Emerging Markets Review 19 (2014) 7795

Contents lists available at ScienceDirect

Emerging Markets Review

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

Can institutions and macroeconomic factors

predict stock returns in emerging markets?
Paresh Kumar Narayan a,1, Seema Narayan b,2, Kannan Sivananthan Thuraisamy a,
a
b

School of Accounting, Economics and Finance, Deakin University, Melbourne, Australia

School of Economics, Finance and Marketing, RMIT University, Melbourne, Australia

a r t i c l e

i n f o

Article history:
Received 13 January 2014
Received in revised form 13 March 2014
Accepted 3 April 2014
Available online 13 April 2014
Keywords:
Predictability
Returns
Mean-variance investor
Institutions

a b s t r a c t
In this paper we test for predictability of excess stock returns for 18
emerging markets. Using a range of macroeconomic and institutional
factors, through a principal component analysis, we nd some
evidence of in-sample predictability for 15 countries. In-sample
predictability is corroborated by out-of-sample tests. Using a
mean-variance investor framework, we show that investors in most
of these emerging markets can make signicant prots from dynamic
trading strategies. Finally, we show that investors in most countries
where short-selling is prohibited could make signicant gains if
limited borrowing and short-selling were allowed.
2014 Elsevier B.V. All rights reserved.

1. Introduction
In this paper our focus is on the predictability of stock market returns in emerging markets. The literature
is voluminous. See, for example, the long list of inuential papers cited in Ferreira and Santa-Clara (2011) and
Westerlund and Narayan (2014a,b). Two directions are popular. One stream of studies considers whether
returns are predictable using macroeconomic indicators, while the other stream considers nancial ratio
predictors and, at best, the results are mixed. The main issue is that in-sample and out-of-sample tests
produce conicting results, which is problematic. The background to the existing tension is as follows.
Generally, in-sample tests of return predictability have found some encouraging results favouring
predictability, which is best summarised by Lettau and Ludvigson (2001, p. 842), It is now widely accepted
that excess returns are predictable by variables such as dividendprice ratios, earningsprice ratios,
dividendearnings ratios, and an assortment of other nancial indicators.
Corresponding author at: 70 Elgar Road, Burwood Highway, Centre for Financial Econometrics, School of Accounting, Economics
and Finance, Deakin University, Australia. Tel.: + 61 3 9244 6913.
E-mail addresses: paresh.narayan@deakin.edu.au (P.K. Narayan), seema.narayan@rmit.edu.au (S. Narayan),
sivananthan.thuraisamy@deakin.edu.au (K.S. Thuraisamy).
1
Tel.: + 61 3 924 46180.
2
Tel.: + 61 3 9925 5890.

http://dx.doi.org/10.1016/j.ememar.2014.04.005
1566-0141/ 2014 Elsevier B.V. All rights reserved.

78

P.K. Narayan et al. / Emerging Markets Review 19 (2014) 7795

This literature has attracted criticism though. On econometric grounds, there is not one main issue but
many: (a) the predictor variables have been highly persistent, and when corrected produce even weaker
evidence of predictability (see Ang and Bekaert, 2007; Stambaugh, 1999); (b) the predictive model's
errors are correlated with predictor innovations (see Lewellen, 2004); (c) data mining (see Foster et al.,
1997; Rapach and Wohar, 2006a); (d) parameter instability, so much so that the hypothesis of a constant
regression coefcient is almost always rejected (see Paye and Timmermann, 2006); (d) heteroskedasticity
(Westerlund and Narayan, 2012, 2013, 2014a); and cross-sectional dependence (Westerlund and
Narayan, 2014b). Persistency and correlation between returns and predictor innovations have the
tendency to bias the regression coefcients and the ensuing t-statistics on which the null hypothesis of no
predictability is based (see, inter alia, Lewellen, 2004; Stambaugh, 1999). Compared with in-sample tests
of return predictability, there are limited studies on out-of-sample tests. Of the limited studies (see Welch
and Goyal, 2008 and the references therein) that exist, the evidence is generally negative. Welch and
Goyal (2008) represent a comprehensive analysis of stock return predictability. They consider a wide
range of predictor variables and perform both in-sample and out-of-sample tests. They conclude that most
models do not reveal predictability of returns, which is true both in-sample and out-of-sample. Thus, they
claim that predictive regression models would not have helped an investor with access only to available
information to protably time the market (p. 1455). The lack of consensus on return predictability has
motivated much of the recent literature, with Ferreira and Santa-Clara (2011, p. 515) claiming that: The
predictability of stock market returns remains an open question.
While the focus of much of the stock return predictability literature has been on nancial ratio
predictors, it is unknown whether other non-nancial predictors also fail to predict returns in-sample and
out-of-sample and, thus, the question remains unanswered.3 It is well-known that emerging market risk
return characteristics are different compared to developed markets. Compared to developed markets, for
instance, emerging markets are highly volatile and provide attractive returns. Harvey (1995) argues that
emerging markets are segmented with high degree of return predictability.4 On the issue of predictability
of returns in emerging markets, Hjalmarsson (2010) nds that fundamentals, such as earningprice ratio
and dividendprice ratio, have reasonable ability to predict stock returns of emerging markets.
The goal of this paper is to examine whether macroeconomic factors and institutional factors predict
excess returns. We consider institutional risks associated with corruption, ethnic tension, internal
country-specic conicts, and law and order, and macroeconomic risks associated with up to 10
macroeconomic indicators. While some studies, such as Ferson and Harvey (1994, 1998), consider
macroeconomic predictors, none of the studies has considered the role of institutions in predicting
returns.5 Our analysis is conducted on time series data and considers 18 developing countries.
The contribution of our paper is three-fold. First, we focus on emerging markets where institutions play
an important role in the performance of stock markets. Therefore, for the 18 countries we choose, there is
a rich data set on institutions. This allows us to gain more insights on the role of institutions. We are also
able to examine whether or not investors can make use of the information content in institutions to make
non-negligible prots in developing countries. In addition, we also learn from the literature that
macroeconomic indicators are successful predictors of returns, although studies on this subject are
limited. Therefore, we also entertain macroeconomic predictors of returns. It follows that our use of both
institutional and macroeconomic variables as predictors of returns for emerging markets is unique and,
therefore, offers a fresh perspective on return predictability. Moreover, with limited emphasis on
developing country markets, very little is known about the role of short-selling. We contribute to this
literature as well. We nd that there are nine countries in our sample in which short-selling is prohibited.
3
The study that comes closest to our work is Mateus (2004), who examines stock return predictability both for individual
countries and for a cross-section of 13 EU accession countries. Related studies on return predictability have used other approaches to
testing for predictability; for an example, see Kinnunen (2013) and Eterovic and Eterovic (2013).
4
Lack of integration of emerging markets with developed markets has also been conrmed in more recent studies (see Bekaert et
al., 2011; Cakici et al., 2013).
5
There are, however, related studies that have generally considered the effects of institutions on specic rm issues. DemirgucKunt and Maksimovic (1999) explain the role of institutions in debt composition for both developed and developing countries. The
impact of religion on market outcomes has been considered by Kumar et al. (2011). The role of governance at the rm level on stock
returns has been considered by Core et al. (2006). The relationship between societal norms and nancial sector development has
been analysed by Garretsen et al. (2004).

P.K. Narayan et al. / Emerging Markets Review 19 (2014) 7795

79

We estimate prots allowing for limited borrowing and short-selling for those nine countries and show
how much better off investors in those countries could be.
Second, our estimation approach is different from the extant literature. We propose a conditional
capital asset pricing model (CAPM)-based predictive regression framework. This approach departs from
the cross-sectional and/or panel data approaches, thus ensuring that we can easily pick up any
heterogeneous effects of institutions and, for that matter, macroeconomic factors on returns in each of the
18 countries. This point is relevant as institutional strength and macroeconomic settings in different
developing countries are different, as several studies have documented, so it is not news. The data we use
reveal that institutional and macroeconomic conditions vary from one developing country to another; and,
as a result, these are likely to have a heterogeneous effect on returns at the country level. However, to-date
this has not been tested, thus, nothing is known about this subject.
Third, nothing is known about the economic signicance of return predictability when one uses institutions
and macroeconomic variables as predictors for emerging markets. We undertake a rigorous analysis of
protability using both passive and dynamic trading strategies. Moreover, drawing on a mean-variance
investor framework, we estimate investor utilities for each of the countries for which predictability exists.
We arrive at a number of new ndings which can be summarised as follows. First, we nd that excess
returns are predictable for 15 countries. Moreover, using popular out-of-sample tests, we are able to nd
reasonable evidence that our proposed predictive regression model outperforms the historical average.
Second, we nd a heterogeneous response of returns to institutional and macroeconomic predictors. In 12
countries, institutions predict returns, while in nine countries macroeconomic indicators predict returns.
Third, based on evidence of predictability, we examine whether investors can gain from such
information. We consider a mean-variance investor, who has a portfolio of risky (stock market) assets and
risk-free (short-term interest rate) assets. The portfolio is decided purely on information that results from
the predictive regression model. The investor will invest little in stocks if the predicted excess return
(volatility) is low (high). In our setup, the investor is allowed to rebalance his portfolio once a month. The
coefcients of the model are re-estimated each month when new information becomes available. This
information is sufcient for the investor to revise her/his beliefs about expected returns and volatility. We
devise trading strategies that allow investors to engage in limited borrowing and short-selling and where
transaction costs are also allowed. We call them dynamic trading strategies from which prots are
generated. These prots are compared to those obtained from passive trading strategies. We also estimate
the certainty equivalent return for an investor, following closely the work of Marquering and Verbeek
(2004). From this, we are able to address the question of how much would an uninformed investor, with a
given risk-aversion, be willing to pay to switch from a static to a given dynamic portfolio? We nd that
while investors prefer dynamic trading strategies the magnitude of prots in each of the 15 countries
differs. This reects the different roles played by institutions and macroeconomic factors and, as a result,
the different scopes for prots in those countries.
Finally, we identify nine countries in our sample where short-selling is prohibited. We show that if
limited borrowing and short-selling were allowed, investors in most of those countries would benet from
higher prots from dynamic trading strategies.
The balance of the paper is organised as follows. In the next section, we discuss the theoretical
framework motivating the empirical analysis in this paper. In Section 3, we discuss the data and results. In
the nal section, we provide concluding remarks.
2. Motivating theoretical framework
2.1. Empirical framework
Following Ferson and Harvey (1998), we begin with the following unconditional CAPM:
n

o



M
M
r t1 Et rt1 t r t1 Et r t1
t1



Et  t1 0;

M
Et t1 r t1 0:

80

P.K. Narayan et al. / Emerging Markets Review 19 (2014) 7795

Here, Et(.) denotes the conditional expectation given a common public information at time t. The return
for country i is represented by rt + 1 and the market portfolio return is represented by rM
t + 1. The
corresponding conditional expected returns model has the following form:





M
Et r t1 t t Et r t1 :

And, following Cochrane (1996), Lettau and Ludvigson (2001), and, more recently, Kang et al. (2011),
we represent the alphas and betas in the following manner:
t 0 1 t

t 0 1 t :

Here, t is a vector of conditioning variables. Given our aim of examining separately the effects of
institutions and macroeconomic factors on stock returns, in this paper we have two sets of distinct
conditioning variables. To separate these effects, we can further decompose the vector of conditioning
variables into a vector of institutional type variables and a vector of macroeconomic variables. We can,
therefore, re-write the conditional alphas and betas as follows:
m

t 0 1 Nt 2 Nt
t 0 1 Nt 2 Nt :

I
Here, Nm
t and Nt are vectors of macroeconomic indicators and institutional indicators, respectively. As
we explain in the data section, we have on hand as many as 10 measures of macroeconomic performance
and four measures of institutional depth. Following Stock and Watson (2002) and Ludvigson and Ng
(2007), we use a diffusion index approach and condense the macroeconomic and institutional risk
components into two summary measures, one for macroeconomic performance and one for institutional
performance, using principal component analysis for each country in our sample. Principal component
analysis is ideal in our case because it is able to sufciently deal with problems of multi-collinearity and
over-parameterization. We, therefore, have a principal component based measure of macroeconomic
indicators, which we denote as PCM, and institutional indicators, which we denote as PCI. It follows that our
econometric model has the following form:

r t1 0 1 PC M t 2 PC I t 0 r t 1 PC M t r t 2 PC I t r t t1 :

This multivariate predictive regression model is consistent with the models of return predictability
considered by Rapach and Wohar (2006b) and Marquering and Verbeek (2004) within a time series
framework.6 The rest of the variables in Eq. (7) are as follows: (a) the excess country return is denoted by rt;
(b) world excess market return is denoted by rW
t ; and coefcients 1 and 2 are associated with the interactive
effect of principal components through the world market return variable. Following Marquering and Verbeek
(2004), we use a GARCH-based predictive regression model, in which Eq. (7) is, typically, the mean equation
and the variance equation has the following form:
2

ht 1 t1 2 ht1

where ht is the conditional variance of returns, and 2t 1 and h2t 1 represent short-term and long-term
news, respectively.
6
Using multivariate predictive regression models, Rapach and Wohar (2006b) examine predictability of returns based on
macroeconomic predictors, while Marquering and Verbeek (2004) use nancial ratio predictors.

P.K. Narayan et al. / Emerging Markets Review 19 (2014) 7795

81

2.2. The relevance of institutional and macro factors for returns

2.2.1. The role of institutions
There is a large body of theoretical literature in corporate nance which shows that the choice of
securities or contracts between the rm and its investors depends, in large part, on the information
available to investors. The ability of investors to protect their investment depends on both nancial and
legal institutions.
Allen (2001) argued in favour of nancial institutions, drawing explicitly the link between nancial
institutions and asset pricing. He claimed that, In standard asset pricing theory, investors are assumed to
invest directly in nancial markets. The role of nancial institutions is ignored (p. 1165). His idea was rooted
in the belief that nancial institutions create an agency problem and have a crucial role in providing liquidity.7
In Qian and Strahan (2007), the focus was on understanding how laws and institutions shaped
nancial contracts with respect to bank loans. They nd that strong creditor rights improve loan
availability because in the presence of better legal protection lenders are more willing to provide credit on
favourable terms. This relationship is well entrenched within the theories of debt based on the transfer of
control rights upon default (see Aghion and Bolton, 1992). The implication is that if creditors possess
greater ability to force repayment or take control of the rm in the event of a default, they will offer credit
on more favourable terms ex ante.
In a more extensive study involving thirty developed and developing countries, Demirguc-Kunt and
Maksimovic (1998) show that a well-developed legal system is imperative for rm growth. More
specically, they report that Firms in countries that have active stock markets and high ratings for
compliance with legal norms are able to obtain external funds and grow faster (p. 2134). La Porta et al.
(1997), using a sample of 49 countries, show that those with poorer investor protection, reected in weak
legal rules and poor quality of law enforcement, tend to be characterised by smaller and narrower capital
markets. For a similar analysis, see La Porta et al. (1998).8
In a relatively recent study, Papaionannou (2009) uses a panel data model consisting of 50 countries to
examine the determinants of international nancial ows from banks. One of his key determinants is
institutional quality. He nds that poorly performing institutions, such as weak protection of property rights,
legal inefciencies, and a high risk of expropriation, are key factors inhibiting growth of foreign bank capital.
While our motivation is the same as in the literature alluded to above, our approach and indeed our aim
with regard to the treatment of institutions and returns are different. We use time series data to capture the
effect of institutions on excess returns, while the extant literature examines the nancial performance of a
rm in relation to corporate governance, which is captured through the characteristics of the rm's board or
the corporate legislation and guidelines. A rm's board characteristics may include amongst other things the
board size, board diligence, and board independence (see, inter alia, Christensen et al., 2010; Lehn et al.,
2009). For studies that focus on corporate legislation and guidelines, see Bragaa-Alves and Shastri (2011),
Bebchuk et al. (2009), and Gompers et al. (2003). By comparison, we have four specic institutional factors.
Including all of them in a regression model will lead to multi-collinearity. Therefore, based on principal
component analysis, we create a single series, which effectively captures the institutional performance.
2.2.2. The role of macroeconomic indicators
There is a large body of literature that considers the relationship between returns and macroeconomic
factors. Compared with the returnsinstitutions literature alluded to above, much of the work on the returns
and macroeconomic factors is based on time series data. Essentially, two directions have been popular and, thus,
commonly explored on the returnsmacroeconomic nexus. First, aggregate consumption growth directly enters
asset pricing models on which there is already a rich literature, owing, in large part, to the consumption-based
CAPM. The consumption effect was rst described by Lucas (1978) and Breeden (1979), who argued that the
risk premium on an asset is determined by the ability to insure it against uctuations in consumption. Later, the
consumption effect, through the ratio of stock market wealth and through risks inherent in cash ows in
inuencing asset prices, was documented by Bansal et al. (2005) and Da (2009), respectively. In much earlier
work, Campbell and Cochrane (1999) proposed a habit formation model, where both surplus consumption and
7
8

An informative discussion on agency problem can be found in Shleifer and Vishny (1997).
An excellent discussion on investor protection and rm performance can be found in La Porta et al. (2000).

82

P.K. Narayan et al. / Emerging Markets Review 19 (2014) 7795

per capita real consumption growth rate are factors that inuence the stochastic discount factor. Moreover, the
effect of labour income on returns has been considered by Santos and Veronesi (2006) and Danthine and
Donaldson (2002); the effects of money on asset returns have been shown by Barberis et al. (2001) and
Giovannini (1989); the effects of ination on stock returns through demand and supply disturbances have been
considered by Hess and Lee (1999); and the effects of government budgets on asset returns have been studied
by Boothe and Reid (1989).
A related strand of the literature considers the effects of macroeconomic news on the performance of
nancial markets. Jones et al. (1998) examine the reaction of daily Treasury bond prices to US
macroeconomic news announcements. Moreover, earlier studies (see, inter alia, Gennotte and Marsh,
1993) have considered the effects of variations in economic uncertainty on capital assets.
What do we learn from this literature and how does our approach differ in terms of modelling the
effects of macroeconomic factors? We learn that macroeconomic factors, such as ination, interest rates,
money, government budget balances, consumption, and income have been used to capture the effects of
macroeconomic risks. Clearly, the list of macroeconomic factors is long and we utilize up to 10
macroeconomic factors. Within a time series framework it is impossible, at least from an econometric
modelling point of view, to include all factors individually. Doing so would expose the model to the
problem of multi-collinearity, making the estimates biased. Our approach is to take a principal component
of the macroeconomic factors as a proxy for the measure of macroeconomic risk.
3. Data
The data used in this study include equity market returns for 18 emerging markets, the Morgan Stanley
Capital Index (MSCI) local currency world market index, and institutional and macroeconomic variables.9
The data on returns are sourced from the Bloomberg Financial Database, and the institutional and
macroeconomic risk components are sourced from the International Country Risk Guide (ICRG) database.
Column 2 of Table 1 identies the sample period covered with respect to each of the 18 emerging markets.
The sample period is dictated by data availability in Bloomberg and ICRG databases. It is important to note
here that the ICRG database has as many as 70 developing countries. However, we excluded those
countries in which there was little variation in institutional factors. This is needed because if institutional
risk components are not changing over long periods of time it is pointless to model them. Our ltering
approach was to exclude all countries which had continuously three years of institutional data that did not
change. In this way, we end up with a sample of 18 countries. It should also be noted that other studies
have also used institutional risk components from the ICRG database and, like us, they note that at least for
some of those countries in the ICRG database there is high within country variation in many of the
institutional measures; see, for instance, Papaionannou (2009) and Glaeser et al. (2004).
We employ four institutional risk components; namely, corruption, ethnic tension, external conict, and
law and order. The corruption component of the institutional risk factor captures the level of risk arising from
corrupt behaviour mainly associated with the length of time the government has been in power continuously.
The next component under this group is ethnic tension, which measures the degree of tolerance and
compromise between various ethnic groups in a country. The third component, internal conict, captures any
institutional risk arising in the country due to civil war, civil disorder, and terrorism. The last component, law
and order, measures the strength and impartiality of the legal system, as well as the popular observance of
law and order in a given country. From these variables, as we explain later, we capture the effects of
institutions through principal component analysis for each of the 18 countries in our sample.
In addition, we employ 10 macroeconomic risk components; namely, budget balance as a percentage of
GDP, current account as a percentage of exports of goods and services (XGS), current account as a
percentage of GDP, debt service as a percentage of XGS, exchange rate stability, foreign debt, GDP growth,
ination, net international liquidity, and per capita GDP. From these variables, using principal component
analysis, we extract a single measure of macroeconomic risk for each of the 18 countries in our sample. We
briey outline how individual risk components are built using different scales to capture a country's
macroeconomic conditions: Budget balance as a percentage of GDP increases the rating when a given
9
Eighteen developing countries included in our sample are: Argentina, Bangladesh, Brazil, Chile, China, Egypt, Kenya, Lebanon,
Malaysia, Mexico, Oman, Pakistan, Peru, Russia, South Africa, Taiwan, Tunisia, and Venezuela.

P.K. Narayan et al. / Emerging Markets Review 19 (2014) 7795

83

Table 1
Descriptive statistics. In this table we report selected descriptive statistics, namely mean, standard deviation, skewness, and kurtosis,
for excess returns. This is followed by a principal component analysis of institutions (PCI) and macroeconomic factors (PCM). For PCI
we report the percentage contribution of the rst three principal components to the standardized variance, while for PCM we report
the percentage contributions of the rst ve principal components to the standardized variance.
Sample period

Argentina
Bangladesh
Brazil
Chile
China
Egypt
Kenya
Lebanon
Malaysia
Mexico
Oman
Pakistan
Peru
Russia
S. Africa
Taiwan
Tunisia
Venezuela

1988.12011.6
1988.112011.6
1988.42011.6
1989.12011.6
2002.12011.6
1993.32011.6
1991.22011.6
1996.12011.6
1987.62011.6
1994.12011.6
1992.12011.6
1990.52011.6
1990.12011.6
1997.92011.6
1995.62011.6
1987.122011.6
1999.42011.6
1993.32011.6

PCI [% of variance]

PCM [% of variance]

Mean

Std. dev.

Skewness

Kurtosis

PC1

PC2

PC3

PC1

PC2

PC3

PC4

PC5

3.808
0.589
6.957
1.347
0.647
1.011
0.307
0.091
0.308
0.973
0.420
0.930
3.975
1.469
0.679
0.134
0.740
1.814

20.311
9.226
19.651
6.267
9.430
8.490
7.078
7.232
7.253
7.670
5.843
9.709
15.368
13.420
6.296
10.126
3.988
10.406

2.887
0.864
0.618
0.457
0.562
0.303
0.964
0.904
0.147
1.027
0.507
0.580
2.587
0.856
1.031
0.354
0.030
0.045

18.127
10.736
6.092
7.146
4.123
5.040
9.093
8.350
5.766
5.872
6.772
6.151
16.854
6.333
7.237
5.269
4.980
7.103

48.7
70.7
66.2
55.3
61.5
47.4
55.4
44.2
54.3
46.2
63.2
53.1
69.8
60.6
75.2
49.9
44.6
73.7

81.6
84.5
85.4
82.5
86.7
75.5
80.8
73.1
89.3
83.2
98.9
75.8
92.4
85.6
89.4
79.9
81.1
96.7

100
94.2
96.8
95.4
96.7
94.3
94.5
88.6
95.9
95.6
100
91.3
96.6
98.1
95.2
96.3
96.4
100

43.3
66.8
40.1
44.3
39
40.5
58.4
39.1
42.5
41
50.4
51.9
64
49.7
29.9
44.4
43.8
43.6

66.7
79.3
61.5
60
62.1
63.5
76.9
64
62.9
59.5
69.7
73.3
78
75.7
52.9
65.3
65.3
61.6

80.1
87.8
76.1
71.6
73.3
77.4
86.2
78.7
75.3
73.2
80.4
82.9
87.6
85.1
64.9
78.5
76.1
72.3

86.4
92.1
84
80.2
82.6
86
91.3
87.1
83.1
81.8
86.4
89.3
92.9
89.7
74.3
88.3
83.4
81.5

91.5
95.4
90.1
86.9
91.4
91.2
94.2
92.9
88.4
88.9
91.8
92.9
95.6
92.9
81.7
93.3
88.5
86.9

country has positive percentage while points are deducted as the country moves to a decit budgetary
condition. A country having a budget balance of 4% or over will be allocated 10 points while a negative 30%
of budget balance will earn zero points indicating higher level of risk. Similarly, current account as a
percentage of XGS earns 15 points for a surplus of 10% or more, while a negative 30% (imbalance) earns
zero points, reecting a deteriorating condition with respect to imports and exports. Similar attribution is
made with respect to current account as a percentage of GDP. Ination rate of less than 2% in a country
earns a maximum of 10 points while ination rate of over 130% earns 0 points, indicating a higher level
of risk. A GDP growth of 4% or above in real terms earns 10 points while a negative growth rate (less than
30%) earns zero points.
Debt related measures, such as foreign debt as a percentage of GDP, earn 10 points if the percentage lies
between 0 and 4.9%. On the other hand, indebtedness rising to 200% of GDP earns zero points. Similar scaling
is used for a county's capacity to service its debt as a percentage of exports of goods and services: a percentage
of 4.9% and below earns 10 points while a higher relative percentage (85% or above), reects constraints in
debt service capacity and, therefore, earns zero points. The exchange rate stability component is a risk
measure that captures the appreciation/depreciation of a country's currency against the US dollar over a
calendar year. If the appreciation changes or the depreciation change is kept at the lower end (up to 9.9% for
appreciation change and 0.1% to 4.9 for depreciation change) then full 10 points are allocated, favouring
the country's currency situation with respect to this risk component. Net international liquidity is the
estimated ofcial reserves for a given year measured in months. A cover of, at least 15 months earns all ve
points while a cover of two weeks or less earns 0 points, signalling deteriorating liquidity conditions.
4. Results
4.1. Preliminary analysis of the data
We begin the results with an analysis of our data set. Table 1 reports the summary statistics on excess
returns. Panel A contains the basic statistics while panels B and C contain the percentage contribution of
each of the principal components to the standardized variance of the institutional factors and macroeconomic
factors, respectively. We begin by examining statistics on excess returns. The rst thing we notice is that there
is wide variation in mean excess returns. It is lowest for Lebanon (0.09%) and highest for Brazil (6.96%).

84

P.K. Narayan et al. / Emerging Markets Review 19 (2014) 7795

Mean excess returns are close to 4% in Peru and Argentina and in excess of 1% in the case of Chile, Egypt,
Russia, and Venezuela. Countries with the highest mean returns also experience higher volatility as measured
by the standard deviation. Skewness is generally found to be negative while evidence based on the kurtosis
statistic suggests a leptokurtic distribution of excess returns for all 18 countries.
Next, we briey summarise the main features of the principal component analysis. With regard to the
institutional indices, where we have four specic variables, we nd that the rst principal component on
average across all 18 countries explains around 58% of the standardized variance. For about 44% of the
countries in our sample, the rst principal component explains over 60% of the standardized variance. By
comparison, the second and third principal components explain only around 27% and 11% of the
standardized variance, respectively. The cumulative percentage contributions of each of the three
principal components are plotted in Fig. 1. We, thus, conclude that the rst principal component explains
the variations in the institutional indices better than any other linear combination of variables.
For the macroeconomic variables, we have 10 indicators. We nd that, on average, the rst principal
component explains around 46% of the standardized variance, while the second and third principal
components explain, on average, around 20% and 12% of the standardized variance, respectively. The rest of
the principal components, by comparison, explain less than 13% of the variance. Again, as for institutional
factors, the rst principal component explains the variance better than any other linear combinations.
The predictable regression should ideally have all variables in stationary form to avoid any bias in the
predictive regression model (see, Westerlund and Narayan, 2012; Narayan et al., 2014). While the
stationary form of excess returns and excess world market returns are rather obvious, the same cannot be
claimed for the principal components of institutions and macroeconomic conditions. We conrm the
integrational property of all the variables for each of the 18 countries by conducting an augmented Dickey
and Fuller (ADF, 1979) test. The ADF test examines the null hypothesis that there is a unit root against the
alternative that the variable is stationary. Our approach in running the ADF test is to utilize a maximum of
eight lags to control for any potential serial correlation in the variable. We then use the Schwarz
Information Criteria (SIC) to obtain the optimal lag length. The results are not tabulated here but are
available upon request. We only summarise the key nding here. First, we nd that the excess returns for
each country, the excess world market returns and the two interactive variables appear to be clearly
stationary. The unit root null hypothesis is rejected comfortably at the 1% level for all 18 countries. Second,
except for Oman and Kenya, the principal component of institutional indicators are non-stationary, as the
unit root null hypothesis cannot be rejected at the 5% level. When we consider the principal component
of macroeconomic indicators, we nd that the unit root null hypothesis cannot be rejected at the 5% level
for any of the 18 countries, suggesting that for all countries the macroeconomic indicator is unit root
non-stationary. The implication here is that except for Oman and Kenya when using the institutional

100
90
80
70
60
50
40
30
20
10
0

PC1

PC2

PC3

Fig. 1. Percentage contribution of principal components to standardized variance of institutions. This gure plots the cumulative
contributions of the rst three principal components of institutional factors to the standardized variance. The bulk of the variations
comes from the rst principal component.

P.K. Narayan et al. / Emerging Markets Review 19 (2014) 7795

85

variable as a predictor, we take the rst difference of the principal components in testing for return
predictability. This approach ensures that our test for return predictability does not suffer from the
inherent bias imposed by persistency of predictorsan issue which has occupied much interest in nancial
economics (see, for instance, Westerlund and Narayan, 2014a).

4.2. Evidence on predictability

4.2.1. In-sample evidence
The results based on the predictive regression model (7) are reported in Table 2. The results are
organised row-wise by country. The main features of our results can be summarised as follows. First, we
nd that PCM predicts excess returns for seven countries (Bangladesh, Brazil, China, Argentina, Kenya,
Oman and Venezuela).

Table 2
Results for return predictability. In this table, we report the results on return predictability based on a GARCH model. The mean
W
W
equation takes the following form: rt + 1 = 0 + 1PCMt + 2PCIt + 0rW
t + 1PCMtrt + 2PCItrt + t. The principal components
of macroeconomic indicators and institutional factors are denoted by PCMt and PCIt, respectively. The excess country returns is
denoted by rt.; world excess market return is denoted by rW
t ; and coefcients 1 and 2 are associated with the interactive effect of
principal components through the world market return. Finally, the variance equation has the following form: ht = + 1t
2
2
2
2
1 + 2ht 1, where ht is the conditional variance of returns, and t 1 and ht 1 represent short-term and long-term news,
respectively. Figures in parenthesis are the p-value and the asterisks indicate statistical signicance at 1% (***), 5% (**) and 10% (*).

Argentina
Bangladesh
Brazil
Chile
China
Egypt
Kenya
Lebanon
Malaysia
Mexico
Oman
Pakistan
Peru
Russia
S. Africa
Taiwan
Tunisia
Venezuela

PCM

PCI

rW

PCMrW

PCIrW

3.2663**
(0.012)
0.1672
(0.706)
2.9995***
(0.001)
1.8839***
(0.000)
14.1267***
(0.007)
0.8865
(0.642)
0.0302
(0.943)
3.1995
(0.505)
0.6813**
(0.034)
1.7432**
(0.025)
0.7113***
(0.003)
1.0151*
(0.071)
3.1343***
(0.001)
2.3924*
(0.010)
0.2502
(0.835)
0.6534
(0.240)
1.7402
(0.321)
2.3412***
(0.006)

1.3369*
(0.052)
0.5316***
(0.007)
1.449**
(0.050)
0.088
(0.839)
5.8768***
(0.002)
0.6498
(0.384)
0.5474***
(0.000)
0.8385
(0.482)
0.1039
(0.554)
0.4044
(0.333)
0.2083*
(0.085)
0.2034
(0.322)
0.0002
(1.000)
0.4269
(0.625)
0.3716
(0.388)
0.1477
(0.620)
1.1091
(0.151)
0.8144*
(0.075)

1.5109*

0.569*

0.2315
(0.485)
0.8188
(0.404)
0.9664**
(0.021)
1.0752
(0.336)
2.7331*
(0.015)
0.7196*
(0.069)

0.0701
(0.459)
0.1684
(0.453)
0.2388**
(0.046)
2.7123*
(0.052)
0.5751
(0.244)
0.1028
(0.273)

(0.571)
0.1468
(0.585)
0.3495
(0.420)
0.4375**
(0.021)
0.6803*
(0.095)
1.0423
(0.213)
0.6744
(0.584)
0.3072
(0.647)
0.3438
(0.474)
1.7051**
(0.032)
0.4566
(0.511)

(0.056)
0.1046
(0.245)
0.1086
(0.706)
0.1782***
(0.006)
0.1362
(0.371)
0.2943
(0.228)
0.4638**
(0.032)
0.5903**
(0.031)
0.0453
(0.756)
0.4971
(0.123)
0.4455*
(0.070)

0.0362
(0.825)
0.04
(0.360)
0.2817
(0.122)
0.2169*
(0.100)
1.0497*
(0.058)
0.002
(0.992)
0.0396
(0.283)
0.6149*
(0.067)
0.0394
(0.468)
0.1139
(0.372)
0.0526
(0.351)
0.0426
(0.516)
0.2066
(0.179)
0.0185
(0.936)
0.1431
(0.270)
0.1107
(0.227)
0.2096
(0.216)
0.1585
(0.189)

0.084
(0.712)
0.0675
(0.379)
0.5322**
(0.038)
0.3042**
(0.012)
0.4752**
(0.050)
0.3061
(0.238)
0.0191
(0.856)
0.8333
(0.294)
0.135*
(0.089)
0.0713
(0.573)
0.0252
(0.555)
0.0987
(0.363)
0.3192*
(0.099)
0.1412
(0.615)
0.294*
(0.077)
0.023
(0.849)
0.1737
(0.338)
0.0098
(0.941)

R2
0.1029
0.0072
0.0067
0.0603
0.0289
0.0746
0.0103
0.0851
0.0349
0.0146
0.0133
0.0242
0.0363
0.0538
0.0111
0.0014
0.0866
0.0345

86

P.K. Narayan et al. / Emerging Markets Review 19 (2014) 7795

Second, we nd that when the principal component of macro indicators is modelled interactively
through the world excess market return, it is able to predict the excess returns of Chile, China and
Lebanon. It follows that, either directly or indirectly (through the excess world market return), macro
indicators predict excess returns for as many as nine countries.
Third, we nd that PCI predicts excess returns for seven countries (Argentina, Chile, Egypt, Kenya,
Oman, Pakistan and Tunisia). Fourth, when we use the principal components of institutions interacted
with the world market excess returns, we nd that they predict excess returns for Brazil, Chile, China,
Malaysia, Peru and South Africa. In all, then, either directly or indirectly, institutions predict excess returns
for 12 countries. Lastly, we discover that for 15 countries, there is some evidence that either macroeconomic
or institutional factors predict excess returns.
The key implication is that one can use our proposed predictability model to forecast returns for these
15 countries. The next question is that if investors in these countries do indeed use our model to forecast
returns, how relevant are the forecasts? This, effectively, is an issue of the out-of-sample forecasting
performance of our model. To demonstrate the forecasting performance of our predictive regression
model, we follow Welch and Goyal (2008) and Rapach et al. (2010). We utilize 50% of the sample to
produce in-sample estimates and use those estimates to forecast returns for the rest of the 50% of the
sample for each of the 15 countries.
4.2.2. Out-of-sample evidence
To measure the out-of-sample performance of our forecasting model, we use commonly used metrics,
namely, the out-of-sample R2, which we denote as OOS_R2, the Theil U statistic, and the mean squared
error test proposed by McCracken (2007), which we denote as MSE F. The OOS_R2 follows the proposal
of Welch and Goyal (2008) and Ferreira and Santa-Clara (2011):
2

OOS R 1

MSEmodel
MSEmean

where MSEmodel is the mean square error of the out-of-sample predictions from our proposed model, while
MSEmean is the mean squared error of the historical sample mean. When OOS_R2 N 0, our proposed
predictive regression model predicts returns better than the historical mean, and vice versa. By
comparison, the Theil U statistic is dened as the ratio of the square roots of the mean-squared
forecasting errors of the predictive regression model relative to the historical average. If the Theil U b 1,
Table 3
Out-of-sample forecasting performance. In this table, we report the Theil U statistic, the OOS_R2, and the MSE-F statistic (denoted
with stars) as a metric for comparing the out-of-sample performance of our multivariate predictive regression model with that of
historical average. *** denotes the statistical signicance at the 1% level at which null hypothesis that the MSE from the predictive
regression model is equal to the historical average. The test statistics are considered only for the 15 countries for which some
evidence of return predictability was found.

Argentina
Bangladesh
Brazil
Chile
China
Egypt
Kenya
Lebanon
Malaysia
Oman
Pakistan
Peru
S. Africa
Tunisia
Venezuela

Theil U

OOS_R2

0.7095***
0.9403***
0.8709***
0.8957***
0.6597***
0.7086***
0.7847***
0.9281***
0.8331***
0.7383***
0.8639***
0.7446***
0.7363***
0.8700***
0.8600

0.0308
0.0334
0.0674
0.0398
0.1094
0.1138
0.0382
0.0222
0.0255
0.0648
0.0069
0.0433
0.3408
0.0405
0.0120

P.K. Narayan et al. / Emerging Markets Review 19 (2014) 7795

87

our proposed predictive regression model outperforms the historical average. Finally, the McCracken
(2007) test examines the equality of the MSE of the unconditional (historical mean) and the conditional
(predictive model) forecasts. Following the proposal in Welch and Goyal (2008), we generate bootstrap
p-values to test the null of equal MSE.
The results are reported in Table 3. The ratio of the Theil U statistic is reported in column 2. We nd
that all ratios are less than 1. This suggests that the proposed predictive regression model outperforms the
historical average for all countries. The McCracken MSE-F statistic corroborates this evidence; the F
statistics are all signicant at the 1% level for all countries (except Venezuela), favouring the predictive
regression model. Finally, when we consider the OOS_R2, the results are mixed. Only for four countries is
the OOS_R2 greater than zero, suggesting that only for those four countries does the predictive regression
model outperforms the historical average. On the whole, then, it is fair to claim the out-of-sample
superiority of our predictive regression model over the historical average. A nal note here is that, where
there is no predictability, such as in the case of Mexico, Taiwan, and Russia, the need for forecasts does not
arise. This also means that in evaluating trading strategies we only concentrate on those 15 countries for
which predictability has been found.
4.3. Prots from passive and dynamic trading strategies
The results from passive trading strategies are reported in Table 4. Essentially, we consider three types
of investments: 100% in the market; 50% in the market and 50% in a risk-free asset; and 100% in a risk-free
asset. It is true that these investment types are arbitrary and that one could come up with a type that best
addresses the research question. Our approach here is motivated by Marquering and Verbeek (2004), who
consider the same three types of investments. For each strategy, we estimate returns and their statistical
signicance, and we compute the standard deviation of returns. We nd that by investing 100% in a risky
asset (market), investors in ve out of 15 countries end up making statistically signicant prots. A 50:50
strategy reduces returns and risk, as expected. The most protable countries turn out to be Malaysia and
Oman, where a 50:50 strategy leads to prots of over 1% per month.
In Table 5, we report prots from dynamic trading strategies. Specically, we consider a dynamic
trading strategy (DTS1) where the portfolio weight is restricted to be between 0 and 1 and another where
the portfolio weight is restricted to be between 0 and 1.5 (DTS2)the rationale for this strategy and the
Table 4
Prots from passive trading strategies. In this table, we report prots from passive trading strategies for each of the 15 countries for
which we discovered evidence for return predictability. Prots are generated for three passive trading strategies: (a) one in which
investors investor 100% in the market (column 2); (b) one in which investors invest 50% in the market and 50% in a risk-free treasury
bill rate (column 3); and (c) one in which investors invest 0% in the market and 100% in a risk-free treasury bill rate (column 4). For
each strategy, prots together with its standard deviation (SD) are reported. *, **, and *** denote statistical signicance at the 10%,
5%, and 1% levels, respectively.
100%

Argentina
Bangladesh
Brazil
Chile
China
Egypt
Kenya
Lebanon
Malaysia
Oman
Pakistan
Peru
S. Africa
Tunisia
Venezuela

50%

0%

Mean

SD

Mean

SD

Mean

SD

5.798***
2.583***
13.446***
1.623***
0.941
3.177***
2.506
0.647**
3.615***
3.309***
0.499**
3.517***
6.232***
0.486***
1.474***

5.826
1.117
4.652
2.036
8.938
5.143
0.683
2.050
0.632
3.007
1.914
1.163
2.651
0.759
2.476

2.797***
1.194***
6.622***
0.715***
0.408**
1.508***
1.170
0.238*
1.910***
1.735***
0.336***
1.669***
3.032***
0.133***
0.818***

2.898
0.589
2.329
1.006
4.502
2.564
0.356
1.049
0.338
1.518
0.968
0.612
1.292
0.397
1.187

0.205***
0.195***
0.203***
0.193***
0.126***
0.161***
0.166***
0.172***
0.206***
0.160***
0.173***
0.178***
0.168***
0.180***
0.162***

0.167
0.163
0.167
0.168
0.159
0.146
0.141
0.157
0.168
0.142
0.144
0.149
0.155
0.173
0.149

88

P.K. Narayan et al. / Emerging Markets Review 19 (2014) 7795

construction of portfolio weights are explained in the next section. Panel A reports results for the case of
no transaction cost, while results in Panel B are based on a transaction cost of 1%. The main results can be
summarised as follows. First, investors in all countries earn statistically signicant prots regardless of the
restrictions on the portfolio weight.
Second, the most protable countries turn out to be Oman, Malaysia, China and Venezuela, where
returns are between 1.4 and 3.7% per month. In the other countries studied, investors earn less than 1% per
month. The weakest returns are experienced by investors in Bangladesh, Brazil, South Africa, Peru, and
Kenya, where returns are in the range of 0.16 to 0.18% per month. Third, with DTS2, the opportunities for
prots increase, and the largest gains are experienced by investors in Oman, Malaysia and China. On the
other hand, no gains in returns are made by investors in Bangladesh, Brazil, Kenya, Peru, and South Africa.
A nal observation we make here relates to the Sharpe ratio. The Sharpe ratio, dened as the mean
excess return on the portfolio relative to the standard deviation of the portfolio return, is reported in the
last column of Table 5. We notice that, generally, Sharpe ratios from DTS1 are higher than from DTS2; the
exception is Chile. This reects the attractiveness of a dynamic trading strategy without borrowing (DTS2).
Similarly, when we compare Sharpe ratios of passive trading strategies (not reported here) with those
from dynamic trading strategies, we nd that dynamic trading strategies are relatively more attractive for
investors.
Table 5
Prots from dynamic trading strategies. In this table, we report prots from two dynamic trading strategies: one in which portfolio
weights are restricted to be between 0 and 1 (no short selling), and one in which limited short-selling and borrowing are allowed.
The results are divided into two panels. Results contained in Panel A are based on zero transaction cost, while results in Panel B take
into account a 1% transaction cost. The nal columns also contain the results on Sharpe ratio, dened as the mean excess return on a
portfolio relative to the standard deviation of the portfolio return.
Optimal (01)

Optimal (01.5)

Sharpe ratio

SD

Mean

SD

(01)

(01.5)

1.405
0.147
0.149
0.416
4.100
1.507
0.145
0.888
0.638
2.292
0.816
0.141
0.161
0.638
1.764

0.695***
0.174***
0.178***
0.306***
2.630***
0.838***
0.161***
0.510***
4.998***
5.235***
0.797***
0.161***
0.179***
0.658***
1.900***

1.720
0.147
0.149
0.573
5.769
2.108
0.145
1.253
1.181
3.444
1.008
0.141
0.161
0.932
2.684

0.310
0.145
0.167
0.193
0.487
0.344
0.032
0.296
5.217
1.53
0.685
0.124
0.071
0.557
0.705

0.285
0.145
0.167
0.198
0.434
0.321
0.032
0.270
4.058
1.479
0.619
0.124
0.071
0.513
0.648

Panel B: With 1% transaction cost (optimal )

Argentina
1.278**
4.492
Bangladesh
0.171***
0.141
Brazil
0.178***
0.149
Chile
0.226***
0.304
China
2.367***
4.506
Egypt
0.179***
0.155
Kenya
0.161***
0.145
Lebanon
0.321***
0.675
Malaysia
3.204***
0.566
Oman
2.967***
2.139
Pakistan
0.857***
0.912
Peru
0.161***
0.141
S. Africa
0.179***
0.161
Tunisia
0.253***
0.337
Venezuela
1.125***
1.652

1.549**
0.171***
0.178***
0.237***
3.087***
0.179***
0.161***
0.367**
4.480***
4.221***
0.941***
0.161***
0.179***
0.265***
1.512***

5.717
0.141
0.149
0.370
6.562
0.155
0.145
0.981
1.109
3.189
1.126
0.141
0.161
0.402
2.498

0.239
0.173
0.167
0.108
0.497
0.114
0.032
0.222
5.299
1.312
0.751
0.124
0.071
0.218
0.583

0.235
0.173
0.167
0.118
0.451
0.114
0.032
0.199
3.854
1.274
0.682
0.124
0.071
0.211
0.540

Mean
Panel A: No transaction cost (optimal )
Argentina
0.641***
Bangladesh
0.174***
Brazil
0.178***
Chile
0.273***
China
2.123***
Egypt
0.680***
Kenya
0.161***
Lebanon
0.435***
Malaysia
3.536***
Oman
3.667***
Pakistan
0.732***
Peru
0.161***
S. Africa
0.179***
Tunisia
0.535***
Venezuela
1.405***

P.K. Narayan et al. / Emerging Markets Review 19 (2014) 7795

89

4.4. Evidence from a utility-based measure

In this section, following Rapach et al. (2010), Campbell and Thompson (2008), and Marquering and
Verbeek (2004), and Narayan et al. (2013)we consider a relatively robust approach in testing the
economic signicance of our results. More specically, we undertake a utility-based measure of economic
signicance, where realized utility gains are computed for a mean-variance investor on a real-time basis. A
mean-variance investor has the following utility function:

 1


Et r t1 Vart r t1 :
2

10

Such that, given a portfolio of t + 1 for the risky asset, the utility simply becomes:
n
o 1
n
o
2
r f ;t1 t1 Et r m;t1 t1 Vart r m;t1
2

11

where rt + 1 is the excess return on the portfolio or security, rf,t + 1 is the risk-free rate of return, rm,t + 1 is
the market excess return, Vart is the rolling variance of the risky asset, is the risk aversion factor, and
t + 1 is the investor's portfolio weight in period t + 1, computed as follows:


t1



Et r t1
n
o:
Vart r m;t1

12

We constrain the portfolio weight on stocks to lie between 0% and 150% each month. We also generate
results for portfolio weights of 50% to 150% but do not report the results here, as Campbell and
Thompson argue that the most realistic weight is 01.5. In particular, they state that such a weight
restriction prevents the investor from shorting stocks or taking more than 50% leverage (p. 1525). In
other words, if forecasts for excess returns are good, investors may borrow up to 50% of their investment
value and invest in risky assets.10
The average utility level, ex post, becomes:
^ 1
U
T

TX
1
t0

1 2
r t1 t1 Vart :
2

13

The average utility is computed for the dynamic portfolio as well as for the passive portfolio. This
allows us to compare the utilities from both types of portfolios. We are, as a result, able to obtain the
maximum fee an investor will be willing to pay for holding the dynamic portfolio over the passive one.
The results on average utilities for the three passive and the two dynamic trading strategies for three
different levels of risk aversion ( = 3, 6, 12) are reported in Table 6. The results are organised as follows. In
Panel A, we report utilities obtained from the passive strategies, while in Panel B we report utilities obtained
from the two dynamic strategies. A number of important ndings emerge from this analysis. First, if one only
considers utilities from passive strategies, investors in 10 countries end up with negative utilities. This implies
that investors would prefer not to invest in those markets. However, if investors use a dynamic trading
strategy, all 10 countries which had negative utilities end up with positive utilities. Therefore, the difference in
utilities obtained from dynamic trading strategies and passive trading strategies is positive, thus always
favouring dynamic strategies. The investor, therefore, would be willing to pay more per month to hold a
dynamic trading strategy rather than a passive trading strategy. As we saw earlier, doing so leads to greater
prots for investors. Under a range of different scenarios, we showed that investors make relatively more
prots from dynamic trading strategies compared to passive trading strategies.
10
Detailed results on prots and utilities for a portfolio with restricted weights of between 0.5 and 1.5 are available upon request
from the corresponding author.

90

P.K. Narayan et al. / Emerging Markets Review 19 (2014) 7795

Second, we notice that when investors were allowed to borrow 50% of the value of their investment
value (DTS2), average utilities are slightly higher than those obtained from passive trading strategies.
Finally, we observe that results in favour of dynamic trading strategies are robust to different levels of risk.
In evaluating trading strategies, it is natural to allow for transaction costs, which is something we have
ignored so far. To see whether average utilities still favour dynamic strategies over passive strategies, we now
re-estimate all economic gains for the 15 countries and report the results in Table 7. Following the vast
literature that uses transaction costs in estimating prots, we consider transaction costs of 0.5% (medium)
and 1% (high). For simplicity and to conserve space, here we only compare the dynamic portfolios with only
one passive strategyone that invests 50% in a risky asset and 50% in a risk-free asset. The results from
alternative passive strategies are broadly similar, and detailed results are available from the corresponding
author upon request. The choice of the 50:50 strategy is ideal here as it generally gives higher average utilities
compared to a strategy that invests 100% in the market, as we saw from the results reported in Table 6.
The following features of the economic gains in the presence of transaction costs stand out and deserve
a mention. First, we notice that for seven of the 15 countries the economic gains from the dynamic strategy

Table 6
Average realized utilities for passive strategies without transaction cost. In this table, we report the average realized utilities
associated with passive (Panel A) and dynamic trading strategies (Panel B). We use three different risk aversion parameters and
consider two passive and two dynamic trading strategies. The results are based on a case where there is zero transaction cost.
Panel A: Passive trading strategies
=3

Argentina
Bangladesh
Brazil
Chile
China
Egypt
Kenya
Lebanon
Malaysia
Oman
Pakistan
Peru
S. Africa
Tunisia
Venezuela

=6

= 12

100% market

50% market

100% market

50% market

100% market

50% market

5.799
2.71
14.092
1.808
3.567
3.943
2.554
0.646
3.561
3.305
0.624
3.739
6.693
0.51
1.198

2.774
1.265
6.925
0.804
1.202
1.729
1.194
0.207
1.874
1.789
0.413
1.786
3.233
0.335
0.699

5.878
2.722
14.166
1.846
4.829
4.307
2.559
0.704
3.556
3.179
0.578
3.746
6.747
0.502
1.162

2.779
1.266
6.925
0.806
1.281
1.752
1.195
0.210
1.874
1.781
0.410
1.786
3.236
0.335
0.697

6.037
2.747
14.312
1.923
7.355
5.035
2.570
0.820
3.547
2.927
0.486
3.761
6.854
0.487
1.091

2.547
1.077
6.638
0.622
1.047
1.514
1.015
0.018
2.030
1.932
0.570
1.598
3.002
0.484
0.861

Panel B: Dynamic trading strategies

=3

Argentina
Bangladesh
Brazil
Chile
China
Egypt
Kenya
Lebanon
Malaysia
Oman
Pakistan
Peru
S. Africa
Tunisia
Venezuela

=6

= 12

Optimal
(01)

Optimal
(01.5)

Optimal
(01)

Optimal
(01.5)

Optimal
(01)

Optimal
(01.5)

0.632
0.174
0.178
0.269
1.872
0.618
0.161
0.422
3.532
3.584
0.724
0.161
0.179
0.534
1.381

0.685
0.174
0.178
0.299
2.279
0.734
0.161
0.485
4.990
5.091
0.787
0.161
0.179
0.655
1.850

0.624
0.173
0.178
0.266
1.621
0.555
0.161
0.409
3.528
3.501
0.717
0.161
0.179
0.532
1.357

0.674
0.173
0.178
0.292
1.929
0.629
0.161
0.459
4.983
4.929
0.777
0.161
0.179
0.652
1.801

0.607
0.173
0.178
0.258
1.118
0.430
0.161
0.383
3.520
3.334
0.702
0.161
0.179
0.528
1.310

0.653
0.173
0.178
0.278
1.228
0.421
0.161
0.408
4.968
4.605
0.758
0.161
0.179
0.645
1.701

P.K. Narayan et al. / Emerging Markets Review 19 (2014) 7795

91

(DTS2) are greater than the passive trading strategy, implying investor preference for holding dynamic
strategies rather than static strategies. To demonstrate this point, let us consider one specic case. For
Malaysia, for instance, we notice that with a 1% transaction cost, investors are willing to pay a fee for all
trading strategies, but investors prefer paying higher fees to maintain a dynamic trading strategy rather
than a passive one. Interestingly, at a higher transaction cost (1%)a cost that most studies considerthe
passive strategy has a negative utility for 10 countries but dynamic strategy produces positive utilities.
This, again, reects the preference of investors for dynamic trading strategies.
For almost all countries, when we allow for a limited amount of borrowing in the presence of
transaction cost under a dynamic trading strategy, investor utility is maximised. The exception is Tunisia,
for which investor utility is higher under a passive trading strategy.
4.5. How benecial is short-selling?
How benecial is short-selling? To address this question, in Table 7 (last three columns) we provide
country-specic information on short-selling. In particular, we identify; (a) whether short-selling is
practised, (b) start date of short-selling, and (c) imposition of any bans on short-selling. We believe this
analysis is important for the following reason. In Table 8, we report results from two dynamic trading
strategies that allow for limited borrowing and short-selling. We consider two cases. In the rst case, we
allow for borrowing and short-selling of 5% and in the second case we allow for borrowing and
short-selling of 10%. From both strategies, we estimate prots and utility for each country in the presence
of transaction cost. What is clear from our results is that short-selling leads to relatively higher prots
compared to a strategy without short-selling. However, we are also aware that not all countries in our
sample allow for short-selling. From Table 7, we identify nine countries (Bangladesh, China, Egypt, Kenya,
Lebanon, Oman, Peru, Tunisia and Venezuela) where short-selling activities are prohibited. This means we
can now answer the question: How much more prot and utility gain could investors make in these
markets if short-selling (of either 5% or 10%) was indeed allowed?
We summarise the prots and utility from dynamic trading strategies with and without short-selling in
Table 9. Column 2 contains prots and utility from a strategy of no short-selling, while the corresponding
results from dynamic strategies with 5% and 10% short-selling are reported in columns 3 and 4,
Table 7
Average realized utilities with transaction cost. In this table, we report the average realized utilities associated with a 50:50 passive
trading strategy (column 2) and the two dynamic trading strategies (column 3). The utilities are based on a risk aversion parameter
of six. The results are based on a case where there are transaction costs of 0.5% and 1%. It should be noted that naked short-selling has
been prohibited in Brazil, Chile, Malaysia, and South Africa.

Argentina
Bangladesh
Brazil
Chile
China
Egypt
Kenya
Lebanon
Malaysia
Oman
Pakistan
Peru
S. Africa
Tunisia
Venezuela

Passive trading
strategies

Dynamic trading strategies

50% market

Optimal (01)

Is short-selling allowed?

Optimal
(01.5)

0.50%

1%

0.50%

1%

0.50%

1%

1.275
0.212
1.186
0.624
21.195
6.113
0.089
0.967
0.083
2.136
0.782
0.144
0.876
0.133
0.57

2.77
1.265
6.921
0.802
1.139
1.711
1.194
0.204
1.874
1.795
0.415
1.785
3.23
0.335
0.701

2.938
0.175
4.882
1.028
6.216
0.31
0.093
8.907
3.609
3.511
1.151
0.079
3.089
0.363
0.684

3.351
0.171
5.179
1.393
8.786
0.418
0.071
10.87
3.095
2.714
0.825
0.07
3.547
0.234
0.628

1.231
0.175
0.178
0.278
2.018
0.194
0.161
0.393
5.312
4.57
1.309
0.161
0.179
0.443
1.689

1.175
0.171
0.178
0.227
1.878
0.178
0.161
0.34
4.471
3.921
0.919
0.161
0.179
0.263
1.398

Yes
No
Yes
Yes
No
No
No
No
Yes
No
Yes
No
Yes
No
No

Start date

Imposition of
ban on short-selling

1999
Never allowed
Since inception
1999
Never allowed
Never allowed
Never allowed
Never allowed
Pre 1997
Never allowed
Since inception
Never allowed
Since inception
Never allowed
Never allowed

Pre-1999
Always
Never imposed
Pre-1999
Always
Always
Always
Always
Sept 97Dec 06
Always
Never imposed
Always
Never imposed
Always
Always

92

P.K. Narayan et al. / Emerging Markets Review 19 (2014) 7795

respectively. The main ndings are as follows. In eight of the nine countries (except for China), the
introduction of limited short-selling of 5% relative to the dynamic trading strategy without short-selling
will lead to a rise in prots by between 8.4% in the case of Venezuela to 118% in the case of Egypt.
Bangladesh records a rise in prots of around 111%, followed by Kenya (104%), Oman (72.6%), Lebanon
(31.4%), Tunisia (15.8%), and Peru (14.7%). A similar trend in the rise in prots is noticed when a higher
level (10%) of borrowing and short-selling is allowed. When we consider utilities, we notice two things.
First, without short-selling, China, Egypt, Lebanon and Oman had negative utilities, reecting lack of
investor preference for dynamic trading strategies. However, with short-selling all utilities become
positive. Second, utilities increase with an increase in the magnitude of short-selling.
On the whole, our analysis suggests that allowing for limited short-selling and borrowing is likely to be
benecial to investors. We have considered our results based on limited short-selling scenarios which can
be extended to consider any specic amount of short-selling that countries would like to implement.
5. Concluding remarks
In this paper we contribute to the literature on the predictability of stock returns. Our analysis
produces four new ndings. First, we consider evidence of predictability for as many as 18 emerging
markets using time series data. This approach ensures that we are in a position to discern country
heterogeneity with respect to institutional and macroeconomic predictors. Our main nding here is that
for 12 countries institutions predict returns, while for nine countries macroeconomic variables predict
returns. The heterogeneity of countries is clear, as, for some countries institutions predict returns, for some
countries macroeconomic variables predict returns, while for others none of the factors are able to predict
returns. In all, we nd that for 15 out of the 18 developing countries studied, there is some evidence that
either institutions, macroeconomic variables, or a combination of those variables, predict excess returns.
Second, unlike the return predictability literature that considers nancial ratios as predictors, we nd that
in-sample evidence of predictability is corroborated by out-of-sample tests. It follows that while an investor
may not be prepared to accept return forecasts based on nancial ratios, no such tension exists if the investor
is willing to draw on information contained in combinations of macroeconomic and institutional factors.
Third, while the literature has considered the determinants of returns, suggesting that returns are
predictable using macroeconomic variables, there is no evidence of economic signicance of such predictability.
We undertake an extensive analysis of economic signicance through a mean-variance investor framework.
Table 8
Prots and utility from limited short selling (0.05 to 1.05 and 0.1 to 1.1). In this table, we report prots and utility from dynamic
trading strategies where we conditionally allow for limited borrowing and short-selling. We consider two restrictions; one where
we allow for 5% borrowing and short-selling and the other where we allow for 10% short-selling and borrowing. We report the mean
prots, standard deviation, and utility associated with each of these two strategies, accounting for a 1% transaction cost. ** and ***
denote statistical signicance at the 5% and 1% levels, respectively.
Limited short-selling: 0.05 to 1.05

Argentina
Bangladesh
Brazil
Chile
China
Egypt
Kenya
Lebanon
Malaysia
Oman
Pakistan
Peru
S/Africa
Tunisia
Venezuela

Limited short-selling: 0.10 to 1.1

Mean

SD

Utility

Mean

SD

Utility

2.057**
0.551***
1.673***
0.499***
1.673***
0.601***
0.494***
0.517***
3.469***
3.273***
0.920***
0.635***
0.842***
0.332***
1.316***

4.6576
0.1772
0.5446
0.3655
0.5446
0.3074
0.1691
0.7047
0.6647
2.2938
0.9294
0.1593
0.3088
0.3348
1.7725

1.8193
2.0091
1.672
0.491
1.672
0.5978
0.4941
0.4978
3.4627
3.0918
0.9024
0.6352
0.8412
0.3301
1.2492

1.668**
0.361***
0.926***
0.362***
0.926***
0.390***
0.328***
0.422***
3.338***
3.121***
0.889***
0.398***
0.511***
0.293***
1.220***

4.569
0.1494
0.319
0.3211
0.319
0.1991
0.1515
0.6842
0.6142
2.2149
0.9189
0.1454
0.2269
0.3346
1.7107

1.4501
1.1096
0.9254
0.3554
0.9254
0.3883
0.3276
0.4044
3.3324
2.9535
0.8721
0.398
0.5105
0.291
1.1591

P.K. Narayan et al. / Emerging Markets Review 19 (2014) 7795

93

Table 9
Impact of short-selling on prots and investor utility. This table reports the gains in prots and utilities for those nine countries
where no short-selling is allowed. The utility and prots, reported in column 2, are based on a strategy of no short-selling, while
prots reported in columns 3 and 4 are based on limited borrowing and short-selling of 5% and 10%, respectively. Utility is based on a
risk-aversion factor of six. All estimates take into account 1% transaction cost. *** denotes statistical signicance at the 1% level.
Countries without short-selling

Bangladesh
China
Egypt
Kenya
Lebanon
Oman
Peru
Tunisia
Venezuela

No short-selling

Short-selling (5%)

Short-selling (10%)

Prots

Utility

Prots

Utility

Prots

Utility

0.171***
2.367***
0.179***
0.161***
0.321***
2.967***
0.161***
0.253***
1.125***

0.171
8.786
0.418
0.071
10.87
2.714
0.07
0.234
0.628

0.361***
0.926***
0.390***
0.328***
0.422***
3.121***
0.398***
0.293***
1.220***

1.109
0.925
0.388
0.328
0.404
2.954
0.398
0.291
1.159

0.551***
1.673***
0.601***
0.494***
0.517***
3.273***
0.635***
0.332***
1.316***

2.009
1.672
0.598
0.491
0.498
3.092
0.632
0.330
1.249

More specically, we consider both passive and dynamic trading strategies (including one where limited
borrowing (50%) is allowed). We show that dynamic trading strategies offer higher returns for investors.
Moreover, we discover that allowing for limited borrowing allows investors to make relatively higher prots.
We also estimate investor utility, which merely signies the amount that an investor will be willing to pay to
maintain a particular trading strategy. We nd similar results when we allow limited short-selling of 5% and
10%. For the 15 countries for which returns are predictable, we compare investor utility from passive strategies
with that from dynamic ones. Our main conclusion is that investors have a predilection for dynamic trading
strategies.
Finally, we identify that there are nine countries in our sample where short-selling is prohibited. We
show that if limited borrowing and short-selling are allowed in these countries, then investors in most of
these countries will benet from higher prots and utilities from dynamic trading strategies.
In terms of directions for future research; given what we have discovered, it would seem ideal to
examine if institutional and macroeconomic factors that we consider for emerging country markets would
hold also for other developing country markets. It may well be the case that there are certain types of
developing markets where institutional factors matter more than macroeconomic factors for predicting
returns and vice versa. Understanding exactly which factors predict stock returns will guide potential
trading strategies for investors in these markets. A second avenue for extending the work we have
presented here is by modelling emerging market stock return predictability using a panel data predictive
regression model proposed recently by Westerlund and Narayan (2014b). There are several appealing
features of this panel data predictive regression model, which has implications for trading strategies as
Westerlund and Narayan (2014b) show. We recommend these issues for future research.

References
Aghion, P., Bolton, P., 1992. An incomplete contracts approach to nancial contracting. Rev. Econ. Stud. 59, 473494.
Allen, F., 2001. Do nancial institutions matter? J. Financ. 56, 11651175.
Ang, A., Bekaert, G., 2007. Stock return predictability: is it there? Rev. Financ. Stud. 20, 651707.
Bansal, R., Dittmar, R.F., Lundblad, C.T., 2005. Consumption, dividends, and the cross section of equity returns. J. Financ. LX, 16391672.
Barberis, N., Huang, M., Santos, T., 2001. Prospect theory and asset prices. Q. J. Econ. 116, 153.
Bebchuk, L., Cohen, A., Ferrell, A., 2009. What matters in corporate governance? Rev. Financ. Stud. 22, 783827.
Bekaert, G., Harvey, C.R., Lundblad, C.T., Siegel, S., 2011. What segments equity markets? Rev. Financ. Stud. 24, 38413890.
Boothe, P.M., Reid, B.G., 1989. Asset returns and government budgets in a small open economy. J. Monet. Econ. 23, 6577.
Bragaa-Alves, M.V., Shastri, K., 2011. Corporate governance, valuation and performance: evidence from a voluntary market reform in
Brazil. Financ. Manag. (Spring), 139157.
Breeden, D.T., 1979. An intertemporal asset pricing model with stochastic consumption and investment opportunities. J. Financ.
Econ. 7, 265296.
Cakici, N., Fabozzi, F.J., Tan, S., 2013. Size, value, and momentum in emerging market stock returns. Emerg. Mark. Rev. 16, 4665.
Campbell, J.Y., Cochrane, J.H., 1999. By force of habit: a consumption based explanation of aggregate stock market behaviour. J. Polit.
Econ. 107, 205251.
Campbell, J., Thompson, S., 2008. Predicting the equity premium out of sample: can anything beat the historical average? Rev. Financ.
Stud. 21, 15091531.

94

P.K. Narayan et al. / Emerging Markets Review 19 (2014) 7795

Christensen, J., Kent, P., Stewart, J., 2010. Corporate governance and company performance in Australia. Aust. Account. Rev. 55 (20),
372386.
Cochrane, J.H., 1996. A cross-sectional test of an investment-based asset pricing model. J. Polit. Econ. 104, 572621.
Core, J.E., Guay, W.R., Rusticus, T.O., 2006. Does weak governance cause weak stock returns? An examination of rm operating
performance and investors' expectations. J. Financ. 61, 655687.
Da, Z., 2009. Cash ow, consumption risk, and the cross-section of stock returns. J. Financ. LXIV, 923956.
Danthine, J.P., Donaldson, J., 2002. Labour relations and asset returns. Rev. Econ. Stud. 69, 4164.
Demirguc-Kunt, A., Maksimovic, V., 1998. Law, nance, and rm growth. J. Financ. III, 21072137.
Demirguc-Kunt, A., Maksimovic, V., 1999. Institutions, nancial markets, and rm debt maturity. J. Financ. Econ. 54, 295336.
Dickey, D.A., Fuller, W.A., 1979. Distribution of the estimators for autoregressive time series with a unit root. J. Am. Stat. Assoc. 74,
427431.
Eterovic, N.A., Eterovic, D.S., 2013. Separating the wheat from the chaff: understanding portfolio returns in an emerging market.
Emerg. Mark. Rev. 16, 145169.
Ferreira, M.A., Santa-Clara, P., 2011. Forecasting stock market returns: the sum of the parts is more than the whole. J. Financ. Econ.
100, 514537.
Ferson, W.E., Harvey, C.R., 1994. Sources of risk and expected returns in global equity markets. J. Bank. Financ. 18, 775803.
Ferson, W.E., Harvey, C.R., 1998. Fundamental determinants of national equity market returns: a perspective on conditional asset
pricing. J. Bank. Financ. 21, 16251665.
Foster, F.D., Smith, T., Whaley, R.E., 1997. Assessing goodness-of-t of asset pricing models: the distribution of the maximal R2. J. Financ.
53, 591607.
Garretsen, H., Lensink, R., Sterken, E., 2004. Growth, nancial development, societal norms and legal institutions. J. Int. Finan.
Markets, Inst. Money 14, 165183.
Gennotte, G., Marsh, T.A., 1993. Variations in economic uncertainty and risk premiums on capital assets. Eur. Econ. Rev. 37, 10211041.
Giovannini, A., 1989. Uncertainty and liquidity. J. Monet. Econ. 23, 239258.
Glaeser, E.I., La Porta, R., Lopez-de-Silanes, F., Shleifer, A., 2004. Do institutions cause growth? J. Econ. Growth 9, 271303.
Gompers, P.A., Ishii, J.L., Metrick, A., 2003. Corporate governance and equity prices. Q. J. Econ. 118, 107155.
Harvey, C.R., 1995. Predictable risk and returns in emerging markets. Rev. Financ. Stud. 8, 773816.
Hess, P.J., Lee, B.-S., 1999. Stock returns and ination with supply and demand disturbances. Rev. Financ. Stud. 12, 12031218.
Hjalmarsson, E., 2010. Predicting global stock returns. J. Financ. Quant. Anal. 45, 4980.
Jones, C.M., Lamont, O., Lumsdaine, R.L., 1998. Macroeconomic news and bond market volatility. J. Financ. Econ. 47, 315337.
Kang, J., Kim, T.S., Lee, C., Min, B.-K., 2011. Macroeconomic risk and the cross-section of stock returns. J. Bank. Financ. 35, 31583173.
Kinnunen, J., 2013. Dynamic return predictability in the Russian stock market. Emerg. Mark. Rev. 15, 107121.
Kumar, A., Page, J.K., Spalt, O.G., 2011. Religious beliefs, gambling attitudes, and nancial market outcomes. J. Financ. Econ. 102,
671708.
La Porta, R., Lopez-de-Silanes, F., Shleifer, A., Vishny, R.W., 1997. Legal determinants of external nance. J. Financ. LII, 11311150.
La Porta, R., Lopez-de-Silanes, F., Shleifer, A., Vishny, R.W., 1998. Law and nance. J. Polit. Econ. 106, 11131155.
La Porta, R., Lopez-de-Silanes, F., Shleifer, A., Vishny, R.W., 2000. Investor protection and corporate governance. J. Financ. Econ. 58,
327.
Lehn, K.M., Patro, S., Zhao, M., 2009. Determinants of the size and composition of the U.S. corporate boards: 19352000. Financ.
Manag. 38, 747780.
Lettau, M., Ludvigson, S., 2001. Consumption, aggregate wealth and expected stock returns. J. Financ. 56, 815849.
Lewellen, J., 2004. Predicting returns with nancial ratios. J. Financ. Econ. 74, 209235.
Lucas, R., 1978. Asset prices in an exchange economy. Econometrica 46, 14291446.
Ludvigson, S.C., Ng, S., 2007. The empirical riskreturn relation: a factor analysis approach. J. Financ. Econ. 83, 171222.
Marquering, W., Verbeek, M., 2004. The economic value of predicting stock index returns and volatility. J. Financ. Quant. Anal. 39,
407429.
Mateus, T., 2004. The risk and predictability of equity returns of the EU accession countries. Emerg. Mark. Rev. 5, 241266.
McCracken, M.W., 2007. Asymptotics for out-of-sample tests of Granger causality. J. Econ. 140, 719752.
Narayan, P.K., Narayan, S., Sharma, S.S., 2013. An analysis of commodity markets: what gain for investors? J. Bank. Financ. 37,
38783889.
Narayan, P.K., Sharma, S., Poon, W.C., Westerlund, J., 2014. Do oil prices predict economic growth? New global evidence. Energy Econ.
41, 137146.
Papaionannou, E., 2009. What drives international nancial ows? Politics, institutions and other determinants. J. Devel. Econ. 88,
269281.
Paye, B.S., Timmermann, A., 2006. Instability of return prediction models. J. Empirical Finance 13, 274315.
Qian, J., Strahan, P.E., 2007. How laws and institutions shape nancial contracts: the case of bank loans. J. Financ. 62, 28032834.
Rapach, D.E., Wohar, M.E., 2006a. In-sample vs. out-of-sample tests of stock return predictability in the context of data mining.
J. Empirical Finance 13, 231247.
Rapach, D.E., Wohar, M.E., 2006b. Structural breaks and predictive regression models of aggregate U.S. stock returns. J. Financ. Econ.
4, 238274.
Rapach, D.E., Strauss, J.K., Zhou, G., 2010. Out-of-sample equity premium prediction: combination forecasts and links to the real
economy. Rev. Financ. Stud. 23, 821862.
Santos, T., Veronesi, P., 2006. Labour income and predictable stock returns. Rev. Financ. Stud. 19, 144.
Shleifer, A., Vishny, R.W., 1997. A survey of corporate governance. J. Financ. 52, 737783.
Stambaugh, R., 1999. Predictive regressions. J. Financ. Econ. 54, 375421.
Stock, J.H., Watson, M.W., 2002. Forecasting using principal components from a large number of predictors. J. Am. Stat. Assoc. 97,
11671179.
Welch, I., Goyal, A., 2008. A comprehensive look at the empirical performance of equity premium prediction. Rev. Financ. Stud. 21,
14551508.
Westerlund, J., Narayan, P.K., 2012. Does the choice of estimator matter for forecasting returns? J. Bank. Financ. 36, 26322640.

P.K. Narayan et al. / Emerging Markets Review 19 (2014) 7795

95

Westerlund, J., Narayan, P., 2013. Testing the efcient market hypothesis in conditionally heteroskedastic futures markets. J. Futur.
Mark. 33, 10241045.
Westerlund, J., Narayan, P., 2014a. Testing for predictability in conditionally heteroskedastic stock returns. J. Financ. Econ. http://dx.
doi.org/10.1093/jjnec/nbu001.
Westerlund, J., Narayan, P., 2014b. A random coefcient approach to the predictability of stock returns in panels. J. Financ. Econ.
http://dx.doi.org/10.1093/jjnec/nbu003.