Modeling the Volatility-Return Trade-Off When Volatility May Be Nonstationary
What drives herding in oil-rich, developing stock markets? Relative roles of own volatility and...
-
Upload
independent -
Category
Documents
-
view
4 -
download
0
Transcript of What drives herding in oil-rich, developing stock markets? Relative roles of own volatility and...
Our reference: ECOFIN 451 P-authorquery-v9
AUTHOR QUERY FORM
Journal: ECOFIN Please e-mail or fax your responses and any corrections to:
E-mail: [email protected]
Article Number: 451 Fax: +353 6170 9272
Dear Author,
Please check your proof carefully and mark all corrections at the appropriate place in the proof (e.g., by using on-screenannotation in the PDF file) or compile them in a separate list. Note: if you opt to annotate the file with software other thanAdobe Reader then please also highlight the appropriate place in the PDF file. To ensure fast publication of your paper pleasereturn your corrections within 48 hours.
For correction or revision of any artwork, please consult http://www.elsevier.com/artworkinstructions.
Any queries or remarks that have arisen during the processing of your manuscript are listed below and highlighted by flags inthe proof. Click on the ‘Q’ link to go to the location in the proof.
Location in Query / Remark: click on the Q link to goarticle Please insert your reply or correction at the corresponding line in the proof
Reference(s) given here were noted in the reference list but are missing from the text – please positioneach reference in the text or delete it from the list.
The reference given here is cited in the text but is missing from the reference list – please make thelist complete or remove the reference from the text: “Nielsen and Olsen (2001)”, “Hammoudeh et al.(2004)”, “Hammoudeh and Choi (2006)”, “Kodres and Pristsker (1998), Patev and Kanaryan (2003),Karunanayake (2010), Kremer and Nautz (2012), Froot et al. (1992), Choe et al. (1999) and Avramovet al. (2006)”.Please complete and update the reference given here (preferably with a DOI if the publication data arenot known): “Zheng and Zuo (2014)”, “Yao et al. (2013)”, “Demirer et al. (2013)”.For references to articles that are to be included in the same (special) issue, please add the words ‘thisissue’ wherever this occurs in the list and, if appropriate, in the text.
Q1 Please confirm that given names and surnames have been identified correctly.Q2 The country name has been inserted for the affiliations ‘b’ and ‘c’. Please check, and correct if necessary.Q3 Refs. “Kodres and Pristsker (1998), Patev and Kanaryan (2003), Karunanayake (2010), Kremer and
Nautz (2012), Froot et al. (1992), Choe et al. (1999) and Avramov et al. (2006)” are cited in the textbut not provided in the reference list. Please provide them in the reference list or delete these citationsfrom the text.
Q4 Ref. “Hammoudeh and Choi (2006)” is cited in the text but not provided in the reference list. Pleaseprovide it in the reference list or delete this citation from the text.
Q5 Ref. “Hammoudeh et al. (2004)” is cited in the text but not provided in the reference list. Please provideit in the reference list or delete this citation from the text.
Q6 Ref. “Nielsen and Olsen (2001)” is cited in the text but not provided in the reference list. Please provideit in the reference list or delete this citation from the text.
Q7 Uncited references: This section comprises references that occur in the reference list but not in the bodyof the text. Please position each reference in the text or, alternatively, delete it. Any reference not dealtwith will be retained in this section.
Page 1 of 2
Location in Query / Remark: click on the Q link to goarticle Please insert your reply or correction at the corresponding line in the proof
Q8 Please update Ref. “Demirer et al. (2013)”.Q9 Please update Ref. “Yao et al. (2013)”.Q10 Please update Ref. “Zheng and Zuo (2014)”.Q11 Please specify the significance of bold values cited in Tables 2 and 3.
Please check this box or indicate your approval ifyou have no corrections to make to the PDF file
Thank you for your assistance.
Please cite this article in press as: Balcilar, M., et al. What drives herding in oil-rich, developing stockmarkets? Relative roles of own volatility and global factors. North American Journal of Economicsand Finance (2014), http://dx.doi.org/10.1016/j.najef.2014.06.009
ARTICLE IN PRESSG ModelECOFIN 451 1–23
North American Journal of Economics and Finance xxx (2014) xxx–xxx
Contents lists available at ScienceDirect
North American Journal ofEconomics and Finance
What drives herding in oil-rich, developing1
stock markets? Relative roles of own volatility2
and global factors3
Mehmet Balcilara, Rıza Demirerb, Shawkat Hammoudehc,∗Q1
a Department of Economics, Eastern Mediterranean University, via Mersin 10, Famagusta, T. R. North4
Cyprus, Turkey5b Department of Economics & Finance, Southern Illinois University Edwardsville, Edwardsville, ILQ26
62026-1102, United States7c Lebow College of Business, Drexel University, Philadelphia, PA 19104, United States8
9
a r t i c l e i n f o10
11
JEL classification:12
C3213
G1114
G1515
Keywords:16
Herding17
Smooth transition regime switching18
Gulf Arab stock markets19
a b s t r a c t
The main goal of this paper is to formally establish the volatility-herding link in the developing stock markets of the oil-rich GCCcountries by examining how market volatility affects herd behav-ior after controlling for global factors. Using a regime-switching,smooth transition regression model (STR), we find significant evi-dence of herding in all Gulf Arab stock markets, with the marketvolatility being the more paramount factor governing the switchesbetween the extreme states of non-herding and herding. The globalvariables comprised of the U.S. stock market performance, the priceof oil and the US interest rate as well as the risk indexes includingthe CBOE Volatility Index (VIX) and the St. Louis Fed’s FinancialStress Index (FSI) are found to be significant factors governing thetransition to herding states. The findings stress the effect of con-tagion in financial markets, despite the restrictions established bythe GCC policymakers in order to protect their markets.
© 2014 Published by Elsevier Inc.
1. Introduction20
The literature on herd behavior in financial markets has been expanding rapidly in recent years,21
partly due to the prolonged market crisis that was originated in the U.S. financial markets and later22
∗ Corresponding author. Tel.: +1 610 949 0133; fax: +1 215 895 6975.E-mail address: [email protected] (S. Hammoudeh).
http://dx.doi.org/10.1016/j.najef.2014.06.0091062-9408/© 2014 Published by Elsevier Inc.
Please cite this article in press as: Balcilar, M., et al. What drives herding in oil-rich, developing stockmarkets? Relative roles of own volatility and global factors. North American Journal of Economicsand Finance (2014), http://dx.doi.org/10.1016/j.najef.2014.06.009
ARTICLE IN PRESSG ModelECOFIN 451 1–23
2 M. Balcilar et al. / North American Journal of Economics and Finance xxx (2014) xxx–xxx
spread to global markets.1 Although earlier studies establish a logical link between market volatility23
and herd behavior (e.g. Bikhchandani, Hirshleifer, & Welch, 1992; Christie & Huang, 1995), none of the24
studies in the literature have empirically examined the relative roles of a market’s own volatility and25
external factors in driving market states where herd behavior is observed. This is especially of concern26
for developing markets that are more prone to global effects. Given the suggestion in the literature27
that herd behavior might contribute to market volatility and pricing inefficiencies (e.g. Bikhchandani28
et al., 1992; Nofsinger & Sias, 1999 and more recently Blasco, Corredor, & Ferreruela, 2012), examining29
the relative roles of domestic market volatility and external factors in developing stock markets can30
provide additional valuable insight to policy makers regarding the development of market mechanisms31
to mitigate the negative effects resulting from herd behavior.32
Earlier studies including Christie and Huang (1995) and Chang, Cheng, and Khorana (2000) suggest33
that investors will be more likely to suppress their own beliefs and copy the behavior of others during34
periods of market stress, implying that market volatility is an important factor that may trigger herding.35
Regime-based tests of Balcilar, Demirer, and Hammoudeh (2013) suggest that market states during36
which herd behavior is observed are indeed associated with crashes and extreme volatility periods.37
Focusing on extreme market movements, studies including Kodres and Pristsker (1998), Patev and38
Kanaryan (2003) and Karunanayake (2010) reiterate that bad news and financial crises contribute toQ339
market volatility and herd behavior.2 Similarly, Kremer and Nautz (2012) argue that herding intensity40
depends on stock characteristics including past returns and volatility in an asymmetric way, that is,41
rising stock volatility leads to increased sell herding while buy herding measures decrease. Overall,42
there is sufficient evidence in the literature associating market volatility with herd behavior, with the43
relationship displaying an asymmetric pattern relative to the sign of the market direction. However,44
the mechanism in which a market’s own volatility influences herd behavior is yet to be explored.45
Furthermore, considering the fact that emerging markets are especially prone to global factors, a46
study that formally distinguishes between a market’s own volatility and global factors provides a new47
perspective to investor behavior in developing markets that has not been presented in prior studies.48
There are several contributions of this study. First, we explore the relative roles of a market’s own49
volatility and global factors in driving herd behavior in developing stock markets, with a focus on50
the cash- and oil-rich Gulf Cooperation Council (GCC) stock markets – Abu Dhabi, Dubai, Kuwait,51
Oman, Qatar and Saud Arabia. Prior studies in the literature base their tests on the assumption of a52
link between a market’s own volatility and herd behavior without explicitly modeling the volatility-53
herding link in their models. Therefore, this study extends the literature on herding by formally54
exploring the role of a market’s own volatility on herd behavior. Second, this study contributes to the55
literature on emerging markets from a new perspective by exploring the effects of the global financial56
environment on herd behavior after controlling for the local volatility factor in the stock market. Sep-57
arating the local and external factors in the model can provide valuable insight to the mechanism in58
which herd behavior develops in a stock market and aid policy makers in their regulatory tasks. Third,59
unlike prior studies in the literature, we propose a smooth transition regime-switching model where60
regime transitions are modeled in a smoothly time-varying framework as a function of a transition61
variable that governs the switching mechanism. The smooth transition regime-switching approach is62
flexible and switching is not abrupt or sharp as in the Markov switching models as will be explained63
later in the paper. Regime-switching is governed by an unobservable Markov chain process and there-64
fore, one can never be sure whether a particular regime has occurred at a particular time; but only65
assign probabilities to its occurrence. From a practical perspective, the smooth transition regression66
(STR) model provides a more realistic approach to herding tests as heterogeneous agents in the market67
with a diverse set of beliefs are unlikely to respond simultaneously to news or economic signals, thus68
leading to non-synchronized responses. Therefore, the herding tests based on the STR model for these69
markets allow one to gain insight into the factors driving herding behavior from a unique perspective70
1 Philippas et al. (2013), Lee et al. (2013), Yao et al. (2013), Zheng and Zuo (2014) and Demirer, Kutan, and Zhang (2013),among others.
2 The literature also examines the effect on volatility of investors that imitate other investors’ trades (Froot, et al. 1992; Choe,et al., 1999; Avramov, et al. 2006).
Please cite this article in press as: Balcilar, M., et al. What drives herding in oil-rich, developing stockmarkets? Relative roles of own volatility and global factors. North American Journal of Economicsand Finance (2014), http://dx.doi.org/10.1016/j.najef.2014.06.009
ARTICLE IN PRESSG ModelECOFIN 451 1–23
M. Balcilar et al. / North American Journal of Economics and Finance xxx (2014) xxx–xxx 3
that has not been done in the literature. The STR model not only accounts for direct effects of the global71
factors on herding behavior, but also provides insight into the variables that govern transitions into72
herding regimes. By doing so, this study contributes to the literature on herding, volatility transmission73
and international asset pricing.74
Our tests yield novel findings regarding the integration of the developing GCC markets with75
global markets with important implications for international diversification. First, herd behavior76
in these markets is a dynamic phenomenon, but not observed in all periods and it evolves in a77
smooth regime-switching fashion. Among the five GCC markets under consideration, Dubai, Kuwait,78
Qatar, and Saudi Arabia are unique with strong and persistent herding almost in all periods. Sec-79
ond, own volatility is the most significant factor driving regime switches between the extreme80
states of non-herding and herding, establishing a direct link between market volatility and herd81
behavior. This finding is consistent with Hammoudeh and Choi (2006) wherein market volatility Q482
in these markets is largely explained by local market factors. Finally, we find that shocks in global83
factors significantly contribute to investor herding in the GCC stock markets, despite the restric-84
tion to access by foreign investors into some of these markets. Global factors including the U.S.85
stock market performance, the price of oil and the US interest rate as well as risk indexes includ-86
ing the VIX and the FSI are found to be significant factors governing the transitions to herding87
states. Interestingly, this evidence comes despite the fact that most GCC markets protect them-88
selves from foreign investors by putting up barriers to entry of foreign investors, partly in the89
hope of reducing the impact of global volatilities on their markets and partly as a matter of90
sovereignty.91
The remainder of the paper is organized as follows. Section 2 briefly summarizes previous studies.92
Section 3 provides description of the data and the benchmark model used in herding tests, while93
Section 4 presents the STR models that are used to test the relative impacts of volatility and global94
effects on herd behavior. Section 5 presents the empirical results and Section 6 concludes the paper95
and discusses implications of the findings.96
2. Previous studies97
Earlier studies in the herding literature start with Christie and Huang (1995) and Chang et al. (2000)98
who report partial evidence of herd behavior in several advanced and developing Asian markets, except99
the U.S. Later studies focus mostly on emerging markets and provide support for herd behavior in a100
number of developing markets including China (Hsieh, Yang, Yang, & Lee, 2011; Lee, Chen, & Hsieh,101
2013; Tan, Chiang, Mason, & Nelling, 2008; Yao, Ma, & He, 2013), Taiwan (Demirer et al., 2010), Gulf102
Arab stock markets (Balcilar et al., 2013), among others. On the other hand, in other recent studies,103
Chiang and Zheng (2010), Economou, Kostakis, and Philippas (2011) and Philippas, Economou, Babalos,104
and Kostakis (2013) report evidence of herd behavior in advanced markets as well. In a recent study,105
Balcilar et al. (2013) propose a regime-switching alternative to the standard testing methodology and106
document evidence of herding in the GCC stock markets during the high and crash volatility states.107
On the other hand, a number of studies on the developing Gulf Arab stock markets have focused108
on the integration of these rapidly developing markets with global markets. Studies including109
Hammoudeh and Li (2008), Yu and Hassan (2008), Marashdeh and Shrestha (2010), Ravichandran110
and Maloain (2010), Cheng, Jahan-Parvar, and Rothman (2010) and more recently Demirer (2013)111
examine the interaction of these markets with global markets from a portfolio diversification point112
of view. The literature in general documents that these markets are partially integrated with global113
markets, suggesting the potential for international diversification benefits for global investors. How-114
ever, the literature on the interaction of these markets with global markets has not yet been extended115
to the herding context. Furthermore, none of the herding tests in the literature differentiate the effect116
of a market’s own volatility from that of global factors in their tests. This paper contributes to the117
literature on developing markets by examining herd behavior in an environment characterized by118
structural breaks, external factor effects and smooth transition regime-switching governed by a tran-119
sition or switch variable. To our knowledge, this is the first study that utilizes formal tests and selection120
criteria to identify the variables that govern the transition between non-herding and herding regimes.121
Please cite this article in press as: Balcilar, M., et al. What drives herding in oil-rich, developing stockmarkets? Relative roles of own volatility and global factors. North American Journal of Economicsand Finance (2014), http://dx.doi.org/10.1016/j.najef.2014.06.009
ARTICLE IN PRESSG ModelECOFIN 451 1–23
4 M. Balcilar et al. / North American Journal of Economics and Finance xxx (2014) xxx–xxx
3. Data and testing methodology122
Several methodologies that utilize price data in herding tests have been proposed in the literature.3123
A popular methodology applied in the herd literature originates from an earlier model by Christie and124
Huang (1995) that was later improved by Chang et al. (2000). The methodology employed in this125
literature focuses on the dispersion of returns in a portfolio of assets with similar characteristics.126
The dispersion of returns in a portfolio is measured by the cross-sectional absolute deviation (CSAD)127
of individual stock returns around the market return, and displays the pattern of return dispersions128
during periods of large market movements. The dispersion of returns is defined as129
CSADt = 1n
n∑
i=1
|Ri,t − Rm,t (1)130
where n is the number of stocks in the portfolio and Ri,t and Rm,t are the return on stock i and the131
market portfolio for period t, respectively. The testing methodology uses the conditional Capital Asset132
Pricing Model (CAPM) as the base model and detects the presence of herd behavior by focusing on133
possible non-linearity in the asset pricing model. For this purpose, Chang et al. (2000) develop the134
following benchmark model135
CSADt = ˛0 + ˛1|Rm,t | + ˛2R2m,t + εt (2a)136
where a significant and negative ˛2 estimate is used as support for the presence of herding behavior.137
This specification, although static in nature, has been applied to herding tests in a number of recent138
studies including Philippas et al. (2013), Lee et al. (2013) and Yao et al. (2013).139
Following prior studies in the literature on the interaction between the GCC stock markets and140
global factors (e.g. Balcilar & Genc, 2010; Khalifa, Hammoudeh, & Otrano, 2013), we augment the141
benchmark model in Eq. (2a) with two global variables, i.e. the oil price and the S&P 500 index, which142
are documented to significantly interact with GCC stock returns to estimate143
CSADt = ˛0 + ˛1|Rm,t | + ˛2R2m,t + ˛3R2
US,t + ˛4R2Ot + εt (2b)144
where RUS,t and RO,t are the returns on the S&P 500 index and US price of WTI crude oil for period t,145
respectively. The augmented model is motivated by a number of studies in the international asset pri-146
cing literature including Stulz (1984, 1995) and Karolyi and Stulz (2003), suggesting that the domestic147
CAPM specification would be incorrect if there are additional risk factors perceived by local investors148
that might contribute to stock returns beyond what can be explained by the domestic market factor149
only. In the case of GCC stock returns, prior studies including Balcilar and Genc (2010) and Khalifa150
et al. (2013) document the presence of global effects on these markets. Therefore, the augmented151
model avoids any specification errors in the domestic pricing model and allows one to examine the152
effects of shocks in global factors on investor behavior in GCC markets. Therefore, in Eq. (2b), estimat-153
ing significant and negative values for ˛3 and ˛4 suggests that large movements in the global factors154
significantly contribute to herding in the GCC markets even after controlling for the domestic market155
factor.156
The dataset consists of weekly closing price series for all stocks listed on five GCC stock exchanges157
including those of Saudi Arabia, Dubai, Abu Dhabi, Kuwait, and Qatar which have consistent and158
adequate data series. These data are sourced from Thompson/Reuters. We use the weekly data in159
our tests because the GCC markets follow different trading days and weekends from the Western160
markets (i.e. Fridays are part of the weekends in the GCC countries and their markets are closed on161
those days). Therefore, in order to avoid any un-synchronization due to differences in trading days,162
we utilize weekly stock returns using Tuesdays as the base day in our weekly return calculations. This163
day avoids the second effect in both groups of markets. As explained earlier, one of the contributions164
of the model used in this analysis is to separate the effect of the market’s own volatility from external165
factors. For this purpose, as will be explained in more detail in Section 4, we also utilize a number of166
3 Demirer, Kutan, and Chen (2010) provide a comparison of the testing methodologies that utilize return data in their tests.
Please cite this article in press as: Balcilar, M., et al. What drives herding in oil-rich, developing stockmarkets? Relative roles of own volatility and global factors. North American Journal of Economicsand Finance (2014), http://dx.doi.org/10.1016/j.najef.2014.06.009
ARTICLE IN PRESSG ModelECOFIN 451 1–23
M. Balcilar et al. / North American Journal of Economics and Finance xxx (2014) xxx–xxx 5
Table 1Descriptive statistics.
Mean S.D. Min Max n Sample period
Abu DhabiRm 0.200% 3.230% −13.870% 16.670% 541
5/29/2001–3/6/2012CSAD 6.230% 4.570% 0.430% 27.480% 541�2
t (Rm) 0.0011 0.0009 0.0002 0.0048 541Dubai
Rm 0.120% 4.870% −21.860% 15.510% 4141/13/2004–3/6/2012CSAD 4.900% 2.230% 1.690% 13.680% 414
�2t (Rm) 0.0025 0.002 0.0009 0.0145 414
KuwaitRm 0.180% 2.240% −13.100% 7.960% 819
12/5/1995–3/6/2012CSAD 6.410% 3.740% 1.690% 29.670% 819�2
t (Rm) 0.0005 0.0006 0.0001 0.0068 819Qatar
Rm 0.280% 4.330% −24.050% 14.370% 4591/14/2003–3/6/2012CSAD 3.880% 1.860% 0.740% 16.570% 459
�2t (Rm) 0.0019 0.0023 0.0005 0.0175 459
Saudi ArabiaRm 0.200% 3.480% −23.220% 13.710% 855
1/17/1995–3/6/2012CSAD 3.470% 2.140% 1.070% 22.800% 855�2
t (Rm) 0.0013 0.0021 0.0002 0.0204 855
Global variablesDollar Index Return 0.010% 0.690% −3.950% 3.070% 894
1/17/1995–3/6/2012
S&P 500 Return 0.120% 2.510% −15.770% 12.370% 894Crude oil return (WTI) 0.190% 5.470% −37.010% 25.180% 894FSI 0.037 1.021 −1.256 5.429 894T-Bill Rate (TB3) 3.032 2.079 0.010 6.230 894VIX 21.735 8.408 9.900 67.640 894
Note: This table reports the descriptive statistics for daily market index returns and cross sectional return dispersions acrossall listed stocks in each exchange, respectively. CSAD is the cross-sectional absolute deviation of returns as a measure of returndispersion. n is the number of observations. FSI is the St. Louis Federal Reserve’s Financial Stress Index, TB3 is the U.S. three-month Treasury bill rate, VIX is the CBOE Volatility Index, and WTI is West Texas Intermediate price. The global variables are usedaccording to their stationarity or lack thereof. Bahrain is not included because its data starts in 2009. �2
t (Rm) is the estimate of theconditional return variance estimated as the recursive one-step ahead forecast from a generalized autoregressive conditionalheteroskedasticity (GARCH) model with GARCH (1,1) specification.
global factors that serve as potential transition variables to govern the switching mechanism between167
non-herding and herding market states since these factors may indirectly drive investor sentiment in168
the GCC markets.169
Table 1 provides the summary statistics and the sample periods for each GCC market as well as the170
global factors utilized. We use a generalized autoregressive conditional heteroskedasticity (GARCH)171
model to estimate the conditional variance values which are later utilized as a proxy for market172
volatility in each GCC stock market.4 In order to generate the market volatility data, we use a recursive173
estimation scheme and obtain the conditional variances, which are denoted as �2t (Rm), at time t. In174
other words, the generated conditional variances are the one-step ahead forecasts from the GARCH175
(1,1) models fitted to observations 1,2, . . ., t − 1. Figs. 1–5 provide the plots of the generated conditional176
volatilities as well as the CSAD values for each GCC market. Interestingly, the figures suggest a close177
association between the return dispersion values and the market volatility estimates, particularly for178
Abu Dhabi and Dubai.179
Examining the return values, all GCC markets yield positive average returns during their sample180
periods, with the highest value for Qatar reflecting its growing oil and natural gas fortunes and the181
lowest for Dubai possibly due to the recent real estate market’s crash and its dependence on tourism.182
The high level of return in GCC markets relative to the S&P 500 index, with the exception of Dubai, is183
4 GARCH (1,1) specification is used in the estimation of conditional volatility terms.
Please cite this article in press as: Balcilar, M., et al. What drives herding in oil-rich, developing stockmarkets? Relative roles of own volatility and global factors. North American Journal of Economicsand Finance (2014), http://dx.doi.org/10.1016/j.najef.2014.06.009
ARTICLE IN PRESSG ModelECOFIN 451 1–23
6 M. Balcilar et al. / North American Journal of Economics and Finance xxx (2014) xxx–xxx
Fig. 1. CSAD and market volatility used in STR herding model for Abu Dhabi stock market.
associated with relatively higher levels of volatility in general. In the case of return dispersion (CSAD)184
values, stock returns in Saudi Arabia and Qatar display the lowest cross-sectional dispersion possibly185
due to the government’s use of their own domestic stock shares to stabilize their own markets (BalcilarQ5186
et al., 2013; Hammoudeh et al., 2004).187
4. Testing the effects of market volatility and global factors on herding behavior188
One of the weaknesses of the standard testing model employed in the literature is that the model is189
static in nature, with the model parameters assumed to be constant over time. Therefore, the bench-190
mark model in Eq. (2a) and the augmented model in Eq. (2b) fail to capture the dynamic relation191
between market conditions and herd behavior. On the other hand, returns and related time series192
data from financial markets are well known to display nonlinearities which manifest themselves in193
various forms such as asymmetric responses and adjustments during crises and recovery periods,194
responses to market shocks depending on the size and sign of the shocks, etc. A number of studies195
including Tyssedal and Tjostheim (1988), Hamilton (1988), Schwert (1989), Pagan and Schwert (1990),196
Sola and Timmermann (1994), Schaller and van Norden (1997), Kim, Nelson, and Startz (1998), Kim197
and Nelson (1998), and Mayfield (1999) document that nonlinear features in financial data are well198
captured by regime switching models. Following these observations, Balcilar et al. (2013) address the199
shortcoming in the benchmark specification by proposing a regime-based model of return dispersions200
where market regimes are identified in terms of the level of volatility. Their tests indeed distinguish201
Please cite this article in press as: Balcilar, M., et al. What drives herding in oil-rich, developing stockmarkets? Relative roles of own volatility and global factors. North American Journal of Economicsand Finance (2014), http://dx.doi.org/10.1016/j.najef.2014.06.009
ARTICLE IN PRESSG ModelECOFIN 451 1–23
M. Balcilar et al. / North American Journal of Economics and Finance xxx (2014) xxx–xxx 7
Fig. 2. CSAD and market volatility used in STR herding model for Dubai stock market.
between herding and non-herding market regimes and confirm that herd behavior is observed during202
high market volatility states only.203
Despite its advantages, the Markov-switching (MS) models assume that the regime-switching is204
governed by an unobservable Markov chain process and therefore, one can never be sure whether205
a particular regime has occurred at a particular time; but only assign probabilities to its occurrence.206
Moreover, the MS models imply sharp regime switching, and therefore utilize a limited number (usu-207
ally two) of regimes in their specifications. However, this assumption might be too restrictive when208
the transition between the regimes is not discrete jumps, but rather gradual changes or in the form209
of a continuum of regimes. Another attractive specification that smoothly captures herding or non-210
herding over the different market phases in a consistent way with the return and volatility structure of211
returns is the smooth transition regression (STR) model in the spirit of Teräsvirta and Anderson (1992),212
Granger and Teräsvirta (1993), and Teräsvirta (1994, 1998). The STR models have the advantage of213
allowing for a smooth or gradual change from one state (e.g. non-herding) to another (e.g. herding)214
where the transition can be captured in the form of a continuum of regimes (Teräsvirta, 1998). From215
a practical perspective, one can argue that heterogeneous agents in the market with a diverse set of216
beliefs are unlikely to respond simultaneously to news or economic signals. Furthermore, a number217
of factors including differences in applicable transaction costs, heterogeneity in investors’ objectives218
due to differences in investment horizons, geographical location, and a variety of different risk profiles219
as Peters (1994) suggests, may lead to non-synchronized responses by agents to market shocks. From220
Please cite this article in press as: Balcilar, M., et al. What drives herding in oil-rich, developing stockmarkets? Relative roles of own volatility and global factors. North American Journal of Economicsand Finance (2014), http://dx.doi.org/10.1016/j.najef.2014.06.009
ARTICLE IN PRESSG ModelECOFIN 451 1–23
8 M. Balcilar et al. / North American Journal of Economics and Finance xxx (2014) xxx–xxx
Fig. 3. CSAD and market volatility used in STR herding model for Kuwait stock market.
this perspective, one can argue that a dynamic model of herding which allows for smooth transitions221
or continuum of states between two extreme market states (non-herding and herding states) provides222
a more realistic setup in herding tests and can better capture the role of market factors on transitions223
to herding states.224
Following the arguments presented above, we propose a dynamic regime-switching specification,225
similar to Balcilar et al. (2013) which not only distinguishes between herding and non-herding mar-226
ket regimes but also allows the cross regime switches to be governed by the market volatility and the227
global factors. On the other hand, Balcilar et al. (2013) assumes the constant transition probabilities228
across the market regimes, and therefore does not provide insights to how the market’s own volatility229
drives the transitions across the herding and non-herding regimes. Furthermore, unlike Balcilar et al.230
(2013) we focus on the role of the global factors in driving herding regimes in the developing GCC mar-231
kets and providing insight into the integration of these markets with global markets from a different232
methodological approach, and we additionally examine the role of volatility. Another distinction of this233
paper which is also methodological is the use of an STR model that provides a more realistic approach234
to the herding tests than in the previous paper. The methodology in Balcilar et al. (2013) implies that235
the herding regime may be related to the volatility, but does not allow a direct way to incorporate236
this implication into the model. Additionally, the methodology used here allows a smooth transition237
between the regimes as well as determining how fast the markets adjust toward a new state. The het-238
erogeneous agents in the markets with a diverse set of beliefs are unlikely to respond simultaneously to239
Please cite this article in press as: Balcilar, M., et al. What drives herding in oil-rich, developing stockmarkets? Relative roles of own volatility and global factors. North American Journal of Economicsand Finance (2014), http://dx.doi.org/10.1016/j.najef.2014.06.009
ARTICLE IN PRESSG ModelECOFIN 451 1–23
M. Balcilar et al. / North American Journal of Economics and Finance xxx (2014) xxx–xxx 9
Fig. 4. CSAD and market volatility used in STR herding model for Qatar stock market.
news or economic signals, thus leading to non-synchronized and non-herding responses, and also mak-240
ing the STR model a more realistic approach. Therefore, this study offers several novelties compared to241
Balcilar et al. (2013) and contributes both to the literature on financial integration as well as to herding.242
For this purpose, we estimate the following two-state STR model of the cross-sectional absolutedispersions as warranted by the data
CSADt = [˛0,1 + ˛1,1|Rm,t | + ˛2,1R2m,t + ˛3,1R2
US,t + ˛4,1R2O,t] + [˛0,2 + ˛1,2|Rm,t | + ˛2,2R2
m,t
+ ˛3,2R2US,t + ˛4,2R2
O,t]G(st, �, c) + εt (3)
where εt ∼ iid(0, �2) is a sequence of iid random variables and G(st, � , c) is the transition function243
(bounded by 0 and 1) that controls the regime shift mechanism as a smooth and continuous function244
of the realized values of the continuous transition variable st. In this specification, � > 0 is the slope245
of the transition function that controls the speed of switching and c is the threshold parameter for246
switching between herding and non-herding regimes. Thus, CSAD is modeled to evolve through a247
smooth transition between regimes depending on the sign and magnitude of past realizations of the248
transition variable st. The nonlinearities are obtained by conditioning the regression parameters ˛i,j,249
i = 0, 1, . . ., 4, j = 1, 2 to change smoothly with st in such a way that the variable st is the transition variable250
governing switching from the linear regime (Regime 1) to the nonlinear regime (Regime 2). Indeed,251
st may be set to d-lagged values of a variable, where d indicates the number of periods the transition252
variable leads the switch in regime dynamics. As will be discussed in the empirical results section,253
Please cite this article in press as: Balcilar, M., et al. What drives herding in oil-rich, developing stockmarkets? Relative roles of own volatility and global factors. North American Journal of Economicsand Finance (2014), http://dx.doi.org/10.1016/j.najef.2014.06.009
ARTICLE IN PRESSG ModelECOFIN 451 1–23
10 M. Balcilar et al. / North American Journal of Economics and Finance xxx (2014) xxx–xxx
Fig. 5. CSAD and market volatility used in STR herding model for Saudi Arabia stock market. Note: (a) Plots the CSAD definedin Eq. (1). (b) Plots the market volatility, i.e., the conditional standard deviation of market return, which is used as transitionvariable in the STR model in. The shaded regions in (a) and (b) correspond to regimes where herding is supported with negativesignificant coefficients on squared returns in Eq. (3).
we find that regime switches may indeed be quite gradual, heavily driven by the market volatility,254
supporting the description by Christie and Huang (1995) that associates herding with higher market255
volatility.256
Teräsvirta and Anderson (1992) specify the transition function G(·) by using two alternative forms,257
namely the logistic smooth transition regression (LSTR) model and the exponential smooth transition258
regression (ESTR) model. In the LSTR model, G(·) is defined by a logistic function in the form259
G(st, �, c) = [1 + exp {−�(st − c}]−1, � > 0 (4)260
On the other hand, in the ESTR framework, G(·) is captured by an exponential function261
G(st, �, c) = 1 − exp{−�(st − c)2}, � > 0 (5)262
In both equations, � is the speed of transition between regimes and c indicates the halfway point or263
the threshold between the two regimes. Eq. (3) combined with Eq. (4) produces the LSTR model, while264
Eq. (3) combined with Eq. (5) yields the ESTR model. In the STR models, expansion and contraction are265
representations of two different economic phases, but transition between the two regimes is smooth,266
controlled by st (Sarantis, 2001). The LSTR and ESTR models describe different dynamic behavior.267
The LSTR model allows the expansion and contraction regimes to have different dynamics, whereas268
the ESTR model suggests that the two regimes have similar dynamics (Sarantis, 2001). We also take269
Please cite this article in press as: Balcilar, M., et al. What drives herding in oil-rich, developing stockmarkets? Relative roles of own volatility and global factors. North American Journal of Economicsand Finance (2014), http://dx.doi.org/10.1016/j.najef.2014.06.009
ARTICLE IN PRESSG ModelECOFIN 451 1–23
M. Balcilar et al. / North American Journal of Economics and Finance xxx (2014) xxx–xxx 11
note that when � → ∞, the model degenerates into the conventional threshold model which has270
discrete instant regime switching. On the other hand, when � → 0, the model degenerates to the271
linear regression model (Teräsvirta & Anderson, 1992). The procedure for constructing an appropriate272
STR model for a specific variable is comprised of three following stages.273
4.1. Testing for linearity274
For a set of transition variables s1t, s2t, . . ., smt one has to first establish whether linearity is rejected275
for any of the transition variables sjt, j = 1, 2, . . ., m. Unfortunately, the linearity restriction that � = 0276
cannot be tested directly since � and c are unidentified and are also nuisance parameters under the null277
hypothesis of linearity. Luukkonen, Saikkonen, and Teräsvirta (1988) suggests taking the Taylor series278
expansion of G(·) with respect to st and constructing LM type tests in the exponential regression equa-279
tions. For the k-th order Taylor expansions (k = 1, 2, 3, 4), the LM tests can be computed as F type tests by280
setting all parameters to zero where skt exists as a variable multiplying xt =
(1, |Rm,t |, R2
m,t, R2US,t, R2
O,t
)′.281
The resulting tests are denoted as LMk for k = 1, 2, 3, 4. Luukkonen et al. (1988) argue that LM1 and LM3282
tests have power against the LSTR alternative, while LM2 and LM4 tests have power against the ESTR283
alternatives. A rejection of any of the denoted LMk tests establishes evidence against the linearity in284
favor of the STR model.285
4.2. Selecting the transition variable286
Teräsvirta (1994) shows that the LM3 tests also have power against the ESTR alternatives and can287
be used to select the best fitting transition variable. The test is performed in the following regression:288
CSADt = ˇ′0xt + ˇ′
1xtst + ˇ′2xts
2t + ˇ′
3xts3t + ut (6)289
where the null of linearity involves the joint restrictions H0 : ˇ1 = ˇ2 = ˇ3 = 0. An appropriate transition290
variable can be determined by first computing the LM3 tests for a set of transition variables s1t, s2t, . . .,291
smt and selecting the transition variable as the one with the smallest p-value of the test.292
4.3. Selecting the form of the transition function293
Having rejected linearity, the choice between the LSTR and ESTR model is then conducted by294
applying the following sequence of nested tests (Teräsvirta, 1994) to Eq. (6)295
H03 : ˇ3 = 0, (7a)296
H02 : ˇ2 = 0|ˇ3 = 0 (7b)297
H01 : ˇ1 = 0|ˇ2 = ˇ3 = 0, (7c)298
A standard procedure, as discussed in Teräsvirta and Anderson (1992), is then followed in the299
selection of the appropriate STR model. There are three possible sequential outcomes:300
i. The rejection of H03 : ˇ3 = 0 implies the selection of the LSTR model.301
ii. If H03 is not rejected, then we proceed to test H02 : ˇ2 = 0|ˇ3 = 0. Rejection of H02 implies the302
selection of the ESTR model.303
iii. If H02 is not rejected, then we proceed to test H01 : ˇ1 = 0|ˇ2 = ˇ3 = 0. A rejection of H01 implies304
selection of the LSTR model.305
As a general selection procedure, if the p-value of the test corresponding to H02 is the smallest, an306
ESTR model is selected; while in all other cases an LSTR model is selected. Escribano and Jordá (1999)307
propose a somewhat different selection procedure based on testing two separate hypotheses within308
Eq. (6) and following equation:309
CSADt = ˇ′0xt + ˇ′
1xtst + ˇ′2xts
2t + ˇ′
3xts3t + ˇ′
4xts4t + ut (7)310
Please cite this article in press as: Balcilar, M., et al. What drives herding in oil-rich, developing stockmarkets? Relative roles of own volatility and global factors. North American Journal of Economicsand Finance (2014), http://dx.doi.org/10.1016/j.najef.2014.06.009
ARTICLE IN PRESSG ModelECOFIN 451 1–23
12 M. Balcilar et al. / North American Journal of Economics and Finance xxx (2014) xxx–xxx
In Escribano and Jordá procedure, H0E : ˇ2 = ˇ4 = 0, and H0L : ˇ1 = ˇ3 = 0 are tested in Eqs. (6) and311
(7), respectively, and LSTR (ESTR) model is selected if the minimum p-value is obtained for H0L (H0E).312
On the other hand, various authors (Eitrheim & Terävirta, 1996; Granger & Teräsvirta, 1993; Sarantis,313
2001; Teräsvirta, 1994) argue that if this sequence of tests is strictly applied, then it may lead to wrong314
conclusions since the higher order terms of the Taylor expansion used in the derivation of these tests315
are ignored. Thus, it is recommended to first compute the p-values for all the F-tests of (i)–(iii) above316
and proceed with the choice of the appropriate STR model based on the lowest p-value or the highest317
F-statistic.318
Regarding the selection of the potential transition variables to be used in the STR specification of319
Eq. (3), we identify six variables that we believe would potentially have a role in driving herd behavior320
in the GCC stock markets:321
1. Domestic market volatility. Following studies that suggest a link between market volatility and herd322
behavior (e.g. Blasco et al., 2012), we include in our models the domestic market volatility (measured323
as the conditional variance of the market return).324
2. Oil price. The economies of GCC markets are highly sensitive to the price of crude oil as most GCC325
countries are net exporters and their stock markets and investments are fueled by oil revenues and326
petrodollars. It is possible that the oil price directly feeds herding tendencies in these markets.327
3. U.S. market return. As experienced during the 2008 global financial crisis, developments in the328
U.S. market can potentially lead to major movements in developing stock markets, either through329
contagion effects or through the activities of international investors. Therefore, it can be argued330
that herd behavior is partially driven by developments in the U.S. stock market.331
4. Three-month T-Bill rate (TB3). GCC countries officially or effectively peg their currencies to the U.S.332
dollar which might lead these markets to be somewhat sensitive to changes in the U.S. T-bill rate.333
Furthermore, the U.S. Treasuries have often been considered a safe haven for global investors dur-334
ing market crisis periods. At times of market stress, investors move away from volatile developing335
markets and currencies and shift their funds into safer assets including the USD-denominated Trea-336
sury securities. Therefore, the T-bill rate can be considered as a variable which may capture this337
sentiment across investors, particularly during periods of market stress. This applies in particular338
to the GCC central banks that park their foreign reserves in those securities.339
5. The dollar exchange rate index. Following a similar argument as in (4), it is possible that movements340
in the value of the U.S. dollar against major currencies reflect investors’ risk appetite as investors341
often move in and out of the USD-denominated assets in response to changes in risk appetite. This342
leads to a possible link between the movements in the U.S. dollar and herd behavior in developing343
markets. The dollar exchange rate has special importance in the GCC countries because they peg344
their currencies officially or effectively to the US dollar.345
6. Global risk measures. These measures encompass risk indexes including the CBOE volatility index346
(VIX), often termed as the fear index, and St. Louis Federal Reserve’s financial stability index (FSI).347
These two indexes are included in the model in order to control for investors’ perception of risk and348
financial distress in the equity market.349
Thus, the set of transition variables in Eqs. (4) and (5) is defined as {USD Return, VIX Return, SP500350
Return, WTI Return, FSI Return, TB3}.5351
5. Empirical results352
In this section, we first start with the LM-STR tests for linearity of CSAD, second we identify the353
appropriate transition variable and finally we proceed with the hypothesis tests to choose between the354
LSTR and ESTR models as explained in the previous section.6 Once we select the appropriate STR model,355
we then estimate this model. The STR models are nonlinear regression models and are estimated with356
5 The unit root tests indicate that TB3 is integrated of degree one, therefore its first difference is used in the estimation.6 The results for the static models in Eqs. (2a) and (2b) are not reported for brevity and are available upon request.
Please cite this article in press as: Balcilar, M., et al. What drives herding in oil-rich, developing stockmarkets? Relative roles of own volatility and global factors. North American Journal of Economicsand Finance (2014), http://dx.doi.org/10.1016/j.najef.2014.06.009
ARTICLE IN PRESSG ModelECOFIN 451 1–23
M. Balcilar et al. / North American Journal of Economics and Finance xxx (2014) xxx–xxx 13
nonlinear least squares method. As will be discussed later in this section, the static linear models given357
in Eqs. (2a) and (2b) are rejected by the LM-STR tests against the STR alternatives. Furthermore, our358
tests indicate that the p-values of the LM3 are the smallest for all GCC markets when the conditional359
variance �t2(Rm) is used as the transition variable, suggesting that the STR model with market volatility360
as the transition variable provide the best explanatory power as will be discussed later.361
5.1. Testing for linearity and model selection362
We begin our analysis by testing for linearity against STR alternatives with a possible set of potential363
transition variables that may govern the regime-switching mechanism. The F variants of the four LM-364
STR linearity tests, explained in Section 4.1, are reported in Table 2 along with their p-values given in365
parenthesis. In order to establish evidence in favor of a smooth transition regime-switching model,366
one has to reject at least one of the four LM-STR tests. We should note at this point that LM-STR367
tests are based on linearized regression models where the transition function does not directly exist.368
Therefore, it does not matter whether the alternative is an LSTR or ESTR model. However, the LM1 test369
has good power against the LSTR alternative, the LM2 and LM4 tests have good power against ESTR370
alternative, and the LM3 tests have good power against both the LSTR and ESTR alternatives (van Dijk,371
Teräsvirta, & Franses, 2002).372
Once a particular transition variable is selected from the list of the potential variables, the next373
step is to test for the presence of nonlinear dynamics in the data given the transition variable. It is of374
interest to test whether nonlinearity adds any explanatory power to the static, constant coefficient375
model in Eq. (2b). Note that rejection of linearity implies that the herding relationship follows a regime376
switching (nonlinear) form and statistical inferences that are based on linear models are likely to be377
misleading. Therefore, due to the time-varying feature of the herding model, the tests based on the378
assumption of the linear single regime model may fail to detect herding behavior since it may not379
be observed in all periods. If linearity is rejected given a particular transition variable, it implies that380
the selected transition variable sufficiently captures the nonlinearity in return dispersions (CSAD),381
and thus can be interpreted as a variable that governs the smooth regime-switching variable, i.e. a382
variable that leads to a switch from a linear regime (Regime 1) to a nonlinear regime (Regime 2), and383
vice versa.384
The findings reported in Table 2 indicate that all LM-STR tests reject linearity for all GCC markets in385
favor of the STR model, when the market volatility defined as the conditional variance of the market386
return �t2(Rm) is the transition variable. Moreover, the rejection is highly significant with all p-values387
well below the 1% level, implying the presence of strong nonlinearity that can be adequately captured388
by the STR model. For example, in the case of Abu Dhabi which is known for its conservative governance389
of its financial markets, the LM2, LM3 and the LM4 tests reject linearity when the USD and FSI are the390
transition variables, and all STR-LM tests reject linearity when TB3 is the transition variable. This391
finding implies that USD, FSI, and TB3 may also generate smooth regime switching for Abu Dhabi.392
However, in the case of Dubai, which depends heavily on tourism and real estate sectors, only the LM2393
test rejects linearity when USD is the transition variable. However, linearity is not rejected for Dubai394
when SP500, VIX, WTI, TB3 and FSI are used as transition variable. Linearity is rejected by all LM-STR395
tests for Kuwait when USD, SP500, VIX, and WTI are used as transition variables. Similarly, for Qatar,396
all STR-LM tests reject linearity when TB3 is the transition variable, but linearity is not rejected when397
USD, SP500, VIX, WTI, and FSI are used as transition variable. In the case of Saudi Arabia, the LM1 and398
LM2 tests reject linearity, when USD and SP500 are the transition variables when only LM1 rejects399
linearity with TB3 as the transition variable.400
Examining the findings for the other global factors, when USD is the transition variable, linearity is401
rejected for all markets except Qatar whereas when the SP500 index is the transition variable, the tests402
reject linearity for Kuwait and Saudi Arabia. In the case of VIX, this risk measure is found to capture403
regime-switching dynamics only for Kuwait; while WTI captures the same effect for Kuwait and Saudi404
Arabia. Finally, using FSI as the transition variable, linearity is rejected for Abu Dhabi and Saudi Arabia.405
In the case of the US interest rate TB3, we find that this variable captures regime-switching dynamics406
for Abu Dhabi and Qatar, and to some extent for Saudi Arabia.407
Pleasecite
this
articlein
press
as:Balcilar,M
.,etal.Wh
atdrives
herd
ing
inoil-rich
,develop
ing
stockm
arkets?R
elativeroles
ofow
nvolatility
and
globalfactors.N
orthA
merican
Journalof
Economics
andFinance
(2014),http
://dx.d
oi.org/10.1016/j.najef.2014.06.009
AR
TIC
LE
IN P
RE
SS
G M
odelEC
OFIN
4511–23
14M
.Balcilaret
al./North
Am
ericanJournalofEconom
icsand
Financexxx
(2014)xxx–xxx
Table 2LM type tests for STR nonlinearity and model identification for the herding models with global factors.Q11
Transition variable
�t2(Rm) USD SP500 VIX WTI FSI TB3
Abu DhabiLM1 15.1097 [0.0000] 1.2617 [0.2791] 0.3094 [0.9074] 1.1058 [0.3562] 0.5408 [0.7454] 1.5712 [0.1664] 5.9810 [0.0000]LM2 8.3891 [0.0000] 2.4518 [0.0072] 1.0113 [0.4324] 0.9885 [0.4522] 1.3544 [0.1984] 2.5277 [0.0056] 3.8309 [0.0001]LM3 5.6729 [0.0000] 2.2044 [0.0056] 0.9426 [0.5162] 0.7273 [0.7575] 1.2934 [0.2011] 1.8072 [0.0308] 4.2541 [0.0000]LM4 4.4478 [0.0000] 2.0216 [0.0056] 1.0541 [0.3960] 0.7845 [0.7336] 1.1571 [0.2873] 1.8663 [0.0129] 3.3636 [0.0000]H01 15.1097 [0.0000] 1.2617 [0.2791] 0.3094 [0.9074] 1.1058 [0.3562] 0.5408 [0.7454] 1.5712 [0.1664] 5.9810 [0.0000]H02 1.5852 [0.1624] 3.6109 [0.0032] 1.7111 [0.1302] 0.8725 [0.4992] 2.1620 [0.0570] 3.4479 [0.0045] 1.6445 [0.1465]H03 0.3450 [0.8855] 1.6780 [0.1381] 0.8089 [0.5436] 0.2196 [0.9541] 1.1670 [0.3242] 0.3953 [0.8521] 4.8223 [0.0003]H0E 0.6399 [0.7798] 1.6698 [0.0847] 1.2499 [0.2563] 0.5926 [0.8205] 0.4098 [0.9421] 2.2620 [0.0136] 3.0312 [0.0010]H0L 1.1250 [0.3410] 1.2015 [0.2871] 1.3118 [0.2207] 0.7123 [0.7132] 1.6862 [0.0808] 2.3342 [0.0107] 2.1715 [0.0183]
DubaiLM1 3.4075 [0.0050] 0.9441 [0.4522] 0.1710 [0.9732] 0.2584 [0.9355] 0.7655 [0.5751] 0.7575 [0.5809] 0.5673 [0.7251]LM2 2.5372 [0.0057] 1.9508 [0.0374] 0.5484 [0.8553] 0.7723 [0.6556] 0.7150 [0.7105] 0.7088 [0.7164] 1.1206 [0.3451]LM3 2.2963 [0.0039] 1.4672 [0.1141] 0.7629 [0.7189] 0.7415 [0.7420] 0.5849 [0.8867] 0.9090 [0.5542] 0.8675 [0.6014]LM4 1.9746 [0.0077] 1.3250 [0.1586] 0.5965 [0.9154] 0.6547 [0.8698] 0.6127 [0.9038] 0.8018 [0.7120] 0.8215 [0.6876]H01 3.4075 [0.0050] 0.9441 [0.4522] 0.1710 [0.9732] 0.2584 [0.9355] 0.7655 [0.5751] 0.7575 [0.5809] 0.5673 [0.7251]H02 1.6400 [0.1483] 2.9348 [0.0129] 0.9259 [0.4640] 1.2853 [0.2693] 0.6676 [0.6482] 0.6632 [0.6516] 1.6691 [0.1409]H03 1.7658 [0.1188] 0.5234 [0.7586] 1.1894 [0.3136] 0.6859 [0.6344] 0.3367 [0.8906] 1.3041 [0.2613] 0.3787 [0.8633]H0E 1.4834 [0.1431] 1.3311 [0.2115] 0.1169 [0.9996] 0.6424 [0.7773] 0.4663 [0.9114] 0.9057 [0.5279] 1.1940 [0.2931]H0L 2.1286 [0.0215] 0.7724 [0.6555] 0.3721 [0.9583] 0.7588 [0.6686] 0.7197 [0.7060] 0.6566 [0.7646] 0.3250 [0.9743]
KuwaitLM1 3.0492 [0.0098] 3.3586 [0.0052] 4.2773 [0.0008] 3.9663 [0.0015] 3.2771 [0.0062] 1.5991 [0.1579] 0.9795 [0.4292]LM2 2.7932 [0.0021] 2.4745 [0.0064] 4.1982 [0.0000] 2.8495 [0.0017] 2.9136 [0.0014] 1.3961 [0.1771] 0.9748 [0.4638]LM3 2.8805 [0.0002] 2.4748 [0.0015] 2.3758 [0.0023] 2.3874 [0.0022] 2.2898 [0.0035] 1.2579 [0.2229] 0.9295 [0.5305]LM4 2.3405 [0.0008] 2.0089 [0.0056] 2.7171 [0.0001] 2.4767 [0.0004] 2.3351 [0.0008] 1.3594 [0.1343] 0.7579 [0.7658]H01 3.0492 [0.0098] 3.3586 [0.0052] 4.2773 [0.0008] 3.9663 [0.0015] 3.2771 [0.0062] 1.5991 [0.1579] 0.9795 [0.4292]H02 2.5087 [0.0289] 1.5784 [0.1637] 4.0387 [0.0013] 1.7152 [0.1286] 2.5193 [0.0283] 1.1913 [0.3116] 0.9702 [0.4350]H03 1.5229 [0.1801] 2.4313 [0.0336] 0.2827 [0.9227] 1.4472 [0.2050] 1.0408 [0.3923] 0.9817 [0.4279] 0.8408 [0.5209]H0E 1.6730 [0.0827] 1.8286 [0.0522] 1.9980 [0.0309] 2.2645 [0.0130] 2.2880 [0.0120] 1.4682 [0.1465] 0.2421 [0.9919]H0L 1.4956 [0.1361] 1.3403 [0.2044] 1.8238 [0.0529] 1.5290 [0.1242] 1.8438 [0.0498] 0.5965 [0.8175] 0.7255 [0.7008]
QatarLM1 11.6081 [0.0000] 0.6132 [0.6898] 0.0937 [0.9932] 1.1887 [0.3137] 0.9255 [0.4641] 1.6610 [0.1427] 2.1283 [0.0610]LM2 7.2924 [0.0000] 0.8141 [0.6152] 0.4683 [0.9103] 0.8403 [0.5899] 1.1053 [0.3563] 1.4374 [0.1609] 1.9793 [0.0339]LM3 5.3427 [0.0000] 1.1210 [0.3344] 0.4732 [0.9535] 0.7252 [0.7594] 0.7547 [0.7281] 1.1516 [0.3075] 1.9286 [0.0190]LM4 4.3336 [0.0000] 1.2062 [0.2441] 0.6614 [0.8642] 0.6034 [0.9109] 0.7908 [0.7256] 0.9928 [0.4696] 2.0493 [0.0050]
Pleasecite
this
articlein
press
as:Balcilar,M
.,etal.Wh
atdrives
herd
ing
inoil-rich
,develop
ing
stockm
arkets?R
elativeroles
ofow
nvolatility
and
globalfactors.N
orthA
merican
Journalof
Economics
andFinance
(2014),http
://dx.d
oi.org/10.1016/j.najef.2014.06.009
AR
TIC
LE
IN P
RE
SS
G M
odelEC
OFIN
4511–23
M.Balcilar
etal./N
orthA
merican
JournalofEconomics
andFinance
xxx(2014)
xxx–xxx15
Table 2 (Continued)
Transition variable
�t2(Rm) USD SP500 VIX WTI FSI TB3
H01 11.6081 [0.0000] 0.6132 [0.6898] 0.0937 [0.9932] 1.1887 [0.3137] 0.9255 [0.4641] 1.6610 [0.1427] 2.1283 [0.0610]H02 2.7504 [0.0184] 1.0150 [0.4082] 0.8430 [0.5197] 0.4985 [0.7774] 1.2823 [0.2704] 1.2100 [0.3034] 1.8110 [0.1093]H03 1.3808 [0.2302] 1.7215 [0.1283] 0.4884 [0.7850] 0.5045 [0.7729] 0.0763 [0.9958] 0.5932 [0.7052] 1.7919 [0.1131]H0E 1.6605 [0.0876] 1.4128 [0.1716] 0.7833 [0.6450] 0.3409 [0.9695] 0.8486 [0.5819] 0.8032 [0.6257] 2.2987 [0.0123]H0L 3.8493 [0.0001] 0.9066 [0.5269] 0.8164 [0.6129] 0.5514 [0.8531] 0.8879 [0.5445] 1.2175 [0.2773] 2.6142 [0.0043]
Saudi ArabiaLM1 48.5041 [0.0000] 2.2426 [0.0483] 1.1201 [0.3480] 0.9352 [0.4573] 1.5050 [0.1857] 2.1308 [0.0598] 2.3655 [0.0381]LM2 35.8408 [0.0000] 1.5175 [0.1281] 2.0580 [0.0255] 0.9279 [0.5065] 2.0445 [0.0266] 1.6744 [0.0823] 1.6539 [0.0873]LM3 26.4863 [0.0000] 1.1600 [0.2981] 1.3937 [0.1431] 0.6736 [0.8120] 1.5220 [0.0908] 1.7935 [0.0314] 1.3187 [0.1837]LM4 21.2642 [0.0000] 1.2285 [0.2223] 1.4236 [0.1026] 0.5683 [0.9349] 1.2674 [0.1925] 2.0761 [0.0038] 1.1245 [0.3177]H01 48.5041 [0.0000] 2.2426 [0.0483] 1.1201 [0.3480] 0.9352 [0.4573] 1.5050 [0.1857] 2.1308 [0.0598] 2.3655 [0.0381]H02 18.2318 [0.0000] 0.7952 [0.5532] 2.9827 [0.0112] 0.9210 [0.4665] 2.5700 [0.0256] 1.2152 [0.2999] 0.9431 [0.4522]H03 5.7504 [0.0000] 0.4549 [0.8099] 0.0876 [0.9942] 0.1741 [0.9722] 0.4895 [0.7843] 2.0116 [0.0748] 0.6551 [0.6577]H0E 4.1241 [0.0000] 1.2299 [0.2675] 1.6783 [0.0814] 0.5596 [0.8473] 0.8297 [0.6000] 2.5115 [0.0056] 0.7164 [0.7095]H0L 7.1580 [0.0000] 1.3002 [0.2257] 0.7024 [0.7228] 0.1840 [0.9974] 0.9669 [0.4709] 1.8839 [0.0440] 1.1943 [0.2908]
Note: This table reports the F variants of the LM type linearity tests and the p-values given in square brackets, as well as the H01, H02, and H03 tests used in the specification procedures andTeräsvirta (1994); and the H0E and H0L tests used in the procedure of Escribano and Jordá (1999), applied to the global herding model given in Eq. (3). The tests are applied with varioustransition variables’ specifications, which include the conditional variance of the market return (�t
2(Rm), the US dollar exchange rate index return (USD), the S&P 500 stock market indexreturn (SP500), the CBOE Volatility VIX index return (VIX), the West Texas Intermediate crude oil price return (WTI), St. Louis Federal Reserve’s financial stability index return (FSI), andweekly changes in the U.S. 3-month Treasury bill rate.Note: See note to Table 3.
Please cite this article in press as: Balcilar, M., et al. What drives herding in oil-rich, developing stockmarkets? Relative roles of own volatility and global factors. North American Journal of Economicsand Finance (2014), http://dx.doi.org/10.1016/j.najef.2014.06.009
ARTICLE IN PRESSG ModelECOFIN 451 1–23
16 M. Balcilar et al. / North American Journal of Economics and Finance xxx (2014) xxx–xxx
In short, the findings suggest that local market volatility is the only consistent transition variable408
that uniformly rejects linearity for all GCC markets by all LM-STR tests. In the case of global factors409
however, we find that the global factors also contribute to regime switching in GCC markets, but not410
as strong as the local market volatility factor. However, Kuwait and Saudi Arabia are found to be more411
prone to be influenced by global factors. Market volatility, as the most significant transition variable412
explaining regime-switching in all GCC markets, is most likely to be influenced by the changes in413
global factors, thus indirectly reflecting their effects.414
Once nonlinearity is established, we proceed with the selection of the desired transition variable.415
The findings in Table 2 indicate that more than one transition variable can explain the regime-416
switching dynamics in the GCC stock markets. In order to determine the transition variable that417
best explains the regime-switching dynamics, we follow Teräsvirta (1994) and perform the LM3 tests418
in Eq. (6) with the restriction H0 : ˇ1 = ˇ2 = ˇ3 = 0 imposed.7 The variable with the smallest p-value419
or the highest F value is selected as the transition variable that best captures the regime-switching420
dynamics. The reported F test values (p-values) for all potential transition variables are at maximum421
(minimum) uniformly for all GCC markets when the transition variable is the conditional variance of422
the market returns, �t2(Rm). Therefore, we conclude that the market volatility factor best governs the423
regime-switching in all GCC markets, and thus we select it as the transition variable in the rest of our424
analysis.425
Regarding the procedure to determine the form of the transition function G(·) that will be used426
in the herding model, we consider the LSTR form given in Eq. (4) and the ESTR form given in Eq. (5).427
We use the two procedures proposed by Teräsvirta and Anderson (1992) and Escribano and Jordá428
(1999), both explained in Section 4.3, in order to determine the form of the transition function. The429
sequential testing procedure of Teräsvirta and Anderson (1992) selects the ESTR if the F-statistic value430
(the p-value) of the H02 test is the largest (the smallest), and selects the LSTR if the F-statistic value431
(the p-value) of the H01 or H03 test is the largest (the smallest). The results reported in Table 2 have the432
largest F statistic value (the smallest p-value) for the H01 test, uniformly for all GCC markets. Therefore,433
based on the Teräsvirta and Anderson (1992) procedure, the LSTR model is selected for all countries.434
The Escribano and Jordá (1999) procedure confirms this result for all GCC markets except Kuwait for435
which H0E has the smallest p-value and the ESTR model is thus preferred. However, we also choose436
the LSTR model for Kuwait since the p-value of the H01 test is smaller than the p-value of the H0E.437
Therefore, the LSTR model with the smooth transition function given in Eq. (4) is identified as the438
preferred model for all GCC markets.439
In order to further check the robustness of the LM-STR tests, we include in the model several440
combinations of dummies that correspond to spikes in the CSAD values exceeding three standard441
deviations of the mean, with the restriction that no more than 18 dummies will be included in anyQ6442
case.8 None of the combinations of the dummies changes the LM-STR test results. In fact, the inclusion443
of the dummies even enhances the test results in favor of the STR alternatives. The additional LM tests444
using the outlier robust M-estimation method (van Dijk et al., 2002) further suggest that the findings445
in Table 2 are robust. Thus, we conclude that the STR specification for the GCC markets is not spurious446
and corresponds to a true regime-switching model. Once again, the results of the robust tests for each447
country are not included for space considerations and are available upon request.448
Having identified the appropriate STR specification, the parameters of the STR herding model449
described in Eqs. (3) and (5) are estimated using the nonlinear least squares method as the model is450
truly nonlinear in the parameters � and c. The initial estimates of the parameters � and c are obtained451
using a grid search. Given the initial parameter estimates, the BFGS algorithm9 is used to optimize over452
7 Teräsvirta (1994) shows that the LM3 test has power against both the LSTR and ESTR alternatives. Therefore, the form ofthe transition function G(·) does not influence the selection of the transition variable by the LM3 test procedure.
8 It is possible that a regime is spurious and corresponds to few spikes in the data. For instance, Nielsen and Olsen (2001) findthat a third regime for the Danish stock market is a figment of the data which disappears when dummy variables correspondingto few spikes in the data are included. But this is not the case in our STR models which is consistent with the MS models inBalcilar et al. (2013).
9 BFGS refers to the well-known Broyden–Fletcher–Goldfarb–Shanno algorithm.
Please cite this article in press as: Balcilar, M., et al. What drives herding in oil-rich, developing stockmarkets? Relative roles of own volatility and global factors. North American Journal of Economicsand Finance (2014), http://dx.doi.org/10.1016/j.najef.2014.06.009
ARTICLE IN PRESSG ModelECOFIN 451 1–23
M. Balcilar et al. / North American Journal of Economics and Finance xxx (2014) xxx–xxx 17
Table 3Estimates for the herding models with global factors under smooth transition regime switching.
Abu Dhabi Dubai Kuwait Qatar Saudi Arabia
� 0.0191** (0.0096) 1.9950*** (0.0789) 0.9900*** (0.2384) 8.8650** (3.7720) 1.3250*** (0.2000)c 0.0023*** (0.0003) 0.0015*** (0.0001) 0.0003*** (0.0000) 0.0010*** (0.0002) 0.0008*** (0.0001)˛0,1 −0.0769 (0.0537) 0.0028 (0.0037) 0.0109 (0.0095) 0.0148*** (0.0044) 0.0206*** (0.0009)˛0,2 1.1100*** (0.1826) 0.0367*** (0.0029) 0.0465*** (0.0026) 0.0206*** (0.0034) 0.1291*** (0.0376)˛1,1 4.2300*** (1.5390) 0.0486 (0.1514) 0.9072*** (0.2262) 0.0598 (0.1907) 0.6897*** (0.0424)˛1,2 −31.5900*** (5.2450) 0.3288* (0.1729) 0.0835 (0.5106) 0.3293 (0.2185) −2.2380 (1.3770)
Herding coefficients˛2,1 27.2000*** (5.4800) 3.5710*** (1.3650) 0.8953 (5.7520) 5.6880** (2.4670) 24.0900** (12.2800)˛2,2 −24.3900** (10.4300) −4.0620*** (1.4640) −3.1320*** (0.7410) −6.4480** (2.5570) −2.2300*** (0.3400)˛3,1 28.2600*** (5.3020) 2.6660 (2.2520) 4.8960*** (1.1240) 1.2860 (1.5800) −0.2585 (0.4289)˛3,2 −1.3860 (1.8210) −1.9850 (2.1400) −6.5680*** (1.6570) −1.6920 (1.4630) −52.3700** (21.4100)˛4,1 11.3000*** (1.4620) 0.0967 (0.5186) 0.5066* (0.2960) 0.1119 (0.3274) 0.0087 (0.0991)˛4,2 −1.183* (0.6741) −0.0137 (0.5069) −0.8340** (0.3704) −0.1258 (0.3115) 10.8700 (6.8950)
Fit statistics�1 0.89 0.37 0.44 0.37 0.62�2 0.11 0.63 0.56 0.63 0.38n 541 414 819 459 855n1 484 155 364 172 526n2 57 259 455 287 329
SSR 0.7715 0.1322 0.9910 0.0821 0.1866� 0.0382 0.0181 0.0350 0.0136 0.0149�(st) 0.0009 0.0020 0.0006 0.0023 0.0021AIC −6.5085 −7.9917 −6.6879 −8.5765 −8.4019BIC −6.4132 −7.8750 −6.6189 −8.4686 −8.3353
Note: This table presents the estimates of the regime switching LSTR model given in Eqs. (3) and (5). The heteroskedasticity andautocorrelation consistent (HAC) standard errors are reported in parentheses, which are obtained using the HAC covariancematrix of Newey and West (1987). n is the total number of observations, nk is the number of observations in regime k, �k ispercentage of observations falling in regime k, SSR is the standard error of the regression, � is the standard deviation of theresiduals, �(st) is the standard deviation of the transition variable st , AIC is the Akaike information criterion, and BIC is the LRtest is the linearity test. Standard errors of the estimates are given in parentheses.
* Significance at the 10% level.** Significance at the 5% level.
*** Significance at the 1% level.
� and c and the remaining parameter estimates, ˛i,j, i = 0, 1, . . ., 4, j = 1, 2 are obtained using ordinary453
least squares at each step of the estimation.454
5.2. Results of the STR model455
Table 3 presents our findings for the global LSTR model specified in Eqs. (3) and (5). The estimates for456
each state clearly differentiate each regime in terms of the sign and size of the coefficients. For example,457
we observe significantly higher intercept estimates in the nonlinear regime (Regime 2), implying458
greater dispersion values during this regime. This suggests that returns are more dispersed during459
periods of high volatility. However, examining the relationship between the return dispersion and460
market return values, we find that the herding coefficients (˛2,2) are all negative and highly significant461
at one percent level, suggesting the presence of herd behavior in all GCC markets during the nonlinear462
regime only.10 The only exception to this is Abu Dhabi for which the coefficient is significant but at the463
5% level.11 Interestingly, the results suggest no evidence of herding during the low volatility regime464
10 The nonlinear regime refers to the high volatility regime during which herding is observed. It can also be called the inefficientor non-equilibrium regime as market prices do not reflect the fundamentals during the bubble/crash periods.
11 As discussed later, regime classification for Abu Dhabi indicates that there are only two long periods, one during 2005 andanother during 2008–2009, when this market is in nonlinear herding. There are a few other short periods of herding but theseare not significantly long. This is also reflected in the parameter estimates only significant at the 5% significance level.
Please cite this article in press as: Balcilar, M., et al. What drives herding in oil-rich, developing stockmarkets? Relative roles of own volatility and global factors. North American Journal of Economicsand Finance (2014), http://dx.doi.org/10.1016/j.najef.2014.06.009
ARTICLE IN PRESSG ModelECOFIN 451 1–23
18 M. Balcilar et al. / North American Journal of Economics and Finance xxx (2014) xxx–xxx
(Regime 1) with none of the estimates for ˛2,1 estimated to be negative and significant. Therefore,465
we conclude that herding is only present in the nonlinear, high volatility regime. These findings are466
consistent with the suggestion in earlier studies including Christie and Huang (1995), Chang et al.467
(2000) and Bikhchandani and Sharma (2001) without regime switching, that investors will be more468
likely to suppress their own beliefs and copy the behavior of others during periods of market stress.469
In the case of global factors, all of the coefficients representing the global factors (i.e. ˛3,j and ˛4,j,470
j = 1,2) in Eq. (3) are found to be negative in the nonlinear regime, whereas these coefficients are mostly471
positive during the linear regime. This suggests that the global factors indeed contribute to herding472
in GCC markets, however, during the nonlinear regime only. Examining the individual markets, we473
observe the strongest results are for Abu Dhabi, Kuwait, and Saudi Arabia. Both the U.S. stock market474
and the oil price are found to significantly contribute to herd behavior in Kuwait, whereas the oil475
price and the U.S. stock market are found to be significant drivers of herd behavior in Abu Dhabi476
and Saudi Arabia, respectively. Finding U.S. and oil market effects for these three countries is not477
surprising as these three countries are major oil-exporters and prominent members of OPEC, making478
their economies highly sensitive to oil prices. Interestingly, none of the global factors are found to479
contribute to herding in Dubai and Qatar. The findings for Dubai can be explained by its relatively480
lesser and decreasing dependence on oil exports and its greater diversification in the real estate and481
tourism sectors. Similarly, the findings for Qatar may be due to the interventions in this market by the482
government through the Qatar Investment Authority (Balcilar et al., 2013). From the perspective of483
financial integration of these markets with global markets, a significant U.S. market effect in the case of484
Kuwait and Saudi Arabia during the high volatility regime provides partial support for the integration485
of these markets with global markets, despite the limited access allowed for foreign investors. This486
suggests that the U.S. market effect is most likely due to contagion effects and is consistent with Khalifa487
et al. (2013) who find evidence of volatility spillover from the U.S. stock market and the oil market to488
GCC stock markets.489
5.3. Timing, frequency, and persistence of market regimes490
Having found evidence of herding in GCC stock markets and partial evidence on global effects, we491
next examine the frequency, persistence, and time evolution of the herding regime in order to gain492
further insight into investor behavior in these markets. Table 3 reports the number of observations in493
each regime (nj, j = 1, 2) and percentage of observations relative to the total number of observations (�j,494
j = 1, 2) for each regime. The percentage of observations classified as falling into the herding regime495
(Regime 2) are 11%, 63%, 56%, 63%, and 62% for Abu Dhabi, Dubai, Kuwait Qatar, and Saudi Arabia,496
respectively. These results imply that more than half of the observations for all GCC markets except497
Abu Dhabi can be classified as falling into the period during which herding is present. Interestingly,498
only a small percentage of observations belong to the herding regime for Abu Dhabi which is fiscally499
governed more conservatively than the other GCC countries. In general, we can conclude that herding500
is a commonly observed phenomenon for four of the GCC markets, with the exception of Abu Dhabi.501
On the other hand, the frequency of the number of observations does not convey any information502
on the duration and timing of the herding and non-herding regimes. Instead, this information can be503
obtained from the regime classification of the periods. Any period where we have st < c is classified504
in the non-herding region or close to the linear non-herding state. Analogously, all observations with505
st > c are assumed to be in the region of nonlinear herding regime or being closer to it. Whether an506
observation belongs to a particular regime is hard to determine unless G(·) = 0 (complete non-herding)507
or G(·) = 1 (complete herding). This is so as the testing methodology uses the signs of the coefficients508
˛2,1 and ˛2,2 in order to label a regime as a herding or non-herding regime and an observation has509
weights from both regimes at points between G(·) = 0 or G(·) = 1. As explained earlier, the STR models510
have the unique feature of smoothly moving from one extreme state to the other and it is normal511
to expect most of the observations to fall into this smooth transition period. The observations in512
the transition period are generated from the coefficients of both extreme regimes, and thus the STR513
model can be thought of as generating a continuum of regimes from two extreme pole regimes. When514
st > c, the weight given to the herding regime will exceed 50% (G(·) > 0.50), and when st < c, the weight515
given to non-herding regime will exceed 50% (1–G(·) > 0.50). The point where st = c, which means that516
Please cite this article in press as: Balcilar, M., et al. What drives herding in oil-rich, developing stockmarkets? Relative roles of own volatility and global factors. North American Journal of Economicsand Finance (2014), http://dx.doi.org/10.1016/j.najef.2014.06.009
ARTICLE IN PRESSG ModelECOFIN 451 1–23
M. Balcilar et al. / North American Journal of Economics and Finance xxx (2014) xxx–xxx 19
G(·) = 0.50, is the halfway between the two poles of the regimes and the observations will receive 50%517
weight from both poles. Thus, we can say that more than half of the time, observations receive more518
weight from the herding regime which can be interpreted as being closer to the extreme herding519
regime.520
A time evolution of the regime classification may well illustrate both the frequency and the per-521
sistence of regimes. Figs. 1–5 plot the CSAD and market volatility, defined as the square root of the522
conditional variance. These figures allow one to examine the dynamic evolution of the market regimes523
over time. In each panel of Figs. 1–5, the observations that are classified in the herding regime are524
shaded. The shaded regions in these figures indicate strong persistence for the herding regime par-525
ticularly for Dubai and Qatar. The least persistent herding regime is observed in Abu Dhabi. There526
are two-year long herding periods observed for Qatar and Dubai. There are several periods where the527
herding regime is observed continuously for more than two years in Kuwait and Saudi Arabia. All these528
results indicate that the herding regime is not only frequent but also quite persistent.529
Panels (b) in Figs. 1 through 5 display market volatility, along with periods where periods of herding530
is indicated with shades. In Fig. 1(a), for example, we see that herding was a persistent phenomenon531
in Abu Dhabi from early 2005 to mid-2005 and from the end of 2008 to mid-2009. This period cer-532
tainly witnessed several crashes and extreme volatilities. In Dubai, Kuwait, Qatar, and Saudi Arabia, the533
periods 2005–2007, 2008, and 2009–2010 are the three periods where herding is detected continu-534
ously and over long periods, as shown in Figs. 2(b)–5(b), respectively. The figures profoundly illustrate535
that the support for herding is stronger and more persistent for the most volatile periods, confirming536
the earlier finding that market volatility is the main factor that drives herd behavior in these mar-537
kets. Overall, among the five GCC markets, Dubai, Kuwait, Qatar, and Saudi Arabia are unique with538
strong and persistent herding almost in all periods. Herding is less frequent for Abu Dhabi, but occurs539
persistently for some periods during 2005–2009.540
5.4. Speed of adjustment between the non-herding and herding regimes541
In Table 3, we also report the slope parameter of the logistic transition function (�), defined in542
Eq. (5), which controls the speed of adjustment from one regime to another. The case when � = 0543
corresponds to a linear model. Lower values of � indicate slower speed of adjustment between regimes.544
Estimates of all � parameters are all significant at the 5% level even though this parameter is usually545
not estimated precisely. This finding reinforces our results of testing for linearity. Estimates of the546
speed parameter are 0.0191, 0.9900, 1.3250, 1.9950, and 8.8650, respectively, for Abu Dhabi, Kuwait,547
Saudi Arabia, Dubai, and Qatar. The regime switching is the fastest in Qatar and the slowest in Abu548
Dhabi. Even though the frequency of regime switching is small in Abu Dhabi, the speed of adjustment549
is the slowest, implying that the frequency of switching and its speed are independent. As indicated550
above, this emirate is run conservatively, follows a moderate economic development and does not551
allow its sovereign wealth fund the Abu Dhabi Investment authority to invest in its domestic market.552
Fig. 6 plots the value of the transition function, evaluated at the estimated parameter values, against553
the transition variable and marks the threshold c with a vertical line in each plot. The visual inspection554
of the figures clearly reveals the fast switching of the transition function from 0 toward 1 for Qatar and555
how slow it is for Abu Dhabi. Indeed, the range of transition function values never comes close to 1556
for Abu Dhabi. As discussed above, the speed of adjustment parameter controls also how observations557
spread between the two extreme regimes of non-herding and herding. The slower the adjustment558
speed, the more likely for observations to remain in the interim regimes as the system stays longer in559
between the two extremes. In order to illustrate this point, we count the percentage of observations560
where G(·)is only 0.15 points away from either 0 or 1. Indeed, the percentage of observations falling561
in the lower end below the value 0.15 is zero percent for Abu Dhabi, Dubai, Kuwait, and Saudi Arabia562
and 3.7% for Qatar. On the other hand, the percentage of observations falling in the upper end above563
0.95 (close to the extreme herding regime) are 0.0%, 1.9%, 6.5%, 4.7%, and 7.7%, respectively for Abu564
Dhabi, Dubai, Kuwait, Qatar, and Saudi Arabia. This implies that most of the observations indeed fall565
into the transition periods, whereas the system stays closer to the extreme herding regime compared566
to the non-herding regime.567
Please cite this article in press as: Balcilar, M., et al. What drives herding in oil-rich, developing stockmarkets? Relative roles of own volatility and global factors. North American Journal of Economicsand Finance (2014), http://dx.doi.org/10.1016/j.najef.2014.06.009
ARTICLE IN PRESSG ModelECOFIN 451 1–23
20 M. Balcilar et al. / North American Journal of Economics and Finance xxx (2014) xxx–xxx
Fig. 6. Transition function against the transition variable in the smooth transition regime switching herding mode. Note: Eachgraph plots the logistic transition function G(st , � , c) given in Eq. (5), evaluated at the estimated parameter values, against thetransition variable st . The vertical line in gray color in each graph is drawn at the estimated value of the threshold parameter cwhich corresponds to G(.) = 0.50. Above 0.50, the market is in the herding regime.
6. Implications and conclusions568
This paper contributes to the literature on herding and international finance by examining the569
relative roles of a market’s own volatility and global factors in driving herd behavior in developing570
markets, with a focus on the cash- and oil-rich GCC stock markets (Abu Dubai, Dubai, Kuwait, Qatar571
Please cite this article in press as: Balcilar, M., et al. What drives herding in oil-rich, developing stockmarkets? Relative roles of own volatility and global factors. North American Journal of Economicsand Finance (2014), http://dx.doi.org/10.1016/j.najef.2014.06.009
ARTICLE IN PRESSG ModelECOFIN 451 1–23
M. Balcilar et al. / North American Journal of Economics and Finance xxx (2014) xxx–xxx 21
and Saud Arabia). By doing so, this study establishes a direct link between market volatility and herd572
behavior which is only assumed in standard tests. Furthermore, unlike prior studies in the literature,573
we propose a smooth transition regime-switching model (STR) where regime transitions are modeled574
in a smoothly time-varying framework as a function of a transition variable that governs the switching575
mechanism. The STR specification provides a more realistic setup for herding tests as it allows for576
smooth transitions or continuum of states between non-herding and herding states and thus provides577
further insight to the role of market factors on transitions to herding states.578
Our results have four distinct features: First, herd behavior is a dynamic phenomenon but not579
observed in all periods and it evolves in a smooth regime switching fashion. Among the five GCC580
markets, Dubai, Kuwait, Qatar, and Saudi Arabia are unique with strong and persistent herding almost581
in all periods. Herding is less frequent for economically conservative Abu Dhabi, but occurs persistently582
for some periods during 2005–2009. The regime switching is the fastest in the fast growing Qatar and583
the slowest in Abu Dhabi. Even though the frequency of regime switching is small in Abu Dhabi,584
the speed of adjustment is also the slowest, implying that the frequency of switching and its speed585
are independent. The findings, however, show that the regime switching is not swift in all five GCC586
markets, probably due to restrictions on foreign investments and other institutional factors in those587
countries. Second, market volatility is the most significant variable driving regime switching between588
the extreme states of non-herding and herding.589
This finding establishes a direct link between market volatility and herd behavior and is consistent590
with earlier studies including Christie and Huang (1995) and Chang et al. (2000), suggesting that591
investors will be more likely to suppress their own beliefs and copy the behavior of others during592
periods of market stress. Finally, shocks in global factors significantly contribute to investor herding593
in the GCC stock markets despite the restriction to access by foreign investors in some of these markets.594
Global factors including the U.S. stock market performance, the price of oil and the US interest rate595
as well as risk indexes including the VIX and the FSI are found to be significant factors governing the596
transition to herding states. Interestingly, this evidence comes despite the fact that most GCC markets597
protect themselves from foreign investors by putting up barriers to entry of foreign investors, partly598
in the hope of reducing the impact of global volatilities on their markets and partly as a matter of599
sovereignty. This finding stresses the effect of contagion in financial markets despite the restrictions600
established by policy makers in order to protect developing markets.601
Uncited references Q7602
Krolzig (1997), Lakonishok, Shleifer, and Vishny (1992) and Wermers (1999).603
Acknowledgements604
We would like to thank the Guest Editors, Shawkat Hammoudeh and Duc Kuong Nguyen, the Editor,605
Hamid Beladi, and a reviewer for thoughtful comments and suggestions.606
References607
Balcilar, M., & Genc, I. (2010). The links between crude oil prices and GCC stock markets: The time varying Granger causality608
tests. In EMU Economic Research Center, Working Paper, No: 2010-001.609
Balcilar, M., Demirer, R., & Hammoudeh, S. (2013). Investor herds and regime-switching: Evidence from Gulf Arab stock markets.610
Journal of International Financial Markets, Institutions and Money, 23, 295–321.611
Bikhchandani, S., Hirshleifer, D., & Welch, I. (1992). A theory of fads, fashion, custom, and cultural change as informational612
cascades. Journal of Political Economy, 100, 992–1026.613
Bikhchandani, S., & Sharma, S. (2001). Herd behavior in financial markets: A review. IMF Staff Papers, 47, 279–310.614
Blasco, N., Corredor, P., & Ferreruela, S. (2012). Does herding affect volatility? Implications for the Spanish stock market.615
Quantitative Finance, 12(2), 311–327.616
Chang, E. C., Cheng, J. W., & Khorana, A. (2000). An examination of herd behavior in equity markets: An international perspective.617
Journal of Banking and Finance, 24(10), 1651–1699.618
Cheng, A., Jahan-Parvar, M. R., & Rothman, P. (2010). An empirical investigation of stock market behavior in the Middle East and619
North Africa. Journal of Empirical Finance, 17, 413–427.620
Chiang, T. C., & Zheng, D. (2010). An empirical analysis of herd behavior in global stock markets. Journal of Banking and Finance,621
34(8), 1911–1921.622
Please cite this article in press as: Balcilar, M., et al. What drives herding in oil-rich, developing stockmarkets? Relative roles of own volatility and global factors. North American Journal of Economicsand Finance (2014), http://dx.doi.org/10.1016/j.najef.2014.06.009
ARTICLE IN PRESSG ModelECOFIN 451 1–23
22 M. Balcilar et al. / North American Journal of Economics and Finance xxx (2014) xxx–xxx
Christie, W. G., & Huang, R. D. (1995). Following the pied piper: Do individual returns herd around the market? Financial Analyst623
Journal, 1995(July–August), 31–37.624
Demirer, R., Kutan, A., & Chen, C. (2010). Do investors herd in emerging stock markets? Evidence from the Taiwanese market.625
Journal of Economic Behavior and Organization, 76, 283–295.626
Demirer, R. (2013). Can advanced markets help diversify risks in Frontier markets? Evidence from Gulf Arab Stock Markets.627
Research in International Business and Finance, 29, 77–98.628
Demirer, R., Kutan, A., & Zhang, H. (2013). Do ADR investors herd? Evidence from advanced and emerging markets. InternationalQ8629
Review of Economics and Finance (in press).630
Economou, F., Kostakis, A., & Philippas, N. (2011). Cross-country effects in herding behaviour: Evidence from four south European631
markets. Journal of International Financial Markets, Institutions and Money, 21(3), 443–460.632
Eitrheim, O., & Terävirta, T. (1996). Testing the adequacy of smooth transition autoregressive models. Journal of Econometrics,633
74, 59–75.634
Escribano, A., & Jordá, O. (1999). Improved testing and specification of smooth transition regression models. In P. Rothman (Ed.),635
Nonlinear time series analysis of economic and financial data (pp. 289–319). Boston: Kluwer.636
Granger, C. W. J., & Teräsvirta, T. (1993). Modelling nonlinear economic relationships. Oxford: Oxford University Press.637
Hamilton, J. D. (1988). Rational-expectations econometric analysis of changes in regime: An investigation of the term structure638
of interest rates. Journal of Economic Dynamics and Control, 12, 385–423.639
Hammoudeh, S., & Li, H. (2008). Sudden changes in volatility in emerging markets: The case of Gulf Arab stock markets.640
International Review of Financial Analysis, 17, 47–63.641
Hsieh, M., Yang, T., Yang, Y., & Lee, J. (2011). Evidence of herding and positive feedback trading for mutual funds in emerging642
Asian countries. Quantitative Finance, 11(3), 423–435.643
Karolyi, A., & Stulz, R. (2003). Are financial assets priced locally or globally? In G. Constantinides, G. Constantinides, et al. (Eds.),644
Handbook of the economics of finance. Amsterdam: North-Holland.645
Khalifa, A., Hammoudeh, S., & Otrano, E. (2013). Patterns of volatility transmissions within regime switching across GCC and646
global markets. International Review of Economics and Finance, 29, 512–524.647
Kim, C., & Nelson, C. R. (1998). Testing for mean reversion in heteroskedastic data II: Autoregression tests based on Gibbs-648
sampling-augmented randomization. Journal of Empirical Finance, 5, 385–396.649
Kim, C., Nelson, C. R., & Startz, R. (1998). Testing for mean reversion in heteroskedastic data based on Gibbs-sampling-augmented650
randomization. Journal of Empirical Finance, 5, 131–154.651
Krolzig, H. M. (1997). Markov switching vector autoregressions. Modelling, statistical inference and application to business cycle652
analysis. Berlin: Springer.653
Lakonishok, J., Shleifer, A., & Vishny, R. V. (1992). The impact of institutional trading on stock prices. Journal of Financial Economics,654
32, 23–43.655
Lee, C., Chen, M., & Hsieh, K. (2013). Industry herding and market states: Evidence from Chinese stock markets. Quantitative656
Finance, 13(7), 1091–1113.657
Luukkonen, R., Saikkonen, P., & Teräsvirta, T. (1988). Testing linearity against smooth transition autoregressive models.658
Biometrika, 75, 491–499.659
Mayfield, E. S. (1999). Estimating the market risk premium Working Paper. Harvard Business School.660
Marashdeh, H., & Shrestha, M. B. (2010). Stock market integration in the GCC countries. International Research Journal of Finance661
and Economics, 37, 102–114.662
Newey, W., & West, K. (1987). A simple positive semi-definite, heteroskedasticity and autocorrelation consistent covariance663
matrix. Econometrica, 55, 703–708.664
Nofsinger, J., & Sias, R. (1999). Herding and feedback trading by institutional and individual investors. Journal of Finance, 54,665
2263–2295.666
Pagan, A. R., & Schwert, G. W. (1990). Alternative models for conditional stock volatility. Journal of Econometrics, 45, 267–290.667
Peters, E. E. (1994). Fractal market analysis: Applying chaos theory to investment and economics. John Wiley and Sons.668
Philippas, N., Economou, F., Babalos, V., & Kostakis, A. (2013). Herding behavior in REITs: Novel tests and the role of financial669
crisis. International Review of Financial Analysis, 29, 166–174.670
Ravichandran, K., & Maloain, A. M. (2010). Global financial crisis and stock market linkages: Further evidence on GCC market.671
Journal of Money, Investment and Banking, 16, 46–56.672
Sarantis, N. (2001). Nonlinearities, cyclical behaviour and predictability in stock markets: International evidence. International673
Journal of Forecasting, 17, 459–482.674
Schaller, H., & van Norden, S. (1997). Regime switching in stock market returns. Applied Financial Economics, 7, 177–191.675
Schwert, G. W. (1989). Business cycles, financial crises, and stock volatility. Carnegie-Rochester Conference Series on Public Policy,676
31, 83–126.677
Sola, M., & Timmermann, A. (1994). Fitting the moments: A comparison of ARCH and regime switching models for daily stock returns.678
London Business School, Centre for Economic Forecasting. Discussion Paper No. DP 6-94.679
Stulz, R. (1984). Pricing capital assets in an international setting: An introduction. Journal of International Business Studies, Winter,680
55–73.681
Stulz, R. (1995). The cost of capital in internationally integrated markets: The case of Nestlé. European Financial Management,682
(March), 11–22.683
Tan, L., Chiang, T. C., Mason, J. R., & Nelling, E. (2008). Herding behavior in Chinese stock markets: An examination of A and B684
shares. Pacific-Basin Finance Journal, 16, 61–77.685
Teräsvirta, T. (1994). Specification, estimation and evaluation of smooth transition autoregressive models. Journal of the American686
Statistical Association, 89, 208–218.687
Teräsvirta, T. (1998). Modeling economic relationships with smooth transition regressions. In A. Ullah, & D. E. A. Giles (Eds.),688
Handbook of applied economic statistics (pp. 507–552). New York: Dekker.689
Teräsvirta, T., & Anderson, H. M. (1992). Characterizing nonlinearities in business cycles using smooth transition autoregressive690
models. Journal of Applied Econometrics, 7, S119–S136.691
Please cite this article in press as: Balcilar, M., et al. What drives herding in oil-rich, developing stockmarkets? Relative roles of own volatility and global factors. North American Journal of Economicsand Finance (2014), http://dx.doi.org/10.1016/j.najef.2014.06.009
ARTICLE IN PRESSG ModelECOFIN 451 1–23
M. Balcilar et al. / North American Journal of Economics and Finance xxx (2014) xxx–xxx 23
Tyssedal, J. S., & Tjostheim, D. (1988). An autoregressive model with suddenly changing parameters and an application to stock692
market prices. Applied Statistics, 37, 353–369.693
van Dijk, D., Teräsvirta, T., & Franses, P. H. (2002). Smooth transition autoregressive models – A survey of recent developments.694
Econometric Reviews, 21, 1–47.695
Wermers, R. (1999). Mutual fund trading and the impact on stock prices. Journal of Finance, 54, 581–622.696
Yao, J., Ma, C., & He, W. P. (2013). Investor herding behaviour of Chinese stock market. International Review of Economics and Q9697
Finance (in press).698
Yu, J. S., & Hassan, M. K. (2008). Global and regional integration of the Middle East and North African (MENA) stock markets.699
Quarterly Review of Economics and Finance, 48(3), 482–504.700
Zheng, T., & Zuo, H. (2014). Reexamining the time-varying volatility spillover effects: A Markov switching causality approach. Q10701
North American Journal of Economics and Finance (in press).702