Are Islamic stock returns predictable? A global perspective

    Are Islamic Stock Returns Predictable? A Global Perspective Paresh Kumar Narayan, Dinh Hoang Bach Phan, Susan Sunila Sharma, Joakim Westerlund PII: DOI: Reference:

S0927-538X(16)30127-5 doi: 10.1016/j.pacfin.2016.08.008 PACFIN 878

To appear in:

Pacific-Basin Finance Journal

Received date: Accepted date:

21 June 2016 29 August 2016

Please cite this article as: Narayan, Paresh Kumar, Phan, Dinh Hoang Bach, Sharma, Susan Sunila, Westerlund, Joakim, Are Islamic Stock Returns Predictable? A Global Perspective, Pacific-Basin Finance Journal (2016), doi: 10.1016/j.pacfin.2016.08.008

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT Are Islamic Stock Returns Predictable? A Global Perspective

T

Paresh Kumar Narayan1, Dinh Hoang Bach Phan2, Susan Sunila Sharma3, Joakim Westerlund4

SC R

IP

Centre for Financial Econometrics, Deakin Business School, Deakin University

This Version: 15 March 2016

NU

Mailing Address

AC

CE P

TE

D

MA

Paresh Kumar Narayan Alfred Deakin Professor Centre for Financial Econometrics Deakin Business School Deakin University 221 Burwood Highway Burwood, Victoria 3125 Australia Telephone: +61 3 9244 6180 Fax: +61 3 9244 6034 Email: [email protected]

1

Alfred Deakin Professor, Centre for Financial Econometrics, Deakin Business School, Deakin University, Australia; Email: [email protected]. 2 Research Fellow, Centre for Financial Econometrics, Deakin Business School, Deakin University, Australia; Email: [email protected]. 3 Senior Lecturer in Financial Econometrics, Centre for Financial Econometrics, Deakin Business School, Deakin University, Australia; Email: [email protected]. 4 : Professor in Financial Econometrics, Lund University (Sweden) and Centre for Financial Econometrics, Deakin University (Australia); Email: [email protected].

ACCEPTED MANUSCRIPT Are Islamic Stock Returns Predictable? A Global Perspective

ABSTRACT

T

Using the sharia-compliant measures, we compile a data set that spans January 1981 to

IP

December 2014 and contains 2,577 Islamic stocks. Using as many as 12 financial and

SC R

macroeconomic predictors, we discover strong evidence of both in-sample and out-of-sample return predictability. There is robust evidence of predictability only when U.S. stock returns

NU

are used as a predictor. We find that investing in regional (industry) portfolios offers on average, across the 12 predictors, meaningful profits of 6.16% (6.03%) per annum. Investing

MA

in a portfolio of Islamic stocks belonging to emerging markets (9.89% per annum) and a portfolio of Islamic stocks belonging to the consumer goods sector (6.37% per annum) offer

Stocks;

CE P

TE

Islamic

AC

Keywords:

D

the most returns amongst regions and industries, respectively. Predictability;

Returns;

Profits.

ACCEPTED MANUSCRIPT I.

Introduction

There is a proliferation of research on Islamic finance. Such has been the interest on Islamic

T

finance that recent years have seen multiple special issues on Islamic finance published in

IP

such journals as Journal of Economic Behaviour and Organization and Pacific-Basin

SC R

Finance Journal. A key challenge which has become obvious from the work on Islamic finance is the lack of historical time-series data; therefore, research questions that revolve around the use of time-series data at the stock-level are limited. Most of the time-series

NU

analyses, as a result, make use of index-level data. In this regard, the Dow Jones Islamic

MA

Stock Price Index has been the most popular. The aim of our paper is to compile a new comprehensive data set on Islamic stocks. Drawing on the Sharia-compliant principles, we

D

screen all Islamic stocks listed on stock exchanges globally. The screening procedure is

TE

aimed at selecting stocks that persistently follow the Sharia-compliant criterion. Over the time period January 1981 to December 2014, we end up with 2,577 Islamic stocks; more

CE P

details are provided later, in Section II. Our concurrent aim is to test whether the time-series of Islamic stock returns are

AC

predictable. This is not a trivial question despite the fact that the literature on stock return predictability is voluminous and, therefore, rich (see, inter alia, Fama, 1981; Campbell, 1987; Fama and French, 1988, 1989; Campbell and Shiller, 1988a,b; Kothari and Shanken, 1997; Pontiff and Schall, 1998; Lamont, 1998; Rapach et al., 2005; Welch and Goyal, 2008; Campbell and Thompson ,2008; Rapach et al. 2010, Westerlund and Narayan, 2012). However, none of these studies considers whether Islamic stock returns are predictable. Ignoring Islamic stock returns can be costly because there are a number of important differences between Islamic stocks and non-Islamic stocks. Islamic stocks that form part of the Dow Jones Islamic Market World Index and its sub-indices, reflecting country-specific, regional and industry attributes, cover investment products that facilitate ethical investing

2

ACCEPTED MANUSCRIPT within the context of Sharia-principles and differ from conventional stocks in two main ways. The first distinguishing feature is that for a stock to be categorized as a Sharia-compliant

T

Islamic investment it must satisfy the business activity criteria. Twenty-three business

IP

activities5 are regarded as inappropriate by Sharia-principles. In addition, a company will not

SC R

qualify if its income source is from either alcohol, tobacco, pork-related products, conventional financial services (such as banking and insurance), weapons and defence, and/or entertainment and exceeds 5% of its total revenue.

NU

The second feature that distinguishes Islamic stocks from conventional stocks relates

MA

to the financial health of the firm with particular emphasis on solvency-related measures. For instance, to qualify as an Islamic stock (a) total debt to market capitalization, (b) cash and

D

interest bearing securities to market capitalization, and (c) accounts receivables to market

TE

capitalization, should all be less than 33% of the 24-month average trailing market capitalization6. Given the screening criteria applicable to business activities and, in particular,

CE P

the financial health of individual stocks, the discriminatory ability of Islamic stocks could offer a different story regarding stock return predictability compared to what we already

AC

know with respect to non-Islamic stocks. Whether or not this is the case is the subject of this paper and, therefore, our paper takes a step in this direction of understanding Islamic stock pricing.

In addition to these features of Islamic finance, it is also imperative to note that the Islamic finance industry has grown rapidly over the last two decades, in the process becoming an alternative (to conventional finance) investment option. The main attractiveness

5

The Dow Jones system, for example, identifies the following business activities as inappropriate for Islamic investments: Defence, Brewers, Distillers & Vintners, Food Products, Recreational Products, Tobacco, Food Retailers & Wholesalers, Broadcasting & Entertainment, Media Agencies, Gambling, Hotels, Recreational Services, Restaurants & Bars, Banks, Full Line Insurance, Insurance Brokers, Property & Casualty Insurance, Reinsurance, Life Insurance, Consumer Finance, Specialty Finance, Investment Services, and Mortgage Finance. 6 The trailing 24-month average market capitalization is used, which avoids any skewed figures arising from factors regarded as seasonal.

3

ACCEPTED MANUSCRIPT of Islamic finance is that it brings greater diversification and financial stability (see Balcilar et al. 2015). The total value of Islamic finance assets under management was estimated to be

T

more than US$2 trillion in 2014 with a compounded average growth rate of over 17% over

IP

2010-2014 (Ibrahim, 2015). Ibrahim (2015) points out that the Islamic financial sector is no

SC R

longer an investment option for faith-based (Muslim) investors—a point echoed by Umar (2015)—and that it is catering for the needs and demands of new customs which are predominantly non-Muslims. The recent global financial crisis has emphasized the

NU

importance of Islamic finance—which has provided the much needed stability to the financial

MA

system at a time it was most needed. Likewise the global emerging sukuk (Islamic bond) market is gaining prominence and is valued at around US$130 billion ((see Balcilar et al. 2015). This achievement is not trivial, a point made by Ibrahim (2015: 189) when he notes

TE

D

that: “The most exciting development is the acceptance of sukuk outside the Muslim world as

Luxembourg”.

CE P

manifested by its issuance in … the UK, Senegal, Hong Kong, South Africa, and

To address our proposed research question, we follow three steps. In the first step, we

AC

handpick all Islamic stocks from around the globe, that is, Islamic stocks that are listed on the stock exchanges from around the world; see section II for details. This search leads to a total of 2,577 stocks. In the second step, we identify a range of predictors, those that have been popular in the stock return predictability literature. We identify 12 predictors of returns. These first two steps allow us to compile a unique stock-level time-series monthly data over the period January 1981 to December 2014 for 2,577 Islamic stocks. In the third step, we undertake econometric tests—both in-sample and out-of-sample tests—for predictability. These statistical tests are complemented by an economic significance test where we estimate investor utility gains and profits for an investor faced with a mean-variance utility function.

4

ACCEPTED MANUSCRIPT Our approaches deliver the following new insights on the behaviour of Islamic stocks. First, out of the 12 commonly used predictors of returns, we find that exchange rate returns,

T

U.S. stock returns and a commodity price index returns consistently predict both regional and

IP

sectoral stock returns. This predictability holds in both in-sample and out-of-sample tests. By

SC R

comparison, inflation and money supply are the weakest predictors of returns. Second, specifically on out-of-sample evidence of predictability, we find that U.S. stock returns beat the constant returns model more consistently than does the combination forecast method.

NU

Third, in economic significance analysis, based on a mean-variance investor utility function,

MA

we discover that utility gains are positive for all regions and sectors when using U.S. returns as a predictor. Moreover, the annualized utility gains fall in the 0.42% to 3.09% range with an

D

average utility gain of 1.73% and 1.61% per annum for regions and sectors, respectively.

TE

Fourth, we also estimate profits for a mean-variance investor and find that the most profitable region is the emerging markets with annualized profits of 9.89% followed by Africa (8.01%).

CE P

The developed countries are the least profitable with an annualized profit of 3.56%. Amongst sectors, the annualized profits fall in the 5.70% to 6.37% range.

AC

Our findings connect with several strands of the literature. Our first finding confirming the leading role of U.S. stock returns connects with a recent study by Rapach et al. (2013), who show that U.S. returns predict stock market returns of 11 industrialized countries. Our findings reveal the important role of U.S. returns for a large time-series data set on Islamic stocks.

Our second finding, establishing the economic significance of

predictability, connects with the broader literature on stock return predictability which shows that forecasting models do beat the constant returns model in providing investors greater utility. We discover this too; however, a caveat is in order. Of the 12 predictors we consider, not all offer investors utility gains over the constant returns model. Therefore, the choice of predictors and indeed forecasting models is important in maximising investor utility. In this

5

ACCEPTED MANUSCRIPT exercise, again the forecasting model based on using U.S. returns as a predictor provides investors utility gains over the constant returns model regardless of whether we consider

T

regional or sectoral portfolios. The key implication of these findings is that in any future

IP

evaluation of asset pricing of Islamic stocks, regardless of whether it is empirical or

our results set the motivation for future studies.

SC R

theoretical, an explicit role for U.S. stock returns would need to be established. In this regard,

Our finding that Islamic stocks are profitable contributes to the existing debate on the

NU

profitability of Islamic stocks. Some studies (see Al-Khazali et al., 2014; and Ho et al., 2014)

MA

find mixed evidence that Islamic stock markets are profitable: these studies show that Islamic stock indices outperform conventional indices during crisis periods but not during non-crisis

D

periods. In addition, Hayat and Kraeussl (2011) discover weak evidence of profitability. On

TE

the other hand, Ashraf and Mohammad (2014), Bialkowski et al. (2012), and Hoepner et al. (2011) find relatively convincing evidence that Islamic stocks are profitable. There are two

CE P

key differences between our study and this literature. First, we consider the largest universe of Islamic stocks over a 30-year period whereas this literature considers not stocks but market

AC

indices to draw conclusions on the profitability of the Islamic stock market. This is not a trivial difference because by considering firm level data for over 2,500 stocks, we provide robust evidence of profitability of Islamic stocks. Second, our approach of estimating profits is different from this literature. We estimate profits using a sound theoretical framework— that is, the mean-variance investor utility function, which has been shown to work well in time-series forecasting of stock returns; see Campbell and Thompson (2008) and studies thereafter. Our study also connects with the debate between in-sample and out-of-sample forecasting evaluations. There is an active debate and one that is not tension-free. There are two groups of studies, one which shows a predilection for in-sample tests (see Welch and

6

ACCEPTED MANUSCRIPT Goyal, 2008; Ashley et al. 1980) while the other argues that out-of-sample tests should be given more prominence (Inoue and Kilian, 2004). Our study is important in this debate

T

because it, we believe, eases this tension. Our empirical setup leads to estimation of a total of

IP

215 time-series predictive regression models. Out of this, in 118 predictive regression models

SC R

(representing 55% of our regressions), we reject the null hypothesis of no predictability. By comparison, we find statistically significant evidence for out-of-sample predictability (that is, where forecasting models beat the constant returns model) in 122/215 cases (58%).

NU

Therefore, with our forecasting story, we find fairly even evidence of predictability from both

MA

in-sample and out-of-sample tests. The key message from our empirical analysis is that when a large number of forecasting models are estimated, the chances of obtaining consistent

D

results between in-sample and out-of-sample tests increase. We are unaware of forecasting

TE

studies that have estimated hundreds of models as we have done. We are able to do this due purely to the richness of our data set.

CE P

The rest of the paper proceeds as follows. Section II is about the data and concludes with a statistical description of the data. Section III presents and discusses the results. A

section.

AC

range of robustness tests occupy Section IV, and concluding remarks are contained in the last

II.

Data and descriptive statistics

Our data set includes 2,577 Islamic stocks which are components of the Dow Jones Islamic Index. The data is monthly and covers the period January 1981 to December 2014. This data are from Datastream. The number of stocks (as noted in parenthesis below) in aggregate markets and sectors are as follow: 

Developed (1636), Emerging (941), Europe (405), Australasia (92), Asia (1294), Americas (729), and Africa (57).

7

ACCEPTED MANUSCRIPT 

Basic materials (258), consumer goods (465), consumer services (249), financials (130), health care (311), industrial (577), oil and gas (160), technology (329),

T

telecommunication (57), and utilities (38).

IP

The stock returns of aggregate markets and sectors are computed as equal-weighted returns of

SC R

stocks belonging to the particular market or sector. The definition of the rest of the variables and their source are noted in Table I. It follows that at the regional level the number of stocks fall in the 57 to 1,636 range whereas at the sectoral level the number of stocks fall in the 38 to

NU

577 range.

MA

In Table II (Panel A), selected descriptive statistics on excess returns for various portfolios, including global (all countries), developed countries, and emerging countries, and

D

portfolios of regional countries (Europe, Australasia, Asia, Africa, and America) are

TE

provided. In panel B, by comparison, we form various sectoral portfolios and report selected descriptive statistics. What we observe from the mean and standard deviation statistics is how

CE P

portfolios of Islamic stocks are likely to be different. Specifically, amongst regional/global portfolios the monthly mean excess returns are in the -0.89 to 0.48% range, whereas amongst

AC

sectoral portfolios the monthly excess returns fall in the -0.51 to 0.24% range. Portfolios of Islamic stocks belonging to Australasia and Africa are most volatile while portfolios that include all stocks belonging to developed countries and a portfolio of all 2,577 stocks are the least volatile. A similar heterogeneity is observed for sectoral portfolios; volatility is in the 4.54% (HC) to 7.79% (TEL) range. The first-order autoregressive coefficients for each of the portfolios of excess returns are all less than 30%, suggesting that shock persistence is low. Except for the American portfolio and sectoral portfolios of CS, FI and UT, the remaining portfolios are all characterized by strong ARCH effects, that is, the null hypothesis of “no ARCH” is rejected with a p-value in the 0 to 0.08 range. Finally, we observe that most portfolio returns are

8

ACCEPTED MANUSCRIPT reasonably persistent as indicated by the first-order autoregressive coefficient. For the global and regional portfolios, return persistency is in the 14% to 28% range, whereas for the

T

sectoral portfolios the persistency is in the 8% to 30% range. The implication is that past

IP

returns influence current returns.

SC R

We now turn to Table III where we report descriptive statistics for each of the predictors. We have both region-specific and common predictors. The nine region-specific (or individual) predictors are: short-term rate (STR), long-term rate (LTR), market rate (MR),

NU

term spread (TS), inflation (INF), industrial production growth (IPG), difference in

MA

unemployment rate (DUN), money growth (M1), and exchange rate return (ER). The three common predictors we use are: a credit risk factor (CRF), the U.S. excess returns (US), and

D

the commodity index returns (CIR). For each of these 12 predictors, we report mean, standard

TE

deviation, AR(1) coefficient, and a test of heteroscedasticity. There are several interesting features of the predictors’ data. First, across all regions the most volatile variables are

CE P

industrial production growth, money growth, and exchange rate returns. Second, most predictors across the regions are characterized by heteroscedasticity; the exceptions are

AC

exchange rate returns, long term interest rate, and the unemployment rate for some of the regions. Third, most predictors are highly persistent regardless of region. In predictive regression models, the endogeneity of the predictor variable has been identified as a key statistical constraint, which has motivated theoretical attention. The aim of this section is to gauge the presence (or otherwise) of the endogeneity of each of the predictors for each of the global/regional set of stocks and for each of the sectors making our data set. We report the p-value of the test that the coefficient 𝛾 in the equation 𝜖𝑡 = 𝛾𝜀𝑡 + 𝜂𝑡 is zero, 𝜖𝑡 is the residual from the predictive regression model 𝑟𝑡 = 𝛼 + 𝛽𝑃𝑡−1 + 𝜖𝑡 , and 𝜀𝑡 is the residual from the AR(1) regression of the predictor 𝑃𝑡 = 𝜇(1 − 𝜌) + 𝜌𝑃𝑡−1 + 𝜀𝑡 .

9

ACCEPTED MANUSCRIPT Rejecting the null that 𝛾 = 0 suggests that endogeneity exists in the predictive regression model.

T

The results for the eight regions are reported in Panel A of Table IV, while Panel B

IP

reports the results for the 10 sectors. The null hypothesis that the slope coefficient is zero is

SC R

rejected in 7/12 for global and emerging; 5/12 for developed, Europe, and America; 8/12 for Australasia and Asia, and 6/11 models for Africa. Amongst the sectors, we find that the slope coefficient is rejected in the case of BM, CG, CS, OG, and TEL for 6/12 predictors; for

NU

FI and UT in the case of 8/12 predictors; for HC in the case of 5/12 predictors; and for IN and

MA

TE in the case of 7/12 predictors.

The implication is the following. Not all predictors are endogenous. In other words,

D

evidence on endogeneity is mixed. For example, of the 95 predictive regression models for

TE

global/regions, in 51/95 (54%) regression models the predictor variable is endogenous. Similarly, for sectoral predictive regression models, in 65/120 (54%) models the predictor

CE P

variable is endogenous. It follows that endogeneity is an issue that we need to explicitly deal with together with predictor persistence and heteroscedasticity, as identified earlier. These

AC

salient features of the data are accommodated by a flexible generalised least squares-based predictive regression model proposed by Westerlund and Narayan (2015), which is what we use to test for predictability.

III.

Results

A. In-sample predictability The predictability results are reported in Table V; panel A has results for each of the global/regional set of countries using each of the 12 predictor variables. The results can be summarized as follows. First, we find that STR, ER, U.S. returns and CIR are the most popular predictors since they predict returns for the global as well as for each of the region-

10

ACCEPTED MANUSCRIPT specific countries. Second, LTR, MR and TS are the second most popular predictors—they are only unable to predict returns of Europe (LTR) and Africa (MR and TS). Third, the least

T

popular predictors are IPG and DUN, which do not predict returns for any of the countries,

IP

followed by INF and M1, which only predict returns for Asia.

SC R

Panel B has results for each of the sectors. The most popular predictors are ER, U.S. returns and CIR; these three predict returns for all 10 sectors. STR, which predicts returns of 8/10 sectors, and TS, which predicts returns of 7/10 sectors, are the second and third most

NU

popular predictors, respectively. Other predictors, such as MR and LTR, are also important,

MA

predicting returns of 6/10 and 4/10 sectors, respectively. Meanwhile, M1 does not predict returns of any of the sectors, and INF and IPG predict returns of only one sector, while DUN

D

predicts returns of two sectors.

TE

The main messages from in-sample tests are twofold. First, not all 12 predictors predict Islamic stock returns—some are more powerful predictors of returns than others, and

CE P

some predictors (ER, U.S. returns, and CIR) predict returns for all regions and sectors. Second, with 12 predictors and 18 regions/sectors, we estimate a total of 215 time-series

AC

predictive regression models. Out of this, in 118 predictive regression models (representing 55% of all regressions), we reject the null hypothesis of no predictability.

B. Out-of-sample results This section reports out-of-sample evaluations of the importance of predictors in forecasting excess returns vis-à-vis the constant returns model. The forecasting regression model is of the following form: 𝐸𝑅𝑡 = 𝛼 + 𝛽𝑃𝑡−1 + 𝜖𝑡

(1)

A 50% in-sample period is used to generate recursive forecasts of excess returns for the remaining 50% of the sample. Besides the forecasts from individual and common predictors,

11

ACCEPTED MANUSCRIPT we also apply three forecasting combination approaches as suggested by Rapach et al. (2010), namely mean, median, and trimmed mean (T-mean). We report the out-of-sample

T

(𝑂𝑅 2 ), which examines the difference in the mean squared errors from the competition model

IP

and the constant returns model. The Clark and West (2007) MSFE-adjusted test statistic,

SC R

denoted with an asterisk, examines the null hypothesis that the 𝑂𝑅 2 = 0 against the alternative that 𝑂𝑅 2 > 0. These statistics for the eight regions (10 sectors) occupy Panel A (Panel B) of Table VI.

NU

Consider first the regional out-of-sample predictability results based on the null

MA

hypothesis that 𝑂𝑅 2 = 0; when forecasts are based on U.S returns as a predictor, the null is rejected for all regions and the coefficient is positive suggesting that a forecasting model that

D

uses U.S. returns to forecast regional returns beats the constant returns model. The second

TE

most popular forecasting model is the one that uses CIR, MR, TS, and M1 as the null is rejected in 5/8 regions, followed by the LTR and INF based models for which the null is

CE P

rejected in 4/8 regions. The other forecasting models that use IPG, DUN, ER and CRF reject the null hypothesis for at most three regions.

AC

We turn to Panel B where sectoral return results are reported. Beginning with the null hypothesis that 𝑂𝑅 2 = 0, it is the U.S. returns- and the TS-based forecasting models that reject the null hypothesis (and the sign on the coefficient is greater than zero) for most sectors (9/10), followed by INF- and STR-based forecasting models (7/10 sectors), and LTR and M1 based forecasting models (6/10 sectors). Four models (MR, IPG, DUN, and ER) reject the null in 50% of the sectors while the weakest models are those that use CIR (4/10 rejections) and CRF (2/10 rejections) as predictors. We conclude the out-of-sample analysis by applying a combination forecast evaluation method proposed by Rapach et al. (2010). These are based on simple average schemes, such as mean, median and trimmed mean. The mean combination forecast sets the

12

ACCEPTED MANUSCRIPT weight as 1/N, where N is the total number of predictive regressions. The median combination forecast is the median of the forecasted returns amongst all N. Finally, the

T

trimmed mean combination forecast generates weights by setting weights to zero for

IP

individual forecasts with the smallest and largest values and a weight of 1/(N-2) for the

SC R

remaining individual forecasts. The results from each of these three test statistics are reported in columns 2-4 of Table VI. All three statistics provide robust evidence that returns of global, emerging, Europe, Australasia, Asia, and Africa regions are predictable. There is no evidence

NU

that returns of developed markets and Americas are predictable. At the sectoral level, we find

MA

that all sectoral returns are predictable with the exception of CS and UT. We make an interesting observation with respect to the performance of the

D

combination forecast method vis-à-vis the U.S. returns-based forecasting model. We find that

TE

neither amongst the eight regions nor amongst the 10 sectors do the combination forecast method beats the U.S. returns-based model. For example, where the combination forecast

CE P

method reveals predictability for 6/8 regions and 8/10 sectors, U.S. returns-based models beat the constant returns model in all eight regions and in 9/10 sectors.

AC

The main messages from this out-of-sample analysis are the following. First, excluding the combination forecast results, we find that in 122/215 forecasting models the constant returns model is beaten, suggesting that the evidence of predictability from insample and out-of-sample tests are fairly even. Second, there is consistency in terms of the most successful and the weakest predictors of returns in both in-sample and out-of-sample evaluations. For example, in both exercises U.S. returns and CIR turn out to be the most powerful predictors whereas the weakest predictors are inflation and M1.

C. Utility gains

13

ACCEPTED MANUSCRIPT These results based on the 𝑂𝑅 2 are not a perfect measure of economic significance because it does not account for investor risk aversion over the out-of-sample period. To circumvent this

T

limitation, the literature (see Rapach et al. 2010; Campbell and Thompson, 2008)

IP

recommends using a utility-based measure. This section is a response to this proposal, where

SC R

we compute utility for a mean-variance (MV) investor on a real time basis. The utility function has the following form:

(2)

NU

1 𝐸𝑡 {𝑟𝑡+1 } − 𝛾𝑉𝑎𝑟𝑡 {𝑟𝑡+1 } 2

For a given portfolio weight (𝜋𝑡+1 ) in period t+1 for the risky asset, the utility simply

MA

becomes:

(3)

TE

D

1 2 𝑟𝑓,𝑡+1 + 𝜋𝑡+1 𝐸𝑡 {𝑟𝑡+1 } − 𝛾𝜋𝑡+1 + 𝑉𝑎𝑟𝑡 {𝑟𝑡+1 } 2

Where 𝑟𝑡+1 is the return on the stocks, 𝑟𝑓,𝑡+1 is the risk-free rate of return, 𝑉𝑎𝑟𝑡 is the rolling

CE P

variance of the risky asset, 𝛾 is the risk aversion factor, and the investor’s portfolio weight is

AC

computed as follows:

∗ 𝜋𝑡+1 =

𝐸𝑡 {𝑟𝑡+1 } − 𝑟𝑓,𝑡+1 𝛾𝑉𝑎𝑟𝑡 {𝑟𝑡+1 }

(4)

Apart from restricting the portfolio weight to between 0 and 1, we also constrain the portfolio weight on stocks to lie between -0.5 and 1.5, representing an investor who undertakes shortselling and borrowing of up to 50%.

The average utility level, ex post, becomes: 𝑇−1

1 1 2 ̂ = ∑ [𝑟𝑡+1 − 𝛾𝜋𝑡+1 𝑈 𝑉𝑎𝑟𝑡 ] 𝑇 2

(5)

𝑡=0

The average utility is computed for the investor’s dynamic portfolio as well as the passive portfolio (one that derives forecasts from the constant returns model). This allows us to 14

ACCEPTED MANUSCRIPT compare the utilities from these two types of portfolios. We are, as a result, able to obtain the maximum fee an investor is willing to pay for holding the dynamic portfolio over a passive

T

one. In generating the utility gain, we need to have a forecast of returns and its variance. Our

IP

approach is to use 50% of the sample as an in-sample period and generate recursive forecasts

SC R

of excess returns for the remaining 50% of the sample. The variance is computed using a 60month rolling window of returns. Our portfolio is a two-asset portfolio, where one is a risky asset and the other is a risk-free asset. The portfolio weight is used to decide the proportion of

NU

a dollar to be invested in the risky asset (which is the stock market) and the risk-free asset

MA

(which is the three-month bill rate). As is common in this mean-variance setup, we make two assumptions: (a) investors only use public information to forecast the one-period ahead

D

excess returns; and (b) investors rebalance their portfolio once a month (since we use

TE

monthly data). The utility gains are the difference between the utility from a trading strategy based on our predictor-based model and the benchmark constant returns model. The utility

CE P

gains are annualized. The average utility gains for each region and sector based on each predictor are reported in the last row and column of Table VII.

AC

We begin with utility gains based on a portfolio weight that does not allow for any short-selling or borrowing. These results are reported in panel A of Table VII. The utility gains are positive for all regions and sectors when using the U.S. returns as a predictor. The annualized utility gains fall in the 0.42% to 3.09% range with an average utility gain of 1.73% and 1.61% per annum for the region and sectors, respectively. There are a total of 18 models, both regional and sectoral models combined. The predictor STR has a positive utility gain in 16/18 return models while for two predictors (ER and CIR) utility gain is positive in at least 15/18 models. Utility gains are positive in 14/18 (TS), 13/18 (MR), 12/18 (INF), and 11/18 (DUN and M1) return models. The least gain in utility is observed for IPG (7/18 models), and LTR and CRF (9/18 models). On average (across the 18 return models) while

15

ACCEPTED MANUSCRIPT the utility gain is maximised when using U.S. returns as a predictor, the second best model is CIR followed by ER with sectoral annualized utility gains of 0.35% and 0.32% per annum,

T

respectively. Of the 12 predictors, eight on average have positive utility gains. In Panel B, we

IP

report results based on 50% short-selling and borrowing. The results are broadly consistent

SC R

with those observed without allowing for short-selling and borrowing.

D. Investor Profits

NU

Using the portfolio weights derived for a mean-variance investor in Section C, we can now compute dynamic profits (P) over time for an investor as follows:

MA

∗ ∗ )𝑟 𝑃𝑡 = 𝜋𝑡+1 × 𝑟𝑡+1 + (1 − 𝜋𝑡+1 𝑓,𝑡+1

(6)

D

The average annualized profits are reported in Table VIII. Panel A has results based on zero

TE

short-selling and borrowing. This table addresses two questions. First, which are the most profitable regions and sectors? The most profitable region is the emerging markets with an

CE P

annualized profit of 9.89% followed by Africa (8.01%). Asia and Europe make annualized profits of 6.13% and 6.11%. The developed countries are the least profitable (3.56%).

AC

Amongst sectors, the annualized profits fall in the 5.70% to 6.37% range; CG, HC, and IN are amongst the most profitable sectors; and TEL is the least profitable sector. The second question is: which is the most successful predictor? In other words, which predictor maximizes profits? To answer this, we average profits across the eight regions and across the 10 predictors. From this exercise, we find that when using U.S. returns as a predictor annualized regional profits are 8.42% followed by CIR predictor (6.32%) while the least profitable predictor is DUN (5.52%). The story is very similar when considering sectoral profitability. Profits are maximized when using U.S. returns at 7.84%, followed by STR (6.17). The least profits are obtained when using IPG (5.65% per annum).

16

ACCEPTED MANUSCRIPT We now re-estimate profits by allowing for 50% short-selling and borrowing. These results are reported in Panel B. The results corroborate those obtained without short-selling

T

and borrowing from the perspective of the most profitable regions/sectors and the most

IP

successful predictors. In summary, the first thing to note is that profits are higher with short-

SC R

selling and borrowing. For example, the emerging markets are still the most profitable with an annualized profit of 10.56% followed by Africa (8.18%). Similarly, among sectors, IN is the most profitable with an annualized profit of 6.48%. Amongst the most successful

IV.

MA

(regions) and 8.72% per annum (sectors).

NU

predictors, it is still U.S. returns, which on average deliver profits of 9.62% per annum

Robustness Test

D

In this section, we undertake a rigorous robustness test analysis of the results obtain so far.

TE

Our attempt at checking the robustness of our results proceeds along several lines, as follows.

CE P

First, one of our key findings is that regions and sectors are profitable. A natural question here is: are these profits are a compensation for risks or simply a result of mispricing. Our approach to testing this is through using the following regression specification:

AC

Profit 𝑡 = 𝛼 + 𝛽𝑀𝐾𝑇 [𝑅𝑚,𝑡 − 𝑟𝑓,𝑡 ] + 𝛽𝑆𝑀𝐵 𝑆𝑀𝐵𝑡 + 𝛽𝐻𝑀𝐿 𝐻𝑀𝐿𝑡 + 𝛽𝐵𝐴𝐵 𝐵𝐴𝐵𝑡 + 𝜀𝑡 Here Profit is the equal-weighted time-series mean-variance portfolio profits obtained when using each of the 12 predictors in the forecasting model. Portfolio profits are regressed on global Islamic risk factors. The risk factors are market excess returns (MKT), small-minusbig (SMB) returns, high-minus-low (HML) book-to-market returns, and betting-against-beta (BAB) returns. These factors have been computed by Narayan, Narayan, Phan, Thuraisamy, and Tran (NNPTT, 2015) and we use these factors here. The regression is run using OLS where standard errors are corrected for heteroskedasticity and autocorrelation based on Newey-West procedure. A maximum of 12 lags are allowed for and the optimal lag length is

17

ACCEPTED MANUSCRIPT chosen using the Schwarz Information criterion. We estimate a total of 269 time-series regressions; there are 12 predictors by 18 regional and sectoral profits, plus there are the

T

mean, medium and trimmed-mean profits for each of the 18 regions/sectors. Therefore, even

IP

excluding these three combination forecast based profits, on individual predictors alone we

SC R

have 215 time-series regressions, resulting into 215 abnormal returns. The coefficient on alpha (abnormal returns) is reported (see Table IX). The first thing to note (unreported) is that alphas are statistically different from zero. The t-statistics (unreported) suggest that all

NU

estimated alphas are statistically different from zero at the 1% level of significance. The

MA

implication is that even after accounting for commonly known risk factors, Islamic stocks, both regional and sectoral, are profitable. The second observation we make is in terms of

D

magnitudes of profits. Without accounting for risks, U.S.-based models provided the most

TE

profits followed by CIR-based models, and the least profit was obtained by DUN-based model. The same trend is noticed when risk factors are accounted for; annualized alpha turns

CE P

out to be 7.61% (U.S. returns), 6.09% (CIR), and 4.81% (DUN). The results are also similar when we consider the most profitable regions. Previously, without adjusting for risks, we

AC

found the emerging markets to be the most profitable, followed by Africa, with developed countries been the least profitable. The same is true even after adjusting for risks; annualized abnormal returns are 9.67% (emerging markets), 7.27% (Africa), and 2.96% (developed countries). The results reported in Panel B of Table IX are the resulting alphas from a regression of time-series profits (equal-weighted portfolio) based on 50% short-selling and borrowing. The results are robust regardless of how we compute profits. We take another attempt at testing the robustness of profits, concerned that in case the Islamic factors we have considered, based on the NNPTT factors, may not be fully capturing all risks. In this case, we replace the Islamic risk factors with those available for the U.S.

18

ACCEPTED MANUSCRIPT market. These factors, except for BAB, are available from Kenneth French’s website 7 while the BAB factor is obtained from Frazzini’s website.8 The 269 regression alphas are reported

T

in Table X. The results are remarkably similar to those obtained when using the NNPTT

IP

Islamic factors. To summarise: (a) the U.S. returns-based model still offers the most profits

SC R

(7.84% per annum), followed by CIR-based model (6.04% per annum); (b) CG and IN remain the most profitable sectors; and (c) emerging countries are the most profitable (9.90%

be the least profitable (2.91% per annum).

NU

per annum) followed by Africa (7.29% per annum), and the developed countries continue to

MA

Our final set of robustness tests, which span 40 pages of results, include: (i) testing investor utility and profits using different risk aversion factors; (ii) testing for profits using

D

different levels of transaction costs; and (iii) testing for risk adjusted profits that are inclusive

TE

of (a) different risk aversion factors and (b) different levels of transaction costs. These results, in tabulated form, are available in an online Appendix. The main outcome from these tests is

CE P

that our main results reported in this paper remain unchanged. On the basis of robustness tests

AC

we have undertaken, it is clear that our findings are robust.

V.

Concluding Remarks

In this paper we analyse how well predictors track returns for Islamic stocks. Our paper is innovative because our analysis represents the most comprehension and detailed analysis of Islamic stocks—that is, we consider 2,577 Islamic stocks from around the globe. We categorize these stocks into different regions, including considering all stocks (global), and various sectors. We have a total of 18 categories of regions/sectors, allowing us to construct equal-weighted time-series data on Islamic stocks and a range of macroeconomic and financial predictors over a monthly time frequency from January 1981 to December 2014. 7 8

http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html http://www.econ.yale.edu/~af227/data_library.htm

19

ACCEPTED MANUSCRIPT We undertake a rigorous empirical analysis of asset pricing using our data set. These tests involve in-sample predictability analysis, out-of-sample forecasting evaluation, and

T

estimation of investor utility and profits from a dynamic trading strategy based on a mean-

IP

variance utility function. Our analysis unravels several new insights on not only the pricing of

SC R

Islamic stocks but also regarding the broader literature on stock return predictability. First, we find that not all 12 predictors we use are able to predict Islamic stock returns. The most successful predictor in both in-sample and out-of-sample evaluations is U.S. returns. Second,

NU

utility gains are maximised for a mean-variance investor when U.S. returns are used as a

MA

predictor. Third, profits are maximized when U.S. returns are used as a predictor and the most profitable region is the emerging countries while the most profitable sector is consumer

D

goods. The key implication is that in pricing Islamic stocks the role of the U.S. market cannot

TE

be ignored. In other words, factoring the role of the U.S. market should form part of both future empirical and theoretical work on Islamic stock pricing. Our study shows that not

AC

CE P

doing so will come at a cost.

20

ACCEPTED MANUSCRIPT References Al-Khazali, O., Lean, H.H., and Samet, A., (2014) Do Islamic stock indexes outperform

T

conventional stock indexes? A stochastic dominance approach, Pacific-Basin Finance

IP

Journal, 28, 29-46.

SC R

Ashley, R., Granger, C.W.J., and Schmalensee, R., (1980) Advertising and aggregate consumption: An analysis of causality, Econometrica, 48, 1149-1167. Ashraf, D., and Mohammad, N., (2014) Matching perception with reality—performance of

NU

Islamic equity investments, Pacific-Basin Finance Journal, 28, 175-189.

MA

Balcilar, M., Demirer, R., and Hammoudeh, S., (2015) Global risk exposures and industry diversification with Shariah-compliant equity sectors, Pacific-Basin Finance Journal,

D

35, 499-520.

TE

Bialkowski, J., Etebari, A., and Wisniewski, T.P., (2012) Fast profits: Investor sentiment and stock returns during Ramadan, Journal of Banking and Finance, 36, 835-845.

CE P

Campbell, J., (1987) Stock returns and the term structure, Journal of Financial Economics, 18, 373-399.

AC

Campbell, J., and Shiller, R., (1988a) The dividend-price ratio and expectations of future dividends and discount factors, Review of Financial Studies, 1, 195-228. Campbell, J., and Shiller, R., (1988b) Stock prices, earnings and expected dividends, Journal of Finance, 43, 661-676. Campbell, J., and Thompson, S., (2008) Predicting excess stock returns out of sample: Can anything beat the historical average? The Review of Financial Studies, 24, 1509-1531. Clark, T., and West, K., (2007). Approximately normal tests for equal predictive accuracy in nested models, Journal of Econometrics, 138, 291-311. Fama, E., (1981) Stock returns, real activity, inflation and money, American Economic Review, 71, 545–565.

21

ACCEPTED MANUSCRIPT Fama, E., and French, K., (1988) Dividend yields and expected stock returns, Journal of Financial Economics, 22, 3-25.

T

Fama, E., and French, K., (1989) Business conditions and expected return on stocks and

IP

bonds, Journal of Financial Economics, 25, 23-49.

SC R

Ho, C.S.F., Rahman, N.A.A., Yusuf, N.H.M., and Zamzamin, Z., (2014) Performance of global Islamic versus conventional share indices: International evidence, Pacific-Basin Finance Journal, 28, 110-121.

NU

Hoepner, A., Rammal, H., and Rezec, M., (2011) Islamic mutual funds’ financial

MA

performance and international investment style: Evidence from 20 countries, European Journal of Finance, 17, 829-850.

D

Ibrahim, M.H., (2015) Issues in Islamic banking and finance: Islamic banks, Shari’ah-

TE

complaint investment and sukuk, Pacific-Basin Finance Journal, 34, 185-191. Inoue, A., and Kilian, L., (2004) In-sample or out-of-sample tests of predictability: Which

CE P

one should we use? Econometric Reviews, 23, 371-402. Kothari, S., and Shanken, J., (1997) Book-to-market, dividend yield and expected market

AC

returns: a time series analysis, Journal of Financial Economics, 44, 169-203. Lamont, O., (1998) Earnings and expected returns, Journal of Finance, 53, 1563-1587. Marquering, W., and Verbeek, M., (2004) The economic value of predicting stock index returns and volatility, Journal of Financial and Quantitative Analysis, 39, 407-429. Narayan, P.K., Narayan, S., Phan, D., Thuraisamy, K., and Tran, V., (2015) Credit quality implied momentum profits for Islamic stocks, Pacific-Basin Finance Journal, DOI: 10.1016/j.pacfin.2015.11.004. Pontiff, J., and Schall, L., (1998) Book-to-market ratios as predictors of market returns, Journal of Financial Economics, 49, 141-160.

22

ACCEPTED MANUSCRIPT Rapach, D.E., Strauss, J.K., and Zhou. G., (2013) International stock return predictability: What is the role of the United States? Journal of Finance, 68, 1633-1662.

T

Rapach, D., Strauss, J., and Zhou, G. (2010) Out-of-sample equity premium prediction:

IP

Combination forecast and links to the real economy, Review of Financial Studies, 23,

SC R

821-862.

Rapach, D., Wohar, M., and Rangvid, J. (2005) Macro variables and international stock return predictability, International Journal of Forecasting, 21, 137-166.

NU

Umar, Z., (2015) Islamic vs conventional equities in a strategic asset allocation framework,

MA

Pacific-Basin Finance Journal, http://dx.doi.org/10.1016/j.pacfin.2015.10.006 Welch, I., and Goyal, A., (2008) A comprehensive look at the empirical performance of

D

equity premium prediction, Review of Financial Studies, 21, 1455-1508.

TE

Westerlund, J., and Narayan, P., (2012) Does the choice of estimator matter when forecasting stock returns, Journal of Banking and Finance, 36, 2632-2640.

CE P

Westerlund, J., and Narayan, P., (2015) Testing for predictability in conditionally

AC

heteroskedastic stock returns, Journal of Financial Econometrics, 13, 342-375

23

ACCEPTED MANUSCRIPT Table I: List of variables

T

This table reports the list of individual predictors (Panel A) and the common predictors (Panel B) used to predict and forecast portfolio returns. Each of these variables are downloaded for each stock in our sample, then once the stocks is assigned to a particular region or sector, an equal-weighted average is taken to compute the portfolio-level predictor. In total, we have 10 sectoral portfolios and eight regional portfolios; these are noted in Table II. All data are downloaded from Datastream, except the U.S. BAA and AAA bond yields, which are downloaded from Goyal’s website: http://www.hec.unil.ch/agoyal/.

IP

SC R

NU

MA

D TE CE P

CRF U.S. CIR

Panel A: Individual predictors Short-term rate is proxied by the three-month bill rate. Long-term rate is proxied by the 10-year bond rate. Market rate. Term spread is calculated as the difference between the long-term rate and the short-term rate. Inflation rate is calculated as the log difference of the consumer price index. Industrial production growth rate is calculated as the log difference of industrial production index. Difference in the unemployment rate. M1 growth is calculated as the log difference in M1. Exchange rate returns is calculated as the log difference in the bilateral exchange rate the U.S. dollar. Panel B: Common predictors Credit risk factor is calculated as the difference between the U.S. BAA and AAA bond yields. U.S. excess returns is calculated as the S&P500 returns in excess of the U.S. three-month Treasury bill. Commodity index returns (Commodity Research Bureau BLS Spot Index).

AC

STR LTR MR TS INF IPG DUN M1 ER

24

ACCEPTED MANUSCRIPT Table II: Descriptive statistics of excess returns

MA D TE CE P 25

Panel B: Sectoral portfolios Mean SD AR(1) ARCH(1) -0.14 5.84 0.24 0.00 0.08 4.72 0.24 0.00 0.23 5.04 0.22 0.27 -0.51 6.07 0.30 0.32 0.20 4.54 0.14 0.01 -0.01 4.99 0.23 0.04 -0.29 6.60 0.18 0.00 0.10 7.54 0.19 0.00 -0.04 7.79 0.13 0.01 0.24 6.16 0.08 0.35

SC R

BM CG CS FI HC IN OG TE TEL UT

NU

ARCH(1) 0.01 0.02 0.00 0.00 0.08 0.00 0.00 0.17

AC

Global Developed Emerging Europe Australasia Asia Africa America

Panel A: Regional portfolios Mean SD AR(1) 0.00 4.89 0.23 0.33 4.88 0.19 -0.89 6.33 0.28 -0.04 5.75 0.19 -0.06 7.13 0.14 0.13 5.41 0.25 -0.46 7.12 0.18 0.48 5.42 0.17

IP

T

This table reports selective descriptive statistics for the regional and sectoral portfolios’ excess returns. The 10 sectoral portfolios are: basic material (BM), consumer goods (CG), consumer services (CS), financials (FI), health care (HC), industrial (IN), oil and gas (OG), technology (TE), telecommunication (TEL), and utilities (UT). The eight regional (including global) portfolios are noted in column 1 (Panel A). The statistics include the mean value, standard deviation, first-order autoregressive (AR(1)) coefficient, and the p-value resulting from a Lagrange multiplier test of a zero slope restriction in an autoregressive conditional heteroscedasticity regression of order 1 (ARCH(1)). Panel A has statistics for regional portfolios while panel B has the corresponding statistics for sectoral portfolios.

ACCEPTED MANUSCRIPT Table III: Descriptive statistics of predictors

CRF U.S. CIR

Mean 0.09 0.29 0.11

MA

NU

Panel A: Individual predictors ARCH(1) Developed 0.00 STR 0.00 LTR 0.00 MR 0.00 TS 0.00 INF 0.00 IPG 0.00 DUN 0.01 M1 0.18 ER ARCH(1) Europe 0.00 STR 0.31 LTR 0.00 MR 0.00 TS 0.00 INF 0.00 IPG 0.02 DUN 0.00 M1 0.00 ER ARCH(1) Asia 0.00 STR 0.01 LTR 0.00 MR 0.00 TS 0.58 INF 0.62 IPG 0.22 DUN 0.26 M1 0.16 ER ARCH(1) America 0.02 STR 0.00 LTR 0.06 MR 0.00 TS 0.00 INF 0.07 IPG N/A DUN 0.12 M1 0.48 ER Panel B: Common predictors SD AR(1) 0.04 0.96 4.41 0.07 2.68 0.25

D

AR(1) 0.95 1.00 0.99 0.91 0.72 -0.30 0.44 -0.07 0.40 AR(1) 0.93 1.00 0.98 0.91 0.67 -0.19 0.30 0.07 0.49 AR(1) 0.99 1.00 0.99 0.88 0.91 0.69 0.24 -0.32 0.23 AR(1) 0.96 0.95 1.00 0.85 -0.05 -0.27 N/A -0.19 0.17

TE

SD 0.44 0.22 0.22 0.35 0.71 1.50 0.16 1.64 1.44 SD 0.99 0.26 0.30 0.90 1.53 3.64 0.21 2.44 1.64 SD 0.39 0.32 0.38 0.13 0.33 0.50 0.37 1.94 2.77 SD 0.36 0.28 0.32 0.24 0.96 3.44 N/A 1.99 2.70

CE P

Mean 0.80 0.63 0.73 -0.18 0.71 0.25 0.01 1.26 0.45 Mean 1.36 0.79 0.79 -0.57 1.25 0.29 0.02 1.83 0.89 Mean 0.70 0.71 0.67 0.01 0.35 0.20 0.00 0.70 0.07 Mean 0.92 1.01 0.91 0.09 0.76 0.15 N/A 1.14 0.56

AC

Global STR LTR MR TS INF IPG DUN M1 ER Emerging STR LTR MR TS INF IPG DUN M1 ER Australasia STR LTR MR TS INF IPG DUN M1 ER Africa STR LTR MR TS INF IPG DUN M1 ER

SC R

IP

T

This table reports selective descriptive statistics for the individual and common predictors of regional and sectoral portfolio returns. The nine individual predictors are short-term rate (STR), long-term rate (LTR), market rate (MR), term spread (TS), inflation (INF), industrial production growth (IPG), difference in unemployment rate (DUN), M1 growth (M1), and exchange rate return (ER); while the three common predictors are: a credit risk factor (CRF), U.S. excess returns (U.S.), and commodity index returns (CIR). The statistics include the mean value, standard deviation, AR(1), and ARCH(1). AR(1) refers to the autoregressive coefficient of order 1, while ARCH (1) refers to a Lagrange multiplier test of a zero slope restriction in an ARCH regression of order 1—the p-value from this test is reported. Finally, N/A implies that DUN data is unavailable for Africa.

26

Mean 0.52 0.57 0.70 0.05 0.29 0.20 0.01 0.72 0.07 Mean 0.88 0.60 0.59 -0.28 0.67 0.18 0.01 0.86 0.40 Mean 0.58 0.55 0.83 -0.03 0.35 0.46 0.00 1.00 0.17 Mean 0.52 0.61 0.63 0.09 1.51 0.15 -0.01 2.44 1.64

SD 0.33 0.25 0.22 0.11 0.28 1.34 0.15 1.23 1.71 SD 0.60 0.22 0.34 0.54 0.86 1.57 0.21 1.69 2.10 SD 0.25 0.18 0.25 0.21 0.33 3.33 0.10 1.60 0.97 SD 0.33 0.24 0.29 0.16 2.20 2.38 0.28 3.98 3.68

AR(1) 0.99 1.00 0.98 0.96 0.66 -0.34 0.39 -0.15 0.35 AR(1) 0.93 1.00 1.00 0.90 0.53 -0.29 0.49 -0.17 0.37 AR(1) 0.94 0.99 0.87 0.92 0.38 -0.44 -0.01 -0.07 0.32 AR(1) 0.97 0.99 0.96 0.92 0.72 -0.19 -0.12 0.16 0.41

ARCH(1) 0.00 0.13 0.00 0.00 0.37 0.00 0.14 0.00 0.44 ARCH(1) 0.00 0.78 0.00 0.00 0.00 0.00 0.00 0.00 0.57 ARCH(1) 0.00 0.01 0.00 0.00 0.00 0.00 0.00 0.44 0.15 ARCH(1) 0.00 0.82 0.34 0.00 0.00 0.00 0.74 0.00 0.00

ARCH(1) 0.00 0.01 0.04

ACCEPTED MANUSCRIPT Table IV: Testing for endogeneity This table reports the results for endogeneity test based on a regression of residuals from the predictive regression model on residuals from the first-order autoregressive predictor regression model. The regression has the following form: 𝜖𝑡 = 𝛾𝜀𝑡 + 𝜂𝑡

STR 0.08 0.07 0.07 0.03 0.12 0.03 0.07 0.00 0.11 0.01

LTR 0.00 0.00 0.00 0.01 0.00 0.00 0.05 0.02 0.00 0.04

MR 0.36 0.34 0.22 0.04 0.62 0.55 0.39 0.63 0.79 0.01

NU

BM CG CS FI HC IN OG TE TEL UT

Panel A: Regional portfolios TS INF IPG DUN 0.05 0.58 0.79 0.69 0.53 0.68 0.61 0.63 0.03 0.49 0.42 0.81 0.27 0.88 0.69 0.67 0.02 0.42 0.70 0.63 0.01 0.44 0.37 0.31 0.34 0.80 0.58 N/A 0.73 0.72 0.70 0.92 Panel B: Sectoral portfolios TS INF IPG DUN 0.13 0.33 0.89 0.55 0.12 0.74 0.86 0.83 0.11 0.69 0.56 0.31 0.04 0.36 0.39 0.60 0.17 0.42 0.60 0.18 0.06 0.80 0.69 0.89 0.14 0.79 0.16 0.55 0.01 0.46 0.84 0.15 0.16 0.27 0.90 0.09 0.01 0.21 0.48 0.74

MA

MR 0.43 0.83 0.12 0.87 0.06 0.86 0.01 0.33

D

LTR 0.00 0.05 0.00 0.00 0.00 0.07 0.24 0.00

AC

CE P

TE

Global Developed Emerging Europe Australasia Asia Africa America

STR 0.02 0.18 0.01 0.18 0.00 0.00 0.04 0.35

SC R

IP

T

where 𝜖𝑡 is the residual from the predictive regression model 𝑟𝑡 = 𝛼 + 𝛽𝑃𝑡−1 + 𝜖𝑡 and 𝜀𝑡 is the residual from the AR(1) regression of the predictor 𝑃𝑡 = 𝜌𝑃𝑡−1 + 𝜀𝑡 . We report the p-value of the test that examines the null that 𝛾 = 0, which if rejected suggests that endogeneity exists in our proposed predictive regression model. The results for the eight regional portfolios are reported in Panel A, while Panel B reports the results for the 10 sectoral portfolios. Finally, N/A implies that DUN data is unavailable for Africa.

27

M1 0.67 1.00 0.54 0.65 0.53 0.05 0.72 0.79

ER 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

CRF 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

U.S. 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

CIR 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

M1 0.81 0.47 0.71 0.90 0.81 0.36 0.91 0.88 0.78 0.31

ER 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

CRF 0.00 0.00 0.01 0.01 0.00 0.00 0.00 0.00 0.00 0.02

U.S. 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

CIR 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

ACCEPTED MANUSCRIPT Table V: In-sample predictability of excess returns

Asia

Africa Americ a

BM

CG

CS

FI

HC

IN

OG

TE

STR 0.17* ** 0.20* ** 0.14* ** 0.18* ** 0.19* ** 0.17* ** 0.12* *

0.10* * 0.11* *

-0.10*

-0.08

0.05

0.10* *

-0.10*

0.10*

0.0 1

0.22* **

0.02

0.0 8

0.13* *

0.05

0.14* ** 0.12* *

0.05

0.05

0.01

0.06

0.06

0.00

NU

0.04

0.0 9

0.05

0.08

0.0 2

-0.03

0.02

0.10 *

0.1 1

-0.05

0.20* **

0.01

0.0 0

N/A

-0.03

0.0 0.00 -0.05 2 Panel B: Sectoral portfolios TS INF IPG DUN M1 0.07

MR 0.11* * 0.10* *

-0.03

-0.03

0.16* **

0.06

0.0 8

0.05

-0.08

0.11* *

0.12* *

0.16* **

0.03

0.0 7

0.06

0.07

-0.07

-0.09*

0.20* **

0.10 *

0.0 2

0.10*

0.06

0.10* * 0.14* **

0.13* * 0.14* **

0.16* **

0.08

0.1 0

0.04

0.08

0.06

0.04

0.1 5*

-0.06

-0.02

-0.04

-0.07

0.06

0.03

0.0 0

0.15* **

0.10

-0.07

LTR

0.16* **

0.08

-0.09*

-0.08

0.16* **

0.07

0.0 4

0.08

0.05

0.21* **

0.08

0.0 4

0.07

0.04

28

ER 0.40* ** 0.26* ** 0.35* ** 0.34* ** 0.57* ** 0.49* ** 0.35* ** 0.20* **

CRF

U.S.

CIR

0.07

0.89* **

0.38* **

0.01

0.92* **

0.35* **

0.09 *

0.61* **

0.41* **

0.01

0.81* **

0.40* **

0.03

0.62* **

0.36* **

0.10 **

0.63* **

0.32* **

0.01

0.50* **

0.32* **

0.00

0.97* **

0.33* **

ER 0.46* ** 0.42* ** 0.29* ** 0.37* ** 0.36* ** 0.40* ** 0.33* ** 0.23* **

CRF

U.S.

CIR

0.08

0.72* **

0.48* **

0.11 **

0.81* **

0.36* **

0.12 **

0.91* **

0.27* **

0.02

0.67* **

0.38* **

0.08 *

0.76* **

0.23* **

0.03

0.89* **

0.40* **

0.06

0.66* **

0.43* **

0.05

0.84* **

0.24* **

IP

0.11* * 0.12* *

0.0 6

SC R

-0.09*

0.07

MA

Austral asia

-0.07

-0.08*

0.17* **

D

Europe

0.14* **

-0.09*

TE

Emergi ng

Panel A: Regional portfolios TS INF IPG DUN M1

MR 0.11* * 0.10* * 0.22* **

CE P

Develo ped

LTR

AC

Global

STR 0.17* ** 0.12* * 0.23* ** 0.14* ** 0.13* ** 0.19* ** 0.11* * 0.12* *

T

This table reports results on excess return predictability for eight regional portfolios (in Panel A) and 10 sectoral portfolios (in Panel B) using a set of nine individual predictors and three common predictors. The predictive regression model is the bias-adjusted FGLS estimator proposed by Westerlund and Narayan (2015).The coefficient of the predictors is reported, and *, **, and *** denote significance at the 10%, 5% and 1% levels, respectively. Finally, N/A implies that DUN data is unavailable for Africa.

ACCEPTED MANUSCRIPT -0.07

-0.03

-0.05

0.08

0.02

0.0 6

0.07

0.06

UT

0.11* *

-0.02

-0.02

0.13* *

0.02

0.0 7

0.10

0.10

0.19* ** 0.27* **

0.03

0.67* **

0.19* **

0.07

0.51* **

0.25* **

SC R

IP

T

TEL

NU

Table VI: Out-of-sample evaluations of forecasting of excess returns

TE

D

MA

This table reports out-of-sample evaluations of the importance of predictors in forecasting excess returns vis-àvis the constant returns model. A 50% in-sample period is used to generate recursive forecasts of excess returns for the remaining 50% of the sample. Besides the forecasts from individual and common predictors, we also apply three forecasting combination approaches as proposed by Rapach et al. (2010), namely, mean, median, and trimmed median (T-mean). We report the out-of-sample (𝑂𝑅2 ), which examines the difference in the mean squared errors from the competition model relative to the constant returns model. The Clark and West (2007) MSFE-adjusted test statistic, denoted with an asterisk, examines the null hypothesis that the 𝑂𝑅2 = 0 against the alternative that the 𝑂𝑅2 > 0; *,**, and *** denote rejection of the null hypothesis at the 10%, 5%, and 1% levels of significance, respectively. The results for forecasting excess returns of eight regional portfolios are reported in Panel A, while Panel B reports the results for 10 sectoral portfolios. Finally, N/A implies that DUN data is unavailable for Africa. Panel A: Regional portfolios

Asia Afric a Amer icas

1.22 ***

0.90 ** 0.26 4.63 *** 0.61 * 0.95 ** 0.45 ** 0.90 * 0.34

0.25 4.95 *** 0.96 ** 1.02 ** 0.68 ** 1.04 * 0.03

ST R

LT R

MR

TS

INF

0.95 *

4.67 ***

0.75 * 0.07 5.64 ***

1.14 ** 0.31 ** 3.91 ***

0.66 * 0.24 4.27 ***

0.66

0.33

0.48

1.11 **

1.01 * 0.12 1.26 * 0.19

1.01 **

0.92 * 0.26 4.59 *** 0.66 * 0.86 **

0.56 0.83 * 0.03

CE P

Me dian

AC

Globa l Devel oped Emer ging Europ e Austr alasia

Me an

TMea n 0.97 ** 0.11 4.86 *** 0.66 * 0.93 ** 0.58 ** 0.98 * 0.21

0.10

0.55 0.71 * 0.05

DU N

M1

ER

CR F

U.S .

0.37 3.88 ***

0.61 * 0.28 4.15 ***

0.73 ** 0.12 4.52 ***

1.15 ** 0.42 5.09 ***

0.13 0.62 3.69 ***

0.41

0.36

0.45

0.49

0.44

0.33

0.96 **

0.78 **

0.84 **

0.23

0.31

0.15

0.21

0.79 * 0.49

0.82 * 0.38

0.72 * 0.69

1.03 ** 0.87 *** 0.75 * 0.52

0.79 * 0.65 0.73 * 0.66

3.60 ** 2.70 ** 6.42 *** 3.62 *** 1.85 ** 2.53 ** 1.37 ** 2.11 *

CR F

U.S .

IPG 0.48

N/A 0.48

0.35 0.21 1.22 ** 0.28

CIR 0.97 * 0.28 5.57 *** 1.47 ** 0.82 * 0.25 0.87 * 0.21

Panel B: Sectoral portfolios

BM CG

Me an

Me dian

1.19 ** 1.14 ***

1.05 ** 0.73 ** 0.04 2.13

CS

0.45

FI

2.40

TMea n 1.13 ** 0.95 **

ST R

LT R

MR

TS

INF

IPG

DU N

M1

ER

0.98 * 0.92 *

0.91 * 0.70 ** 0.60 1.95

1.18 * 1.09 ** 1.27 2.33

1.01 ** 1.02 ** 0.36 ** 2.03

0.81 * 0.57 *

0.59 *

0.77 *

0.86 **

0.22

0.46

0.45

1.49 ** 1.13 **

0.22 1.67

0.05 1.86

0.20

0.46

2.28

1.97

29

0.02 1.90

0.01

0.47

1.81

2.91

0.19 1.02 1.75 1.67

2.20 ** 3.56 ** 3.50 ** 4.27

CIR 1.58 ** 0.71 0.08 3.05

ACCEPTED MANUSCRIPT

TE TEL

0.19

0.13

0.16

** 0.15 0.80 * 0.83 *

** 0.10 1.19 * 1.17 *

** 0.89 ** 0.93 ** 1.19 ** 0.31 *

** 0.51 * 0.68 * 1.21 ** 0.29 *

** 0.09 0.76 * 1.17 ** 0.20

**

** 0.37 * 0.72 ** 1.14 ** 0.78 ***

0.13 0.63 * 1.17 ** 0.34 *

0.14

0.20

0.38

0.30 *

0.75

0.50

0.48

0.66 *

0.51

0.27

0.16

0.01

0.35 *

0.05

0.00

0.16

*** 0.34 1.18 ** 1.45 **

** 1.58 0.17 1.22 *

0.28

0.09

0.52

0.48

0.85

0.04

0.36

0.33

*** 1.84 * 4.69 *** 1.34 ** 1.98 ** 1.38 * 0.48

*** 0.26 1.42 ** 1.61 ** 0.15 0.14 0.12

NU

UT

0.51

** 1.21 ** 1.01 * 1.48 ** 0.30 **

T

OG

** 0.45 ** 1.08 ** 1.34 ** 0.35 ** 0.60 *

IP

IN

** 0.36 * 0.97 ** 1.34 ** 0.34 *

SC R

HC

** 0.54 ** 1.46 *** 1.33 ** 0.49 ** 0.70 **

Table VII: Mean-variance investor utility gains

AC

CE P

TE

D

MA

This table reports utility gains for a mean-variance investor based on our proposed forecasting model vis-à-vis the constant returns model. A 50% in-sample period is used to generate recursive forecasts of excess returns for the remaining 50% of the sample. Utilities are computed based on first estimating the portfolio weight, which is an increasing function of return forecasts and a decreasing function of the return variance and the risk aversion factor. The risk aversion factor equals six and the variance is computed using a 60-month rolling window of returns. The portfolio weight is restricted to be between 0 and 1, implying that there is no short-selling and no borrowing (Panel A) while the portfolio weight is also restricted to be between -0.5 and 1.5, which allows for 50% short-selling and borrowing (Panel B). Our portfolio is a two-asset portfolio, where one is a risky asset and the other is a risk-free asset. The portfolio weight is used to decide the proportion of a dollar to be invested in the risky asset (which is the stock market) and the risk-free asset (which is the three-month bill rate). As is common in this mean-variance setup, we make two assumptions: (a) investors only use public information to forecast the one-period ahead excess returns; and (b) investors rebalance their portfolio once a month. The utility gains are the difference between the utility from a trading strategy based on our proposed model and the constant returns model. The reported utility gains are annualized. The average utility gains based on the predictors are reported in the last column, while the average utility gains of regional and sectoral portfolios are reported in the two last rows of each panel. Finally, N/A implies that DUN data is unavailable for Africa.

Global Develope d Emerging

Europe Australasi a Asia

Panel A: No short-selling and no borrowing 𝜋(0,1) TD ST LT M IN IP M mea TS U R R R F G 1 n N 0.1 0.1 0.4 0.2 0.0 0.0 0.0 0.0 0.24 2 5 1 7 6 3 1 8 0.0 0.1 0.3 0.2 0.2 0.4 0.2 0.1 0.09 9 6 4 3 7 3 7 1 0.0 0.0 0.1 0.0 0.0 0.0 0.0 0.0 0.03 6 1 6 0 0 5 0 6 0.1 0.2 0.0 0.1 0.0 0.0 0.0 0.0 0.15 4 7 2 5 5 8 0 1

Me an

Med ian

0.3 6

0.18

0.3 9

0.29

0.0 5

0.01

0.3 0

0.13

0.1 3

0.04

0.04

0.2 0

0.3 9

0.18

0.34

0.2 4

0.0 9 0.5 5

0.1 2 0.1 0

0.0 0 0.0 1

30

0.0 5

0.0 0

0.0 6

0.0 1

0.0 0 0.0 3

ER

CR F

U. S.

CI R

Aver age

0.5 3 0.3 5

0.0 0 0.3 5

2. 59

0.7 4

0.41

3. 09

0.0 6

0.12

0.1 5

0.0 0

0. 95

0.4 6

0.12

0.9 7

0.25

0. 90

0.0 5

0.09

0.5 7

0.0 2

0.0 4 0.1 4 0.0 2

2. 02

0.1 6

0.0 7 0.3 6

1. 82

0.2 9

0.21

ACCEPTED MANUSCRIPT

HC IN OG

TE TEL UT Regions average Sectors average

0.4 0

0.3 6

0.18

0.24

0.2 2

0.0 5

0.12

0.03

0.1 0

0.02

0.05

0.03

0.03

0.0 9 0.5 1 0.1 9 0.2 8 0.1 6 0.0 5 0.2 2 0.2 1

Me an Global Develope d Emerging Europe Australasi a Asia

0.1 4

0.5 2

0.1 6

0.0 0

0.6 5 0.1 2

0.5 3

0.3 2

0.2 0 0.1 5

0.1 8

0.1 0 0.1 1

0.1 1

0.0 2

0.0 6

0.0 0

0.0 0

0.8 3 0.1 9

0.0 0 0.0 2 0.1 7

0.0 0 0.3 6 0.1 4 0.0 2

0.0 0 0.2 3 0.0 0

0.1 3 0.0 1 0.1 1 0.0 6 0.0 4 0.0 2

1.7 4

0.0 0.4 0.5 0.1 4 3 0 1 0.1 0.4 0.2 0.0 0.28 0.34 3 6 1 5 0.1 1.0 0.5 0.0 0.0 0.0 0.0 0.12 0.14 2 1 5 1 4 0 0 0.1 0.0 0.0 0.1 0.1 0.0 0.1 0.5 0.22 0.20 9 3 3 7 4 9 8 4 0.0 0.0 0.1 0.1 0.1 0.1 0.1 0.1 0.13 0.13 7 1 8 3 3 5 4 3 0.1 0.0 0.0 0.1 0.0 0.0 0.0 0.0 0.01 0.03 6 1 1 6 1 6 0 2 0.0 0.1 0.0 0.0 0.0 0.1 0.1 0.0 0.00 0.07 9 3 9 3 6 5 0 4 0.2 0.0 0.1 0.2 0.0 0.0 0.0 0.0 0.10 0.14 6 7 7 0 4 5 1 6 Panel B: Short-selling and borrowing 𝜋(−0.5,1.5) TD Med ST LT M IN IP M mea TS U ian R R R F G 1 n N

0.4 3 3.8 8 1.4 5 1.0 8 0.7 1

1.1 7

1.31

1.42

0.28

0.08

3.59

3.80

0.94

1.02

1.01

0.98

0.0 9 3.6 4 0.9 5 1.1 6

0.47

0.61

0.5 3

1.0 3 0.1 6 4.2 9 0.4 4 1.0 5 0.2 6

1.5 2

0.0 3 0.2 5

T

0.23

0.0 0 0.4 1

0.6 1

IP

0.15

N/ A 0.4 3

SC R

0.2 8

0.0 4

0.0 5 0.6 3 0.1 3 0.1 8 0.1 7

NU

FI

0.28

0.0 0 0.3 3

MA

CS

0.1 5

0.0 0 0.4 2

D

CG

0.1 2 0.2 0

0.0 7

0.15

0.1 1 0.0 2

TE

BM

0.01

0.03

CE P

Americas

0.0 1

AC

Africa

0.7 3 0.2 4 0.2 0 0.3 2 0.4 3 0.1 4 0.1 6 0.1 5 0.2 2 0.0 3 0.3 2

ER

0.3 4 3.0 4 0.7 0 1.0 6

1.3 4 0.2 3 3.5 4 1.0 1 0.9 1

0.9 9 0.2 7 3.2 5 0.6 6 1.0 2

0.7 5 0.4 5 2.9 0 0.5 5 0.8 3

0.9 6 0.2 7 3.1 6 0.7 4 0.8 9

1.1 2 0.0 1 3.4 9 0.8 0 1.1 2

1.3 8 0.4 7 3.9 9 0.6 9 0.2 8

0.3 9

0.2 4

0.3 2

0.1 5

0.2 2

0.9 4

0.2 9

31

0.0 0 0.5 0

0. 42

0.0 2

0.01

2. 05

0.1 1

-0.04

0.0 0 0.0 4 0.2 1

1. 86

0.9 4

0.39

2. 65

0.7 2

0.46

2. 27

0.2 7

0.19

0.0 0 0.2 8 0.0 0 0.0 9

0. 81

0.4 5 0.4 1 0.9 7

0.1 7 0.1 6 0.0 1 0.1 2 0.0 3

1. 62 0. 57

CR F 0.1 1 0.7 1 2.6 9 0.5 7 0.8 1 0.6 6

2. 31 3. 06 0. 44

0.13

0.27 0.47

0. 49

0.4 8 0.0 3 0.1 3 0.0 1

1. 73

0.3 4

0.15

1. 61

0.3 5

0.24

U. S.

CI R

Aver age

3. 75 2. 55 5. 16 3. 32 1. 36

1.1 4 0.6 1 4.1 3 1.5 9 0.7 9

2. 43

0.2 2

-0.04

0.27 0.16

0.08

1.25

-0.02 3.61 1.00 0.94

0.40

ACCEPTED MANUSCRIPT

HC IN OG TE TEL UT Regions average Sectors average

0.3 7 2.1 2 0.6 1 2.1 2 2.2 8 0.6 1 0.3 7 0.0 8 1.2 8 1.1 2

0.85

1.01

0.8 7

0.96

1.25

0.9 7

0.06

0.16

1.89

2.02

0.40

0.50

1.46

1.62

2.30

2.29

0.44

0.45

0.28

0.32

0.04

0.06

0.93

1.05

0.86

0.97

0.4 3 1.7 9 1.2 9 1.3 2 2.5 1 0.3 9 0.2 1

0.7 7

1.3 9

0.7 7

0.4 3

0.1 3

0.4 4

0.6 3

1.6 7

0.8 3 0.5 9 1.7 3 0.1 8 1.1 1 1.2 3 0.1 5 0.1 6

1.3 2 1.1 0 2.0 5

1.2 9

0.7 5

0.3 0 1.7 9

0.0 0 1.6 8

0.6 4 0.0 7 1.6 4

0.9 8 1.4 0 2.1 1 0.4 0 0.2 7 0.2 1 0.8 7

0.5 6 1.0 5 2.1 3 0.3 8 0.2 6 0.0 4 0.7 9

0.3 1 0.2 2 1.4 8 0.1 0 1.1 4 2.0 6 0.2 6 0.3 8 0.0 5 0.5 7

0.9 5

0.7 2

0.5 4

0.2 0 1.0 1

0.0 5 0.9 0

0.0 6 1.5 7 1.6 8 0.2 5 0.3 9 0.0 2 0.9 7

1.0 0

0.5 2

0.7 6

32

0.4 4

T

1.4 8

0.5 6 0.4 6

IP

1.1 1

N A 0.4 3

0.9 6

0.0 4

0.5 5 0.6 9

0.6 3 0.0 1 1.5 9

1.2 2

0.2 9 1.4 7 2.4 2 0.3 3 0.2 6

0.0 5 0.7 5

0.4 3 1.1 5 2.0 4 0.9 9 0.2 8 0.0 1 0.9 4

0.6 7

0.7 7

1.0 8

SC R

0.28

0.6 3 0.3 3

NU

FI

0.1 6

0.6 1 0.4 3

MA

CS

1.0 0 0.2 0

0.6 4

0.12

0.5 5 0.0 2

D

CG

0.77

TE

BM

0.70

CE P

Americas

0.8 2

AC

Africa

0.1 6 1.0 1 2.0 8 0.4 4 0.2 7

0.3 9 2.5 2

0.2 2 0.9 4

0.5 5 0.6 3 0.0 1 1.3 0 1.5 0 1.4 8 1.7 6 0.3 5 2.1 1 0.0 8 0.4 6 0.2 9 0.3 1 0.0 4

1. 27 1. 66

0.7 1 0.3 8

-0.12

3. 88

1.4 7

1.03

3. 48

0. 33 2. 69

0.7 3 0.0 7 2.4 6 0.3 9 1.6 6 2.4 8 0.1 9 0.1 0 0.1 2 0.9 5

2. 51

0.8 5

2. 54 3. 35 2. 20 4. 52 2. 18 1. 75 0. 92

0.73

0.91

0.01 1.96

0.30 1.48 2.08 0.47 0.33

0.04 0.97

0.86

ACCEPTED MANUSCRIPT Table VIII: Mean-variance investor profits

Europe Australasi a Asia Africa Americas BM CG CS FI HC IN OG TE TEL UT Regions average Sectors average

9.73 5.86 5.16 5.91

NU

MA

2.97

D

Emerging

5.86

TE

Develope d

6.0 5 3.6 6 9.7 7 6.0 4 5.2 5 6.1 3 7.9 6 4.3 9 5.9 6 6.0 9 5.9 5 5.7 8 6.0 6 6.2 1 5.8 8 5.8 0 5.6 9 5.7 5 6.1 6 5.9 2

7.97 3.95 5.83

AC

Global

Med ian

Panel A: No short-selling and no borrowing 𝜋(0,1) TD ST LT M IN IP mea TS U M1 R R R F G n N 6.1 5.9 6. 5.9 5. 5. 5.6 5.7 5.92 7 0 27 6 74 65 8 6 3.5 3.4 3. 3.0 3. 2. 2.9 3.1 3.18 8 1 87 0 00 83 6 5 9.7 9.7 9. 9.7 9. 9. 9.7 9.7 9.75 8 4 78 2 72 67 2 8 5.9 5.5 5. 5.8 5. 5. 5.7 5.7 5.88 5 3 74 8 68 66 3 4 5.3 5.2 5. 5.1 5. 5. 5.1 5.2 5.15 6 8 25 1 17 11 1 7 6.4 5.8 6. 5.7 5. 5. 5.6 6.3 6.08 0 7 25 1 79 73 9 2 7.8 7.9 7. 7.9 7. 8. N 7.9 7.94 9 0 95 6 96 02 A 6 4.4 4.3 4. 3.7 3. 3. 3.7 3.8 4.09 3 1 38 9 91 59 8 2 6.0 5.8 6. 5.8 5. 5. 5.6 5.8 5.91 8 2 20 4 68 55 8 1 6.5 5.9 6. 6.2 5. 5. 5.6 5.7 5.96 9 4 45 9 81 56 9 0 6.4 5.7 5. 6.0 5. 5. 5.7 5.7 5.86 3 4 78 9 79 73 2 8 5.8 5.7 5. 5.6 5. 5. 5.6 5.6 5.73 2 2 78 8 68 68 8 2 7.4 6.0 6. 6.5 6. 5. 5.7 5.9 6.00 1 1 44 1 09 61 3 4 6.2 5.9 6. 5.9 5. 5. 5.6 5.7 6.03 1 5 52 0 73 82 8 1 5.8 5.2 5. 5.6 5. 5. 5.6 5.6 5.83 7 3 96 7 64 66 8 8 5.7 5.6 5. 5.6 5. 5. 5.6 6.0 5.72 1 2 63 8 66 61 9 8 5.6 5.5 5. 5.6 5. 5. 5.6 5.6 5.66 3 5 70 5 65 68 6 5 5.8 5.6 5. 5.8 5. 5. 5.7 5.6 5.73 9 8 70 6 68 63 0 8 6.2 5.9 6. 5.8 5. 5. 5.5 5.9 6.00 0 9 19 9 87 78 2 7 6.1 5.7 6. 5.9 5. 5. 5.6 5.7 5.84 7 3 02 2 74 65 9 6

CE P

Me an

SC R

IP

T

This table reports profits for a mean-variance investor based on our proposed forecasting model. A 50% insample period is used to generate recursive forecasts of excess returns for the remaining 50% of the sample. Profits are computed based on first estimating the portfolio weight, which is an increasing function of return forecasts and a decreasing function of the return variance and the risk aversion factor. The risk aversion factor equals six and the variance is computed using a 60-month rolling window of returns. The portfolio weight is restricted to be between 0 and 1, implying that there is no short-selling and no borrowing (Panel A) while the portfolio weight is also restricted to be between -0.5 and 1.5, which allows for 50% short-selling and borrowing (Panel B). Our portfolio is a two-asset portfolio, where one is a risky asset and the other is a risk-free asset. The portfolio weight is used to decide the proportion of a dollar to be invested in the risky asset (which is the stock market) and the risk-free asset (which is the three-month bill rate). The reported profits are annualized. The average profits based on the predictors are reported in the last column, while the average profits of regional and sectoral portfolios are reported in the two last rows of each panel. Finally, N/A implies that DUN data is unavailable for Africa.

5.90 5.77 5.70 5.93 5.96 5.81 5.73 5.66 5.71 5.92 5.80

33

ER 6.2 6 2.8 9 9.8 8 5.8 1 4.8 4 5.7 7 7.9 3 4.4 9 6.3 0 6.5 2 6.1 6 5.8 9 6.3 2 6.1 7 5.8 3 5.6 8 5.6 7 5.9 1 5.9 8 6.0 4

C R F 5. 68 2. 88 9. 72 5. 77 4. 99 5. 70 7. 96 3. 70 5. 68 5. 68 5. 68 5. 68 5. 68 5. 68 5. 62 5. 68 5. 68 5. 68 5. 80 5. 68

U. S.

CI R

9.5 4 7.7 2 10. 96 9.0 2 6.2 4 8.3 1 8.5 6 7.0 4 7.5 4 9.7 4 8.9 9 6.7 1 8.9 0 10. 24 6.2 3 7.5 5 6.2 7 6.2 2 8.4 2 7.8 4

6.4 7 3.4 3 10. 21 6.7 9 5.2 1 6.0 3 7.9 9 4.4 3 6.6 2 6.4 9 6.1 9 6.1 6 5.6 6 6.7 3 6.1 9 5.5 4 5.6 5 5.6 9 6.3 2 6.0 9

Aver age 6.26 3.56 9.89 6.11 5.24 6.13 8.01 4.31 6.07 6.37 6.17 5.84 6.36 6.36 5.77 5.84 5.70 5.78 6.16 6.03

ACCEPTED MANUSCRIPT

Africa Americas BM CG CS FI HC IN OG TE TEL UT Regions average Sectors average

5.16 5.91 8.02 3.95 6.06 5.95 5.72 5.96 5.91 6.04 5.79 5.70 5.65 5.69 6.02 5.85

ER

34

C R F 5. 15 2. 53 9. 93 5. 61 4. 99 5. 12 7. 90 3. 58 5. 20 4. 70 4. 80 5. 66 4. 25 5. 27 5. 59 5. 39 5. 89 5. 42 5. 60 5. 22

T

6.4 0 2.7 8 10. 79 5.7 4 4.5 7 5.8 2 8.3 6 4.7 1 6.8 8 6.6 1 6.3 0 6.7 8 6.0 1 6.4 4 5.9 9 5.6 0 5.6 3 5.8 8 6.1 4 6.2 1

IP

SC R

Asia

5.90

NU

Australasi a

MA

Europe

2.97 10.2 6

D

Emerging

5.98

TE

Develope d

6.4 9 3.7 2 10. 60 6.4 8 5.2 4 6.1 7 8.1 6 4.4 1 6.3 3 6.5 8 6.2 1 6.2 1 6.1 5 6.8 1 5.7 5 5.8 8 5.7 5 5.7 4 6.4 1 6.1 4

AC

Global

Med ian

CE P

Me an

Panel B: Short-selling and borrowing 𝜔(−0.5,1.5) TD ST LT M IN IP mea TS U M1 R R R F G n N 6.1 5.8 6. 5.9 5. 5. 5.6 5.8 6.11 7 5 56 9 63 42 3 2 3.5 3.4 3. 3.0 3. 2. 2.9 3.2 3.19 8 1 87 0 00 81 6 5 10.5 10. 11. 9. 10. 9. 9. 9.8 10. 0 25 87 78 19 97 61 3 20 5.9 5.4 5. 5.9 5. 5. 5.7 5.7 5.98 7 8 72 7 62 53 1 7 5.3 5.2 5. 5.0 5. 4. 5.0 5.2 5.14 6 8 23 7 18 99 5 9 6.4 5.8 6. 5.6 5. 5. 5.6 6.4 6.06 0 7 25 8 76 59 5 2 7.8 8.5 8. 7.9 7. 7. N 7.8 8.09 9 0 00 2 95 91 A 9 4.4 4.3 4. 3.7 3. 3. 3.7 3.7 4.12 3 1 38 8 91 54 8 8 6.0 5.9 6. 5.9 5. 5. 5.6 5.8 6.23 8 8 60 9 64 34 5 4 6.5 6.2 6. 6.3 5. 5. 5.6 5.6 6.29 9 1 93 0 73 36 4 3 6.4 5.4 5. 6.1 5. 5. 5.7 5.7 5.96 3 5 44 0 79 56 1 8 5.8 5.8 6. 5.8 5. 5. 5.7 5.7 6.09 4 0 21 4 76 56 3 0 7.4 5.6 6. 6.5 6. 5. 5.6 5.9 6.03 1 3 07 4 08 42 6 8 6.2 5.8 6. 5.9 5. 5. 5.6 5.7 6.22 1 4 54 7 62 75 0 7 6.0 5.2 5. 5.6 5. 5. 5.5 5.5 5.76 9 3 96 1 64 58 8 4 5.6 5.5 5. 5.6 5. 5. 5.6 6.3 5.71 5 4 59 5 64 51 9 1 5.6 5.5 5. 5.6 5. 5. 5.6 5.6 5.69 1 6 81 5 63 76 5 5 5.8 5.7 5. 5.8 5. 5. 5.7 5.6 5.71 9 7 80 7 63 60 1 5 6.2 6.3 6. 5.9 5. 5. 5.5 6.0 6.15 6 2 22 5 88 68 1 5 6.1 5.7 6. 5.9 5. 5. 5.6 5.7 5.97 8 0 10 5 72 54 6 8

U. S.

CI R

10. 97 8.5 7 13. 00 10. 97 6.7 0 9.4 5 9.5 6 7.7 7 9.0 9 11. 14 9.9 9 8.3 2 9.1 5 12. 20 6.1 5 8.2 7 6.7 8 6.0 8 9.6 2 8.7 2

6.0 1 2.9 1 11. 24 6.9 9 5.0 5 5.7 0 8.0 6 4.1 1 6.6 8 5.8 9 5.8 1 6.7 0 5.3 2 6.5 7 6.1 5 5.5 2 5.4 8 5.5 5 6.2 6 5.9 7

Aver age 6.30 3.56 10.5 6 6.26 5.23 6.14 8.18 4.34 6.25 6.39 6.10 6.16 6.13 6.48 5.76 5.86 5.76 5.74 6.32 6.06

ACCEPTED MANUSCRIPT Table IX: Results from regression model using global Islamic risk factors This table reports results from regressing the portfolio profits (generated using a risk aversion factor of six) on global Islamic risk factors. The risk factors are market excess returns (MKT), small-minus-big (SMB) stock returns, high-minus-low (HML) book-to-market returns, and betting-against-beta (BAB) factor. The regression model has the following form:

T

Profit 𝑡 = 𝛼 + 𝛽𝑀𝐾𝑇 [𝑅𝑚,𝑡 − 𝑟𝑓,𝑡 ] + 𝛽𝑆𝑀𝐵 𝑆𝑀𝐵𝑡 + 𝛽𝐻𝑀𝐿 𝐻𝑀𝐿𝑡 + 𝛽𝐵𝐴𝐵 𝐵𝐴𝐵𝑡 + 𝜀𝑡

SC R

IP

The regression is run using OLS where standard errors are corrected for heteroskedasticity and autocorrelation based on Newey-West procedure. A maximum of 12 lags are incorporated and the optimal lag length is chosen using the Schwarz Information criterion. The annualised alpha coefficient is reported. Panel A is based on a profit dependent variable that is free of short-selling and borrowing whereas Panel B has profits (dependent variable) that accounts for 50% short-selling and borrowing. The average alphas based on the predictors are reported in the last column, while the average alphas of regional and sectoral portfolios are reported in the last two rows of each panel. Finally, N/A implies that DUN data is unavailable for Africa. All alphas are statistically different from zero at the 1% level or better.

Europe Australasi a Asia Africa Americas BM CG CS FI HC IN OG TE TEL UT Regions average

9.73 5.76 5.03

NU

MA

2.59

D

Emerging

5.74

TE

Develope d

5.8 6 3.2 4 9.7 4 5.8 3 5.0 9 5.8 8 8.0 3 4.1 9 5.7 9 5.9 2 5.8 4 5.7 6 5.8 6 5.9 5 5.7 1 5.9 4 5.6 8 5.6 9 5.9 8

5.88 8.09 3.76

AC

Global

Med ian

CE P

Me an

Panel A: No short-selling and no borrowing 𝜋(0,1) TD ST LT M IN IP mea TS U M1 R R R F G n N 4.5 5.3 5. 5.6 5.6 5.6 5.7 5.7 5.77 4 6 44 9 5 4 0 6 2.4 2.0 2. 2.8 2.4 2.5 2.8 2.9 2.68 1 8 71 5 8 1 0 2 9.7 9.5 8. 9.7 9.7 9.6 9.7 9.7 9.74 1 1 31 1 2 6 2 6 5.1 4.6 5. 5.6 5.5 5.5 5.6 5.6 5.72 7 1 18 2 7 0 8 7 4.7 4.4 4. 5.1 5.0 5.1 5.1 5.2 4.99 9 8 96 1 2 2 1 3 4.6 3.8 4. 5.8 5.7 5.7 5.8 6.2 5.85 6 6 57 0 3 9 1 1 7.5 7.5 7. 8.1 8.1 8.1 N/ 8.1 8.04 1 1 58 0 0 1 A 0 3.8 3.4 3. 3.7 3.6 3.5 3.6 3.6 3.85 2 6 96 6 8 6 6 7 3.7 5.0 4. 5.5 5.5 5.4 5.6 5.7 5.75 3 0 86 8 7 9 9 9 3.8 5.6 6. 5.7 5.7 5.3 5.7 5.7 5.82 7 9 08 7 1 9 0 0 5.2 5.7 5. 5.7 5.4 5.6 5.6 5.6 5.78 6 0 74 1 8 3 7 4 5.2 5.4 5. 5.6 5.6 5.6 5.6 5.6 5.73 9 4 39 9 9 9 9 0 5.8 5.8 6. 6.0 5.7 5.4 5.7 5.9 5.82 8 0 12 0 2 8 2 2 4.4 5.1 5. 5.6 5.6 5.8 5.6 5.7 5.82 4 1 20 7 2 0 9 1 5.4 3.0 3. 5.6 5.6 5.6 5.6 5.6 5.71 6 4 54 8 4 6 9 9 5.8 6.2 6. 5.7 5.8 5.7 5.7 6.2 5.88 4 3 04 2 2 6 8 4 5.5 5.6 5. 5.6 5.6 5.6 5.6 5.6 5.68 8 2 67 5 7 5 7 4 5.3 5.6 5. 5.6 5.6 5.5 5.6 5.5 5.68 6 2 68 4 9 8 6 8 5.3 5.1 5. 5.8 5.7 5.7 4.8 5.9 5.83 3 1 34 3 4 4 1 1

5.72 5.77 5.67 5.71 5.75 5.86 5.78 5.84 5.67 5.66 5.82

35

ER

CR F

U. S.

CI R

5.9 8 2.6 5 9.7 9 5.6 7 4.4 8 5.5 4 7.9 7 4.3 2 6.0 1 6.1 1 5.8 7 5.7 9 5.9 3 5.8 4 5.8 0 5.7 5 5.6 9 5.8 1 5.8 0

5.6 9 2.8 4 9.7 2 5.7 1 4.6 0 5.8 4 8.1 0 3.7 4 5.6 9 5.6 9 5.6 9 5.6 9 5.6 9 5.6 9 5.2 6 5.6 9 5.7 0 5.6 9 5.7 8

8.3 4 6.4 6 10. 43 8.1 8 5.6 2 7.2 1 8.1 0 6.5 3 6.5 0 8.1 9 7.8 0 6.2 5 7.9 6 8.8 1 5.9 1 7.8 3 6.2 1 5.9 5 7.6 1

6.2 7 2.8 4 10. 07 6.4 8 4.8 9 5.9 9 8.0 9 4.0 7 6.4 4 6.2 8 5.9 7 6.0 2 5.2 9 6.5 0 6.1 2 5.4 8 5.6 5 5.5 0 6.0 9

Aver age 5.84 2.96 9.67 5.75 4.95 5.59 7.27 4.02 5.53 5.85 5.85 5.69 5.96 5.84 5.29 6.02 5.70 5.65 5.76

ACCEPTED MANUSCRIPT

Africa Americas BM CG CS FI HC IN OG TE TEL UT Regions average Sectors average

5.86 5.09 5.92 8.46 3.76 6.43 6.10 5.63 6.39 5.81 6.13 6.28 5.79 5.66 5.67 6.09 5.99

5.6 5

7.1 4

5.9 2

5.74

ER

CR F

U. S.

CI R

Aver age

6.6 5 2.5 1 11. 51 5.7 6 4.5 3 5.8 9 9.1 2 4.5 8 7.5 0 6.8 7 6.1 7 7.2 6 5.9 9 6.7 1 6.5 6 5.6 7 5.6 7 5.8 4 6.3 2 6.4 2

6.8 5 2.6 3 11. 91 5.8 6 4.6 0 6.8 0 8.6 6 3.6 0 7.9 5 7.7 2 6.4 0 6.5 8 5.5 6 6.6 8 5.2 5 5.4 0 5.9 1 5.8 9 6.3 6 6.3 3

11. 25 8.3 4 14. 15 11. 60 7.0 3 9.6 1 10. 60 7.8 3 10. 21 11. 24 9.7 3 8.8 6 9.1 4 12. 51 6.6 0 8.3 8 7.0 1 6.0 2 10. 05 8.9 7

6.0 0 2.1 8 12. 33 7.0 0 4.8 6 5.7 9 8.5 4 3.4 3 7.3 4 5.9 2 5.5 1 7.0 7 5.1 5 6.8 3 6.5 2 5.3 8 5.5 0 5.4 1 6.2 7 6.0 6

T

5.8 6

IP

2.59 10.9 8

SC R

Asia

6.06

NU

Australasi a

6.6 1 3.2 5 11. 34 6.3 8 5.1 3 5.9 7 8.5 9 4.1 3 6.6 7 6.9 5 6.3 3 6.5 4 6.1 5 6.8 6 5.7 1 6.0 1 5.7 7 5.7 5 6.4 3 6.2 8

MA

Europe

Med ian

D

Emerging

Me an

5.0 5.3 5. 5.7 5.6 5.6 5.7 5.7 7 2 43 1 6 1 0 5 Panel B: Short-selling and borrowing 𝜋(−0.5,1.5) TD ST LT M IN IP mea TS U M1 R R R F G n N 4.5 5.7 6. 5.8 5.5 5.8 6.0 6.1 6.20 4 1 16 2 7 1 0 4 2.4 2.0 2. 2.8 2.4 2.5 2.8 3.0 2.68 1 8 71 5 8 2 0 8 11.2 10. 13. 8. 10. 10. 10. 10. 11. 4 55 00 31 75 87 52 65 03 5.1 4.5 5. 5.7 5.5 5.6 5.8 5.8 5.81 9 2 10 5 2 6 5 0 4.7 4.4 4. 5.2 5.0 5.1 5.2 5.6 5.02 9 8 96 2 5 5 0 1 4.6 3.8 4. 5.9 5.7 5.7 5.8 6.5 5.83 6 6 57 3 2 6 4 1 7.5 8.8 7. 8.4 8.4 8.4 0.0 8.4 8.46 1 4 75 4 0 7 0 4 3.8 3.4 3. 3.7 3.6 3.5 3.6 3.6 3.85 2 6 96 5 8 1 6 1 3.7 5.8 5. 5.9 5.7 6.0 6.3 6.5 6.56 3 0 89 6 5 6 6 6 3.8 7.2 8. 5.8 5.6 5.7 6.0 5.9 6.64 7 7 12 2 9 4 4 9 5.2 6.2 6. 5.7 5.4 5.5 5.7 5.6 6.03 6 3 91 1 8 8 0 7 5.3 5.8 6. 6.1 6.2 6.1 6.2 6.2 6.44 0 0 23 1 9 4 9 4 5.8 6.0 6. 6.0 5.7 5.4 5.7 6.1 6.00 8 2 27 2 1 1 5 3 4.4 5.1 5. 5.8 5.5 6.2 6.0 6.1 6.20 4 1 33 6 6 3 4 8 6.2 3.0 3. 6.2 6.2 6.1 6.1 6.1 5.87 2 4 54 2 2 9 5 3 5.7 6.1 6. 5.6 5.7 5.6 5.8 6.3 5.86 5 2 00 9 9 9 0 6 5.5 5.5 5. 5.6 5.6 5.7 5.6 5.6 5.70 5 6 64 4 6 9 7 3 5.3 6.0 6. 5.6 5.8 5.5 5.6 5.5 5.72 6 0 27 4 1 6 8 7 5.4 5.7 5. 6.0 5.9 5.9 5.0 6.2 6.14 3 4 44 6 1 2 0 8 5.1 5.6 6. 5.8 5.8 5.8 5.9 6.0 6.10 4 9 02 7 0 4 5 5 5.77

TE

Develope d

5.74

CE P

Global

5.8 1

AC

Sectors average

36

6.37 3.05 11.3 0 6.13 5.12 5.91 7.90 4.08 6.59 6.69 6.20 6.51 6.08 6.46 5.72 6.00 5.77 5.75 6.23 6.18

ACCEPTED MANUSCRIPT Table X: Results from regression model using U.S. risk factors

T

This table presents results from regressing the portfolio profits (generated using a risk aversion factor of six) on U.S. risk factors. The risk factors are market excess return (MKT), small-minus-big (SMB) stock returns, highminus-low (HML) book-to-market returns, which are all downloaded from French’s website; and bettingagainst-beta (BAB) factor which is downloaded from Frazzini’s website. The regression model has the following form:

IP

Profit 𝑡 = 𝛼 + 𝛽𝑀𝐾𝑇 [𝑅𝑚,𝑡 − 𝑟𝑓,𝑡 ] + 𝛽𝑆𝑀𝐵 𝑆𝑀𝐵𝑡 + 𝛽𝐻𝑀𝐿 𝐻𝑀𝐿𝑡 + 𝛽𝐵𝐴𝐵 𝐵𝐴𝐵𝑡 + 𝜀𝑡

Europe Australasi a Asia Africa Americas BM CG CS FI HC IN OG TE TEL UT Regions average

9.93 5.94 5.09

MA

2.54

D

Emerging

5.84

TE

Develope d

5.9 8 3.3 0 9.9 5 6.0 3 5.2 0 6.0 1 8.0 3 4.0 5 5.9 1 6.0 5 5.9 6 5.8 8 5.9 8 6.0 8 5.8 0 5.9 1 5.8 0 5.8 2 6.0 7

5.90 8.06 3.58

AC

Global

Med ian

Panel A: No short-selling and no borrowing 𝜋(0,1) TD ST LT M IN IP mea TS U M1 R R R F G n N 4.6 5.3 5. 5.7 5.7 5.7 5.8 5.8 5.87 7 7 35 6 4 4 2 5 2.3 2.0 2. 2.8 2.4 2.4 2.7 2.8 2.67 0 2 38 8 4 9 7 3 9.9 9.7 8. 9.9 9.9 9.8 9.9 9.9 9.94 1 1 73 2 2 6 2 6 5.2 4.8 5. 5.7 5.7 5.6 5.8 5.8 5.90 5 8 33 5 6 3 7 4 4.9 4.6 5. 5.1 5.0 5.1 5.1 5.3 5.07 1 8 01 9 7 9 9 0 5.0 4.4 5. 5.8 5.7 5.8 5.8 6.2 5.95 8 6 20 1 4 3 2 1 7.5 7.5 7. 8.0 8.0 8.0 N/ 8.0 8.02 9 4 60 7 7 8 A 7 3.2 2.9 3. 3.6 3.4 3.2 3.5 3.5 3.65 1 2 43 3 7 9 2 3 4.3 5.0 4. 5.6 5.6 5.6 5.8 5.9 5.87 5 3 80 6 7 0 1 1 4.5 5.7 6. 5.8 5.8 5.5 5.8 5.8 5.93 5 6 05 4 1 2 2 2 5.3 5.7 5. 5.7 5.5 5.7 5.7 5.7 5.88 8 9 83 1 8 2 8 7 5.4 5.5 5. 5.8 5.8 5.8 5.8 5.7 5.85 5 4 47 1 1 1 1 3 5.7 5.8 6. 5.9 5.7 5.5 5.8 5.9 5.92 3 7 12 7 7 8 3 5 4.6 5.1 5. 5.7 5.7 5.8 5.8 5.8 5.94 9 2 06 4 2 7 1 0 5.4 3.4 3. 5.8 5.7 5.7 5.8 5.8 5.80 4 5 72 0 6 9 1 1 5.7 5.6 5. 5.8 5.7 5.7 5.8 6.1 5.84 8 6 53 0 6 2 3 8 5.6 5.6 5. 5.7 5.7 5.7 5.7 5.7 5.78 2 2 79 2 7 5 8 5 5.5 5.7 5. 5.7 5.8 5.6 5.7 5.7 5.80 4 3 79 6 1 9 8 0 5.3 5.2 5. 5.8 5.7 5.7 5.5 5.9 5.88 7 0 38 8 8 6 6 5

CE P

Me an

NU

SC R

The regression is run using OLS where standard errors are corrected for heteroskedasticity and autocorrelation based on Newey-West procedure. A maximum of 12 lags are incorporated and the optimal lag length is chosen using the Schwarz Information criterion. The annualised alpha coefficient is reported. Panel A is based on a profit dependent variable that is free of short-selling and borrowing whereas Panel B has profits (dependent variable) that accounts for 50% short-selling and borrowing. The average alphas based on the predictors are reported in the last column, while the average alphas of regional and sectoral portfolios are reported in the last two rows of each panel. Finally, N/A implies that DUN data is unavailable for Africa. All alphas are statistically different from zero at the 1% level or better.

5.82 5.88 5.77 5.83 5.83 5.97 5.89 5.85 5.77 5.77 5.86

37

ER

CR F

U. S.

CI R

6.1 1 2.6 3 10. 01 5.8 8 4.4 4 5.6 8 7.9 8 4.1 6 6.1 6 6.2 9 5.9 9 5.9 4 6.1 1 6.0 1 5.9 2 5.7 6 5.7 9 5.9 5 5.8 6

5.8 1 2.8 8 9.9 2 5.8 9 4.7 3 5.8 4 8.0 7 3.7 2 5.8 1 5.8 1 5.8 1 5.8 1 5.8 1 5.8 1 5.2 9 5.8 1 5.8 2 5.8 1 5.8 6

8.6 6 6.7 1 10. 75 8.4 0 5.9 5 7.6 1 8.3 5 6.2 5 6.9 7 8.7 4 8.1 8 6.5 0 8.1 8 9.2 0 6.1 0 7.5 0 6.3 0 6.1 3 7.8 4

6.2 4 2.6 5 10. 24 6.5 7 4.8 6 5.9 5 8.0 6 3.7 7 6.4 5 6.2 8 5.9 7 6.1 1 5.2 0 6.5 0 6.1 4 5.2 8 5.7 4 5.6 0 6.0 4

Aver age 5.93 2.91 9.90 5.92 5.04 5.77 7.29 3.74 5.69 6.03 5.96 5.82 6.01 5.94 5.42 5.89 5.79 5.78 5.81

ACCEPTED MANUSCRIPT

Africa Americas BM CG CS FI HC IN OG TE TEL UT Regions average Sectors average

6.07 5.12 5.92 8.31 3.58 6.30 6.11 5.74 6.45 5.86 6.18 6.12 5.83 5.78 5.78 6.09 6.01

5.7 6

7.3 8

5.9 3

5.83

ER

CR F

U. S.

CI R

Aver age

7.2 9 2.6 1 11. 96 6.3 3 4.4 3 6.1 2 8.9 8 4.6 5 8.0 3 7.4 3 6.6 4 7.8 3 6.4 8 7.3 7 6.7 0 5.8 7 5.7 9 6.0 4 6.5 5 6.8 2

6.9 4 2.6 7 11. 80 6.1 6 4.7 3 6.4 1 8.3 9 3.6 3 7.2 6 7.2 7 6.5 9 6.5 5 5.8 1 6.6 4 5.2 5 5.8 4 6.2 1 5.9 4 6.3 4 6.3 3

12. 08 9.4 4 14. 54 12. 62 8.0 7 10. 14 10. 81 8.9 5 10. 90 11. 91 10. 75 9.7 8 9.8 6 13. 25 7.1 3 9.5 3 7.5 0 6.3 3 10. 83 9.6 9

6.3 8 2.5 2 12. 69 7.7 4 5.0 0 5.8 3 8.4 3 3.8 7 7.5 9 6.1 5 5.8 6 7.5 7 5.1 7 7.2 8 6.7 0 5.3 8 5.5 9 5.5 3 6.5 6 6.2 8

T

5.9 9

IP

2.54 11.0 5

SC R

Asia

6.12

NU

Australasi a

6.9 4 3.4 9 11. 52 6.8 8 5.2 9 6.1 6 8.4 7 4.1 6 6.7 6 7.1 2 6.6 1 6.7 9 6.3 2 7.2 4 5.7 5 6.1 2 5.9 3 5.8 6 6.6 2 6.4 5

MA

Europe

Med ian

D

Emerging

Me an

5.2 5.3 5. 5.7 5.7 5.7 5.8 5.8 5 6 42 8 4 1 1 4 Panel B: Short-selling and borrowing 𝜋(−0.5,1.5) TD ST LT M IN IP mea TS U M1 R R R F G n N 4.6 5.5 5. 5.8 5.6 5.7 6.0 6.2 6.35 7 3 92 2 7 9 8 3 2.3 2.0 2. 2.8 2.4 2.4 2.7 3.0 2.70 0 2 38 8 4 9 7 0 11.3 10. 12. 8. 10. 10. 10. 10. 11. 7 60 62 73 76 91 51 73 09 5.2 4.8 5. 5.8 5.7 5.7 6.0 6.0 6.16 6 6 34 6 7 7 5 4 4.9 4.6 5. 5.2 5.0 5.1 5.2 5.5 5.09 1 8 00 4 9 7 3 6 5.0 4.4 5. 5.8 5.7 5.7 5.8 6.3 5.94 8 6 20 5 2 7 2 7 7.5 8.4 7. 8.2 8.2 8.3 N/ 8.2 8.32 9 4 70 7 5 2 A 5 3.2 2.9 3. 3.6 3.4 3.2 3.5 3.5 3.69 1 2 43 1 7 5 2 1 4.3 5.4 5. 5.8 5.7 5.8 6.2 6.3 6.54 5 2 44 9 3 0 1 8 4.5 6.6 7. 5.8 5.7 5.6 6.0 6.0 6.65 5 6 33 6 7 7 5 0 5.3 5.9 6. 5.7 5.5 5.6 5.8 5.8 6.17 8 7 64 2 8 0 0 2 5.4 5.7 6. 6.1 6.3 6.1 6.3 6.3 6.62 6 8 15 8 4 2 5 5 5.7 5.8 6. 5.9 5.7 5.4 5.8 6.1 6.09 3 9 23 9 6 7 3 5 4.6 5.0 5. 5.8 5.6 6.2 6.0 6.2 6.40 9 7 16 7 5 1 8 3 5.8 3.4 3. 6.0 6.0 6.0 6.0 6.0 5.84 3 5 72 5 0 6 5 2 5.7 5.5 5. 5.8 5.7 5.6 5.8 6.5 5.87 5 8 49 0 4 5 5 0 5.6 5.6 5. 5.7 5.7 5.9 5.7 5.7 5.83 0 2 89 2 8 1 9 7 5.5 5.9 6. 5.7 5.8 5.6 5.8 5.6 5.82 4 8 21 6 8 6 0 9 5.4 5.6 5. 6.0 5.9 5.8 5.7 6.2 6.20 5 9 46 4 1 8 4 6 5.2 5.5 5. 5.8 5.8 5.8 5.9 6.0 6.18 9 4 83 8 2 1 8 9 5.86

TE

Develope d

5.84

CE P

Global

5.9 2

AC

Sectors average

38

6.53 3.13 11.4 1 6.48 5.26 6.06 8.49 4.00 6.58 6.72 6.36 6.70 6.20 6.62 5.75 6.08 5.93 5.86 6.33 6.28

ACCEPTED MANUSCRIPT Are Islamic Stock Returns Predictable? A Global Perspective

HIGHLIGHTS We compile a new data set (January 1981 to December 2014) containing 2,577

T



IP

Islamic stocks.

We find that financial and macroeconomic variables predict stock returns.



Robust evidence of predictability exists only when U.S. stock returns are used as a

SC R



CE P

TE

D

MA

Regional (industry) portfolios offers average profits of 6.16% (6.03%) per annum.

AC



NU

predictor.

39