Sentiment-Augmented Asset Pricing in Bursa Malaysia: A Time-Varying Markov Regime-Switching Model

Abstract

This paper examines the nonlinear effects of investor sentiment on asset pricing in Bursa Malaysia. The Fama and French three-factor model is re-augmented within a time-varying Markov regime-switching framework to investigate the three risk premiums, conditioned by four different proxies for investor sentiment (i.e. marketwide indicators). The study finds evidence that the stock returns movement of Bursa Malaysia exhibits a nonlinear two regimes pattern. Besides, changes in the investor sentiment to some extent function as a mediator in the regime switching dynamics between bear and bull market cycles in Malaysian stock returns. It is also found that an increase in positive sentiment of investors leads to a higher transition probability of regime switching during bear markets. In addition, the three risk premiums are timevariant, contingent upon the fluctuation of the proxies for investor sentiment within discrete regimes. The study finds that in general, the market premium falls when the stock market switches from bull to bear markets. On the contrary, both the size and value premiums increase when the stock market moves from bull to bear markets.

Full text

Malaysian Journal of Economic Studies Vol. 55 No. 2, 2018 285
Malaysian Journal of Economic Studies 55(2): 285–300, 2018 ISSN 1511-4554
Senment-Augmented Asset Pricing in Bursa Malaysia:
A Time-Varying Markov Regime-Switching Model
Han Hwa Goha
Lee Lee Chongb
Ming Ming Laic
Mulmedia University
Abstract: This paper examines the nonlinear eects of investor senment on asset
pricing in Bursa Malaysia. The Fama and French three-factor model is re-augmented
within a me-varying Markov regime-switching framework to invesgate the three risk
premiums, condioned by four dierent proxies for investor senment (i.e. market-
wide indicators). The study nds evidence that the stock returns movement of Bursa
Malaysia exhibits a nonlinear two regimes paern. Besides, changes in the investor
senment to some extent funcon as a mediator in the regime switching dynamics
between bear and bull market cycles in Malaysian stock returns. It is also found that
an increase in posive senment of investors leads to a higher transion probability of
regime switching during bear markets. In addion, the three risk premiums are me-
variant, conngent upon the uctuaon of the proxies for investor senment within
discrete regimes. The study nds that in general, the market premium falls when the
stock market switches from bull to bear markets. On the contrary, both the size and
value premiums increase when the stock market moves from bull to bear markets.
Keywords: Asset pricing, Bursa Malaysia, investor sentiment, time-varying Markov
regime-switching model
JEL classicaon: G120, G410, C580
1. Introducon
Extreme market volality in global nancial markets is becoming more common. Such
volale uctuaons have been observed during the Brexit, U.S. presidency elecon,
price drop in crude oil and more other world events in recent years. Chen, Tian and
Zhao (2017) portrayed 2016 as the year of global black swan events. UK’s Brexit
(leaving the European Union) vote and Donald Trump’s unexpected win in the US
presidenal elecon have caused the decline of several stock market indices in the
a Faculty of Management, Mulmedia University, Persiaran Mulmedia, 63100 Cyberjaya, Selangor Darul
Ehsan, Malaysia. E-mail:hhgoh@mmu.edu.my (Corresponding author)
b Faculty of Management, Mulmedia University, Persiaran Mulmedia, 63100 Cyberjaya, Selangor Darul
Ehsan, Malaysia. E-mail:llchong@mmu.edu.my
c Faculty of Management, Mulmedia University, Persiaran Mulmedia, 63100 Cyberjaya, Selangor Darul
Ehsan, Malaysia. E-mail:mmlai@mmu.edu.my
Arcle Info: Received 25 April 2018; Revised 12 September 2018; Accepted 20 September 2018
hps://doi.org/10.22452/MJES.vol55no2.8

286 Malaysian Journal of Economic Studies Vol. 55 No. 2, 2018
Han Hwa Goh, Lee Lee Chong and Ming Ming Lai
European nancial markets and global nancial markets over a short period of me.
The Malaysian stock market is not isolated from such external shocks and coupled
with the 1MDB scandal, investors’ confidence deteriorated while the ringgit has
been steadily declining over the past one and a half years, according to prominent
Malaysian economist Jomo Kwame Sundaram (Idris & Aziz, 2016). Up to date, no
convenonal nance theories, which perceive market to be raonal, can fully explain
the irraonalies of behaviour in the stock market. Standard asset pricing theories state
that asset prices are determined purely by investors’ unbiased cognive evaluaon
and maximisation of expected utility, and there is no role for investor sentiment
(Xu & Green, 2013). However, behavioural theories claim that investors may hold
fallacious stochasc beliefs, either with excessive opmism or pessimism, and therefore
inaccurately esmate asset value; hence asset prices to digress from their intrinsic
values (De Long, Shleifer, Summers, & Waldmann, 1990; Kumar & Lee, 2006; Lee,
Shleifer, & Thaler 1991). Posive senment spurs investors to be more condent about
their competence to assess situaons and thus more unhesitant to take risks; and vice
versa (Kuhnen & Knutson, 2011).
Asset valuaons are vital for investors to determine the value of a rm and thus
make an investment strategy. The capital asset pricing model (CAPM), propounded by
Sharpe (1964) and Lintner (1965), claims that there is a posive eect of the market
risk on the risk premium of nancial assets. Nonetheless, CAPM does not reect the
share return actually obtained on the equity market (Reinganum, 1981; Rosenberg,
Reid, & Lanstein, 1985). As such, Fama and French three-factor (hereaer F-F) model
(1993) is designed to augment the convenonal CAPM by including two factors of rm
size and book-to-market value. As evidenced by them in the US stock markets, the
returns of stocks of small capitalisaon and high book-to-market values are higher than
those of the CAPM. There has subsequently been extensive empirical work carried out
to evaluate the soundness of the F-F model, among others, Lawrence, Geppert and
Prakash (2007); Simpson and Ramchander (2008). In recent years, Fama and French
(2015) further developed a 5-factor model to explain average returns. However, the
two new factors (protability and investment) they have added to improve explanatory
power are relavely recent discoveries which are premature and the research of these
factors in dierent markets and me periods is sll limited. Thus far, Fama-French
5-factor model has been tested to Indian market (Harshita & Yadav, 2015) and Japan
market (Kubota & Takehara, 2017) besides the U.S.
Investors’ senment is ignored and not measured in tradional nance theories.
Nevertheless, recent studies have shown that markets are senment driven (Baker,
Wurgler, & Yuan, 2012; Yang & Zhang, 2014). Investors’ senment drives asset values
away from its fundamentals as evidenced by re-occurrence of market anomalies
and nancial turmoil. It is therefore important to quanfy investors’ senment and
capture it in asset valuaon in order to have ecient capital allocaon and opmal
cost of capital. However, very lile research have been done in modelling asset pricing
with investors’ senment in emerging markets which are perceived to be young and
underdeveloped with more noise traders compared to developed markets.

Malaysian Journal of Economic Studies Vol. 55 No. 2, 2018 287
Senment-Augmented Asset Pricing in Bursa Malaysia
Unlike those mature markets in developed countries, the Malaysian stock market,
which is an emerging market, is inecient in the weak form and signicantly overreacts
to surprises in economic crisis and polical events (Ali, Nassir, Hassan, & Abidin, 2010).
The study of Lai, Tan and Chong (2013) also showed evidence with surveyed data that
both the instuonal and retail investors in Malaysia tend to overreact. The role of
investor senment is pernent in explaining market ineciency. Posive senment
encourages investors to take more risk as they have more condence whereas negave
senment will have the reverse impact (Kuhnen & Knutson, 2011). The linkage between
senment and stock return is hardly constant under dierent market circumstances
(Karakatsani & Salmon, 2008).
In the literature, investor senment consists of two measures, direct and indirect.
While indirect measures are proxies from market data, direct measures are the
surveyed data with direct contact with investors such as the American Associaon of
Individual Investors (AAII) and Investor Intelligence (II) which are only available in the
United States. We use the indirect measures in our study. Although there are a variety
of proxies for senment measures suggested by the behavioural pricing literature,
Baker and Wurgler (2006) stated that there are no uncontroversial and definitive
measures of senments. Our choice of senment indicators has, to a large extent,
been constrained by lack of data availability. As a result, we employ the following
proxies: number and return of inial public oerings (i.e. NIPO and RIPO) as employed
by Baker and Wurgler (2007), rao of advancers to decliners (ADR) by Brown and
Cli (2004) and consumer senment index (CSI) by Chen (2011), Fisher and Statman
(2003), Ho and Hung (2012), Hsu, Lin and Wu (2011), Jansen and Nahuis (2003), and
Schmeling (2009).
The role of senment on stock prices remains vague in theory and disputable in
empirical tests. This study is therefore crucial and mely in areas of asset valuaon and
porolio management. In this study, we aim to ulise a me-varying Markov regime-
switching model to examine the risk premiums of the F-F model when we employ four
dierent market senment proxies separately as a mediator of regime switches. We use
the F-F model in our study as it is globally recognised as a useful descripon of stock
returns. It is hoped that the empirical outcomes of this study oer an insighul view
and are able to help investors in the stock market assess risk premiums and thus stock
returns all the more precisely.
Overall, the contributions of this paper to the behavioural finance literature
especially from the Malaysian perspective are in the following aspects. Firstly, we
use a me-varying Markov-switching model to examine whether investor senment
changes play a mediang role in explanaon of regime-switching dynamics between
bear and bull market cycles in Malaysian stock returns. Besides, we unearth the
nonlinear associaon between stock return and three risk premiums of the F-F model.
Lastly, we test directly the market senment proxies individually instead of a single
composite index of senment as shown in most previous researches. The reasons for
doing so are twofold: rst to disclose dissimilaries in senment eects; and second,
to circumvent the replicaon problem over me that is usually an issue whenever the
principal components are employed to work out a composite index. The remainder of
this paper is structured as follows. The following secon presents the literature review

288 Malaysian Journal of Economic Studies Vol. 55 No. 2, 2018
Han Hwa Goh, Lee Lee Chong and Ming Ming Lai
while secon 3 explains the data and methodology involved. Secon 4 illustrates and
discusses the empirical results and Secon 5 concludes the ndings.
2. Literature Review
The standard asset pricing theory explains that prices (and returns) of stocks should be
equal or close to their expected fundamental values and the returns rely only on the
cross-seconal systemac risks (Fama & French, 1993, 1995, 2012, 2015). Fama and
French (1993) ulised the overall market risk premium, size factor and book-to-market
rao to account for excess stock returns and they further developed a ve-factor model
to explain average returns in 2015.
A substanal body of empirical evidence in nance however reveals that stock
returns persistently deviate from their fundamentals, and further discloses that investor
senment aects the stock returns (Baker & Wurgler, 2006, 2007; Kim & Ha, 2010; Liao,
Huang, & Wu, 2011; Yang & Zhang, 2014). Behavioural theories postulate that investors
may develop erroneous beliefs and/or behavioural bias, either with undue opmism
or pessimism, about the future distribuon of returns on assets and subsequently
inaccurately assess the asset values, causing anomalies of prices from their intrinsic
values (De Long et al., 1990; Kumar & Lee, 2006; Lee et al., 1991). Only limited studies
have incorporated behavioural dimension in asset pricing models, among others,
Shefrin and Statman (1994), Statman, Fisher and Anginer (2008), Xu and Green (2013)
and Yang and Li (2013).
Prior empirical studies show that there is a significant influence of investor
senment on stock returns due to limited arbitrage acvies and investor irraonality.
Ho and Hung (2009) indicated that the effectiveness of asset pricing models in
explaining stock returns can be enhanced when investor senment is incorporated into
modelling the dynamics of risk exposures. The impacts of investor senment on stock
prices are studied by researchers in some countries who employ regression models with
aggregate or industrial-level data (Chen, Chen, & Lee, 2013; Schmeling, 2009; Zhang
& Semmler, 2009). Their empirical results have pointed to investor senment as the
cardinal agent of nonlinear and asymmetric stock returns. The link between senment
and stock returns is geng more complicated as suggested by mounng evidence. The
minimal eects of senment on the stock return may well be varying between regimes
of high and low senments (opmism and pessimism). Besides, McMillan (2003) and
Lee and Chiu (2012) suggested that owing to the existence of market fricon, cost of
transacon, as well as the interacve behaviour of informed and ‘noise’ investors, the
nancial markets may exhibit behaviour of nonlinearity.
A review of literature reveals that most of the studies done are focussed on
developed markets which have dierent market structure and characteriscs as com-
pared to emerging markets. Unlike those mature markets in developed countries, the
Malaysian capital market (an emerging economy) overreacts to surprises in domesc
polical events and is inuenced by external shocks such as SARS outbreak, September
11th terrorist aacksas cited by Ali et al. (2010). Schmeling (2009) revealed that the
impacts of senment on stock returns are more substanal for a naon which has less
market integrity or less ecient regulatory instuons.

Malaysian Journal of Economic Studies Vol. 55 No. 2, 2018 289
Senment-Augmented Asset Pricing in Bursa Malaysia
3. Data and Methodology
3.1 Data
In conducng empirical esmaon, we collect the data on stocks which have been listed
for at least a year and traded on main board of Bursa Malaysia over the period January
2001 to December 2015. The market senment proxy data is also collected over the
same period. The rm and senment proxy data is largely taken from the data service
providers, i.e. Datastream and Bloomberg Inc. The 1997 Asian nancial crisis had caused
the stock markets of South East Asian region to collapse. Although there is no specic
date of full economic recovery, Angabini and Wasiuzzaman (2010) believed that the
economy almost recovered by the middle of year 2000. Hence, the invesgaon of
stock returns performance by adapng Fama and French three-factor model (1993) only
starts in January 2001.
3.2 Senment Proxies
For lack of data availability, our choice of market senment proxies has to a large
extent been constrained by the following: number and return of inial public oerings
(i.e. NIPO and RIPO), rao of advancers to decliners (ADR) and consumer senment
index (CSI).
It is oen assumed the underlying demand for inial public oerings (IPOs) is highly
sensive to investor senment. The prices of IPOs are normally put up aer having
consultaon with investment bankers who are well-versed in market situaons. Yet,
the puzzle of IPOs being under-priced to such a great extent sll remains unresolved.
Meanwhile, the volality of average rst-day returns can be strongly linked to the
number of IPOs and other proxies of senment that are not fundamentally related.
Brown and Cli (2004) stated that the relave market robustness measured in
buying–selling imbalance can be reected through the rao of advancers to decliners
(ADR). The number of advancing issues (advancers) is referred to as the total number
of shares in the Malaysian stock market whose closing prices at month-end are higher
than their opening prices at the beginning of the month, while the number of declining
issues (decliners) measures the opposite. In general, a larger value of the ADR in-
dicates a broader base of an upward trend, and thus a stronger underlying senment
of the market.
CSI is basically a fundamental indicator of economic senment. It has been issued
by the Malaysian Institute of Economic Research (MIER) since 1988. This index is
developed from quarterly surveys on over 1,200 domesc households. The range of
quesons in the survey covers the respondents’ both contemporary and ancipated
nancial condions, economic and job prospects as well as their purchase intenon
on houses and other major consumer durables. The CSI is also known as the consumer
condence index (CCI), which has been gaining in popularity as one of the proxies of
investor senment in the stock market. In recent years, there have been many empirical
studies carried out in developed markets showing that CCIs and stock returns are
signicantly related especially amid bear economic situaons.

290 Malaysian Journal of Economic Studies Vol. 55 No. 2, 2018
Han Hwa Goh, Lee Lee Chong and Ming Ming Lai
3.3 Empirical Model
In the traditional Fama and French model, the expected excess returns on stock
porolios are a funcon of the porolios’ risk of three factors, i.e. the market, size and
value premiums, which can be wrien as follows:
(1)
where Ri,t – RFRt is the excess stock returns while RMt – RFRt, market premium,
measures the market porolio’s value weighted excess returns. The coecient αi is
alpha of the F-F model. SMBt is the size premium, measured by the variaons in returns
between a small capitalisaon porolio and a large capitalisaon porolio. HMLt is the
value premium, measured by the returns’ dierences between a porolio of high book-
to-market (value) stocks and a porolio of low book-to-market (growth) stocks. The
term ei,t is the residual. In the F-F model, it is esmated the coecients of SMBt and
HMLt to be posive, suggesng that small stocks and value stocks carry higher risks and
hence higher expected returns as compared to large stocks and growth stocks.
Starng from January 2001, for the size factor, all stocks listed on the main board
of Bursa Malaysia are sorted in descending order according to the values of market
capitalisaon from December 2000. The stocks are separated into 2 porolios, that is,
small (S) and big (B) market values. This forms the SMB factor. For the book-to-market
value (BM) factor, the previously formed two porolios of small and big market value
are further divided into 3 porolios, respecvely, that comprise high (H), medium (M),
and low (L). As a result, there are 6 size/BM porolios produced, i.e. S/H, S/M, S/L, B/H,
B/M, B/L. The value-weighted monthly returns are then esmated over next 12 months
for these 6 porolios and becomes the HML factor. This process of porolio formaon is
reformed and repeated each year unl December 2015.
For the sake of invesgang the dynamic nonlinear associaon between stock
return and 3 risk premiums of the F-F model, we adopt the two-state Markov switching-
AR(p) where the specicaons echo the original model of Hamilton (1989) but allow
the term of constant, slope coecients to be state-dependent (also known as regime-
dependent) as follows:
(2)
where rt indicates the excess stock returns, RMt is the market premium, SMBt is the
size premium, HMLt is the value premium and Ɛt is the innovaon process. The state
independent autoregressive component with an opmal lag order p, AR(p), is employed
to render the innovaon process white noise. While , and are respecvely the
state-dependent coecients of market premium, size premium and value premium,
is a regime-varying constant term. The unobservable state variable st is a latent indi-
cator variable which can take only 2 values: either 1 or 2. It is assumed the state variable
observes a rst-order Markov chain with a transion probability matrix as follows:
(3)
 
tititittiitti eHMLSMBRFRRMRFRR,3,2,1,, 





P
i
tititStStSSt
rHMLSMBRMr
tttt
1

,
),0(...
2

Ndii
t

t
S

t
S

t
S

t
S

t
S

t
S

t
S

t
S

t
S

t
S

t
S

t
S

t
S

t
S

t
S

t
S

P=
PP
PP
11 11
22 22
1
1
−
−






Malaysian Journal of Economic Studies Vol. 55 No. 2, 2018 291
Senment-Augmented Asset Pricing in Bursa Malaysia
where and .
These transion probabilies are held to be invariant with me as in the original
model of Hamilton. Nevertheless, in order to examine whether the changes in
senment play a mediang role in the regime switching dynamics between two states
of the stock market cycles in Bursa Malaysia, we consider a me-varying transion
probability Markov-switching (hereafter TVTP-MS) model, where the transition
probabilies are described as follows:
(4)
where and . Zt is the vector of
senment proxies that aect the likelihood of regime-switching. In our study, we use
four specicaons of the Z vector, whereby each Z vector includes the lagged values of
ADR, CSI, NIPO and RIPO, respecvely. In the TVTP-MS model, the transion probability
is specied in the following logisc funcon:
and (5)
Due to the presence of two regimes or states, we obtain two separate esmates of
 – one for each regime. The regimes-switching probabilies are allowed to be varying
over me with the changes in investor senment, which are represented by four market
senment proxies.
4. Results and Discussions
4.1 Descripve Stascs
Table 1 summarises descripve stascs of stock excess returns (r), market premium
(RM), size premium (SMB), value premium (HML) and four market senment proxies.
Two variables of senment proxies (i.e. ADR and CSI) are transformed into logarithmic
form to compress the scale. Interesngly, the stock excess returns exhibit the mean
value below zero with the minimum of -22.935 and the maximum of 12.290 for the
sample period from January 2001 to December 2015. While the mean values of
1)1(
11

1tt
SSPP
2)2(
22  1tt SSPP
1)1(
11

1tt
SSPP
2)2(
22  1tt SSPP
P=
PP
PP
11 11
22 22
1
1
()
()
() ()
ZZ
ZZ
tt
tt
−
−





)11()(
11
t1ttt
ZSSPZP 

)2,2()(
22
t1ttt
ZSSPZP 

)11()(
11
t1ttt
ZSSPZP 

)2,2()(
22
t1ttt
ZSSPZP 



















2
0
,1
2
0
,1
11
exp1
exp
)(
K
KtK
K
KtK
t
CI
CI
ZP




















2
0
,2
2
0
,2
22
exp1
exp
)(
K
KtK
K
KtK
t
CI
CI
ZP




















2
0
,1
2
0
,1
11
exp1
exp
)(
K
KtK
K
KtK
t
CI
CI
ZP




















2
0
,2
2
0
,2
22
exp1
exp
)(
K
KtK
K
KtK
t
CI
CI
ZP



292 Malaysian Journal of Economic Studies Vol. 55 No. 2, 2018
Han Hwa Goh, Lee Lee Chong and Ming Ming Lai
the RM and SMB are negave, the remainder show a posive value. Kurtosis is the
measurement of the “fatness” for a tail distribuon. All the series display a posive
excess kurtosis (i.e., leptokurtic distribution) over the sample period. This implies
that all the series’ distributions have higher peaks around the mean than normal
distribuon, which causes thick tails on both sides of the distribuon. The r, RM, ADR
and CSI series indicate negave skewness while SMB, HML and another two senment
proxies (NIPO and RIPO) series indicate posive skewness. In addion, Jarque-Bera
stasc denotes the goodness-of-t on whether the sample data has the skewness and
kurtosis to match a normal distribuon. From Table 1, the Jarque-Bera stasc signies
that the normality test is rejected at 1 percent signicance level for all the series during
the sample period.
4.2 Esmaon and Diagnoscs
We begin our empirical analysis by first testing whether the TVTP-MS model with
inclusion of four separate market senment proxies (i.e. ADR, NIPO, CSI and RIPO)
in its transion equaon provides a beer characterisaon of the stock returns in
Bursa Malaysia than the linear model and the Markov switching model with fixed
regime transion probabilies (FTP-MS). For the specicaons of both the TVTP-MS
and FTP-MS models, in accordance with the Akaike Informaon Criterion (AIC) and
Schwarz Criterion (SBC), we use TVTP-MS AR(4) specicaon, which is found to be
adequate to make the residuals white noise, for ADR and NIPO. Meanwhile, for CSI
and RIPO, the specicaon of TVTP-MS AR (3) is chosen. Besides, the number of lags
K for esmates of  (i.e. four market senment proxies) is also selected based on the
Akaike Informaon Criterion (AIC) and Schwarz Criterion (SC). While the ADR and NIPO
enter the transacon equaon with 2 months lag, there is a 1 month lag for both CSI
and RIPO. Our models are in line with Schmeling’s (2009) study where the author
invesgated the relaon between investor senment and future stock returns for 18
industrialised countries and found that the predicve power of senment was most
pronounced for short-term horizons of 1 to 6 months.
Table 1. Summary stascs
r RM SMB HML ADR CSI NIPO RIPO
Mean -2.727 -2.398 -0.088 0.938 0.097 4.635 2.311 0.025
Maximum 12.290 10.770 9.765 14.047 1.798 4.821 13.000 3.782
Minimum -22.935 -20.110 -4.823 -6.166 -4.112 4.156 0.000 -0.976
Std. Dev. 5.323 4.231 2.282 2.688 0.991 0.148 2.430 0.633
Skewness -0.406 -0.500 0.744 0.663 -1.206 -1.472 1.838 1.830
Kurtosis 4.674 5.054 4.550 5.523 5.762 4.333 7.268 10.199
Jarque-Bera 25.965*** 39.131*** 34.628*** 60.949*** 100.824*** 78.319*** 237.944*** 489.119***
Observaons 180 180 180 180 180 180 180 180
Note: *** denotes signicance at 1 percent level.

Malaysian Journal of Economic Studies Vol. 55 No. 2, 2018 293
Senment-Augmented Asset Pricing in Bursa Malaysia
Comparison among three models using model selection criteria is reported
in Tables A1, A2, A3 and A4 (see Appendix) for the four market senment proxies
respecvely. All the model selecon criteria conclude that TVTP-MS model is more
superior to both the linear and the FTS-MS models across four market sentiment
proxies. In addition, Table 2 shows that the data intensely rejects at 1 percent
signicance level the linear model in favour of the TVTP-MS model for all the panels, in
accordance with standard Likelihood Rao (LR) test.
As the sample is dichotomising into regimes that show increasing stock excess
returns and decreasing stock excess returns, the Markov-switching models are able to
idenfy two regimes which can be labelled as low mean (i.e. bear market) and high
mean (i.e. bull market) in which the eects of three risk factors (RM, SMB and HML)
of the F-F model on the stock excess returns dier signicantly, as exhibited in Table
2. Interesngly, the two mean regimes idened by the model show negave values
but dierences in the magnitude. The negavity of the mean is in agreement with
the results of Lai et al. (2013) in their examinaon of the F-F model in Bursa Malaysia
between January 1996 and December 2005.
The esmaon results in Table 2 show that on the whole all the senment proxies,
with the only excepon of RIPO, do somewhat aect the probabilies of switching
between regimes. Specically, the senment proxy of NIPO has exerted an inuence on
both regime transion probabilies of the bull and the bear markets. Meanwhile, ADR
and CSI are seen to aect only the regime transion probabilies of the bull market and
the bear market, respecvely.
In high mean regime (i.e. bull market), all the separate sentiment proxies’ co-
ecients () are posive, implying that increases in investors’ posive senment would
raise the probability of being in the high mean regime (i.e. bull market). Conversely, the
negave coecient of  as exhibited by all the senment proxies in the regime of low
mean (i.e. bear market) signies that an increase in investors’ posive senment would
cause a decline in the probability of being in the low mean regime (i.e. bear market).
When comparing the absolute values of the coefficients of  for all the sentiment
proxies between the high mean regime (i.e. bull market) and low mean regime (i.e. bear
market), it is revealed that the laer is higher than the former. The result implies that
when there is an increase in posive senment of the investors, the probability of the
stock returns moving from bear market into bull market is higher than the probability
of the stock returns staying in the bull market. In other words, an increase in posive
sentiment of investors leads to a higher transition probability of regime switching
during bear market.
For the esmated market premiums as shown in Panels A to D in Table 2, the
posive market premiums are not only in line with the convenonal empirical results
of the F-F model but also varying with me. The esmated market premiums in the
bull market are signicantly larger than that of the bear market. Besides, the market
premiums shown in the four panels are all greater than one in the bull market, implying
that the stock prices in the Bursa Malaysia encounter higher volality than the market
porolio index.
With regard to the estimation results of size premium, the size premiums are
posive and signicant in both regimes, consistent with the line of reasoning in the

294 Malaysian Journal of Economic Studies Vol. 55 No. 2, 2018
Han Hwa Goh, Lee Lee Chong and Ming Ming Lai
Table 2. TVTP-MS model results (with four market senment proxies in the separate transion
equaon)
Panel A: ADR Panel B: NIPO
TVTP-MS-AR(4) model TVTP-MS-AR(4) model
Regime 1: Regime 2: Regime 1: Regime 2:
High mean Low mean High mean Low mean
Parameters Esmate Esmate Esmate Esmate
α -0.47 (0.24)* -1.85 (0.35)*** -0.72 (0.33)** -1.34 (0.29)***
βRM 1.01 (0.05)*** 0.70 (0.08)*** 1.00 (0.07)*** 0.79 (0.06)***
δSMB 0.40 (0.08)*** 1.37 (0.18)*** 0.27 (0.15)* 0.88 (0.10)***
θHML 0.41 (0.08)*** 0.63 (0.13)*** 0.24 (0.09)** 0.74 (0.09)***
ρ1 -0.14 (0.10) 0.03 (0.09)
ρ2 0.05 (0.09) 0.02 (0.08)
ρ3 0.16 (0.08)** 0.20 (0.08)**
ρ4 -0.05 (0.08) -0.14 (0.08)*
σ 0.54 (0.06)*** 0.58 (0.06)***
Regime parameters

(ADRt–2) 2.50 (1.08)** -8.61 (11.52)

(NIPOt–2) 0.46 (0.26)* -0.57 (0.32)*
Log-likelihood -357.42 -357.73
LR test 73.95 [0.00]*** 71.62 [0.00]***
J-B 3.26 [0.20] 2.04 [0.36]
Q(10) 4.02 [0.67] 9.15 [0.17]
Q2(10) 7.76 [0.65] 5.56 [0.85]
Panel C: CSI Panel D: RIPO
TVTP-MS-AR(3) model TVTP-MS-AR(3) model
Regime 1: Regime 2: Regime 1: Regime 2:
High mean Low mean High mean Low mean
Parameters Esmate Esmate Esmate Esmate
α -0.32 (0.18)* -1.57 (0.41)*** -0.34 (0.33) -1.54 (0.3 8)***
βRM 1.07 (0.06)*** 0.69 (0.08)*** 1.08 (0.07)*** 0.69 (0.08)***
δSMB 0.44 (0.09)*** 1.40 (0.18)*** 0.41 (0.10)*** 1.25 (0.17)***
θHML 0.52 (0.09)*** 0.59 (0.10)*** 0.41 (0.14)*** 0.60 (0.09)***
ρ1 0.05 (0.09) 0.02 (0.09)
ρ2 -0.08 (0.09) -0.06 (0.09)
ρ3 0.08 (0.09) 0.12 (0.10)
σ 0.54 (0.07)*** 0.54 (0.07)***

Malaysian Journal of Economic Studies Vol. 55 No. 2, 2018 295
Senment-Augmented Asset Pricing in Bursa Malaysia
F-F model that small capitalisaon stocks generate bigger returns than those of large
capitalisaon. The esmated size premiums are also found to be me-varying instead of
me-invariant. Addionally, the results show that small rms seem to be able to obtain
more size premium in the low mean regime (i.e. bear market) compared to the high
mean regime (i.e. bull market). It is probably jusable that in the midst of a negave
market senment in the bear market, small rms are able to adjust their operaonal
strategies more swily and easily than large rms.
The value premiums exhibited in Table 2 are all significantly positive in both
regimes, in line with the standard empirical analyses of F-F model and most of its
advocates that value stocks (which are of more wholesome operang structure and
system) generate higher returns than growth stocks. Again, the value premiums are
seen to be me-varying. Interesngly, the results of the esmated value premium show
that the value premiums are increasing in the low mean regime (i.e. bear market).
To put it simply, investors are more aracted to the value stocks in the situaon of
negave market senment during the bear market.
Figure 1 plots the ltered transion probabilies for TVTP-MS model with inclusion
of four separate market senment proxies in its transion equaon in the low mean
regime (i.e. bear market). When the probabilies are above 0.5, the Bursa Malaysia is
more likely to be in a bear market and vice versa. As noced in Figure 1, the TVTP-MS
model is able to well capture some major and crical episodes in the regional or global
economy such as the 2003 Iraq war, the 9.3 magnitude earthquake in Southeast Asia in
December 2004, the 2007-2009 Global Financial Crisis (GFC) as well as the 2011-2012
European Sovereign Debt Crisis.
Table 2. Connued
Panel C: CSI Panel D: RIPO
TVTP-MS-AR(3) model TVTP-MS-AR(3) model
Regime 1: Regime 2: Regime 1: Regime 2:
High mean Low mean High mean Low mean
Parameters Esmate Esmate Esmate Esmate
Regime parameters

(CSIt–1) 2.56 (3.19) -28.46 (11.81)**

(RIPOt–1) 0.21 (1.09) -1.01 (1.12)
Log-likelihood -369.22 -369.96
LR test 62.16 [0.00]*** 60.13 [0.00]***
J-B 1.42 [0.49] 1.98 [0.37]
Q(10) 10.47 [0.16] 5.29 [0.63]
Q2(10) 10.84 [0.37] 7.45 [0.68]
Notes: J-B stands for the Jarque-Bera normality test, Q(10) and Q2(10) indicates the Box-Pierce serial
correlaon test for residuals and squared residuals, respecvely. Figures in parentheses and square
brackets are standard errors and ρ-values, respecvely. ***, ** and * denote stascal signicance at
1%, 5% and 10% levels, respecvely. The number of lag K for esmates of

is selected based on the
Akaike Informaon Criterion (AIC) and Schwarz Criterion (SC).

296 Malaysian Journal of Economic Studies Vol. 55 No. 2, 2018
Han Hwa Goh, Lee Lee Chong and Ming Ming Lai
Figure 1. Filtered transion probabilies for TVTP-MS model (with four separate market senment proxies)
Panel A: ADR Panel B: NIPO
Panel C: CSI Panel D: RIPO

Malaysian Journal of Economic Studies Vol. 55 No. 2, 2018 297
Senment-Augmented Asset Pricing in Bursa Malaysia
5. Conclusion
In this study, we empirically look into the nonlinear eects of investor senment on
asset pricing in Bursa Malaysia. In parcular, we employ two-state TVTP-MS model for
re-examining the F-F model’s three risk premiums when four dierent market senment
proxies are ulised as a mediator of regime switches. To the best of our knowledge,
our study is the rst in Malaysian literature to focus explicitly on the role of investor
sentiment using market sentiment proxies individually in explaining stock market
regimes and regime switches.
On the whole, all the model selecon criteria conclude that the two-state TVTP-MS
model is more superior to both the linear and the FTS-MS models across four market
senment proxies. The empirical results of the TVTP-MS model indicate that to some
extent changes in senment play a mediang role in the regimes-switching dynamics
betweenbear and bull market cycles in Malaysian stock returns. Moreover, our study
shows that an increase in positive sentiment of investors would result in a higher
transion probability of regime switching during bear markets. It is also found that
the lead mes from the four market senment proxies to the stock returns in Bursa
Malaysia are on average within one to two months.
In addion, the results from our TVTP-MS model illustrate that the F-F model’s
three risk premiums are me-varying. The market premium falls as the stock market
switches from bull to bear periods. On the contrary, both the size and value premiums
increase when the stock market moves from bull to bear periods. It is therefore well
advised that investors in Bursa Malaysia should adjust their investment portfolios
conngent on each period’s esmated risk premiums.
References
Ali, N., Nassir, A.M., Hassan, T., & Abidin, S.Z. (2010). Short run stock overreacon: Evidence from
Bursa Malaysia. Internaonal Journal of Economics and Management, 4(2), 319-333.
Angabini, A., & Wasiuzzaman, S. (2010). GARCH models and the nancial crisis – a study of the
Malaysia stock markets. Internaonal Journal of Applied Economics and Finance, 5(3), 226-
236. hps://doi.org/10.3923/ijaef.2011.226.236
Baker, M., & Wurgler, J. (2006). Investor senment and the cross-secon of stock returns. Journal
of Finance, 61(4), 1645-1680. hps://doi.org/10.1111/j.1540-6261.2006.00885.x
Baker, M., & Wurgler, J. (2007). Investor senment in the stock market. Journal of Economic
Perspecves, 21(2), 129-152. hps://doi.org/10.1257/jep.21.2.129
Baker, M., Wurgler, J., & Yuan, Y. (2012). Global, local and contagious investor senment. Journal
of Financial Economics, 104(2), 272-287. hps://doi.org/10.1016/j.jneco.2011.11.002
Brown, G.W., & Cli, M.T. (2004). Investor senment and the near-term stock market. Journal of
Empirical Finance, 11(1), 1-27. hps://doi.org/10.1016/j.jempn.2002.12.001
Chen, S-S. (2011). Lack of consumer condence and stock returns. Journal of Empirical Finance,
18(2), 225-236. hps://doi.org/10.1016/j.jempn.2010.12.004
Chen, M-P., Chen, P-F., & Lee, C-C. (2013). Asymmetric eects of investor senment on industry
stock returns: Panel data evidence. Emerging Markets Review, 14(March), 35-54. hps://doi.
org/10.1016/j.ememar.2012.11.001
Chen, X., Tian, Y., & Zhao, R. (2017). Study of the cross-market eects of Brexit based on the
improved symbolic transfer entropy GARCH model – An empirical analysis of stock-bond
correlaons. PLoS ONE, 12(8): e0183194. hps://doi.org/10.1371/journal.pone.0183194

298 Malaysian Journal of Economic Studies Vol. 55 No. 2, 2018
Han Hwa Goh, Lee Lee Chong and Ming Ming Lai
De Long, J.B., Shleifer, A., Summers, L.H., & Waldmann, R.J. (1990). Noise trader risk in nancial
markets. Journal of Polical Economy, 98(4), 703-738. hps://doi.org/10.1086/261703
Fama, E.F., & French, K.R. (1993). Common risk factors in the returns on stocks and bonds. Journal
of Financial Economics, 33(1), 3-56. hps://doi.org/10.1016/0304-405X(93)90023-5
Fama, E.F., & French, K.R. (1995). Size and book-to-market factors in earnings and returns. Journal
of Finance, 50(1), 131-155. hps://doi.org/10.1111/j.1540-6261.1995.tb05169.x
Fama, E.F., & French, K.R. (2012). Size, value, and momentum in internaonal stock returns.
Journal of Financial Economics, 105(3), 457-472. https://doi.org/10.1016/j.jfineco.
2012.05.011
Fama, E.F., & French, K.R. (2015). A ve-factor asset pricing model. Journal of Financial Economics,
116(1), 1-22. hps://doi.org/10.1016/j.jneco.2014.10.010
Fisher, K.L., & Statman, M. (2003). Consumer condence and stock returns. Journal of Porolio
Management, 30(1), 115-127. hps://doi.org/10.3905/jpm.2003.319925
Hamilton, J.D. (1989). A new approach to the economic analysis of nonstaonary me series and
the business cycle. Econometrica, 57(2), 357-384. hps://doi.org/10.2307/1912559
Harshita, Singh, S., & Yadav, S.S. (2015). Indian stock market and the asset pricing models.
Procedia Economics and Finance, 30, 294-304. https://doi.org/10.1016/S2212-5671(15)
01297-6
Ho, C., & Hung, C-H. (2009). Investor senment as condioning informaon in asset pricing.
Journal of Banking & Finance, 33(5), 892-903. hps://doi.org/10.1016/j.jbankn.2008.10.004
Ho, J.C., & Hung, C-H. (2012). Predicting stock market returns and volatility with investor
senment: Evidence from eight developed countries. Journal of Accounng and Finance,
12(4), 49-66. hps://doi.org/10.2139/ssrn.2279339
Hsu, C-C., Lin, H-Y., & Wu, J-Y. (2011). Consumer confidence and stock markets: The panel
causality evidence. Internaonal Journal of Economics and Finance, 3(6), 91-99. hps://doi.
org/10.5539/ijef.v3n6p91
Idris, A.N., & Aziz, A. (2016, December 8). Jomo: Ringgit’s fall also due to loss of condence in
government. The Edge Financial Daily.
Jansen, W.J., & Nahuis, N.J. (2003). The stock market and consumer confidence: European
evidence. Economics Leers, 79(1), 89-98. hps://doi.org/10.1016/S0165-1765(02)00292-6
Karakatsani, N.V., & Salmon, M.H. (2008). Senment and price formaon: Interacons and regime
shis (SSRN working paper series). hps://doi.org/10.2139/ssrn.1102209
Kim, T., & Ha, N. (2010). Investor senment and market anomalies. 23rd Australasian Finance
and Banking Conference 2010 Paper (SSRN working paper series). hps://doi.org/10.2139/
ssrn.1663649
Kubota, K., & Takehara, H. (2017). Does the Fama and French Five-Factor model work well in
Japan? Internaonal Review of Finance, 18(1), 137-146. hps://doi.org/10.1111/ir.12126
Kuhnen, C.M., & Knutson, B. (2011). The inuence of aect on beliefs, preferences and nancial
decision. Journal of Financial and Quantave Analysis, 46(3), 605-626. hps://doi.org/
10.1017/S0022109011000123
Kumar, A., & Lee, C. (2006). Retail investor sentiment and return comovements. Journal of
Finance, 61(5), 2451-2486. hps://doi.org/10.1111/j.1540-6261.2006.01063.x
Lai, M.M., Tan, S.H., & Chong, L.L. (2013). The behavior of instuonal and retail investors in
Bursa Malaysia during the bulls and bears. Journal of Behavioral Finance, 14(2), 104-115.
hps://doi.org/10.1080/15427560.2013.790822
Lawrence, E.R., Geppert, J., & Prakash, A.J. (2007). Asset pricing models: A comparison. Applied
Financial Economics, 17(11), 933-940. hps://doi.org/10.1080/09603100600892863
Lee, Charles M.C., Shleifer, A., & Thaler, R.H. (1991). Investor senment and the closed-end
fund puzzle. Journal of Finance, 46(1), 75-109. hps://doi.org/10.1111/j.1540-6261.1991.
tb03746.x

Malaysian Journal of Economic Studies Vol. 55 No. 2, 2018 299
Senment-Augmented Asset Pricing in Bursa Malaysia
Lee, C.C., & Chiu, Y.B. (2012). The impact of real income on insurance premiums: Evidence from
panel data. International Review of Economics & Finance, 21(1), 246-260. https://doi.
org/10.1016/j.iref.2011.07.003
Liao, T-L., Huang, C-J., & Wu, C-Y. (2011). Do fund managers herd to counter investor senment?
Journal of Business Research, 64(2), 207-212. hps://doi.org/10.1016/j.jbusres.2010.01.007
Lintner, J. (1965). Security prices, risk, and maximal gains from diversicaon. Journal of Finance,
20(4), 587-615. hps://doi.org/10.1111/j.1540-6261.1965.tb02930.x
McMillan, D.G. (2003). Non-linear predictability of UK stock market returns. Oxford Bullen of
Economics and Stascs, 65(5), 557-573. hps://doi.org/10.1111/j.1468-0084.2003.00061.x
Reinganum, M.R. (1981). A new empirical perspecve on the CAPM. Journal of Financial and
Quantave Analysis, 16(4), 439-462. hps://doi.org/10.2307/2330365
Rosenberg, B., Reid, K., & Lanstein, R. (1985). Persuasive evidence of market ineciency. Journal
of Porolio Management, 11(3), 9-16. hps://doi.org/10.3905/jpm.1985.409007
Schmeling, M. (2009). Investor senment and stock returns: Some internaonal evidence. Journal
of Empirical Finance, 16(3), 394-408. hps://doi.org/10.1016/j.jempn.2009.01.002
Sharpe, W.F. (1964). Capital asset prices: A theory of market equilibrium under condions of risk.
Journal of Finance, 19(3), 425-442. hps://doi.org/10.2307/2977928
Shefrin, H., & Statman, M. (1994). Behavioral capital asset pricing theory. Journal of Financial and
Quantave Analysis, 29(3), 323-349. hps://doi.org/10.2307/2331334
Simpson, M.W., & Ramchander, S. (2008). An inquiry into the economic fundamentals of the
Fama and French equity factors. Journal of Empirical Finance, 15(5), 801-815. hps://doi.
org/10.1016/j.jempn.2008.02.003
Statman, M., Fisher, K.L., & Anginer, D. (2008). Aect in a behavioral asset-pricing model. Financial
Analysts Journal, 64(2), 20-29. hps://doi.org/10.2469/faj.v64.n2.8
Xu, Y.H., & Green, C.J. (2013). Asset pricing with investor senment: Evidence from Chinese stock
markets. Manchester School, 81(1), 1-32.hps://doi.org/10.1111/j.1467-9957.2011.02260.x
Yang, C., & Zhang, R. (2014). Does mixed-frequency investor senment impact stock returns?
Based on the empirical study of MIDAS regression model. Applied Economics, 46(9), 966-972.
hps://doi.org/10.1080/00036846.2013.864041
Yang, C.P., & Li, J.F. (2013). Investor senment, informaon and asset pricing model. Economic
Modelling, 35(September), 436-442.hps://doi.org/10.1016/j.econmod.2013.07.015
Zhang, W., & Semmler, W. (2009). Prospect theory for stock markets: Empirical evidence with
me-series data. Journal of Economic Behavior and Organizaon, 72(3), 835-849.hps://doi.
org/10.1016/j.jebo.2009.08.003

300 Malaysian Journal of Economic Studies Vol. 55 No. 2, 2018
Han Hwa Goh, Lee Lee Chong and Ming Ming Lai
Appendix
Table A1. A comparison of models via model selecon criteria for stock returns in Bursa Malaysia
Linear model FTP-MS AR(4) TVTP-MS AR(4) with ADR
Log-likelihood -403.8853 -366.9890 -357.4154*
AIC 4.587614 4.340784 4.254721*
SBC 4.747262 4.610996 4.560961*
HQ 4.652344 4.450380 4.378930*
Note: * The best model with relavely small AIC/SBC/HQ but relavely high log-likelihood. The TVTP-MS
model specicaon and the esmatesare reported in Panel A, Table 2. The esmates for both the
FTP-MS and the linear models are available upon request.
Table A2. A comparison of models via model selecon criteria for stock returns in Bursa Malaysia
Linear model FTP-MS AR(4) TVTP-MS AR(4) with NIPO
Log-likelihood -403.8853 -366.9890 -357.7290*
AIC 4.587614 4.340784 4.258284*
SBC 4.747262 4.610996 4.564524*
HQ 4.652344 4.450380 4.382493*
Note: * The best model with relavely small AIC/SBC/HQ but relavely high log-likelihood. The TVTP-MS
model specicaon and the esmates are reported in Panel B, Table 2. The esmates for both the
FTP-MS and the linear models are available upon request.
Table A3. A comparison of models via model selecon criteria for stock returns in Bursa Malaysia
Linear model FTP-MS AR(3) TVTP-MS AR(3) with CSI
Log-likelihood -403.8898 -381.2864 -369.2201*
AIC 4.576554 4.466513 4.352769*
SBC 4.718463 4.717734 4.639879*
HQ 4.634092 4.568399 4.469210*
Note: * The best model with relavely small AIC/SBC/HQ but relavely high log-likelihood. The TVTP-MS
model specicaon and the esmates are reported in Panel C, Table 2. The esmates for both the
FTP-MS and the linear models are available upon request.
Table A4. A comparison of models via model selecon criteria for stock returns in Bursa Malaysia
Linear model FTP-MS AR(3) TVTP-MS AR(3) with RIPO
Log-likelihood -403.8898 -381.2864 -369.9547*
AIC 4.576554 4.466513 4.325144*
SBC 4.718463 4.717734 4.648180*
HQ 4.634092 4.568399 4.477511*
Note: * The best model with relavely small AIC/SBC/HQ but relavely high log-likelihood. The TVTP-MS
model specicaon and its esmates are reported in Panel D, Table 2. The esmates for both the
FTP-MS and the linear models are available upon request.