Investment homework help

Journal of Banking & Finance 34 (2010) 236–245

Contents lists available at ScienceDirect

Journal of Banking & Finance

j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / j b f

International diversification strategies: Revisited from the risk perspective

Ye Bai a,*, Christopher J. Green b

a School of Accounting, Economics and Statistics, Edinburgh Napier University, Craiglockhart Campus, 219 Colinton Road, Edinburgh EH14 1DJ, UK b Department of Economics, Loughborough University, UK

a r t i c l e i n f o a b s t r a c t

Article history: Received 23 July 2008 Accepted 24 July 2009 Available online 6 August 2009

JEL classification: G15

Keywords: Emerging equity markets Diversification Cross-sectional variation Risks

0378-4266/$ - see front matter � 2009 Elsevier B.V. A doi:10.1016/j.jbankfin.2009.07.026

* Corresponding author. Tel.: +44 131 455 4423; fa E-mail addresses: [email protected] (Y. Bai), C.J.G

Following Roll [Roll, R., 1992. Industrial structure and comparative behaviour of international stock mar- ket indices. Journal of Finance 47, 3–42] and Heston and Rouwenhorst [Heston, S.L., Rouwenhorst, G.K., 1994. Does industrial structure explain the benefits of international diversification. Journal of Financial Economics 36, 3–27], researchers have decomposed stock returns into country and industry components. Evidence suggests that industry components have become more important in recent years, but the rea- sons for this are unclear. Existing research concentrated mainly on stock returns in industrial countries. In this paper we consider instead the decomposition of stock risks within emerging equity markets. We provide a rationale for this procedure and its relationship to return decompositions. The results provide new firm-specific evidence on the debate over country and industry components, their stability over time, and the implications for portfolio diversification.

� 2009 Elsevier B.V. All rights reserved.

1. Introduction

It is generally believed that potential gains from international diversification are generated from the imperfect correlation among national stock market returns. However, with the increasing trend of financial liberalization in emerging stock markets (ESMs), we would expect to observe shifts in portfolio holdings of investors away from domestic assets in the major industrial countries and towards ESMs. Theoretically, during this process, risk-sharing among international investors should increase, and therefore ESMs become less sensitive to domestic economic shocks and more inte- grated with the rest of the world. Bekaert et al. (2008) report that although there is a significant decrease in segmentation over time, ESMs are still not as effectively integrated as developed markets.

One line of research has focused particularly on identifying the factors that impede perfect integration among national equity markets and quantifying the extent of integration. Following the methodological framework proposed by Heston and Rouwenhorst (1994), there has been renewed interest in using country and industry factors to explain stock returns and their low interna- tional co-movement. Country factors may include local monetary and fiscal policies, business cycle indicators, and the institutional and legal regimes; industry factors may include industry composi- tion, size and other industry-specific characteristics. Examples in- clude: Serra (2000), Campa and Fernandes (2004), Phylaktis and

ll rights reserved.

x: +44 131 455 3475. [email protected] (C.J. Green).

Xia (2006) and Bai et al. (2008). In the absence of integration, we would expect country factors to be more important than industry factors in explaining stock returns. As integration proceeds, coun- try factors should become less important and (trans-national) industry factors more important. Ultimately this may lead to the reduction of portfolio diversification benefits from ESMs and a greater focus on industry diversification.

Existing studies of industry and country factors have concen- trated exclusively on decompositions of the cross-section of stock returns. However, since one of the main benefits expected from portfolio diversification is risk reduction, movements in stock re- turns should reflect in part the outcome of portfolio shifts in re- sponse to perceived changes in stock risks. In a fully integrated world market, theory would suggest that market risk is the most important factor driving stock returns, as individual risk is diversi- fied away. However, there are several reasons for thinking that it would be useful to look directly at the cross-section of individual stock risks, using the same country-industry decomposition as for returns.

First, as noted at the outset, there is ample evidence that inter- national stock markets are not fully integrated, especially when ESMs are included in the analysis (Bai, 2008). Lack of integration implies that country and individual risks will be important factors in determining stock returns, and therefore investor behaviour (Chandar et al., 2009; Chung et al., 2009). Second, many groups of investors are constrained in their portfolio selection by asymmetric information, domestic regulations and corporate poli- cies (Kalev et al., 2008; Silva and Chavez, 2008). Pension funds and

2

Y. Bai, C.J. Green / Journal of Banking & Finance 34 (2010) 236–245 237

others have fiduciary obligations which often limit the currencies and firms in which they can invest. Such investors are inevitably concerned with individual as well as market risk. Third, as Camp- bell et al. (2001) observe, individual risk is important in areas other than portfolio selection, for example in event studies and in the pricing of stock options. Fourth, if risk differences in segmented markets are an important underlying driving force of stock return movements, we might expect there to be some correspondence be- tween the return decomposition and the risk decomposition. But, if this were so, investors who pursue a strategy of diversifying among countries (for example) based exclusively on the return decompo- sition may simply be shifting towards securities with greater total risk (i.e. including country-specific and company-specific risk), without necessarily achieving the desired overall risk reduction.

In this paper therefore, we study the decomposition of stock risks following broadly the same methodology as that typically used to study stock returns (Heston and Rouwenhorst, 1994; Bai et al., 2008). We estimate monthly measures of conditional total risk for a large sample of individual stocks, using stock-specific models of the conditional mean return, with (G)ARCH-type errors to model the time-varying risks. We then calculate month-by- month cross-sectional decompositions of these risks into global, country, industry and idiosyncratic components. We focus particu- larly on the country and industry effects and examine changes over time in the relative contributions and significance of these effects using F-tests in the monthly cross-section regressions. We com- pare our results with analogous return decompositions, and finally consider how far the time series of country and industry risks may be explained by more fundamental data from which the risks may derive.

In addition to its focus on risk decomposition, the present paper is novel in several ways. First, previous results on stock return decompositions are mostly based on industry components of na- tional stock market indices rather than individual companies. The use of indices has several limitations, for example: first, invest- ment managers buy individual shares not industrial indices; and second, the decomposition results will depend in part on the weighting and changing composition of the industry indices. Therefore, we use firm-level data for the analysis. Specifically, we use monthly data to analyse a sample of over 1500 firms in 13 ESMs and 11 industry groups covering the period August 1984 to July 2004.1 Second, the analysis concentrates exclusively on ESMs and thus gives information about risk and diversification within ESMs and not also as between ESMs and other countries. The re- sults of Grisolia and Navone (2007) suggest that it may be reason- able to look separately at different regional levels of asset allocation. Our sample excludes Europe and North America but includes firms from Africa, Asia and Latin America. Third, since it is agreed that ESMs are not well-integrated into international capital markets, we would expect all the elements of total risk to be important in understanding diversification possibilities, whereas in countries that are more fully integrated into world markets, it may be sensible to focus more on market risk. Fourth, we use a more aggregated industry classification than in most re- turn decomposition studies. We argue that an unduly fine classifi- cation can involve arbitrary allocations of multi-activity firms to specific sectors, a particularly important problem in ESMs where conglomerates are more prevalent than in the industrial countries.

The general finding on ESMs’ return data is that the importance of industry effects has been increasing over time in determining the cross-sectional variation of stock returns, but country effects still tend to dominate industry effects (Bai et al., 2008). In this pa- per, we employ the same underlying dataset as in Bai et al (2008),

1 As far as we are aware, only Campa and Fernandes (2004) cover more ESMs.

Tables of results using the variance are available from the authors on request. 3 The vast majority of the (G)ARCH models are GARCH (1,1). In a few cases, the

(G)ARCH estimation does not converge within a reasonable number of iterations These firms are dropped from the dataset.

and therefore we can more easily compare the results of the risk and return decompositions.

The rest of the paper is organized as follows. In Section 2 we discuss the measurement and calculation of total risk using the conditional variance or standard deviation, and the methodology for risk decomposition as among world, country, industry and idi- osyncratic effects. In Section 3 we provide results of the decom- positions; Section 4 outlines possible determinants of the industry and country effects; and Section 5 contains some con- cluding remarks and discussion. Bai et al. (2008) describe the dataset in detail; in this paper we provide a brief summary in an Appendix.

2. Methodology

The common measure of share and portfolio risk used in asset pricing theory is either the standard deviation or variance of re- turns. The primary use of the risk measure in this paper is to cal- culate linear decompositions into country and industry effects. Thus, deciding between the two measures amounts to deciding whether the ‘‘true” decomposition is linear in the standard devi- ation or the variance. Since this is unknown a priori, we take an agnostic approach and calculate the decompositions using first one and then the other. There are few major differences in the important elements of the results. For compactness therefore, we report and discuss only the results using the standard deviations.2

For each company, we estimate time series of the conditional standard deviation of returns (CSDs) using a separate univariate ARMA-(G)ARCH model for each firm (Engle and Bollerslev, 1986). The conditioning information is therefore firm-specific in each case. The model is estimated in three steps. First, we used correlo- gram diagnostics to identify a low-order ARMA (p,q) model for the conditional mean. In fact, 75% of all firms’ data are consistent with a constant mean return model (Table 1). We then estimated the conditional mean model and checked the residuals for (G)ARCH ef- fects using the standard LM test (Engle, 1982). Finally, we esti- mated a combined model with low-order ARCH or GARCH errors, and recovered the time series of conditional variances and implied CSDs from these estimates. The main purpose of these calculations is to identify a time series of conditional risks. Therefore the ARMA-(G)ARCH models are estimated primarily with a view to parsimony as well as model accuracy and coefficient significance, and we were able to fit plausible low-order ARMA-(G)ARCH mod- els to virtually all the firms in the dataset.3

To identify the separate country and industry influences on risk, we apply the standard dummy variable (fixed-effect) model of Heston and Rouwenhorst (1994) to the cross-section of CSDs in each month:

s$ijk ¼ d þ X11 j¼1

/jDi;j þ X13 k¼1

uk Ci;k þ ei ð1Þ

Here: s$ijk is the CSD of security i belonging to industry j and country k at time t;Di,j and Ci,k are the (11) industry and (13) country dum- mies, and:

Di;j ¼ 1 if security i belongs to industry j 0 otherwise

�

Ci;k ¼ 1 if security i belongs to country k 0 otherwise

�

.

Table 1 Conditional standard deviation (CSD) and variance calculations by country. Table gives summary statistics for the calculation of the firm-and-time-specific conditional standard deviations and variances. Two main types of return models have been identified from the correlogram statistics of each stock return: a constant mean model (including zero mean) and an ARMA (p,q) model. Panel A shows for each country, the percentages of the two types of mean model in the total sample. ARCH LM tests have been applied to test for ARCH behaviour of the residuals of the conditional mean model. These are based on the Lagrange multiplier principle with the null hypothesis of no ARCH errors versus the alternative hypothesis that the conditional variance can be represented by an ARCH (q) process. Following identification of the (G)ARCH process an ARMA-(G)ARCH model is fitted to the data. Panel B shows the percentages of the firms represented by either ARCH or GARCH errors. Each percentage is calculated by dividing the number of the firms corresponding to each cell by the total number of the firms in the sample.

Countries No. of firms Panel A Panel B Conditional mean models Conditional variance models

Constant mean (% models) ARMA (p,q) (% models) ARCH (% models) GARCH (% models)

Brazil 64 1.63 2.54 1.69 2.34 Chile 64 0.79 3.38 0.91 3.12 China 144 2.08 7.29 3.58 5.53 India 178 6.51 5.14 2.41 8.98 Israel 29 0.72 1.11 0.39 1.30 Malaysia 237 13.86 1.30 0.72 14.38 Mexico 39 2.02 0.52 0.26 2.21 Pakistan 44 2.15 0.72 0.59 2.08 South Africa 69 4.10 0.26 0.59 3.51 South Korea 244 15.29 0.52 1.37 14.31 Taiwan 178 10.86 0.72 1.04 10.15 Thailand 185 11.58 0.46 0.91 11.13 Taiwan 62 3.71 0.33 0.46 3.12 Sum 1537 75.30 24.29 14.92 82.16

Country Dummies F-test Probability

0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00

1/ 8/ 84

1/ 10 /8 5

1/ 12 /8 6

1/ 2/ 88

1/ 4/ 89

1/ 6/ 90

1/ 8/ 91

1/ 10 /9 2

1/ 12 /9 3

1/ 2/ 95

1/ 4/ 96

1/ 6/ 97

1/ 8/ 98

1/ 10 /9 9

1/ 12 /0 0

1/ 2/ 02

1/ 4/ 03

1/ 6/ 04

Industry Dummies F-test Probability

0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00

/8 /8 4

/1 0/ 8

/1 2/ 8

/2 /8 8

/4 /8 9

/6 /9 0

/8 /9 1

/1 0/ 9

/1 2/ 9

/2 /9 5

/4 /9 6

/6 /9 7

/8 /9 8

/1 0/ 9

/1 2/ 0

/2 /0 2

/4 /0 3

/6 /0 4

238 Y. Bai, C.J. Green / Journal of Banking & Finance 34 (2010) 236–245

Eq. (1) states that each CSD may be decomposed into the sum of a constant (the global effect), an industry component, a country component, and an error term (the idiosyncratic effect). Since both sets of dummy variables are perfectly collinear, we identify the coefficients by normalising their value-weighted sums to be equal to zero. This is equivalent to measuring industry and country ef- fects in comparison to the ‘average firm’ in the sample (Heston and Rouwenhorst, 1994). Specifically, we define wj and v k as the shares of industry j and country k, respectively in total sample cap- italization and impose the restrictions:

X11 j¼1

wj/j ¼ 0 and X13 k¼1

v kuk ¼ 0 where X

j

wj ¼ X

k

v k ¼ 1 ð2Þ

Under these restrictions, the Ordinary Least Squares estimators (OLS) of / and u are fully identified. In addition, the estimated dis- turbances are by construction orthogonal to all the industry and country dummies so that the average residual is zero in every country and industry. Applying the constraints (2) to Eq. (1) gives

s$ijk ¼ d þ X10 j¼1

/jdi;j þ X12 k¼1

ukci;k þ ei ð3Þ

with di;j ¼ðDi;j �ðwj=wJÞDi;JÞ; ci;k ¼ðCi;k �ðv k=v KÞCi;KÞ; and J and K are the (arbitrary) specific industry (J) and country (K) on which the restrictions are normalised. Estimation of (3) is equivalent to estimating a model of excess risks over an average benchmark (d̂).4 Following estimation, we apply separate F-tests to the industry and country dummies each month. These give a general picture of the significance of the industry and country effects over time but also build a foundation for the decomposition analysis.

To construct the industry and country decompositions follow- ing Heston and Rouwenhorst (1994), we multiply (3) by the share of each firm (pi,k) in the total market value of its country and sum over all firms in each country (k) to get for example:

S$k ¼ d̂ þ X10 j¼1

pk;j/̂jdi;j þ ûk ð4Þ

4 The averages in the decomposition models are not portfolio averages because they neglect the covariances among securities.

(pk,j is the share of the total market value of country k included in industry j.) Eq. (4) decomposes the value-weighted sum of country risks (S$k) into a common (global) component (d̂), an average of the industry effects of the stocks that make up its marketP10

j¼1pk;j/̂j di;j � �

, and a country-specific component (ûk). It states that the risk in country k (e.g. Brazil) may be different from the overall sample average risk for two reasons: first, the industry dis- tribution of Brazil’s market is different from that of ESMs as a whole; second, the risk of Brazilian stocks may differ from that of stocks in the same industry but located in a different country.

1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

Fig. 1. F-test Probabilities for CSD. Figure shows probability values of the F-test for both industry dummies and country dummies from August 1984 to July 2004. The dominant country effects and the increasing importance of industry effects that have been further confirmed from the point of view of stock risks.

Table 2 Dummy variables model F-tests for CSD. Table summarises the F-tests done separately for the industry and country dummies in each month. It shows the percentage of significant and insignificant months within each sample period. From August 1984 to February 1986, industry and country effects both have significant influences on risks; country effects play the dominant role in explaining the variations in stock risk during the second sub-period; industry and country factors are again almost equally important since mid-1996.

Sample period Industry dummies Country dummies

No. of months % Sig. at 5% % Sig. at 10% % Insig. Total % Sig. at 5% % Sig. at 10% % Insig. Total

08/84-02/86 19 21.1 52.6 26.3 100.0 63.2 21.1 15.8 100.0 03/86-03/96 120 3.3 3.3 93.3 100.0 95.0 0.8 4.2 100.0 04/96-07/04 101 96.1 3.9 0.99 100.0 100.0 0.0 0.0 100.0

Y. Bai, C.J. Green / Journal of Banking & Finance 34 (2010) 236–245 239

Similarly, each value-weighted sum of industry risks can be expressed as the sum of the common component (d̂), the average of the country components of the stocks in industry jP12

k¼1pj;kûk ci;k � �

, and an industry-specific component (/̂j):

S$j ¼ d̂ þ X12 k¼1

pj;kûk ci;k þ /̂j ð5Þ

(pj,k is the share of the total market value of industry j included in country k.)

Eqs. (4) and (5) show the industry and country effects for one particular month. Estimating (3) for every month yields two sets of time series giving: the geographically-diversified risks of firms in industry j (d̂ þ /̂j), each of which has the same country compo- sition as the average sample firm; and the industrially-diversified risks of firms in country k (d̂ þ ûk ), each of which has the same industry composition as the average firm.

The standard deviation of the pure industry effect time series (SD ð/̂jÞ) gives an absolute measure of how important the pure industry effects are in determining the variation of geographi- cally-diversified industry index risks. Relative measures are given by the ratios:

SDð/̂jÞ SD /̂j þ

P12 k pj;kûk ci;k

� � ð6aÞ

SD P12

k pj;kûkci;k � �

SD /̂j þ P12

k pj;kûk ci;k � � ð6bÞ Eq. (6a) shows how much of the variation in the excess risk of

each industry portfolio is explained by the pure industry effect; correspondingly (6b) shows how much of the variation is ex- plained by the sum of country effects. Analogous measures may be computed for the country time series. Further details are given in Bai et al. (2008).

The foregoing method is used to gauge the relative importance of country and industry effects. We use Mean Absolute Deviations (MADs) to measure the magnitudes of the industry and country ef- fects (Cavaglia et al., 2000; Bai et al., 2008).

MADk;t ¼ X13 k¼1

v k;tjuk;tj ð7Þ

Eq. (7) gives the absolute values of estimated pure country ef- fects at time t, multiplied by the corresponding share of country k in total sample capitalization. The larger is the MAD, the greater are the magnitudes on average and the more dispersed they are around the sample average in that month. An analogous measure may be derived for the industry effect.

For further analysis, following Bai et al. (2008), we grouped the 11 industry sectors into 5 traded-goods and 6 non-traded goods groups.5 For these two broad groups we compare the means and

5 Traded goods are: Industrial goods and services (IGS), Consumer goods (CG) Consumer services (CS), Basic materials (BM), Utilities (UT) and Health care (HC) non-traded goods are: Technology (TEC), Telecoms (TEL), Financials (FI), Personal and household goods (PHG) and Basic Industrials (BI).

, ;

medians of the SDs of the sum of country effects, the pure industry effects, and the ratios of the two components over excess risks ((6a) and (6b)). Due to their different sensitivities to international shocks, we anticipate that more variation in traded-goods indus- tries’ risk will be explained by pure industry effects, whereas we expect more variation in non-traded goods industries to arise from the sum of country effects.

3. Analysis of risk decompositions

We split the sample into 3 sub-periods (at March 1986 and April 1996) based on an inspection of the results of F-tests on Eq. (3) (Fig. 1 and Table 2). It is apparent that country effects are an important factor in determining stock risks for the whole sample period. In contrast, industry effects are important from August 1984 to February 1986, but almost negligible from about March 1986 until March 1996. From April 1996 to the end of the sample, industry effects again exhibit much greater and more uniform significance.

Results for the first sub-period may be affected by the relatively small number of companies and countries. Before February 1986 there are just three countries in the sample (Malaysia, South Africa and South Korea) but ten industries. However, 1986 also marks the start of a period of market liberalization. According to Henry (2000), many of the ESMs in our sample officially liberalized their stock markets during or after 1986. This is reflected to some extent in the relative importance of country and industry effects. Although many ESMs were officially liberalized, the dominant country effects after 1986 show that there was still long way to go from official liberalization to fuller integration with other emerging and developed stock markets.

The second break in April 1996 is close to but pre-dates the East Asian financial crisis. Countries and regions that were directly af- fected by the crisis contribute 64.15% of all sampled firms and 57.02% of market capitalization by December 2003. Our sample also includes non-Asian markets, but it would not be unreasonable to expect the crisis to be associated with changes in the underlying determinants of stock risks. Furthermore, the pattern of these re- sults is remarkably similar to those reported for stock return decompositions for industrial countries and emerging markets (Bai et al., 2008). Country effects dominate before the Asian crisis, but industry effects become more important as the crisis develops, and following. However, industry effects became important for stock risks from about April 1996, whereas they did not become important for returns until about a year later (Bai et al., 2008). This suggests that market fundamentals may have been changing be- fore the Asian crisis materialized. The greater integration in stock risks evidenced by the increased industry effects could have been a factor foreshadowing the contagion that appeared during the cri- sis itself, especially as the industry effects on returns lag the effect on risks in the way that they do.

Turning to the decomposition results, we see that for the period as a whole (Table 3) most of the cross-sectional variation in the country and industry CSDs can be attributed to country-specific ef-

Table 3 Decomposition of CSD into country and industry effects. Table gives the standard deviation (SD) of the components of the value-weighted country and industry CSDs. Each country CSD is decomposed into a pure country effect and a sum of 11 industry effects. Each industry CSD is decomposed into the sum of 13 country effects and a pure industry effect. The ratio relative to the market gives the ratio of the SD of that component to the SD of the sum of that pure effect and the sum of the value-weighted industry (country) effects. (i) The pure country effect measures the average CSD of firms in a country relative to firms that are in the same industry but located in a different country. (ii) The sum of the eleven industry effects represents the component of a country’s CSD that can be attributed to the difference between the industrial composition of its market and the industrial composition of the emerging markets. (iii) The pure industry effect measures the average CSD of firms in an industry relative to firms which are located in the same country but belong to a different industry. (iv) The sum of the thirteen country effects represents the component of an industry’s CSD that can be attributed to the difference between the geographical composition of its index and the geographical composition of the emerging market sample as a whole.

Country Pure country effect Sum of 11 industry effects Industry Sum of 13 country effects Pure industry effect

Std. Dev. Ratio Std. Dev. Ratio Std. Dev. Ratio Std. Dev. Ratio

Brazil 0.086 1.074 0.007 0.091 Industry goods 0.265 0.721 0.105 0.286 Chile 0.077 1.111 0.006 0.088 Con goods 0.406 0.869 0.062 0.133 China 0.072 1.156 0.005 0.080 Con services 0.183 0.773 0.056 0.237 India 0.080 1.133 0.003 0.043 Basic mats 0.165 1.792 0.091 0.993 Israel 0.355 0.927 0.043 0.111 Utilities 0.424 0.903 0.049 0.105 Malaysia 0.425 0.906 0.046 0.098 Health care 0.420 0.896 0.051 0.108 Mexico 0.075 1.083 0.005 0.070 Technology 0.418 0.873 0.064 0.134 Pakistan 0.083 1.209 0.005 0.066 Telecoms 0.423 0.888 0.059 0.124 South Africa 0.402 0.844 0.075 0.158 Financial 0.427 0.816 0.104 0.198 South Korea 0.426 0.916 0.040 0.086 Pers house 0.409 0.871 0.063 0.134 Taiwan 0.064 1.061 0.008 0.129 Basic industry 0.381 0.852 0.068 0.153 Thailand 0.703 1.018 0.060 0.087 Turkey 0.227 1.090 0.004 0.021 Cross-country average 0.236 1.041 0.024 0.087 Cross-industry average 0.356 0.932 0.070 0.237

Table 4 Decomposition of excess value-weighted CSD into country and industry effects in three sub-sample periods. Table gives the same information as Table 3 but with the results split into 3 sub-periods: 08.84–02.86; 03.86–03.96; and 04.96–07.04.

08.84–02.86 03.86–03.96 04.96–07.04

Pure country effect Sum of 11 industry effects

Pure country effect Sum of 11 industry Pure country effect Sum of 11 industry effects

Std. Dev. Ratio Std. Dev. Ratio Std. Dev. Ratio Std. Dev. Ratio Std. Dev. Ratio Std. Dev. Ratio

Country Brazil – – – – 0.110 1.044 0.006 0.055 0.017 0.950 0.006 0.304 Chile – – – – 0.112 1.179 0.004 0.043 0.014 0.814 0.006 0.323 China – – – – 0.110 1.414 0.004 0.049 0.023 0.897 0.004 0.167 India – – – – 0.112 1.174 0.003 0.032 0.015 0.927 0.003 0.197 Israel – – – – 0.465 0.894 0.057 0.109 0.027 0.912 0.006 0.218 Malaysia 0.013 0.912 0.003 0.183 0.572 0.902 0.064 0.101 0.049 0.986 0.004 0.079 Mexico – – – – 0.105 1.094 0.004 0.040 0.012 0.935 0.004 0.283 Pakistan – – – – 0.129 1.367 0.003 0.029 0.017 0.992 0.004 0.247 South Africa 0.025 0.987 0.002 0.063 0.541 0.842 0.103 0.161 0.015 0.967 0.003 0.195 South Korea 0.010 1.011 0.002 0.218 0.570 0.915 0.055 0.088 0.049 1.020 0.003 0.065 Taiwan – – – – 0.086 1.081 0.004 0.045 0.021 0.954 0.005 0.245 Thailand – – – – 0.910 0.989 0.084 0.091 0.233 1.001 0.002 0.009 Turkey – – – – 0.106 1.076 0.003 0.031 0.299 0.999 0.005 0.017 Average 0.016 0.970 0.002 0.155 0.302 1.075 0.030 0.067 0.061 0.950 0.004 0.181

Sum of 13 country effects

Pure industry effect Sum of 13 country effects

Pure industry effect

Sum of 13 country effects

Pure industry effect

Industry Ind goods 0.010 0.928 0.002 0.592 0.356 0.717 0.144 0.290 0.012 0.861 0.005 0.393 Con goods 0.004 0.577 0.002 0.632 0.557 0.868 0.085 0.133 0.017 0.925 0.006 0.301 Con serves 0.004 0.491 0.004 1.112 0.252 0.774 0.077 0.236 0.012 0.848 0.006 0.441 Basic mats 0.018 0.850 0.002 0.290 0.229 1.792 0.126 0.984 0.015 0.836 0.007 0.376 Utilities 0.013 1.110 0.001 0.715 0.578 0.900 0.068 0.106 0.013 0.801 0.009 0.532 Health care 0.010 0.028 0.002 0.019 0.575 0.894 0.070 0.110 0.013 0.708 0.009 0.478 Technology 0.010 0.018 0.002 0.017 0.574 0.877 0.083 0.126 0.015 0.936 0.010 0.619 Telecoms – – – – 0.575 0.885 0.079 0.121 0.010 0.802 0.009 0.739 Financial 0.003 0.013 0.003 0.030 0.598 0.816 0.144 0.197 0.022 0.891 0.014 0.563 Pers house 0.010 0.044 0.002 0.026 0.566 0.873 0.085 0.132 0.014 0.876 0.011 0.684 Basic Ind 0.004 0.007 0.005 0.015 0.530 0.852 0.094 0.151 0.023 1.062 0.009 0.420 Average 0.009 0.407 0.003 0.345 0.490 0.931 0.096 0.235 0.015 0.868 0.009 0.504

240 Y. Bai, C.J. Green / Journal of Banking & Finance 34 (2010) 236–245

fects. Each pure country effect explains most of the excess risk for any country, in comparison with the sum of the industry effects. Conversely, the pure industry effects explain less than the sum of the country effects. Country effects also tend to dominate industry

effects in explaining the variation in the CSDs in each of the 3 sep- arate sub-periods studied (Table 4), but, in parallel with the F-test results, industry effects increase in importance in the final sub-per- iod in the run-up to, during and after the East Asian crisis. Once

Fig. 2. Mean Absolute Deviations for CSDs. Figure plots the industry and country MADs for value-weighted risks. The four figures show the MADs for the sample period as a whole, and then for the three sub-periods.

Y. Bai, C.J. Green / Journal of Banking & Finance 34 (2010) 236–245 241

again these results are similar to the findings for return decompo- sitions, but with industry effects becoming more important some- what earlier in time than for the returns.

The calculated industry and country MADs (Fig. 2) broadly sup- port the F-tests results, in that country MADs are mostly apprecia- bly larger than industry MADs, suggesting that country effects generally play the dominant role in determining the conditional standard deviation. Nevertheless, the increased importance of industry effects in the third sub-period is as apparent in the MAD plots as in the significance levels of the F-tests.

In Table 5 we distinguish between traded and non-traded goods industry components of the CSDs. In the full sample the re-

Table 5 Industry effects for traded and non-traded goods industries, CSDs. Table shows the pure in traded goods industries. ‘‘Std. Dev.” is the cross-sectional standard deviation of the pure in the industry effect the proportion of the variation in each industry portfolio excess return e the proportion of the variation in each industry portfolio excess return explained by the su median (in parentheses) give the time mean and median of the Std. Dev. and Ratio statist

Mean (Median) Whole sample: 08.84–07.04

Sum of 13 country effects Pure industry effec

Std. Dev. Ratio Std. Dev. R

Non-traded goods 0.375b 0.850 0.067a 0 (0 .416b) (0 .861) (0.061a) (0

Traded goods 0.335b 1.030 0.075a 0 (0.406b) (0.873) (0 .064a) (0

03.86–03.96

Non-traded goods 0.517b 0.850 0.091a 0 (0.570b) (0.862) (0.082a) (0

Traded goods 0.458b 1.029 0.102a 0 (0.557b) (0.877) (0.085a) (0

a Denotes statistics where a larger proportion of return variation is explained by pure b Denotes statistics where a larger proportion of return variation is explained by th

industries.

sults are mostly as expected, with more variation in traded-goods industries explained by pure industry effects, and more variation in non-traded goods industries arising from the sum of the coun- try effects. We find somewhat similar results for the three sepa- rate sub-samples, although there are some anomalies, especially in the first sub-period. There are several potential explanations for these findings which are less favourable to the traded/non- traded goods hypothesis than is the analysis of Griffin and Karolyi (1998) using the mean return decomposition. For example, given the higher proportion of conglomerates in emerging markets, it may be that the classification of a firm as a traded or non-traded goods producer is somewhat less robust than in the major indus-

dustry effect and the sum of the 13 country effects decomposed into traded and non- dustry effect or the sum of the 13 country effects, respectively. The ‘‘Ratio” shows for xplained by the pure industry effect (Eq. (6a)). For the 13 country effects, the ‘‘Ratio” is m of the country effects (Eq. (6b)). These are cross-sectional statistics. The mean and ics, respectively.

08.84–02.86

t Sum of 13 country effects Pure industry effect

atio Std. Dev. Ratio Std. Dev. Ratio

.158a 0.375b 0.850b 0.008 0.158a

.143) (0 .416b) (0 .861b) (0.008) (0.143a)

.331a 0.010b 0.480b 0.007 0.310a

.134) (0. 010b) (0.577b) (0 .006) (0.290a)

04.96–07.04

.157a 0.016b 0.880b 0.010 0.563

.142) (0.014) (0.862b) (.009) (0.547)

.329a 0.014b 0.853b 0.007 0.433

.133) (0.015) (0.861b) (0.007) (0.393)

industry effects in traded goods industries than in non-traded goods industries. e sum of the country effects in non-traded goods industries than in traded goods

Table 6 Time Series Determinants of Country and Industry Effects, CSDs. Table shows the results of estimating Eqs. (8a) and (8b) over the available time series of each calculated country and industry effect (excluding China: see the Appendix). For each explanatory variable, the F tests show the significance of the 12 lags in explaining the country or industry effect. For each country or industry and each variable, the table gives: the degrees of freedom of the F test, the F value, and its probability. Country effect regressions omit one at a time: inflation volatility; output volatility; country effect (CCSD). Industry effect regressions omit one at a time: size volatility; level of concentration; industry effect (ICSD). ‘‘Chow” gives the result of a Chow test on the simplest version of the model determined by the foregoing F tests. The break date is 1996.03.

Countries Industries

Inflation Output CCSD Chow Size Conc. ICSD Chow

Brazil F(12,113) F(12,113) F(12,113) F(37,76) Ind goods F(12,171) F(12,171) F(12,171) F(25,100) F value 10.7907 1.0128 7.8263 3.4955 0.0555 0.0874 33.1714 0.0199 Prob 0.000** 0.442 0.000** 0.000** 1.000 1.000 0.000** 1.000 Chile F(12,131) F(12,131) F(12,131) F(49,70) Con goods F(12,190) F(12,190) F(12,190) F(37,100) F value 2.8024 1.8016 15.1775 4.7586 2.4477 1.3572 37.1568 1.0242 Prob 0.002** 0.054* 0.000** 0.000** 0.006** 0.190 0.000** 0.438 India F(12,133) F(12,133) F(12,133) F(49,72) Con services F(12,190) F(12,190) F(12,190) F(49,100) F value 2.5842 2.4280 20.4351 2.6303 5.8861 3.5103 24.5609 2.7580 Prob 0.004** 0.007** 0.000** 0.000** 0.000** 0.000** 0.000** 0.000**

Israel F(12,172) F(12,172) F(12,172) F(37,135) Basic mats F(12,190) F(12,190) F(12,190) F(25,100) F value 1.2042 1.8632 39.4498 0.6550 4.1397 1.7903 26.9272 0.3216 Prob 0.284 0.042** 0.000** 0.893 0.000** 0.052* 0.000** 1.000 Malaysia F(12,190) F(12,190) F(12,190) F(37,153) Utilities F(12,190) F(12,190) F(12,190) F(25,100) F value 0.7797 2.0157 18.4819 0.2737 15.0179 1.6482 24.5303 4.1040 Prob 0.671 0.025** 0.000** 1.000 0.000** 0.081* 0.000** 0.000**

Mexico F(12,148) F(12,148) F(12,148) F(37,111) Health care F(12,162) F(12,162) F(12,162) F(25,100) F value 1.7355 0.5804 22.8765 0.7885 0.4003 0.7122 19.1833 0.9629 Prob 0.065* 0.855 0.000** 0.751 0.962 0.738 0.000** 0.490 Pakistan F(12,113) F(12,113) F(12,113) F(25,100) Technology F(12,161) F(12,161) F(12,161) F(25,100) F value 0.8175 1.0069 17.1686 2.4415 1.2191 0.6530 58.9510 0.8429 Prob 0.632 0.447 0.000** 0.006** 0.274 0.794 0.000** 0.614 South Africa F(12,190) F(12,190) F(12,190) F(37,153) Telecoms F(12,162) F(12,162) F(12,162) F(25,100) F value 1.9967 0.5802 23.0071 1.1492 0.7546 0.5640 19.3931 0.9024 Prob 0.026** 0.857 0.000** 0.294 0.696 0.868 0.000** 0.552 South Korea F(12,190) F(12,190) F(12,190) F(37,153) Financial F(12,163) F(12,163) F(12,163) F(37,100) F value 0.5392 2.1495 29.6704 0.7317 1.7548 1.2105 24.4667 1.0622 Prob 0.887 0.016** 0.000** 0.820 0.060* 0.280 0.000** 0.393 Taiwan F(12,152) F(12,152) F(12,152) F(25,139) Pers house F(12,173) F(12,173) F(12,173) F(25,100) F value 0.5485 0.6402 15.4665 0.3459 0.4786 0.4450 30.2734 0.3924 Prob 0.880 0.805 0.000** 0.983 0.925 0.943 0.000** 0.971 Thailand F(12,160) F(12,160) F(12,160) F(37,123) Basic Ind F(12,190) F(12,190) F(12,190) F(49,100) F value 1.5663 0.4395 34.1440 1.5464 7.2698 4.3499 53.8825 1.1874 Prob 0.106* 0.945 0.000** 0.059* 0.000** 0.000** 0.000** 0.234 Turkey F(12,148) F(12,148) F(12,148) F value 0.4944 0.9044 0.0504 Prob 0.916 0.544 1.000 12 6 5 11 5 11 6 4 11 2

The number of F tests and Chow test rejections for each variable is shown in the last row of the table. * Significant at the 10% level. ** Significant at the 5% level.

242 Y. Bai, C.J. Green / Journal of Banking & Finance 34 (2010) 236–245

trial countries. See Bai et al. (2008) for further discussion of this point.

Comparing these results for stock risks to the mean return decompositions reported by other researchers, we see a broadly similar pattern of country effects being dominant in the data through the mid-/late-1990s followed by a gradual increase in the importance of industry effects. This is also the case for the mean return decompositions reported by Bai et al. (2008). If total risk does have a similar decomposition to the returns, investors who pursue a diversification strategy based exclusively on the re- turn decomposition may simply be shifting towards securities with greater total risk, without necessarily achieving the desired overall risk reduction from diversification, especially if the markets con- cerned are not fully integrated in world capital markets, as would appear to be the case with the ESMs in our study.

6 Bai, 2008 studies quarterly movements in industry and country components; Campa and Fernandes (2004) studies annual data.

4. Time series determinants of stock risks

Variations in stock risks have implications for portfolio selec- tion, stock pricing and for the economy as a whole. It is therefore important to understand whether the underlying risk factors bear any systematic relationship to variations in other economic vari- ables. Given the decomposition of stock risks among global, coun-

try, industry and idiosyncratic factors, we would expect the systematic components of this decomposition to be related to fun- damental factors in the economy. Thus for example, we might ex- pect the country effect to be related to country-specific factors causing greater or less volatility, and the industry effect to be re- lated to corresponding industry-specific factors. In this section, we investigate these relationships by studying the time series properties of the calculated country (uk) and industry (/j) effects.

We would expect country-specific risks to be related to macro- economic variables reflecting economic policy and the business cy- cle, but also to structural factors determining the general framework within which the economy operates, such as the legal, regulatory and political system (La Porta et al., 1997). Likewise, we would expect industry-specific risks to be related to time series changes in industry characteristics, and also to the structure of the international framework within which the industry operates, such as trading activity and intellectual property rights. Low fre- quency structural determinants of the industry and country com- ponents of stock returns have been studied by Bai (2008) and Campa and Fernandes (2004).6 In this paper, we are concerned with

Table A1 Industry and Country Composition of the Sample: December 2003. Panel A gives the number of stocks included in the sample cross-tabulated by country and industry. Panel B gives the average market capitalization of stocks cross-tabulated by country and industry, and expressed as a percentage of the total sampled market value.

Country Industry

IG CG CS BM UT HC TC TL FI PH BI Total

A: Number of stocks by country and industry Brazil 12 14 5 8 9 1 – 9 2 2 4 66 Chile 7 16 8 6 12 1 – 4 – – 10 64 China 36 23 32 5 7 20 6 3 – 3 9 144 India 33 38 9 42 9 18 9 3 – 5 13 179 Israel 7 4 3 5 – 2 2 1 3 – 2 29 Malaysia 34 82 26 16 8 4 4 5 5 – 53 237 Mexico 1 16 12 2 – – – 2 – 1 4 38 Pakistan 2 14 – 9 5 4 – 1 2 – 7 44 South Africa 16 13 19 3 – 2 6 1 3 – 6 69 South Korea 24 75 18 33 5 14 12 2 7 2 51 243 Taiwan 37 51 6 29 1 – 29 1 – – 24 178 Thailand 16 67 24 21 2 8 6 6 14 4 16 184 Turkey 5 22 3 5 1 2 3 – 3 1 17 62 Sum 230 435 165 184 59 76 77 38 39 18 216 1537

B: Market capitalization of stocks by country and industry Brazil 0.25 4.43 0.34 0.29 1.17 – – 0.99 0.03 0.01 0.02 7.51 Chile 5.45 0.37 0.70 0.60 1.69 0.04 – 0.54 – – 0.23 9.62 China 1.28 0.77 0.95 0.48 1.05 0.66 0.16 0.11 – 0.05 0.28 5.77 India 1.87 1.00 0.26 2.40 0.78 1.42 2.18 0.30 – 1.12 0.58 11.90 Israel 0.17 0.08 0.10 0.28 – 2.01 0.06 0.30 0.30 – 0.08 3.39 Malaysia 0.64 2.05 1.39 0.13 1.99 0.03 0.13 0.96 0.06 – 0.77 8.15 Mexico 0.68 0.68 2.23 0.01 – – – 0.63 – 0.18 0.07 3.81 Pakistan 0.01 0.07 – 0.16 0.29 0.04 – 0.28 0.01 – 0.03 0.89 South Africa 0.74 0.60 1.09 0.54 – 0.14 0.11 0.81 0.17 – 0.23 4.42 South Korea 1.73 3.24 0.51 0.57 1.60 0.11 7.23 1.56 0.43 0.14 0.95 18.08 Taiwan 3.20 2.14 0.21 3.42 0.01 – 10.11 – 0.02 – 0.69 19.78 Thailand 0.47 0.49 0.44 0.89 0.13 0.07 0.41 0.74 0.09 0.08 1.44 5.24 Turkey 0.12 0.80 0.09 0.02 0.06 0.02 0.05 – 0.06 – 0.22 1.43 Sum 15.93 16.70 8.33 9.78 8.76 4.52 20.44 7.22 1.17 1.57 5.59 100

IG, Industrial goods and services; CG, Consumer goods; CS, Consumer services; BM, Basic material; UT, Utilities; HC, Health care; TC, Technology; TL, Telecom; FI, Financials, PH, Personal and household goods; BI, Basic Industrials.

Y. Bai, C.J. Green / Journal of Banking & Finance 34 (2010) 236–245 243

stock risks, and we concentrate instead on higher frequency deter- minants of the time series of stock risks. The conditional volatility of individual stock returns tends to be very persistent (Engle and Patton, 2001). Is this persistence derived from the systematic coun- try and industry effects, as would seem reasonable, and if so, do more ‘‘fundamental” variables also impact on country and industry effects?

We investigate these questions in a preliminary way by using our constructed monthly data to test for one-way Granger-causal- ity of the country and industry effects. Specifically, we estimate the following regressions for each country and industry, respectively:

uk;t ¼ kk þ X12 h¼1

k0;k;huk;t�h þ X12 h¼1

k1;k;hf1;k;t�h þ X12 h¼1

k2;k;hf2;k;t�h þ gk;t

ð8aÞ

/j;t ¼ kj þ X12 h¼1

p0;j;h/j;t�h þ X12 h¼1

p1;j;h g1;j;t�h þ X12 h¼1

p2;j;h g2;j;t�h þ nj;t

ð8bÞ

In Eq. (8a), f1,k is the volatility of country k’s industrial output; f2,k is the volatility of country k’s inflation rate. These are standard sum- mary measures of the volatility of an economy. For the industry ef- fects (8b), size and market power have been documented as influencing share price volatility (Campa and Fernandes, 2004; Roll, 1992), and we therefore use the volatility of industry market value (g1,j) for the size effect, and the Herfindahl index (g2,j) to measure industry concentration. Further details are given in the Appendix. We use 12 lags for each variable in the model to allow for possible occurrence of seasonal effects.

We first tested all the variables for order of integration and found them to be I(0). We therefore estimate (8a) and (8b), and

perform F tests on each of the 3 blocks of variables separately. Where 2 blocks are insignificant, we test them jointly. In addition, given the evidence of a change in the pattern of industry and coun- try effects in 1996, we estimate for each country and industry a simplified version of the model determined by the F tests, and car- ry out Chow tests on these models for a break in 1996.03. The re- sults are shown in Table 6.

These regressions confirm our expectations that country and industry risks are highly persistent for all countries (except Turkey) and all industries. However, we also see that either inflation or out- put volatility, or both, help predict country stock risks in 9 of the 12 countries in the sample, implying that real fluctuations do affect stock market volatility. The Chow tests suggest that the regression parameters have to be treated with caution with 5 countries giving evidence of a break in the relationship in 1996. However, there is no clear regional pattern in the Chow tests, for example of a break in Asian country regressions and no break in non-Asian countries. The industry regressions are somewhat less clear-cut, with size and concentration being significant in just 6 of the 11 industries. The Chow tests mostly suggest that there was no break in these relationships in 1996, but these results could in part be attributable to the relative weakness of industry effects overall in explaining stock risks in the early part of the sample. There are no distinct dif- ferences as between traded and non-traded goods industries.

5. Discussion and conclusions

One of the main benefits from international portfolio diversifi- cation is risk reduction. With progress in financial liberalization in most ESMs, increasing cross-border investment should increase risk-sharing among ESMs, and between ESMs and the rest of the

244 Y. Bai, C.J. Green / Journal of Banking & Finance 34 (2010) 236–245

world. As ESMs become more integrated with other markets, we expect there to be less scope for risk reduction for investors by including ESM companies in their portfolios. However there is cur- rently no clear answer to the question as to how far the integration process has gone. Existing research has focussed on identifying and quantifying the relative importance of country and industry effects in explaining cross-sectional variation in stock returns. It is gener- ally agreed that industry effects have become more important in the last two decades, but there is less agreement on whether this is due to increased integration or other factors such as the global IT bubble. Even so, country effects remain the more dominant force driving the low co-movement of international equity returns, sug- gesting that investors can still gain by diversifying across countries.

The focus of our investigation is on the relative importance of country and industry effects in determining the cross-sectional variation in stock risks, rather than returns. The analyses, including F-tests, risk decompositions, MADs, and sub-sample break-downs, have revealed a broadly similar pattern of results from different perspectives. For the 1986–2004 period as a whole, country effects dominate industry effects, as we would expect in a sample consist- ing exclusively of ESM companies. However, there are interesting variations over time. From 1984 to 1986, industry and country fac- tors both have a significant influence on stock risks but industry factors play a somewhat more important role. We suspect that these initial results may be attributable partly to the relatively small sample of companies in the early part of the data. From 1986 to 1996 country factors play the dominant role in line with earlier results from studies on stock returns. However, from 1996 to 2004, industry effects become more consistently significant, although still remaining less important than the country effects.

The changing time pattern of country and industry effects is consistent with the argument that increasing liberalization in emerging markets will tend to increase industry effects over time. This pattern of results is also broadly consistent with those for re- turn decompositions, except that industry effects on stock risks start becoming important about a year earlier (in 1996) than they do on returns. It may be expected that contagion effects during and after the 1997 Asian financial crisis would increase correlations be- tween firms more subject to global industry shocks than to domes- tic shocks, thus increasing the importance of industry effects.7

However, the one-year lead of industry effects in the risk decompo- sition over that in the return decomposition is suggestive that risk fundamentals may have anticipated to some extent the forthcoming crisis.

Time series analysis of the decompositions underlines the per- sistence of the country and industry components of stock risks. Nevertheless, we also find clear evidence that real factors help pre- dict many of the country and industry time series, especially the former. However, break (Chow) tests on the model are inconclusive as to whether the Asian crisis reflected the ongoing response of stocks risks and prices to changes in real factors or to a discrete change in the responses themselves.

Perhaps the most important finding is that the determinants of stock risks in ESMs follow a broadly similar pattern over time to those of ESM stock returns, apart from the apparent lead in the in- creased industry effect. At first glance this may appear unremark- able given that theory would suggest that stock returns evolve primarily so as to price risk. However, the decomposition in our pa- per is of total risk. A correspondence between returns and total risk is less expected. It suggests that a diversification strategy across countries (for example) based on the return decomposition may

7 Sachs and Warner (1995) argues that one may in fact expect stronger country effects as a consequence of contagion.

simply involve a shift towards securities with greater total risk which in turn could mitigate the expected reduction in covariance risk. Establishing more precisely the relationship between risk and return decompositions is an important subject for future research.

Acknowledgement

The authors would like to thank an anonymous reviewer for many helpful suggestions and comments that greatly improved this paper. We also benefit from comments of participants at the 6th INFINITI Conference on International Finance held in Trinity College Dublin.

Appendix A

A.1. Stock Price data (Sections 2 and 3)

The raw share price data for this study are collected from Data- stream and consist of monthly adjusted prices of 1537 ordinary shares from August 1984 through July 2004. Monthly returns are calculated as log price differences excluding dividends, and ex- pressed in US dollar terms, converting from local currency to US dollars at the ruling monthly exchange rates. The data are derived from 13 countries and are classified into 11 industry sectors, based on the Dow Jones/FTSE Industry Classification Benchmark (ICB) le- vel 3. Table A1 describes industry and country composition of the sample. Further details are available in Bai et al (2008).

A.2. Industry and country effect, time series regressions (Section 4)

Inflation and industrial output data for each country are ex- tracted from the IMF’s International Financial Statistics. Taiwanese data are taken from the National Statistics Bureau website at: http://www.eng.stat.gov.tw/mp.asp?mp=5. Data for China are insufficient; and China is therefore omitted from the calculations. Industry market value and Herfindahl concentration indices are calculated from the raw share price data. The volatility of output, inflation and industry size are estimated as the modulus of devia- tions of the change in output (or rate of inflation, or change in industry market value) from its time mean. The Herfindahl indices are used directly in the regressions.

References

Bai, Y., 2008. The Sources of Cross-Sectional Variation in Stock returns and Risk: An Empirical Analysis of Emerging Markets. Unpublished Ph.D. Dissertation, Loughborough University, Loughborough.

Bai, Y., Green, C.J., Leger, L., 2008. Industry and country factors in stock returns revisited: Are emerging markets different? Working Paper, Loughborough University.

Bekaert, G., Harvey, C.R., Lundblad, C.T., Siegel S., 2008. What segments equity markets? Working Paper, University of North Carolina.

Campa, J., Fernandes, N., 2004. Sources of gains from international portfolio diversification. Journal of Empirical Finance 13, 417–443.

Campbell, J.Y., Lettau, M., Malkiel, B.G., Xu, Y., 2001. Have individual stocks become more volatile? An empirical exploration of idiosyncratic risk. Journal of Finance 56, 1–43.

Cavaglia, S., Brightman, C., Aked, M., 2000. The increasing importance of industry factors. Financial Analysts Journal 56, 41–54.

Chandar, N., Patro, D.K., Yezegel, A., 2009. Crises, contagion and cross-listings. Journal of Banking and Finance 33, 1709–1729.

Chung, K.H., Smith, W.T., Wu, T.L., 2009. Time diversification: Definitions and some closed-form solutions. Journal of Banking and Finance 33, 1101–1111.

Datastream Advance, Version 5, 2009. Thomson Reuters. Engle, R.F., 1982. Autoregressive conditional heteroscedasticity with estimates of

the variance of United Kingdom inflation. Econometrica 50, 987–1008. Engle, R.F., Bollerslev, T., 1986. Modelling the persistence of conditional variances.

Econometric Reviews 5, 1–50. Engle, R.F., Patton, A.J., 2001. What good is a volatility model? Quantitative Finance

1, 237–245.

Y. Bai, C.J. Green / Journal of Banking & Finance 34 (2010) 236–245 245

Griffin, J.M., Karolyi, G.A., 1998. Another look at the role of the industry structure of markets for international diversification strategies. Journal of Financial Economics 50, 351–373.

Grisolia, E., Navone, M., 2007. Cross-industry diversification: Integration or bubble? Working Paper, Bocconi University.

Henry, P.B., 2000. Stock market liberalization, economic reform, and emerging market equity prices. Journal of Finance 55, 529–564.

Heston, S.L., Rouwenhorst, G.K., 1994. Does industrial structure explain the benefits of international diversification? Journal of Financial Economics 36, 3–27.

IMF, 2009. International Financial Statistics, Washington D.C. Kalev, P.S., Nguyen, A.H., Oh, N.Y., 2008. Foreign versus local investors: Who knows

more? Who makes more? Journal of Banking and Finance 32, 2376–2389.

La Porta, R., Lopez-de-Silanes, F., Shleifer, A., Vishny, R.W., 1997. Legal determinants of external finance. Journal of Finance 52, 1131–1150.

Phylaktis, K., Xia, L., 2006. The changing role of industry and country effects in the global equity markets. European Journal of Finance 12, 627–648.

Roll, R., 1992. Industrial structure and comparative behaviour of international stock market indices. Journal of Finance 47, 3–42.

Sachs, J.D., Warner, A., 1995. Economic reform and the process of global integration. Brooking Papers on Economic Activity 1, 1–118.

Serra, A.P., 2000. Country and industry factors in returns: evidence from emerging markets stocks. Emerging Markets Review 1, 127–151.

Silva, A.C.S., Chavez, G.A., 2008. Cross-listing and liquidity in emerging market stocks. Journal of Banking and Finance 32, 420–433.