Emerging Markets Review: Antonina Waszczuk

Emerging Markets Review 15 (2013) 186210
Contents lists available at SciVerse ScienceDirect
Emerging Markets Review

journal homepage: www.elsevier.com/locate/emr
A risk-based explanation of return

patternsEvidence from the Polish stock market
Antonina Waszczuk
European University Viadrina Frankfurt (Oder), Grosse Scharrnstr. 59, 15023 Frankfurt (Oder), Germany
a r t i c l e i n f o a b s t r a c t
Article history: Using both sorting and cross-sectional tests, this paper investigates the
Received 30 November 2011 patterns in the average stock returns related to stock fundamentals, past
Received in revised form 18 October 2012 return performance, idiosyncratic risk, and turnover in the Polish equity
Accepted 16 December 2012
market for the period 20022011. To examine the persistence of
Available online 22 December 2012
the patterns, we apply the Monotonic Relation test of Patton and
Timmermann (2010). The results favour the book-to-market ratio as a
Keywords:
determinant of the cross-sectional variation of stock returns while
CAPM anomalies
Return patterns
momentum remains insignicant. The Fama and French (1993) three-
Warsaw Stock Exchange factor model, which uses local size and value risk premiums adjusted for
Three-factor model the skewed size distribution of the sample, captures most of the recognised
Momentum anomalies. Further, we show that Polish domestic SMB and HML factors
Local risk factors are not correlated with their U.S. and German counterparts.
2012 Elsevier B.V. All rights reserved.
Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
2. Data and methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
3. Empirical ndings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
3.1. Patterns in average stock returns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
3.2. FamaMacBeth (1973) cross-sectional test . . . . . . . . . . . . . . . . . . . . . . . . 201
3.3. A (multifactor) risk-based explanation of return patterns . . . . . . . . . . . . . . . . . 204
3.4. Relation between domestic (Polish) and global risk factors . . . . . . . . . . . . . . . . . 206
4. Summary and conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
This study was nancially supported by the Graduate College Risk Analysis in Baltic States and Central and Eastern Europe of
the European University Viadrina in Frankfurt (Oder). The author also thanks the participants of the Sixth International Conference
on Money, Investment and Risk organised by Nottingham Business School in April 2011 and Sven Husmann, Michael Soucek and
Micha Przykucki for their useful comments and support. The author also thanks the anonymous reviewer for further suggestions
which surely enriched the paper.
E-mail address: euv36052@europa-uni.de.
1566-0141/$ see front matter 2012 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.ememar.2012.12.002
A. Waszczuk / Emerging Markets Review 15 (2013) 186210 187
1. Introduction
For over twenty years, detecting the determinants of disperse cross-sectional performance of stocks
has been one of the most rapidly expanding areas in the asset pricing literature. The popularity of this
branch of modern nance is at least twofold. First, numerous patterns in average stock returns, i.e., the
dependencies between stock characteristics and stock returns, have been documented since the early
eighties, and the research in this eld still continues. 1 Second, a failure of the empirical tests on the Capital
Asset Pricing Model (CAPM) questioned the importance of mean-variance based beta factor as a unique
risk measure and intensied the debate about the multidimensional character of risk. 2 The resulting
incorporation of additional arbitrage premium-based risk factors into asset pricing models proposed by,
e.g., Fama and French (1993) or Carhart (1997), considerably weakened the dominance of the CAPM as a
pricing model. Recently, numerous studies use the three-factor model as a return-generating process for
the U.S. equity market. 3
The empirical success of the multifactor models lacks, however, a theoretical foundation and
therefore remains a controversial and intensively discussed issue. The risk-based explanation for the
existence of predictable patterns in average returns relies on disperse sensitivity of stocks (portfolios) to
underlying risk factors. It also concentrates on linking of risk factors mimicking premiums with
fundamental risk sources. At the same time, many scholars argue that market inefciency and
behavioural biases are responsible for the existence of return patterns. Yet another reason might be
market microstructure frictions or data snooping. To exclude the latter argument and deliver further
robustness tests, many out-of-sample studies that use non-U.S. data were published. Firstly, researchers
concentrated primarily on advanced markets and delivered evidence for the common risk factors in, e.g.,
Japan, United Kingdom (Chan et al., 1998) or for a sample of few European markets (Heston et al., 1995;
Rouwenhorst, 1998). With time, the research on emerging markets (EMs) developed, as they were
considered to be an attractive independent sample. In general, as shown in Table 1, these studies deliver
results that are consistent with the U.S. evidence documenting excess returns that are related to
attributes such as size, value, liquidity or momentum.
EMs attract international investors for two reasons: the potential to generate high average returns and
the opportunity to diversify portfolios. 4 The theoretically desirable reduction of non-systematic risk comes
not only from the expansion of the efcient frontier, i.e., having access to a wider spectrum of investment
opportunities, but also from the segmentation of EMs from the developed capital markets. Harvey (1995)
and Bekaert and Harvey (1997) show that, even after emerging markets opened for foreign investors and
capital, they remained dominated by domestic market participants who select their portfolios based on
local economic and market movements rather than global trends. Bekaert and Harvey (2002) conclude that,
despite increasing integration between EMs and developed markets in the post-liberalisation period, the EMs
1
Banz (1981) nds that investors require a higher return for holding small rms (measured by capitalisation) rather than large
ones. Stattman (1980) and Rosenberg et al. (1985) show that rms with high levels of book-to-market ratio are mostly in poor
nancial condition, which decreases stock price; Basu (1977, 1983) shows that stocks with low priceearnings ratios earn more than
predicted by the market model and vice versa; DeBondt and Thaler (1985) nd that long-term past returns (three- to ve-year
horizon) tend to reverse; Jegadeesh and Titman (1993, 2001) investigate portfolios built on short-term past returns of common
stocks and show that stocks with higher average past returns (up to one year) tend to generate higher returns over a three- to
twelve-month holding period. Amihud and Mendelson (1986), Amihud (2002) and others nd a positive relationship between a
stock's illiquidity and the average returns. Ang et al. (2006) and Ang et al. (2009) argue that stocks with low idiosyncratic volatility
generate higher returns. For further examples, see, e.g., Fama and French (2008).
2
Fama and French (1992) show that the positive relationship between the market beta and the average return disappears in the
post-1970s period. The CAPM is also unable to explain the returns of portfolios that are related to the predictable cross-sectional
variation in stock returns. The unexplained patterns are therefore called the CAPM anomalies.
3
See, e.g., the literature discussing the role of idiosyncratic volatility in asset pricing, where individual risk of a given stock is
calculated relative to a three-factor model, e.g., Ang et al. (2006). There is also much research investigating the momentum effect on
equity markets worldwide. The literature lacks, however, corresponding evidence about performance of the four-factor model of
Carhart (1997) when explaining expected returns. For most recent results, see Fama and French (2012).
4
The high average returns are achieved because of the risk related to political instability, forms of government, level of corporate
governance and/or geographical location. These sources of uncertainty are weakly correlated with global economic performance.
EMs also have superior growth prospects.
188 A. Waszczuk / Emerging Markets Review 15 (2013) 186210
Table 1
Literature overview evidence regarding the existence of patterns in the average returns from European emerging markets.
Paper Countries Period Frequency Europe (CEE) MV BV/MV P/E MOM LIQ DY
Panel A: International
Claessens and Dasgupta 19 19861993 Monthly Yes (no) Yes Yes Yes No Yes
(1995)
Fama and French (1998) 16 19871995 Monthly Yes (no) Yes
Rouwenhorst (1999) 20 19821997 Monthly Yes (no) Yes Yes Yes No
Barry et al. (2002) 35 19852000 Monthly Yes (yes) Yes/no Yes
van der Hart et al. (2003) 32 19882004 Monthly Yes (yes) Yes Yes Yes Yes
Serra (2003) 21 19901996 Weekly Yes (yes) Yes Yes Yes Yes Yes Yes
Panel B: Regional
Time-series
Kowerski (2010a) 1 19942007 Monthly PL Yes Yes
Lischewski and Voronkova 1 19962009 Monthly PL Yes Yes No
(2012)
Cross-section
arnowski (2007) 1 19972007 Monthly/weekly PL Yes No Yes
Kowerski (2010b) 1 19952005 Monthly PL Yes
Time-seriescross-section
Borys and Zemcik (2011) 4 19932007 Monthly PL Yes Yes
Czapkiewicz and Skalna 1 20022009 Monthly PL Yes Yes
(2010)
Czapkiewicz and Skalna 1 20022010 Monthly PL Yes Yes
(2011)
This table summarises the literature devoted to patterns and risk premiums in stock returns from EM. Listed papers consider data from
EM but do not necessarily focus on these markets exclusively. Panel A shows the studies based on international data from several
countries, while Panel B focuses on studies on the Polish market. Column Paper refers to the particular study, column Countries gives
the number of countries used in a study and column Period shows the sample period used in empirical tests of given study. Column
Frequency describes the type of data used. The fth column in Panel A tells whether European Markets were used in a study using yes
to indicate that it is the case and no otherwise. The information about inclusion of countries from Central and Eastern Europe is given in
brackets and uses the same self explanatory convention. In Panel B we use PL to mark studies working with Polish data. MV, BV/MV, P/E,
MOM, LIQ and DY represent the effects related to market capitalisation, book-to-market ratio, priceearnings ratio, momentum
(short-term past return performance), liquidity and dividend-yield, respectively. The exact denition and construction method for a
given variable depends upon the paper. Yes indicates that a relationship between the variable and average returns has been found in a
sample; otherwise, no is used; marks variables that have not been discussed in a given paper.
are still segmented enough to contribute to portfolio diversication.5 However, as presented in Table 1,
prominent studies very seldom consider a single EM. They generally include several countries from different
geographical regions, each with a different political and economic background, and deliver common evidence
for the entire sample.6 Although such an extension of the dataset increases the formal statistical power of
tests, it does not allow the formulation of strong conclusions concerning single markets. Further deciency of
studies of particular importance for this paper is the fact that only Barry et al. (2002), van der Hart et al.
(2003) and Serra (2003) include Central and Eastern European (CEE) countries in their samples. Additionally,
these sample periods last maximally until the year 2004 and therefore cover only the rst years of a markets'
existence when transmission process could strongly bias the results.
This paper aims to ll in this gap in the international evidence. According to the monthly reports
published by the Federation of European Securities Exchanges (FESE), at the end of June 2011, the Warsaw
Stock Exchange (WSE) was the leader of not only the CEE but also of Southern and Eastern Europe, with a
5
Voronkova (2004) delivers evidence in favour of signicant and stronger long-run relationships between the emerging CEE and
global markets than has previously been reported. Fedorova and Vaihekoski (2009), in contrast, conrm the signicance of regional,
rather than global, market risk factors for CEE.
6
This is due to a limited number of stocks and data covering a relatively short time-series when compared to the U.S. market, for
which the time-series going back to 1929 can be provided, see, e.g., Davis (1994) and Davis et al. (2000).
market capitalisation of over 145 billion EUR and over 420 companies included in the Warsaw Stock
Exchange Index (WIG). WSE also outperformed the former region leader, the Austrian Wiener Brse. The
selection of the Polish Market is therefore not an accidental one. We believe it is representative of the CEE,
particularly with its recently introduced strategy to take the leading position in the region and attract capital
searching companies from the entire CEE. 7 Our choice is also motivated by the lack of internationally available
studies devoted to CAPM anomalies and the performance of the three-factor model on the Polish market
while, according to the reports of the WSE, at the end of 2011 47% of overall turnover was caused by foreign
investors.8 Further, the amount of foreign initial price offerings (IPOs) on the WSE is rising systematically, as
presented in Table 2. These ndings clearly show that the group of investors interested in understanding asset
pricing mechanisms of the Polish market increases constantly.9
This paper contributes to the current literature in several ways. First, we give the most comprehensive
picture of the Polish stock market to date in context of asset pricing. We work with an up-to-date dataset that
covers the period from July 2002 through June 2011.10 For this sample period, we investigate if the
relationship between the stock characteristics and their returns documented for developed markets holds
true for the WSE. We sort stocks into portfolios according to various attributes, most of which have never
been referred to for WSE. Complementarily, we apply cross-sectional regression tests for individual stocks.
Fama and French (1996) argue that most of the regularities observed in average returns are caused by the
homogeneous character of risk. The premium factors in the three-factor model are therefore expected to
explain patterns if they indeed serve as an approximation of the fundamental sources of risk. Basing on
international and Polish evidence, presented in Table 1, we include the well established size and value
premiums in a time-series regression of the average returns of portfolios that are sorted by stocks' attributes.
Our goal is to test if the three-factor model outperforms the CAPM when explaining portfolios other than
those constructed two-dimensionally on size and book-to-market ratio. This procedure allows us to respond
to the critique of asset pricing tests delivered by Lewellen et al. (2010) because we extend the variety of the
portfolios for which we compare the explanatory power of the CAPM and the three-factor model. A further
element of innovation in this study is the articial cut-off we apply when calculating the risk premiums
related to the rm's size and book-to-market ratio. The WSE is characterised by the large number of micro
stocks. The cut-off helps to concentrate the small stocks into one group, allowing the other to accumulate
larger ones. Additionally, we carry out the Monotonic Relation (MR) test of Patton and Timmermann (2010).
We test the hypothesis that the given pattern is signicantly positive (negative) considering the performance
of all portfolios and not exclusively the signicance of hedge premiums constructed as difference in returns of
marginal portfolios.
Our empirical ndings document the presence of a majority of the investigated patterns in at least one
type of returns. Some of them, however, are not robust to the return weighting scheme and diminish in
simple or cap-adjusted returns. Reversal, liquidity, and idiosyncratic risk neither produce monotonic
patterns in portfolio returns nor generate signicant arbitrage premiums. The cross-sectional test of Fama
and MacBeth (1973) favours the book-to-market ratio as the determinant of the cross-sectional variation
of single stock returns. None of the other attributes explain the returns signicantly, neither when used
alone nor in multiple regressions when used together with the book-to-market ratio. We conrm the
superiority of the three-factor model over the CAPM in explaining the time-series variation of average
portfolio returns. Our comparative analysis of the explanatory power of the models takes into account the
most recent discussion and argumentation of the efciency of time-series tests. We present additional
statistics which supplement our conclusions with statistical properties of the regressions. To the best of
7
Visegrad countries (Poland, Czech Republic, Slovakia and Hungary) took a relatively similar transformation path and are
historically and economically close due to geographical location. However, the cumulative market capitalisation of the three latter
countries constitutes only about 48% of the WSE equity value.
8
Specically, the literature lacks internationally published English language papers on this topic. For overview of relevant
literature, see Panel B of Table 1.
9
The reports on the investor structure on WSE are available under http://www.gpw.pl/analizy.
10
The use of a relatively current dataset is of particular importance for young equity markets, such as those in the CEE, because of
their short history and subsequent rapid growth. Only a few years ago did the number of listed stocks reach the minimal level
needed for powerful statistical tests and reasonable conclusions. Furthermore, only since the year 2000 Polish equity market has
attracted important institutional investors, e.g., pension funds, which had a big inuence on development and characteristics of WSE
as stated in Schotman and Zalewska (2006).
Table 2
Overview of the WSE development in years 20022011.
Year Stocks listed Foreign stocks Stocks used IPOs Capitalisation
2002 217 0 177 6 25 260

2003 203 1 159 5 29 660
2004 230 5 164 36 52 500
2005 255 7 195 35 79 900
2006 284 12 210 38 114 280
2007 351 23 250 81 141 240
2008 374 25 300 33 64 110
2009 379 25 315 13 102 470
2010 400 27 321 34 137 000
2011 426 39 xxx 38 100 939
This table shows the basic characteristics of the WSE for the period 20022011. It gives the number of stocks listed at the end of each
year in the second column, the number of listed foreign stocks in the third column, and the number of debut stocks in the fth
column. The last column gives the capitalisation of all domestic stocks in millions of EUR (exchange rate at the end of the calendar
year). The fourth column presents the number of stocks used when building the risk mimicking factors with the exclusion of extreme
returns (yearly, at the end of June). Statistics come from the WSE website: http://www.gpw.pl/analizy_i_statystyki.
our knowledge, we also deliver the rst tests for the momentum pattern and four-factor model of Carhart
(1997) on the WSE. The momentum model helps explain the momentum pattern but has no superiority
over the three-factor model for the remaining grouping attributes.
The remainder of this paper is organised as follows: Section 2 presents the dataset and discusses the
methodology. Section 3 documents the patterns in the average returns of portfolios in Subsection 3.1 and
cross-sectional regression tests in Subsection 3.2. Subsection 3.3 compares the performance of asset
pricing models in explaining CAPM anomalies and Subsection 3.4 delivers the comparative analysis of local
and global risk factors. The nal part gives a summary and the conclusions.
2. Data and methodology
The data sample used in this study covers the period from July 2002 to June 2011. The monthly values for
the accounting data (capitalisation, book-to-equity and priceearnings ratio) were collected from monthly
reports published by the WSE. The daily returns and volume were generated by the WSE information portal
GPW-InfoStrefa.11 The creation of this unique dataset allows us to largely avoid survivorship bias by including
companies which were suspended, merged or delisted during the relevant period. We exclude nancial and
non-domestic rms because they are subject to different accounting standards. The returns are adjusted for
dividends. We argue that our sample includes an acceptable number of stocks and time-series observations
for further empirical study. For an overview of the number of stocks used in this study, see Table 2.
The rst aim of this study is to investigate whether the Polish stock market exhibits monotonic
relationships between some stocks' characteristics and their average returns. We apply two standard
approaches designed to identify patterns in average stock returns. First, we sort the stocks according to the
investigated characteristics, and examine the return differences between the marginal portfolios. Second, we
apply the Fama and MacBeth (1973) test, which uses the characteristics as the explanatory variables in
cross-sectional regressions of single stocks. Both methodological concepts have their advantages and
shortcomings, recently discussed in Fama and French (2008). Due to the methodological diversity they are
supplementary approaches providing a good cross-check.
In the sort approach, we group stocks on size, book-to-market and priceearnings ratio, short- and
long-term past returns, liquidity and idiosyncratic volatility. While the rst two variables have been
already examined due to their central role in the multifactor asset pricing, sorting on the remaining ones
presents the rst evidence for the WSE. 12 We provide further evidence for the priceearnings ratio that
goes beyond the sample period examined in arnowski (2007). Ball (1978) describes the reciprocal of the
priceearnings ratio as a catch-all variable which is supposed to predict future returns if current earnings
11
http://www.gpw.pl and http://www.gpwinfostrefa.pl.
12
An exception is liquidity, see Lischewski and Voronkova (2012).
proxy for future earnings. Low price relative to earnings characterises distressed stocks with high discount
rates. Furthermore, we incorporate momentum factor as a new center stage anomaly which gains this
prominent position from the diminishing and non-persistent size effect. Return reversal is a natural
supplement of momentum evidence. The mean-reversion of returns over the long-term is a consequence of a
breakdown in short-term continuation, as winners become too expensive and losers attract investors with
low price and high expected return. We also deliver the rst evidence for one of the most discussed patterns
recently, the puzzling idiosyncratic risk pattern. While, e.g., Merton (1987) and Malkiel and Xu (2002) argue
that stocks with high idiosyncratic volatility should earn more due to their riskiness, Ang et al. (2006) and Ang
et al. (2009) show the exact opposite relationship. Idiosyncratic risk studies are, however, methodologically
very inconsistent. Because it is not uncommon to calculate stocks' individual volatility relative to pricing
model the results might be sensitive to the choice of the return generating process and therefore, differ from
paper to paper.13 Such a mixed picture encourages us to deliver further empirical evidence from emerging
markets where relatively high variation in stock returns belongs to one of the market specics. Furthermore,
we investigate the liquidity and its relationship to returns covariation. Bekaert et al. (2007) argue that
liquidity is particularly important when evaluating emerging markets because such markets are characterised
by fewer investors and fewer securities. The concept of liquidity is not straightforward and scholars use
various proxies to capture its relevant dimension. In this paper, we focus on the trading quantity proxied by
turnover ratio, which allows us to distinguish between stocks that are traded sporadically and those that
change owners regularly and so have higher inuence on the performance of the local market. These latter
stocks are, at least theoretically, attractive to investors that do not plan for long-term investments and
therefore do not generate excess returns, the liquidity premium. 14
We investigate patterns described above using a traditional quintile-based portfolio formation. For each
characteristic, we allocate stocks into ve portfolios according to the value of the sorting attribute. 15 To judge
the signicance of differences in the performance of marginal portfolios, we construct the arbitrage premium
for each pattern. To do so, we mimic the following trading strategy: the investor buys a portfolio which is
riskier according to theory or past evidence and goes short on portfolio with lower expected returns. This
zero-investment portfolio is held for one year (month) and rebalanced at the end of each June (at the end of
each month). The returns are recorded on a monthly basis. The holding periods are non-overlapping. We do
not adjust the portfolios for delisted stocks during the holding period, so if a stock leaves the portfolio due to
delisting, merger or other reason (e.g., missing value) the averages are calculated for the remaining sample
without regrouping across portfolios. Then, we quantify the protability of the hedge portfolio return using
the difference-in-means test. We distinguish between equal- and value-weighted portfolios. For details of
portfolio construction and attribute measures, see Tables 5 and 6.
We also address the ndings of Barry et al. (2002) who show that the size effect in EMs disappears when
controlling for extreme return observations. To eliminate the potential penny stock effect we exclude stocks
with a price lower than 0.50 PLN. We also remove observations with an absolute return over 50%. The majority of
returns ltered out by this procedure is strongly negative and is accumulated around the years 20072009.
Results presented in the following sections refer to a sample without extreme observations, unless stated
otherwise. In general, we conrm the inuence of extreme returns on the size effect. However, stronger
differences are visible also for the technical anomalies. Second lter set to control for known microstructural
concerns eliminates the rst month of each year to check robustness to January effect. We conrm the inuence
of extraordinary good performance of small stocks in Januaries on marginal premiums. Size strategy does not
generate signicant payoffs when this month is omitted. The inuence on other variables is mixed.
13
Fu (2009) shows that negative relation between rm-specic risk and average return is related to reversal in returns of micro
stocks which is captured in Ang et al. (2006) and Ang et al. (2009) idiosyncratic risk proxy.
14
For a full overview of the liquidity measures and their relevance for the Polish market, see Lischewski and Voronkova (2012).
Note that their turnover proxy differs from the one applied in this study because they use monthly values that are constructed from
daily volume divided by shares outstanding. Furthermore, the authors work with a different sample period.
15
We construct only ve portfolios, in contrast to the commonly used decile procedure, due to the trade-off between the number
of stocks per portfolio and the grid resolution. High grid resolution guarantees bigger differentiation between the stock
characteristics across portfolios. However, a small number of stocks per group could cause biases in the results. Thus, larger groups
guarantee the diversication and elimination of idiosyncratic biases. In the original study of Fama and French (1993) the number of
stocks per portfolio exceeded 200. For the same reason, we omit the suggestion of Lakonishok et al. (1994), who argue that sorting
on two attributes at the same time allows a better differentiation of the rms' specics.
Furthermore, we expand our evidence for the existence of patterns in returns with the MR test of
Patton and Timmermann (2010). This non-parametric test delivers a good supplement to the sort
approach and the zero-investment strategy because it investigates the complete cross-sectional pattern in
the expected stock returns and so allows conclusions about, e.g., the risk-proxy properties of sorting
variables. The MR test is based on a bootstrap approach in which new samples are randomly drawn with
replacement from the original sample of the average monthly portfolio returns. We draw new samples
b = 10.000 times in our study. After simulation the mean portfolio returns are calculated for all drawn b
samples over the full study period. Further, they are demeaned by subtracting the original time-series
portfolio averages. We test the null hypothesis that there is no difference in portfolio returns (at pattern)
against the alternative hypothesis suggesting that there is increasing monotonicity related to the sorting
attribute:
H0 : 0 vs: H 1 : min i > 0: 1

i2;;5
We construct the return differences between the adjacent portfolios which we mark as i r t;Pi r t;Pi1
for i = 2 to 5. Following the notation of Patton and Timmermann (2010) the test statistic is given as:
J T min i 2
i2;;5
for the original sample. To obtain the p-value for the test we count the number of cases for which JT b JTb
(where JTb is calculated analogously to JT for all b draws) and divided it by the number of bootstraps, b.16
Next, we look for the risk-based explanation for the return patterns. We check if the standard version
of the CAPM is able to explain the differences in the average returns between portfolios. Further, we
investigate the contribution of additional risk factors to explanation of Polish expected stock returns as
proposed by Fama and French (1993). The authors argue that not only the sensitivity of portfolios return
to market risk premium, MRP, (as for CAPM) explains the expected excess return of these portfolios, but
that also the portfolios sensitivity to additional factors is relevant. These factors are mimicked by the
premiums related to the size, SMB, and the book-to-market ratio, HML, of the listed stocks. We also
introduce momentum factor, WML, as proposed by Carhart (1997).
To obtain the relevant slopes on risk premiums, as well as to quantify the explanatory power of the
asset pricing models discussed above, we run the following simple OLS regression for the CAPM:
r t;p r t;f p p MRPt t;p 3
and three (four) factor model

r t;p r t;f p bp MRPt sp SMBt hp HMLt wp WMLt t;p : 4
v
As a proxy for a market return, rm , we take the return of the WIG-PL Index. 17 Risk free rate, rf, is proxied
v
with one-month intrabank interest rate WIBOR. Market risk premium MRP is then rm rf, calculated at the
end of each month. SMB, HML and WML are portfolios mimicking size, book-to-market and momentum
risk factors respectively.
The failure of the CAPM in explaining the average stock (portfolio) returns is often justied by its
untestability, as shown by Roll (1977). This is because the market risk premium is unobservable and the used
16
To test the decreasing pattern one should change the order of portfolios to get increasing relationship. For more details on the
methodology of the MR test see Patton and Timmermann (2010). We thank the authors for making the MatLab code for the test
available under http://public.econ.duke.edu/~ap172/code.html.
17
Being aware of the importance of risk factor proxies, we use the WIG-PL index containing all domestic stocks as a market return. The
correlation between this index and the value-weighted sample market return is 0.98 (WIG20 is 0.97). We argue that the use of an
externally calculated market return is more plausible and allows us to avoid the suspicion that the model works well because of a
manipulated market portfolio proxy. The WIG-PL Index is analogous to the WIG (the overall index), with the exclusion of foreign stocks. It
includes all stocks that full the minimal restrictions concerning the amount of stocks in free turnover and nancial conditions.
proxy might be inappropriate. At the same time, the construction methods for other risk premiums (i.e.,
mainly SMB and HML) used in multifactor models have not been broadly discussed. The original methodology
proposed in Fama and French (1993) ts the American market environment, but even the authors point out
that it is arbitrary. Nevertheless, their procedure was implemented in almost all of the empirical tests
following their initial paper. The potential pitfalls resulting from imitation of the original methodology have
been addressed only in most recent papers working with international data, see Fama and French (2012).
Already in Fama and French (2008) the authors note that extremes in sorts that use all stocks are dominated
by small stocks because they are more plentiful than big stocks and the fundamentals of small rms are
typically more disperse. In our case, we proceed similarly as Fama and French (1993), but with some
adjustment to Polish market specics. Generally, we also divide the available stocks into two size and three
book-to-market groups. In Poland, however, there is no exchange comparable to NASDAQ.18 The WIG index is
still signicantly dominated by smaller stocks (though this tendency has been reduced in the last few years)
while its performance covaries strongly with the WIG20 index.19
For that reason, we introduce an innovation in the SMB and HML construction method. First, we extract
the half of the sample with the largest stocks (by market capitalisation). After, we build an articial cut-off
equal to the median of this half of the sample. The cut-off is used to group stocks into two size and three
book-to-market portfolios. In this way, we can differentiate clearly between small and large stocks, which
would not be the case if we used the median of the entire sample. Because the variation in methodology may
result in deviating results, in Table 3 we compare the premiums for both the standard and the new
construction methods.
In Panel A of Table 3, we show the size and value risk factors calculated with and without the cut-off and
weighted by the stock capitalisation at the end of the current month. As expected, the differentiation between
the construction methods inuences the magnitude of the premiums. Using all of the stocks results in a high
SMB and a much lower HML premium. When the cut-off is introduced, the opposite result appears the
value premium dominates the size premium. In both cases, the value premium is more volatile. Looking at the
average returns of the single portfolios that are used to calculate premiums, we see that the quantitative
difference comes mainly from portfolios with small stocks. Fama and French (2007) show that the higher
average returns of small stocks arise from the extraordinary performance of small stocks which generate
higher prices, become bigger and migrate to the big caps portfolio. To further investigate this postulate for our
sample, we calculate the premiums with xed weightings from the formation period. The results shown in
Panel B of Table 3 reveal the insignicant and slightly negative SMB, while HML retains its magnitude. By
weighting using the capitalisation from the formation period, we neglect the primary source of the size
premium, i.e. increase in the capitalisation of small stocks. This nding conrms the nature of size premium
shown by Fama and French (2007).20 Based on these ndings, we use premiums that are calculated with the
cut-off and weighted with the changing capitalisation in our time-series tests. This choice is also motivated by
the fact that our methodology supports the main idea behind six two-dimensional sizebook-to-market
portfolios, i.e., separation between size and value effects. In fact, premiums calculated with cut-off show
smaller standard deviations. WML construction follows Carhart (1997). We split the stocks into winners and
losers according to the 3rd or the 7th decile of the short-term past performance, estimated as the
compounded-return over the last twelve months lag one. The regrouping of the portfolios for the winners
minus losers premium occurs monthly. All risk premiums are value-weighted.21
Panel A of Table 4 presents the descriptive statistics for all risk premiums discussed in this paper. We
observe, beyond the results presented in Table 3, that all of the premiums but the market premiums are
18
The only alternative equity market in Poland, New Connect, was opened in autumn 2007.
19
The distribution of rm size on WSE is signicantly skewed. The average market capitalisation over the sample is 768 million
with a standard deviation of 2802 and a median of 122.
20
We do not further investigate the migration of stocks between 2 3 portfolios because it is beyond the scope of this paper. We
solely stress the importance of value-weighting the returns of those portfolios by elucidating the source of the size premium.
21
Fama and French (1993) explain the use of value-weighted portfolios with a negative relationship between the return variance
and the size. This procedure is consistent with investor's interest in minimising volatility. Furthermore, using value-weighted
components results in mimicking portfolios that capture the different return behaviors of small and big stocks or high- and low-BE
stocks, in a way that corresponds to realistic investment opportunities.
Table 3
Average returns of six sizebook-to-market portfolios and the SMB and HML risk premiums from the different construction methods.
No cut-off Cut-off
Portfolio Mean Std Mean Std
Panel A: Flexible weighting

Small-growth 2.93 11.73 1.45 9.04
Small-neutral 1.79 9.15 1.83 8.02
Small-value 2.46 10.51 2.14 7.95
Big-growth 0.57 7.64 0.54 8.32
Big-neutral 0.69 7.80 2.00 8.41
Big-value 1.52 7.03 2.04 7.51
SMALL 2.40 9.14 1.59 7.49
BIG 0.55 6.51 0.57 6.78
HIGH 1.99 7.90 2.09 7.05
LOW 1.18 8.54 0.45 7.72
SMB 1.85 5.79 1.02 4.10
HML 0.81 6.05 1.63 5.15
Panel B: Fixed weighting

Small-growth 0.65 10.04 0.91 8.33
Small-neutral 0.05 8.35 0.48 7.38
Small-value 0.06 9.78 0.23 7.76
Big-growth 0.86 7.64 0.92 7.81
Big-neutral 0.01 7.74 0.16 8.04
Big-value 0.71 7.39 1.13 7.14
SMALL 0.25 8.67 0.39 7.19
BIG 0.04 6.78 0.02 6.84
HIGH 0.33 7.93 0.68 6.98
LOW 0.75 8.10 0.91 7.39
SMB 0.21 5.04 0.40 3.79
HML 1.08 4.99 1.59 4.54
Panel A presents the value-weighted average returns (in percent) of six size and book-to-market portfolios using 108 monthly
observations between July 2002 and June 2011. In the case of construction without a cut-off, we split the stocks into two size groups based
on the median size of the full sample and, independently, into three book-to-market portfolios based on the 3th and 7th decile of the
book-to-market ratio values. We follow the same procedure for construction with a cut-off, with the difference that the quintiles come
from the half of the sample that includes the largest stocks, but they are applied on the full sample. Panel A shows the average monthly
returns weighted with the current (exible) capitalisation while Panel B weights stocks with the capitalisation from the formation period,
i.e., in June of year t. SMALL and BIG are the simple averages of returns from three book-to-market portfolios for small and large stocks,
respectively. HIGH and LOW are simple averages of two size portfolios for value and growth stocks, respectively.
signicantly different from zero over the sample period. The equal-weighted market return is only slightly
negative, while the value-weighted return is only slightly positive. This is because the sample period
incorporates intervals of poor economic performance, such as the Lehman Brothers crisis of 2008. The
potentially simultaneous signicance of multiple risk factors in the regression tests could be due to the
multicollinearity. Panel B of Table 4, presenting the correlation coefcients between the different risk
premiums, does not show any signicant co-movements between the premiums used in the same
regression tests. The equal-weighted market return is relatively highly correlated with other factors, with
the highest correlation (except the obvious correlation with value-weighted counterpart) with size
premium constructed without cut-off (0.48). There is evidence suggesting a countercyclical character of
the value premium and a procyclical character of the momentum premium. However, for WSE momentum
is negatively correlated with market performance during the sample period.
When discussing the results of the time-series tests, we again address the criticism of Lewellen et al.
(2010). Generally, the main quality indicator of a given model is the statistical insignicance of the
intercepts accompanied by a high root square error (RSQ). Because both the single and the multifactor
models indicate that the expected returns are fully explained by the linear sensitivity of the stocks
(portfolios) to the risk factors, all of the regression intercepts (i.e., pricing errors) should be equal to zero.
Table 4
Descriptive statistics and correlation coefcients for risk premiums.
Panel A: Statistics Mean Std Min Max Skewness Kurtosis t-value
SMBno 1.85 5.79 10.37 21.29 0.99 1.48 3.32

SMByes 1.02 4.10 12.94 13.25 0.10 1.09 2.60
HMLno 0.81 6.05 15.39 22.95 0.34 2.21 1.39
HMLyes 1.64 5.15 9.80 19.32 0.77 1.26 3.30
MRPe 0.42 7.61 24.84 18.13 0.12 0.74 0.58
MRP 0.85 6.74 24.80 18.63 0.56 1.46 1.31
WML 1.28 6.22 21.84 14.89 0.94 2.68 2.14
Panel B: Correlation SMBno SMByes HMLno HMLyes MRPe MRP WML
SMBno 1.00
SMByes 0.67 1.00
HMLno 0.24 0.01 1.00
HMLyes 0.00 0.05 0.58 1.00
MRPe 0.48 0.34 0.01 0.02 1.00
MRP 0.17 0.01 0.04 0.05 0.86 1.00
WML 0.05 0.11 0.15 0.20 0.23 0.18 1.00
Panel A presents the mean average return (Mean), the standard deviation (Std), and the minimal (Min) and the maximal (Max)
values of the premiums over the sample period from July 2002 until June 2011. Skewness, kurtosis and t-values for the zero-mean
test for risk mimicking premium are shown in the three last columns of Panel A. Panel B presents the correlation coefcient between
all of risk premiums. no behind SMB and HML refers to premiums calculated without cut-off while yes with cut-off. MRPe stands for
equal-weighted market return including all stocks being used in calculation of SMB and HML. MRP represents the value-weighted
WIG-PL Index. WML stands for momentum risk factor. Construction of all premiums is described in Section 2 and Table 3.
The standard procedure tests the hypothesis whether the set of alpha parameters is jointly zero. The
so-called GRS test proposed by Gibbson et al. (1989) uses the following statistic for one-factor model: 22
W ^ ^ 1 ^
5
1 E^ ff
T
where ^ is the vector of portfolio intercepts generated in time-series tests, ^ is the covariance
matrix of regression residuals and f stands for model factor. The numerator is equal to
2
2
q = q ET f =^ f SR 2 . Authors argue that (q/q) 2 is square of the maximum Sharpe
ratio of the ex post tangency portfolio formed from all of the test assets plus the model factor f, and the
second part of the above equation is the square maximum Sharpe ratio of the factor alone. The
difference can be interpreted as how far the return f is inside the ex post frontier, see Cochrane (2005).
Following the recommendation of Lewellen et al. (2010) and the empirical examples from Fama and
French (2012), we show the average absolute value of the intercepts from each set of regressions and
the average of the standard errors of the intercepts as well as SR(). The measures of precision are the
RSQ and the average standard error of intercepts. As mentioned by Fama and French (2012), it is useful
to add information such as the magnitude and the estimation precision for the intercepts beyond the
SR() value to get a complete picture of the regression quality.
22
The GRS test can be extended for multiple factors, see, e.g., Cochrane (2005) for details.
Our methodology refers also to other critical aspects mentioned by Lewellen et al. (2010). They argue
that judging the power of alternative models by their ability to explain the sizebook-to-market
portfolios is misleading because of the analogous building character of both the left and the right hand
side of Eq. (4). They propose to expand the tests by examining portfolios sorted by other variables. 23
Grouping the stocks not only on fundamentals but also on, e.g., past returns or volatility, is consistent
with this suggestion. 24
3. Empirical ndings
3.1. Patterns in average stock returns
We refer to patterns in the average stock returns when we are able to document a monotonic relationship
between the portfolio average returns and the magnitude of a sorting variable. A risk (multi-)factor
framework predicts a linear relation between the expected returns and the portfolios loadings on the models'
risk factors. This relationship is not predicted for the expected returns and the sorting criteria.25 However, it is
helpful to know whether the returns follow the pattern systematically or if the differences are given only for
the portfolios with extreme values of sorting attributes.26
Table 5 shows the basic statistics of quintile portfolios sorted by size, book-to-market and priceearnings
ratios. In the case of portfolios formed on market capitalisation, small stocks earn more than large stocks for
both types of returns, but the hedge premium between marginal groups is signicant only in the case of the
cap-weighted returns. Value-adjustment delivers higher returns, which indicates the inuence of big,
well-performing stocks within size portfolios. Value-weighting produces also higher standard deviation of
returns for the small stocks, but similar for the larger stocks. The hedge return premium is mostly driven by
abnormally high returns that are generated in January, so its signicance diminishes when Januaries are
excluded from the sample. When comparing the subperiods, we observe that for both the equal- and the
value-weighted returns there is no evidence of abnormal performance for the small stocks relative to the
large stocks for the second subperiod. 27 The full period results are inuenced by the rst subperiod, when
there is a signicant premium for both types of portfolio returns. This nding is consistent with international
evidence suggesting the demise or even the reversal of the size effect in more recent periods, see van Dijk
(2011). When releasing the control for extreme returns, we obtain qualitatively identical results. The hedge
premiums are, however, slightly higher. For value-weighted portfolios, the marginal portfolio's premium is
signicant, even after exclusion of the January months. For the full sample it remains signicant for both types
of return. These results are partially consistent with the nding of Barry et al. (2002) who show that the
magnitude of the size effect in emerging markets is inuenced by extreme returns.
Similarly, as in many earlier international studies, we show that the value effect is stronger than the
size effect in an economic sense and more persistent in the robustness tests for sample limitations. The
expected decreasing pattern in average post-formation returns is visible in both equal- and
value-weighted returns, and both premiums are signicant. The value pattern is also less sensitive to
the January effect. Although the arbitrage premiums are slightly smaller for the full sample, they are
signicant in both cases. Further, the inclusion of extreme returns does not inuence the returns. The
23
As shown in Table 1 mainly tests for the size and book-to-market ratio on the LHS were performed for the WSE, with the
exception of Lischewski and Voronkova (2012), who additionally consider liquidity portfolios.
24
We do not run time-series tests for the sizebook-to-market portfolios for two reasons. First, it can be expected that because the
portfolios on the LHS of Eq. (4) are constructed analogically as SMB and HML factors, those factors are able to capture the variation in
the portfolio returns. Second, such tests have already been performed for the Polish market. For examples, see Panel B of Table 1.
25
The alternative, characteristic-based explanation of the empirical success of multifactor models states that it is not the risk
loadings, but the characteristics themselves, which are priced in the cross-section of average returns, see Daniel and Titman (1997).
We do not further investigate this proposition in this paper.
26
On the other hand, the lack of signicant payoffs of the hedge portfolios does not mean that a characteristic is not important for
return covariation. Instead, the (potential) source of correlation may not be of a special hedge concern for investors.
27
Results for subperiods and a sample including extreme returns are not shown due to space limitations but available upon
request.
Table 5
Arithmetic and value-weighted averages of monthly returns on ve portfolios sorted by fundamentals.
Attribute Portfolio Premium MR test

Lowhigh p-value
1 2 3 4 5
Low High
MV
EW average return 0.57 0.15 0.05 0.44 0.25
Std. deviation 8.88 8.59 8.26 7.12 7.28 0.85 0.77
VW average return 3.11 1.88 2.24 0.99 0.77
Std. deviation 13.00 9.50 8.55 7.29 6.85 2.35** 0.86
Average size 41 80 186 489 5 042
CV(size) 1.32 2.17 2.04 1.97 0.67
MV non January
Std. deviation 8.78 8.54 8.12 7.03 7.15 0.25
Std. deviation 1.11 9.61 8.46 7.10 6.72 1.35
Average size 41 81 187 491 5 079
BV/MV
Std. deviation 7.62 7.30 7.96 8.01 8.99 1.95*** 0.01**
Std. deviation 8.50 9.48 7.56 7.44 7.74 2.31*** 0.24
Average size 1 131 1 704 1 316 884 720
CV(size) 0.62 0.42 0.37 0.37 0.33
BV/MV non January

Std. deviation 7.66 7.28 7.73 7.86 8.86 1.63***
Std. deviation 8.62 9.60 7.24 7.33 7.50 1.98***
Average size 1 155 1 777 1 337 891 715
P/E
Std. deviation 9.38 7.69 7.87 7.56 7.40 0.93* 0.99
Std. deviation 10.03 7.28 8.01 9.76 7.23 1.62** 0.95
Average size 1 671 1 454 1 876 1 202 703
CV(size) 0.42 0.40 0.47 0.48 0.61
P/E non January

Std. deviation 9.30 7.53 7.91 7.50 7.44 0.87
Std. deviation 10.01 7.14 7.91 10.06 7.05 1.84**
Average size 1 664 1 495 1 935 1 241 725
Portfolios considered in Table 5 are constructed annually using the quantile values of the characteristics at the end of June of year t
(market capitalisation, MV, and priceearnings ratio, P/E, or at the scal year end (to calculate book-to-market ratio, BV/MV, we use
the March book value of equity and market value in December t-1)). Monthly returns are calculated for the next twelve months. The
time series include 108 monthly returns (99 when Januaries are excluded). The statistical signicance of the difference-in-mean test
is marked with ***, ** and * at the 1%, 5% and 10% level, respectively. The p-value for the MR test is shown in the last column. EW and
VW stand for equal- and value-weighted average portfolio returns, respectively. Average size gives the mean capitalisation of the
portfolio over the period considered (in millions PLN). CV(size) is the average coefcient of variation, i.e., average ratio of standard
deviation of size to its mean.
premiums are only slightly higher when the outliers are included. The subsample comparison shows that
the value effect is more signicant in the second period, while marginally non-signicant (p-value slightly
above 0.10) in the rst.
Table 6
Arithmetic and value-weighted averages of monthly returns on ve portfolios sorted by past returns, idiosyncratic volatility and
liquidity.

LowHigh p-value
1 2 3 4 5
Low High
MOM
Std. deviation 7.88 7.08 6.90 6.64 6.94 1.45*** 0.01***
Std. deviation 8.08 7.14 7.03 6.22 7.35 1.92*** 0.03**
Average size 625 720 734 825 890
CV(size) 0.51 0.46 0.41 0.43 0.61
MOM non January

Std. deviation 7.88 6.96 6.85 6.66 6.86 1.48***
Std. deviation 8.25 7.24 7.06 6.21 7.14 1.83***
Average size 638 709 726 825 898
REVERSAL
Std. deviation 7.87 6.92 7.06 6.87 6.72 0.49 0.66
Std. deviation 7.95 6.69 7.58 7.30 6.60 0.06 0.61
Average MV 604 909 963 999 867
CV(size) 0.51 0.45 0.41 0.43 0.61
REVERSAL non January

Std. deviation 7.73 6.77 7.03 6.80 6.70 0.25
Std. deviation 7.86 6.83 7.69 7.18 6.62 0.07
Average MV 600 903 989 995 865
IV
Std. deviation 5.28 6.58 7.44 7.93 7.74 0.94** 0.97
Std. deviation 5.12 7.08 8.18 8.21 9.14 0.35 0.94
Average size 1 871 937 467 216 266
CV(size) 0.52 0.48 0.41 0.44 0.61
IV non January
Std. deviation 5.24 6.56 7.38 7.91 7.56 1.17***
Std. deviation 5.17 7.13 7.97 8.18 9.09 0.32
Average size 1 946 980 446 231 278
LIQ
Std. deviation 5.46 6.76 6.76 7.41 8.49 0.50 0.97
Std. deviation 5.45 6.36 6.02 7.51 8.16 0.10 0.15
Average size 809 705 1 100 886 545
CV(size) 0.56 0.74 0.66 0.37 0.76
Table 6 (continued)
LowHigh p-value
1 2 3 4 5
Low High
LIQ non January

Std. deviation 5.23 6.73 6.72 7.40 8.48 0.17
Std. deviation 5.29 6.37 6.20 7.54 8.17 0.70
Average size 847 734 1 165 887 555
The portfolios built on momentum (MOM) and reversal (REVERSAL) are regrouped monthly based on the quantile values of their
continuously compounded past returns between t-12 and t-2 and t-36 and t-13, respectively. The idiosyncratic volatility, IV, is
proxied by the standard deviation of the individual stock returns over the past 36 months. As the liquidity measure, LIQ, we take the
average monthly turnover (trading volume divided by market capitalisation) over the last twelve months. The monthly returns are
calculated for the period between July 2002 and June 2011. The time series include 108 monthly returns (99 when Januaries are
excluded). The statistical signicance of the difference-in-mean test is marked with ***, ** and * at the 1%, 5% and 10% level,
respectively. The p-value for the MR test is shown in the last column. EW and VW stand for equal- and value-weighted average
portfolio returns, respectively. Average MV gives the mean capitalisation of the portfolio over the period considered (in millions
PLN). CV(size) is the average coefcient of variation, i.e., average ratio of standard deviation of size to its mean.
The results in the bottom part of Table 5 show that stocks with a high priceearnings ratio earn less than
those with a low ratio. The hedge portfolio premium is signicant for both types of returns, although it is
higher for the value-weighted return. The quantitative difference is mostly driven by low priceearnings ratio
stocks, where the weighted return is much higher, suggesting that within this portfolio the larger stocks
generate higher returns. The same effect is present in the non-January sample and in the subsamples. The
exclusion of Januaries results in insignicant equal-weighted arbitrage premium while the value-weighted
premium remains signicant and quantitatively similar. A sample expanded by return outliers delivers
similar results with slightly higher premiums for both types of returns in both January and non-January
samples. Evidence from subsamples suggests that the priceearnings ratio pattern diminishes in more recent
period. When looking at the whole pattern, we observe the U-shaped relation between priceearnings ratio
and subsequent returns, documented also by Fama and French (1992). The lack of monotonicity is conrmed
by extremely high p-values of MR test.
Table 6 summarises results for momentum, contrarian, idiosyncratic risk and liquidity portfolios. As
shown in the top part of the table, the stocks with poor performance over the last twelve months lag one
generate the lowest returns in the post-formation month. 28 The arbitrage premium is signicant for both
(non)-January samples. The value-weighted average returns are higher. Examining the subperiods, the
continuation in performance appears to be weaker in the more current months. Momentum pattern is
visible in both subperiods for both types of returns, but the signicance of payoffs to strategy diminishes
with time. Similarly as in Bhootra (2011), momentum payoffs are higher for value-weighted returns and
the extreme returns appear to have a strong inuence on the presence of momentum in our sample. When
included, the relationship between past performance and returns is not visible. Our result is mainly due to
the large difference between returns generated by past winners, which are much lower when extreme
returns remain in the sample. In our study, controlling for extreme returns means primarily eliminating
the highly negative outliers. When they are included, they strongly affect the past winners and drive down
their performance.
In the second part of Table 6 we present results for reversal in long-term past returns. Similarly as Fama
and French (1996), we present returns of portfolios sorted on past returns with exclusion of last year
before formation period. Traditional construction methodology using three to ve years of past returns
with one month delay delivers outcomes which suggest a at relation to subsequent returns. Our building
method does not produce signicant arbitrary premium but a visible pattern in equal-weighted returns.
The same pattern is present in cap-adjusted returns when the highest portfolio is omitted. However, MR
test rejects the hypothesis about potential monotonicity. The high return of past long-term winners may
28
The last month is omitted due to the rst-order autocorrelation documented by Jegadeesh (1990) and Lehmann (1990) and to
reduce the effect of the bidask bounce.
be caused by relatively high variation in size of stocks in this portfolio. The exclusion of January months
has the strongest inuence on low portfolios for simple returns and the pattern diminishes. In the Polish
market, a reversal effect is therefore not robust either to extreme returns control or to cap-adjustment.
Similarly as Rouwenhorst (1999), we show that past losers are on average smaller than past winners.
Losers tend to be also more volatile than average stocks, what is only partially in line with previous
evidence of Rouwenhorst (1999) showing that both marginal portfolios have more disperse returns.
For idiosyncratic risk portfolios we state that the results are highly dependent on the weighting
scheme. While equal-weighted return of portfolio with low idiosyncratic risk is signicantly higher than
the return of portfolios with high idiosyncratic risk, the effect disappears with value-weighting of returns.
The main reason for these inconclusive results is a clear relationship between idiosyncratic risk and size.
Most volatile stocks are on average six times smaller than most stable stocks. Value-weighting inuences
mainly returns of volatile portfolios. As expected, the inclusion of extreme returns changes the results.
While for the equal-weighted portfolios they remain qualitatively similar (the premium loses its
signicance), the pattern reverses for value-weighted portfolios. This result is mainly due to the lower
returns generated by stocks with low idiosyncratic volatility. This evidence presents a puzzle for Polish
stock market. Equal-weighted returns cleaned of extreme observations support the ndings of Ang et al.
(2006, 2009) regarding the negative relationship between idiosyncratic risk and expected returns.
However, in contrast to their robustness tests, the pattern is not persistent on the WSE. Our evidence
stands also in opposition to Bali and Cakici (2008) who show that the negative pattern vanishes for simple
average returns. The negative simple average returns of high-IV portfolio could be explained by the
reasoning of Fu (2009) who shows that the negative pattern documented by Ang et al. (2006, 2009) is
driven by a reversal in returns of highly volatile and small stocks. The high return of the middle portfolio
may suggest that investors are willing to buy stocks with average price uctuations over time. Due to
rising demand for such assets, the price increases in the post-formation period and so does the return.
We close our pattern investigation with liquidity portfolios. Liquidity concept is elusive, as mentioned by
Amihud (2002), and therefore hard to quantify using one single liquidity measure. We use turnover because it
allows for the separation between liquidity and the size effect which are otherwise difcult to distinguish. We
show the relatively smooth spread in size across the turnover portfolios, with the biggest stocks in the middle
portfolio. The last section of Table 6 reports that second and fth portfolios vary very strongly in average size.
Stocks with a high turnover over the last twelve months earn less, on average, than those which are less
liquid. This pattern is monotonic when the lowest portfolio is omitted. The MR test rejects the monotonicity
for both kinds of returns, however, the p-value of 0.15 is considerably smaller for cap-adjusted returns than
for simple returns (0.97). The hedge premium is insignicant for both the simple and the weighted returns,
and the results remain unaffected by the exclusion of Januaries. In the case of value-weighted portfolios, the
difference between the second lowest and the highest portfolio returns is signicantly positive (not shown).
Marginally low turnover portfolio is characterised by the lowest standard deviation. The volatility of returns
rises monotonically from illiquid to liquid stocks. Liquid stocks are also the smallest which is consistent with
the ndings of Lo and Wang (2000) for the subperiods of their study. Also Lee and Swaminathan (2000) show
that stocks with high turnover behave more like small stocks rather than those with low turnover which
would be expected if turnover proxied for liquidity. 29
Furthermore, Tables 5 and 6 show the statistics of the MR test. This test is designed to investigate the
monotonicity of the average returns across attribute-sorted portfolios. Thus it gives additional insight into the
relationship between the sorting variables and the expected portfolio returns. Only book-to-market (for
equally-weighted returns) and momentum patterns have clear monotonicity in their returns. These patterns
have dominated the research in asset pricing in recent years. This is because the value effect has not suffered
the deterioration characteristic for size effect, and momentum has been called centre stage anomaly in the
recent years. Therefore our study delivers an additional test for their robustness across markets.
As mentioned above, the return weighting scheme can inuence the return values and the picture of
the given pattern. Many studies do not deliver the results for both the simple and the cap-weighted
29
A possible explanation for the negative relationship between size and turnover, standing in opposition to general evidence
nding small stocks to be illiquid, is the as yet insufciently developed mutual fund industry on the WSE. As noticed by Lo and Wang
(2000), an increasing number of large institutional investors makes small-stock investments more difcult, due to the high
concentration of their invested capital.
average returns; instead, they mention solely the qualitative analogy between them. We argue that,
especially in the case of emerging markets where the distribution of assets size is dominated by micro and
small stocks, both methods require a direct comparative analysis. Empirical research delivers different
results for simple and value-weighted returns. Relying on our ndings we conclude that value-weighted
returns are persistently higher than equal-weighted returns. This effect may emerge from the fact that
bigger stocks have a tendency to earn more within portfolios after controlling for given attributes.
However, it may also result from extreme differences in capitalisation of stocks within portfolios that
causes high stock concentration. In such cases the construction of cap-adjusted returns mitigates the
inuence of a considerable portion of stocks within the portfolios. Contrary to the expectations, the
cap-adjustment does not decrease the variation in returns within portfolios but often increases it.
Above being said, in the time-series tests presented in the next section we focus on value-weighted
returns. This is a common procedure in the majority of studies since it allows for the reection of the real
investment opportunities that investors face.
3.2. FamaMacBeth (1973) cross-sectional test
In this subsection, we provide a supplementary test for the sorting approach presented above. The
main advantage of the Fama and MacBeth (1973) test is its ability to capture the simultaneous inuence of
many rm characteristics used as explanatory variables in the cross-sectional regressions. The test consists
of two steps. In the rst step, every month the expected stock returns (here proxied by realised returns)
are regressed on the past values of the chosen stock characteristic(s). We use log size, log book-to-market
ratio, priceearnings ratio, momentum, reversal, log turnover, and idiosyncratic volatility as regressors. As
in the previous subsection, we omit the individual beta coefcients, which is equivalent to assuming that
all of them are equal to unity. 30 In the second step, the simple mean of the coefcients obtained from the
regressions is calculated. The outcome of the difference-in-mean test indicates whether the given
coefcients help explain the cross-sectional variance across stocks, both in one- and multivariate
constellation. Each coefcient in the cross-sectional regression can be considered as the return of a
zero-cost (zero-investment) minimum-variance portfolio, with a weighted average of the corresponding
regressor equal to unity and the weighted averages of all the other regressors equal to zero. The weights
are tilted toward rms with more volatile returns. Because the returns on individual stocks can be
extreme, the potential for inuential observation problems emerges in Fama and MacBeth (1973)
regressions. For that reason, the sorts and cross-sectional test are used together.
Table 7 presents the results of the univariate and multivariate cross-sectional regression tests. The rst
line shows the average slopes and their corresponding t-values of the univariate regressions. When every
characteristic is used alone in the test, the signs of the coefcients are in line with the results of the sorts
test. The strongest dependency between a variable and the expected returns is observed for the
book-to-market ratio. The slope of 1.13% is almost 5 standard errors away from zero. Turnover also helps
explain the cross-section of the average stock returns at the 10% signicance level. Standard deviation
produces a high coefcient slope, which is 1.84 standard errors away from zero. The negative sign
indicates that stocks with small past volatility earn more than those with more time-varying returns. This
is in line with the ndings of Ang et al. (2009). In contrast to the sorting approach, which delivers
relatively strong evidence for the momentum effect, the short-term past performance with a slope of
0.31% and t-value of 0.83 does not help explain the cross-sectional variation in stock returns. This might
occur if the arbitrage premium from Table 6 is driven by extreme returns, which is probably due to the
high standard deviation of the returns in both marginal portfolios. Since the Fama and MacBeth (1973)
regression of the single stock returns is biased by the extreme observations, it is possible that systematic
relation over the whole sample is not found in this test. 31 As in many international studies working with
30
Use of the individual beta estimates, imprecise in their nature, is not expected to add to the understanding of the cross-section of
returns, see also Fama and French (2008) for further argumentation. The use of portfolio-based estimation is related to extra
complications due to the small number of stocks in our sample. Fama and French (1992) sort stocks into 100 portfolios, which would
result in 13 stocks per portfolio in our sample.
31
Both winners and losers tend to be more volatile than the average stock in a sample, because the ranking on past returns is
correlated with volatility, see Rouwenhorst (1999).
Table 7
Results of the cross-sectional Fama and MacBeth (1973) test.
Regression ln(MV) ln(BV/MV) P/E(+) MOM REVERSAL ln(LIQ) IV
Panel A: Univariate
0.13 1.13 0.00 0.31 0.30 0.22 16.72
(1.20) (4.72) (1.31) (0.83) (1.30) (1.44) (1.84)
Panel B: Multivariate
0.16 1.06 0.01 0.52 0.17 0.14 0.96
(1.10) (3.64) (1.38) (0.89) (0.58) (0.86) (0.06)
0.04 1.01
(0.34) (4.29)
1.06 0.20
(4.63) (1.30)
0.27 0.36
(2.38) (2.29)
1.06 0.21
(4.26) (2.27)
0.23 0.22
(2.07) (2.40)
0.19 1.17
(0.41) (1.05)
0.05 1.08 0.68
(0.44) (4.18) (1.71)
1.15 0.64
(4.54) (1.61)
This table shows the results for the Fama and MacBeth (1973) two-stage regression test, including seven different stock
characteristics. We run the univariate (Panel A) and multivariate (Panel B) regressions using size (ln(MV)), book-to-market ratio
(ln(BV/MV)), priceearnings ratio with positive earnings only (P/E(+)), momentum (MOM), reversal (REVERSAL), liquidity ln(LIQ)
and idiosyncratic risk (IV). Every month, we run a cross-sectional OLS regression, assigning lagged stock characteristics Kj (for J 1, 2,
3, 7) to their simple returns Ri,t as follows.
X
J
j
Ri;t 0;t j;t K i;t1 t (6)
j1
We run the test with up to seven predictors. We use the winsorize values for (ln(BV/MV)) and (P/E(+)), putting the smallest and
largest values equal to the 0.005 and 0.995 fractiles. The rst row for each representation shows the coefcient estimates in
percentages per month. The t-statistics of the Fama and MacBeth (1973) coefcient estimates are reported in parentheses.
more recent datasets, the market capitalization is negatively but insignicantly related to the stocks'
cross-sectional variation, see van Dijk (2011) for a comparative analysis.
Next, we consider different combinations of characteristics and the regression variant using all of them.
The results reveal the dominant role of the book-to-market ratio as the variable with signicant and robust
explanatory power for the cross-sectional variation of returns. The slopes on this characteristic are
consistently close to or over four standard errors from zero, independent on whichever other regressors are
used in the test. Moreover, the book-to-market ratio seems to absorb much of the size effect in the Polish
market. When size and the book-to-market ratio are used together, the slope on ln(MV) is only 0.04% with
a t-statistic of 0.34. On the other hand, if ln(MV) is used with liquidity or idiosyncratic volatility, it becomes
signicant. The signicance of loadings on liquidity or idiosyncratic volatility remains, although these
characteristics are often considered as competing proxies for the same source of correlation, i.e., the size
effect. However, as shown in the case of portfolios sorted on those variables, the relation between size and our
proxy for liquidity (turnover) is not strong. The increase in ln(MV) slope, in comparison to univariate
regression, suggests that both liquidity and idiosyncratic risk capture some extreme effects, preventing size
from being inuential in a univariate regression. Size also partially captures the effects otherwise covered by
the book-to-market ratio.
Table 8
The CAPM time-series regression results for the ve portfolios formed on size, value, priceearnings ratio, past-returns, idiosyncratic
volatility and liquidity.
Attribute Portfolio GRS p(GRS) || Std() SR() avRSQ
1 2 3 4 5
Low High
MV
0.0223 0.0101 0.0137 0.0019 0.0010 2.07 0.0746 0.0098 0.0064 0.32 0.59
t() 1.81 1.32 2.38 0.39 0.60
1.01 0.96 0.98 0.82 0.96
t() 6.6 9.04 15.41 12.94 26.23
RSQ 0.28 0.49 0.64 0.61 0.95
BV/MV
0.0137 0.0102 0.0011 0.0146 0.0093 3.88 0.0029 0.0098 0.0051 0.43 0.61
t() 2.25 1.75 0.27 2.79 2.13
0.88 1.05 0.94 0.79 0.92
t() 11.64 11.75 9.53 6.55 16.61
RSQ 0.51 0.60 0.74 0.54 0.68
P/E
0.0147 0.0075 0.0106 0.0151 0.0004 4.55 0.0010 0.0097 0.0056 0.47 0.57
t() 2.37 1.90 2.36 1.77 0.08
1.19 0.85 0.98 0.79 0.79
t() 11.99 9.77 8.86 7.45 12.97
RSQ 0.68 0.66 0.72 0.31 0.58
MOM
0.0063 0.0004 0.0042 0.0047 0.0135 2.48 0.0365 0.0058 0.0045 0.35 0.66
t() 1.18 0.08 1.22 1.30 2.69
0.95 0.85 0.85 0.73 0.82
t() 11.21 11.28 15.14 7.26 8.41
RSQ 0.67 0.68 0.71 0.67 0.60
REVERSAL
0.0073 0.0057 0.0046 0.0023 0.0090 2.35 0.0463 0.0058 0.0041 0.34 0.65
t() 1.80 1.44 1.07 0.68 1.86
0.97 0.72 0.86 0.91 0.74
t() 19.37 8.01 9.84 14.71 12.60
RSQ 0.72 0.56 0.62 0.74 0.60
IV
0.0014 0.0020 0.0135 0.0015 0.0032 1.83 0.1137 0.0043 0.0046 0.30 0.68
t() 0.55 0.59 2.86 0.28 0.44
0.65 0.88 0.96 0.93 1.01
t() 12.70 12.19 16.72 10.83 10.59
RSQ 0.77 0.75 0.67 0.61 0.58
LIQ
0.0034 0.0069 0.0049 0.0014 0.0010 1.14 0.3450 0.0035 0.0041 0.23 0.66
t() 1.02 1.63 1.29 0.39 0.18
0.62 0.75 0.69 0.93 0.92
t() 9.24 12.41 8.41 15.52 9.41
RSQ 0.61 0.66 0.64 0.74 0.61
The time-series regressions work with monthly, value-weighted portfolio returns constructed as described in Tables 5 and 6. and
stand for intercept and estimator of the sensitivity coefcient to the market-risk premium, respectively. t(.) gives the t-values for
estimator in brackets. || is the average absolute intercept for a set of regressions, std() is the average standard error of the
intercepts and SR() is the rooted numerator of Eq. (5). RSQ is the adjusted R2 for single portfolios and avRSQ is the average of RSQ
for set of portfolios. GRS gives the value of Gibbson et al. (1989) test and p(GRS) its p-value.
3.3. A (multifactor) risk-based explanation of return patterns
With the approach of Fama and MacBeth (1973) presented above, it is possible to recognise the
cross-sectional determinants of average returns. The objective of the time-series approach, discussed in
this subsection, is to examine the explanatory power of the asset pricing models that incorporate risk
premiums related to those determinants. Given that a pattern is deemed an anomaly if the CAPM is unable
to capture the differences in a portfolios performance, we rst confront this benchmark model with
portfolios sorted by different characteristics. Table 8 shows the intercepts and beta slopes for the portfolios
ranked by all the attributes discussed in the previous subsection. The CAPM is able to explain the variation
in the returns of liquidity and idiosyncratic volatility portfolios. In our sample, sorting on liquidity does not
produce clear patterns. This may be the reason for the success of CAPM. The single-factor model explains,
on average, 66% of the variation in returns and produces the smallest SR() as well as the lowest absolute
value of the intercepts. For idiosyncratic volatility, where the relationship between the sorting attribute is
followed by the relationship in size, we observe the increase in beta loadings from the least volatile to the
most volatile portfolio. This increase indicates that the differences in portfolio size are partially captured
by the beta coefcient. Furthermore, the returns on the lowest portfolio are measured with more
precision, as judged by the RSQ. Although it is on the marginal level for the p-value (0.11), the GRS test
does not reject the hypothesis that all the intercepts are jointly equal to zero. The marginal p-value
possibly results from the fact that the model is not able to capture the extra high return of a portfolio with
average (middle) historic volatility. For liquidity portfolios, we also observe a positive pattern in the CAPM
betas, which is in line with the ndings of Rouwenhorst (1999). The relation is, however, opposite to the
return pattern previously presented, where the less liquid portfolios earn more than the liquid ones. For
that reason, we observe a decrease in the intercepts from illiquid to liquid assets. The magnitude of the
pricing errors is, however, small. The ability of CAPM to explain liquidity-sorted portfolios is in line with
the results presented by Lischewski and Voronkova (2012), who conrm the insignicance of
risk-adjusted returns (alphas) in the CAPM time-series tests.
For size groups, the stocks in the lowest portfolio load stronger on the market risk premium than the
larger stocks. Nevertheless, the model signicantly underestimates the riskiness of this portfolio and
misses over 70% of its variance. It is not surprising since value-weighted market index is strongly biased
toward large stocks. This is visible in the high t-values for betas in highest size portfolio and its high RSQ.
For the portfolios ranked on book-to-market ratio CAPM overestimates the riskiness of low ratio stocks
and underestimates those with high ratio leaving the strong pattern in the intercepts. In the case of the
priceearnings ratio we document the widest spread in loadings on the market risk factor. The model
predicts higher average returns for low priceearnings ratio stocks. It underestimates, however, the
returns for low portfolios and overestimates those for higher portfolios and leaves the pattern in
intercepts observed in value-weighted returns. The intercepts in the momentum strategy also follow a
given pattern, where the model overestimates the returns for the past losers. These stocks load more on
the market premium. On the other hand, the model underestimates returns of winners, which are less
sensitive to MRP but generate higher returns due to the continuation in the stocks' short-term
performance. CAPM also fails by long-term reversal because it predicts too low return for long-term
winners. This is in line with the reversal pattern. However, the historical data deliver high average returns
for the long-term winners' portfolio.
After showing that the CAPM is not able to capture variation in stock returns related to most patterns,
in the next step, we confront the three-factor model with the same portfolio sets. This approach assumes
that risk has a multidimensional character that is captured by the common risk factors built into the
multifactor model. Thus, the model is expected to improve the description of the left-hand-side average
returns left unexplained by CAPM. For both the size and the value portfolios, the three-factor model
delivers more precise results and signicantly decreases the magnitude of the intercepts and increases the
RSQ, as shown in Table 9. The absolute value of the intercepts decreases from 0.81 to 0.36 for the size
portfolios and from 1.03 to 0.51 for the value portfolios. For the priceearnings ratio, the model also
improves statistics by increasing RSQ, particularly for the portfolios with higher values for the ratio.
Nevertheless, the GRS test leads to the rejection of the hypothesis about the cumulative zero-magnitude of
pricing errors. The momentum portfolios produce results qualitatively similar to those delivered by Fama
and French (1996). The past losers load more on the size premium, while the past winners load negatively
Table 9
Three-factor model time-series regression results for ve portfolios formed on value, priceearnings ratio and past-returns.
Portfolio Quantiles GRS p(GRS) || Std() SR() avRSQ
1 2 3 4 5
Low High
MV
0.0130 0.0064 0.0033 0.0000 0.0015 0.71 0.6209 0.0048 0.0048 0.20 0.71
t() 1.40 1.05 0.97 0.01 1.07
RSQ 0.43 0.62 0.82 0.72 0.96
BV/MV
0.0031 0.0011 0.0000 0.0113 0.0045 1.38 0.24 0.0040 0.0045 0.28 0.66
t() 0.84 0.21 0.00 2.05 1.06
RSQ 0.65 0.66 0.74 0.55 0.69
P/E
0.0151 0.0104 0.0089 0.0004 0.0025 3.41 0.0070 0.0075 0.0045 0.44 0.63
t() 2.68 2.57 2.05 0.09 0.59
RSQ 0.67 0.66 0.72 0.46 0.62
MOM
0.0086 0.0033 0.0042 0.0068 0.0159 3.03 0.0138 0.00786 0.0045 0.41 0.68
t() 1.56 0.64 1.18 1.92 3.41
b 0.95 0.83 0.86 0.74 0.84
t(b) 11.58 11.13 15.04 7.30 8.40
s 0.23 0.20 0.2400 0.05 0.14
t(s) 2.22 1.82 3.15 0.41 1.21
h 0.00 0.11 0.15 0.10 0.24
t(h) 0.01 1.38 2.27 1.16 2.47
RSQ 0.67 0.69 0.74 0.67 0.63
REVERSAL
0.0066 0.0050 0.0078 0.0003 0.0061 1.70 0.1417 0.0052 0.0042 0.31 0.67
t() 1.55 1.15 1.60 0.09 1.52
b 0.97 0.72 0.87 0.92 0.73
t(b) 18.34 7.87 10.96 14.82 13.92
s 0.23 0.01 0.19 0.01 0.45
t(s) 2.48 0.10 1.93 0.13 3.72
h 0.10 0.05 0.08 0.13 0.10
t(h) 1.55 0.69 0.86 1.77 1.32
RSQ 0.74 0.56 0.62 0.75 0.68
IV
0.0021 0.0038 0.0080 0.0035 0.0028 0.88 0.4975 0.0040 0.0046 0.22 0.71
t() 0.80 1.11 1.92 0.73 0.36
RSQ 0.78 0.75 0.73 0.68 0.59
LIQ
0.0045 0.0060 0.0073 0.0035 0.0023 1.48 0.2038 0.0047 0.0041 0.29 0.68
t() 1.18 1.62 1.85 0.92 0.43
RSQ 0.62 0.73 0.65 0.74 0.63
The time-series regressions work with monthly, value-weighted portfolio returns constructed as described in Tables 5 and 6. , b, s
and h stand for intercept and estimators of the sensitivity coefcient to the market-risk, the size and the value premiums,
respectively. t(.) gives the t-values for estimators in brackets. || is the average absolute intercept for a set of regressions, std() is
the average standard error of the intercepts and SR() is the rooted numerator of Eq. (5). RSQ is the adjusted R2 for single portfolios
and avRSQ is the average of RSQ for set of portfolios. GRS gives the value of Gibbson et al. (1989) test and p(GRS) its p-value.
on the value premium. In this way, the continuation in performance is missed because the model predicts
returns that are too high for the losers and too low for the winners. The intercepts become signicant and
all of the statistics for the momentum portfolios become worse. The opposite occurs for reversals, where
the model is able to capture the higher return of the past winners which load more on the size premium.
Also in the case of idiosyncratic volatility the three-factor model better describes returns of the upper
portfolios. It is consistent with the earlier nding that highly volatile stocks are, on average, smaller than
stocks with more stable returns. We see improvement in the regression ttings, particularly for the middle
portfolios. For liquidity portfolios, the three factor model rises RSQ from 0.66 to 0.73 for the second
portfolio, but does not show higher explanatory power for other portfolios. In contrast, it decreases the
estimation precision of the intercepts visible in higher values of || and SR(). This is in line with the
evidence of Lischewski and Voronkova (2012).
Summing up the performance of the three-factor model, we argue that it captures more variation in
portfolio returns. As expected, the most notable improvement is shown for size and value portfolios. It also
catches the reversal in returns on Polish market, although the regression statistics show a rather marginal
improvement in precision of regression t. The superiority of the three-factor model over CAPM in the
case of idiosyncratic risk is mostly due to the relationship between stocks' individual risk and size.
Similarly as for the U.S. market, the three-factor model misses the momentum effect. The portfolios built
on priceearnings ratio also present a challenge for the model. This characteristic remains insignicant in
the cross-sectional test but is presumably related to some other pattern-related attribute.
To complete the investigation of risk-based explanation of return patterns we apply the four-factor
model suggested by Carhart (1997). With the exception of the momentum portfolios, the model does not
capture more variation in the average portfolio returns than the three-factor model. The loadings on the
WML mostly remain insignicant. 32 However, the WML has a highly negative signicant loading on the
losers and a highly positive signicant loading on the winners for the momentum portfolios. The F-statistic
for the GRS test is 1.20, the intercept's absolute value and average standard error are 0.42 and 0.41,
respectively, the SR() is 0.27, and the average RSQ increases to 0.77 in comparison to the three-factor
model. Additional risk factor, therefore, contributes to the explanatory power solely in the case when the
left hand side of the regression in Eq. (4) is constructed with respect to the momentum factor. This
evidence is not surprising. Momentum portfolios are the biggest challenge for the three-factor model
which misses the short-term continuation in the stock's performance. The momentum is a dynamic, fast
changing phenomenon which is unlikely to be captured by slowly moving size and value effects.
In general, the three-factor model increases the precision of the time-series regressions and explains
almost all of the signicant adjusted returns (alphas) left by the CAPM, as observed in the lower GRS
statistics. In all cases, the multifactor model ts better the time-series data. Additionally, the average of the
standard errors for the intercepts is lower for the multifactor model, and the SR() statistics decrease in all
but the momentum and liquidity portfolios. In the case of momentum, it results from the general failure of
the three-factor model in describing momentum patterns. For sorts on turnover, the strongly negative
loadings of the value premium lead to a clearer underestimation of returns in comparison to the CAPM.
3.4. Relation between domestic (Polish) and global risk factors
After documenting the superiority of the multifactor model over CAPM, we investigate the interrelation
between the Polish risk proxying factors and their analogues in other countries. The question about the local
and global nature of the risk factors is one of the largest concerns in the modern day international asset
pricing literature. An increasing number of empirical papers have introduced global risk factors into pricing
models in order to compare their performance with purely domestic versions. In a world of perfectly
integrated efcient markets there is one common set of risk factors explaining the expected stock returns.
However, numerous studies have shown that local risk premiums capture more of the variation in stock
returns than their global counterparts. For example, van der Hart et al. (2005) show that global risk factors do
not contribute to the explanation of patterns in international portfolio returns and Grifn (2002) argues that
only local factors help explain global stock returns. A recently published study by Hou et al. (2011) shows that
international versions of multifactor asset pricing models including both global and local risk factors
outperform purely global alternatives. This is especially true for the emerging market subsample (including
Poland) considered in their study.
32
For that reason, the results are not shown, but are available upon request.
In what follows, we run the correlation tests between Polish, U.S., and German Fama and French (1993)
risk factors to examine the common movements in premiums in those markets. 33 We consider U.S. risk
factors due to the dominant role of the American nancial market on the international scene. Furthermore,
there is evidence of an inuence of the European stock markets, particularly the London Stock Exchange and
the Deutsche Brse, on the performance of the WSE, as shown by Schotman and Zalewska (2006). To include
regional (European) risk factors, we use those estimated for the German market. 34 As displayed in Table 10,
our proxy for the market risk premium is highly correlated (0.75 and 0.72) with both its U.S. and German
counterparts. At the same time our SMB risk factor is insignicantly (0.03 and 0.06) correlated with its U.S.
and German counterparts. There is no correlation between the Polish and German HML factor (0.02), and the
correlation with the American HML is 0.16. 35 The co-movement of three market returns is not surprising since
the U.S. economy, being the world's largest, has direct effects on the economies and nancial markets of most
other countries. Thus the prices of securities in U.S. nancial markets presumably incorporate information
about the economic fundamentals or market liquidity that is relevant to a broad cross-section of countries.36
At the same time, Germany is the largest economy in Europe and the most important trading partner for
Poland, while the Deutsche Brse is the second largest nancial market on the Continent. For these reasons,
the high correlation between market premiums is economically plausible.
The concern about the rationale behind cross-country differences in the time-series development of
risk premiums related to size and the book-to-market ratio may converge with the concern about the
sources of those premiums. Hou et al. (2011) show that for size, value and momentum factors the
magnitude and direction are country-specic. Fama and French (2012) deliver similar results for broader
regions of North America, Asia Pacic, Europe, and Japan. The descriptive statistics, displayed in Table 10,
show analogous differences in the average premiums for our sample. The weak co-movements between
Polish, American, and German risk factors indicate that they are driven by different sources and are rooted
more locally than globally. Therefore their link to the underlying universal state variables might be
questioned. Also the proposition about the segmentation of the Polish market is not convincing in the light
of similar ndings for the German market. Similarly low correlations between size and value-based risk
premiums across markets can be found in, e.g., Fama and French (1998) and Grifn (2002). Nevertheless,
the ndings presented here require further investigation which is beyond the scope of this paper.
4. Summary and conclusions
Our study focuses on two aspects of asset pricing on WSE. Its rst goal is to investigate the existence of
internationally recognised patterns in average portfolio returns and deliver additional evidence for an
unexplored but regionally important Polish equity market. We apply a sorting approach and cross-sectional
test of Fama and MacBeth (1973). We conrm the presence of patterns in average returns of portfolios sorted
by size, book-to-market ratio, priceearnings ratio, momentum and idiosyncratic risk in at least one type of
returns, simple or value-weighted. We argue, however, that the magnitude of the resulting hedge premiums,
i.e., the difference in returns of the marginal portfolios, depends on the return weighting scheme. The MR test
designed to track the persistence of a given pattern across portfolios conrms the monotonicity only for the
value- and momentum-sorted portfolios. This result is consistent with international evidence regarding the
special importance of these two attributes. The Fama and MacBeth (1973) test stresses the role of
book-to-market ratio as an determinant of cross-sectional variation in stock returns. However, none of other
attributes, although of the expected sign corresponding to sorts results, is found signicant in the test.
The second goal of this study is to challenge the CAPM with the documented patterns and investigate
whether they present anomalies or are a result of a higher sensitivity of stocks (portfolios) to the market
risk factor. We show that the CAPM leaves unexplained patterns in the intercepts in the case of all sorting
33
The American and German data come from Kenneth French and the Centre for Financial Research Cologne websites, and are
accessible at http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html and http://www.cfr-cologne.de/version06/
html/research.php?topic=paper&auswahl=data&v=data.
34
We limit the evidence for the European market to Germany since there is no equivalent database for the UK market.
Furthermore, the Fama and French (1993) model fails to reliably describe returns in the UK, as shown in Gregory et al. (2011).
35
We show the correlations for risk premiums constructed with the cut-off. The results for non cut-off factors are qualitatively the
same.
36
For example Roll (1988) shows that, in the crash of 1987, 19 out of 23 countries in the sample had a drop exceeding 20%.
Table 10
Correlation between Polish, U.S. and German Fama and French (1993) risk factors over 20022010.
Correlation test Descriptive statistics
Risk factor Country Poland Germany Country Mean Std
MRP Poland 0.92 7.14

Germany 0.72 Germany 0.49 6.09
USA 0.75 0.87 USA 0.50 4.83
SMB Poland 1.12 4.18
USA 0.06 0.06 USA 0.39 2.46
HML Poland 1.73 5.28
USA 0.16 0.18 USA 0.15 2.61
Table 10 presents the correlation coefcients between Polish, U.S. and German Fama and French (1993) risk factors over the period
July 2002 and December 2010 (102 observations). The sample is limited because, for the moment of publication, the German risk
factors are available only up to the end of 2010.
criteria except liquidity and idiosyncratic risk. For the rest of the sorting attributes, the GRS test rejects the
hypothesis that intercepts are jointly equal to zero. Applying the Fama and French (1993) three-factor
model with risk premiums constructed with respect to the specics of capitalisation disproportion on
Polish market, results in insignicant pricing errors for the size, value and reversal grouped portfolios. For
each sorting variable the multifactor model increases the precision of the regression estimates, leaving the
adjusted returns unexplained only in the case of momentum and priceearnings ratio. Continuation in
short-term returns is explained by four-factor model including momentum risk factor. However, this
expanded model does not capture additional variation in the time-series of expected returns for any other
grouping criteria.
This paper examines also the correlation between Polish and both American and German risk
mimicking premiums. While market risk factors exhibit high correlation across markets, size and value
premiums do not comove signicantly. This nding suggests the regional character of the two latter
premiums. The results are consistent for all three markets. However, they require further more detailed
investigation which is beyond the scope of this paper.
Our evidence sheds more light on the behaviour and pricing of equity stocks on WSE giving a
comprehensive overview of current trends in the asset pricing literature and using the most recent
descriptive methods. Further, our ndings uncover the issues regarding asset pricing and valuation on young
emerging markets. The skewness in market capitalisation among listed stocks seems to have a strong
inuence on results of standard methodology used in relevant empirical studies. With our paper, we also
hope to open a discussion about persistence and protability of momentum strategy which is known to be
very costly due to necessity of frequent portfolio regrouping. Since momentum remains insignicant in the
cross-section it seems to be a short lasting phenomenon which cannot possibly be exploited by market
players. The unexplained pattern in portfolios sorted on priceearnings ratio leaves a eld for search of
further explanation, both risk-based and behavioural. Finally, mixed results concerning idiosyncratic
volatility demand further exploration. Firstly, as shown by Umutlu et al. (2010), the degree of nancial
liberalisation is negatively related to total and idiosyncratic volatility in emerging economies. Secondly,
literature related to that topic uses very disperse methodology and, therefore, gives an opportunity for various
robustness tests which were beyond the scope of this paper. Regarding the discrepancy between domestic
and global size and value premiums further research on the application of global asset pricing models on
Polish data could be conducted.
References
Amihud, Y., 2002. Illiquidity and stock returns: cross-section and time-series effects. Journal of Financial Markets 5, 3156.
Amihud, Y., Mendelson, H., 1986. Asset pricing and the bidask spread. Journal of Financial Economics 17, 223249.
Ang, A., Hodrick, R., Xing, Y., Zhang, X., 2006. The cross-section of volatility and expected returns. Journal of Finance 61, 259299.
Ang, A., Hodrick, R., Xing, Y., Zhang, X., 2009. High idiosyncratic volatility and low returns: international and further U.S. evidence.
Journal of Financial Economics 91, 123.
Bali, C., Cakici, N., 2008. Idiosyncratic volatility and the cross-section of expected returns. Journal of Financial and Quantitative
Analysis 43, 2958.
Ball, R., 1978. Anomalies in relationships between securities' yields and yield-surrogates. Journal of Financial Economics 6, 103126.
Banz, R.W., 1981. The relationship between return and market value of common stocks. Journal of Financial Economics 9, 318.
Barry, C., Goldreyer, E., Lockwood, L., Rodriguez, M., 2002. Robustness of size and value effects in emerging equity markets,
19852000. Emerging Markets Review 3, 130.
Basu, S., 1977. Investment performance of common stocks in relation to their priceearnings ratios: a test of efcient market
hypothesis. Journal of Finance 32, 663682.
Basu, S., 1983. The relationship between earnings yield, market value and return for NYSE common stocks: further evidence. Journal
of Financial Economics 12, 129156.
Bekaert, G., Harvey, R., 1997. Foreign speculators and emerging equity markets. Working Paper 1441, Graduate School of Business,
Stanford University.
Bekaert, G., Harvey, R., 2002. Research in emerging markets nance: looking to the future. Emerging Markets Review 3, 429448.
Bekaert, G., Harvey, R., Lundblad, C., 2007. Liquidity and expected returns: lessons from emerging markets. The Review of Financial
Studies 20, 17831831.
Bhootra, A., 2011. Are momentum prots driven by the cross-sectional dispersion in expected stock returns? Journal of Financial
Markets 14, 494513.
Borys, M., Zemcik, P., 2011. Size and value effects in Visegrad countries. Emerging Markets Finance and Trade 47, 5068.
Carhart, M., 1997. On persistence in mutual fund performance. Journal of Finance 52, 5782.
Chan, K.C., Karceski, J., Lakonishok, J., 1998. The risk and return from factors. Journal of Financial Quantitative Analysis 33, 159188.
Claessens, S., Dasgupta, S., 1995. The cross-section of stock returns: evidence from the emerging markets. World Bank Policy
Research Working Paper 1505.
Cochrane, 2005. Asset Pricing. Princeton University Press, Princeton.
Czapkiewicz, A., Skalna, I., 2010. The CAPM and the FamaFrench models in Warsaw Stock Exchange. Przegld Statystyczny 57, 128141.
Czapkiewicz, A., Skalna, I., 2011. Uyteczno stosowania modelu Famy i Frencha w okresach hossy i bessy na rynku akcji GPW
w Warszawie. Bank i Kredyty 42, 6180.
Daniel, K., Titman, S., 1997. Evidence on characteristics of cross-sectional variation in stock returns. Journal of Finance 52, 133.
Davis, J., 1994. The cross-section of realized stock returns: the pre-Compustat evidence. Journal of Finance 49, 15791593.
Davis, J., Fama, E., French, K., 2000. Characteristics, covariances, and average returns: 19291997. Journal of Finance 55, 389406.
DeBondt, W.F.M., Thaler, R.H., 1985. Does the stock market overreact? Journal of Finance 40, 793805.
Fama, E., French, K., 1992. The cross-section of expected stock returns. Journal of Finance 47, 427465.
Fama, E., French, K., 1993. Common risk factors in the returns on stocks and bonds. Journal of Financial Economics 33, 356.
Fama, E., French, K., 1996. Multifactor explanation of asset pricing anomalies. Journal of Finance 51, 5584.
Fama, E., French, K., 1998. Value versus growth: the international evidence. Journal of Finance 53, 19751999.
Fama, E., French, K., 2007. Migration. Financial Analysts Journal 63, 4858.
Fama, E., French, K., 2008. Dissecting anomalies. Journal of Finance 63, 16531678.
Fama, E., French, K., 2012. Size, value and momentum in international stock returns. Journal of Financial Economics 105, 457472.
Fama, E., MacBeth, J., 1973. Risk, return, and equilibrium: empirical tests. Journal of Political Economy 71, 607636.
Fedorova, E., Vaihekoski, M., 2009. Global and local sources of risk in Eastern European emerging stock markets. Czech Journal of
Economics and Finance 59, 119.
Fu, F., 2009. Idiosyncratic risk and the cross-section of expected stock returns. Journal of Financial Economics 91, 2437.
Gibbson, M.R., Ross, S.A., Shanken, J., 1989. A test of the efciency of a given portfolio. Econometrica 57, 11211152.
Gregory, A., Tharyan, R., Christidis, A., 2011. Constructing and Testing Alternative Versions of the FamaFrench and Carhart Models in
the UK. Paper No. 11/02. University of Exeter.
Grifn, J.M., 2002. Are the FamaFrench factors global or country specic? Review of Financial Studies 15, 783803.
Harvey, C., 1995. Predictable risk and returns in emerging markets. Review of Financial Studies 8, 773816.
Heston, S., Rouwenhorst, G., Wessels, R., 1995. The structure of international stock returns and the integration of capital markets.
Journal of Empirical Finance 2, 173197.
Hou, K., Karolyi, G., Kho, B., 2011. What factors drive global stock returns? Review of Financial Studies 24, 25272574.
Jegadeesh, N., 1990. Evidence of predictable behavior of security returns. Journal of Finance 45, 881898.
Jegadeesh, N., Titman, S., 1993. Returns to buying winners and selling losers: implications for stock market efciency. Journal of
Finance 48, 6591.
Jegadeesh, N., Titman, S., 2001. Protability of momentum strategies: an evaluation of alternative explanations. Journal of Finance 56,
699720.
Kowerski, M., 2010a. Efekt wartoci ksigowej do rynkowej na Giedzie Papierw Wartociowych w Warszawie. Finansowy
Kwartalnik Internetowy e-Finanse 5.
Kowerski, M., 2010b. The analysis of an investment risk within emerging capital markets. The case of the Warsaw Stock Exchange.
Finansowy Kwartalnik Internetowy e-Finanse 6, 123.
Lakonishok, J., Shleifer, A., Vishny, R., 1994. Contrarian investment, extrapolation, and risk. Journal of Finance 49, 15411578.
Lee, C., Swaminathan, B., 2000. Price momentum and trading volume. Journal of Finance 55, 20172069.
Lehmann, B.N., 1990. Fads, martingales, and market efciency. Quarterly Journal of Economics 105, 128.
Lewellen, J., Nagel, S., Shanken, J., 2010. A sceptical appraisal of asset pricing tests. Journal of Financial Economics 96, 175194.
Lischewski, J., Voronkova, S., 2012. Size, value and liquidity. Do they really matter on an emerging stock market? Emerging Markets
Review 13, 825.
Lo, A., Wang, J., 2000. Trading volume: denitions, data analysis and implications of portfolio theory. The Review of Financial Studies 13,
257300.
Malkiel, B.G., Xu, Y., 2002. Idiosyncratic Risk and Security Returns. Working Paper, Princeton University.
Merton, R., 1987. A simple model of capital market equilibrium with incomplete information. Journal of Finance 42, 483510.
Patton, A.J., Timmermann, A., 2010. Monotonicity in asset returns: new tests with applications to the term structure, the CAPM and
portfolio sorts. Journal of Financial Economics 98, 605625.
Roll, R., 1977. A critique of the asset pricing theory's tests part I: on past and potential testability of the theory. Journal of Financial
Economics 4, 129176.
Roll, R., 1988. The international crash of October 1987. Financial Analysts Journal SeptemberOctober 1988, 1935.
Rosenberg, B., Reid, K., Lanstein, R., 1985. Persuasive evidence of market inefciency. Journal of Portfolio Management 11, 917.
Rouwenhorst, G., 1998. International momentum strategies. Journal of Finance 53, 267284.
Rouwenhorst, G., 1999. Local factors and turnover in emerging markets. Journal of Finance 54, 14391464.
Schotman, P., Zalewska, A., 2006. Non-synchronous trading and testing for market integration in Central European emerging
markets. Journal of Empirical Finance 13, 462494.
Serra, A.P., 2003. The cross-sectional determinants of returns: evidence from emerging markets' stocks. Journal of Emerging Market
Finance 2, 123162.
Stattman, D., 1980. Book values and stock returns. The Chicago MBA: A Journal of Selected Papers 4, 2545.
Umutlu, M., Akdeniz, M., Altay-Salih, A., 2010. The degree of nancial liberalization and aggregated stock-return volatility in
emerging markets. Journal of Banking and Finance 34, 509521.
van der Hart, J., Slagter, J., van Dijk, D., 2003. Stock selection strategies in emerging markets. Journal of Empirical Finance 10,
105132.
van der Hart, J., de Zwart, G., van Dijk, D., 2005. The success of stock selection strategies in emerging markets: is it risk or behavioral
bias? Emerging Markets Review 6, 238262.
van Dijk, M., 2011. Is size dead? A review of the size effect in equity returns. Journal of Banking and Finance 12, 32633274.
Voronkova, S., 2004. Equity market integration in Central European emerging markets: a cointegration analysis with shifting
regimes. International Review of Financial Analysis 13, 633647.
arnowski, J., 2007. Prba ustalenia przydatnoci zmiennych fundamentalnych do konstrukcji wielowskanikowego modelu rynku
w warunkach polskich. Conference Proceedings Chair of Financial Markets Jagielonski University Cracow. http://janek.ae.
krakow.pl/~fkrk/pliki/jz.pdf.

Emerging Markets Review: Antonina Waszczuk

Enviado por

Dados do documento

Título original

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

Emerging Markets Review: Antonina Waszczuk

Enviado por

Direitos autorais:

Formatos disponíveis

Emerging Markets Review 15 (2013) 186210

Contents lists available at SciVerse ScienceDirect

Emerging Markets Review

A risk-based explanation of return

Year Stocks listed Foreign stocks Stocks used IPOs Capitalisation

2002 217 0 177 6 25 260

2. Data and methodology

H0 : 0 vs: H 1 : min i > 0: 1

r t;p r t;f p p MRPt t;p 3

and three (four) factor model

Portfolio Mean Std Mean Std

Panel A: Flexible weighting

Panel B: Fixed weighting

Panel A: Statistics Mean Std Min Max Skewness Kurtosis t-value

SMBno 1.85 5.79 10.37 21.29 0.99 1.48 3.32

Panel B: Correlation SMBno SMByes HMLno HMLyes MRPe MRP WML

3.1. Patterns in average stock returns

Attribute Portfolio Premium MR test

BV/MV non January

P/E non January

Attribute Portfolio Premium MR test

MOM non January

REVERSAL non January

LIQ non January

3.2. FamaMacBeth (1973) cross-sectional test

Regression ln(MV) ln(BV/MV) P/E(+) MOM REVERSAL ln(LIQ) IV

Attribute Portfolio GRS p(GRS) || Std() SR() avRSQ

3.3. A (multifactor) risk-based explanation of return patterns

Portfolio Quantiles GRS p(GRS) || Std() SR() avRSQ

3.4. Relation between domestic (Polish) and global risk factors

4. Summary and conclusions

Correlation test Descriptive statistics

Risk factor Country Poland Germany Country Mean Std

MRP Poland 0.92 7.14

Você também pode gostar