Herding behavior in vietnamese stock market empirical evidence from quantile regression analysis

International School of Business --- Phan Dang Bao Anh HERDING BEHAVIOR IN VIETNAMESE STOCK MARKET: EMPIRICAL EVIDENCE FROM QUANTILE REGRESSION ANALYSIS MASTER OF BUSINESS Honours H

International School of Business

-

Phan Dang Bao Anh

HERDING BEHAVIOR IN VIETNAMESE STOCK

MARKET: EMPIRICAL EVIDENCE FROM

QUANTILE REGRESSION ANALYSIS

MASTER OF BUSINESS (Honours)

Ho Chi Minh City – Year 2015

International School of Business

-

Phan Dang Bao Anh

HERDING BEHAVIOR IN VIETNAMESE STOCK

MARKET: EMPIRICAL EVIDENCE FROM

QUANTILE REGRESSION ANALYSIS

ID: 22130006

MASTER OF BUSINESS (Honours)

SUPERVISOR: A.Pro.Dr VO XUAN VINH

Ho Chi Minh City – Year 2015

Firstly, I would like to express my gratefulness to my supervisor A.Prof Dr.Vo Xuan Vinh for his professional guidance, intensive support, valuable suggestions, instructions and continuous encouragement during the time of research and writing this thesis

I would like to express my deepest appreciation to ISB Research Committee for their valuable time as their insightful comments and meaningful suggestions were contributed significantly for my completion of this research

My sincere thanks also go to all of all of my lecturers at International Business School- University of Economics Ho Chi City for their teaching and guidance during

my Master course

Last but not least, I would like to thanks my family, whom were always

supporting me and encouraging me with their best wishes

TABLE OF CONTENT

CHAPTER 1: INTRODUCTION 1

1.1 Research background 2

1.2 Research gap 4

1.3.Research objectives 5

1.4 Research methodology and scope 6

1.5 Research structure 7

CHAPTER 2: LITERATURE REVIEW 8

2.1 Theoretical literature review 8

2.2 Empirical literature review 11

2.3 Measuring herding in financial markets 19

2.4 Hypothesis development 25

CHAPTER 3: RESEARCH METHODOLOGY 27

3.1 Data collection and sample description 27

3.2 Regression model for testing the hypotheses 28

3.2.1 Regression model for testing the presence of herding bahavior in Vietnamese stock market: 28

3.2.2 Regression model for estimation the degree of herd in rising and falling market: 29

3.3 Regression methodology 30

3.3.1 Research process 31

3.3.2 Quantile regression analysis 31

CHAPTER 4: EMPIRICAL RESULT 34

4.1 Decriptive statistics 34

4.2 Correlation analysis among variables 35

4.3 Regression result 36

4.3.1 Evidence on herd presence in Vietnamese stock market 36

4.3.2 Herding behavior in up and down markets 38

4.4 Regression result from Quantile regression analysis 39

5.1 Conclusion 45

5.2 Implications of herding behavior in Vietnamese stock market 46

5.3 Limitations and further research direction 48

REFERENCES 50

APPENDICES 55

Table 1.1: A summary of empirical evidence on herding behavior 15

Table 3.1: Summary of data observations used in the study 27

Table 4.1: Descriptive statistics for daily market return and cross-sectional absolute deviation (CSAD) for the Vietnamese stock market from 1/2005 to 4/2015 34

Table 4.2 Correlation among main variables 35

Table 4.3: Regression result of herding behavior in Vietnamese stock market 36

Table 4.4: Regression results of herding behavior in rising and declining market 38

Table 4.5: Analysis of herding behavior in Vietnamese stock market by quantile regression 40

Table 4.6: A summary of research results 43

ABSTRACT

This study examines the herding behavior of investors in Vietnamese stock market using data sample of 299 companies listed on Ho Chi Minh City Stock

Exchange Using a least square method, the author finds evidence of herding presence

in rising and falling market when considering over the period of 2005 – 4/2015 as well

as in the periods of pre-crisis and post-crisis By applying quantile regression analysis

to estimate the herding equation, the author find supporting evidence of herding during the period studied as well as when splitting the market into two sub-periods; however, the level of this trend is somewhat different conditional on quantile region

Key words: herding behavior, Vietnamese stock market, quantile regression,

asymmetry

CHAPTER 1: INTRODUCTION

This chapter presents the introduction of the study It contains the research background, research gap, research objectives, research methodology and scope and research structure

1.1 Research background

Traditional financial framework understands financial market by using models which meet four foundation conditions: (i) investors are assumed to be rational, (ii) market is efficient, (iii) investors make a decision on portfolios based on the rules of mean-variance portfolio theory, and (iv) the expected returns are a function of risk (Statman, 2014) Among them, the condition of rational investor is considered a

central assumption in which people make decisions reasonably and no biases in their future prediction However, the world economy has been shaken by the global

financial crisis in 2008, which originated from US and then expanded globally As soon as the crisis began, many economists and financial forecasters were no longer able to analyze the bankruptcy of a variety of enterprises or banks in an intensive way

The Vietnamese stock market is not an exception From its foundation in 2000, the Vietnamese stock market experiences “hot” growth and drastical fluctuation

without stability causing virtual stock matter The value of VN-Index in 2000 of 100 points increases to 571 points after just one year and a half which astonishes economic experts; however, this increment does not last long and rush to fall under 140 points in

2003, 150-200 points in 2004 The peak of growing phase is in the period of

2006-2007 as Vietnamese stock market has the highest growth of 1100 points

(approximately 145%) in Asia – Pacific region, even exceeding the Shanghai stock market growth of 135% Particularly, the VN-Index reaches to the record of 1170.67 points on March 12th, 2007 – the highest level in the world This event makes stock experts and market managers difficult to understand, thereby bring out the fear of bubble formation in the stock market

After a long time of increasing prices, the Vietnamese stock market has signal

to considerably decrease with the lowest record of 236 points on February 24th, 2009 The happening in the market during this period is very complicated to anticipate Once again, economics experts doubt the precision of the efficient market theory A paradox

is present that when the stock price is driven further from the fundamental value of 30% investors still trade constantly; whereas, when the stock prices decrease at an attractive level in declining market investors massively sell stocks instead of buying Is

it true that the Vietnamese stock market operation does not abide by any rules or there are phenomena dominating the market which cause an unusual fluctuation? Failure of the economists as well as their theories leads to a list of different questions in different context: Are people rational? Or are they influenced by emotion such as fear, greed which caused wrong decision?

Then, a new branch of financial research appears beside traditional financial framework which helps economic experts and finance researchers partial explain unusual fluctuation Behavioral finance is a new strand of finance which investigates the behavior of investors in financial market; in other word, it is a combination

between psychology and finance It considers psychological factors as essential input

to financial analysis Behavioral finance can elucidate several financial reactions that contrast with standard financial theory and can thus make a contribution to avoidance

of mistakes as well as advancing investment strategies (Fromlet, 2001)

Previous researchers put sustained effort to understand investors’ behaviour in the market as well as its impact on stock price These investment behaviors are

influenced by some factors such as investors’ insight, criterion to measure investment efficiency or market instability… In terms of psychology, investors are assumed to be rational and always strive to optimize their actions but the fact that the rationality appears to be inhibited by numerous cognitive biases, such as overconfidence, over-optimism, herding, representativeness … and so on In this research, the author focuses

on the investment behavior of market participants regarding to their tendency to follow the actions of others, which engages in herd behavior

Herding behavior is defined as the trend of investors to imitate the actions of others (Luu, 2013) This tendency is considered an inherent psychology of investors but it becomes stronger as they have to make decision in a market condition with high uncertainty and low transparency Over last decades, research regarding this topic receives an attention from scientists and empirical researchers A numerous theories are developed and empirical investigations are conducted to examine the presence and reasons of this phenomenon in financial market Researchers in this field believe that the presence of herding behavior has impact on results derived from asset pricing model because it influences stock price fluctuation, thus influencing risk and return of stocks (Tan et al, 2008) Similar to speculation, herding behavior may be rational or irrational If market participants follow market consensus, the fluctuation is more and more serious that can leads to instability in financial system, particularly in the period

of global crisis In addition, herding behavior lasting so long can drive the stock prices further fundamental value which causes destabilization If investors are dominated by sentiment such as greedy or fear of loss, they can trade in a “frenzied” way; as a

results, economic bubbles are created and may collapse the stock market In sum, herding behavior can lead to bad consequences of reducing the efficiency of market, even result in the market instability and financial collapse

Basing on these arguments, doing research about herding behavior can help investors have an objective overview and be prudent when making investment

decision Therefore, the author decides to do a research of “Herding behavior in

Vietnamese stock market: an empirical evidence from Quantile regression analysis” The study applies research model proposed by Chang, Cheng and Khorona (2000) and modified by Chiang et al (2010) to investigate the presence of herd in Vietnamese stock market

1.2 Research gap

Several empirical studies have examined and detected the herding behavior in many region throughout the world, form developed to emerging countries For

example: Chang et al (2000) find evidence of herding in Japan and South Korea but

no detection of herd in US and Hong Kong, Fu and Lin (2010) detect the presence of herd in Chinese stock market which was in accordance with the results from Zhou (2007) and Zhao (2011), Caparrelli, D’Arcangelis and Cassuto (2010) find the

evidence of herd in Italian stock market and Caporale, Economou and Phillipas (2008) prove the existence of this trend in Athen, Greece… In Vietnam, there are a little research test for the existence of this tendency applying different model for detection Nevertheless, previous research concentrates on using Ordinary Least Square (OLS) method to regress their model, which consists of a few drawbacks that may lead to wrong results Tran and Truong (2011) overcome this situation by using a more

powerful approach called GARCH; however, this method is still not optimal and

complicated in processing data

The new point of this study is the choice of methodology Instead of standard method of OLS and dummy variable models which are pronounced in earlier literature review, this research utilizes the model of Chang et al (2000) and applies quantile regression analysis in empirical investigation Quantile regression is considered a valid alternative to the estimation of herding model such as Christie and Huang (1995) and Chang et al (2000) (Jani, 2008)

In addition, preceding research investigated herding during very old period, usually from the formation of stock market in Vietnam to five years later At that time, the number of securities was extremely small, even there were only 5, 10 and 20

stocks listed in the Vietnamese stock market in 2000, 2001 and 2002, respectively The data set of this research is collected from 2005, when Vietnamese stock market marked

an impressive growth, until now, in which the market is gradually matured The

purpose of gathering the data set over the new period is to indicate a precise result with current condition of Vietnamese stock market

1.3.Research objectives

The substantial objective of this research is to test for the existence of herding behavior in Vietnamese stock market through examining the relationship between the level of equity return dispersion which measured by CSAD and the overall market return

The study proceeds to test for the asymmetric effect in case herd exists through investigating whether to have differences in degree of herding behavior under various market conditions

1.4 Research methodology and scope

With the research objectives are set earlier, the thesis uses indirect approach by examining the relationship between market return and stock return dispersion in order

to determine the existence as well as the level of herding behavior in Vietnamese stock market This research methodology is proposed by Chang et al (2000) and then

modified by Chiang et al (2010), which is applied commonly in research regarding herd later

The thesis conducts research based on data sample of 2568 daily observations

of 299 companies listed on Ho Chi Minh City Stock Exchange (HSX) over the period

of 2005-4/2015 Besides the traditional OLS method is used to test for the hypotheses, quantile regression analysis is also used as a new point in choosing research

methodology to test for the regression model The selection of quantile regression has more advantages compared to OLS

Ordinary least square approach is highly effective for intensive comprehension about the central tendencies in a data set But OLS analysis will be less useful for understanding about the points which lie closer to the upper or lower tail of the

distribution within a population (Gowlland, Xiao & Zeng, 2009) Meanwhile,

Gowlland et al (2009) assert that quantile regression allows researchers to describe more accurately behavior of the points lying towards the tails of the distribution

Moreover, this method makes contribution to alleviating some of the statistical

problem of traditional OLS method such as outlier sensitivity, non-normal distribution, errors in variables and so on (Zhou & Anderson, 2011)

CHAPTER 2: LITERATURE REVIEW

In this chapter, the theoretical framework and empirical investigation will be discussed more clearly Theoretical framework indicates general information about herd such as definition of this trend in financial market, particularly in stock market Empirical investigation shows the detection of herding in numerous countries as well

as asymmetric effect consideration Moreover, measurement of herding behavior in financial market is also presented In addition, the hypothesis is developed based on some argument

2.1 Theoretical literature review

Generally speaking, in economics and finance, herd or herding behavior means the process participants are imitating other actions or base their decision on others’ actions (Spyrou, 2013) Nofsinger and Sias (1999) define herd as a tendency where investors in the market trade in the same direction during the same time Avery and Zemsky (1998) suggest that investors who ignore their initial assessment and trade by following the trend in the previous trade Many other definitions regarding herd are proposed by other researchers For example, it is a mutual imitation (Welch, 2000), a behavior converge to the consensus (Hirshleifer and Teoh, 2003) or a form of

correlated behavior (Hwang and Salmon, 2004) Luu (2013) suggests that herd effect

in financial market is identified as a phenomenon in which investors tend to mimic others’ behavior

In the security market, investors herd by making their investment decisions based on the masses’ decisions of buying or selling stocks If rational investors usually follow their own belief that makes the market efficient, herding; in contrast, causes the inefficiency of the market In general, herding investors tend to act the same ways as the groups who have little inadequate information about the surrounding circumstances but behave in the same direction to get more secure (Caparrelli et al, 2010) Detecting the existence of herd is considered carefully by practitioners because of the fact that investors tend to ignore their own private information and usually follow majority of

the market, this leads to a consequence that the stock price deviates further from the fundamental value; therefore, impacting on many good chances for investment

What drives herding behavior is an important issue that needs addressing The reason behind this phenomenon may be diverse Scharfstein and Stein (1990) argue that the reputation concerns in labor market without perfect information environment and insurance for any bad decisions may lead managers to follow each other’s actions Scharfstein and Stein (1990) come up with a model in which the labors are able to understand their manager’ ability through investment decisions that manager makes Thus, in order to preserve their reputation before the employees when things go bad, managers may rationally herd by ignoring their own private information and follow the decisions of others In other words, herding behavior is considered as an insurance against manager underperformance (Rajan, 2006)

Trueman (1994) indicates that analyst abilities affect analyst compensation The expert tends to forecast earnings closer to prior earnings previously announced by other analysts in an effort to mimic higher ability and achieve higher compensation Graham (1990) suggests that analysts more likely to herd are the one who

characterized by high reputation and low ability Specifically, high reputation analysts have greater motives to follow market consensus in order to protect their reputation while low ability analysts are confused when there is strong public information

inconsistent with their private information and they know if they make forecasts

different from the consensus, they are more likely to be fired

Another reason driving herding behavior is information-based issues An

informational cascade takes place when individuals feel optimal to follow actions of individuals before them without considering their own information (Bikhchandani et

al, 1992) For example, for investors who are newly enter the market at the later stage,

it may be an optimal decision to imitate trading behavior of previous investors since they surmise that previous investors possess more accurate information A decision model analyzed by Banerjee (1992) shows that decision makers tend to look at the

decisions made by previous decision makers since previous decision makers may hold related important information

Herding behavior caused by these above reasons has one point in common that individual investors/fund managers/analysts mimic action of market consensus

rationally; even in terms of social standpoint, it is inefficient This kind of herding behavior is called “intentional” herding Bikhchandani and Sharma (2000) distinguish between “intentional” herding and “spurious” herding “Intentional” herding is an extent in which the decisions are purely imitative; as such, investors copy other market participants and ignore their own information (Caparrelli, 2010) The rest is “spurious” herding where investors take similar investment decision since they confront with similar information sets and react to the same changes in fundamental factors (Spyrou, 2013) For example, rapidly rising bank interest rates make stock less attractive

investment Investors under the changed situations may want to open a bank account to earn interest instead of holding a larger percentage of stocks in their portfolio This is not herding in accordance with the definition above because investors are not making their decision after observing others Instead, they are reacting to commonly known public information, which is the increment of interest rates

Bikhchandani and Sharma (2000) suggest that “spurious herding” may lead to

an efficient outcome while “intentional herding” may not, even causes fragile market, excess volatility and systematic risk Spyrou (2013) argues that herding behavior spreading among institutional and individual investors is often cited as the main reason behind periods of extreme volatility and market instability Herd drives stock price away the fundamental value; as a consequences, herd lasting in the long-term can destabilize the market even lead to collapse Bikhchandani and Sharma (2001) indicate that when there are many investors imitating action of others who receive mis-

information, which is very popular in any emerging stock markets (i.e Vietnam),

herding behavior is formed Once there is any investor realizing that information is wrong, the imitation causes them to react back; as a result, this leads to extreme

volatility and pushes the market into crisis Research results found in Nakagawa, Oiwa

and Takeda (2010) assert that relatively strong herding behavior in financial institution

in 1980 may result in the formation and collapse of asset-price bubble

In summary, there are many reasons leading to herding behavior As above discussion, market participants may infer information from the actions of previous participants, fund managers may herd to protect their reputation, analysts herd for the reason related to remuneration or investors are simply non-rational and herding

behavior arises from the consequences of psychological or social conventions It is argued that herding may result in efficient outcome (spurious herding); however, some economists suggest that it may destabilize prices and give rise to bubble-like

phenomena in financial markets (intentional herding)

2.2 Empirical literature review

Empirical studies have focused on testing the existence of herding behavior as well as its effect on financial market Since 1990s, many researchers have made

intensive efforts to detect this trend in terms of various market participants within different markets of the world, such as individual or institutional investors in both emerging and developed countries Nevertheless, the evidence is found contrary

One branch of the empirical literature shows weak or no evidence of herd in stock market Lakonishok, Shleifer and Vishny (1991) use a quarterly portfolio of 769 equity pension funds between 1985 and 1989 as a new data set to evaluate how their trading affects stock prices The results indicate that no herding behavior is found in smaller stocks as well as no cross-sectional relationship between changes in pension funds’ holding of a stock and its abnormal return within the US

Christie and Huang (1995) propose a model to test the herd by using sectional standard deviation (CSSD) with a set of daily and monthly data from July

cross-1962 to December 1988 The findings show that no evidence regarding the presence of herding behavior in US markets In the same approach, Chang et al (2000) extend the model to a new and more powerful level in which CSSD is replaced by cross-sectional absolute deviation (CSAD) at time t to detect the herd They analyze US, Hong Kong,

Japan, South Korea and Taiwan and find no evidence of herding in US and Hong Kong, Japanese investors herd partially and especially notice the existence of this trend in two Asian countries (South Korea and Taiwan)

Fu and Lin (2010) examine herding behavior in Chinese stock market by

applying both Christie and Huang’s model and Chang et al’ and determine that herding

is not found in this country However, asymmetric effect exists which reveals that herding is greater in downward market than in the upward

Houda and Mohamed (2013) investigate how returns behavior involved the movement of the MSCI world index; in other words, herding during market upturns and downturns in Africa, Asia, Europe and America stock market by using another new approach in the spirit of CH and CCK model called EGARCH(1, 1) The results show that herd is significantly higher during rising market than declining market and this asymmetry is confirmed by ARCH model

On the other hand, another branch shows the evidence of herding presence Nofsinger and Sias (1999) find the relationship between institutional investors and herding formation in US stock market using Sias model with monthly data from 1977

to 1996 Similarly, Wermers (1999) also find the empirical evidence of herd in US mutual fund industry over the period of 1975 to 1994 Hwang and Salmon (2001) investigate herding in US, UK and South Korea and show an interesting results that advanced markets such as US and UK reveal less herd behavior than emerging

markets such as the South Korea

Caparrelli et al (2010) examine the herd existence in Italian stock market and found a non-linear relationship between dispersion and returns The results support Christie and Huang’s conclusions that herding is present in extreme market condition

in this country

Zhou (2007) indicates the presence of herding in both A shares and B shares in Chinese stock market applying Sias model With the same results, Tan et al (2008) and Zhao (2011) find the existence of herding in both Shanghai and Shenzhen stock

exchange though using different model in which Tan et al (2008) use Chang et al’s modified model; meanwhile, Zhao applies Var modelling for his detection

Economou, Kostakis and Philippas (2010) examine herd behavior in extreme market conditions using daily data from four Mediterranean stock markets including Greece, Italy, Portugal and Spain for the years 1998-2008 using model proposed by Chang et al (2000) The findings show that there is evidence for the presence of herd

in Italy and Greece during this period of time, this conclusion suites the results from examination of Caparrelli et al (2010) and Caporale et al (2008) for the Italian and Greek stock market, respectively In addition, asymmetric herding behavior is also investigated in terms of market returns, trading volume and return volatility Several outcome are found but noticeably, there is no evidence of herding in the Spanish stock market During the global financial crisis of 2008, herding is found only in Portuguese stock market; investors of three rest Mediterranean countries seem to have been

rational

A recent research of Choi and Skiba (2015) applying Sias’s model extended literature review about herding measure on international markets which includes 41 countries Using a set of quarterly institutional holdings data during the span of the last quarter of 1999 to the first quarter of 2010, the study finds statistically significant herding propensities in 41 target countries that have significant presence of

institutional investors

Tran and Truong (2011) examine the presence of herding behavior in

Vietnamese stock market as well as its asymmetric effect conditional on the direction

of market movements Using the data of daily price series of all securities in Ho Chi Minh City Stock Trading Center during the period of 2002 to 2007, applying

GARCH(1,1) model to reduce drawbacks in OLS method, the findings found the existence of herding in this emerging market; however; there is no evidence for

asymmetric effect in this case In conformity with these above authors, Kallinterakis (2007) investigates herding in Vietnamese stock market applying Hwang and Salmon (2004) model using market direction and market volatility as control variables Then,

the cross-sectional standard deviation of all individual betas was used as the input for the estimation of the herding measure If these betas have smaller value, herd exhibits Finally, he finds evidence supporting the existence of herding in Vietnamese stock market

There are other studies doing research regarding herding behavior in

Vietnamese stock market such as Tran (2010) and Nguyen (2013) Both authors apply indirect approach relied on aggregate price and market activity data to test for the presence of herding behavior In spite of being conducted over different periods, both studies find the empirical evidence of this trend in Vietnamese stock market However, when testing for the level of this phenomenon in rising and falling market, there are somewhat different Namely, Tran’s findings (2010) indicate that degree of herding behavior in up market is considerably greater than that in down market whereas

findings obtained from research of Nguyen (2013) conversely conclude that the level

of this phenomenon is more dominant in declining market than in rising market

Besides, Nguyen (2013) proceeds to divide the analyzed period into four sub-stages as well as categorize stocks into two types of large market capitalization stocks and small market capitalization stocks to test for the existence and the level of this tendency His final conclusion is that herding behavior is present in all sub-periods with different levels depending on the development characteristic of each phase and there is no distinction in herding degree in terms of large or small market capitalization

Table 1.1: A summary of empirical evidence on herding behavior

Panel A: Studies showing weak or no evidence of herding behavior

Lakonishok et al

(1991)

US Daily data from 1985 to 1989 LSV No herd by fund managers except in

smaller stocks, but no destabilizing influence of the stock prices

Christie and

Huang (1995)

7/1962 to 12/1988

CH The results indicate the absence of herd for

both daily and monthly returns

Chang et al

(2000)

US, Hong Kong, Japan, South Korea, and Taiwan

Daily data:

US: 1/1963-12/1997 Hong Kong: 1/1981 – 12/1995

Japan/Taiwan: 1/1976 – 12/1995

Southe Korea: 1/1978 – 12/1995

CCK No evidence of herd is found in the US and

Hong of Kong, partial evidence of this trend is found in Japan It is noticed that the significant evidence of herd has been found

in South Korea and Taiwan

Herd is not found in Chinese stock market but asymmetric effect is found; namely, herd is higher in up market than down market

Houda and

Mohamed (2013)

Africa, Asia, Europe and the US (28 countries)

Daily data from 1/2006 to 2/2009

CH, CCK and E-GARCH model

There is no herding behavior found over the sample; however, it is found that investors have asymmetric reaction to the good and bad news when separating the market into the upturns and the downturns

Panel B: Studies showing the evidence of herding behavior

Study has been summarized the impact of institutional investors herding on the stock market return

Wermers (1999) US Quarterly data from 1975 to

1994

LSV Study found the presence of herding

behavior in US mutual fund industry over the period studied

Hwang and

Salmon (2001)

US, UK and South Korea

Monthly data from 1/1990 to 10/2000

HS Study suggests that developed markets such

as US and US show less herding behavior than emerging markets such as South Korea Caparrelli et al

(2010)

Italy Daily data from 9/1988 to

1/2011

CH, CCK and HS

Herd is present during extreme market conditions both in terms of sustained growth rate and high stock levels according

to CH’ model and interpretation

Zhou (2007) China Daily data from 1993 to 2002 HS Zhou documents existence of herding

behavior in both the Shanghai and Shenzhen A-share and B-share markets Tan et al (2008) China Daily data from 7/1994 to

12/2003

CCK and modified CCK

Paper found the presence of herd within the Chinese stock market as well as the

asymmetric effect in terms of stock returns, trading volume and volatility

Zhao (2011) China Daily data from 1/2004 to

7/2007

Var modelling

Herding behavior has been observed in the Chinese stock market

Economou,

Kostakis and

Phillipas (2010)

Greece, Italy, Portugal and Spain

Daily data from 1998 to 2008 CCK The paper investigates the presence of her

during the global financial crisis of 2008 as well as examining the existence of

asymmetric herding behavior associated with market returns, trading volume and return volatility

There is the existence of herd in the Athens stock market

Choi and Skiba

(2015)

41 countries in over the world

Quarterly institutional holdings data from 1999 to

2010

Sias The paper documents that herding behavior

in international market is widespread

Panel C: Studies showing the evidence of herding behavior in Vietnamese stock market

Tran and Truong

(2011)

Vietnam Daily data from 3/2002 to

7/2007

CCK Herding behavior exists in Vietnamese

stock market and no asymmetric effect is found in this case

CCK Paper found the presence of herd and

asymmetric effect as well Specifically, herding behavior in up market is

considerable than that in down market Nguyen (2013) Vietnam Daily data from 2004 to

3/2012

CCK Paper found evidence of herding presence

and its level in down market is greater than

in up market In addition, no difference is found in terms of market capitalization stocks

(Source: Synthesized by the author) Note: LSV, CH, CCK and HS refers to Lakonishok-Schleifer-Vishny Model; Christie and Huang Model; Chang, Cheng and Khorana Model and Hwang and Salmon Model

2.3 Measuring herding in financial markets

Generally speaking, empirical methodologies to measure herding behavior can

be classified into two main categories: studies that rely on micro-data or proprietary data and investigate whether specific investor types herd and the other are studies that rely on aggregate price and market data and investigate whether herd presents or not based on the market consensus Lakonishok et al (1992) and Sias (2004) are the

representatives of two of the most commonly used measures of the former which be presented firstly; then this study proceeds with a discussion of two of the most

commonly used the latter (Christie and Huang, 1995; Chang et al, 2000)

A common herding measure of institutional investors is proposed by

Lakonishok et al, (1992; hereafter LSV) The notion behind their metric is that herding behavior is concluded to exist at the level of individual stocks if money managers have

a tendency to disproportionately buy (sell) an individual stock (i.e end up on the same side of the market) They compute herd as the fraction of net buyers (money managers who increase their holding in a stock during a given quarter) to the total money

managers who trade that stock excluding an adjustment factor that declines as the number of money managers active in that stock rises The LSV herding measure (H) is calculated as:

H(i) = | 𝐵(𝑖)

𝐵(𝑖)+𝑆(𝑖)− 𝑝(𝑡)| − 𝐴𝐹(𝑖) (1)

In the above specification, B(i) is the number of money managers who are net buyers, S(i) is the number of money managers who are net sellers (who decrease their holding), p(t) is the expected proportion of money managers buying in that quarter relative to the number active, AF(i) (is called the adjustment factor) is the expected value of |B/(B+S) – p| under the null hypothesis of no herding LSV point out that, for any stock, the more declining the AF is, the more rising the number of money

managers active in that stock is

LSV conclude that if herd exists, there should be significant cross-sectional variation in this measure On the other hand, if no herd exists, the expected value of this metric should not vary from period to period

In the same spirit, Sias (2004) argues that the proportion of institutional

investors buying this quarter and last quarter will covary with each other, if

institutional investors herd or follow their own trades into and out of the same

securities; thus, herd can be evaluated by estimating the cross-sectional correlation between demand for an asset by institutional investors last quarter and demand for an asset by institutional investors this quarter The calculation begins with an estimation

of every institutional investors’ position in every asset as a ratio of asset’s shares outstanding at both the beginning and the end of each quarter If an institutional

investors increases (decreases) ownership in the stock, this investor is defined as a buyer (seller) For each stock quarter, the portions of investors that are buyers are estimating This ratio is estimated as in the following equation:

Raw∆𝑘,𝑡= 𝐵𝐼𝑘,𝑡

𝐵𝐼𝑘,𝑡+ 𝑆𝐼𝑘,𝑡 (2)

Where BI is the number of institutional buying asset k during quarter t, SI is the

number of institutional selling asset k during quarter t Sias then standardizes the

fraction of institutional investors buying asset k in quarter t in order to allow

aggregation over time and comparison for different market capitalizations and

investors types as follows:

∆𝑘,𝑡= 𝑅𝑎𝑤∆𝑘,𝑡− 𝑅𝑎𝑤∆𝑘,𝑡

𝜎(𝑅𝑎𝑤∆𝑘,𝑡) (3) Sias (2004) proceeds to evaluate the cross-sectional regression as below:

∆𝑘,𝑡= 𝛽𝑡∆𝑘,𝑡−1+ 𝜀𝑘,𝑡 (4) The notion is that “If institutional investors follow each other into and out of the same securities (herd), or if individual institutional investors follow their own last-

quarter trades, then the fraction of institutions buying in the current quarter will be positively correlated with the fraction of institutions buying in the previous quarter” (Sias, 2004, p 172) The key difference between the LSV and the Sias measure is that while the former test indirectly for the cross-sectional temporal dependence within periods, the latter is a direct test of whether institutional investors follow each other’s trades during the following periods (Spyrou, 2013)

Other different approaches which are the most commonly used are proposed by Christie and Huang (1995) (hereafter CH) and Chang et al (2000) (hereafter CCK) using stock return as a data set to identify herding behavior CH suggest that overall market conditions are the element which major of market participants rely on before making any investment decision According to rational asset pricing model, these two authors insist that return dispersion will rise with the absolute value of market return throughout conventional phases, because individual investors use their own various private information as a basis for trading in stock market Nevertheless, during periods

of tremendous market movements, individuals have a tendency to ignore their own beliefs and incline to form investment decision based on majority in the market Under these conditions, individual stock returns lean toward bunching around the overall market return Therefore, CH suggest that herding will be more dominant during

periods of market stress, which is defined in terms of extreme returns The following equation is used in their herding test:

𝑅𝐷𝑡 = 𝛼 + 𝛽𝐿𝐼𝑡𝐿 + 𝛽𝑈𝐼𝑡𝑈+ 𝜀𝑡 (5) where

𝑅𝐷𝑡 is the return dispersion at time t

𝐼𝑡𝐿 = 1, if the return on the market for time period t lies in the extreme lower tail of the returns distribution, and zero otherwise

𝐼𝑡𝑈 = 1, if the return on the market for time period t lies in the extreme upper tail of the returns distribution, and zero otherwise

In order to quantify the return dispersion, CH proposed the use of cross

sectional standard deviation (CSSD) method, which is defined as:

𝐶𝑆𝑆𝐷𝑡 = √∑ (𝑅𝑖,𝑡− 𝑅𝑚,𝑡)

2 𝑁

𝑖=1

(𝑁−1) (6) where N is the number of firms in the portfolio, 𝑅𝑖,𝑡 is the observed stock return of firm i at time t and 𝑅𝑚,𝑡 is the cross – sectional average stock of N returns in the

aggregate market portfolio at time t According to the model, if herd exists, return dispersion will be smaller during period of market stress because investors will make similar decision Thus, statistically significantly negative values of 𝛽𝐿 and 𝛽𝑈 in Eq (5) would indicate the presence of herding Hwang and Salmon (2004) who employ the cross-sectional dispersion of asset sensitivity to various fundamental factors rather than asset returns propose a comparable methodology This measure can discriminate between price adjustments to information about fundamentals from herding behavior due to shifts in market sentiment

However, Christie and Huang (1995) method exists a few drawbacks Firstly, the definition of extreme return is not objective CH’s study used a value of 1% and 5% as the cut - off points to define the extreme market movements As such, to be considered as an extreme movement, an observation of market return must lie in the first and fifth lower or upper percentile of the market return distribution (Garg & Gulati, 2013) In reality, the feature of return distribution may change over time and depend on investors’ opinion, even differ from each of them Secondly, CH model recognizes herding under the condition of extreme return only However, herding behavior may be present at the entire return distribution and become more dominant during period of market stress (Chiang et al, 2010)

An alternative model to Christie and Huang model to test for herding was

proposed by Chang et al (2000) These two authors claimed that CH approach

requires defining what is meant by market stress in order to find evidence of herding, which is a strict test; therefore, they construct a methodology in spirit to the CH

approach They argue that if investors have a tendency to follow aggregate market activity during periods of large price swing, the linear and increasing relation between dispersion and market return will no longer hold; instead of that, it becomes non-

linearly increasing or even decreasing They develop a non-linear regression

specification to estimate the relation between the cross-sectional absolute deviation of stock returns and the market return Beginning with the conditional version of CAPM model to define the expected cross-sectional absolute deviation of stock return

(ECSAD) in period t, they rewrite the equation as follows:

𝐸𝐶𝑆𝐴𝐷𝑡 = 1

𝑁∑𝑁𝑖=1|𝛽𝑡− 𝛽𝑚| 𝐸𝑡(𝑅𝑚− 𝛾0) (7)

In (7), Rm is the return on the market portfolio, 𝛾0 is return on the zero- 𝛽 portfolio, 𝛽𝑚 is the systematic risk of an equally weighted market portfolio, and 𝛽𝑡 is the systematic risk on the asset i CCK then demonstrate that the increasing and linear relation between dispersion and time-varying market expected return is:

𝐶𝑆𝐴𝐷𝑡 = 𝛾0+ 𝛾1|𝑅𝑚,𝑡| + 𝛾2𝑅𝑚,𝑡2 + 𝜀𝑡 (10) 𝐶𝑆𝐴𝐷𝑡 is a cross – sectional absolute deviation and constructed as a method to measure return dispersion, which is calculated by the following formula:

𝐶𝑆𝐴𝐷𝑡 = 1

𝑁∑𝑁𝑖=1|𝑅𝑖,𝑡− 𝑅𝑚,𝑡| (11)

where 𝑅𝑚,𝑡 is the return of market portfolio and 𝑅𝑖,𝑡 is the return of stock i at time t It should be noted that both |𝑅𝑚,𝑡| and its quadratic form 𝑅𝑚,𝑡2 are appeared as

independent variables in Eq (10) Chang et al (2000) assert that under normal

conditions, the relationship between return dispersion and market volatility, as

determined by the rational pricing model, is linear As such, an increase in the absolute value of the market return will lead to a rise in the dispersion of individual investor return However, the notion behind this approach is that during periods of relatively large price movement, if participants incline to make decision based on aggregate market behavior, such increasingly linear relationship no longer holds; instead, is more likely to be non-linear increasing or even decreasing The point should be noted here is that CSAD is not the metric used to test for the existence of herd, the fact that herd is identified through the relationship between CSAD and the market return

CCK proceed to run these following regressions in order to examine whether the degree of herding behavior is asymmetric in rising and falling markets:

𝐶𝑆𝐴𝐷𝑡𝑈𝑃 = 𝛼 + 𝛾1𝑈𝑃|𝑅𝑚,𝑡𝑈𝑃| + 𝛾2𝑈𝑃(𝑅𝑚,𝑡𝑈𝑃)2+ 𝜀𝑡 (10)

𝐶𝑆𝐴𝐷𝑡𝐷𝑂𝑊𝑁 = 𝛼 + 𝛾1𝐷𝑂𝑊𝑁|𝑅𝑚,𝑡𝐷𝑂𝑊𝑁| + 𝛾2𝐷𝑂𝑊𝑁(𝑅𝑚,𝑡𝐷𝑂𝑊𝑁)2+ 𝜀𝑡 (11)

In summary, CCK suggest that during periods of extreme market conditions, if investors tend to herd toward the market consensus, a negative non-linear relation between CSAD and the average market return should exit which is performed by the statistically significant negative coefficient of the quadratic term

The first category of herding measure which is based on trading pattern of a group of investors is later applied by many researchers but nearly conduct in countries with developed stock market However, gathering data of institutional investors in immature and developing Vietnamese stock market is more difficult than in other developed markets; thus, this method is hardly applied Meanwhile, the second

category measures herd indirectly depending on return dispersion of individual stocks compared with market return, which these two data are easily to collect because they

are always available Besides, CH method existing some drawbacks makes the

application of this model to Vietnamese stock market inappropriate The relatively short history of this market is one of difficulties in collecting data; thus, in identifying the occurrence of extreme return In addition, stock market in Vietnam is evaluated as drastically fluctuated; therefore, using the same level of dispersion as other countries can lead to inaccurate regression results For that reason, the study applies CCK’s model to analyze level of return dispersion as an approach to measure herding

behavior in Vietnamese stock market

2.4 Hypothesis development

In recent years, Vietnamese economy has experienced notable economic growth and gradually integrated with the global market but its financial market has been still small and in process of development Being established in 2000 with only two stocks, the Vietnamese stock market has matured to consist of 192 listed companies in 2006 (Tran and Truong, 2011) and till now it includes 308 stock stickers In addition to its limited size and number of participants, the Vietnamese stock market is governed by incompletely improved regulatory environment which is likely to cause a lack of

transparency This could result in a greater likelihood of making investment decision based on market consensus among investors

Moreover, relatively high degree of government intervention in equity market operation with the imposition of many restrictions through monetary policy of central bank as an instrument may limit investment opportunities of market participants Meanwhile, there are few alternatives for investors as the bond and the real estate markets are underdeveloped As a result, investors are more likely to speculate on Vietnamese stock market which may lead to a larger price movement This potentially results in a tendency of herding behavior

Given the structure of Vietnamese stock market, investors are more likely to imitate others’ action to feel more secure This phenomenon is also similar to the tendency of investors in many other emerging markets which are confirmed the

herding presence by Chang et al (2000), Hwang and Salmon (2004) In accordance with findings of some authors who do the research topic regarding herding behavior in Vietnamese stock market such as My and Truong (2011), Kallinterakis (2007), Tran (2010) and Nguyen (2013), the author also assumes the existence of herding behavior

in Vietnamese stock market Therefore, the following hypothesis is tested:

Hypothesis 1: Herding behavior exits in the Vietnamese stock market

Apparently, direction of the market seems to impact on investor behavior Several studies have found that there are some changes in the level of herding under different market circumstances, conditional on the market is rising or declining (Tran

& Truong, 2011) In the study context, it is interesting to examine whether herding behavior presents an asymmetric effect on days when the market is up vis-à-vis days when the market is down Therefore, it is assumed that there is distinction in degree of herding bahavior in case of market ups and downs This argument leads to the second hypothesis as follow:

Hypothesis 2: There is difference in herding level between rising market and falling market.

CHAPTER 3: RESEARCH METHODOLOGY 3.1 Data collection and sample description

In order to detect the presence of herd in Vietnamese stock market, the author used daily closed price of VN-Index of all stocks listed in Ho Chi Minh City Stock Exchange (HSX) as the research data over the period from 1/2005 to 22/4/2015 The daily stock price was secondary data which were collected in the website of HSX and included 308 companies in total at the end of 22/4/2015 However, nine stock stickers were eliminated from the entire sample because of the lack of data caused by newly listed on the Exchange or intermittent data caused by company operation After the data collection, collected sample included 299 stocks equivalent to 299 firms which provided 2568 daily observations over period studied

In addition, the author also divided the entire sample into two separated periods: before and after the global financial crisis Specifically, the period of 2005-

sub-2007 was pre-crisis which provided 749 observations and the period of 2008-2015 was the interval of during and after global financial crisis (hereafter post-crisis) which provided 1819 daily observations

The reason for the division into two sub-periods helped to provide more

accurate insight regarding the nature of herding bahavior in Vietnamese stock market before and after influenced by the global financial crisis which began from the sub-prime crisis in the US

Table 3.1: Summary of data observations used in the study

Period analyzed Daily observations Entire period (2005 - 4/2015) 2568

Pre-crisis (2005 - 2007) 749

Post-crisis (2008 - 4/2015) 1819

(Source: Synthesized by the author)

3.2 Regression model for testing the hypotheses

3.2.1 Regression model for testing the presence of herding bahavior in

Vietnamese stock market:

As mentioned above, the author used the model in spirit of Chang et al (2000)

to test for the existence of herd in Vietnamese stock market In this part, CCK

approach is presented briefly regarding the detection of herding over the entire

distribution of market returns The herding equation is set as followed:

𝐶𝑆𝐴𝐷𝑡 = 𝛾0+ 𝛾1|𝑅𝑚,𝑡| + 𝛾2𝑅𝑚,𝑡2 + 𝜀𝑡

If 𝛾2 <0, herd exists in Vietnamese stock market

In normal market condition, investors make their investing decision

independently so the correlation between each individual stock return is low As a result, an increase in market return can lead to an increase in stock return dispersion because each asset has its own sensitivity to market return; as such, CSAD and Rm,t go with the same direction Therefore, if 𝛾2 is greater than 0 and significantly statistical,

no evidence of herd found in Vietnamese stock market

In opposition, investors tend to react similarly during periods of large price swing This action results in an increase in correlation and a decrease in dispersion between individual stocks It means the linear relationship between CASD and Rm,t is

no longer correct; instead, it is a non-linear relation (proportionately increase and then decrease) Hence, a significantly statistical and negative 𝛾2 proves the existence of herd

The study proceeded to the next step of variables construction Regression model included independent and dependent variables which were described the

calculation as follows:

Independent variables: As mentioned earlier, this research used indirect

approach depending on return dispersion in relation with market return to detect herd;

therefore, the independent variable in this research was market return From the

original data were closed price of VN-Index, the author conducted the calculation of daily market return as the following formula:

𝑅𝑚,𝑡 = 100 𝑥 (ln(𝑃𝑡) − ln(𝑃𝑡−1)) (12) where: Pt is closed price of VN-Index at time t and Pt-1 is closed price of VN-Index at

time t-1

Dependent variable: the dependent variable in this research was return

dispersion measured by cross-sectional absolute deviation, which was expressed as:

𝐶𝑆𝐴𝐷𝑡 = 1

𝑁∑|𝑅𝑖,𝑡 − 𝑅𝑚,𝑡|

𝑁

𝑖=1

where N is the number of firms in the portfolio, 𝑅𝑚,𝑡 is the return of market portfolio

at time t and 𝑅𝑖,𝑡 is the return of stock i at time t

Similarly, Ri,t in the formula was calculated as follow:

𝑅𝑖,𝑡 = 100 𝑥 (ln(𝑃𝑡) − ln(𝑃𝑡−1)) where Pt is closed price of stock i at time t and Pt-1 is closed price of stock i at time t-1

Note that closed price of stock was the fixed price after adjustment of dividend payment such as cash dividend, stock dividend and bonus stock dividend This

adjustment helped to estimate more precisely the return that investors received as well

as reflected exact nature of volatility

3.2.2 Regression model for estimation the degree of herd in rising and falling market:

In this part, through testing of herd under different market conditions, the

research investigated whether to have difference in level of this phenomenon in up and down market or not Regression model modified by Chiang et al (2010) was used for this estimation Instead of dividing the sample into two parts, Chiang et al (2010)

added dummy variable to the model to test for the asymmetric effect The model was

as follow:

𝐶𝑆𝐴𝐷𝑡 = 𝛾0 + 𝛾1(1 − 𝐷)𝑅𝑚,𝑡 + 𝛾2𝐷𝑅𝑚,𝑡+ 𝛾3(1 − 𝐷)𝑅𝑚,𝑡2 + 𝛾4𝐷𝑅𝑚,𝑡2 + 𝜀𝑡 (14)

where D is a dummy variable and D = 1 if 𝑅𝑚,𝑡<0, D = 0 otherwise

By using dummy variable, model coefficients included two separate parts In detail, 𝛾1 and 𝛾3 express the relationship between CSAD and Rm,t on days when the market is up or D = 0 Whereas, D = 1 or when the market is down, this relationship is expressed by 𝛾2 and 𝛾4

Furthermore, 𝛾3 and 𝛾4 express the non-linear relationship between CSAD and

𝑅𝑚,𝑡 in market ups and market downs, respectively The research conducted

considering the sign of each coefficient to prove the presence of herd; moreover, the magnitude of 𝛾3 and 𝛾4 was also taken into account If 𝛾3 < 𝛾4, CSAD in up market is smaller than in down market As such, with the same level of volatility in market return, the return dispersion will strongly decrease on days in down market vis-à-vis days in up market

In conclusion, the author tested hypothesis from Chiang et al (2010) model: 𝛾3<

0 and 𝛾4 < 0 to find evidence of herd in Vietnamese stock market and considered accepting or rejecting hypothesis of 𝛾3 = 𝛾4 to investigate the asymmetry in different market conditions

3.3 Regression methodology

OLS and Quantile Regresson (QREG) were two approaches chosen for model estimation For the former, this was a very popular method used in econometrics so the author presented the research process conducted in this study when using OLS instead

of introducing clearly about this method For the latter, the author would present basic theory, equation foundation and regression model used to test for the hypothesis

3.3.1 Research process

Step 1: Descriptive statistics was used to describe basic characteristics of data set Some features such as mean, min, max, standard deviation,… were described preliminarily The purpose of this step not also built a foundation for the subsequent quantitative analysis but also provided an overview of the observed data series

Step 2: Test for correlation between all variables was conducted This step showed the basic overview regarding the correlative relationship between independent and dependent variables

Step 3: Testing hypothesis by investigating the relationship between

independent and dependent variables applying CCK’s model for the presence of herd and modified model by Chiang et al (2010) for herding in different market condition

If herd was detected, Wald test was performed to examine the similarity in the level of herd in up and down market

Step 4: Hypothesis test used Quantile Regression as a new measure to detect herding behavior in different percentiles

3.3.2 Quantile regression analysis

To test for the hypotheses, quantile regression was an analysis used beside OLS method because of its usefulness in measuring the dispersion, particularly in non-stable environment; thus, could lead to more efficient estimation since it was enable us

to perform the regression over the entire distribution of dependent variable (Zhou & Anderson, 2011)

Briefly defined, “quantile regression is a statistical procedure designed to

estimate conditional quantile functions” (Chiang et al, 2010 as cited in Koenker, 2005; Alexander, 2008, p.119) The linear conditional quantile function is as follow:

QYi(τIX = x) = xi′γ (15) where Yi isdependent variable, Xi is a vector of independent variables and γ is a vector

of coefficients

By minimizing weighted deviations from the conditional quantile, the formula

is substituted as follows:

𝛾̂𝑞𝑢𝑎𝑛𝑡𝑖𝑙𝑒,𝜏 = arg 𝑚𝑖𝑚 ∑𝑛𝑖=1𝜌𝜏 (𝑦𝑖 − 𝑥𝑖′𝛾 ) (16)

where the conditional distribution of the dependent variable yi is characterized

by different values of the τth quantile given xi (Koenker, 2005), and ρτ is a weighting factor called a check function For any τ ∈(0,1), a check function is defined as:

ρτ (ui)= { τ ui if ui ≥ 0

where ui =yi−xi′γ Eqs (16) and (17) imply that

𝑌̂𝑞𝑢𝑎𝑛𝑡𝑖𝑙𝑒,𝜏 = arg min {∑ τ|yi− xi′γ| + ∑i: y (1 − τ)

i >xi′

According to expression (18), by minimizing a weighted sum of absolute errors, where the quantile values control the weights, the quantile regression estimators can be achieved The quantile regression becomes the median regression if τ = 0.5 The

interrelationship between dependent and independent variables can be estimated at any specific quantile because it does not have limitation at median level (Chiang et al, 2010) Hence, the correlation between CSADt and R2m,t is tested easily

Quantile regression for measuring dependent variable CSADt and a set of

independent variables Xt, for τ quantiles are formularized as:

𝑄𝑟 (𝜏|Xt) = 𝛾0,𝜏 + 𝛾1,𝜏 (1 − D)Rm,t+ 𝛾2,𝜏 𝐷 Rm,t+ 𝛾3,𝜏(1 − D)R2m,t + 𝛾4,𝜏𝐷 R2m,t +

ετ,t

Xt: a vector of the right-hand-side variables of Eq

D: a dummy variable by setting D = 1 if Rm,t<0 and D = 0 otherwise

This formula above is useful to test for the herding behavior in case of rising and falling market condition In this situation, the coefficient 𝛾4 is expected to

significantly negative for expression of asymmetric effect as in hypotheses 2

OLS is a regression method based on the mean of the conditional distribution of stock return dispersion; thus, it is difficult to discriminate between different quantile and even cause the ignorance of herding that exists only in certain quantiles Whereas, quantile regression provides a more complete view of how this trend performs over different quantiles In addition, Barnes and Hughes (2002) argue that quantile

regression can alleviate some of the statistical problem of the standard OLS method, such as non – normal distribution, outlier sensitivity, errors in variables and omitted variable bias Many financial datasets contain one or more of these challenges so quantile regression may be useful for employing such data (Gowlland et al, 2009) It is concluded that quantile regression builds on more efficiency for estimators than OLS