence on

us
asset prices. To understand whether this turmoil increases investors’ fear and whether it

corresponds to their investment behavior, we considered the period from 2007 to 2009 as the
an
global financial crisis (GFC) period and from 2010 to 2014 as the non-global financial crisis

(NGFC) period. We defined the GFC period on the basis of the results of Ben-David, Franzoni,
M

and Moussawi (2012) and Flannery, Kwan, and Nimalendran (2013).


ed

3. Methodology

3.1. CSSD Method


pt

According to the method of Christie and Huang (1995), individual stock returns converge
ce

toward the market return in presence of herding behavior. This reduces the dispersion of stock

returns from market returns. In this study, we used the cross-sectional standard deviation (CSSD)
Ac

method of Christie and Huang (1995) to evaluate the dispersion of returns, presented as follows:

2
1 N
CSSD =  (Ri ,t − Rm,t )
N − 1 i =1
(1)

where Ri ,t is the observed return of an individual stock i on day t , and Rm , t is the

cross-sectional average of N stock returns in the portfolio on day t . This model,

adopted from the study of Christie and Huang (1995), indicated that herding behavior will

6
be more widespread during the period of market stress. Thus, they regressed the CSSD

method against an intercept and two dummies to verify extreme market conditions. The

details of the regression equation are given as follows:

L
CSSD t = + D tL + U
D tU + t (2)

where, DtL = 1 , if the market return on day t lies in the extreme lower tail of the return

t
distribution, and zero, otherwise; moreover, DtU = 1 , if the market return on day t lies in

ip
the extreme upper tail of the return distribution, and zero, otherwise.

cr
In the presence of herding behavior, CSSD t is lower during the period of market stress,

us
indicating that individual stock returns would converge to the market return. Thus, significantly

L U
negative estimates for and suggest the presence of herding behavior.
an
3.2. CSAD Method
M

Chang, Cheng, and Khorana (2000) proposed an alternative measure of return dispersion,

namely the cross-sectional absolute deviation of returns (CSAD), shown as follows:


ed

1

N
CSAD t = i =1
R i ,t − R m ,t (3)
N
pt

where Ri ,t denotes the observed return of an individual stock i on day t and Rm , t denotes
ce

the cross-sectional average of N stock returns in the portfolio on day t . They argue that the

CSAD and market returns should have a linear relationship under the CAPM context. Thus, the
Ac

presence of herding behavior in the period of market stress indicates a non-linear relationship.

Chang, Cheng, and Khorana (2000) use a model to examine the non-linear relationship, shown as

follows:

CSAD t = + 1 R m ,t + 2 R m2 , t + t (4)

If herding behavior exists, then 2 would be significantly negative. This implies that the

7
deviation of stock returns from market returns would decrease over the period of stress.

Specifically, as reported by Gleason, Mathur, and Peterson (2004), the non-linear component can

also be acquired for CSSD if herding behavior is present during the period of market stress. Thus,

on the basis of the findings of Gleason, Mathur, and Peterson (2004), we adopt two additional

regression models by swapping the dependent variables in Equations (2) and (4) to acquire a

supplementary test, which is shown as follows:

t
ip
L
CSAD t = + D tL + U
D tU + t (5)

cr
CSSD t = + 1 R m ,t + 2 R m2 , t + t (6)

Because the time series of the CSSD and CSAD appear to be highly auto-correlated, we

us
adjusted the standard error of estimated regression coefficients of heteroskedasticity and
an
auto-correlation by using the method of Newey and West (1987) and added a 1-day lag of

dependent variables as a regressor in our equations. Thus, we investigated herding behavior by


M
using the following equation:

L
Vari t = + DtL + U
DtU + Vari t −1 + t (7)
ed

Vari t = + 1 Rm ,t + 2 R m2 , t + 3Vari t −1 + t (8)


pt

In Equations (7) and (8), Vari denoted the dependent variable in the regression model and

subscript i denotes the CSSD or CSAD.


ce
Ac

3.3. Asymmetric Reactions to News

Chang, Cheng, and Khorana (2000) reported that the market reaction to both good and bad

news is asymmetric in three developed markets, namely the United States, Hong Kong, and

Japan. They divided these samples into two categories: up and down. The up market is regarded

as trading days with positive market returns, and the down market is regarded as trading days

with negative market returns. In their empirical evidence, the CSAD increases with R m , t , and

8
the rate of increase in the down market is lower than that in the up market, supporting the

presence of an asymmetric reaction to good and bad macroeconomic news. In addition, investors

are more fearful in the down market than in the up market (Kahneman and Tversky, 1979;

McQueen et al., 1996; Low, 2004). Hence, they are more likely to exhibit herding behavior

during these periods. We investigated this proposition of an asymmetric reaction in the up versus

down markets by using the following equations:

t
ip
Vari tUP = + UP
1 R mUP, t + UP
2 ( R mUP, t ) 2 + UP
3 Vari tUP
−1 + t (9)

cr
Vari tDOWN = + 1
DOWN
RmDOWN
,t + DOWN
2 ( RmDOWN
,t )2 + DOWN
3 Vari tDOWN
−1 + t (10)

us
where, rmUP, t and rmDOWN
,t denote the absolute values of market returns when the markets

are up and down, respectively, and ( rmUP, t ) 2 and ( rmDOWN


,t ) 2 denote the respective square
an
values. Vari is the dependent variable in the models, and subscript i denotes the CSSD
M
or CSAD.

UP DOWN
If 1 is higher than 1 , the rate of increase in the dispersion is higher in up markets
ed

than in down markets, indicating the presence of an asymmetric reaction to good and bad news.

This conjecture is consistent with the finding of directional asymmetry reported by McQueen et
pt

al. (1996), which indicated that all stocks tend to respond quickly to negative macroeconomic
ce

news, but that small stocks tend to show a delayed response to positive macroeconomic news.

4. Empirical Results
Ac

4.1. Descriptive Statistics

Table 1 presents descriptive statistics of the daily return dispersions of the CSSD and

CSAD and the daily VIX for the full sample period and eight sub-periods. We used only the daily

return to compute the CSSD and CSAD for all our empirical data. The estimated mean value of

the CSSD is 0.021 in the full sample period, with a high of 0.026 in 2008 and a low of 0.017 in

2014 in yearly sub-periods. The minimum value of the daily CSSD is 0.012 in all periods,

9
indicating that the magnitude of CSSD estimates is positive across the periods. The estimates of

CSSD are remarkably similar to those estimates of CSAD, signifying that all returns of

individual stocks are not completely dependent on market returns based on both dispersion

measures. In addition, the mean values of the CSSD (CSAD) in 2007, 2008, and 2009 are 0.024

(0.017), 0.026 (0.019), and 0.024 (0.018), respectively, which are higher than the mean values of

the CSSD (CSAD) in the whole sample period and other years. This finding indicates an increase

t
ip
in the deviation of stock returns from market returns during the financial crisis in 2007–2009,

which may reflect various investment behaviors among traders during the crisis.

cr
Regarding the VIX, the mean and corresponding standard deviation are 22.04 and 9.31 in

us
the full sample period, respectively. A higher VIX value represents a higher degree of investors’

fear. We observe higher VIX values in 2007–2009 than in other years, with the average VIX
an
values reaching 23.49, 34.39, and 34.43 in 2007, 2008, and 2009, respectively, and then

gradually decreasing until the lowest level in 2013–2014; these results indicate that investors’
M

degree of fear is higher in the financial turmoil period.


ed

Concerning the auto-correlation coefficients of the CSSD, CSAD, and VIX, we observed

that the time series of the CSSD and CSAD in the full sample period appeared to be highly
pt

auto-correlated. For example, the first-order auto-correlation of the CSSD (CSAD) is 0.792

(0.796). Hence, we adjusted the standard error of the estimated regression coefficients of
ce

heteroskedasticity and auto-correlation by using the method of Newey and West (1987).
Ac

Moreover, the unit root test results of the Augmented Dickey–Fuller test demonstrated that the

CSSD, CSAD, and VIX series are stationary in the sample periods. In addition, we used Pearson

correlation analysis to examine the correlation of the CSSD with the CSAD as well as with the

VIX in the full sample period. We determine that the CSSD (CSAD) is positively related to the

VIX5.

5
Here, we did not report these results; however, they are available upon request from the author.

10
Figure 1 presents the plot of the VIX for the full sample period. The VIX value

significantly increases from the middle of 2007 and remained at a high level until the end of

2009. In general, VIX values are higher in 2007–2009, particularly at the end of 2008, than in

other yearly periods. This finding indicates that the global financial turmoil caused by the

subprime mortgage crisis in July 2007 affected VIX values; this finding is consistent with results

listed in Table 1.

t
ip
4.2. Effect of Changes in Investors’ Fear on Herding Behavior

cr
Table 2 presents the estimation results of Equation (7) for the full sample period under

us
small and large changes in VIX values. The group with large changes includes days with

unusually high positive and negative changes (10%) in VIX values, and the group with small
an
L U
changes includes other days. With regard to small changes in VIX values, the and
M
coefficients of the CSSD and CSAD measures are not significantly negative based on 5% and

L U
10% criterion levels. For example, the estimate of ( ) for the CSSD is −0.0001 (−0.0004)
ed

with a t-statistic of −0.36 (−1.51) under the 5% criterion level.

U
Regarding the large changes in VIX values, the coefficients of the CSSD and CSAD
pt

measures are negative based on 5% and 10% criterion levels; however, they do not reach the
ce

conventional significance levels. These findings indicate that two return dispersions of the CSSD

and CSAD do not significantly decrease over the period of extreme market movements. This
Ac

finding indicates that herding behavior is generally not observed among Taiwanese stock

investors when the market has abnormally large price movements.

The estimation results of Equation (8) under positive and negative changes in VIX values

are presented in Panel A of Table 3. Concerning positive changes in VIX values, the 2

coefficient of the CSSD (CSAD) is −5.004 (−5.438) with a t-statistic of −6.57 (−7.27) during the

full sample period, which is significantly less than zero. This finding does not support the linear

11
model, suggesting that herding behavior occurs when investors’ fear increases. Furthermore, in

the eight sub-periods, excluding 2012–2014, significantly negative 2 coefficients provide

evidence of the presence of a nonlinear model.

With respect to negative changes in VIX values, the 2 coefficients of the CSSD and

CSAD are −3.868 and −4.198, respectively, in the full sample period, and both these coefficients

are significantly less than zero at the level of 1%. This finding indicates that herding behavior

t
ip
occurs over the whole sample when investors’ fear decreases. In addition, on the basis of the

cr
results in eight sub-periods, we observe that the 2 coefficients of the CSAD are significantly

negative in 2007–2009, supporting the existence of herding behavior. In general, these results

us
indicate that herding behavior exists in the full sample period and most sub-periods based on the
an
increase in investors’ fear, whereas herding behavior occurs in the full sample period and

sub-periods of 2007, 2008, and 2009 based on the decrease in investors’ fear. This finding
M
provides evidence that herding behavior is encouraged by an increase in investors’ fear and that it

decreases after the financial crisis in 2007–2009.


ed

To examine the impact of the global financial crisis on herding behavior, the period of

2007–2009 is regarded as the GFC period and the period of 2010–2014 is regarded as the NGFC
pt

period. Moreover, during the period of market stress, a high trading volume is often accompanied
ce

by high volatility of stock prices. Hence, we evaluate herding behavior by repeating the

regression of Equation (8) under small and large trading volumes and present the results in Panel
Ac

B of Table 3.

In the small trading volume, the 2 coefficient of the CSSD (CSAD) is −3.029 (−3.857)

with a t-statistic of −3.61 (−4.56) in the full sample period and −3.442 (-4.627) with a t-statistic

of −3.41 (−4.73) in the GFC period, which are significantly less than zero. However, in the

NGFC period, the 2 coefficients of the CSSD and CSAD are −0.919 and −0.889, respectively,

and both these coefficients are not statistically significantly less than zero at conventional

12
significance levels. This result indicates that herding behavior exists in the full sample and GFC

periods under the small trading volume. In the large trading volume, except for the CSSD in the

GFC period, the 2 coefficients of the CSSD and CSAD are significantly less than zero across

the periods, supporting the presence of herding behavior under a large trading volume.

4.3. Regression Results within the Quintiles of Changes in the VIX

t
ip
To investigate the influence of different degrees of changes in the VIX on herding behavior,

cr
we again repeat the regression of Equation (8) for the full sample period. The sample is

categorized into three quintiles for positive and negative changes in the VIX. Group 1 denotes

us
the smallest changes in the VIX and Group 3 denotes the largest changes in the VIX. The

estimation results are presented in Table 4. Panels A and B present the results of positive and
an
negative changes in the VIX, respectively.
M
In Group 1 of Panel A in Table 4, the 2 coefficient of the CSSD (CSAD) is −7.569

(−7.273), with a t-statistic of −5.17 (−4.55) in the hypothesis of 2 = 0 , which significantly less
ed

than zero. This finding indicates the presence of herding behavior when Group 1 does not follow

the linear model. In addition, we observe similar results in Group 2 and Group 3. These results
pt

support the notion that herding behavior occurs under various levels of positive changes in the
ce

VIX.

In Panel B, the 2 coefficient is significantly negative only in Group 1. Furthermore, in


Ac

Group 1, the 2 coefficient of the CSSD is −3.592, with a t-statistic of −3.34, whereas the 2

coefficient of the CSAD is −4.281, with a t-statistic of −5.05. However, in Groups 2 and 3, the

2 coefficients of the CSSD and CSAD measures are less than zero; however, these coefficients

do not reach conventional significance levels. This finding indicates that herding behavior exists

only in the group with the smallest negative changes in the VIX; that is, herding behavior does

not occur when investors’ fear decreases.

13
The aforementioned results demonstrate that herding behavior exists within various levels

of positive changes in the VIX but is absent when the degree of negative changes in the VIX is

higher, providing supplementary evidence that investors’ fear does affect herding behavior. In

addition, the degree of herding behavior in small changes in the VIX is higher than that in large

changes in the VIX. This means that compared with large changes in the VIX, investors in the

Taiwan stock market are more sensitive to small changes in the VIX and this therefore exerts a

t
ip
higher impact on their investment decisions, presenting a decreasing marginal influence of the

VIX.

cr
us
4.4. Asymmetrical Herding Behavior

Chang, Cheng, and Khorana (2000) argued that, if market responses to good and bad news
an
are asymmetric, the rate of increase in the dispersion is higher in up markets than in down

markets. Furthermore, they report that an asymmetric response is present in three developed
M

markets but absent in two emerging markets. On the basis of the results of Chang, Cheng, and
ed

Khorana (2000), we examine the asymmetric reaction in up and down markets by using the

non-linear model for the full sample period and eight sub-periods. The estimation results of
pt

Equation (9) for the up market and of Equation (10) for the down market are presented in Panel

A of Table 5.
ce

UP
In Panel A of Table 5, in the full sample period, the 1 estimate of the CSSD (CSAD) is
Ac

DOWN
0.201 (0.265) with a t-statistic of 5.48 (7.24), and the 1 estimate of the CSSD (CSAD) is

0.251 (0.291) with a t-statistic of 7.26 (8.68), which are significantly higher than zero. This

strongly indicates that return dispersion increases with Rm . In the eight sub-periods, most cases

exhibit similar results. Regarding the CSAD measure, the rate of increase in the up market is the

highest in 2008 (0.305) and the lowest in 2014 (−0.031), whereas that in the down market is the

highest in 2011 (0.349) and the lowest in 2012 (0.156). In particular, we determine that the

14
UP DOWN
hypothesis of 1 − 1 cannot be rejected at conventional significance levels in most cases.

This finding suggests the absence of an asymmetric reaction in the Taiwan stock market, which

is consistent with the finding of Chang, Cheng, and Khorana (2000). However, this finding is

inconsistent with the directional asymmetry proposed by McQueen et al. (1996) and weakly

supports the theory of loss aversion reported by Kahneman and Tversky (1979).

UP DOWN
Concerning the estimates of the non-linear term of Rm2 , the and coefficients of

t
2 2

ip
the CSSD (CSAD) are −4.563 (−4.928) and −3.512 (−3.817), respectively, in the full sample

cr
period and are significantly less than zero, indicating that herding behavior is observed in the up

UP
and down markets during the whole sample period. In the sub-sample periods, the

us
2

DOWN
coefficients are statistically negative during 2007–2010, whereas the 2 coefficients are
an
statistically negative during 2007–2011. This result indicates that herding behavior decreases

after the financial crisis in 2007–2009, which is consistent with the results in Panel A of Table 3.
M
UP DOWN
Moreover, nonsignificant signs of 2 − 2 in most cases signify that herding behavior does

not significantly differ between the up and down markets.


ed

We examine the asymmetric reaction under small and large trading volumes for the full
pt

sample, GFC, and NGFC periods. The estimation results are provided in Panel B of Table 5. In

UP DOWN
the small trading volume, the and estimates of the CSSD and CSAD are
ce

1 1

significantly higher than zero across the periods, signifying that return dispersion increases with
Ac

UP DOWN
Rm . However, the hypothesis of 1 − 1 cannot be rejected at conventional significance

levels in all periods. This result indicates that an asymmetric reaction is absent in the small

trading volume. Regarding the non-linear term of Rm2 , significantly negative coefficients are

observed for the up and down markets, except for in the NGFC period, indicating that herding

behavior occurs in the full sample and GFC periods, which is similar to the results in Panel B of

Table 3. Furthermore, the degree of herding behavior does not significantly differ between the up

15
and down markets in the small trading volume.

UP DOWN
In the large trading volume, the 1 and 1 estimates of the CSSD and CSAD are

significantly higher than zero across the periods, except for the CSSD in the GFC period,

UP
indicating that return dispersion increases with Rm . Moreover, 1 is significantly lower than

DOWN
1 in the NGFC period, revealing that the increasing rate in the dispersion is higher in the

t
up market than in the down market, suggesting an inverse asymmetric response.

ip
To investigate the impact of changes in investors’ fear on the asymmetric reaction to good

cr
and bad news, we use the Equation (9) for the up market and Equation (10) for the down market

us
under positive (negative) changes in the VIX; Table 6 presents the estimation results. The left

panel of Table 6 presents the results under positive changes in the VIX. The estimates on the
an UP
linear term of Rm are significantly positive for both return dispersions. The 1 coefficient of

DOWN
the CSSD (CSAD) is 0.381 (0.431) with a t-statistic of 7.23 (9.25), and the coefficient
M
1

of the CSSD (CSAD) is 0.251 (0.294) with a t-statistic of 8.75 (11.63); therefore, a significantly
ed

positive relationship exists between return dispersion and Rm . Specifically, the coefficients of

the CSSD and CSAD measures are higher in the up market than in the down market, and the
pt

UP DOWN
hypothesis of 1 − 1 can be rejected at the 5% significance level. The fact that the
ce

increasing rate in the dispersion is significantly higher in the up market than in the down market

supports the presence of an asymmetric response. This implies that when investors’ fear
Ac

increases, they react more rapidly to bad news, verifying prior conjectures on an asymmetric

reaction to news in developed markets (McQueen et al., 1996; Chang, Cheng, and Khorana,

2000). In addition, this finding supports the theory of loss aversion reported by Kahneman and

Tversky (1979).

The right part of Table 6 presents the results obtained under negative changes in the VIX.

Similar to the results under the positive changes in the VIX, coefficients in the absolute term of

16
UP
market returns are significantly positive. However, the 1 coefficients of the CSSD and CSAD

DOWN
measures are lower than the 1 coefficients of both measures, and their differences can be

realized at the 5% significance level. This finding demonstrates that the rate of increase in return

dispersion is lower in the up market than in the down market, signifying that investors react more

quickly to good news than to bad news. This phenomenon is in contrast to the results obtained

under the positive changes in the VIX, a phenomenon termed as an inverse asymmetric reaction

t
ip
in this study. This finding is inconsistent with those of McQueen et al. (1996) and Chang, Cheng,

cr
and Khorana (2000). The inverse asymmetric reaction implies that when investors’ fear decreases,

they would have a faster response to good news than to good news, indicating that the theory of

us
loss aversion in Kahneman and Tversky (1979) would not be valid.
an
5. Conclusions and Remarks

In this study, we investigate the investment behavior of market participants with respect to
M

changes in the VIX in the Taiwan stock market by using two measures of return dispersions and

two methods. By using the method of Christie and Huang (1995), we observe that return
ed

dispersions do not decrease during the periods of extreme market movements, revealing that

stock investors do not generally tend to exhibit herding behavior as the market has abnormally
pt

large price movements. However, by using the non-linear method of Chang, Cheng, and Khorana
ce

(2000), we determine that herding behavior occurs in the full sample period and most

sub-periods under positive changes in the VIX, whereas herding behavior occurs only in the full
Ac

sample period and sub-periods of 2007–2008 under negative changes in the VIX. This result

demonstrates that herding behavior is encouraged by an increase in investors’ fear, and it

decreases after the period of financial turmoil. Moreover, we examine the impact of different

degrees of changes in the VIX on herding behavior, and we find that herding behavior exists

across various levels of positive changes in the VIX but not under higher levels of negative

changes in the VIX. This provides supplementary evidence that herding behavior would decrease

17
with a decrease in investors’ fear.

We demonstrate that if the VIX factor is ignored, no asymmetric reaction occurs, which is

consistent with the finding of Chang, Cheng, and Khorana (2000) derived in emerging markets.

However, an asymmetric reaction to news is observed, and it exhibits distinct patterns under

different classifications of the VIX. Under positive changes in the VIX, investors more quickly

react to bad news than to good news, verifying previous conjectures on an asymmetric reaction

t
ip
to news (McQueen et al., 1996) and supporting the theory of loss aversion reported by

Kahneman and Tversky (1979). Under negative changes in the VIX, investors tend to react more

cr
quickly to good news, indicating an inverse asymmetric reaction, which is inconsistent with the

us
finding of McQueen et al. (1996). Moreover, the notion of an inverse asymmetric reaction

implies that the theory of loss aversion reported by Kahneman and Tversky (1979) would not be
an
valid when investors’ fear decreases.

We also examine herding behavior under different trading volumes. Our empirical results
M

reveal that herding behavior is present in most periods under a large trading volume but is absent
ed

after the global financial crisis period under a small trading volume. This suggests that herding

behavior can be observed on days with a large trading volume. Moreover, we examine the
pt

asymmetric reaction under the different trading volumes and observe little support for the

presence of an asymmetric reaction under the small trading volume. Under the large trading
ce

volume, stock investors react more quickly to good news than to bad news, presenting an inverse
Ac

asymmetric reaction.

In Taiwan stock market, there are some characteristics, such as a numerous individual

investors and a larger daily price limit in listing stocks, which could enhance herding behavior.

The individual investors are commonly regarded to be more speculative and have relatively short

investment horizons. And, a larger daily price limits would influence the speed that stocks prices

incorporate the news, which may cause informational inefficiency, especially during financial

18
crisis. In this study, our empirical results provide an evidence that herding behavior exists in

Taiwan stock market. The stylized fact presents that those individual investors possess the

speculative characteristic. In addition, our empirical evidence shows that those speculative

investors would enhance herding behavior when their fear increases.

In general, our results reveal that an increase in investors’ fear encourages herding behavior.

This may be because the major participants of the Taiwan stock market are individual investors;

t
ip
therefore, they are often risk averse and have less information. They consequently become more

irrational and nervous while making investment decisions and follow others’ decisions, thereby

cr
exhibiting herding behavior.

us
Traditional asset pricing models, for example, Capital Asset Pricing Model (CAPM),

indicate that based on rational expectation, the investors can correctly price financial assets.
an
However, herding behavior would cause asset prices to deviate from their efficient prices. In our

empirical results, the presence of herding behavior implies that many investors in Taiwan stock
M

market have an irrational asset pricing. Moreover, specifically, in Taiwan stock market, when the
ed

investors’ fear increases, the level of herding behavior would raise. This would lead to a larger

misprice assets and increase the difficulty for risk management.


pt

Besides, many management implications also can be obtained from our empirical findings.

Hsieh (2013) indicates that information-driven herding behavior might result in the mispricing of
ce

a stock, which consequently confuses investors. Therefore, investors should remain rational and
Ac

fearless while making investment decisions. In addition, government should undertake protective

mechanisms to make the market more stable and thus enable investors to be fearless.

Accordingly, management can achieve the most crucial goal of maximizing stockholders’ wealth.

Furthermore, the increase and decrease in investors’ fear indicate an opposite asymmetric

response to news. Future studies should focus on developing investment and hedging strategies

associated with investors’ fear.

19
Reference

Avramov, D., T. Chordia and A. Goyal. 2006. “The Impact of Trades on Daily Volatility.” Review

of Financial Studies 19, no 4: 1241-1277.

Bekaert, G. and C.R. Harvey. 1997. “Emerging Equity Market Volatility.” Journal of Financial

Economics 43, no 1: 29-77.

t
Ben-David, I., F. Franzoni and R. Moussawi. 2012. “Hedge Fund Stock Trading in the Financial

ip
Crisis of 2007–2009.” Review of Financial Studies 25, no 1: 1-54.

cr
Blasco, N. and S. Ferreruela. 2008. “Testing Intentional Herding in Familiar Stocks: An

Experiment in an International Context.” Journal of Behavioral Finance 9, no 2: 72-84.

us
Chang, E.C., J.W. Cheng and A. Khorana. 2000. “An Examination of Herd Behavior in Equity
an
Markets: An International Perspective.” Journal of Banking & Finance 24, no 4:

1651-1679.
M
Chang, E.C. and S. Dong. 2006. “Idiosyncratic Volatility, Fundamentals, and Institutional

Herding Evidence from the Japanese Stock Market.” Pacific-Basin Finance Journal 14, no
ed

2: 135-154.

Chiang, T.C., J. Li, L. Tan and E. Nelling. 2013. “Dynamic Herding Behavior in Pacific-Basin
pt

Markets: Evidence and Implications.” Multinational Finance Journal 17, no 3: 165-200.


ce

Choi, N. and R.W. Sias. 2009. “Institutional Industry Herding.” Journal of Financial Economics

94, no 4: 469-491.
Ac

Christie, W.G. and R.D. Huang. 1995. “Following the Pied Piper: Do Individual Returns Herd

around the Market?” Financial Analysts Journal 51, no 3: 31-37.

Dasgupta, A., A. Prat and M. Verardo. 2011. “The Price Impact of Institutional Herding.”

Review of Financial Studies 24, no 1: 892-925.

Dasgupta, A., A. Prat and M. Verardo. 2011. “Institutional Trade Persistence and Long-Term

Equity Returns.” Journal of Finance 66, no 1: 635-653.

20
Flannery, M.J., S.H. Kwan and M. Nimalendran. 2013. “The 2007–2009 Financial Crisis and

Bank Opaqueness.” Journal of Financial Intermediation 22, no 1: 55-84.

Gleason, K.C., I. Mathur and M.A. Peterson. 2004. “Analysis of Intraday Herding Behavior

among the Sector ETFs.” Journal of Empirical Finance 11, no 4: 681-694.

Hsieh, S.F. 2013. “Individual and Institutional Herding and the Impact on Stock Returns:

Evidence from Taiwan Stock Market.” International Review of Financial Analysis 29, no 4:

t
ip
175-188.

Huang, T.C., B.H. Lin and T.H. Yang. 2015. “Herd Behavior and Idiosyncratic Volatility.”

cr
Journal of Business Research 68, no 2: 763-770.

us
Hwang, S. and M. Salmon. 2004. “Market Stress and Herding.” Journal of Empirical Finance 11,

no 4: 585-616.
an
Kahneman, D. and A. Tversky. 1979. “Prospect Theory: An Analysis of Decision under Risk.”

Econometrica 47, no 1: 263-292.


M

Lakonishok, J., A. Shleifer and R.W. Vishny. 1992. “The Impact of Institutional Trading on
ed

Stock Prices.” Journal of Financial Economics 32, no. 3: 23-43.

Low, C. 2004. “The Fear and Exuberance from Implied Volatility of S&P 100 Index Options.”
pt

Journal of Business 77, no. 3: 527-546.

McQueen, G., M. Pinegar and S. Thorley. 1996. “Delayed Reaction to Good News and the
ce

Cross-Autocorrelation of Portfolio Returns.” Journal of Finance 51, no. 3: 889-919.


Ac

Natividad, B., C. Pilar and F. Sandra. 2012. “Does Herding Affect Volatility?Implications for

the Spanish Stock Market.” Quantitative Finance 12, no. 4: 311-327.

Newey, W. K. and K. D. West. 1987. “A Simple, Positive Semi-Definite, Heteroskedasticity and

Autocorrelation Consistent Covariance Matrix.” Econometrica 55, no. 3: 703-708.

Nofsinger, J.R. and R.W. Sias. 1999. “Herding and Feedback Trading by Institutional and

Individual Investors.” Journal of Finance 54, no. 4: 2263-2295.

21
Sias, R.W. 2004. “Institutional Herding.” Review of financial Studies 17, no. 1: 165-206.

Venezia, I., A. Nashikkar and Z. Shapira. 2011. “Firm Specific and Macro Herding by

Professional and Amateur Investors and Their Effects on Market Volatility.” Journal of

Banking & Finance 35, no. 3: 1599-1609.

Whaley, R.E. 2000. “The Investor Fear Gauge.” Journal of Portfolio Management 26, no.4:

12-17.

t
ip
cr
us
an
M
ed
pt
ce
Ac

22
Table 1 Descriptive statistics

t
Table 1 presents the descriptive statistics on daily cross-sectional standard deviation (CSSD), the daily cross-sectional absolute deviation of

rip
returns (CSAD), and the daily VIX for full sample period and eight sub-periods. The Stdev is the standard deviation; The Min and Max are the

c
us
minimum and maximum, respectively. The “No.” indicates the number of trading days in studying sample. In addition, for full sample period,

an
the serial correlation of daily CSSD, CSAD, and VIX are reported for lag 1, 3, 10, and 20 along with t-statistics of the Augmented Dickey-Fuller

M
test. The whole sample period has 1,989 trading observations from January 1, 2007 to December 31, 2014.

d
Serial correlation lag
Full period 2007 2008 2009
te 2010 2011 2012 2013 2014
(Full period)
c ep
Ac

CSSD CSSD

No. 1,989 247 249 251 251 247 250 246 248 1 0.792

23
Mean 0.021 0.024 0.026 0.024 0.021 0.019 0.018 0.018 0.017 3 0.707

Stdev 0.004 0.004 0.004 0.004 0.003 0.003 0.003 0.002 0.002 10 0.633

t
rip
Min 0.012 0.017 0.015 0.016 0.014 0.015 0.014 0.012 0.013 20 0.547

c
us
Max 0.043 0.035 0.043 0.038 0.029 0.036 0.029 0.024 0.031 ADF-test -4.29***

an
CSAD CSAD

M
No. 1,989 247 249 251 251 247 250 246 248 1 0.796

d
Mean 0.015 0.017 0.019 0.018 te0.014 0.014 0.012 0.012 0.011 3 0.705
ep
Stdev 0.004 0.003 0.004 0.004 0.002 0.003 0.003 0.001 0.002 10 0.631
c

Min 0.008 0.011 0.009 0.012 0.009 0.011 0.009 0.008 0.008 20 0.553
Ac

Max 0.035 0.028 0.035 0.031 0.024 0.029 0.023 0.017 0.022 ADF-test -4.27***

24
VIX VIX

t
rip
No. 1,989 247 249 251 251 247 250 246 248 1 0.987

c
us
Mean 22.04 23.49 34.39 34.43 19.45 22.63 19.14 13.69 11.87 3 0.965

Stdev 9.31 7.46 9.18 6.55 2.76 7.31 3.58 1.57 1.76 10 0.925

an
Min 8.37 11.74 22.62 18.12 13.29 13.71 14.61 9.79 8.37 20 0.864

M
Max 60.41 44.51 60.41 46.72 27.14 40.61 33.39 18.57 17.51 ADF-test -2.65*

d
te
c ep
Ac

25
Table 2 Regression results of CSSD and CSAD on market dummy variables for small and

large changes in the VIX

This table reports the estimation results of Equation (7) for the full sample conditional on small and large changes

in the VIX. The large changes group contains the days with unusually high positive and negative changes (10%) in

t
ip
the VIX, and the small changes group contains the other days. The Equation (7) is shown as follows:

cr
L
Vari t = + DtL + U
DtU + Vari t −1 + t
, where Vari is the dependent variables in the regression, and the

us
subscript i denotes the CSSD or CSAD measure. DtL = 1 , if the market return on day t lies in the extreme

lower tail of the returns distribution; and equal to zero otherwise, and DtU = 1 , if the market return on day t lies
an
in the extreme upper tail of the returns distribution; and equal to zero otherwise. The criterions for extreme are
M
arranged at 5% and 10% of the market returns observations. t-statistic is in parentheses. ***, **, and * denote the

significant level at 1%, 5%, and 10% level, respectively.


ed

Small changes in VIX Large changes in VIX (10%)


pt

L U L U
Adj-R2 Adj-R2
ce

Criterion = 5%
Ac

CSSD 0.004*** -0.0001 -0.0004 0.794*** 0.629 0.008*** 0.0002 -0.0008 0.643*** 0.411

(10.97) (-0.36) (-1.51) (41.43) (8.52) (0.33) (-1.33) (15.72)

CSAD 0.003*** -0.0001 -0.0002 0.797*** 0.636 0.006*** 0.0003 -0.0003 0.637*** 0.404

(10.75) (-0.03) (-0.43) (41.58) (8.19) (0.52) (-0.48) (15.21)

26
Criterion = 10%

CSSD 0.005*** 0.0002 0.0001 0.795*** 0.632 0.007*** -0.0001 -0.0007 0.645*** 0.413

(10.65) (0.73) (0.58) (41.48) (8.45) (-0.03) (-1.26) (15.76)

t
ip
CSAD 0.003*** 0.0001 0.0003 0.798*** 0.634 0.005*** 0.0003 -0.0003 0.637*** 0.402

cr
(10.51) (0.71) (0.67) (41.79) (8.05) (0.59) (-0.52) (15.09)

us
an
M
ed
pt
ce
Ac

27
Table 3 Regression results of CSSD and CSAD on the absolute and squared returns of market indices within the changes in the VIX

and trading volume

t
rip
Panel A reports the estimation results of Equation (8) conditional on the positive and negative changes in the VIX for the full sample, and eight sub-periods, and Panel B

c
us
reports the result of Equation (8) conditional on the trading volume for full sample period and GFC period and NGFC period. The Equation (8) is shown as follows:

an
Vari t = + 1 R m ,t + 2 R m2 , t + Vari
3 t −1 + t
, where Vari denotes the dependence variable in the regression model, and subscript i denotes the CSSD or CSAD.

is absolute value of average returns of market portfolio on day t , and R m2 ,t is its squared value. T-statistic is in parentheses. ***, **, and * denote the significant

M
R m ,t

level at 1%, 5%, and 10% level, respectively.

d
Panel A: The estimation results of Equation (8) conditional on VIX
te
ep
ΔVIX ≥ 0 ΔVIX < 0
c
Ac

1 2 3 Adj-R2 1 2 3 Adj-R2

28
t
Full CSSD 0.006*** 0.292*** -5.004*** 0.651*** 0.605 0.005*** 0.203*** -3.868*** 0.721*** 0.623

rip
period (10.42) (9.05) (-6.57) (22.69) (8.24) (5.79) (-3.56) (23.54)

c
us
CSAD 0.004*** 0.339*** -5.438*** 0.611*** 0.648 0.003*** 0.264*** -4.198*** 0.675*** 0.667

an
(10.02) (12.09) (-7.27) (22.39) (7.54) (8.16) (-3.93) (20.72)

M
2007 CSSD 0.008*** 0.269*** -5.075*** 0.601*** 0.527 0.007*** 0.292*** -6.662*** 0.608*** 0.427

d
(4.24) (5.56) (-5.58) (7.28) (4.41) (3.92) (-4.18) (7.63)

CSAD 0.006*** 0.294*** -4.803***


te0.573*** 0.601 0.005*** 0.362*** -7.231*** 0.569*** 0.483
ep
(4.05) (7.64) (-6.56) (6.56) (3.93) (5.08) (-4.17) (6.82)
c
Ac

2008 CSSD 0.012*** 0.216** -4.131** 0.459*** 0.225 0.012*** 0.092 -1.958 0.507*** 0.243

(5.84) (2.43) (-2.38) (6.07) (4.49) (1.14) (-0.94) (3.46)

29
CSAD 0.009*** 0.325*** -5.668*** 0.418*** 0.261 0.008*** 0.209*** -3.565* 0.475*** 0.301

(5.64) (3.81) (-3.45) (5.73) (3.62) (2.74) (-1.87) (3.51)

t
rip
2009 CSSD 0.008*** 0.351*** -6.872*** 0.543*** 0.441 0.007*** 0.141* -4.129 0.665*** 0.443

c
us
(3.53) (3.41) (-2.85) (4.84) (4.64) (1.89) (-1.51) (10.11)

CSAD 0.006*** 0.419*** -7.795*** 0.523*** 0.512 0.004*** 0.216*** -4.584** 0.659*** 0.546

an
(3.97) (4.51) (-3.61) (5.79) (4.42) (3.51) (-2.03) (11.51)

M
2010 CSSD 0.011*** 0.252*** -4.025*** 0.422*** 0.346 0.009*** 0.174** -2.676 0.491*** 0.386

d
(5.01) (5.11) (-3.05) te(4.26) (4.51) (2.33) (-1.21) (4.52)
ep
CSAD 0.007*** 0.254*** -3.468*** 0.387*** 0.419 0.007*** 0.201*** -1.753 0.363*** 0.448
c

(6.14) (6.58) (-3.56) (4.96) (4.93) (2.94) (-0.94) (3.04)


Ac

2011 CSSD 0.012*** 0.251*** -3.363** 0.329*** 0.321 0.012*** 0.227*** -2.691 0.314*** 0.361

30
(5.89) (3.31) (-1.99) (3.45) (7.98) (3.61) (-1.56) (4.08)

CSAD 0.008*** 0.312*** -4.001** 0.293*** 0.445 0.008*** 0.252*** -1.903 0.245*** 0.506

t
rip
(5.99) (4.15) (-2.27) (3.45) (9.32) (4.93) (-1.46) (3.46)

c
us
2012 CSSD 0.009*** 0.143** 0.973 0.434*** 0.523 0.011*** 0.175** -0.009 0.372*** 0.348

(6.73) (2.12) (0.57) (5.78) (5.02) (2.45) (-0.01) (3.14)

an
CSAD 0.006*** 0.191*** 0.797 0.348*** 0.668 0.006*** 0.216*** 0.241 0.343*** 0.502

M
(7.05) (3.54) (0.57) (4.69) (4.91) (3.92) (0.15) (3.01)

d
2013 CSSD 0.011*** 0.161** 0.416 te
0.316*** 0.235 0.009*** 0.081 2.267 0.382*** 0.203
ep
(7.43) (2.19) (0.13) (3.87) (7.89) (1.05) (0.61) (5.43)
c

CSAD 0.008*** 0.193*** -0.442 0.237*** 0.325 0.008*** 0.058 6.479** 0.275*** 0.279
Ac

(8.23) (3.49) (-0.21) (3.01) (6.75) (1.03) (2.31) (2.87)

31
t
continued

c rip
us
2014 CSSD 0.011*** 0.147** 4.89** 0.325*** 0.397 0.011*** 0.017 8.645 0.408*** 0.332

an
(5.81) (2.06) (2.15) (2.95) (7.51) (0.16) (1.45) (5.64)

M
CSAD 0.007*** 0.143*** 3.483** 0.299*** 0.454 0.007*** 0.051 6.933 0.359*** 0.392

d
(5.96) (2.64) (2.08) (2.73) (8.63) (0.69) (1.43) (5.69)

te
ep
Panel B: The estimation results of Equation (8) conditional on trading volume
c
Ac

Small trading volume Large trading volume

32
1 2 3 Adj-R2 1 2 3 Adj-R2

t
rip
Full CSSD 0.005*** 0.185*** -3.029*** 0.713*** 0.613 0.005*** 0.198*** -3.193*** 0.713*** 0.628

c
us
period (9.41) (5.45) (-3.61) (27.29) (7.37) (5.64) (-2.85) (19.94)

an
CSAD 0.003*** 0.264*** -3.857*** 0.659*** 0.651 0.003*** 0.228*** -3.381*** 0.709*** 0.689

(8.74) (7.83) (-4.56) (24.98) (6.62) (6.72) (-2.87) (20.38)

M
d
GFC CSSD 0.009*** 0.187*** -3.442***
te0.592*** 0.434 0.008*** 0.093* -1.612 0.654*** 0.451
ep
period (9.05) (4.16) (-3.41) (13.71) (5.16) (1.83) (-1.35) (10.13)
c
Ac

CSAD 0.006*** 0.282*** -4.627*** 0.564*** 0.513 0.005*** 0.157*** -2.142* 0.645*** 0.512

(8.51) (6.52) (-4.73) (13.78) (4.81) (3.35) (-1.75) (10.11)

33
t
NGFC CSSD 0.007*** 0.149*** -0.919 0.578*** 0.564 0.008*** 0.227*** -3.149** 0.469*** 0.398

rip
period (9.15) (4.82) (-1.01) (14.21) (12.11) (5.62) (-2.12) (12.23)

c
us
CSAD 0.005*** 0.194*** -0.889 0.505*** 0.661 0.005*** 0.251*** -3.445** 0.458*** 0.487

an
(11.09) (7.34) (-1.17) (14.48) (11.43) (6.79) (-2.41) (11.98)

M
d
te
c ep
Ac

34
Table 4 Estimation results within the quintiles of changes in the VIX

This table reports the estimation results of Equation (8) within the quintiles of the positive and negative

changes in the VIX during the full sample period. Group 1 denotes the smallest changes in the VIX and

Group 3 denotes the largest changes in the VIX. See the note to Table 3 for a detailed description of

t
ip
Table 4.

cr
Panel A: The estimation results of Equation (8) under the ΔVIX ≥ 0

us
CSSD CSAD

Adj-R Adj-R
an
1 2 3 1 2 3
2 2
M
ed

Grou 0.018** 0.501** -7.569** -0.015 0.246 0.012** 0.512** -7.273** -0.024 0.329

p1 * * * (-0.37) * * * (-0.62)
pt

(18.59) (8.75) (-5.17) (17.39) (9.33) (-4.55)


ce

Grou 0.007** 0.267** -4.961** 0.608** 0.506 0.004** 0.315** -5.505** 0.588** 0.556
Ac

p2 * * * * * * * *

(7.43) (5.85) (-5.61) (13.09) (7.09) (6.95) (-5.49) (12.53)

Grou 0.008** 0.273** -3.892** 0.541** 0.464 0.005** 0.332** -4.539** 0.486** 0.509

p3 * * * * * * * *

35
(6.15) (4.82) (-3.49) (8.81) (5.72) (6.04) (-4.03) (7.91)

Panel B: The estimation results of Equation (8) under the ΔVIX <0

CSSD CSAD

t
Adj-R Adj-R

ip
1 2 3 1 2 3
2 2

cr
us
Grou 0.005** 0.243** -3.592** 0.677** 0.619 0.004** 0.306** -4.281** 0.631** 0.656
an
p1 * * * * * * * *
M
(7.64) (5.97) (-3.34) (17.63) (7.65) (8.68) (-5.05) (15.82)

Grou 0.004** 0.234** -4.698 0.487** 0.569 0.003** 0.291** -4.104 0.615** 0.612
ed

p2 * (2.49) (-1.39) * * * (-1.27) *


pt

(4.02) (7.46) (5.35) (3.47) (11.68)


ce

Grou 0.006** 0.071 -0.829 0.685** 0.516 0.004** 0.155** -1.803 0.646** 0.563

p3 * (1.18) (-0.72) * * * (-1.60) *


Ac

(6.85) (12.64) (5.49) (3.08) (13.98)

36
Table 5 Asymmetric reaction examination in up and down markets

t
Panel A reports the examination results of asymmetric reaction by using Equation (9) for the up market and Equation (10) for the down market during the full sample period

rip
and eight sub-periods. Panel B reports the empirical results of asymmetric reaction based on trading volume over full sample, GFC, and NGFC periods. The Equation (9)

c
us
and (10) are shown as follows, respectively: Vari tUP = + UP
1 R mUP,t + UP
2 ( R mUP,t ) 2 + UP
3 Vari tUP
−1 + t
,

an
Vari tDOWN = + 1
DOWN
R mDOWN
,t + DOWN
2 ( R mDOWN
,t )2 + DOWN
3 Vari tDOWN
−1 + t
, where R mUP is the absolute value of the cross-sectional average return of all available stock

when the market is up, while RmDOWN is the absolute value of the cross-sectional average return of all available stock when the market is down. ( RmUP ) 2 and ( RmDOWN ) 2

M
are the squared value of market return in up and down markets, respectively. t-statistics is in parentheses. Differences in coefficients are tested by utilizing the t-statistic. ***,

d
te
**, and * denote significant level at the 1%, 5%, and 10% level, respectively.
ep
Panel A: The examination results of asymmetric reaction
c
Ac

UP market Down market

UP UP UP DOWN DOWN DOWN UP DOWN UP DOWN


1 2 3 Adj-R2 1 2 3 Adj-R2 1 − 1 2 − 2

37
Full CSSD 0.201*** -4.563*** 0.734*** 0.621 0.251*** -3.512*** 0.672*** 0.649 -0.049 -1.049

t
rip
period (5.48) (-4.43) (31.29) (7.26) (-4.52) (22.97) (-0.96) (-0.82)

c
us
CSAD 0.265*** -4.928*** 0.701*** 0.671 0.291*** -3.817*** 0.619*** 0.683 -0.026 -1.112

an
(7.24) (-4.72) (29.96) (8.68) (-5.01) (19.65) (-0.53) (-0.86)

2007 CSSD 0.201*** -4.953*** 0.502*** 0.308 0.271*** -4.551*** 0.665*** 0.609 -0.069 -0.402

M
(2.76) (-2.92) (5.24) (3.36) (-2.91) (7.97) (-0.63) (-0.17)

d
CSAD 0.267*** -5.196*** 0.484*** te
0.398 0.313*** -4.709*** 0.612*** 0.651 -0.045 -0.487
ep
(4.51) (-4.11) (5.12) (4.08) (-3.02) (7.11) (-0.46) (-0.24)
c

2008 CSSD 0.186** -4.172** 0.516*** 0.262 0.069 -0.817 0.486*** 0.239 0.116 -3.354
Ac

(2.04) (-2.03) (7.32) (0.68) (-0.38) (5.49) (0.84) (-1.13)

38
CSAD 0.305*** -5.679*** 0.464*** 0.314 0.178*** -2.334* 0.439*** 0.271 0.127 -3.346

(3.48) (-2.94) (6.91) (2.63) (-1.69) (5.87) (1.15) (-1.41)

t
rip
2009 CSSD 0.187*** -6.009*** 0.629*** 0.423 0.255*** -3.589** 0.684*** 0.658 -0.068 -2.419

c
us
(3.03) (-6.54) (7.62) (4.28) (-2.45) (12.87) (-0.79) (-1.39)

CSAD 0.281*** -6.859*** 0.641*** 0.526 0.314*** -4.483*** 0.614*** 0.667 -0.033 -2.375

an
(4.81) (-7.54) (8.84) (6.19) (-3.83) (11.68) (-0.42) (-1.61)

M
2010 CSSD 0.174*** -5.639** 0.557*** 0.384 0.291*** -4.405*** 0.491*** 0.466 -0.117 -1.233

d
(2.61) (-2.27) (6.28) te (4.57) (-3.11) (8.12) (-1.26) (-0.43)
ep
CSAD 0.179*** -3.357* 0.483*** 0.426 0.289*** -3.737*** 0.427*** 0.529 -0.109 0.379
c

(3.06) (-1.68) (6.42) (5.05) (-3.16) (7.45) (-1.34) (0.16)


Ac

2011 CSSD 0.164*** -1.481 0.269*** 0.301 0.301*** -4.296*** 0.362*** 0.371 -0.136 2.815

39
(2.71) (-1.01) (3.06) (4.05) (-2.64) (4.39) (-1.42) (1.28)

CSAD 0.195*** -0.837 0.184** 0.474 0.349*** -4.714*** 0.334*** 0.493 -0.155* 3.878*

t
rip
(3.79) (-0.67) (2.35) (4.79) (-2.78) (4.45) (-1.73) (1.84)

c
us
2012 CSSD 0.211*** -1.514 0.446*** 0.443 0.107* 1.853 0.471*** 0.509 0.104 -3.367

(3.29) (-0.79) (4.27) (1.71) (1.16) (4.97) (1.16) (-1.35)

an
CSAD 0.236*** -0.735 0.445*** 0.619 0.156*** 1.606 0.391*** 0.643 0.079 -2.341

M
(5.16) (-0.52) (5.11) (3.08) (1.25) (4.57) (1.16) (-1.22)

d
2013 CSSD 0.233* -14.099 0.249*** te
0.091 0.147* 1.089 0.375*** 0.318 0.086 -15.179
ep
(1.71) (-1.23) (2.95) (1.87) (0.35) (4.74) (1.04) (-1.33)
c

CSAD 0.231** -10.733 0.194** 0.128 0.159** 1.723 0.296*** 0.443 0.073 -12.453
Ac

(2.21) (-1.26) (2.29) (2.54) (0.67) (3.95) (0.04) (-1.56)

40
2014 CSSD -0.072 12.806** 0.338*** 0.208 0.216** 2.771 0.398*** 0.504 -0.287** 10.035

(-0.68) (2.27) (4.09) (2.55) (1.04) (5.11) (-2.14) (1.61)

t
rip
CSAD -0.031 10.608** 0.291*** 0.253 0.208*** 1.516 0.368*** 0.567 -0.239** 9.093*

c
us
(-0.39) (2.42) (3.59) (3.21) (0.77) (4.56) (-2.34) (1.89)

an
Panel B: The examination results of asymmetric reaction conditional on trading volume

M
UP market Down market

d
UP
1
UP
2
UP
3
te
Adj-R2 1
DOWN DOWN
2
DOWN
3 Adj-R2 UP
1 − DOWN
1
UP
2 − DOWN
2
c ep

Small trading volume


Ac

Full CSSD 0.211*** -4.825*** 0.706*** 0.554 0.206*** -2.507** 0.662*** 0.598 0.005 -2.318

41
period (4.38) (-3.86) (21.79) (4.34) (-2.37) (18.28) (0.08) (-1.42)

CSAD 0.298*** -5.605*** 0.664*** 0.603 0.269*** -3.179*** 0.587*** 0.615 0.029 -2.426

t
rip
(6.42) (-4.59) (21.04) (5.87) (-3.08) (14.02) (0.45) (-1.52)

c
us
GFC CSSD 0.212*** -5.061*** 0.583*** 0.387 0.183*** -2.601* 0.551*** 0.414 0.029 -2.459

period (3.61) (-3.82) (11.56) (2.78) (-1.86) (9.19) (0.32) (-1.28)

an
CSAD 0.313*** -6.215*** 0.565*** 0.479 0.267*** -3.687*** 0.487*** 0.464 0.046 -2.528

M
(5.67) (-4.97) (12.42) (4.22) (-2.74) (7.25) (0.54) (-1.37)

d
NGFC CSSD 0.134*** -0.801 0.471*** te
0.383 0.223*** -1.838 0.521*** 0.558 -0.088 1.037
ep
period (3.03) (-0.58) (6.36) (3.83) (-1.25) (12.09) (-1.21) (0.52)
c

CSAD 0.176*** -0.379 0.402*** 0.525 0.252*** -1.794 0.443*** 0.634 -0.076 1.416
Ac

(4.89) (-0.35) (6.47) (4.94) (-1.42) (9.48) (-1.22) (0.85)

42
Large trading volume

t
rip
Full CSSD 0.133*** -2.239 0.708*** 0.571 0.313*** -5.304*** 0.653*** 0.606 -0.179*** 3.066

c
us
period (3.17) (-1.47) (14.29) (7.57) (-4.65) (11.27) (-3.06) (1.61)

CSAD 0.164*** -1.885 0.703*** 0.631 0.318*** -5.021*** 0.643*** 0.662 -0.153*** 3.136

an
(4.22) (-1.21) (14.35) (8.17) (-4.03) (11.07) (-2.79) (1.57)

M
GFC CSSD 0.112 -2.791 0.468*** 0.203 0.171** -2.621 0.676*** 0.516 -0.059 -0.171

d
period (1.42) (-1.59) (4.36) te (2.08) (-1.38) (7.15) (-0.54) (-0.07)
ep
CSAD 0.196*** -3.533** 0.455*** 0.265 0.181*** -1.898 0.646*** 0.543 0.015 -1.636
c

(2.98) (-2.45) (4.71) (2.71) (-1.28) (6.67) (0.16) (-0.79)


Ac

NGFC CSSD 0.098** 1.785 0.405*** 0.294 0.321*** -5.054*** 0.422*** 0.403 -0.223*** 6.839***

43
period (2.03) (0.95) (8.04) (7.18) (-4.12) (7.59) (-3.37) (3.04)

CSAD 0.116*** 2.382 0.405*** 0.393 0.328*** -5.031*** 0.401*** 0.492 -0.212*** 7.413***

t
rip
(2.95) (1.61) (7.92) (8.56) (-4.27) (7.51) (-3.87) (3.91)

c
us
an
M
d
te
c ep
Ac

44
Table 6 Asymmetric reaction examination in up and down markets within the

changes in the VIX

This table reports the estimation results of Equation (9) for the up market, and Equation (10) for the

down market conditional on the positive (negative) changes in the VIX during the full sample period.

t
ip
See the note to Table 5 for a detailed description of Table 6.

cr
ΔVIX ≥ 0 ΔVIX < 0

us
Up market CSSD CSAD Up market CSSD CSAD
an
0.006*** 0.004*** 0.005*** 0.003***
M
(7.61) (7.03) (10.11) (9.82)

UP UP
0.381*** 0.431*** 0.179*** 0.238***
ed

1 1

(7.23) (9.25) (6.18) (9.49)


pt

UP UP
2 -7.079*** -7.424*** 2 -3.187*** -3.451***
ce

(-6.14) (-7.26) (-4.17) (-5.53)


Ac

UP UP
3 0.614*** 0.591*** 3 0.686*** 0.649***

(15.23) (15.58) (26.61) (27.24)

Adj-R2 0.581 0.632 Adj-R2 0.576 0.641

45
Down market Down market

0.007*** 0.005*** 0.006*** 0.004***

(11.52) (11.82) (7.41) (7.45)

DOWN DOWN
1 0.251*** 0.294*** 1 0.389*** 0.427***

t
(8.75) (11.63) (6.58) (8.08)

ip
DOWN DOWN
-3.472*** -3.825*** -6.733*** -6.601***

cr
2 2

(-5.47) (-6.86) (-4.36) (-4.79)

us
DOWN DOWN
3 0.596*** 0.537*** 3 0.613*** 0.558***
an
(21.39) (20.18) (15.69) (14.89)
M
Adj-R2 0.573 0.616 Adj-R2 0.589 0.625
ed

UP DOWN UP DOWN
1 − 1 0.128** 0.137** 1 − 1 -0.209*** -0.187***
pt

(2.15) (2.56) (-3.19) (-3.23)


ce

UP DOWN UP DOWN
2 − 2 -3.607*** -3.598*** 2 − 2 3.546** 3.152**

(-2.74) (-3.09) (2.05) (2.07)


Ac

46
Ac
ce
pt
ed

47
M
an
us
cr
ip
t

You might also like