Emerging Markets Review

3 Ž2002. 1᎐30

Robustness of size and value effects in

emerging equity markets, 1985᎐2000
Christopher B. Barry a,U , Elizabeth Goldreyer b,
Larry Lockwooda , Mauricio Rodriguez a
a
M.J. Neeley School of Business, Texas Christian Uni¨ ersity, Campus Box 298530, Fort Worth,
TX 76129, USA
b
Daniels College of Business, Uni¨ ersity of Den¨ er, Den¨ er, CO 80208, USA

Received 21 November 2001; received in revised form 10 December 2001;

accepted 14 December 2001

Abstract

We examine the robustness of size and book-to-market effects in 35 emerging equity

markets during 1985᎐2000. Mean returns for high book-to-market firms significantly exceed
mean returns for low book-to-market firms. These findings are robust to tests that control
for size effects and that remove extreme returns. Similarly, mean returns for small firms
exceed mean returns for large firms. But, the firm size results lack robustness to the removal
of extreme returns. Moreover, significant size effects are found in tests that define firm size
relative to the local market average, but generally are not found in tests that use absolute
firm size. Our findings are confirmed by cross-sectional regressions that control for syste-
matic risk at the global and local levels. 䊚 2002 Elsevier Science B.V. All rights reserved.

Keywords: Book-to-market; Market capitalization; Emerging equity markets; Robustness

1. Introduction

Fama and French Ž1992, 1996, 1998. examine a number of securities markets and
show that the common stocks of small firms generally provide higher mean returns
than do the stocks of large firms and that stocks with low market price compared to
book value, earnings, dividends, or cash flow generally outperform those at the

U
Corresponding author. Tel.: q1-817-257-7550; fax: q1-817-619-0194.
E-mail address: c.barry@tcu.edu ŽC.B. Barry..

1566-0141r02r$ - see front matter 䊚 2002 Elsevier Science B.V. All rights reserved.
PII: S 1 5 6 6 - 0 1 4 1 Ž 0 1 . 0 0 0 2 8 - 0
2 C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30

opposite end of the value scale. These empirical observations are known as the
firm size Žor ‘size’. and book-to-market equity Ž‘BErME’ or ‘value’. effects. The
work of Fama and French, and of others who have addressed size and BErME
issues, challenges the validity of the Capital Asset Pricing Model ŽCAPM. because
under the CAPM the only risk factor that affects the expected return of a security
is the security’s systematic risk, commonly measured by beta. Size and BErME
effects are anomalies relative to the CAPM. Accordingly, the validity of size and
BErME effects Žas well as other empirical anomalies. is a controversial issue in
empirical finance.1
The validity of size and BErME effects has been questioned by a number of
scholars.2 One approach has been to ask whether the effects hold generally or if
they sample-specific; for example, Black Ž1993. and MacKinlay Ž1995. raise these
issues. Early evidence on the size and BErME effects relied on Compustat-based
samples. Davis Ž1994. provides evidence consistent with the value and size factors
based on data that predates the Compustat tapes. Kim Ž1997. finds value and size
effects remain after testing a sample that includes all non-Compustat firms.
Fama and French Ž1998. address the sample-specific nature of their results by
studying global equity markets. They provide evidence on 13 developed countries
over 1975᎐1995 and find statistically significant BErME and other value effects in
12 of them.3 They also examine data on 16 emerging markets. They point out that

1
Empirical research has identified a number of factors that are associated with observed stock
returns. These factors include: firm size ŽBanz and Rolf, 1981; Reinganum, 1982; Keim, 1983 and Fama
and French, 1992.; the ratio of price to earnings ŽBasu, 1997, 1983; Reinganum, 1981 and Cook and
Rozeff, 1984.; the ratio of price to book value ŽRosenberg et al., 1985; Fama and French, 1992 and
Hawawini and Keim, 1991, trading volume ŽRoll, 1981.; and momentum ŽBrennan et al., 1998., among
others. Loughran Ž1997. observes seasonality in the BErME factor and notes that small growth stocks
have especially low returns in the United States.
2
For example, Berk and Jonathan Ž1995, 1997. argue that a firm’s market value measures risk and,
thus, that the existence of market value effects does not invalidate the CAPM. He develops a theory
arguing that the market value of the firm predicts expected returns because market value is negatively
related to the unmeasured risk of the firm, and he shows that after controlling for size as measured by
market value, other size proxies such as book value of assets and sales are unrelated to average returns.
Black Ž1995. warns against using models based solely on empirical observations but which lack
theoretical motivation. To provide a rationale for the value premium, Fama and French Ž1995. show
that firms with high BErME ratios often are distressed firms, characterised by depressed earnings and
highly uncertain future earnings. Chen and Zheng Ž1998. find that value stocks are characterized by
high financial risk, high earnings uncertainty and high distress Žproxied by frequency of dividend cuts..
In contrast, Daniel and Titman Ž1997. contend that the risk model cannot be distinguished from a
behavioral overreaction model Žsee De Bondt and Thaler, 1985, 1987. since both models are consistent
with relative distress compensation factors. Davis et al. Ž2000. address this dilemma by showing that the
size and BErME patterns in average returns are better explained by rational compensation for risk
than by an overreaction hypothesis. Taking a different approach, Chan et al. Ž1998. examine the ability
of different factors to explain correlations across stocks returns. The authors find that firm size and
BErME are two of the more important sources of covariation among stock returns for the US, UK and
Japanese equity markets. These findings support the argument that size and BErME proxy for priced
risk factors.
3
Similarly, Arshanapalli et al. Ž1998. examine 18 global stock markets and find superior performance
associated with size and BErME.
 C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30 3

security returns in emerging markets are generally leptokurtic and right skewed, so
they stress that statistical inference can be ‘hazardous.’ Nevertheless, they conclude
that the evidence is generally consistent with the existence of value effects based
on the BErME measure. They note that equity portfolios with low BErME
outperformed high BErME portfolios in 12 of the 16 countries they examined.4
Evidence on size in their sample is less pronounced Ždifferences in returns across
extreme size portfolios are less than two standard deviations from zero..
Patel Ž1998. and Rouwenhorst Ž1999. support the findings of Fama and French
Ž1998. by finding a premium for value stocks and small firms in emerging markets.
Hart et al. Ž2001. also report a significant value effect as well as momentum and
earnings revision effects in emerging markets, and they find that these strategies
could be implemented in practice by large investors even after considering the
practical difficulties and costs of trading in emerging markets. However, they find
no significant firm size effect, and the direction of the effect changes when they
impose a minimum capitalization requirement.5 Achour et al. Ž1998, 1999a,b,c.
provide a comprehensive examination of stock selection strategies based on funda-
mental, expectation and technical factors within Mexico, South Africa and
Malaysia.6 In their 1998 paper, they find strong value effects in South Africa and
Malaysia. No value effects are found in Mexico. Their value factors include
BErME, trailing earnings yield, prospective earnings yield and cash earnings-to-
price yield.
Claessens et al. Ž1993, 1995, 1998. also examine market anomalies in emerging
markets.7 They find that portfolio returns in most of the markets are non-normal

4
Fama and French’s results on value are less prominent when the proxy for ‘value’ is the
price-to-earnings ratio instead of BErME; see Fama and French Ž1998, p. 1996..
5
Hart et al. Ž2001. also find selection strategies of high earnings to price and stocks with positive
analysts’ earnings revisions to carry a premium. Results from their momentum strategy are mixed
depending on the type of investment strategy employed. A multivariate strategy combining value,
momentum and earnings revisions improves results. They find it is best to apply these results to all
emerging markets, as opposed to country by country.
6
They consider BErME, cash flow-to-price, earnings-to-price, earnings-to-price, dividends-to-price,
earnings growth, revenue growth, debtrequity ratios, return on equity, market capitalization, prospec-
tive earnings-to-price, IBES revisions, prospective earnings growth and a number of momentum
measures.
7
Fama and French Ž1998. address the differences between their results and those of Claessens et al.
Ž1998. for emerging markets. They suggest that differences in sample periods might explain some of the
differences in results but that differences in estimation techniques are likely the main explanation. They
argue in particular that cross-sectional regressions Žthe main methodology used by Claessens et al.,
1998. are sensitive to outliers and that such outliers are common in emerging markets. In contrast, the
methodology used by Fama and French is based on differences in returns across portfolios formed on
the basis of a single factor at a time. Another difference between tests based on differences in returns
for ranked portfolios and cross-sectional regressions is that the regressions make an explicit assumption
about functional form Že.g. effects are linear. while portfolio differences do not assume such forms.
Durham Ž2000. argues that studies that have reported statistically significant anomalies in emerging
equity markets suffer from specification bias. He suggests applying extreme bound analysis as a remedy.
He examines 16 emerging markets over 1988᎐1995, finding price-to-book, country risk and relative
market size to have robust effects on returns.
4 C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30

and that beta is generally insignificant in explaining returns. They observe inconsis-
tent size effects and conclude that the size effect does not seem to prevail in
emerging markets to the degree that it has in developed markets. They find
evidence of strong BErME effects, but the direction of the effect is often the
reverse of Fama and French Ž1998., Patel Ž1998., Rouwenhorst Ž1999., Hart et al.
Ž2001. and of earlier work on developed markets.8 Fama and French Ž1998. point
out that the differences may be due to the different sample periods used, but is
more likely due to the different methodologies employed and the effects of
outliers. Shumway Ž1997. and Shumway and Warther Ž1999. examine the bias in
returns caused by omitting the post-delisting returns of stocks that are delisted for
negative performance reasons from the NYSE or NASDAQ. Shumway Ž1997. finds
that delisting has especially important effects on size-based returns but not on
book-to-market-based returns for NYSE stocks. Shumway and Warther Ž1999. find
that the size effect disappears entirely from NASDAQ data when they correct for
the delisting bias. They conclude, ‘Consequently, there is no evidence that there
ever was a size effect among Nasdaq stocks.’ ŽShumway and Warther Ž1999, p.
2378..
A similar argument is made in Bossaerts and Fohlin Ž2000. in their study of
German stocks over the period 1881᎐1913. They observe a size effect within the
German data, but they argue the effect is likely caused by selection bias. It
disappears later in their sample. Moreover, they find a BErME effect, but they
point out that the direction of the effect is opposite the effect observed in recent
US data Ži.e. in their sample, growth stocks outperform value stocks..
Controversy remains over the existence and importance of size and book-to-
market effects in financial markets. In this paper, we provide new evidence about
the size and BErME effects in emerging market returns. We perform a compre-
hensive analysis, focusing on the robustness of the estimated premiums for size and
BErME, and use a more extensive sample than has been used in prior work on
emerging markets. Our sample uses data over a 15-year period from 35 emerging
markets; previous tests of size and value effects have been limited to 10 years of
data and examine one market at a time. Because of the high volatility of emerging
markets, it is especially important to examine longer sample periods. We also
provide evidence based on a variety of empirical methods designed to overcome
concerns raised about the methodologies used in earlier tests. We perform our
tests on portfolios, which lessens the effects of outliers. We provide evidence based
on standard t-tests, but we also conduct non-parametric statistics that are unaf-
fected by non-normality in the data. We use: univariate portfolio tests; multivariate
portfolio tests Žportfolios based on one factor are formed within portfolios based
on the other factor.; tests including all data vs. tests excluding extreme returns;

8
Additional evidence from emerging markets includes Herrera and Lockwood Ž1994., who show that
the relationship among average returns, beta and firm size for Mexican firms is nearly identical to that
of a matched set of NASDAQ firms, and Chui and Wei Ž1998. who examine five Pacific Basin emerging
markets over 1977᎐1993 and report a weak beta effect and significant BErME and size effects.
 C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30 5

tests on subperiods within the entire sample period and that control for seasonality
effects; tests based on cross-sectional returns Žwith and without the extreme
returns . before and after controlling for beta risk relative to local and global
markets; and tests employing relative as well as absolute size measures. Thus, we
provide a comprehensive set of results designed to examine the robustness of the
conclusions we can reach regarding size and value effects in emerging markets.
We find evidence of a strong and persistent book-to-market effect in which mean
returns for high book-to-market Žvalue. firms significantly exceed mean returns for
low book-to-market Žgrowth. firms. These findings hold consistently across para-
metric and non-parametric tests that examine book-to-market independently and
that control for firm size. Similarly, when defining firm size relative to local market
average size, mean returns for small firms significantly exceed mean returns for
large firms. The results are not driven by January effects. Our findings hold after
controlling for book-to-market effects using parametric tests, but the results using
non-parametric tests are mixed. When using absolute firm size, the firm size
premium often disappears. Our findings are confirmed by cross-sectional regres-
sions that control for systematic risk at the global and domestic levels.
We re-examine all results after omitting observations with returns in the upper
or lower 1% tails of the distribution of returns. This approach addresses the
concerns of Fama and French Ž1998. with outliers. In univariate and multivariate
portfolio tests with extreme returns removed, we continue to find strong BErME
effects, but the size effects are appreciably diminished or missing altogether.
Our results consistently point to a robust BErME effect in emerging markets
with and without extreme returns. The size effect is generally present when
measuring size in terms relative to a firm’s local market, and when extreme returns
are included, but is generally not observable if extreme returns are omitted, or if
size is measured in absolute terms. Therefore, our findings provide important
out-of-sample evidence of a return premium for value stocks. The robustness of
our results, combined with results of studies on other markets, demonstrates that
value effects persist around the world and that BErME partially explains differ-
ences in average returns. These results are particularly important because the
BErME factor also has been shown to explain covariation among stocks in various
markets and has been used to proxy for distress risk.9 Thus, the BErME factor
appears to pass the robust acid test for inclusion in multifactor pricing models.
The plan of the paper is as follows. In Section 2 we identify the data sources
used in the study and present an overview of the properties of the sample. Section
3 provides results based on analyses of portfolios formed on the basis of only one
factor at a time. Section 4 presents results for portfolios formed jointly on the basis
of size and BErME. Section 5 describes results of our cross-sectional regression
tests. Section 6 provides results in which absolute size is used rather than relative
size. Section 7 summarizes the results of the paper.

9
See Fama and French Ž1995. and Chan et al. Ž1998..
6 C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30

2. Data and methodology

We use Standard and Poor’s Emerging Markets Data Base Ž2000. ŽEMDB.,
previously distributed by the International Finance Corporation. We use monthly
data from this database on individual stocks in 35 emerging markets. The database
contains historical prices, currency exchange rates, numbers of shares outstanding,
earnings, dividends and other accounting and market data. We observe market
capitalization Žsize. and book-to-market ŽBErME. ratios at the individual firm
level and use those data to construct portfolios across the full set of emerging
markets in the EMDB to test for size and BErME effects. We also provide
cumulative returns over the sample period for portfolios based on size and
BErME.
The sample period spans January 1985 through July 2000. The database provides
coverage for approximately 2000 EM firms, but it is not exhaustive by any means.
For example, the Emerging Stock Markets Factbook Ž2000. indicates that 487
companies were listed on the two principal Brazilian exchanges at the end of 1999,
but only 90 were included in the EMDB.
We compute monthly returns for each firm in the EMDB as follows:

Ž Pt = Cap.Adj. t % Exch.Rt. t . q Ž Div. t % Exch.Rt. t .

Rt s y1 Ž1.
Ž Pty 1 % Exch.Rt. ty1 .

where the adjustment for stock splits and rights offerings ŽCap.Adj.., Exchange
Rate ŽExch.Rt.., Dividend ŽDiv.. and stock price ŽP. are all obtained from the
EMDB. All returns are denominated in US dollars, so historical prices in local
currencies are adjusted for the exchange rate between the local currency and the
US dollar. We use the EMDB to calculate size, and the database includes BErME
ratios.
We report results based on relative BErME and both relative and absolute size.
‘Relative BErME’ for a stock is BErME for the stock divided by the average
BErME ratio in the same market as the stock and for the same month. We use
relative BErME instead of absolute BErME for a stock because accounting
systems vary widely across the 35 markets in our sample.10 For example, Latin
American markets tend to use inflation accounting in constructing book values of
assets and liabilities. Many markets do not. Thus, book value is measured differ-
ently in different markets, and it follows that while a given BErME value in one
market is comparable to other BErME values in the same market, it is generally
not comparable to a BErME value in a distinct market. Accordingly, we adjust
BErME to account for these differences in accounting treatment. We refer to the
resulting value as ‘relative BErME’ or just BErME for brevity.

10
This is similar to the relative BErME measure used by Cohen and Polk Ž1997. to adjust for
differences in accounting practice across industries within the US market.
 C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30 7

Initially, we define size for each firm relative to each firm’s local market average.
Unlike BErME, which suffers from incomparability across markets due to differ-
ences in accounting, size is entirely comparable across markets: $100 million in
Brazil is exactly the same value as $100 million in Zimbabwe. However, if size
matters at the local level, then using absolute size in portfolios may mask the effect
that size has in distinct markets and in fact could capture differences in perfor-
mance across markets rather than the effect of size per se. For example, the
average size of a stock in Brazil in our data is $1.7 billion. The largest stock from
Zimbabwe in our data is $1.5 billion. If there is a local size effect due to the
behavior of local investors in Zimbabwe and also a local size effect due to the
behavior of local investors in Brazil, the use of absolute size in forming composite
portfolios based on size could confound those effects. The use of relative size
allows us to aggregate the relatively large stocks in Zimbabwe with those from
Brazil, and so on, and that will provide us with more powerful tests of the size
effect.
In Section 6 of the paper, we also report results using absolute size, or size not
adjusted by the average size in the local market. The choice of relative size vs.
absolute size is ultimately a question of the degree to which emerging capital
markets are integrated globally. In a perfectly integrated global capital market in
which investors freely select from among all securities in global markets, absolute
size would be the preferred measure of size. If, alternatively, global capital markets
were completely segmented, relative size would be the appropriate measure to
employ in testing for the existence of a size effect. We provide results based on
both measures.11
Portfolios are constructed as follows. First, for each month, the size of each stock
is computed as the product of stock price and number of shares outstanding.12
Second, the relative size of each stock is computed as the ratio of the stock’s size to
the average size of all stocks Žin the EMDB. in that stock’s local market. Third,
relative size quintiles are formed across all 35 markets. That is, the first size
quintile consists of the 20% of all firms in the database that rank smallest in
relative size as measured in US dollars. We refer to the first portfolio as the small
firm portfolio.
We also create relative BErME portfolios. Similar to the relative size calcula-
tion, for each firm and for each month, we calculate the relative BErME as the
ratio of the firm’s BErME to the average BErME in the firm’s local market. For
each month all stocks in the database are ranked by their relative BErME ratios
and assigned to BErME quintiles. The lowest BErME portfolios are referred to
as ‘growth’ portfolios, and the highest are referred to as ‘value’ portfolios.

11
Solnik Ž2000, p. 172. concludes on the basis of recent research in international finance that the
assets of developed securities markets are priced in an integrated global capital market but that
constraints in emerging markets are still serious and limit the degree of integration of these markets.
12
If a firm has multiple classes of stock, we sum the market capitalizations of all classes to arrive at
the overall firm size.
8 C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30

In addition to univariate tests, we present results based on portfolios that are

formed using both relative size and relative BErME rankings. In these results,
BErME effects can be examined controlling for size and vice versa. Each holding
period, firms within each size quintile are ranked on BErME and assigned to
BErME quintiles. This process is repeated for each size quintile. Thus, 25
portfolios are produced per year: Five size categories are established, and then five
BErME quintiles are constructed within each size quintile.13 Returns are equally
weighted across firms within each portfolio. Differences in returns across the size
BErME portfolios are used to infer the independent effects of firm size and
book-to-market on equity returns.
Results are presented for portfolios rebalanced once per year. For portfolios
formed on firm size, stocks are classified in May of year t, and the portfolios are
held for the next 12 months through May of year t q 1. This process is repeated
every year. We use May rather than December to classify stocks on firm size to
avoid possible year end firm size bias when forming portfolios. To avoid conflicts
with reporting lags in accounting data, we use a 7-month lag for all BErME tests
to assure that the book value data were available to market participants.14 Thus, all
BErME portfolios are formed using 7-month-old relative BErME values and they
are then held for the subsequent 12-month holding period.
Table 1 presents descriptive data for 35 emerging markets. Fifteen of the
markets have data available starting in or after 1993; only four markets have data
all the way back to 1985. The number of firms with data available varied widely
both within and across markets. For example, data were available for only one
stock at the start of China’s data period, but by the end of 1996, China had more
stocks in the EMDB Ž180. than any other market. Thus, over time a great deal
more information has become available. Our results generally treat each month’s
data equally, i.e. we do not weigh the returns across time in any way to account for
differences in number of portfolio firms at different points in time. Of course, since
return variances will be lower on more diversified portfolios, there is an implicit
time weighting effect. Unfortunately, the effect may be the reverse of what we
would hope. In earlier periods, when there is less information, portfolio returns
tend to be more extreme and, hence, can affect the final results more than would
later, lower risk and less extreme portfolio returns.
Firm sizes are reported in millions of US dollars. Sizes vary widely. For seven
markets, the average market capitalization is less than $100 million, while for
South Africa the average firm size is slightly more than $2 billion. Taiwan and

13
To illustrate, consider a month in which 1000 stocks are in the database and have all necessary
data. We first form size portfolios of 200 stocks each based on size, and then within each size category
we form BErME portfolios of 40 stocks each. Of course, in most cases the number of stocks is not
conveniently divisible by 25, so there will be minor differences in numbers of stocks in the different
portfolios.
14
Fama and French Ž1992. also use a similar lag for identical reasons in their examination of US
markets.
Table 1
Descriptive statistics for 35 emerging markets, 1985᎐2000

Market Start Mean Mean size Median size S.D. S.D. size Min Min size Max Max size
date BErME Ž$US. Ž$US. BErME Ž$US. BErME Ž$US. BErME Ž$US.

Europe

C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30

Czech Rep. Jul 94 2.14 198.24 54.41 2.96 490.70 Ž0.26. 0.38 33.33 5111.53
Greece Apr 87 0.53 530.03 179.45 0.47 1320.76 0.00 6.04 6.25 15 162.81
Hungary Jun 93 0.85 387.51 76.61 0.77 1023.58 0.08 2.54 5.56 9643.34
Poland Jun 93 0.70 310.22 125.27 0.53 866.40 0.04 5.04 7.69 12 815.11
Portugal Jul 86 0.76 526.70 169.72 0.65 1145.86 0.00 5.57 8.33 15 785.15
Russia Jul 96 7.23 1481.01 271.78 14.37 3417.56 0.01 2.92 100.00 35 307.54
Slovakia Jul 96 3.88 60.01 21.24 6.79 95.71 Ž33.33. 0.47 50.00 552.39
Turkey Mar 89 0.40 559.21 172.46 0.43 1074.71 Ž0.11. 1.55 5.88 15 392.49

Latin America
Argentina Oct 86 2.26 535.68 138.63 5.20 1117.19 Ž5.56. 0.16 100.00 13 240.53
Brazil Oct 86 2.33 1766.98 330.72 3.18 4111.78 Ž10.00. 0.72 50.00 48 723.55
Chile Jul 86 0.91 836.16 356.28 0.87 1228.10 0.04 4.69 16.67 8324.52
Colombia Jul 85 1.97 228.99 80.92 4.13 332.19 0.01 1.77 100.00 2957.22
Mexico Oct 86 1.01 1509.17 441.08 2.40 3141.09 Ž50.00. 3.68 50.00 35 490.64
Peru Jun 93 0.93 304.97 48.37 1.41 530.29 0.05 0.65 20.00 3921.88
Venezuela Sep 86 1.53 274.45 122.27 3.31 389.17 0.00 1.80 50.00 3278.92

Middle East
Bahrain Jul 99 1.82 361.84 167.01 3.78 475.99 0.07 13.93 20.00 1671.09
Israel Jul 97 0.85 726.00 417.12 0.44 828.37 0.08 27.34 3.45 6942.05
Jordan Jan 87 0.82 82.22 21.92 0.60 255.54 Ž2.63. 0.79 4.35 2762.36
Oman Jul 99 1.00 95.15 40.09 0.58 116.21 0.24 3.93 4.00 555.20
Saudi Arabia Jun 98 0.89 1817.09 749.00 0.61 2572.66 0.21 26.00 5.56 12 478.34
Egypt Jul 96 0.49 177.23 78.84 0.48 233.97 0.02 4.23 8.33 4096.03

9
 10
Table 1 Ž Continued.

Market Start Mean Mean size Median size S.D. S.D. size Min Min size Max Max size
date BErME Ž$US. Ž$US. BErME Ž$US. BErME Ž$US. BErME Ž$US.

C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30

Asia
China Jul 93 0.61 539.07 232.36 2.15 2753.57 Ž0.21. 8.72 100.00 125 918.66
Sri Lanka Jul 93 1.11 30.85 18.50 1.63 35.36 Ž1.19. 0.38 33.33 309.17
Taiwan Jul 85 0.46 1575.12 593.09 0.34 2853.18 Ž0.07. 3.95 7.69 50 910.85
India Oct 86 0.74 475.49 191.96 1.17 904.94 Ž11.11. 2.07 16.67 14 802.86
Indonesia Jun 90 0.84 430.76 110.22 2.32 1084.45 Ž100.00. 0.94 33.33 16 691.21
Korea Jul 86 1.49 747.12 271.58 2.79 2078.30 Ž50.00. 0.42 100.00 49 938.38
Malaysia Jul 85 0.68 814.87 356.63 0.82 1665.87 Ž7.14. 2.66 11.11 21 500.19
Pakistan Jul 86 0.89 69.88 24.31 1.15 169.20 Ž4.55. 0.22 16.67 3645.39
Philippines Apr 88 0.99 718.90 157.24 1.49 1188.19 Ž2.13. 1.71 25.00 6813.49
Thailand Jul 86 0.93 715.12 279.58 2.31 1202.21 Ž20.00. 0.83 100.00 13 368.76

Africa
Morocco Jul 96 0.33 590.28 503.15 0.16 473.97 0.09 27.87 0.97 2295.01
Nigeria Jul 85 0.75 55.84 40.03 0.81 55.27 0.02 0.52 9.09 344.04
South Africa Jun 93 0.61 2018.57 1299.86 0.75 2331.37 0.00 14.97 7.69 22 841.75
Zimbabwe Jul 86 1.78 58.83 25.13 3.01 111.66 0.06 0.96 50.00 1482.30

This table presents descriptive statistics for monthly book-to-market values and firm sizes for 35 emerging markets over January 1985᎐June 2000. The
start date is the first month for which valid return, size and lagged book-to-market values are available on individual firms within the local market. ‘Mean
BErME’ and ‘mean size Ž$US.’ are the average book-to-market value and average firm size computed across all firms and months included in tests for the
local market. Firm size is in $US millions. ‘S.D. BErME’ and ‘S.D. size’ are standard deviations of book-to-market values and firm sizes computed over all
firms and months within the local market. ‘Min BErME’ and ‘max BErME’ are local market low and high values for book-to-market for all sampled firms.
Similarly, ‘min size’ and ‘max size’ are local market low and high values for firm size for all sampled firms.
 C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30 11

Mexico, two of the more established emerging markets, have average firm sizes of
$1.58 and $1.51 billion, respectively.
Book-to-market ratios also vary widely across markets, consistent with differ-
ences in accounting conventions. The average BErME ratios vary from a low of
0.33 for Morocco to a high of 7.23 for Russia.

3. Univariate tests

In this section we report results based on relative size without controlling for
relative BErME and on BErME without controlling for size. Table 2 presents the
results of the univariate tests for five portfolios formed on firm size ŽPanel A. and
for five portfolios formed on BErME ŽPanel B.. The number of firms in each
portfolio averaged over the entire sample period equals 197 and for the last month
of the sample period equals 373. Fig. 1 plots the returns for each of the size and
BErME portfolios. Fig. 1 shows that average returns for the full data set conform
to empirical results reported for developed markets. Mean returns decline across
size categories Žfrom 0.0294 for small firms to 0.0074 for large firms. and rise
across BErME categories Žfrom 0.0104 for low BErME, or growth firms, to 0.0368
for high BErME, or value firms..
Table 2 presents t-tests that compare the mean returns for the smallest vs.
largest firm size portfolios and for value vs. growth portfolios. We perform t-tests
on the mean difference of the returns. Therefore, the tests are performed on the
return difference between small and big firms, SMB, and between high and low
BErME firms, HML. These tests explicitly account for cross-correlation in the
returns. The results presented in the line called ‘all data’ of Panel A of Table 2
show that the size effect is statistically significant at the 1% level using all data
included the EMDB Ž t-statistic s 4.42.. In light of the non-normality of emerging
market returns, we also report results on the percentage of months for which the
returns of small stock portfolio exceeded those of the large stock portfolio. Small
stocks outperformed large stocks in over 63% of the months, a result that is
significant at the 1% level. The annualized SMB return is approximately 30%.
Similar results for the BErME portfolios are shown in the ‘all data’ line in Panel
B of Table 2. The t-test of mean difference in returns between the highest and
lowest BErME portfolios Žvalue vs. growth. is significant at the 1% level for both
parametric and non-parametric tests using all data. Value stocks outperformed
growth stocks in more than 74% of the months in our sample. The annualized
HML return is approximately 36%. These annual excess returns are considerably
higher than those reported by Fama and French Ž1998. for developed markets for
their 1974᎐1995 sample period. These results demonstrate the importance of
investment style analysis for emerging markets Žsee Sharpe, 1992.. Our univariate
tests are consistent with Claessens et al. Ž1993. who also document a strong
BErME effect for emerging markets.
We re-examine the results after controlling for January effects. Davis Ž1994.,
Loughran Ž1997. and Chan et al. Ž1998. show that firm size and book-to-market
12 C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30

Table 2
Univariate tests of market capitalization and book-to-market effects

Small 1 2 3 4 Large 5 t-Statistic % Small ) large

Panel A. Mean monthly returns by firm size quintile

All data 0.02942 0.02185 0.01792 0.01363 0.00741 4.42a 63.43%a
Removing 0.01446 0.01607 0.01416 0.01343 0.00914 1.80 56.47%
extremes

Panel B. Mean monthly returns by book-to-market quintile

Low 1 2 3 4 High 5 t-Statistic % Value ) growth
Žgrowth. Žvalue.
All data 0.01041 0.01305 0.01566 0.02372 0.03628 5.71a 74.19%a
Removing 0.00790 0.01157 0.01290 0.01609 0.02312 4.69a 61.77%a
extremes

This table reports mean monthly returns for the emerging market sample for various market value
equity Žsize. and book-to-market ŽBErME. equity quintiles. Portfolios are rebalanced once per year.
Panel A presents mean monthly returns for each market capitalization quintile. Stocks are ranked on
the basis of relative size defined as the firm’s market capitalization divided by the average market
capitalization for the firm’s local market. Firms are assigned to one of five quintiles Žfrom smallest to
largest.. Group 1 Žsmall. consists of the smallest relative size firms and group 5 Žlarge. consists of the
largest relative size firms. Panel B presents mean monthly returns for each relative BErME quintile. A
relative BErME ratio is calculated by dividing the firm’s BErME ratio by the average BErME ratio in
the firm’s local market for that month. Then firms are ranked and portfolios are formed on the basis of
these relative BErME ratios. The stocks are then placed into five relative BErME quintiles from
lowest relative BErME Žor ‘growth’ firms. through highest relative BErME Žor ‘value’ firms..
Portfolios are formed on the basis of relative size or relative BErME are rebalanced once per year.
Mean returns are presented for each portfolio. The t-statistics in Panel A test for differences between
the smallest and largest relative size portfolios. The last column reports the percentage of months for
which the return on the small size portfolio exceeded that of the large size portfolio. Similarly, the
t-statistics in Panel B test for differences between the value Žhigh BErME. and growth Žlow BErME.
portfolios. The last column reports the percentage of months for which the return on the value stock
portfolio exceeded that of the growth stock portfolio. Tests of significance on the difference between
the observed percentage and 50% are also performed.
a
Significant at the 0.01 level.

effects in US markets are especially strong during January but are weak during the
remainder of the year. In contrast, we find that the HML and SMB returns rise
when Januaries are deleted. For example, we find that the mean monthly non-
January HML return is 0.0155 vs. 0.0124 for the January return. Moreover, the
SMB return is negative during Januaries Žy0.0331. vs. 0.0129 during non-Januaries.
These findings indicate that the effects we document for emerging markets are not
driven by January effects.15

15
We do find a January effect for emerging market stock returns when no sorts are performed. In
particular, the mean monthly emerging market composite return during January is more than triple the
non-January return using market value weighting and is nearly twice the non-January return using equal
weighting.
 C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30 13

Fig. 1. Mean monthly returns by size or value portfolios. This figure illustrates the change in mean
monthly returns for equities in 35 emerging markets across five portfolios formed on the basis of
relative size and five portfolios formed on the basis of book equity to market equity ŽBErME. ratios.
Means are computed over June 1986᎐July 2000. Portfolios are rebalanced annually. The figure show
that equity returns in emerging markets fall as firm size increases and returns fall as BErME ratios
falls.

We examine the degree to which size and value effects are due to extreme
returns. We repeat the tests discussed above from Table 2 after deleting returns
falling in the lower and upper 1% tails of the returns distributions over all sampled
months. The results after removing extreme returns Žoutliers. are presented in
Panel A for size portfolios and in Panel B for BErME portfolios in the lines
labeled ‘removing extremes.’ We find that outliers make up approximately 26% of
the average returns of the equally weighted composite portfolio of EMDB stocks
during the 1985᎐2000 sample period. After removing outliers, the size effect is not
significant Ž t-statistic s 1.80.. The BErME effect remains significant Ž t-statistic s
4.69.. The difference in returns between small and large firms drops more than
75% from 220 basis points per month Ž0.02942᎐0.00741. to 53 basis points per
month Ž0.01446᎐0.00914. or to an annualized 6.6%, after removing extreme re-
turns. The difference in returns between value and growth stocks drops 41% from
259 basis points per month to 152 basis points per month, or to approximately 20%
annually. These results show that the size effect in emerging markets is driven by a
small set of returns comprising the upper 1% tail of the returns distribution. The
BErME effect, however, remains highly significant after removing the extreme
returns. The lack of size effects after deleting outliers is similar to the results
documented by Knez and Ready Ž1997. for US markets. Therefore, our results
indicate that for investors to have realized return enhancements by overweighing
14 C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30

small EM firms, they would have had to follow a strict portfolio policy of buying
every small stock so as to include those that led to extremely high returns.
Earlier work has shown that size, value and other factors may have effects that
are associated with each other. For example, Loughran Ž1997. reports a combined
effect wherein higher returns were achieved by investments in small firms that
were value stocks. Therefore, in the following section, we form portfolios account-
ing for size and BErME simultaneously, thus controlling in part for the effects of
the one variable while examining the effects of the other.

4. Results based on portfolios formed on both size and BE r ME

For the results in this section Žshown in Table 3., portfolios are first formed on
the basis of relative firm size. Then, within size categories, portfolios are formed on
the basis of relative BErME. This approach permits us to examine value effects
among small firms and among large firms and provides a different sample in which
to test the claims in Loughran Ž1997.. To be sure that the results are not driven by
the order of portfolio formations, we also conducted tests Žnot reported here. in
which the reverse procedure was applied.16 The results were qualitatively the same
as those presented here.
The t-statistics reported in the last column of Table 3 test the significance of the
mean difference between high and low BErME portfolio returns within each size
quintile. The t-statistics reported in the bottom row test the significance of the
mean difference between the smallest and largest firm portfolio returns within
each BErME quintile. The percentages shown are the fractions of months for
which the high BErME portfolio outperformed the low BErME portfolio and for
which the small size portfolio outperformed the large size portfolio. The number of
firms in each of the 25 portfolios equals 40 when averaged over the entire sample
period and equals 75 for the last month of the sample period.17 The findings
reported in the table indicate significant BErME effects within each size quintile.
The t-statistics for the mean differences range from 2.34 to 3.99. These results
contrast with the findings of Loughran Ž1997. for the US, that value effects do not
appear to apply to large firms. Panel B shows that BErME effects remain

16
See Berk and Jonathan Ž2000. for consequences of sorting procedures for tests of asset pricing
models.
17
We also compare the sizes of our portfolios against those used in Fama and French Ž1992.. Their
portfolio sizes ranged from 70 to 177 for the 10 size-beta portfolios in their smallest size decile. We also
compare the sizes of our portfolios against those used in Fama and French Ž1992.. Their portfolio sizes
ranged from: 70᎐177 for the 10 size-beta portfolios in their smallest size decile; 15᎐41 for the 20
size-beta portfolios in their 2nd and 3rd smallest size deciles; and 11᎐22 for the 10 size-beta portfolios
in their largest firm size decile. Fama and French Ž1992. formed their size breakpoints using NYSE
firms only, which is the reason for the non-uniformity of number of firms in their portfolios: 15᎐41 for
the 20 size-beta portfolios in their 2nd and 3rd smallest size deciles; and 11᎐22 for the 10 size-beta
portfolios in their largest firm size decile. Fama and French Ž1992. formed their size breakpoints using
NYSE firms only, which is the reason for the non-uniformity of number of firms in their portfolios.
 C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30 15

Table 3
Mean monthly returns for size BErME portfolios

Size Book-to-market quintiles t-Statistic

quintiles Low 1 2 3 4 High 5
Ž%.
Žgrowth. Žvalue.

Panel A. All data

Small 1 0.02267 0.02543 0.03614 0.03426 0.05164 3.59a
59.41%b
2 0.01575 0.01772 0.02303 0.02170 0.03162 2.34b
61.18%a
3 0.01036 0.01294 0.01689 0.01365 0.03569 4.18a
63.12%a
4 0.008102 0.00660 0.01256 0.01587 0.02467 3.07a
55.19%
Large 5 y0.00206 0.00195 0.00627 0.00937 0.02118 3.99a
68.82%a
t-Statistic 3.85a 2.82a 3.94a 3.94a 3.62a 5.82a
Ž%. 67.07%a 57.65%b 55.29% 55.29% 55.29% 66.47%a

Panel B. Deleting extreme returns

Small 1 0.01209 0.01388 0.01741 0.01960 0.02305 2.28b
55.29%
2 0.01121 0.01219 0.01928 0.01513 0.02314 2.43b
57.06%
3 0.01071 0.01038 0.01718 0.01053 0.02199 2.41b
62.35%a
4 0.00756 0.00879 0.01005 0.01869 0.02195 3.19a
60.00%a
Large 5 0.00026 0.00445 0.00879 0.01121 0.02091 4.32a
65.29%a
t-Statistic 2.49b 1.85 2.07b 1.87 0.45 4.03a
Ž%. 62.35%a 51.77% 54.12% 52.35% 48.82% 62.35%a

This table presents mean returns for 25 size BErME portfolios of all EMDB firms. A two-pass
classification is used to create the portfolios. First, stocks are classified by relative market value of
equity Žsize., into one of five portfolios. Second, stocks within each relative market capitalization group
are further subdivided into five book-to-market relatives. Portfolios are rebalanced once per year. Mean
monthly mean returns are presented for each of the 25 portfolios. The last column presents t-statistics,
within each size group, for tests of significance of the mean difference in returns between high BErME
Žvalue. portfolio and low BErME Žgrowth. portfolio. The percent of months Ž%. in which returns are
greater for the high BErME portfolio vs. the low BErME portfolio is also reported in the last column.
Similarly, the last row presents t-statistics, within each size group, for tests of significance of the mean
difference in returns between small Ž1. portfolios and large Ž5. portfolios. The last row also reports the
percent of months in which returns are higher for small Ž1. portfolios vs. large Ž5. portfolios within each
BErME group. Significance for the percentage is indicated if the observed percentage differs signifi-
cantly from 50%. Panel A presents results using all data. Panel B presents results after deleting returns
falling in the upper and lower 1% tails of the returns distribution formed over all sampled months.
a
Significant at the 0.01 level.
b
Significant at the 0.05 level.
16 C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30

significant after removing the tails of the returns distribution. The t-statistics range
from 2.28 to 4.32. Thus, BErME effects persist both after controlling for size and
after removing extreme returns.
Table 3 also shows that the size effect is positive in all BErME quintiles. Panel
A Žfor all data. shows a significant size effect for each BErME quintile using
parametric tests. The non-parametric statistics show mixed results for the size
effect. Panel B shows that the size effect is diminished materially after removing
outliers.
Speidell and Stone Ž1997. assert that four major portfolio management ‘style’
categories have emerged in the US market: small value, small growth, large value
and large growth. They argue that it is often difficult to distinguish value vs. growth
firms in emerging markets, and they emphasize the opportunities in small stocks.
Fig. 2 shows the mean returns for all data, over the months of our study for each of
the size᎐value style combinations mentioned by Speidell and Stone Ž1997.. These
are the ‘corner’ portfolios in our analysis of five-by-five portfolio formation based
on size and BErME quintiles. The figure emphasizes that average returns for the
emerging markets were highest in the small value category and lowest in the large
growth category. Tests of the portfolio of high BErME stocks within the small size
category Žvalue portfolio within small size. against the portfolio of low BErME
stocks within the large size category Žgrowth portfolio within large size. show that
the difference in returns was highly significant. The value of the t-statistic for the
mean difference in returns between the small value and the large growth portfolio
was 5.82. The annualized return spread between the small firm, high BE portfolio
and the large firm, low BE portfolio was 87% using all data and was 31% after
removing the extreme returns.18
Fig. 3 shows that the results based on arithmetic mean returns are consistent
with compound returns. The mean differences are not merely an artifact of the
calculation of arithmetic means. Substantial differences in compound returns are
also earned in the four style portfolios that are based on extreme values of size and
BErME. Again, the portfolio consisting of value Žhigh BErME. stocks within
small firms outperforms the other combinations. Our results indicate that the

18
The inclusion of transactions costs might affect the return spreads if the turnover is significantly
different across portfolios. Since our portfolios are rebalanced only once per year, transactions costs will
be controlled to some extent. As shown by Willoughby Ž1997., transactions costs Žthe sum of commis-
sions, fees and market impact. averaged approximately 100 basis points for emerging market stocks
Žusing 1996 trades., which was double the transactions costs of US. OTC stocks for the same time
period. Bekaert et al. Ž1997. provide estimates of transactions costs using trades during 1995 for 21
emerging markets. The transactions costs averaged 110 basis points across the 21 markets. We also
replicated our portfolios using quarterly rebalancing Žwhich would involve higher transactions costs..
The annualized return spread for the quarterly rebalanced portfolios was approximately 1.4 times the
spread derived for the annually rebalanced portfolios, after removing extreme returns. Thus, a strategy
to lessen transactions costs by postponing trades from quarterly to annual would have entailed a
substantial loss of gross return spread in exchange for an also substantial reduction in transactions
costs.
 C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30 17

Fig. 2. Mean monthly returns for ‘corner’ portfolios with annual rebalancing. This figure illustrates the
differences in mean monthly returns for equities in 35 emerging markets across small-cap value,
small-cap growth, large-cap value and large-cap growth portfolios. The portfolios are rebalanced
annually. The four portfolios are the corner portfolios from the five-by-five contingency table presented
in Table 3, Panel A.

spread between small value and large growth stocks in emerging markets has grown
systematically through our sample period.19

5. Cross-sectional regression tests

As Fama and French Ž1998. point out, the results of tests of alternative factors
can be sensitive to the methodology used. In this section, we provide tests of the

19
We also re-ran the multivariate tests separately for two subperiods. The first subperiod ends
December 1992 and the second subperiod starts January 1993. Results using all data for both time
periods show statistically significant value effects within many of the size quintiles. Similarly for relative
size, using all data, we find significant size effects in most of the BErME quintiles in both periods. The
mean monthly difference between the corner portfolios equals 6.6% Ž t-statistic s 4.40. for the first
subperiod and equals 4.3% Ž t-statistic s 3.83. for the second subperiod. These findings further support
the robustness of the findings derived over the entire sample period using all data. After removing
extreme returns, we find significant size effects in the first time subperiod Žwithin four of the five
BErME quintiles., but no significant size effects in the second subperiod. After removing extreme
returns, we find significant BErME effects in both subperiods.
 18
C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30
Fig. 3. Portfolio performance based on relative size and relative book-to-market with annual rebalancing. The lines plot the compound values from
investing one dollar in June 1986 in small-cap value, large-cap value, small-cap growth and large-cap growth portfolios aggregated over 35 emerging
markets. Portfolios are rebalanced annually.
 C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30 19

effects of size and BErME via cross-sectional regressions that simultaneously

account for the effects of both factors. The tests are similar to the Fama and
MacBeth Ž1973. procedures commonly used in examining multiple factors in asset
pricing. Such tests examine the ability of each hypothesized factor to ‘explain’
returns while ‘controlling’ for the others to some degree.
We conduct cross-sectional tests on annually rebalanced portfolios. Initially, we
include only size and BErME as explanatory variables. Then, we introduce local
betas and global betas. These latter tests examine effects after accounting for the
firm’s systematic risk. For each stock, the local beta is estimated by regressing the
stock’s returns against the local market index returns, using all months for which
returns are available for the stock. For each stock, the global beta is estimated by
regressing all the stock’s returns against the Morgan Stanley Capital International
ŽMSCI. World index returns, using all months for which returns are available for
the stock. In cross-sectional regressions that include both the local and global
betas, first we orthogonalize the world index returns against returns for each local
index separately. Next, we regress the stock returns against the orthogonalized
world index returns and corresponding local market index returns to derive the
global and local betas, respectively. We required at least 30 observations to
calculate betas. Therefore, for the cross-sectional regression tests, we exclude
stocks with less than 30 months of return data.20
Our dual beta specification is motivated by Bekaert and Harvey Ž1995., who
contend that standard assumptions that capital markets are, and always were, fully
integrated or fully segmented are of very limited applicability. They develop a
framework that allows for time varying conditional market integration, and they
test their methodology on market indexes from both developed countries and
emerging markets. Their results provide evidence of time varying integration for a
number of countries. During our sample period, a number of emerging market
countries in our sample liberalized their economies, including the increased open-
ing of their securities markets to global investors. As more foreign investors began
to invest directly in securities in those markets, or in securities from those markets
that became listed on global exchanges, it is likely that the world beta would
become increasingly important.21
There are various criticisms of cross-sectional regression tests. One is that the
tests assume that the relationship between the dependent variable and indepen-

20
Duarte and Mendes Ž1997. demonstrate the potential effects of outlier returns on beta estimates
for Emerging Market stocks. Moreover, infrequent trading causes known biases for individual stock beta
estimation Žsee Scholes and Williams, 1977 and Dimson, 1979.. Our portfolio approach Žaveraging the
individual betas. should diminish these effects.
21
A weakness of our regression is that it forces the same cross-sectional slope coefficient on the
characteristics across the 25 portfolios. For example, for the more integrated countries, we would expect
the world beta to be important. But our portfolio formation and regression approach will not distinguish
among the country effects. The smoothing methods may obscure the differential effects of the betas
across countries.
20 C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30

dent variables is of a specific functional form Žin this case, linear.. If the relation-
ship is not of that form, then the tests will lose power to establish factor effects. A
second criticism is that outliers can drive regression tests, i.e. extreme observations
will tend to have a large effect on the outcomes of the tests. Knez and Ready
Ž1997. suggest robust regression procedures for observing the effects of such
outliers. Extreme returns are more prevalent in emerging markets than in devel-
oped markets and, therefore, may have a greater effect on the findings.
The regression tests proceed as follows. For each month, we run a cross-sec-
tional regression using various versions of the following model:

R pt s ␭ 0 t q ␭ 1t REL Ž SIZE p1 . q ␭ 2 t REL Ž BE r ME p7 . q ␭ 3t Local ␤ p

q ␭ 4tGlobal ␤ p q ␧ pt Ž2.

where R pt is the return to portfolio p in month t. ␭ 1t is the slope associated with

firm size, ␭ 2t is the slope associated with book-to-market value, ␭ 3t is the slope
associated with local beta and ␭ 4t is the slope associated with global beta. The time
subscripts on SIZE and BErME correspond to the lagged month Ž1 for size and 7
for BErME. relative to the beginning month in the 12-month holding period. We
run each cross-sectional regression using the following explanatory variables: Ž1.
size and BErME; Ž2. size, BErME and local beta; and Ž3. size, BErME, local
beta and global beta. For these tests we define BErME and size for each firm
relative ŽREL. to each firm’s local market. For each version, a total of 170
regressions are performed Žone regression for each month, June 1986᎐July 2000..
Each regression Žfor any month t. has 25 observations, representing the returns
Žfor month t. for each of the 25 portfolios. The monthly estimates of lambda values
are averaged over the 170 months, and t-statistics are used to test for significance.
The t-statistics are calculated from the means and standard deviations across the
170 lambda values for a given factor.
Table 4 presents the mean lambdas. The values in parentheses are t-statistics.
The last two columns present the percent of months in which the size or BErME
lambda estimates were positive.
Consistent with our earlier multivariate results using all data ŽPanel A., the
monthly size and BErME premiums are significant at the 0.01 level Ž t-statistics s
y2.68 for size and 6.10 for BErME.. However, the percentage of positive monthly
size premiums Ž ␭ 1 . is not significantly different from 50% at the 0.05 level
Žpercentage s 56.47%, t-statistic s 1.69.. In contrast, the percentage of positive
BErME premiums is significantly different from 50% Žpercentage s 71.76%,
t-statistic s 5.67.. Thus, from the cross-sectional tests, the BErME effect is more
persistent over time vs. the size effect for emerging markets.
The signs and significance of the size and BErME slope coefficients remain
unchanged in the three and four factor regressions that control for local and global
systematic risk. Furthermore, the magnitudes of the size and BErME coefficients
Table 4
Risk premiums from cross-sectional regressions

Intercept REL Žsize. REL Local ␤ Global ␤ % Pos. size % Pos. BErME
ŽBErME. premiums premiums

Panel A. All data

Rpt s ␭0t q ␭lt RELŽSizep . q ␭2t RELŽBErMEp . q ␭3t Local ␤ q ␭4tGlobal ␤ q ␧pt
y0.0063 71.76%a

C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30

0.0121 0.0109 43.53%
Ž2.76a . Žy2.68a . Ž6.10a .
0.0106 y0.0065 0.0107 0.0148 45.29% 70.59%a
Ž2.20b . Žy2.48b . Ž5.90a . Ž0.91.
0.0048 y0.0062 0.0107 0.0243 0.0079 44.12% 69.41%a
Ž1.05. Žy2.65a . Ž5.96a . Ž1.36. Ž1.15.
Panel B. Extreme returns deleted
Rpt s ␭0 q ␭1 RELŽSizep . q ␭2 RELŽBE r MEp . q ␭3 Local ␤ q ␭4 Global ␤ q ␧p
0.0103 y0.00266 0.0059 46.47% 62.94%a
Ž2.82a . Žy1.51. Ž3.37a .
0.01045 y0.0029 0.0054 0.0041 47.65% 61.76%a
Ž2.66a . Žy1.41. Ž3.98a . Ž0.49.
0.0043 y0.0035 0.0055 0.0040 0.0117 46.47% 61.76%a
Ž1.25. Žy1.71. Ž4.17a . Ž0.41. Ž2.03b .

This table reports risk premiums estimated from versions of the cross-sectional regression: R pt s ␭ 0 t q ␭ 1t RELŽSize p . q ␭ 2 t RELŽBE r ME p . q
␭ 3t Local ␤ p q ␭ 4tGlobal ␤ p q ␧ pt . Panel A presents results using all data for portfolios rebalanced annually. Panel B presents results after deleting
returns falling in the upper and lower 1% tails of the returns distribution formed over all sampled months. Portfolios are formed on the basis of relative size
and relative BErME and are rebalanced once per year. This process is repeated every year. Thus, 12 Žmonthly. cross-sectional regressions are run before
stocks are reclassified and before the portfolio characteristics are recalculated. The dependent variable in each regression is the vector of monthly returns
for the 25 portfolios. A total of 170 monthly cross-sectional regressions are run ŽJune 1986᎐July 2000.. The Local ␤ p is the mean local market beta for
stocks included in portfolio p. Each stock’s local beta is calculated as the slope from the regression of the stock’s returns against the local market returns,
using all available months for the firm. The Global ␤ p is the mean global beta for stocks included in portfolio. In cross-sectional regressions that include
both the local and global betas, first we orthogonalize the world index returns against returns for each local index separately. Next, we regress the stock
returns against the orthogonalized world index returns and corresponding local market index returns to derive the local and global betas, respectively.
Stocks are included that had at least 30 months reported returns in the EMDB. The table presents the lambda estimates averaged over all months.
t-Statistics are reported in parentheses. The last two columns present the percent of months in which the lambda estimate for size and for BErME were

21
positive, respectively. Significance for the percentage is indicated if the observed percentage differs significantly from 50%.
b
Significant at the 0.01 level.
a
Significant at the 0.05 level.
22 C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30

are similar across regressions. Significant size and BErME effects are found in
each regression, suggesting that size and BErME effects are not captured by beta.
Neither local nor global beta are priced at the 0.05 level of significance. These
findings are consistent with Heston et al. Ž1999. who find global size effects for the
aggregate of 12 European stock markets during 1980᎐1995. They find that small
firms outperform large firms while controlling for global beta effects. Our findings
are also consistent with Rouwenhorst Ž1999. who documents value and size effects,
but finds no local beta effects for emerging markets. Rouwenhorst’s results are
based on univariate single-factor tests similar to our tests reported in our Table 2.
In contrast, our multivariate methodology used in Tables 3 and 4 is based on 25
portfolios formed on the basis of both size and BErME. As we explained earlier,
the multivariate approach provides a cleaner test of the independent effects of firm
size and BErME.
We replicated all tests for individual markets. From the cross-sectional tests
using all data, we found that 72% of the individual markets had a positive BErME
effect Ži.e. higher returns for larger values of BErME., and 56% had a negative
size effect Ži.e. had higher returns for smaller firms.. The percentage is significantly
different from 50% for BErME, but it is not statistically significant for size. There
is a lot of noise present in the returns, which will greatly affect country-specific
returns and observed factors. A major limitation of country-by-country tests is that
there are not a sufficient number of stocks in the separate markets with which to
reliably form the 25 portfolios used in the cross-sectional regression tests. This
limitation is especially acute during the earlier years in the sample. Consequently,
the country-by-country results should be interpreted with caution.
We replicated the cross-sectional regression tests from Eq. Ž2. after deleting
returns falling in the upper and lower 1% of the returns distribution formed over
all sampled months. Results are presented in Panel B of Table 4. The monthly size
premium Ž ␭ 1 . falls dramatically Žfrom 0.0063 to y0.0027. and is no longer
statistically significant Ž t-statistic s y1.51.. These findings are consistent with our
earlier results suggesting that the size effect was driven by extreme returns.
Although the sign remains negative, none of the size premiums in the cross-sec-
tional regressions are statistically significant at the 0.05 level after removing
outliers. In contrast, all the BErME premiums Ž ␭ 2 . remain positive and statisti-
cally significant at the 0.01 level. The global beta premium becomes significant at
the 0.05 level in the four-factor regression. Apparently, the removal of the extreme
returns reduced the standard deviation of the beta premium estimates, causing the
t-statistic for the global beta premium to rise. These global beta findings are
consistent with Heston et al. Ž1999. who also find significant global beta effects.
The non-parametric sign tests reported in the last two columns confirm that the
size effect is not significant but that the BErME effect is significant.
Overall, the cross-sectional tests in Table 4 are consistent with the results
presented in Table 3. Furthermore, the cross-sectional tests show the BErME
effects persist after accounting for size and local and global beta effects. These
findings are consistent with Arshanapalli et al. Ž1998. who show that superior
performance within 18 developed global equity markets for value stocks is not due
 C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30 23

to their greater sensitivity to market movements compared with growth stocks.

They also find value effects independent of size effects. 22 Moreover, we also find
significant BErME effects in tests both with and without extreme returns.

6. Tests based on absolute size

Results based on relative size are appropriate if capital markets are segmented.
If global capital markets are fully integrated, however, there is no basis for
adjusting size for the average size in a market. Thus, as a further test of the
robustness of the earlier results we have reported, in this section we report the
results of tests that employ absolute size and relative BErME. The use of relative
BErME is dictated by differences in accounting measurement of book values
across markets and not on the degree of segmentation or integration across
markets. However, the use of absolute size vs. relative size is not mandated by any
measurement concerns. Since we do not know the extent of capital market
integration globally, especially for emerging markets, we include tests here based
on absolute size.
The results are presented in Tables 5᎐7 and correspond to the previous results
based on relative size that were presented in Tables 2᎐4, respectively. The results
are markedly different for absolute size than they were for relative size. For
example, consider the univariate results for size presented in Panel A of Table 5.
There is no significant size effect for the parametric test using all data Žalthough
the difference between small and large stock portfolios is 90.8 basis points on a
monthly basis, which annualizes to approx. 11%.. The t-statistic is 1.81, which is
just short of statistical significance at the 0.05 level. However, the non-parametric
results are significant: small stocks had higher returns in 59.41% of the observation
months. Deleting extreme returns, there was no size effect. In fact, the sign of
small stock average returns minus large stock average returns became negative
after removing extreme returns.
Table 6 presents the results for size portfolios within BErME portfolios and vice
versa. In Panel A, showing results for all data, we observe significant size effects
based on t-tests Žand in the expected direction. within two BErME quintiles,
specifically the extreme value portfolios and extreme growth portfolios, but not for
the more blended BErME portfolios. Results deleting extreme observations are
shown in Panel B: There, we observe no significant size effects among any of the
BErME categories except for the sign test within the fourth BErME category,
and in that case the sign is negative, i.e. large stocks outperformed small stocks.
Thus, the evidence of a size effect based on absolute size is weak at best in tests
using average returns from size portfolios. In contrast, significant value effects are
present within every size category based on t-tests and sign tests.
Table 7 presents the results of cross-sectional regressions when size is measured

22
Bauman et al. Ž1998. find value effects in all but the smallest firm size category for a sample of
developed markets.
24 C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30

Table 5
Univariate tests of market capitalization and book-to-market effects

Small 1 2 3 4 Large 5 t-Statistic % Small ) large

Panel A. Mean monthly returns by firm size quintile

All data 0.02509 0.01883 0.01811 0.01433 0.01601 1.81 59.41%b
Removing 0.01287 0.01227 0.01373 0.01334 0.01600 y0.82 51.18%
extremes

Panel B. Mean monthly returns by book-to-market quintile

Low 1 2 3 4 High 5 t-Statistic % Value ) growth
Žgrowth. Žvalue.
All data 0.01041 0.013053 0.01566 0.02372 0.03628 5.71a 74.19%a
Removing 0.00790 0.011567 0.01290 0.01609 0.02312 4.69a 61.77%a
extremes

This table reports mean monthly returns for the emerging market sample for various market value
equity Žsize. and book-to-market ŽBErME. equity quintiles. Portfolios are rebalanced once per year.
Panel A presents mean monthly returns for each market value of equity Žsize. quintile. Group 1 consists
of the smallest size firms and group 5 consists of the largest size firms. Panel B presents mean monthly
returns for each relative BErME quintile. The t-statistics in Panel A test for differences between the
smallest and largest size portfolios. The last column reports the percentage of months for which the
return on the small portfolio exceeded that of the large portfolio. Similarly, the t-statistics in Panel B
test for differences between the value Žhigh BErME. and growth Žlow BErME. portfolios. The last
column reports the percentage of months for which the return on the value stock portfolio exceeded
that of the growth stock portfolio. Tests of significance on the difference between the observed
percentage and 50% are also performed.
a
Significant at the 0.01 level.
b
Significant at the 0.05 level.

by absolute size. Panel A shows results for all data, and Panel B shows the results
with extreme observations omitted. In the results with all data, size is not signifi-
cant in any of the regressions. BErME effects continue to be highly significant in
all of the regressions Ž t-statistics are y0.89, y0.60 and y0.03 in the three
cross-sectional regressions.. Panel B provides results with extreme observations
omitted. Again, size effects are not significant in any of the regressions, and
BErME effects are significant in every regression. The sign tests confirm the
results of the regression t-tests.
We can contrast the results in this section using absolute size against results in
the previous two sections based on relative size. When size is measured relative to
other securities in the same market, size effects appear to be present Žwhen all
data are used.. When size is measured in absolute terms, there do not appear to be
significant size effects. The results suggest Žbut do not show. that the relative lack
of integration of emerging markets with global capital markets may limit access of
global investors to the smaller securities of the emerging markets. This is consis-
tent with the observation that it is generally the larger and more visible securities
of emerging markets that issue American or Global Depositary Receipts ŽADRs or
GDRs., thus easily providing foreign investors with access to those securities and
to information regarding those securities. To the extent that it is more difficult or
 C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30 25

Table 6
Mean monthly returns for size BErME portfolios

Size Book-to-market quintiles t-Statistic

quintiles Low 1 2 3 4 High 5 %
Žgrowth. Žvalue.

Panel A. All data

Small 1 0.01965 0.01335 0.02177 0.03332 0.05479 4.33a
57.06%
2 0.00800 0.01220 0.02081 0.02462 0.02794 3.31b
60.59%a
3 0.00737 0.01493 0.01580 0.02084 0.03122 4.04a
62.94%a
4 0.00193 0.00960 0.01426 0.01831 0.02733 3.20a
57.65%b
Large 5 0.00738 0.01315 0.00119 0.02077 0.02735 2.96a
61.18%a
t-Statistic 2.08b 0.03 1.27 1.60 2.82a 5.21a
% 58.82%b 52.94% 51.18% 52.94% 54.71% 64.12%a

Panel B. Deleting extreme returns

Small 1 0.01188 0.00937 0.01136 0.01488 0.02706 3.18a
61.12%a
2 0.00533 0.01056 0.01203 0.01542 0.01756 2.75a
58.24%b
3 0.00637 0.01102 0.01593 0.01391 0.02180 3.54a
64.12%a
4 0.00286 0.01112 0.01570 0.01604 0.02140 3.15a
56.47%
Large 5 0.00746 0.01519 0.01075 0.02074 0.02628 3.24a
61.12%a
t-Statistic 0.85 y1.09 0.11 y1.05 013 3.23a
% 51.77% 50.00% 51.77% 42.35%b 47.65% 60.59%a

This table presents mean returns for 25 size BErME portfolios of all EMDB firms. A two-pass
classification is used to create the portfolios. First, stocks are classified by market value of equity Žsize.,
into one of five portfolios. Second, stocks within each market capitalization group are further subdivided
into five book-to-market relatives. Portfolios are rebalanced once per year. Monthly mean returns are
presented for each of the 25 portfolios. The last column presents t-statistics, within each size group, for
tests of significance of the mean difference in returns between high BErME Žvalue. portfolio and low
BErME Žgrowth. portfolio. The percent of months Ž%. in which returns are greater for the high
BErME portfolio vs. the low BErME portfolio is also reported in the last column. Similarly, the last
row presents t-statistics, within each size group, for tests of significance of the mean difference in
returns between small Ž1. portfolios and large Ž5. portfolios. The last row also reports the percent of
months in which returns are higher for portfolio Ž1. vs. Ž5. within each BErME group. Significance for
the percentage is indicated if the observed percentage differs significantly from 50%. Panel A presents
results using all data. Panel B presents results after deleting returns falling in the upper and lower 1%
tails of the returns distribution formed over all sampled months.
a
Significant at the 0.01 level.
b
Significant at the 0.05 level.
Table 7

26
Risk premiums from cross-sectional regressions

Intercept LN Žsize. REL Local ␤ Global ␤ % Pos. size % Pos. BErME

ŽBErME. premiums premiums

Panel A. All data

Rpt s ␭0t q ␭lt LNŽSizep . q ␭2t RELŽBErMEp . q ␭3t Local ␤ q ␭4tGlobal ␤ q ␧pt
0.0139 y0.00109 0.0118 48.82% 67.65%a
Ž2.18b . Žy0.89. Ž6.07a .

C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30

0.0137 y0.00074 0.0115 y0.0102 48.83% 65.29%a
Ž1.86. Žy0.60. Ž5.74a . Žy0.55.
0.0098 y0.00003 0.0109 0.0054 y0.0015 55.29% 67.06%a
Ž1.38. Žy0.03. Ž5.57a . Ž0.25. Žy0.20.

Panel B. Extreme returns deleted

Rpt s ␭0 q ␭1 LNŽSizep . q ␭2 RELŽBE r MEp . q ␭3 Local ␤ q ␭4 Global ␤ q ␧p
0.0034 0.0010 0.0071 51.76% 60.59%a
Ž0.73. Ž1.02. Ž5.10a .
0.0012 0.0011 0.0068 0.0137 52.94% 59.41%b
Ž0.24. Ž1.17. Ž4.74a . Ž1.45.
0.0013 0.0006 0.0067 0.0230 0.0025 57.06% 62.35%a
Ž0.30. Ž0.70. Ž4.80a . Ž2.18b . Ž0.47.

This table reports risk premiums estimated from versions of the cross-sectional regression: R pt s ␭ 0 t q ␭ 1t LNŽSize p . q ␭ 2 t RELŽBE r ME p . q
␭ 3t Local ␤ p q ␭ 4tGlobal ␤ p q ␧ pt . Panel A presents results using all data for portfolios rebalanced annually. Panel B presents results after deleting
returns falling in the upper and lower 1% tails of the returns distribution formed over all sampled months. Portfolios are formed on the basis of size and
relative BErME and are rebalanced once per year. Thus, 12 Žmonthly. cross-sectional regressions are run before stocks are reclassified and before the
portfolio characteristics are recalculated. The dependent variable in each regression is the vector of monthly returns for the 25 portfolios. A total of 170
monthly cross-sectional regressions are run ŽJune 1986 through July 2000.. The Local ␤ p is the mean local market beta for stocks included in portfolio p.
Each stock’s local beta is calculated as the slope from the regression of the stock’s returns against the local market returns, using all available months for
the firm. The Global ␤ p is the mean global beta for stocks included in portfolio. In cross-sectional regressions that include both the local and global betas,
first we orthogonalize the world index returns against returns for each local index separately. Next, we regress the stock returns against the orthogonalized
world index returns and corresponding local market index returns to derive the local and global betas, respectively. Stocks are included that had at least 30
months reported returns in the EMDB. The table presents the lambda estimates averaged over all months. t-Statistics are reported in parentheses. The last
two columns present the percent of months in which the lambda estimate for size and for BErME were positive, respectively. Significance for the
percentage is indicated if the observed percentage differs significantly from 50%.
a
Significant at the 0.01 level.
b
Significant at the 0.05 level.
 C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30 27

more costly for foreigners to access the smaller securities of emerging markets, it
would not be surprising that the smaller emerging market securities would fail to
provide excess returns at a global level. However, our results on this point are
merely suggestive. It may be an interesting avenue for further investigation.

7. Summary and conclusions

This paper provides new and important evidence on size and book-to-market
value ŽBErME. effects by examining data from emerging capital markets. Our
tests offer an out-of-sample investigation of the importance of size and value
factors as determinants of subsequent portfolio returns. Our approach differs from
existing research by focusing on the robustness of the results. We examine US
dollar-denominated stock returns over the 1985᎐2000 period for stocks listed in the
35 emerging equity markets for which data were available from the Standard and
Poor’s Emerging Markets Data Base. We provide various tests for the effects of
the size and value factors. Earlier work has led to differences in results based on
methodology. Earlier work has also shown that returns in emerging markets are
non-normal and has suggested that results may be sensitive to extreme returns.
Accordingly, we examine size and value effects using a variety of methods.
A difference between our work and that of earlier researchers is that we perform
tests employing measures of size and BErME defined relative to each firm’s local
market average. The use of relative measures facilitates the aggregation of data
across markets instead of being limited to examining data on a market-by-market
basis. The use of relative BErME is motivated by the fact that accounting systems
differ widely across emerging markets. For instance, accounting standards in some
emerging markets specifically address inflation in the measurement of book values
and some do not. Accordingly, a given BErME ratio may have a very different
interpretation in one market than in another. The use of relative BErME
addresses that concern.
The use of relative size is motivated by considerations of whether the emerging
markets that we study are fully integrated with global capital markets. If not, and if
there is nevertheless a size effect within markets, the use of absolute size could
obscure the size effect. On the other hand, if emerging markets are fully integrated
with global markets, then size is size and we would recommend the use of an
absolute size measure.
Our results point to a robust BErME effect that is not driven by extreme
returns in emerging markets. The size effect is not robust to removal of extreme
returns for emerging market equities. BErME effects were pervasive through:
parametric tests; non-parametric tests; and after controlling for size, outliers and
local and global systematic risk. Additional tests indicate a lack of a size effect
when absolute rather than relative size is used.
Size and value factors have generated a great deal of research interest in recent
years. Emerging markets provide a distinct setting in which to develop new
evidence about the factors. The results presented here add to the body of
28 C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30

knowledge about these factors and their roles in security returns. These results
weaken arguments that BErME effects may be due to data mining in the US
market. But they do not resolve once and for all the issues that they address: They
raise a number of issues that remain for future research. One of them is the extent
to which size effects are associated with the degree of integration of emerging
markets with developed markets.
Jensen et al. Ž1997. find that size and value effects depend heavily on the
monetary policy stance of the Federal Reserve: e.g. small firm and high BErME
effects are significant only in expansive monetary policy periods. Since emerging
markets tend to have changes in direction of economic policy that are more
extreme than in developed markets, these changes could presumably explain some
of the extreme returns. This is particularly relevant for emerging markets in which
extreme returns constitute over 25% of the total return. Are small, large, value, or
growth stocks affected more or differently by these shifts than are other firms? To
what extent are emerging market equity returns and investment style shifts pre-
dictable? To what extent can multi-factor models be used to enhance emerging
market portfolio returns?
Relatively little is known about the effects that portfolio flows have upon
extreme returns in emerging markets. There is a tendency for foreign investors to
prefer holding the stocks of larger firms in emerging markets since information is
usually more readily available and they have greater liquidity. Do those factors
interact in affecting the extreme returns? Do they perform in a systematic way, i.e.
do they tend to cause unusually positive or negative outcomes for particular classes
of securities?
This paper has examined an important database of emerging market equities to
determine whether size and book-to-market effects are present in those markets.
The evidence strongly supports the presence of a book-to-market effect of the
same kind as in other markets Ži.e. the so-called ‘value’ stocks tend to outperform
‘growth’ stocks., but the evidence is considerably weaker regarding a size effect.
We look forward to subsequent research that sheds light on the economic forces
that underlie our findings and implications.

Acknowledgements

The authors are grateful to Ken French, Tim Loughran, Bijoy Sahoo, Steve
Mann, Andy Waisburd and an anonymous referee for helpful comments on earlier
drafts of this paper.

References
Achour, D., Campbell, H., Hopkins, G., Lang, C., 1998. Stock selection in emerging markets: portfolio
strategies for Malaysia, Mexico and South Africa. Emerging Markets Q. 2, 39᎐91.
Achour, D., Campbell, H., Hopkins, G., Lang, C., 1999a. Stock selection in Malaysia. Emerging Markets
Q. 3, 54᎐91.
 C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30 29

Achour, D., Campbell, H., Hopkins, G., Lang, C., 1999b. Stock selection in Mexico. Emerging Markets
Q. 3, 1᎐38.
Achour, D., Campbell, H., Hopkins, G., Lang, C., 1999c. Firm characteristics and investment strategies
in Africa: The case of South Africa. Afr. Finance J. 1, 1᎐39.
Arshanapalli, B., Coggin, T.D., Doukas, J., 1998. Multifactor asset pricing analysis of international value
investment strategies. J. Portfolio Manage. 24, 10᎐23.
Banz, R.W., 1981. The relationship between return and market value of common stocks. J. Financ.
Economics 9, 3᎐18.
Basu, S., 1997. The investment performance of common stocks in relation to their price-earnings ratios:
A test of the efficient market hypothesis. J. Finance 32, 663᎐682.
Basu, S., 1983. The relationship between earnings’ yield, market value and return for NYSE common
stocks: further evidence. J. Financ. Economics 12, 129᎐156.
Bauman, W.S., Conover, C.M., Miller, R.E., 1998. Growth versus value and large-cap stocks in
international markets. Financ. Analysts J. 54, 75᎐89.
Bekaert, G., Harvey, C.R., 1995. Time-varying world market integration. J. Finance 50, 403᎐444.
Bekaert, G., Erb, C.B., Campbell, R.H., Viskanta, T.E., 1997. What matters for emerging equity market
investments. Emerging Markets Q. 1, 17᎐46.
Berk, J.B., 1995. A critique of size-related anomalies. Rev. Financ. Stud. 8, 275᎐286.
Berk, J.B., 1997. Does size really matter? Financ. Analysts J. 53, 12᎐18.
Berk, J.B., 2000. Sorting out sorts. J. Finance 55, 407᎐427.
Black, F., 1993. Beta and return. J. Portfolio Manage. 20, 8᎐18.
Black, F., 1995. Estimating expected return. Financ. Analysts J. 51, 168᎐171.
Bossaerts, P., Fohlin, C., 2000. ‘Has the cross-section of average returns always been the same?
Evidence from Germany, 1888᎐1913,’ Social Science Working Paper 1084, Division of the Humani-
ties and Social Sciences, California Institute of Technology, July, 2000.
Brennan, M.J., Chordia, T., Avanidhar, S., 1998. Alternative factor specification, security characteristics
and the cross-section of expected stock returns. J. Financ. Economics 49, 345᎐373.
Chan, K.C., Karceski, J., Lakonishok, J., 1998. The risk and return from factors. J. Financ. Quantitative
Anal. 33, 159᎐188.
Chen, N.-F., Zheng, F., 1998. Risk and return of value stocks. J. Bus. 71, 501᎐535.
Chui, A., Wei, K.C.J., 1998. Book-to market, firm size, and the turn-of-the-year effect: Evidence from
Pacific-Basin emerging markets. Pacific Basin Finance J. 6, 275᎐293.
Claessens, S., Dasgupta S., Glen J., 1993. Stock price behavior in emerging markets. Paper presented at
the World Bank Symposium, Portfolio Investment in Developing Countries.
Claessens, S., Dasgupta, S., Glen, J., 1995. Return behavior in emerging stock markets. World Bank
Econ. Rev. 9, 131᎐151.
Claessens, S., Dasgupta, S., Glen, J., 1998. The cross-section of stock returns: evidence from the
emerging markets. Emerging Markets Q. 2, 4᎐13.
Cohen, R.B., Polk, C.K., 1997. The impact of industry factors in asset-pricing tests. Working Paper,
University of Chicago.
Cook, T.J., Rozeff, M.S., 1984. Size and earningsrprice ratio anomalies: one effect or two? J. Financ.
Quantitative Anal. 19, 427᎐466.
Daniel, K., Titman, S., 1997. Evidence on the characteristics of cross sectional variation in stock returns.
J. Finance 52, 1᎐33.
Davis, J., 1994. The cross-section of realized stock returns: the pre-Compustat evidence. J. Finance 49,
1579᎐1593.
Davis, J., Fama, E.F., French, K.R., 2000. Characteristics, covariances and average returns: 1929᎐1997.
J. Finance 55, 389᎐406.
De Bondt, W.F.M., Thaler, R., 1985. Does the market overreact? J. Finance 40, 793᎐805.
De Bondt, W.F.M., Thaler, R., 1987. Further evidence on investor overreaction and stock market
seasonality. J. Finance 42, 557᎐581.
Dimson, E., 1979. Risk measurement when share are subjected to infrequent trading. J. Financ.
Economics 7, 197᎐226.
Duarte Jr., A.M., Vaz de Melo Mendes, B., 1997. Robust estimation of systematic risk in emerging stock
markets. Emerging Markets Q. 1, 85᎐95.
30 C.B. Barry et al. r Emerging Markets Re¨ iew 3 (2002) 1᎐30

Durham, J.B., 2000. Extreme bound analysis of emerging stock market anomalies. J. Portfolio Manage.
26, 95᎐103.
EMDB, 2000. ‘Emerging Markets Data Base: EMDB 2000’, Standard and Poors Corporation, New
York, Version 6.0.
Emerging Stock Markets Factbook 2000. Standard & Poor’s, New York.
Fama, E.F., French, K.R., 1992. The cross-section of expected stock returns. J. Finance 47, 427᎐465.
Fama, E.F., French, K.R., 1995. Size and book to market factors in earnings returns. J. Finance 50,
131᎐155.
Fama, E.F., French, K.R., 1996. Multifactor explanations of asset pricing anomalies. J. Finance 51,
55᎐84.
Fama, E.F., French, K.R., 1998. Value versus growth: the international evidence. J. Finance 53,
1975᎐1999.
Fama, E.F., MacBeth, J.D., 1973. Risk, return and equilibrium: empirical tests. J. Polit. Economy 81,
607᎐636.
Hart, V.D.J., Slagter, E., van Dijk, D., 2001. Stock selection strategies in emerging markets. Working
Paper, Econometric Institute. Erasmus University.
Hawawini, G., Keim, D., 1991. On the predictability of common stock returns: worldwide evidence.
Finance. North-Holland, Amsterdam.
Herrera, M.J., Lockwood, L.J., 1994. The size effect in the Mexican stock market. J. Banking Finance
18, 621᎐632.
Heston, S.L., Rouwenhourst, K.G., Wessels, R., 1999. The role of beta and size in the cross-section of
European stock returns. Eur. Finance J. 5, 9᎐27.
Jensen, G.R., Johnson, R.R., Mercer, J.M., 1997. New evidence on size and price-to-book effects in
stock returns. Financ. Analysts J. 53, 34᎐42.
Keim, D.B., 1983. Size-related anomalies and stock return seasonality: further empirical evidence. J.
Financ. Economics 12, 13᎐32.
Kim, D., 1997. A reexamination of firm size, book-to-market and earnings price in the cross-section of
expected stock returns. J. Financ. Quantitative Anal. 32, 463᎐489.
Knez, P.J., Ready, M.J., 1997. On the robustness of size and book-to-market in cross-sectional
regressions. J. Finance 52, 1355᎐1382.
Loughran, T., 1997. Book-to-market across firm size, exchange and seasonality: Is there an effect? J.
Financ. Quantitative Anal. 32, 249᎐268.
MacKinlay, A.C., 1995. Multifactor models do not explain deviations from the CAPM. J. Financ.
Economics 38, 3᎐28.
Patel, S.A., 1998. Cross-sectional variation in emerging markets equity returns, January 1988᎐March
1997. Emerging Markets Q. 2, 57᎐70.
Reinganum, M.R., 1981. Misspecification of capital asset pricing: empirical anomalies based on
earnings’ yields and market values. J. Financial Economics 9, 19᎐46.
Reinganum, M.R., 1982. A direct test of Roll’s conjecture on the firm size effect. J. Finance 37, 27᎐35.
Roll, R., 1981. A possible explanation of the small firm effect. J. Finance 36, 879᎐888.
Rosenberg, B., Reid, K., Lanstein, R., 1985. Persuasive evidence of market inefficiency. J. Portfolio
Manage. 11, 9᎐17.
Rouwenhorst, K.G., 1999. Local return factors and turnover in emerging stock markets. J. Finance 54,
1439᎐1464.
Scholes, M., Williams, J., 1977. Estimating betas from nonsynchronous data. J. Financ. Economics 5,
309᎐327.
Sharpe, W.F., 1992. Asset allocation: management style and performance measurement. J. Portfolio
Manage. 18, 7᎐19.
Shumway, T., 1997. The delisting bias in CRSP data. J. Finance 52, 327᎐340.
Shumway, T., Warther, V.A., 1999. The delisting bias in CRSP’s Nasdaq data and its implications for the
size effect. J. Finance 54, 2361᎐2379.
Solnik, B., 2000. International Investments, 4th edition Addison᎐Wesley, New York.
Speidell, L.S., Stone, D., 1997. The case for global small stocks. Emerging Markets Q. 2, 75᎐83.
Willoughby, J., 1997. Trade secrets. Institutional Investor November, 69᎐75.