Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

IIMB Management Review (2021) 33, 50–63

available at www.sciencedirect.com

ScienceDirect

journal homepage: www.elsevier.com/locate/iimb

Time-varying beta, market volatility and stress:


A comparison between the United States and
India
Gagari Chakrabartia,*, Ria Dasb

a
Department of Economics, Presidency University, Kolkata, West Bengal, India
b
Deloitte Haskins & Sells, Kolkata, West Bengal, India

Received 19 February 2018; revised form 24 December 2018; accepted 6 March 2021; Available online 23 March 2021

KEYWORDS Abstract This study examines the time-varying nature of industry betas in India and the United
Time-varying beta; States to explore whether their observed behaviours are independent of the extent of development of
Market risk; the financial market. Such betas relate to the movements, particularly volatility and stresses, in the
Volatility; relevant markets. During 1999–2017, we found significant transmission of volatility from the domestic
Stress index; market to the time-varying betas in both countries. The emerging market betas are further found to
MV GARCH model increase under the domestic market stress. The developed market betas, however, were able to avoid
market stresses or fall under stresses, thereby reducing the investment risk.
© 2021 Published by Elsevier Ltd on behalf of Indian Institute of Management Bangalore. This is an open
access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)

Introduction follows: (1) systematic or non-diversifiable market risk and


(2) unsystematic or asset-specific unique risk that might be
The standard portfolio construction theory recognises that var- diversified by increasing the number of assets in a portfolio.
ious financial assets respond differently to diverse economic In case a portfolio is well diversified, it is the market risk,
events because they possess unique risk–return profiles. A rather than a unique risk that is considered. Such market
diversified portfolio is based on the idea of combining various risks, following Sharpe (1964), may be quantified by the cor-
such asset classes, each with distinct attributes. However, a relation between asset returns or asset betas.
rational and risk-averse investor's best choice is always a port- The widely used capital asset pricing model (CAPM)
folio composed of less-risky assets that provide high return: a (Sharpe, 1964) considered an asset's expected excess return
“perfect” portfolio that is very difficult, if not impossible, to to be proportional to the expected excess return to the mar-
construct. Hence, an investor is concerned about how to pick ket and, thus, introduced “beta” as the covariance between
up assets to construct a portfolio that replicates or closely the selected asset return and the market return in propor-
resembles the perfect one? tion to the market return variance. Hence, beta measures
Literature shows that expected excess return to any asset the responsiveness of such asset return to market move-
is considered a function of its risk, which is categorised as ments or market risk, which the model considers constant
over time. Given that investors are concerned about the
market risk, betas help construct portfolios by allowing
*Corresponding author: Contact number: 9830243185. selection among various assets.
E-mail address: gagari_chakrabarti@yahoo.com (G. Chakrabarti).
https://doi.org/10.1016/j.iimb.2021.03.003
0970-3896 © 2021 Published by Elsevier Ltd on behalf of Indian Institute of Management Bangalore. This is an open access article under the CC
BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Time-varying beta, volatility and stress 51

Subsequent literature challenged the ideas advocated by Studies that suggest the second alternative estimate the
CAPM by questioning its effectiveness to explain cross-sec- time-varying beta using multivariate GARCH models in vari-
tional anomalies in the expected asset returns. Specifically, ous forms (Bollerslev, Engle, & Wooldridge, 1988;
the assumption of a constant beta was aggressively criticised Choudhry, 2005; Choudhry & Wu, 2008; De Santis &
and strong cases for considering beta as time dependent Gerard, 1998; Ng, 1991). There is, however, no conclusive
were established (Agarwal & Mangla, 2014; Basu & proof to justify the choice of one technique over the other
Stremme, 2007; Bos & Newbold, 1984; Brooks, Faff, & Lee, (Choudhry & Wu, 2008; Ebner & Neumann, 2005;
1992; Collins, Ledolter, & Rayburn, 1987; Fabozzi & Fran- Faff, Hillier, & Hillier, 2000).
cis, 1978; Ferson & Harvey, 1991; Jagannathan & Accepting a beta to be time variant adds another dimen-
Wang, 1996; Ohlson & Rosenberg, 1982). sion to the investment decision behaviour (Gonzales-Riv-
Such modifications are important as the models with a era, 1996, 1997). A time-varying beta might represent
time-varying beta can explain much of the observed cross- additional market risk or idiosyncrasies that could jeopar-
sectional dispersion in an expected asset return without dise the choice of the perfect portfolio. Efficient prediction
introducing additional factors, as emphasised by Fama and of a beta is, hence, necessary to help investors make effi-
French (1993, 1995) and Carhart (1997). Different stock mar- cient choices.
ket anomalies, such as the book-to-market effects, size pre-
mium or the profitability of momentum, can be considered
once beta is varied over time (Avramov & Chordia, 2004; The present study
Ball & Kothari, 1989; Black & Fraser, 2000; Campbell & Vuol-
teenaho, 2004; Chan & Chen, 1988; Fama & French, 1997; The present study examines the time-varying nature of beta,
Ferson & Harvey, 1999; Fraser, Hamelink, Hoesli, & Macgre- if any, in the context of two countries, namely, India and the
gor, 2000; Groenewold & Fraser, 1999). Furthermore, the United States, to explore whether the observed behaviours
time-varying betas can better explain the description of of beta are independent of the extent of development in the
return behaviour (Longstaff, 1989). financial markets. Ronzani et al. (2017) stated that different
Berk, Green, and Naik (1999) modelled the time-varying stocks follow (usually within a single country) different
betas of firms in the presence of growth opportunities and dynamism in the evolution of their betas over time. While
found that betas depend on the business cycles. However, such a dynamic nature of market risk, as evidenced by their
the betas of the different assets move differently over such dynamic betas, is important for investors who seek to mini-
business cycles (Jagannathan & Wang, 1996; Lettau & Lud- mise market risks, further dimensions may be added to their
vigson, 2001; Petkova & Zhang, 2005). Gomes, Kogan, and investment decisions if such risks vary among markets, par-
Zhang (2003) found that the betas depend on the business ticularly among those that differ significantly in terms of
cycle and factors such as firm size and market-to-book ratio. financial development. Efficiency in international portfolio
Although there is a need for a time-varying beta, the dis- diversification requires a clear comprehension of the asset
cussions on the choice of a suitable method for estimation price movements in the global markets. An investor, who
remain inconclusive (Nieto, Orbe, & Ainhoa, 2014). intends to keep a particular asset in the portfolio, is likely to
Lewellen and Nagel (2006) stated that allowing for condi- design his or her investment decision differently if the asset
tional time-varying dynamics may not overcome the prob- behaves similarly, and not otherwise, in different markets.
lems associated with the traditional CAPM. It is crucial to Hence, examining various possible differences in the
make appropriate assumptions about the unobservable con- observed behaviour of betas across financial markets has sig-
ditional information set of the investors for such estimation. nificant implications for academicians and investors who
Literature suggests two frameworks: one makes assumptions seek to operate at a global level. Hence, this study selects
about the dynamics of beta, while the other relates to the two financial markets that are at the two extreme ends of
conditional covariance matrix of the return series (Nieto financial development. According to the World Bank (2018)
et al., 2014). Some studies suggesting the first alternative report on financial development ranking of countries, the US
have used the Kalman filter to estimate beta on the assump- market scores high in terms of indexes of Business Freedom
tion that beta dynamics are determined by stochastic pro- (82.7), Trade Freedom (86.7), Investment Freedom (85) and
cesses (Mergner & Bulla, 2008; Moonis & Shah, 2003; Financial Freedom (80) compared with India's scoring of
Wells, 1994), while others have followed parametric 56.4, 72.4, 40 and 40, respectively. The choice of the coun-
approaches to estimate beta as depending on either state tries in this study is further justified by their world ranking
variables (Shanken, 1990) or firm-level parameters (Jagan- in terms of the Index of Economic Freedom that stands at 18
nathan & Wang, 1996; Lettau & Ludvigson, 2001). and 130 for the United States and India, respectively.
Ferreira, Gil, and Orbe (2011) used a non-parametric This study contributes to the literature by relating such
approach for estimating the same. Lin, Chen, and time-varying asset betas to the movements, particularly to
Boot (1992) and Lin and Lin (2000) considered betas as linear the volatility and stresses in the underlying or relevant mar-
and parabolic functions of time. Literature, however, fails kets. While literature finds additional idiosyncratic risks
to comment on the relative desirability of these methods. associated with the time-dependent betas, it overlooks the
Esteban and Orbe (2010), Li and Yang (2011), and Ang and possibility that such betas might respond, even asymmetri-
Kristensen (2012) non-parametrically estimated time-vary- cally, to the movements in the underlying market. There-
ing betas using the rolling least squares methods. This fore, the implications for investors to have changes in time-
approach is likely to overcome the problem of misspecifica- varying betas to be sensitive to market movements would
tion that arises from choosing a particular functional form. differ from the cases where the beta is either considered a
However, it involves the crucial choice of window length. constant or independent of market movement even if it
52 G. Chakrabarti, R. Das

varies. The possibilities of having a series of significantly stressed market. An equity market is under stress when its
higher betas during extreme market volatility or market return at any time falls below a floor established by past
stress would add to the investment risk, which investors returns defined over a window. An increase in investment
might find difficult, if not impossible to hedge. Moreover, risk might be feared when the betas are related to market
with growing financial integration, the global market along stress, particularly if they increase when a market plunges
with its domestic counterpart will probably have serious into stress. Although several studies (e.g.,
implications for asset betas and, hence, for the resultant Cardarelli, Elekdag, & Lall, 2009; Hakkio & Keeton, 2009;
investment risk. Hollo, Kremer, & Lo Duca, 2012) have analysed the issues
This study focuses on six important sectors or industries related to market stress, no attempt has thus far been made
from both countries using an 18-year daily database from 1 to relate the time-varying betas to the local and global mar-
February 1999 to 31 January 2017. This period is characterised ket stresses.
by two significant financial crises, namely, the 2000–2001 dot- Q5: Are the observed behaviours of beta independent of
com crisis and the 2007–2008 global financial meltdown. The the extent of development in the financial markets? The
study examines the movements in the time-varying beta during question gains significance when investors operate globally.
such volatility and stress in the financial market to understand When the time-varying beta of a particular industry behaves
the significance of considering the beta to be time varying differently in two markets at different levels of financial
rather than constant, from an investor's perspective. Specifi- market development, there are possibilities of arbitrage.
cally, the study addresses the following issues: Investors may choose among the sectors not only within a
Q1: Should the industry betas be considered time varying country but also across countries. This would help them min-
rather than constant? This question finds relevance in the avail- imise the otherwise non-diversifiable risks.
able literature that suggests considering beta as time variant
(Bos & Newbold, 1984; Brooks et al., 1992; Collins et al., 1987; Data and methodology
Fabozzi & Francis, 1978; Gonzales-Rivera, 1996, 1997; Ohlson
& Rosenberg, 1982). Such assertion seems valid as the static This study selects six industries, namely, transportation,
beta CAPM finds very little empirical support and applicability healthcare, information technology, industrials, banks, and
(Agarwal & Mangla, 2014; Alam, Chowdhury, & Chowdhury, telecommunications, from the Indian and the US market.
2015; Holthausen & Hughes, 1978). The Bombay Stock Exchange has introduced sectoral indices
Q2: If the beta is indeed time varying, how do we calcu- since 1999. The study uses daily data on such indices from 1
late it? This issue is much debated in the presence of alter- February 1999 to 31 January 2017. The US data have been
native models in the field. In literature, bivariate GARCH, collected from the official website of the National Associa-
BEKK, GARCH-GJR or the GARCH-X is used to model time- tion of Securities Dealers Automated Quotations (NASDAQ).
varying betas (Bodurtha & Mark, 1991; Bollerslev et al., Data on sectoral indices introduced after 1999 have been
1988; Brooks et al., 1992; Engle & Rodrigues, 1989; Gianno- collected from the date of their inception. The choice of
poulos, 1995; Gonzales-Rivera, 1997; Koutmos, Lee, & Theo- sectors or industries is contingent on the availability of the
dossiou, 1994; Ng, 1991; Yun, 2002). Other studies have used common sector data for both countries.
an alternative modelling technique, namely, the Kalman fil- The risk-free return has been obtained from the 91-day
ter (Black, Fraser, & Power, 1992; Faff et al., 2000; treasury bill rate (annualised) from RBI and Federal Reserve
Well, 1994). Schwert and Seguin (1990) suggested an alter- websites. BSE SENSEX, the benchmark index for Indian mar-
native market model with time-varying beta. ket, comprising the 30 most valuable stocks, is the domestic
Faff et al. (2000), however, found no significant difference market proxy for India. The domestic market proxy for the
when the method of computing differed. US sectors has been the Dow Jones Industrial Average of 30
Q3: If the beta changes over time, how does its move- stocks. The study takes Global Dow (GDOW) Index, the 150-
ment relate to the different phases of instability in the mar- stock index of corporations from around the world, as the
ket? Such markets involve both local and global markets global market proxy.
because in an era of financial integration, ignoring possibili- The daily returns are calculated using the logarithmic fil-
ties of volatility spillovers from other markets might affect ter Rt = ln(Pt/Pt1), where Pt is the industry index value at
the robustness of the results derived. The question finds rel- the t-th period.
evance because in the cases where market volatilities
directly affect a beta so that the relevant asset return
responds more intensely to the changes in market returns The constant beta CAPM
when market returns are highly volatile, might add to the
investment risk. While studies by Berk et al. (1999), The constant beta CAPM (Sharpe, 1964) is the basic model
Jagannathan and Wang (1996), Lettau and Ludvigson (2001), that is popularly used across the world to calculate expected
Gomes et al. (2003) and Petkova and Zhang (2005) found returns of stocks. The model in its original form may be
that betas depend on business cycle, these rarely associated described as follows.
the betas with market volatilities. The impacts from local  
Ri  Rf ¼ bi Rm  Rf ð1Þ
and global markets were not separately analysed either.
The issue is discussed in further detail in the following where Ri is the ith stock return; Rm is the market return; Rf is
question. the risk-free return and bi is the market sensitivity of the ith
Q4: How does a beta behave when the underlying rele- stock. However, some studies have questioned the use of the
vant market/s is (are) under stress? The question finds its OLS method to estimate such a model to comment on non-
relevance because a volatile market does not always mean diversifiable risks, particularly because the assumption of
Time-varying beta, volatility and stress 53

homoscedasticity may be violated, leading to the ineffi- suitable GARCH model in our study that works with 4551
ciency of the OLS estimator and to the betas not being accu- observations.
rately measured (Blume, 1971; Lin, Chen, & Boot, 1992; We use a suitably lagged diagonal VECH model for calcu-
Theil, 1971). Subsequently, studies have addressed the prob- lating the beta (Bollerslev, 1990; Bollerslev et al., 1988). A
lem by estimating a suitable ARCH model (Bera, Bubnys, & bivariate specification is used with the following specifica-
Park, 1988; Diebold, Im, & Lee, 1988) or GARCH model tion:
(Corhay, Rad, & Urbain, 1993). This study uses the GARCH
VECH ðHt Þ ¼ C þ A  VECH ðξ t1 ξ 0 t1 Þ þ B  VECH ðHt1 Þ ð3Þ
(1,1) model, based on the minimum Akaike Information Cri-
terion (AIC) to estimate (1). 

The GARCH (1,1) specification includes mean and vari- ξ t ct1 e Nð0; Ht Þ
ance equations as follows:
Mean equation : Yt ¼ xt0 u þ et ð2aÞ where VECH (Ht) is a vector half operator, stacking the lower
triangular of the symmetric Ht , a N £ N time-varying condi-
tional variance–covariance matrix.
Variance equation : s 2t ¼ v þ ae2t1 þ bs 2t1 ð2bÞ
2 3
h11t . . . h1Nt
The mean equation includes (Rm  Rf) as an exogenous
Ht ¼ 4 . . . . . . . . . 5
variable. The variance equation includes the conditional
variance s 2t that depends on three terms: (i) the one-period- hN1t ... hNNt
ahead forecast variance as a function of the lag of the If N = 2, a VECH(1,1) can be written as
squared residual from the mean equation e2t1 (the ARCH 2 3 2 3 2 32 3
term), (ii) the last period's forecast variance s 2t1 (the h11;t c11 a11 a12 a13 21; t1
GARCH term) and (iii) a constant. 4 h21;t 5 ¼ 4 c21 5 þ 4 a21 a22 a23 56 4 1; t1 ; 2: t1
7
5
h22;t c22 a31 a32 a33 22; t1
The phases of sectoral movement and the betas 2 32 3
b11 b12 b13 h11;t1
þ 4 b21 b22 b23 54 h21;t1 5
To justify the use of the time-varying beta, we isolate different b31 b32 b33 h21;t1
phases in the movements of the chosen indexes on the basis of
introspection. To check whether betas vary over such phases of aii, the diagonal elements, show the own spillover or the
market movements, the constant CAPM betas are estimated impact of own past innovations on the present volatility. The
separately for each phase using suitable GARCH models on the cross-diagonal terms (aij, i6¼j) show the impact of past inno-
basis of AIC criterion. If the betas differ significantly from one vation in one asset return on the present volatility of other
phase to another, the market risk may be assumed to be returns. Similarly, bii shows the impact of own past volatility
depending on market movement, justifying, thereby, the need on the present volatility and bij represents the cross-volatil-
to make the beta time varying rather than constant. To check ity spillover, that is, the impact of past volatility of one asset
for the redundancy of the constant beta CAPM, we focus on return on the present volatility of others’ returns.
the decomposition of the mean squared forecast error into ct1 is the information set at time t1, C is a N(N+1)/2 £
bias, variance, and covariance proportions. In case of a good 1 parameter vector, A and B are N(N+1)/2 £ N(N+1)/2
forecasting, the bias and variance proportions should be small parameter matrices. Considering the off-diagonal of the
such that most of the bias should be concentrated on the coefficient matrices A and B as 0, the conditional variances
covariance proportions. Hence, if the constant CAPM beta can be expressed as
model appears with a poor covariance proportion of the mean h11;t ¼ c11 þ a11 e21; þ b11 h11;t1
t1
squared forecast error and if the betas are found to differ sig-
nificantly over the phases, a stage for establishing time-variant
betas may be set. h22;t ¼ c22 þ a33 e22; t1 þ b33 h22;t1

and by assuming the correlations between the conditional


The time-varying beta variances constant
 1=2
Time-varying betas are calculated using a diagonal VECH h12;t ¼ r12 þ h11;t X h22;t
multivariate GARCH model. The selection of the model is Additionally, assuming all aii 0, all bii 0 and all cii> 0, the
justified on two grounds. First, studies (e.g., Faff et al., conditional beta is calculated as
2000) have documented no significant difference in the  
results with a change in the method for computing betas to Conditional Cov Rgreen ; Rmarket
bt ¼ ð4Þ
the Kalman filter; second, different GARCH family models Conditional VarðRmarket Þ
are found to capture the volatility dynamics well in the
Conditional Var(Rmarket) is estimated from h22;t and Condi-
Indian and US stock markets (e.g., Joshi, 2011; Joshi & Pan-
tional cov (Rgreen, Rmarket) is estimated from h12;t .
dya, 2008; Malik, 2011; Mohanty & Kamaiah, 2000;
This study calculates the domestic beta for each industry
Kaur, 2004; Padhi, 2006; Schwert, 1990; Whitelaw, 1994).
by estimating (1). For the Indian market, BSE SENSEX is con-
Moreover, Koutmos (2012) and Ronzani et al. (2017) showed
sidered the proxy for the domestic market and Dow Jones
that the method for applying the GARCH models to compute
Industrial Average is considered the proxy for the US domes-
the time-varying betas yields consistent results with a large
tic market.
number of data points. This further justifies the use of a
54 G. Chakrabarti, R. Das

Beta and market volatility Table 1 Constant CAPM beta for Indian sectors.

To verify whether the betas are related to market volatility, this Industry Domestic Covariance
study uses a suitable bivariate GARCH model as the one given in market beta proportion
(3), specifically to explore whether volatility in the relevant Bank 1.17* 0.25
markets spills over to the time-varying beta. Hence, the mean Automobile 0.86* 0.19
equation contains a time-varying beta as the explained variable, Industrial sector 0.10* 0.13
with the conditional variance of (Rm  Rf) as the explanatory Telecommunication 0.05** 0.12
variable. The variance–covariance equation, however, is not IT 0.02 0.11
modified. Any significantly positive Bij coefficient would docu- Healthcare 0.01 0.05
ment the presence of positive volatility spillover from the mar- *
and ** imply significance at 1% and 5% level, respectively.
ket such that increased market volatility would make time-
varying beta more volatile. The concerned industry might be
subject to additional risk when increased market volatility
makes asset return more sensitive to market movements.
This study examines whether the two types of betas, banking sector emerges as the only aggressive sector, while
namely, the domestic and global betas, are related to the the others remain defensive. The covariance proportions of
local and global market volatilities to explore all types of the mean squared forecast errors are found to be low for all
volatility transmission channels and identify market risk of cases, indicating the inefficiency of a constant beta CAPM.
investment at the industry level. We identified the phases considering the trends in price
Finally, this study maintains that all volatilities may not rep- movements in each industry separately and calculated the
resent a stressed situation. Investment risk might be magnified constant CAPM beta for each phase (Table 2). The results
if asset return's market sensitivities depend on market stress. suggest that beta changes over different phases of market
Any increase in market sensitivity (or, beta) with an increase in movements. Healthcare, for example, shows an insignificant
stress would make the investment in the concerned asset riskier. beta for the entire period on the basis of a constant CAPM,
while its betas varied significantly over the phases of price
The equity market stress index and time-varying beta movements observed for the sector and remained significant
in each phase, except for the first phase, similar to the IT
sector. In some phases, it even emerged a significantly
The study constructs a stress index following the CMAX
aggressive sector. The industrial sector with a very low over-
method as follows:
    all beta became an aggressive sector, particularly after
 2015. Telecommunication betas fluctuate over time and, in
CMAX ¼ Xt =max X 2 Xtj j ¼ 0; 1; . . . T ð5Þ
many cases, assume significantly high values. The betas of
automobile sector also fluctuate over phases and often
where Xt is the return from the concerned market (local or
assume values significantly higher than its constant beta.
global). The moving window is determined by T. Thus, CMAX
The banking sector remains aggressive throughout the
compares the current value of a variable with its maximum
period, except for the first two phases. However, in all
value over the previous T periods. Vila (2000) used this
cases, the estimated betas are either higher or lower than
method to identify the periods of slide in the stock market.
the constant beta.
The trigger level is considered at either 1.5 or 2 standard
The results obtained thus far are consistent with the lit-
deviations below the mean of the series.
erature (e.g., Berk et al., 1999; Jagannathan & Wang, 1996;
This study selects four window periods, namely, 15, 30,
Lettau & Ludvigson, 2001; Gomes et al., 2003; Petkova &
45 and 60, to define stress. Hence, it compares the current
Zhang, 2005) that highlights the inefficacy of constant beta
return from the market in proportion to the maximum return
CAPM to capture the true market risk of assets as the betas
over the previous 15, 30, 45 or 60 days. A market is in stress
vary over different phases of price movement of the asset
if the index value is less than 2 standard deviations below
concerned. The model's inefficiency in dynamic forecasting
the average return for that market. A trigger price is thus
further complicates the issue. Hence, while it is important
defined as the average market return less than 2 standard
to make asset betas time dependent, estimating the con-
deviations of the market return.
stant betas separately over different phases and their com-
The stress index then takes the value 1 (or the market is
parison would lead to further inefficiency. The method,
in stress) if CMAX<Trigger Price; and 0 otherwise.
while suffering from the fundamental shortcomings of the
To confirm whether time-varying betas relate to market
constant beta CAPM in each phase (including an error in
stress, we estimate a suitable GARCH model as the one in 2
forecasting as evidenced by a lower covariance proportion
(a) and 2(b). The mean equation considers time-varying
of the mean squared forecast error), might be sensitive to
betas as dependent on the relevant market stress.
how the phases are defined and is likely to be additionally
burdened by the loss of data in the process of such a phase-
Results and discussion wise estimation. An error in estimating the true market risk
of the asset concerned might thus be inevitable. The study
Case for India estimates a suitable MV GARCH model [as in 4(a) and 4(b)]
based on all 4551 data points to calculate the time-varying
Table 1 reports the constant CAPM betas for six Indian indus- betas for Indian industries. The movements in such betas
tries. Healthcare and IT have insignificant betas. The with their trends are shown in Fig. 1.
Time-varying beta, volatility and stress
Table 2 Indian industry betas over different phases of industry–price movements

Industry beta Phases


1 2 3 4 5 6 7 8
Healthcare 1.2.99–24.1.00 25.1.00–2.4.03 3.4.03–17.6.08 18.6.08–17.3.09 18.3.08–27.3.14 28.3.14–31.1.17
0.11 0.58* 0.67* 0.41* 0.48* 0.74*
Covariance 0.03 0.1 0.12 0.09 0.10 0.13
proportion
Automobile 2.2.99–18.1.00 19.1.00–12.5.03 13.5.03–2.11.07 5.11.07–27.10.08 28.10.08–3.2.2014 4.2.2014–31.1.17
0.82* 0.58* 0.93* 0.65* 0.87* 1.07*
Covariance 0.25 0.17 0.27 0.17 0.24 0.28
proportion
Banking 2.1.02–10.12.02 11.12.02–14.6.06 15.6.06–14.1.08 15.1.08–9.3.09 12.3.09–10.1.14 13.1.14–31.1.17
0.67* 1.00* 1.15* 1.16* 1.23* 1.26*
Covariance 0.12 0.21 0.22 0.22 0.22 0.23
proportion
Tele communi- 19.9.05–9.6.06 12.1.06–9.1.08 10.1.08–12.3.09 13.3.09–18.9.09 22.9.09–1.6.10 2.6.10–10.8.12 13.8.12–10.7.15 13.7.15–31.1.15
cation 0.52* 0.08 0.88* 0.56* 0.01 0.67* 0.04 0.89*
Covariance 0.15 0.05 0.21 0.15 0.03 0.18 0.04 0.21
proportion
Industrials 19.9.05–7.1.08 8.1.08–9.3.09 12.3.09–9.11.10 10.11.10–21.8.13 22.8.13–26.2.15 27.2.15–12.2.16 15.2.16–31.1.17
0.19* 0.95* 0.59* 0.10** 0.33* 1.15* 1.23*
Covariance 0.07 0.25 0.18 0.05 0.11 0.27 0.29
proportion
IT 2.2.99–21.2.00 22.2.00–21.9.01 24.9.01–30.4.03 2.5.03–19.2.07 20.2.07–5.3.09 6.3.09–26.4.13 29.4.13–31.1.17
0.03 1.77* 1.73* 1.02* 0.83* 0.84* 0.61*
Covariance 0.03 0.31 0.32 0.22 0.19 0.19 0.14
proportion
*
and ** imply significance at 1% and 5% level, respectively.

55
56 G. Chakrabarti, R. Das

Fig. 1 Time-varying beta (Indian healthcare).

Fig. 3 Time-varying beta (an Indian bank)


The polynomial trend fitted to the data shows that
healthcare betas are mostly positive and dropping towards
the latter half of the study period. The actual range of betas
is between 0.12 and 0.20, which differs from the range of
0.11 to 0.74 obtained when constant betas were calculated
for different phases of healthcare sector price movements.
While the latter method overestimates the market risk of
the sector, it overlooks the possibilities of having negative
betas over some phases, implying a negative covariance
between the sectoral and market return. An observation of
the timeline shows such phases (e.g. 1999–2000, 2007, 2008–
2009) as periods of disturbances in the Indian market. Nega- Fig. 4 Time-varying beta (Indian industries).
tive healthcare betas for those periods and an overall low
beta range for the sector might establish it as a “safe” place
for investment. This issue is discussed further in the later
sections.
The automobile betas (Fig. 2) showed a rising trend after
the 2007–2008 financial meltdown. The industry faced rising
betas during the dot-com crisis, followed by a period of fall-
ing betas. The range of 0.35–1.46 obtained for the Indian
automobile time-varying beta differs from the range of
0.58–1.07 obtained, whereas constant betas were calculated
for different phases of automobile sector price movements. Fig. 5 Time-varying beta (Indian telecommunications).
The banking index, since its inception in 2002, has shown
a rising trend in its positive betas in the range of 0.28 to with market returns. Moreover, some of these phases of neg-
1.62. The constant CAPM beta calculated for different ative betas for both sectors are associated with the stock
phases ranges between 0.67 and 1.26, making the risk- market downturns. The presence of low or even negative
averse investor underestimate the maximum risk and over- betas, however, may be insufficient to consider these sectors
estimate the minimum risk (Fig. 3). as safe places for investment (see Figs. 4 and 5).
The polynomial trend fitted to the data shows that Indian Although the IT sector started aggressively, its beta
IT, telecommunication and industrial sector betas vary near declined after the 1999–2000 dot-com crisis and has
zero. Since their inception in 2005, industrial and telecom- remained low ever since. This result contradicts our study
munication indexes experienced falling betas, which then results while using constant CAPM beta for different periods
continued to move in low range. Betas increased during (Fig. 6).
2012–2013 and then fell, eventually becoming negative in The results bear significant implications for investment
2016. The industrial sector has relatively more phases of behaviour. Although the betas vary across phases and thus
negative betas when its returns were negatively associated the need to make them time varying, the use of constant

Fig. 2 Time-varying beta (Indian automobile) Fig. 6 Time-varying beta (Indian IT).
Time-varying beta, volatility and stress 57

Table 3 Indian industry betas under alternative methods of computation.

Industry Range of beta when Range of time- Implication for risk-


constant beta CAPM is varying beta (2) averse investor
used across phases (1)
Bank 0.67 to 1.23 0.28 to 1.62 Range wider in (2). (1) underestimates the
Automobile 0.58 to 1.07 0.35 to 1.46 (1) underestimates the risk and benefit
Telecommunication 0.04 to 0.89 0.40 to 1.03 aggressiveness
and defensiveness of the asset
Industrial sector 0.10 to 1.23 0.40 to 0.72 (1) overestimates the asset's (1) overestimates
Healthcare 0.11 to 0.74 0.12 to 0.20 aggressiveness and u market risk
nderestimates its defensiveness
IT 0.03 to 1.77 0.03 to 1.30 Range shorter under (2).
(1) overestimates the
aggressiveness of the asset

CAPM betas for the different phases fails to capture the true in the time-varying beta. Second, it explores whether vola-
market sensitivity of Indian industries (Table 3). Market sen- tility in domestic and/or global market spills over to the
sitivities are either overestimated or underestimated with time-varying betas and the influence that stress in the rele-
serious implications for investors who seek to minimise the vant markets has on the time-varying betas. The possible
non-diversifiable risk on their investment. The market risks presence of volatility spillover from markets is explored by
and benefits of investing in the banking, automobile and estimating two separate bivariate GARCH models as spelt
telecommunication sectors are underestimated, whereas out in Eq. (3).
the risks for industrial, healthcare and IT sectors are overes- Table 4 reports the coefficients of estimated Bij from the
timated. Hence, deciding on the basis of time-varying beta three models.
affects the choice of sectors for possible investment. The study finds no transmission channel from which vola-
The values of beta, however, may be related to the move- tility in a particular sector may be transmitted to its time-
ments in the relevant markets or the sector. Sectoral betas varying beta.
that are closely related to the movements in the underlying The Indian sectoral betas, however, receive significant
sector would signify cyclicality so that market risks may be volatility spillover from the domestic and global markets,
predicted to remain at a particular level during upturns or with the extent of spillover being stronger from the former.
downturns of the sectors. This might add further risks to Within the sectors, spillover from the domestic market has
investment decisions, for example, in cases where market been stronger for the healthcare and IT sectors, followed by
risk increases during downturns. that for the telecommunication, banking, and automobile
A consistently lower beta may not be adequate to imply a sectors. The IT and telecommunication sector betas, how-
lower risk if the betas show greater volatility in a volatile ever, receive the strongest spillover from the global market,
market. To explore the possible presence of such idiosyn- followed by healthcare, banking, and automobile.
cratic risks, this study relates time-varying betas to market A volatile market leads to fluctuation of the market sensi-
movements. tivity of assets, thereby increasing the risk of investment.
To resolve these issues, the study examines a possible vol- This study, however, maintains that a volatile market does
atility transmission channel from the sector to its time-vary- not always indicate market stress. While volatility is mere
ing beta. Any affirmative answer would establish cyclicality fluctuations in returns, stress may be defined as a situation

Table 4 Volatility spillover from domestic and global markets to Indian time-varying beta.

Industry Bij (coefficient showing Bij (coefficient showing Bij (coefficient showing
volatility spillover volatility spillover from volatility spillover
from global market) domestic market) from sector)
Healthcare 0.38* 0.98* 0.14
Transportation 0.22* 0.90* 0.07
Banks 0.29* 0.93* 0.08
Telecommunication 0.42* 0.96* 0.03
Industrials 0.37* 0.93* 0.10
Information technology 0.42* 0.98* 0.10
*
and ** imply significance at 1% and 5% levels, respectively.
58 G. Chakrabarti, R. Das

where market returns fall significantly below some floor set

(60 days, 2SD)


by its historical average. Although investors find it difficult

0.0027*
to design hedging strategies in situations of stress, the risk

0.0005
0.0003

0.0020
0.0003

0.0023

0.0004
0.0030
0.0031

0.0018

0.0020
0.0000
of investment would further increase when the sensitivity of
asset returns to market movements is affected (or, particu-
larly, increases) in situations of market stress because in a
stressed situation, market moves downwards, and with an

(60 days, 1.5SD)


associated increasing beta, asset returns would continue to
respond to such falls by falling more than proportionately.

0.0029*
Table 5 shows how the local and global market stress affect

0.0007
0.0003

0.0024
0.0001

0.0026

0.0001
0.0000
0.0034

0.0018

0.0020
0.004*
Indian sectoral time-varying betas.
The Indian healthcare sector with an insignificant con-
stant CAPM beta faced significant market risks when the
movement of the beta over different phases of its price

(45 days, 2SD)


movements was considered. The volatility in the local and

0.0045*
0.0027*

0.0025*
global markets significantly spills over to raise its market

0.0006
0.0002

0.0024
0.0000

0.0022
0.0031

0.0024

0.0001
0.0002
sensitivity. Stress in the domestic market has a significant
positive impact on healthcare beta, at least during the 15-
and 30-day window. Hence, for healthcare sector, the
domestic market stress is considered significant. However,

(45 days, 1.5SD)


the global market stress does not affect both healthcare and
IT sectors. Telecommunication beta showed an erratic move-

0.0045*
0.0027*

0.0025*
ment over the different phases of its price movement.

0.0006
0.0002

0.0024
0.0002
0.0036

0.0024
0.0023

0.0001
0.0002
Although there are significant volatility spillovers from the
domestic and global markets to its time-varying beta, no
market stress affects such betas. The industrial and tele-
communication sectors are safe, however, with different
(30 days, 2SD)

characteristics. The telecommunication beta is not affected


by any market stress, whereas the industrial sector betas,
0.0013*

0.0027*

0.0024*
0.0011*
0.0003

0.0024
0.0031
0.0012

0.0028
0.0002
0.0019

0.0002
which are unaffected by domestic market stress, drop when
the global market is under stress. The automobile sector
betas behave similarly to those for the industrial sector.
However, the effect of the global market stress is more
(30 days, 1.5SD)

prominent on the industrial sector. The banking sector


Indian time-varying beta and domestic (D) and global (G) market stress.

remains risky because regardless of how stress is defined for


domestic market, such stresses always affect its betas.
0.0013*

0.0027*

0.0023*
0.0010*
0.0002
0.0031
0.0038

0.0022
0.0002
0.0026
0.0021

0.0002
Hence, for the Indian market, healthcare, banking and IT
sectors are subject to higher market risks as their time-vary-
ing betas increase significantly when the Indian market
sharply declines. Such sectoral betas are seldom related to
(15 days, 2SD)

global market risks, but whenever they do the phenomenon


favours the investors.
0.0026*

0.0010*
0.0006
0.0002
0.0027
0.0007

0.0030
0.0003
0.0021
0.0021
0.0023

0.0002

The results obtained for an emerging market, namely,


India, are compared with those obtained for the financially
developed market, such as the United States.
(15 days, 1.5SD)

Case for the United States


0.0013*

0.0029*

0.0025*
0.0010*
0.0002
0.0029
0.0002

0.0028
0.0001
0.0022
0.0022

0.0002

Table 6 shows the constant CAPM beta obtained for the six
US industries. While all betas are significantly positive at the
implies significance at 1% level.

1% level of significance, the IT sector has the highest value


of beta, followed by the telecommunication sector. These
two sectors, however, behave almost like the market, as is
Telecommunication (G)
Telecommunication (D)

evident from their beta values that remain close to 1. The


banking sector exhibits minimum market sensitivity. This is
Automobile (G)
Automobile (D)

in sharp contrast with the results obtained for the Indian


Healthcare (G)
Healthcare (D)

Industrials (G)
Industrials (D)

market.
Banks (G)
Banks (D)

Table 7 shows the movements in the US sectoral betas


Table 5

over the respective phases of their price movements. The


IT (G)
IT (D)

phases are identified considering the trends in the sectoral


*

index movements. For all sectors, the betas vary


Time-varying beta, volatility and stress 59

Table 6 Constant CAPM beta for the US sectors. respectively, with the assumption that a constant CAPM beta
would not be able to capture the true market sensitivity of
Industry Domestic Covariance the sectors.
market beta proportion The time-varying betas for the US sectors estimated using
Bank 0.44* 0.11 a suitable bivariate GARCH model, as shown in 4(a) and 4(b),
Transportation 0.65* 0.14 are shown in Figs. 7 to 12.
Industrial sector 0.82* 0.15 The US time-varying beta shows a declining trend (Fig. 7)
Telecommunication 0.98* 0.13 before the 2007–2008 global meltdown. It then shows a rising
IT 1.08* 0.14 trend until December 2015 and declines again. For a compa-
Healthcare 0.78* 0.14 rable period, a similar trend was observed for the Indian
* ** healthcare sector. The time-varying beta for the US trans-
/ implies significance at 1%/5% level.
portation sector declined during the dot-com crisis of 1999–
2000 (Fig. 8). It increased steadily thereafter until the finan-
cial meltdown of 2008, followed by a mild downward trend.
significantly over the phases and for some sectors, such vari- The Indian automobile sector beta, on the other hand, has
ation is quite significant. For example, the betas of health- shown greater variation with an increasing trend recently.
care sector with a constant CAPM beta 0.78 vary in the range The US banking sector shows a rising trend in beta from
of 0.68 to 1.10, with an average beta of 0.84. The transpor- 1999 to 2009 and starts declining thereafter (Fig. 9). How-
tation sector beta varies in the range of 0.37 to 1.06, with ever, there has been a recent increasing trend. The Indian
an average of 0.78. The constant CAPM, however, underesti- banking sector beta demonstrates a continuously rising
mates its beta at 0.65. The average beta of the banking sec- trend throughout the study period.
tor over phases is 0.60, which is much higher than its The US telecommunication beta shows a mild rising trend
constant CAPM value of 0.44. For sectors such as telecommu- until the financial meltdown of 2007–2008 (Fig. 10). It then
nication, IT and industrial sector, the constant CAPM beta is falls until 2015 and demonstrates a rising trend thereafter.
more than the average beta for the sectors over the phases. The trends are almost similar to those followed by its Indian
Similar to the results obtained for the Indian sectors, the counterpart, with the only exception that the betas of the
average betas over the phases are 0.95, 1.01 and 0.79, latter remained close to zero.

Table 7 The US industry betas over different phases of industry–price movements

Industry beta Phases


1 2 3 4 5 6
Healthcare 14.7.05–12.5-08 13.5.08–27.2.09 28.2.09–14.7.15 14.7.15–31.1.17
0.68* 0.74* 0.85* 1.10*
Covariance 0.15 0.17 0.19 0.21
proportion
Transportation 4.1.99–12.07.07 13.7.07–2.3.09 3.3.09–4.6.14 5.6.14–24.8.15 25.8.15–31.1.17
0.37* 1.06* 0.97* 0.82* 0.70*
Covariance 0.18 0.29 0.27 0.21 0.20
proportion
Banking 4.1.99–21.6.01 24.6.01–20.1.04 21.1.04–15.1.08 16.1.08–1.9.09 2.9.09–8.8.11 9.8.11–31.1.17
0.18* 0.20* 0.51* 0.92* 0.99* 0.77*
Covariance 0.11 0.11 0.15 0.18 0.18 0.16
proportion
Tele communi- 4.1.99–3.3.00 4.3.00–15.2.02 16.2.02–7.10.08 8.10.08–30.12.08 30.12.08–31.1.17
cation 0.82* 0.86* 0.98* 1.10* 1.00*
Covariance 0.18 0.18 0.19 0.21 0.21
proportion
Industrials 4.1.99–3.3.00 4.3.00–23.7.02 24.7.02–30.5.07 31.5.07–19.11.08 20.11.08–31.1.17
0.63* 0.66* 0.81* 0.92* 0.95*
Covariance 0.15 0.15 0.17 0.17 0.17
proportion
IT 4.1.99–9.3.00 10.3.00–22.7.02 23.7.02–8.10.07 9.10.07–12.11.08 13.11.08–6.7.12 7.7.12–31.1.17
0.97* 0.98 1.00* 1.00* 1.06* 1.04*
Covariance 0.21 0.22 0.24 0.24 0.26 0.26
proportion
*
and ** imply significance at the 1% and 5% levels.
60 G. Chakrabarti, R. Das

Fig. 7 Time-varying beta (US healthcare).


Fig. 11 Time-varying beta (US Industrial)

Fig. 8 Time-varying beta (US transportation)

Fig. 12 Time-varying beta (US IT)

concerned beta might take. For the banking, transportation


(automobile in context of India) and telecommunication sec-
tors, the results are identical.
The extent of volatility spillover from the sector, the
domestic and global markets to the time-varying beta for
the US industries is shown in Table 9. This table reports the
Bij coefficients when the relationship between the US time-
varying beta and relevant market's conditional variance
of return are modelled using a suitable model as is given by
Fig. 9 Time-varying beta (US banks)
Eq. (3).
No significant volatility spillover is found from the sectors
or global market to US sectoral betas. Volatility in the
domestic market, however, significantly spills over to such
betas. The extent of the domestic market volatility spillover
is maximum for the IT sector beta and minimum for the
healthcare beta.
Table 10 shows how the stress in domestic and global mar-
kets affects the US time-varying betas. The results differ
from those obtained for the Indian market. Time-varying
Fig. 10 Time-varying beta (US telecommunication). betas for industries such as healthcare, transportation,
banking and industrial sectors are significantly negatively
The US industrial sector beta showed a rising trend until related to domestic market stress. Hence, an increase in
2012 and then fell to rise again since latter half of 2016 domestic market stress leads to a fall in beta. The extent of
(Fig. 11). Such trends differ from those demonstrated by its drop in asset returns following a slide in market would con-
Indian counterpart. tinue to decline, thereby reducing the risk of investment.
Finally, the US IT sector beta is consistently close to the The telecommunication and IT sector betas, however,
value of 1 (Fig. 12). This is in sharp contrast to the Indian IT remain completely dissociated with domestic and global
sector beta movements that remained consistently close to market stress.
zero.
Table 8 justifies the use of time-varying beta rather than Conclusion
using constant beta CAPM across the phases of sectoral price
movements. In all cases, the latter underestimates the mar- While exploring the time-varying nature of beta in India and
ket risk of the industry concerned as it underestimates the the United States, this study found the inefficacy of the con-
maximum value of beta. Concurrently, it underestimates the stant beta CAPM to identify the true nature of market sensi-
benefit as it overestimates the minimum value that the tivity of different sectors. Moreover, the use of such a model
Time-varying beta, volatility and stress 61

Table 8 US industry betas under alternative methods of computation

Industry Range of beta when Range of time- Implication for risk-


Constant beta CAPM varying beta (2) averse investor
is used across phases (1)
Banking 0.18 to 0.99 0.01 to 1.52 Range wider in (2). (1) underestimates risk
Transportation 0.37 to 1.06 0.03 to 1.49 (1) underestimates and benefit (same as in India)
Telecommunication 0.82 to 1.10 0.53 to 1.30 the aggressiveness and
Industrial sector 0.63 to 0.95 0.51 to 1.15 defensiveness of the asset (1) underestimates risk
Healthcare 0.68 to 1.10 0.38 to 1.44 and benefit
IT 0.97 to 1.06 0.81 to 1.18

Table 9 Volatility spillover from domestic and global markets to US time-varying beta

Industry Bij (coefficient showing Bij (coefficient showing Bij (coefficient


volatility spillover from volatility spillover showing volatility
global market) from domestic market) spillover from sector)
Healthcare 0.04 0.91* 0.03
Transportation 0.11 0.93* 0.09
Banks 0.08 0.93* 0.04
Telecommunication 0.04 0.93* 0.02
Industrials 0.02 0.93* 0.12
Information Technology 0.11 0.95* 0.08
*
and ** imply significance at the 1% and 5% level, respectively.

separately for different phases of the sectoral price move- betas that are independent of stress in the domestic market
ment and a comparison of betas thus obtained failed to cap- decline under global market stress. IT and healthcare betas
ture the true dynamics of beta movement. This remains true increase significantly under domestic market stress. Hence,
irrespective of the level of financial development of the within India, telecommunication, automobile and industrial
countries considered, although with some differences in the betas might attract investors who seek to minimise the mar-
nature of the problem. In an emerging and financially less- ket risk of investment. For the US market, IT and healthcare
developed country such as India, the use of such a method betas are the worst affected by domestic market volatility
leads to an overestimation of market risk for sectors such as spillover. Global market volatility, however, does not spill
healthcare, industrial sector, and IT. It, however, underesti- over to any of the betas. The US IT and telecommunication
mates the risks and benefits of investing in banking, automo- betas remain completely dissociated from domestic market
bile, and telecommunication sectors. In the US market, in stress. The betas for the other sectors fall when the domes-
contrast, this method leads to an underestimation of the tic market is under stress. The sectoral upheavals do not
risk and benefit for all sectors. Quantitative techniques such affect the US time-varying betas.
as the iterative cumulative sum of squares (ICSS) test intro- The results bear significant implications for investors who
duced by Inclan and Tiao (1994) used in identifying the seek to minimise the market risk of investment. These per-
phases endogenously might be unsuitable because the prob- haps modify the endemic concern that time-variant beta
lems associated with constant beta CAPM would persist even adds to the risk of investment. Such time-varying sectoral
when used in phases separately. The study thus calculates betas in India and the United States face significant volatility
the time-varying beta for the two countries and relates transmission from their respective domestic markets. The
those to market volatility and particularly to market stress. emerging market sectoral betas face volatility spillover
All Indian sector betas face a significant volatility spill- from global markets, but their developed market counter-
over from the domestic market and a relatively mild trans- parts are not susceptible to these spillovers. The developed
mission from the global market. The upheavals at the market betas can avoid market stress or even collapse,
sectoral levels, however, cannot affect the time-varying thereby making the asset under consideration less risky. This
betas. Of the chosen sectors, such a spillover greatly affects is often true for the emerging market betas, but there are
the IT beta. The impact on the banking and automobile sec- cases where such betas increase under stress. This is particu-
tor betas has been less pronounced. While all betas face larly true for the Indian IT sector. Hence, making betas time
spillover from markets, they behave differently under mar- variant does not always lead to additional market risk for
ket stress. Telecommunication sector betas are not affected investors and might help in hedging. However, concerns per-
by any market stress, whereas the automobile and industrial sist about the emerging market.
62 G. Chakrabarti, R. Das

References

(60days, 2SD)
0.0038*

0.0043*
0.0046*
0.0059*

0.0038*
Agarwal, R., & Mangla, J. (2014). Testing practical applicability of

0.0014

0.0020

0.0020
0.0002
0.0020
0.0018

0.0004
CAPM: A study of stocks of automobile sector using CNX auto
index in NSE. International Journal of Advanced Research in
Management and Social Science 3 (1) [online] Available at: www.
garph.co.uk.
Alam, M.R., Chowdhury, E.K., & Chowdhury, T.U. (2015). Application
(60days, 1.5SD)

of capital asset pricing model: Empirical evidences from Chitta-


gong stock market. The Cost of Management 43 (3), 38–44.
0.0031*

0.0043*
0.0045*
0.0052*

0.0031*
0.0009

0.0024

0.0020
0.0002
0.0016
0.0018

0.0000
Ang, A., & Kristensen, D. (2012). Testing conditional factor models.
Journal of Financial Economics 106 (1), 132–156.
Ball, R., & Kothari, S.P. (1989). Nonstationary expected returns:
Implications for tests of market efficiency and serial correlation
in returns. Journal of Financial Economics 25, 51–74.
(45days, 2SD)

Basu, D., & Stremme, A. (2007). CAPM and time-varying beta: The
cross-section of expected returns. http://ssrn.com/
0.0027*

0.0039*
0.0045*
0.0042*

0.0027*
0.0025*
0.0009

0.0024

0.0004
0.0010
0.0024

0.0002
abstract=972255
Bera, A.K., Bubnys, E., & Park, H. (1988). Conditional heteroscedas-
ticity in the market model and efficient estimates of betas.
Financial Review 23 (2), 201–214.
Berk, J., Green, R.C., & Naik, V. (1999). Optimal investment,
(45days, 1.5SD)

growth options, and security returns. Journal of Finance 54 (5),


1553–1608.
Black, A. & Fraser, P. (2000). Are stock prices too volatile and
–0.0033*

0.0042*
0.0045*
0.0049*

0.0033*
0.0025*
0.0008

0.0024
0.0014
0.0024

0.0002
0.0002

returns too high? A reassessment of the empirical evidence using


a dynamic version of the CAPM. Working Paper, University of
Aberdeen.
Black, A., Fraser, P., & Power, D. (1992). UK unit trust performance
1980-1989: A passive time-varying approach. Journal of Banking
(30days, 2SD)

and Finance 16 (5), 1015–1033.


Blume, M.E. (1971). On the assessment of risk. The Journal of
0.0031*

0.0039*
0.0045*
0.0055*

0.0031*
0.0024*
0.0010

0.0028
0.0017
0.0019

0.0002
0.0002

Finance 26 (1), 1–10.


Bodurtha, J.N., & Mark, N. (1991). Testing the CAPM with time-vary-
ing risks and returns. Journal of Finance 46 (4), 1485–1505.
Bollerslev, T. (1990). Modelling the Coherence in Short-Run Nominal
Exchange Rates: A Multivariate Generalized ARCH Approach.
(30days, 1.5SD)

Review of Economics and Statistics 72, 489–505.


US Time-varying beta and domestic (D) and global (G) market stress

Bollerslev, T., Engle, R., & Wooldridge, J. (1988). A capital asset


0.0031*

0.0040*
0.0044*
0.0050*

0.0031*
0.0023*
0.0008

0.0022
0.0013
0.0026

0.0082
0.0002

pricing model with time-varying covariances. Journal of Political


Economy 96 (1), 116–131.
Bos, T., & Newbold, P. (1984). An empirical investigation of the pos-
sibility of stochastic systematic risk in the market model. Jour-
nal of Business 57 (1), 35–41.
Brooks, R.D., Faff, R.W., & Lee, J.H. (1992). The form of time varia-
(15days, 2SD)

tion of systematic risk: Some Australian evidence. Applied Finan-


0.0029*

0.0039*
0.0046*
0.0042*

0.0029*
0.0023*
0.0009

0.0030
0.0010
0.0021

0.0004
0.0002

cial Economics 2, 191–198.


Campbell, J., & Vuolteenaho, T. (2004). Bad beta, good beta. Amer-
ican Economic Review 94, 1249–1275.
Cardarelli, R., Elekdag, S., & Lall, S. (2009). Financial stress, down-
turns, and recoveries. IMF Working Papers, WP/09/100.
(15days, 1.5SD)

Carhart, M.M. (1997). On persistence in mutual fund performance.


The Journal of Finance 52 (1), 57–82.
Chan, K.C., & Chen, Nai-Fu (1988). An unconditional asset-pricing
0.0032*

0.0039*
0.0043*
0.0052*

0.0032*
0.0025*
0.0005

0.0028
0.0010
0.0022

0.0003
0.0002

test and the role of firm size as an instrumental variable for risk.
The Journal of Finance 43 (2), 309–325.
Choudhry, T. (2005). Time-varying beta and the Asian financial crisis:
implies significance at 1% level.

Evidence from Malaysian and Taiwanese firms. Pacific-Basin


Finance Journal 13 (1), 93–118.
Telecommunication (G)
Telecommunication (D)

Choudhry, T., & Wu, H. (2008). Forecasting ability of GARCH vs Kal-


man filter method: Evidence from daily UK time-varying beta.
Transportation (G)
Transportation (D)

Journal of Forecasting 27 (8), 670–689.


Healthcare (G)
Healthcare (D)

Collins, D., Ledolter, J., & Rayburn, J. (1987). Some further evi-
Industrials (G)
Industrials (D)
Banking (G)

dence on the stochastic properties of systematic risk. Journal of


Banking (D)

Business 60 (3), 425–448.


Table 10

Corhay, A., Rad, A.T., & Urbain, J.P (1993). Common stochastic
IT (G)
IT (D)

trends in European stock markets. Economics Letters 42 (4),


385–390.
*
Time-varying beta, volatility and stress 63

De Santis, G., & Gerard, B. (1998). How big is the premium for cur- Koutmos, D. (2012). Time-Varying Behavior of Stock Prices, Volatility
rency risk? Journal of Financial Economics 49 (3), 375–412. Dynamics and Beta Risk in Industry Sector Indices of the Shanghai
Diebold, F.X., Im, J., & Lee, C.J. (1988). Conditional heteroscedas- Stock Exchange. Accounting and Finance Research 1 (2), 109–125.
ticity in the market. Finance and Economics Discussion Series, Koutmos, G., Lee, U., & Theodossiou, P. (1994). Time-varying betas
42, Division of research and Statistics, Federal Reserve Board, and volatility persistence in international stock markets. Journal
Washington, D.C. of Economics and Business 46 (2), 101–112.
Ebner, M., & Neumann, T. (2005). Time-varying betas of German stock Lettau, M., & Ludvigson, S. (2001). Resurrecting the (C)CAPM: A
returns. Financial Markets and Portfolio Management 19 (1), 29–46. cross-sectional test when risk premia are time-varying. Journal
Engle, C., & Rodrigues, A. (1989). Tests of international CAPM with of Political Economy 109, 1238–1287.
time varying covariances. Journal of Applied Econometrics 4 Lewellen, J., & Nagel, S. (2006). The conditional CAPM does not
(1989), 119–138. explain asset-pricing anomalies. Journal of Financial Economics
Esteban, M.V., & Orbe-Mandaluniz, S. (2010). A nonparametric 82, 289–314.
approach for estimating betas: The smoothed rolling estimator. Li, Y., & Yang, L. (2011). Testing conditional factor models: A non-
Applied Economics 42 (10), 1269–1279. parametric approach. Journal of Empirical Finance 18, 975–992.
Fabozzi, F.J., & Francis, J.C. (1978). Beta as a random coefficient. Lin, H.J., & Lin, W.T. (2000). A dynamic and stochastic beta and its
Journal of Financial and Quantitative Analysis 13 (1), 101–116. implications in global capital markets. International Finance 3
Faff, R.W., Hillier, D., & Hillier, J. (2000). Time varying beta risk: An (1), 123–160.
analysis of alternative modelling techniques. Journal of Business Lin, W.T., Chen, Y.H., & Boot, J.C. (1992). The dynamic and stochas-
Finance and Accounting 27 (5), 523–554. tic instability of betas: Implications for forecasting stock
Fama, E.F., & French, K.R. (1993). Common risk factors in the returns on returns. Journal of Forecasting 11, 517–541.
stocks and bonds. Journal of Financial Economics 33, 3–56. Longstaff, F.A. (1989). A non-linear general equilibrium model of the
Fama, E.F., & French, K.R. (1995). Size and book-to-market factors term structure of interest rates. Journal of Financial Economics
in earnings and returns. The Journal of Finance 50 (1), 131–155. 23, 195–224.
Fama, E.F., & French, K.R. (1997). Industry costs of equity. Journal Malik, F. (2011). Estimating the impact of good news on stock mar-
of Financial Economics 43, 153–193. ket volatility. Applied Financial Economics 21, 545–554.
Ferreira, E., Gil, J., & Orbe, S. (2011). Conditional beta pricing Mergner, S., & Bulla, J. (2008). Time-varying beta risk of pan-Euro-
models: A nonparametric approach. Journal of Banking and pean industry portfolios: A comparison of alternative modelling
Finance 35, 3362–3382. techniques. European Journal of Finance 14, 771–802.
Ferson, W.E., & Harvey, C.R. (1991). The variation of economic risk Mohanty, P. K., & Kamaiah, B. (2000). Volatility and its persistence
premiums. Journal of Political Economy 99 (2), 385–415. in Indian stock market: A case of 30 scrips. Working Paper 53,
Ferson, W.E., & Harvey, C.R. (1999). Conditioning variables and the Institute for Social and Economic Change, Bangalore.
cross section of stock returns. The Journal of Finance 54 (4), Moonis, S.A., & Shah, A. (2003). Testing for time-variation in beta in
1325–1360. India. Journal of Emerging Market Finance 2, 163–180.
Fraser, P., Hamelink, F., Hoesli, M., & Macgregor, B. (2000). Time- Ng, L. (1991). Tests of the CAPM with time-varying covariances: A
varying betas and the cross-sectional return-risk relation: Evi- multivariate GARCH approach. The Journal of Finance 46 (4),
dence from the UK. Working Paper, University of Aberdeen. 1507–1520.
Giannopoulos, K. (1995). Estimating the time varying components of Nieto, B., Orbe, S., & Ainhoa, Z. (2014). Time-varying market beta:
international stock markets' risk. European Journal of Finance 1 Does the estimation methodology matter? SORT 38 (1), 13–42.
(2), 129–164. Ohlson, J., & Rosenberg, B. (1982). Systematic risk of the CRSP equal-
Gomes, J., Kogan, L., & Zhang, L. (2003). Equilibrium cross section weighted common stock index: A history estimated by stochastic-
of returns. Journal of Political Economy 111 (4), 693–732. parameter regression. Journal of Business 55 (1), 121–145.
Gonzalez-Rivera, G. (1996). Time-varying risk. The case of the Padhi, P. (2006). Stock market volatility in India: A case of select
American computer industry. Journal of Empirical Finance 2 scripts. https://papers.ssrn.com/sol3/papers.cfm?
(1996), 333–342. abstract_id=873985
Gonzalez-Rivera, G. (1997). The pricing of time-varying beta. Petkova, R., & Zhang, L. (2005). Is value riskier than growth? Jour-
Empirical Economics 22 (1997), 345–363. nal of Financial Economics 78, 187–202.
Groenewold, N., & Fraser, P. (1999). Time-varying estimates of CAPM Ronzani, A.R.de Pinho, Candido, O., & Maldonado, W.F.L (2017).
betas. Mathematics and Computers in Simulation 48, 531–539. Goodness-of-Fit versus Significance: A CAPM Selection with
Hakkio, C.S., & Keeton, W.R. (2009). Financial stress: What is it, Dynamic Betas Applied to the Brazilian Stock Market. Interna-
how can it be measured, and why does it matter? Federal tional Journal of Financial Studies 5 (4), 1–21.
Reserve Bank of Kansas City Economic Review 94 (2), 5–50. Schwert, G., & Seguin, P. (1990). Heteroscedasticity in stock
Hollo, D., Kremer, M., & Lo Duca, M. (2012). CISS – A composite indi- returns. Journal of Finance 45, 1129–1155.
cator of systemic stress in the financial system. ECB Working Schwert, G.W. (1990). Stock returns and real activity: A century of
Paper No. 1426, March. evidence. Journal of Finance 45, 1237–1257.
Holthausen, D.M., & Hughes, J.S. (1978). Commodity returns and Shanken, J. (1990). Intertemporal asset pricing: An empirical inves-
capital asset pricing. Financial Management 7 (2), 37–44. tigation. Journal of Econometrics 45, 99–120.
Inclan, C., & Tiao, G.C. (1994). Use of cumulative sums of squares Sharpe, W.F. (1964). Capital asset prices: A theory of market equilibrium
for retrospective detection of changes of variance. Journal of under conditions of risk. The Journal of Finance 19 (3), 425–442.
the American Statistical Association 89 (427), 913–923. Theil, H. (1971). Principles of econometrics. Wiley, New York.
Jagannathan, R., & Wang, R. (1996). The conditional CAPM and the Vila, A. (2000). Asset price crises and banking crises: Some empirical
cross-section of expected returns. Federal Reserve Bank of Min- evidence. BIS Conference Papers 8 (March), 232–252.
neapolis Research Department Staff Report 208. Wells, C. (1994). Variable betas on the Stockholm exchange 1971–
Joshi, P., & Pandya, K. (2008). Exploring movements of stock price vola- 1989. Applied Financial Economics 4, 75–92.
tility in India. The ICFAI Journal of Applied Finance 14 (3), 5–32. Whitelaw, R.F. (1994). Time variations and covariations in the
Joshi, P. (2011). Return and volatility spillovers among Asian stock expectation and volatility of stock market returns. Journal of
markets. Sage Open 1–8. Finance 49, 515–541.
Kaur, H. (2004). Time varying volatility in the Indian stock market. Yun, J. (2002). Forecasting volatility in the New Zealand stock mar-
Vikalpa 29 (4), 25–42. ket. Applied Financial Economics 12 (2002), 193–202.

You might also like