country and industry factors in returns: evidence from ... · a.p. serra remerging markets re¤iew...

Emerging Markets ReviewŽ .1 2000 127]151

Country and industry factors in returns:evidence from emerging markets’ stocks

Ana Paula SerraU

CEMPRE, Faculdade de Economia do Porto, Portugal

Received 21 February 2000; received in revised form 30 March 2000; accepted 1 May 2000

Abstract

This study examines the influence of country and industry factors on the cross-sectionalvariance and correlation structure of returns. I use new data on emerging markets’ stocksobtained from the Emerging Markets Data Base. I find that emerging markets’ returns aremainly driven by country factors, as it was shown previously in studies for mature markets,and that cross-market correlation is not affected by the industrial composition of the indices.These results have important implications in regard to international portfolio diversification:cross-market diversification seems to be a better bet than cross-industry diversification. Afiner industry partition shows, however, that ignoring the industrial mix leads to animportant loss of diversification benefits. Q 2000 Elsevier Science B.V. All rights reserved.

JEL classification: G15 International Financial Markets

Keywords: International asset pricing; International portfolio diversification; Segmentation; Emergingmarkets

1. Introduction

Evidence shows that market returns tend to be relatively uncorrelated with eachŽ .other e.g. Akhogan, 1995 . Furthermore, and in spite of an increasing globalisation

of the economies and liberalisation of capital markets in recent years, there is only

U Rua Dr Roberto Frias, 4200 Porto, Portugal. Tel.: q351-22-5571100; fax: q351-22-5505050.Ž .E-mail address: [email protected] A.P. Serra .

1566-0141r00r$ - see front matter Q 2000 Elsevier Science B.V. All rights reserved.Ž .PII: S 1 5 6 6 - 0 1 4 1 0 0 0 0 0 0 7 - 8

( )A.P. Serra r Emerging Markets Re¨iew 1 2000 127]151128

Žweak evidence supporting increasing cross-market co-variation in returns e.g..Longuin and Solnik, 1995 .

Previous literature has looked at why cross-country correlation of returns is low.The discussion of this issue is not settled. Some studies claim that the lowcorrelation of returns between countries results from the diverse industrial struc-tures in each country that are mirrored by different industrial composition of theirstock market indices. As industries are imperfectly correlated, equity markets with

Ž .different industry composition will also be imperfectly correlated. Roll 1992claims that industrial factors play a determinant role but Heston and RouwenhorstŽ .1994 show that the influence of pure industry factors is very small. A strongnational market force seems to dominate industry and other stock-specific influ-ences.

Contrary to what happens with the mature markets, little is known about themain factors that drive the structure of returns for emerging markets.1 A lot ofstudies show that the correlation of returns between emerging markets and maturemarkets is low, and that portfolio diversification into emerging markets would have

Žprovided increased returns and lower risks e.g. Errunza and Pabmanabhan, 1988;.Harvey, 1993 . Yet the literature has not examined whether those results are

driven by different industrial compositions of the market indices, or by differentialeconomic and technological development, or by the existence of formal or informalbarriers to foreign investors.

In this paper, I re-evaluate the importance of industry and country specificeffects in explaining the structure of returns for the case of emerging markets’stocks. I look at different industry classifications and whether these results are thesame when I use regions instead of countries.

This study is important because it provides new evidence based on extensiveemerging markets’ data and it has central implications for international assetmanagement. With a sample of 364 weekly series for between 629 stocks in January1990 and 1702 stocks in December 1996, from 26 markets, I show that countryeffects are the most important factors driving the behaviour of emerging markets’stock returns as shown previously for mature markets. Cross-market correlationdoes not seem to be affected by the industrial composition of the indices. Theseresults suggest that cross-market diversification seems to be a better bet thancross-industry diversification. A finer industry partition shows, however, that ig-noring the industrial mix will lead to an important loss of diversification benefits.

The outline of the paper is as follows. Section 2 discusses the alternativeexplanations that have been suggested to account for the low correlation ofreturns. Section 3 describes the data and briefly defines the empirical methodology.Section 4 presents the main findings and discusses the implications of the results.Section 5 concludes.

1 Ž . Ž . Ž .Exceptions are Divecha et al. 1992 , Claessens et al. 1998 , Fama and French 1998 andŽ .Rouwenhorst 1999 .

( )A.P. Serra r Emerging Markets Re¨iew 1 2000 127]151 129

2. Understanding the correlation of returns

Many studies have studied the correlation of returns within emerging marketsŽ .and with mature markets, in detail e.g. Bekaert, 1993; Bekaert and Harvey, 1995 .

The main findings are that:

v Correlation of returns across countries is low for emerging markets andbetween these and mature markets.

v Correlation coefficients do not seem to be stable but there is only weakevidence supporting an increasing trend.

Table 1 shows the correlation coefficients of weekly returns for the IFC emerg-ing markets’ indices and with the FTrS & P-A World index for the period fromJanuary 1990 to December 1996.

The average weekly dollar return pair-wise correlation for 26 emerging marketsw x 2is 0.07 and can be found in the range y0.15, 0.52 . There seems to be present

some regional effect but not as strong as one would expect given the increasingeconomic links in some geographical areas. The average correlation with the world

w xindex is 0.14 and the coefficients are in the range y0.13, 0.45 . Within a selectedset of mature markets, the average weekly correlation is 0.43.3 The averagecorrelation with the world index is 0.62 and correlation coefficients vary between0.40 and 0.94.

Why is cross-market correlation so low?

2.1. Industrial composition of market indices

The first explanation says that cross-market correlation is low because of the waymarket indices are constructed. Given that the choice of the market indices’constituents is usually based on some criteria such as market capitalisation or valuetraded, it turns out that some indices are concentrated in a few firms. From 1992 to1996, the IFC Global indices represented between 40% and 70% of the underlyingmarket capitalisation. However, this coverage drops to between 2% and 50% if welook at the number of firms. Thus, it is true that for some markets, IFC indicesfocus on some large firms and therefore concentrate on a few sectors. It should bereminded, however, that one of the IFC criteria to select stocks to integrate itsindices, is industrial comprehensiveness. On the other hand, the fact that indicesare concentrated on large stocks does not mean that correlation would be higher ifall stocks were included. On the contrary, increasing the number of firms wouldlead to increasing the number of small stocks that usually carry more idiosyncraticrisk, and it is likely that this enlargement would result in even lower correlation.

2 Ž .Heston and Rouwenhorst 1994 report an average sample correlation of monthly returns of 0.41 for12 European markets over the period 1978]1982.

3These figures refer to the following mature markets: France, Germany, Italy, Japan, UK and USA. Iuse FTrS & P-A indices.

()

A.P

.Serrar

Em

ergingM

arketsR

e¨iew1

2000127

]151130

Table 1aMarket indices’ correlation

Mean S.D. Comp Arg Bra Chi Col Gre Ind Ind Jord Kor Mal Mex Phi Por

Composite 0.0006 0.0224 1.00Argentina 0.0047 0.0678 0.19 1.00Brazil 0.0030 0.0758 0.40 0.15 1.00Chile 0.0045 0.0296 0.25 0.17 0.19 1.00Colombia 0.0058 0.0387 0.10 0.04 0.09 0.02 1.00Greece 0.0013 0.0412 0.16 0.02 0.08 0.09 0.09 1.00India 0.0010 0.0433 0.10 0.01 0.01 0.00 0.01 0.08 1.00Indonesia 0.0007 0.0325 0.36 0.04 0.09 0.06 0.02 0.12 0.09 1.00Jordan 0.0015 0.0217 0.14 0.05 0.00 y0.02 0.05 0.05 0.05 0.05 1.00Korea y0.0017 0.0339 0.37 0.06 0.01 0.11 y0.07 y0.06 0.00 0.06 0.05 1.00Malaysia 0.0025 0.0292 0.49 0.00 0.06 0.04 0.02 0.17 0.05 0.38 0.12 0.18 1.00Mexico 0.0024 0.0410 0.44 0.31 0.24 0.21 0.02 0.09 y0.03 0.10 y0.02 0.13 0.21 1.00Phil 0.0020 0.0366 0.40 0.09 0.06 0.12 0.11 0.14 0.01 0.37 0.05 0.01 0.41 0.21 1.00Portugal 0.0005 0.0248 0.34 0.08 0.20 0.11 0.12 0.38 0.07 0.09 0.08 0.10 0.27 0.16 0.18 1.00Taiwan y0.0008 0.0531 0.74 0.00 0.08 0.08 0.06 0.08 y0.05 0.14 0.14 0.09 0.23 0.14 0.28 0.16Thailand 0.0010 0.0422 0.48 0.12 0.10 0.07 0.08 0.12 0.10 0.30 0.17 0.16 0.52 0.17 0.34 0.28Turkey y0.0013 0.0776 0.20 y0.04 0.07 y0.01 0.10 0.23 0.07 0.12 0.09 y0.01 0.17 y0.01 0.17 0.17Venezuela 0.0053 0.0668 y0.06 0.07 y0.04 y0.02 0.03 y0.05 0.03 y0.04 y0.02 y0.01 y0.07 y0.01 y0.03 0.00China 0.0002 0.0746 0.19 0.07 0.15 0.06 0.02 0.15 0.13 0.16 0.00 0.04 0.15 0.05 0.01 0.11Hungary 0.0020 0.0369 0.15 0.15 0.14 0.22 0.12 0.17 0.15 0.07 0.00 0.06 0.00 0.12 0.09 0.21Pakistan 0.0020 0.0344 0.08 y0.02 y0.01 y0.03 0.16 0.06 y0.03 0.09 0.05 y0.02 0.07 0.04 0.13 0.06Peru 0.0041 0.0422 0.36 0.43 0.28 0.38 0.12 0.01 y0.04 0.10 0.03 0.11 0.03 0.31 0.08 0.09Poland 0.0099 0.0800 0.13 0.12 0.08 0.01 0.09 0.08 y0.04 0.15 0.14 y0.01 y0.01 0.14 0.01 0.18SAfrica 0.0040 0.0316 0.30 0.22 0.05 0.20 y0.09 0.21 0.00 0.13 0.11 0.24 0.23 0.06 0.12 0.14Sri Lanka y0.0005 0.0341 y0.01 y0.05 y0.09 y0.02 0.11 0.03 0.04 0.00 0.00 0.05 y0.02 y0.07 y0.01 0.11Nigeria 0.0064 0.1088 0.01 y0.15 0.16 0.08 0.05 0.03 0.08 0.02 0.00 0.12 y0.17 y0.10 y0.13 y0.08Zimbabwe 0.0097 0.0338 0.11 0.06 0.02 0.07 0.16 y0.07 0.13 0.11 y0.02 0.05 0.12 0.11 0.15 0.08World 0.0014 0.0163 0.45 0.15 0.20 0.06 0.03 0.26 y0.04 0.05 0.09 0.21 0.45 0.30 0.22 0.43

()

A.P

.Serrar

Em

ergingM

arketsR

e¨iew1

2000127

]151131

Ž .Table 1 Continued

Mean S.D. Comp Arg Bra Chi Col Gre Ind Ind Jord Kor Mal Mex Phi Por

Composite 0.0006 0.0224 1.00Argentina 0.0047 0.0678 0.19Brazil 0.0030 0.0758 0.40Chile 0.0045 0.0296 0.25Colombia 0.0058 0.0387 0.10Greece 0.0013 0.0412 0.16India 0.0010 0.0433 0.10Indonesia 0.0007 0.0325 0.36Jordan 0.0015 0.0217 0.14Korea y0.0017 0.0339 0.37Malaysia 0.0025 0.0292 0.49Mexico 0.0024 0.0410 0.44Phil 0.0020 0.0366 0.40Portugal 0.0005 0.0248 0.34Taiwan y0.0008 0.0531 0.74 1.00Thailand 0.0010 0.0422 0.48 0.22 1.00Turkey y0.0013 0.0776 0.20 0.10 0.16 1.00Venezuela 0.0053 0.0668 y0.06 y0.08 y0.02 0.07 1.00China 0.0002 0.0746 0.19 0.00 0.09 y0.01 0.04 1.00Hungary 0.0020 0.0369 0.15 y0.10 0.04 0.01 0.05 y0.03 1.00Pakistan 0.0020 0.0344 0.08 0.02 0.10 0.10 0.08 y0.04 0.00 1.00Peru 0.0041 0.0422 0.36 0.11 0.15 0.06 0.09 0.06 0.18 y0.02 1.00Poland 0.0099 0.0800 0.13 0.00 0.15 0.05 0.00 y0.01 0.20 0.13 0.00 1.00SAfrica 0.0040 0.0316 0.30 0.08 0.16 0.11 0.10 0.08 0.14 0.14 0.15 0.14 1.00Sri Lanka y0.0005 0.0341 y0.01 y0.01 0.05 0.08 0.03 0.02 0.18 0.17 y0.03 0.15 y0.12 1.00Nigeria 0.0064 0.1088 0.01 0.00 y0.11 0.01 y0.01 y0.04 0.06 0.03 0.10 0.01 0.01 0.05 1.00Zimbabwe 0.0097 0.0338 0.11 y0.10 0.07 y0.15 0.01 y0.04 0.15 0.02 0.15 0.23 0.07 0.22 y0.06 1.00World 0.0014 0.0163 0.45 0.27 0.32 0.17 y0.04 y0.01 0.17 0.02 0.26 0.08 0.25 y0.06 y0.13 0.05

a Ž .This table shows the correlation between US $ total returns’ indices weekly data, January 1990]December 1996 . Markets listed above China haveŽ .complete series. The sample data were obtained from Emerging Markets Data Base EMDB , International Finance Corporation, World Bank. FT & SP-A

indices were obtained from Datastream.


The previous argument is related to another position: cross-market correlation islow because it reflects the differential industrial composition of the market indices.

Ž . Ž .On the assumptions that 1 firms within a sector are reasonably correlated; and 2that industry cross-correlation is imperfect, then markets with different industrialstructures should correlate imperfectly.

I have investigated how different the industrial compositions of IFC emergingmarket indices are. In 1996, for 20 out of the 26 markets, the indices were

Ž .concentrated in two or three sectors out of the nine broad sectors that accountedfor more than 80% of the total market capitalisation. Similarly, some industryindices are completely dominated by one market and there are several industryindices that do not have constituents from all markets.

I have examined whether it was possible to establish a simple relation betweenthe diversity in industrial structures of any two-market indices and the correlationof returns. I find that the more two indices differ, the lower the pair-wisecorrelation and the average difference is statistically significant.4

The comparison of the structure of these indices for emerging markets and theone observed for European markets as reported in Heston and RouwenhorstŽ .1994 indicates, a priori, that a strategy of investing domestically in emergingmarkets would provide lower diversification benefits. Emerging markets are moreconcentrated and the diversification potential for investing internationally withinone sector is also more limited. Table 2 shows the summary statistics for the IFCindustry indices5. Standard deviations of the industry indices are not very far fromthe IFC Composite Index, suggesting that there are important geographical diversi-fication benefits regardless of the sector elected for diversification.

To find out more about the importance of the different industrial composition ofthe indices in correlation of returns, I first looked at the correlation of returnsbetween portfolios of stocks in the same industry across markets. Cross-countryreturns within a particular industry never correlate more than market indices. Onthe contrary, for some industries, the correlation is even lower, suggesting that it is

Žbetter to diversify across countries within a particular industry for example,.Agriculture or Services than to adopt a traditional geographical market index

diversification strategy. Two reasons could justify this low co-movement: industriesŽ .are not truly global or the broad sector classification SIC one-digit shadows the

true relations by aggregating many industries that bear no relation with each other.Ž .Some industries are expected to be more global for example, paper than others

Ž .for example, real estate but the correlation here is always low, whatever theindustry considered.6

4 To measure the diversity I have used a simple sum of the absolute differences between the weightsof each industry in the market capitalisation of the indices by the end of 1996.

5I only present here the results of the indices for the nine SIC one-digit sectors because the statisticsat a two-digit level are difficult to analyse, given that the sample covers more than 60 groupings.

6 The underlying assumption when one computes an industry index is that markets are integrated. Ifmarkets were not integrated then international industry factors would mean very little and industry

Ž .effects could no longer be considered as common global factors.

()

A.P

.Serrar

Em

ergingM

arketsR

e¨iew1

2000127

]151133

Table 2aIndustry indices’ correlation

Mean S.D. Agricult Mining Constr Manuf TranpUt Trade Fi,In,RE Services Other

Agriculture 0.0000 0.0272 1.00Mining y0.0009 0.0374 0.41 1.00Construction y0.0016 0.0293 0.47 0.40 1.00Manufacturing y0.0017 0.0206 0.60 0.56 0.56 1.00Transport, utilities 0.0012 0.0233 0.45 0.48 0.49 0.66 1.00Wholesale retail trade y0.0011 0.0248 0.41 0.41 0.68 0.66 0.53 1.00Finance, insurance, RE y0.0017 0.0315 0.62 0.52 0.75 0.73 0.62 0.72 1.00Services 0.0007 0.0300 0.59 0.45 0.60 0.58 0.43 0.53 0.73 1.00Other 0.0005 0.0273 0.68 0.48 0.59 0.62 0.59 0.54 0.71 0.69 1.00

a Ž .This table shows the correlation coefficients between IFC industry indices returns weekly data for the period of 1990]1996 . The sample data wereŽ .obtained from Emerging Markets Data Base EMDB , International Finance Corporation, World Bank.


Second, I looked at the cross-industry correlation at the emerging marketsaggregate level.7 Table 2 shows that cross-industry correlation is always higher thancross-market correlation. The average weekly dollar return correlation acrosssectors is 0.54 against the 0.07 obtained across markets.8,9 On an individual countrybasis, that is free from any potential geographical bias, I find that the cross-industry correlation is also very high, suggesting that a common national factor isdriving the returns.10 This preliminary result suggests that strategies recommend-ing geographical diversification either by investing on aggregate indices or within asingle sector leave an investor better off than strategies that bet on diversification

Ž .across sectors within only one market. Griffin and Karolyi 1998 suggest than oneshould look at a narrower industry classification so that the information about thecross-sectional variation of returns due to industry effects is not lost. When I use a

Ž . Žtwo-digit SIC industry classification 60 categories against one-digit nine cate-.gories , the average cross-industry correlation across industries drops from 0.54 to

0.18. This result can be misleading given that some classifications have very fewconstituents.11

Even if a market index reflects the stock market industrial structure, that indexdoes not reflect the country industrial structure. Market capitalisation can repre-sent as little as 10% of GNP for some of these developing countries. Especially forthose countries whose stock markets have recently emerged following a process ofprivatisation, the firms that are listed on the stock exchange are not always arepresentative sample of the industrial structure of that country. One would expectthat the stock markets in countries that have similar underlying industrial struc-tures move together; one may not observe that, simply because only a subset offirms are publicly traded on the stock market. Arguing that the correlation ofreturns reflects the true underlying economic structures is thus a different argu-ment and leads us to the third suggested explanation.

2.2. Real economic integration

A simple discounted cash flow model can be used to identify the links betweenmacroeconomic variables and stock returns. Economic variables influence stockreturns through the underlying cash flows or the discount rates. If the relevanteconomic forces are international then they should simultaneously affect all equity

7These are home-made equally and value-weighted portfolios based on the stocks in my sample.8 Ž .Heston and Rouwenhorst 1994 report average sample correlations across broad industry sectors of

0.71.9 Ž .Meric and Meric 1989 also find that international industry portfolios returns, based on 17

countries, are more closely correlated than national stock markets returns.10 Statistics and correlation matrices of returns across countries within a particular sector and across

sectors within a particular market were not included here due to space limitations.1117 out of the 60 two-digit SIC industry portfolios in my sample comprise five or less firms.


Ž .returns across the world global shocks . However, if the relevant economic forcesare mainly domestic or regional, then the correlation of returns across countrieswill only be high if business cycles move in tandem. Previous research has foundthat correlation is high when both countries are in a common recession, but it is

Žlow when they are recovering or when business cycles are out of phase e.g. Solnik. 12et al., 1996 .

In recent years, most economies became more open and integrated especially atŽ .a regional level through economic and monetary unions . Given the globalisation

of the economy and of some industries in particular, we would expect that, moreand more, shocks would be global, i.e. that they affect the world economy globallyand every country business cycle. If real integration is the main determinant ofcorrelation, then we should be already observing higher correlation of returns,especially at a regional level, and it should rise even more in the future. Yet theevidence supporting a rise in correlation of returns in recent years is weak.

2.3. Integration of capital markets

The last explanation does not exclude the validity of the previous arguments butfocuses on financial assets as a separate and additional cause for the level ofcorrelation of returns across countries.13 If markets are segmented the rule, toprice andror the price of risk are different across countries. Markets can besegmented because of formal or informal barriers that preclude free investmentworld-wide. Even if the underlying economics are linked, returns may not moveclosely because stock prices are established in separate worlds. Previous researchhas shown that correlation between stock returns increases when markets becomemore integrated as barriers fall. Foreign investors influence leads to common

Ž .priced factors and common risk premiums. Bekaert and Harvey 2000 find weakevidence that the liberalisation of stock markets, the introduction of country funds,and the dual listing of local shares on international exchanges impact positively onthe correlation of returns between those markets and the world market.

In sum, a first examination of the three main explanations that have beenproposed to understand the low cross-market correlation of returns is not conclu-sive. It is true that market indices reflect different industrial compositions. Yetindustry indices are more closely correlated than market indices, and cross-marketcorrelation within a particular industry is low. The increase in correlation in recentyears is also not such as economic integration and the fall of investment barrierswould suggest. The analysis that follows provides further evidence on the impor-tance of these arguments.

12 Ž .Ammer and Mei 1996 show that the correlation between two countries’ stock returns is substan-tially greater than the correlation of measures of real output growth.

13 This does not mean that the reverse is true. Two markets could be highly correlated}as is to beexpected if there is real economic integration}and yet their financial markets could still be segmented.


3. Data and methodology

3.1. Sources

Ž .The main source of my data is the Emerging Markets Data Base EMDB . I usetwo modules: the stock series and the index series, on a weekly basis, from the

Ž .beginning of 1990 to the end of 1996. I use Friday to Friday total continuous logreturns.14

High frequency data could introduce a downward bias in contemporaneouscorrelation because of time differences between markets and lags in informationtransmission. I use weekly data because of sample size and data availability.

My analysis focuses on US returns but I dedicate one separate section to localreturns, when I investigate the effects associated with exchange rates.

ŽFirms are assigned to one of the SIC broad industry categories one and two. Ždigits as in the IFC database. Previous research has used FT Actuaries sectors see

. Ž .Roll, 1992 or the Dow Jones industry see Griffin and Karolyi, 1998 classifica-tions.

3.2. Sample description

My sample excludes some firms that had missing or meaningless data for prices.I have also excluded those firms originating from emerging markets whose cover-age started after 1993. Finally, I have also excluded all firms that seemed to havean important thin trading problem: I have removed all the firms that did not show

Ž .any price changes zero returns for 10 consecutive weeks or more.Overall there are between 629 stocks in 1990 and 1702 stocks in 1996 from 26

emerging markets.15,16

The sample period consists of 364 weeks and includes the 8 months of theŽ .Kuwait invasion starting 2 August 1990 and the Mexican crisis in December

1993.17 A longer period would allow finding out if the results are driven by thisspecific sample period and to what extend the correlation of returns are constantover time. I perform some tests and discuss this issue further below.

14 Ž .See ‘The IFC Indices}Methodology, Definitions and Practices’ 1996 for details on the computa-tion.

15 Ž .Grinold et al. 1989 use monthly data from 1983 to 1988 for 24 countries in a total of 2454Ž . Ž . Ž . Ž .securities including Malaysia 36 , Mexico 13 and South Africa 69 . Heston and Rouwenhorst 1994

use monthly data from 1978 to 1992, covering 829 firms in 12 European countries. Griffin and KarolyiŽ .1998 use weekly data from 25 countries covering 2400 stocks over the period December 1991 to April

Ž . Ž .1995. This latter sample includes stocks from 4 emerging markets: Mexico 31 ; Thailand 70 ; IndonesiaŽ . Ž .32 and Malaysia 76 . Appendix A describes my sample market coverage.

16 My sample ignores ownership restrictions and includes all classes of shares. In terms of equallyweighted portfolios, this procedure necessary results in giving more weight to those stocks that havemore than one class of shares.

17 I have checked the influence of these observations by repeating the analysis on the series excludingthe ‘crises’ observations. My results are robust to this procedure.


3.3. Methodology

The first step is to obtain ‘pure’ country and ‘pure’ industry factors. Manystudies proxy these factors with aggregate indices but these indices are not adjusted

Žfor the differential industrial composition across markets or for the differential.geographical composition across industries . I decompose individual stock returns

Ž .into industry and country components using the procedure in Grinold et al. 1989Ž .and used by Heston and Rouwenhorst 1994 . I run a cross-sectional regression of

individual security returns on industry and country dummies for each month, andobtain a time series of estimated ‘pure’ industry and country effects. I thenmeasure how these effects account for the variation of emerging markets’ aggre-gate returns. Please refer to Appendix A for a brief summary of the methodology.

4. Empirical results

4.1. Country and industry effects

Table 3 shows the time-series variances of the several components for each ofthe marketrindustry equally weighted index, in excess of the emerging markets’

Ž .average. That same decomposition for the value-weighted VW indices, is avail-able upon request. The variance of an excess return market index can be decom-posed into the sum of the effects of its constituent industries and the ‘pure’ countryeffect. Similarly, the variance of an excess return industry index can be decom-posed into the sum of the effects of its constituent markets and the ‘pure’ industryeffect.18

There are several results of interest. First of all, the country effects account foralmost all the variance of market indices’ returns. Indeed, the average variance ofthe sum of the constituent industries’ effects is, on average, only 0.7% of the

Ž . Ž .variance of the excess equally value weighted EW market indices’ returns. Theratio has a maximum for the Philippines, 7%, followed by South Africa with 2%.

Ž .For the value-weighted VW indices, the average ratio is 1.1%. Heston andŽ .Rouwenhorst 1994 } HR from here on } found 0.6% and 7%, respectively, for

equally and value-weighted indices.The relatively small variability of the industry pure effects compared with the

market pure effects explains these results. On average, the variance of pureŽ .industry effects is 1.3 1.9 for VW percent-squared which is much smaller than the

Ž .average weekly variance of pure market effects, 24 22 for VW percent-squared.Ž .This is a ratio of 18:1 12:1 for VW making clear why pure country effects

dominate market indices. HR found pure industry monthly variances of 5.4 and 6.5percent-squared, respectively, for EW and VW industry indices. For the average

18 The effects do not sum to one because country and industry effects are not uncorrelated.

()

A.P

.Serrar

Em

ergingM

arketsR

eïew1

2000127

]151138

Table 3aŽ .Variance decomposition of country and industry EW indices

b cEM average Pure country Sum of industry effectsI. Country indicesCountry Variance Ratio to Variance Ratio to market Variance Ratio to market

market index in excess index in excessindex of EM average of EM average

Argentina 0.0589 0.0048 0.9969 0.0000 0.0010Brazil 0.0652 0.0038 0.9998 0.0000 0.0005Chile 0.3920 0.0009 1.0229 0.0000 0.0152Colombia 0.2368 0.0014 1.0030 0.0000 0.0011Greece 0.2107 0.0013 1.0047 0.0000 0.0037India 0.3059 0.0006 1.0050 0.0000 0.0023Indonesia 0.1688 0.0017 1.0012 0.0000 0.0050Jordan 0.5069 0.0007 1.0226 0.0000 0.0043Korea 0.2561 0.0011 0.9975 0.0000 0.0035Malaysia 0.2599 0.0008 0.9648 0.0000 0.0117Mexico 0.1937 0.0014 0.9991 0.0000 0.0018Philippines 0.2650 0.0009 0.9690 0.0001 0.0657Portugal 0.4433 0.0006 1.0183 0.0000 0.0055Taiwan 0.1082 0.0022 1.0026 0.0000 0.0006Thailand 0.1669 0.0013 0.9934 0.0000 0.0039Turkey 0.0506 0.0055 1.0045 0.0000 0.0007Venezuela 0.0757 0.0044 1.0065 0.0000 0.0006

( )Cross-country aërage a17 0.0003 0.2214 0.0020 1.0007 0.0000 0.0075( )Cross-country median a17 0.2107 0.0013 1.0026 0.0000 0.0035

China 0.0379 0.0045 1.0007 0.0000 0.0003Hungary 0.1354 0.0016 1.0046 0.0000 0.0039Pakistan 0.2178 0.0009 1.0018 0.0000 0.0016Peru 0.1092 0.0018 1.0167 0.0000 0.0087Poland 0.0337 0.0058 0.9979 0.0000 0.0007South Africa 0.2154 0.0008 0.9725 0.0000 0.0202Sri Lanka 0.1797 0.0012 0.9999 0.0000 0.0026Nigeria 0.0187 0.0117 1.0034 0.0000 0.0004Zimbabwe 0.1995 0.0012 1.0094 0.0000 0.0057

( )Cross-country aërage a26 0.0003 0.1889 0.0024 1.0007 0.0000 0.0066( )Cross-country median a26 0.1867 0.0013 1.0220 0.0000 0.0031


Ž.

Tab

le3

Con

tinue

d

de

EM

aver

age

Pur

ein

dust

rySu

mof

geog

raph

ical

effe

cts

II.I

ndus

try

indi

ces

Indu

stry

Var

ianc

eR

atio

toV

aria

nce

Rat

ioto

indu

stry

Var

ianc

eR

atio

toin

dust

ryin

dust

ryin

dex

inex

cess

inde

xin

exce

ssin

dex

ofE

Mav

erag

eof

EM

aver

age

Agr

icul

ture

0.57

770.

0002

0.60

180.

0001

0.47

94M

inin

g0.

4488

0.00

030.

7240

0.00

020.

5191

Con

stru

ctio

n0.

4585

0.00

010.

4169

0.00

020.

5076

Man

ufac

turi

ng0.

8913

0.00

000.

3247

0.00

000.

6487

Tra

nspo

rt,U

tilit

ies

0.62

320.

0001

0.65

580.

0001

0.43

41W

hole

sale

,Ret

ailT

rade

0.74

110.

0001

0.58

930.

0001

0.48

74F

inan

ce,I

nsur

ance

,RE

0.79

860.

0001

0.56

790.

0000

0.52

91Se

rvic

es0.

4242

0.00

020.

4658

0.00

020.

5286

Oth

er0.

4715

0.00

010.

3013

0.00

020.

6443

Cro

ss-i

ndus

try

a¨er

age

0.00

030.

6039

0.00

010.

5164

0.00

010.

5309

Cro

ss-i

ndus

try

med

ian

0.57

770.

0001

0.56

790.

0001

0.51

91

aŽ

Thi

sta

ble

give

sth

eva

rian

ceof

glob

al,

coun

try

and

indu

stry

effe

cts

for

the

equa

llyw

eigh

ted

coun

try

and

indu

stry

indi

ces

wee

kly

data

,Ja

nuar

y19

90or

late

rto

.D

ecem

ber

1996

.M

arke

tslis

ted

abov

eC

hina

have

com

plet

ese

ries

.E

ach

coun

try

inde

xre

turn

isde

com

pose

din

aco

mm

onef

fect

plus

aco

untr

y‘p

ure’

effe

ctan

da

sum

ofth

eco

nsti

tuen

tin

dust

ryef

fect

s.E

ach

indu

stry

inde

xre

turn

isde

com

pose

din

aco

mm

onef

fect

plus

anin

dust

ry‘p

ure’

effe

ctan

da

sum

ofth

eco

nsti

tuen

tco

untr

yef

fect

s.C

ount

ryan

din

dust

ryef

fect

sar

eth

edi

ffer

enti

alef

fect

sre

lati

veto

the

emer

ging

mar

kets

’ave

rage

.The

rati

osgi

veth

epr

opor

tion

ofth

eva

rian

ceof

apa

rtic

ular

com

pone

ntre

lati

veto

the

vari

ance

ofth

ein

dex

retu

rn,r

awor

inex

cess

ofth

eem

ergi

ngm

arke

ts’a

vera

ge.A

llre

turn

sar

em

easu

red

inU

Sdo

llars

per

wee

k.b A

vera

gere

turn

offi

rms

ina

coun

try

inex

cess

ofth

eem

ergi

ngm

arke

ts’

aver

age

afte

rad

just

ing

for

its

indu

stri

alco

mpo

siti

on.

c Com

pone

ntof

aco

untr

yin

dex

that

isdu

eto

adi

ffer

enti

alin

dust

rial

com

posi

tion

inre

lati

onto

the

emer

ging

mar

kets

’av

erag

e.d A

vera

gere

turn

offi

rms

inan

indu

stry

inex

cess

ofth

eem

ergi

ngm

arke

ts’

aver

age

afte

rad

just

ing

for

its

geog

raph

ical

com

posi

tion

.e C

ompo

nent

ofan

indu

stry

inde

xth

atis

due

toa

diff

eren

tial

geog

raph

ical

com

posi

tion

inre

lati

onto

the

emer

ging

mar

kets

’av

erag

e.


Žvariance of pure market effects, HR found a monthly 24 percent-squared both for. 19EW and VW indices .

Second, Table 3 shows that, on average, the pure industry effects account forŽ .52% 49% for VW of the industry indices variance. In HR, the pure industry

effects accounted for a more important part of the variance of the industry indicesŽ .91% and 89% for EW and VW indices . My results show that, geographical effectsthus account for a more important stake of the industry indices’ variation than inprevious studies and this result stems from larger variances in the returns ofemerging markets’ indices. Most of these industry indices could be rich diversifica-tion tools for diversifying internationally as a large part of its variation is due tospecific country effects.

Third, I looked at the intercept of the cross-sectional regressions over the sampleperiod. The intercept series are, by construction, the emerging markets’ equally orvalue-weighted index with the constituents in sample. The variance of the interceptrelative to the total variance of the indices gives an idea of the importance of acommon factor among emerging markets. This aspect will be further discussedbelow when I look at the results by region and within region. The first two columns

Žin Table 3 show that effect. The common factor variance 0.0003 and 0.0005,. Ž .respectively, for EW and VW indices accounts, on average, for 19% 37% for VWŽof the average weekly index variance that are 0.0026 and 0.0027, respectively, for

.EW and VW . For equally weighted indices, this factor has a maximum for Jordan,with 50% and a minimum of 5% for Turkey. These figures suggest that a commonfactor exists, is important, but small. These numbers are mirrored in the lowcorrelation within emerging markets, highlighted in the previous section. For theindustry equally weighted indices, the common effect accounts for as much as 60%of industry indices variance, reflecting simply, as remarked before, that the vari-ances of industry indices are small. In HR, the intercept variance represented, onaverage, as much as 73% for the EW market indices and 94% for the EW industryindices.

When we look at the fit of the cross-sectional regressions of these estimates, wecan see that, as a whole, the common factor and the country and industry factorsaccount for as much as 38% of the variability of the time-series cross-sectional

20 2 Ž 2 . Ž .returns. The median R adjusted R over the 364 week period is 32% 30% .The fit of the regression gives us an idea of how important other omitted commonrisk factors could be. No special pattern is observed over time.21 The F tests on theglobal significance of the regression are always significant and the same is true forthe joint tests on the subset of country effects’ coefficients. The common factor }

19 Ž .In annualised terms, the emerging markets’ indices in my sample show an average medianŽ .standard deviation of 35% 26% against 17% for HR.

20 Ž .The EP Explanatory Power statistic is obtained as one minus the ratio of the sum of errors for allthe weekly cross-sectional regressions divided by the sum of total returns variation, again for all theweekly regressions.

21 The maximum R2 occurs in April 1995 and the minimum in April 1991.


captured by the intercept } is significant 291 times out of 364. The subset ofindustry effects’ coefficients is only significant around one third of the times.

For value-weighted indices, the fit is even better. The explanatory power is 46%,the median adjusted R2 is 41% and both the common factor and the set of industryeffects’ parameters are almost always significant.

The inference using the Generalised Method of Moments, to overcome het-eroscedasticity, produces very similar results.

The correlation matrix can now be computed using the estimated market returnsadjusted for their industrial structure and compared to the statistics provided forraw market indices’ returns in Table 1. As discussed previously, if industrycomposition were important in explaining the indices’ return behaviour, then theadjusted indices would now show more uniform mean returns and standarddeviations, and cross-market correlations for adjusted returns should be higher. Asin HR, I find that the differences between the statistics for raw and adjustedindices are tiny.22 Countries that had high volatility continue to do so and theaverage cross-market correlation does not increase. The average cross-market

Ž .correlation is now 0.06 0.07 for VW indices against 0.07 before and the averageŽ . 23correlation with the World index drops from 0.14 to 0.12 0.15 for VW indices .

For industry indices, the adjustment for geographical effects is not very impor-tant, but mean returns are different for some industries and overall the adjustedindustry indices show lower standard deviations. There is an important increase inthe average cross-industry correlation, from 0.54 to 0.72. Thus, these resultsconfirm the diagnosis that came out from the preliminary analysis: cross-industrycorrelation is on an aggregate basis, higher than cross-market correlation. Thisresult could stem from the fact that the SIC-one-digit industry classification is verywide.

To give a better idea of the diversification benefits of different investmentstrategies suggested by these results, I have compared the variance of fourdifferent portfolios. The first one is here identified as ‘No Diversification’ and itsvariance is obtained as the average of the individual stocks’ variance:

N2

sÝ iis12 Ž .s s 1ND N

The second portfolio is a global portfolio that invests in all the emergingmarkets’ firms. The variance of the equally weighted portfolio provides the vari-

22 Ž . Ž .In HR, the average correlation of the adjusted equally value weighted market indices is 0.74 0.77Ž . Ž .against 0.71 0.76 for raw returns. The average correlation of the adjusted equally value weighted

Ž . Ž .industry indices is 0.42 0.40 against 0.41 0.43 for raw returns.23 The world index is not adjusted. Therefore, there remains here a differential between the industrial

structure of the world index and that of the average emerging market.


ance estimate of this global portfolio and it is here identified as ‘MaximumDiversification’.

2 2 Ž .s s s 2MD EW

The two other portfolios involve:

v diversifying across industries within a country and it is here identified as‘Industry Diversification’. The variance of this portfolio is given by the weightedaverage of the variances of the 26 emerging markets’ indices:

K mk2 2 Ž .s s s 3ÝID kMks1

where the weights are given by the average ratio of the number of companies in amarket relative to the total number of firms in the sample.24

v diversifying across countries within an industry and it is here identified as‘Geographical Diversification’. The variance of this portfolio is given by theweighted average of the variances of the nine emerging markets’ industryindices:

L nl2 2 Ž .s s s 4ÝGD lMls1

where the weights are given by the average ratio of the number of companies in anindustry relative to the total number of firms in the sample. Please refer toAppendix B for notation.

Table 4 reports the results and relates the variance of the latter three portfoliosrelative to the scenario of ‘No Diversification’. The scenario of ‘Maximum Diversi-fication’ indicates that potentially, risk could be eliminated to 6% of the averageindividual risk. In practice, it would be very difficult to hold such a portfolio,especially given the restrictions that still exist in some of these markets and theirsmall size. The two other more realistic strategies, that involve investing in aparticular sector across markets, or diversifying across industries in a sole country,show that the variance can be reduced to, respectively, 8% and 43% of the ‘NoDiversification’ strategy.25,26 Both these strategies provide important diversification

24 For the value-weighted indices, the weights are given by the average ratio of the capitalisation of amarketrindustry relative to the total capitalisation of the firms in the sample over the sample period.

25 HR find that by ‘Maximum Diversification’ portfolio variance could be reduced to 18% of theaverage individual stock’s variance. Industry Diversification within a single country and GeographicalDiversification within a single industry could reduce the portfolio variance to, respectively, 38% and29% of the average individual stock’s variance.


Table 4aŽ .Diversification benefits } country vs. industry allocation equally weighted

Variance Proportion

Average individual stocks 0.0051 100% No di ersificationAverage country indices 0.0022 43% Industry di ersificationAverage industry indices 0.0004 8% Geographical di ersificationEW emerging markets’ index 0.0003 6% Maximum potential di ersification

a Ž .This table compares the average variance of individual stock returns No Di ersification with theŽvariance of an equally weighted portfolio of the emerging markets’ firms in sample Di ersifying Across

.All Emerging Markets’ Stocks and the variances attainable by Di ersifying Across Emerging MarketsWithin a Single Industry and by Di ersifying Across industries Within a Single Country. These lattervariances are weighted by the average ratio of the number of companies of a market or industry relativeto the total number of firms in the sample. The proportions are computed relative to the scenario of nodiversification. Weekly variances of total returns are measured in US dollars for the period of January1990 to December 1996.

benefits but it seems more fruitful to focus on geographical diversification. What-ever industry you pick, and not only on average, if you diversify across markets youare almost always better off } in terms of risk reduction } than if you choose toinvest domestically in any of these markets.27

One would expect that emerging markets would provide less diversificationpotential, on average, than European markets because market and industry indicesare more concentrated. While for the Industry Diversification that is marginallytrue, for Geographical Diversification , the potential diversification with emergingmarkets is much more substantial than the one observed for the mature marketsstudied by HR. The reason for this difference could lie in the more idiosyncraticcharacter of each of these markets and therefore the lower importance of commonfactors within the group of emerging markets.

One of the reasons why results are stronger for emerging markets is the fact thatmature markets’ indices include multinationals or more global stocks. Yet emerg-

Ž .ing markets’ indices also include large firms some are conglomerates and someare dual-listed on international exchanges.

In summary, results so far, using a broad industrial classification, are consistentwith the findings of previous literature by HR and contrast with previous studiesthat found an important role for the industry effects. For the particular case ofemerging markets, country specific effects are also the main driving force ofcountry indices and industry composition cannot account for the low cross-market

26 When I use value-weighted averages, the ‘Maximum Diversification’, the ‘Industry Diversification’and the ‘Geographical Diversification’ strategies reduce risk to, respectively, 13%, 52% and 18% of the‘No Diversification’ scenario.

27 To have an idea how these strategies would do in terms of performance I have looked at thehistorical returns of these three strategies. All of them involved historical average negative returnsduring the sample period. Yet the ‘Industrial Diversification’ strategy has earned only 0.01% more thanthe ‘Geographical Diversification’ for an additional standard deviation of 4%.


correlation. Yet, and as suggested by the significance tests on the subset of industryparameters, it would be unwise to simply forget these effects. Finally, and becausethe cross-market correlation does not seem to be affected by the industrialcomposition of the indices, I confirm the idea suggested by the preliminarydiagnosis, that it pays to diversify internationally but it is more interesting to do sowithin some particular sectors instead of others.

I have repeated the analysis for regions instead of country indices and using atwo-digit industry classification instead of the one-digit classification. I brieflydiscuss those results below.28

4.2. Regional effects

Markets are increasingly more and more influenced by the trading activity ofinternational investors that treat emerging markets or some emerging markets’regions as a single asset class. I have investigated whether that fund allocation waseffectively reflected on the returns’ structure.

I have repeated the regressions using region dummies, instead of countrydummies and, in addition, I have analysed the results within two regions.29 Idiscuss how much diversification can be obtained by investing in only one regionand, if one decides to invest in one market in a region, instead of investing in allthe markets in that region.

4.2.1. Across regionsThe inter-region analysis yields some interesting results:

v The fit of the regressions decreases dramatically reflecting that regional factorshave a very small explanatory power compared with country factors suggestingthat regional commonality is not very strong.

v Cross-regional correlation is low but higher than cross-market correlation. Onaverage, the cross-regional correlation is 0.17 against the average of 0.07between single markets.

v As with the country]industry regressions, the region]industry effects’ decom-position shows that the industry effects play a trivial role in explaining regionalindices.

v For the industry indices, however, there is an important difference. Whilebefore pure industry effects’ variability was, on average, 52% of the variance ofexcess industry index returns, the average variance ratio is now 92%. Again, thisresult reflects the small importance of region effects.

v The average variance of the regional indices is 0.0006 compared with anaverage of 0.0022 for countries. This result shows that investing in a region is

28 From here on, I only report the results for the equally weighted indices. Results for thevalue-weighted indices are, overall, similar and are available under request.

29 I have only looked at Asia and Latin America because these are the two regions in my sample witha reasonable number of constituent markets, respectively, 10 and 7.


not the same as investing in one particular country. Not only do you get moreindustrial diversification but, mainly, you gain through international diversifica-tion because countries within a region are not all alike. While previouslyinvesting across industries in one single country allowed to reduction of totalrisk to 43% of the ‘No Diversification’ scenario, now investing across industrieswithin a single region allows reducing risk to 12% of the ‘No Diversification’

Ž .scenario 10% in Asia, 20% in Latin America . Still, on average, it is better toŽ .invest across regions in one single industry as before, up to 8% of total risk .

These results highlight that there is a lot of diversity within a region and that, atleast over the past 7 years, picking up only one region, on average, almosteliminated the need to diversify across the universe of emerging markets. This is apartial statement that looks at one aspect of the game: risk. This finding combinedwith observed returns could explain the regional strategies of asset managersobserved in practice. To attain risk diversification, it is enough to choose only oneregion. Capital flows migrations from one region to another could reflect theinvestors’ returns forecasts.

4.2.2. Within a regionTo study further how regional benefits can be reaped, I have looked at what

happened within two regions: Asia and Latin America. A priori, I expect to findstronger common industry factors at a regional level, as industries are expected tobe more integrated on a regional basis rather than on a world-wide basis. Theresults confirm that, even within a particular region, the benefits of geographicaldiversification are larger than those of industry diversification. Thus, for example,an investor that has decided to invest only in Asia, will be better off investing in

Žany single industry across the 10 Asian markets risk reduction of up to 14% of the.total risk of the ‘No Diversification’ scenario than to confine herself to cross-

Ž .industry diversification in any single Asian country risk reduction up to 39% .

4.3. A finer industry partition

Another proposed aim of this study was to find out if the results derived herecould be driven by the use of a broad industrial classification.30 It is fallacious tosay that investing across-markets, in the broad Manufacturing sector, does notinvolve any industry diversification. To settle that, I have repeated the analysisusing a finer industry affiliation. The danger of this approach is that the affiliationcould become too narrow to meaningfully interpret the results.

Whether one uses the one or the two-digit industry classification, market indicesare driven by pure country effects. Yet, for industry indices, there are non-trivialchanges. As expected, because these categories include less constituents, the

30 Many emerging markets’ firms are conglomerates. Industry effects mean very little for this type offirms. To assess the impact of these firms, I have excluded the firms in SIC 9. The results are verysimilar to the core results.


variance is larger as a percentage of the total index, both the sum of the countryconstituents’ adjustments and the pure industry effects are more important butpure industry effects take a stronger share of that percentage. Thus it seems thatusing a finer partition unveils that the effect of industrial common factors is moreimportant.

Ž .Griffin and Karolyi 1998 report that a finer partition does not result in adifferent evaluation of industrial diversification benefits. My results contradict thisevidence: on average, the geographical diversification reduces risk to only 34% of

Žthe variance of the ‘No Diversification’ scenario against the 8% with a broader.industry classification . The figure of 8% included some industry diversification.

The importance of these results is not trivial: the loss in risk reduction if oneŽadopts a ‘Geographical Diversification’ strategy alone involves losing 28% 34%

.minus 6% while before, with a broader industry classification, it seemed that oneŽ .was losing only 2% 8% minus 6% . Of course, the 34% I get now also reflect that

the two-SIC affiliation generates very narrow indices with sometimes only twoconstituent firms.31 In sum, a narrow industrial affiliation shows that ignoring theindustrial mix will lead to an important loss of diversification benefits and that thebroad classification approach could be misleading, making what is industry diversi-fication look like country diversification.

4.4. Other results

Increasing economic or financial integration should lead to the dominance ofglobal factors over local factors. Tests of whether the country and industry effectsestimated above are constant over time could be very revealing because in manyemerging markets there have been important changes in recent years and ongoingprocesses of liberalisation of capital markets. It is not possible to pinpoint exactlywhen these markets became integrated into global capital markets, meaning thatthere the change in marginal pricing occurred. The dates used here are the officialliberalisation dates and the estimates of structural breaks of increase in net US

Žcapital flows given by Bekaert and Harvey 2000, Table 1: The opening of equity.markets in emerging countries . I split up the sample in two parts, prior and after

the liberalisation.The results over the two periods are not significantly different from the results

reported above: country effects dominate market indices and industry effects aretiny. In terms of diversification benefits, the loss in risk reduction if one adopts an‘Industry Diversification’ strategy alone involves losing 36% in the period afterliberalisation while before one was losing only 19%. The loss in risk reduction ifone adopts a ‘Geographical Diversification’ strategy alone involves losing 8% in theperiod after liberalisation while in the period prior to the liberalisation, one waslosing only 2%. These results are surprising and seem to suggest that global pricing

31 When I exclude those industries for which there is no geographical diversification, my resultschange slightly. On average, investing in a particular industry, diversifying across countries, reduces risk

Ž .to 27%. Ignoring industrial diversification still involves a loss of 21% 27% y 6% .


had no perverse effect in the benefits of diversification. These results should beread carefully because the fact that the samples for the two sub-periods do not

Ž .include the same number of markets and stocks could lead to incorrect inferencesI have also investigated whether the importance of country factors is the same

over the sample period. I estimated annual variances and the resulting varianceratios. Pure country effects always dominate the behaviour of market indices andthis is true for all markets, regardless of having observed or not increasingeconomic or financial integration over the sample period.

Ž .Previous evidence e.g. Dumas and Solnik, 1995 has shown that the correlationbetween stock returns and currency fluctuations is low and that currency variabilityis very small relative to stock market volatility. If exchange rates reflect onlymonetary policies, returns measured in US dollars will offset those policies.Consequently, country effects would be less strong for US dollars than in localcurrency returns. Alternatively, if exchange rates changes carry real effects or ifstock prices are either positively or negatively correlated in any other way withexchange rates changes, then the analysis in US dollars will unveil additionalfactors.

The arguments above seem to syndicate that the analysis in US dollars or inlocal currency could not produce the same results. I have examined the sensitivityof the results for local currency returns. Results are available upon request. Adifference t-test comparing the variance ratios for US dollars and local currencyreturns shows that the difference is not significantly different from zero. Theimportant role of country effects does not seem to arise from the fact that returnsare expressed in US dollars.

5. Conclusions

Previous evidence has shown that portfolio diversification into emerging marketsimproves its risk-adjusted performance. Yet no research has been conclusive aboutwhat is behind the correlation structure of returns that is the basis for those gains.There is still debate among academicians and practitioners about the influence ofindustrial composition of the indices in the less-than-perfect correlation betweenmarket indices. Another important issue is the effect of the globalisation of theeconomy and liberalisation of capital on the correlation of returns. My researchexpands the empirical evidence on these issues.

I assess the importance of global, country and industry factors by examining asample of emerging markets’ stocks. My results show that country pure effects arethe most important factors driving the behaviour of emerging markets’ individualstock returns. Emerging markets’ indices are driven by country factors, as shownpreviously for mature markets, and cross-market correlation does not seem to beaffected by the industrial composition of the indices. These results have importantimplications for portfolio diversification: regardless of the industry considered,geographical diversification dominates, in terms of risk reduction, domestic indus-trial diversification. For a finer industry classification, the country factors’ domi-


nance is still valid, but ignoring industrial diversification represents a much moreimportant loss of diversification benefits.

Another important result is the small role played by regional factors, showingthat, even within one region, the constituent markets are driven by local ratherthan regional effects. For the period analysed, focusing on one particular regionoffered diversification benefits similar to those obtainable by investing over theuniverse of emerging markets.

Acknowledgements

I am grateful to Narayan Naik for his helpful supervision and encouragement. Ialso thank Richard Brealey, Pradeep Yadav, Wayne Ferson, participants at IFAŽ .LBS seminars, the 1999 EFMA meetings in Paris, an anonymous referee and theeditor for useful comments and suggestions. All remaining errors are my responsi-bility. I thank Peter Wall from International Finance Corporation, for providingthe data. Faculdade de Economia da Universidade do Porto, Portugal providedgenerous financial support.

( )Appendix A: EMDB Emerging Markets Data Base coverage

ŽIndividual stocks’ weekly data in sample in parentheses, the number of firms for.each market at the end of 1996

From 1990Ž . Ž . Ž .Argentina 38 Brazil 99 Chile 51Ž . Ž . Ž .Colombia 27 Greece 69 India 151Ž . Ž . Ž .Indonesia 110 Jordan 58 Korea 185Ž . Ž . Ž .Malaysia 179 Mexico 114 Philippines 71Ž . Ž . Ž .Portugal 46 Taiwan 113 Thailand 115

Ž . Ž .Turkey 64 Venezuela 23

From 1992Ž .Pakistan 87

From 1993Ž . Ž . Ž .China 174 Hungary 16 Nigeria 16

Ž . Ž . Ž .Peru 40 Poland 28 Sri Lanka 51Ž . Ž .South Africa 65 Zimbabwe 24

Appendix B: Dummy variable regression methodology

I briefly describe here the analytical framework used in this study.32 I define the


following return generating model:

Ž .R s a q b q g q e B]1i t t jt k t i t

where R is the return for the firm i in period t that belongs to industry j andi tcountry k. a is a base or global level return in period t, b is the industry effect,t jtg the country effect and e is a firm specific disturbance.33

k t i tA stock is assumed to have either zero or unit exposure on a set of dummy

variables indicating country or industry affiliation. N is the total number of firms inthe sample. M is the number of firms in a particular week. The cross-sectionalregression, for each period, can be stated as follows:

J K

Ž .R s a q b I q g C q e B]2Ý Ýi j i j k ik ijs1 ks1

where I is a dummy variable that equals to one if the security belongs to industryi jj or zero otherwise and C the country dummy that equals to one if the securityi kbelongs to country k or zero otherwise. J is the number of industry categories andK is the number of emerging markets.

To overcome multicollinearity between the regressors, the effects are measuredrelative to the average firm in the sample, instead of measuring the effect of eachcountry and industry. This procedure is equivalent to measuring industry andcountry effects relative to the portfolio of emerging markets’ firms in the sample.For that, two restrictions have to be imposed:

v the weighted sum of industry dummies coefficients equals to zero, i.e.

J

Ž .n b s 0 B]3Ý j jjs1

where n is the number of firms in industry j;j

v the weighted sum of country coefficients equals zero, i.e.

K

Ž .m g s 0 B]4Ý k kks1

Ž .where m is the number of firms in country k. Please see Suits 1984 or KennedykŽ .1985 for more details.

32 Ž . Ž .Please refer to Suits 1984 or Kennedy 1985 for precise derivation.33 This formulation rules out any interaction between industry and country effects. This interaction

could be very important if markets are segmented and consequently industry effects are local incharacter.


The cross-sectional regression is run for each period in the sample, obtaining atime series of estimates for the industry and country effects.

The estimate of the intercept a gives the return on the equally weightedportfolio of firms in my sample. This portfolio has neither country nor industryeffects, in the way they were defined above. The estimates of the coefficients of thecountry dummy variables, g , show the extent to which the behaviour in thatk

Ž .market averaged over all industries is different from the emerging markets’average. The estimates of the coefficients of the industry dummy variables, b ,j

Žshow the extent to which the behaviour in that industry averaged over all.countries is different from the emerging markets’ average.

The sum of the average estimates of a and b yields the return on a portfoliojthat is diversified geographically in industry j. This sum tells how well industry j didin pure terms. Similarly, the sum of the average estimates of a and g yields thekreturn on a portfolio that is diversified across industries in country k. It tells howwell country k did in pure terms.

This estimation procedure allows us thus to reinterpret the individualcountryrindustry indices corrected for industryrgeographic composition.34 Theequally weighted index for any country k can be stated as:

m Jk1EW ˆ Ž .R s a q b I q g B]5ˆ ˆÝ Ýk j i j kmk is1 js1

and similarly for the equally weighted index for any industry j:

nj K1EW ˆ Ž .R s a q g C q b B]6ˆ ˆÝ Ýj k ik jnj is1 ks1

For the value-weighted indices, the regression is estimated using weighted leastsquares. The estimates are obtained in a similar way.

References

Akhogan, H., 1995. The Integration of International Capital Markets } Theory and EmpiricalEvidence. Edward Elgar.

Ammer, J., Mei, J., 1996. Measuring international economic linkages with stock market data. J. FinanceDecember, 1743]1763.

Bekaert, G., 1993. Market integration and investment barriers in emerging markets. World BankDiscussion Papers, 228, December, 221]251, Stijn Claessens and Sudarshan Gooptu.

Ž .Bekaert, G., Harvey, C., 1995. Time-varying world market integration. J. Finance 50 June , 403]444.Ž .Bekaert, G., Harvey, C., 2000. Foreign speculators and emerging equity markets. J. Finance 55 April ,

565]603.

34 Ž . Ž .The fact that B-5 and B-6 hold exactly results from the OLS assumption that errors are notŽ .correlated with the independent dummy variables. It directly results from that assumption that the

average error in any country or in any industry is zero. And, consequently, the error for the average firmin the sample is also zero.


Claessens, S., Dasgupta, S., Glen, J., 1998. The cross-section of stock returns: evidence from theŽ .emerging markets. Emerging Markets Q. 2 Winter , 4]15.

Divecha, A, Drach, J., Stefek, D., 1992. Emerging markets: a quantitative perspective. J. PortfolioManage. Fall, 41]50.

Dumas, B., Solnik, B., 1995. The world price of foreign exchange risk. J. Finance 50, 445]479.Errunza, V., Pabmanabhan, P., 1988. Further evidence on the benefits of portfolio investments in

Ž .emerging markets. Financial Anal. J. 44 JulyrAugust , 76]78.Ž .Fama, E., French, K., 1998. Value versus growth: the international evidence. J. Finance 53 December ,

1975]1999.Griffin, J., Karolyi, G., 1998. Another look at the role of the industrial structure of markets for

Ž .international diversification strategies. J. Financial Econ. 50 December , 351]373.Grinold, R., Rudd, A., Stefek, D., 1989. Global factors: fact or fiction. J. Portfolio Manage. Fall, 79]88.Harvey, C., 1993. Portfolio enhancement using emerging markets and conditioning information. World

Bank Discussion Papers, 228, December, 110]144, Stijn Claessens and Sudarshan Gooptu.Heston, S., Rouwenhorst, G., 1994. Does industrial structure explain the benefits of international

Ž .diversification. J. Financial Econ. 36 August , 3]27.Kennedy, P., 1985. Interpreting dummy variables. Rev. Econ. Stat. 174]175.Longuin, F., Solnik, B., 1995. Is the correlation in international equity returns constant: 1960]1990? J.

Int. Money Finance 14, 3]26.Meric, I., Meric, G., 1989. Potential gains from international portfolio diversification and inter-temporal

stock market relationships. J. Banking Finance 13, 627]640.Roll, R., 1992. Industrial structure and the comparative behavior of international stock market indices.

J. Finance 42, 3]41.Rouwenhorst, K., 1999. Local return factors and turnover in emerging stock markets. J. Finance 54

Ž .August , 1439]1464.Solnik, B., Boucrelle, C., Le Fur, Y., 1996. International market correlation and volatility. Financial

Anal. J. SeptemberrOctober, 17]34.Suits, D., 1984. Dummy variable mechanics vs. interpretation. Rev. Econ. Stat. 66, 177]180.

country and industry factors in returns: evidence from ... · a.p. serra remerging markets re¤iew...

Documents