fama and french factors in australia€¦ · michael a. o’brien⁄ uq business school the...
TRANSCRIPT
Fama and French Factors in Australia
Michael A. O’Brien∗
UQ Business School
The University of Queensland
Qld Australia 4072
Phone: +61 7 3346 9327
e-mail: [email protected]
October 2007
Abstract
This study analyses the size and book to market effects and the ability of the Fama and
French (1993) three factor model to explain the cross-section of returns. Previous studies in
Australia have suffered from data limitations due to the difficulty in obtaining a comprehensive
series of accounting data. This study overcomes these limitations by hand collecting accounting
information on over 98% of all listed companies during the period 1981 to 2005. This study
finds that the Fama and French (1993) model provides increased explanatory power in explaining
the cross-section of returns in Australia when compared to the Capital Asset Pricing Model
(CAPM). In contrast to previous Australian studies this is due to both size and book to market
effects playing a role in asset pricing.
∗I wish to thank my advisers Tim Brailsford, Clive Gaunt and Jamie Alcock who have provided constructive insights,
David Forster who has provided valuable managerial advice and skills during the construction of the database and Alina
Hale with constructive comments. The author gratefully acknowledge financial assistance provided by Dimensional
Fund Advisors (DFA) Australia and the Australian Research Council through ARC Linkage Grant (LP0453913)
1
Preface
Title of Thesis: Risk and Growth in Australia: Are SMB and HML Proxies for Extreme Risk?
Advisers: Professor Tim Brailsford, Dr Clive Gaunt and Dr Jamie Alcock.
Numerous studies have documented the ability of financial variables to explain the cross-section of
returns. These variables include market capitalisation, past stock returns, earnings yield, leverage
and book to market ratios. Bringing this evidence together Fama and French developed a three factor
model to explain the cross-section of returns. This model has been highly successful in explaining
the cross-section of returns but as it is an empirical model the reason why it has preformed so well
is unknown. In Australia only a few studies have analysed whether book to market ratios or the
Fama and French three factor model can explain the cross-section of returns. This thesis attempts to
rectify these gaps by analysing the book to market ratio in Australia and test the Fama and French
model over a 25 year period. One reason for the lack of studies is due to the lack of a comprehensive
accounting data in Australia. To rectify this accounting data is hand collected from annual reports
over the period 1981 to 2005. The thesis also adds to the debate on why the Fama and French factors
can explain asset returns by seeing whether they proxy for risk associated with extreme downside
movements.
The thesis is structured as follows:
• Chapter 1: Introduction.
• Chapter 2: Data.
• Chapter 3: Value and Growth Anomalies.
• Chapter 4: Disentangling Size from Momentum in Australian Stock Returns.
• Chapter 5: Fama and French Factors in Australia.
• Chapter 6: Asymmetries and the Fama and French Factors.
• Chapter 7: Conclusion.
The following study is based on Chapter 5.
2
1 Introduction
Over the last twenty-five years several studies have documented the ability of certain variables to
explain the cross-sectional variation in returns which can not be explained by the Capital Asset
Pricing Model (CAPM).1 These variables include past stock returns (De Bondt and Thaler, 1985,
1987; Jegadeesh and Titman, 1993, 2001), size (Banz, 1981; Brown, Keim, Kleidon, and Marsh,
1983; Reinganum, 1981), earnings yield (Basu, 1977, 1983; Jaffe, Keim, and Westerfield, 1989),
leverage (Bhandari, 1988) and book to market value (Chan, Hamao, and Lakonishok, 1991; Fama
and French, 1992; Rosenberg, Reid, and Lanstein, 1985). Bringing this evidence together Fama
and French (1992) found that size and book to market values are the variables with the strongest
relationship to returns and that many of the other variables explanatory power vanished. Building on
this work Fama and French (1993) propose an empirically driven asset pricing model that captured
the returns to the size and book to market premiums by forming two mimicking factors, small
minus big (SMB) and high minus low (HML). By including these two factors with the market risk
premium they show that the three factor model captures the majority of the common variation in
returns. Subsequently the model has become the benchmark of asset pricing models, and subsequent
studies have tried to understanding why the factors explain the cross-sectional variation of returns.
Three main explanations are proposed. First, the variables are proxies for underlying risk in the
market (Davies, Fama, and French, 2000; Fama and French, 1993, 1995, 1996, 1998) in the spirit
of the Arbitrage Pricing Theory (APT) (Ross, 1976) or the Intertemporal Capital Asset Pricing
Model (ICAPM) (Merton, 1971, 1973). Second, that they are capturing behaviourial biases of the
investors and inefficiencies in the market (Daniel and Titman, 1997; Daniel, Titman, and Wei, 2001;
Lakonishok, Shleifer, and Vishny, 1994; LaPorta, Lakonishok, Shleifer, and Vishny, 1997; Skinner
and Sloan, 2002; Teo and Woo, 2004). Third, that they are a result of data-snooping (Black, 1993a,b;
Kothari, Shanken, and Sloan, 1995; Lo and Mackinlay, 1990).
In Australia a number of market anomalies have also been reported, including past stock returns
(Demir, Muthuswamy, and Walter, 2004; Durand, Limkriangkrai, and Smith, 2006b; Gaunt and
Gray, 2003; Hurn and Pavlov, 2003) and size (Beedles, Dodd, and Officer, 1988; Brown, Keim,
Kleidon, and Marsh, 1983; Durand, Juricev, and Smith, 2007; Gaunt, Gray, and McIvor, 2000).
These studies are generally consistent with US evidence, indicating that these variables may be
proxies for systematic risk or behavioural bias. Unfortunately few studies have studied how earnings
yield (Allen, Lisnawati, and Clissold, 1998), leverage and book to market effects (Gaunt, 2004;
Gharghori, Chan, and Faff, 2006; Halliwell, Heaney, and Sawicki, 1999) influence stock returns in
Australia. One of the reasons for this paucity of research in this area has been due to the lack
of a comprehensive accounting database in Australia. This has also meant that there are only
a few studies that have tested the Fama and French (1993) three factor model in Australia and1Black (1972); Lintner (1965a,b); Mossin (1966); Sharpe (1964)
3
that these studies have generally focused on the mid to late 1990’s when accounting data is more
available (Durack, Durand, and Maller, 2004; Durand, Limkriangkrai, and Smith, 2006a; Gaunt,
2004; Gharghori, Chan, and Faff, 2006; Halliwell, Heaney, and Sawicki, 1999). The few studies
generally find that the three factor model has higher explanatory power compared to the CAPM in
explaining the cross-section of returns in Australia, but this is primarily driven by the SMB factor.
This result must be viewed with caution because of two reasons. First, the co-efficent on the SMB
factor is negative in some studies (Faff, 2001, 2004) and positive in others (Durack, Durand, and
Maller, 2004; Durand, Limkriangkrai, and Smith, 2006a; Gaunt, 2004; Gharghori, Chan, and Faff,
2006) even though similar time periods are studied. Second, these studies only cover 35% of the
market, on average, because book values of the other companies were unavailable. This could lead
to in-correct formation of the HML factor leading to its insignificance in these studies.
The purpose of this study is three fold. First, to significantly increase the coverage of accounting
information in Australia, and to expand this coverage into the 1980’s. Previous studies in Australia
have had limited access to accounting variables limiting the scope and potentially influencing the
results of these studies. This study hand collects accounting information from over 98% of all
companies that produced an annual report over the period 1981 to 2005. This is a significant
increase over previous studies in Australia and allows a through investigation into the size and value
premiums in Australia.
By increasing the coverage and expanding the time period of previous studies we address the second
purpose of this study, which is in response to Lo and Mackinlay (1990) who argue that results of
asset pricing tests need to be examined out-of-sample to ensure that data snooping has not occurred.
The primary location to test the Fama and French (1993) three factor model has been the US, with
only limited studies performed outside the US (Bagella, Becchetti, and Carpentieri, 2000; Daniel,
Titman, and Wei, 2001; Fama and French, 1998). These international studies generally find that
the Fama and French (1993) three factor model has increased explanatory power over the CAPM.
By analysing the size and value premium in a relatively unexplored data set we can form a more
comprehensive understanding of whether the model success is a result of data-snooping.
The final motivation for this research is to broaden the debate on the appropriate asset pricing
model to be used in Australia. Australian research on the CAPM and competing asset pricing
models have found that competing models provide additional power over the CAPM in explaining
the cross-section of returns, but they do not produce drastic improvements over the CAPM. Data
limitations and the length of the period studied has also generated doubt regarding the appropriate
model to apply. By increasing the time period and coverage of companies in the sample we will allow
a more informed debate to develop on the appropriate asset pricing model to use.
Consistent with overseas evidence, results indicate that returns in Australia are positively related
to book to market values. Results also indicate that small firms earn a premium over large firms,
4
but there is a strong non-linearity between size and returns. This non-linearity causes stocks in
the middle size portfolios to under-perform both the stocks in the small and large portfolios. We
further analyse the SMB and HML factors in Australia and find that there is a strong June and
July seasonality in the SMB factor consistent with previous evidence in Australia (Brown, Keim,
Kleidon, and Marsh, 1983; Durand, Juricev, and Smith, 2007). We also find an unusually large
negative return in the HML factor occuring in 1989, which seems to be driven by severe distress in
value stocks leading into 1989 and extremely large negative returns from a few large stocks in the
value portfolio. The results indicate a low correlation between the market risk premium and the two
factors, indicating that they are capturing other underlying risk or behaviourial bias.
We then compare the CAPM against the Fama and French (1993) three factor model over the
period 1982 to 2006. These results demonstrate that the Fama and French (1993) model explains
a significantly higher proportion of the cross-section of returns in Australia when compared to the
CAPM. In contrast to previous studies in Australia we find that both the SMB and HML are
influential in explaining the cross-section of returns. These results also indicate that the Fama and
French (1993) model can not explain the under-performance of middle size portfolios and can only
explain, on average, 70% of the cross-section of returns.
The remainder of this paper is organised as follows. Section 2 reviews the international and Aus-
tralian literature relevant to this research. Section 3 presents the data and discusses our testing
framework. Section 4 analyses the SMB and HML factors in Australia, while Section 5 presents our
results. Section 6 summaries and concludes.
2 Literature Review
The capital asset pricing model (CAPM) is the foundation for most asset-pricing models in finance.
This model specifies that expected return is the product of the risk free rate and the expected
premium for risk. The expected risk premium is a function of the asset’s covariance with the market
return. This model provided the finance community with an elegant method to value risky assets.
Unfortunately the model relies on a number of assumptions including; using a single period model,
perfect information and frictionless markets. Since these restrictions are unrealistic in financial
markets new theoretical models were developed that relaxes some of these assumptions. Fama (1970)
extends the original CAPM into an intertemporal setting and demonstrates that if preferences and
future investment opportunity sets are constant, then an intertemporal utility maximiser can be
treated as if they have a single period utility function. This means that an intertemporal CAPM has
the same structure as the single-period model. However, these assumptions on the utility maximiser
are restrictive. Merton (1971, 1973) demonstrates that if opportunity sets are stochastic then an
intertemporal CAPM (ICAPM) would be a linear multi-factor model. Ross (1976) develops a new
5
approach on how to value risky assets. His model, the Arbitrage Pricing Theory (APT), is developed
from a statistical charateristic of returns. This model assumes that there are a common component to
stock returns plus an idiosyncratic component (Cochrane, 2005). Using the principle of diversification
and arbitrage Ross (1976) demonstrate that the idiosyncratic component should not be priced.
Hence, the expected returns of each stock will be determined by its covariance with each common
component or “factor.” Therefore the model proposes that the expected return on an asset will be a
linear function of k factors. The model does not specifically state what these factors are, but using
factor analysis and economic arguments these factors could be selected and the model tested.
Despite being theoretically elegant, the CAPM has preformed poorly in empirical studies probably
because its assumptions are not meet in financial markets. Early empirical studies indicate that the
security market line is flatter than predicted (Black, Jensen, and Scholes, 1972; Fama and MacBeth,
1973). Subsequent research indicates that other factors are successful in explaining the proportion of
excess returns not explained by beta. Banz (1981); Reinganum (1981) demonstrate that the size of a
stock has explanatory power in explaining the cross-section of expected returns. These studies finds
that average returns on small stocks are higher than large stocks given the stock’s respective beta.
Bhandari (1988) analyses the debt/equity ratio. The debt/equity ratio could be related to the risk
of a stock, but the CAPM argues that the leverage effect should be captured by beta. In contrast
to the predictions of the CAPM, Bhandari (1988) finds a positive relationship between debt/equity
ratio and returns. Basu (1983) finds that high earnings/price ratio (E/P) stocks earn a statistically
significant positive return after controlling for beta and size. Chan, Hamao, and Lakonishok (1991)
find that in the Japanese market similar anomalies occur with the book to market ratio (B/M) and
cash flow yields have a significant positive impact on expected returns. While Chen, Roll, and Ross
(1986); Roll and Ross (1980) suggest that economic variables influence excess return. De Bondt and
Thaler (1985, 1987) demonstrate that stock returns over three to five years have explanatory power
over future returns. These studies demonstrates that stocks that have out performed the market,
over the last three to five years, subsequently under perform the market. In contrast, stocks that
have under performed the market, over the last three to five years, subsequently out perform the
market. Jegadeesh and Titman (1993, 2001) demonstrate that returns over 3 to 12 months also have
predictive power over future returns with winner portfolios continuing to outperform stocks that had
underperformed. This evidence indicates that stocks are priced with factors other than beta.
The empirical failure of the CAPM, with evidence suggesting that other factors can explain a pro-
portion of excess returns, lead to the development of empirically driven multi-factor models based
on the theoretical arguments of the APT and ICAPM. Ball (1978) proposes that yield surrogates,
such as E/P and dividend yields, are correlated with returns because they proxy for underlying risks
not accounted for by traditional risk measures. Using this argument Fama and French (1992, 1993)
analysed yield surrogates, size and B/M in an attempt to develop an empirically driven model. Fama
and French (1992) demonstrate that average stock returns are not positively related to their market
6
betas when portfolios are formed on size and beta. These results indicate that size and B/M play
a significant role in explaining the cross-section of expected returns, while yield surrogates do not
add explanatory power once size and B/M effects are taken into account. Fama and French (1993)
extend Fama and French (1992) by creating two portfolios based on market capitalisation and B/M
ratios that they call SMB (small minus big) and HML (high minus low). These portfolios are de-
signed to proxy for the underlying factor that drives the size and B/M effect. They demonstrate
that the market risk premium, SMB and HML play an important role in explaining the cross-section
of expected returns for stocks. Using this model they subsequently show that the model can explain
the size, book to market, earnings yield, leverage and long term reversal anomalies, but it can not
explain medium term momentum (Fama and French, 1996).
The success of the Fama and French (1993) three factor model lead to a number of studies analysing
why it works. Three major explanations have emerged. First, that the factors SMB and HML
are additional risk factors in the spirit of the ICAPM (Merton, 1971, 1973) or APT (Ross, 1976).
A number of studies argue that the success of the model in explaining a majority of the market
anomalies is proof that they are proxies for underlying risk (Davies, Fama, and French, 2000; Fama
and French, 1995, 1996). Studies have also indicated that SMB and HML are can predict future
GDP in some countries (Liew and Vassalou, 2000) and that consumption wealth ratio is related
to the SMB and HML factor (Lettau and Ludvigson, 2001). Second, that behaviourial bias in
investors and market inefficiency leads to these persistent anomalies in the market (Daniel and
Titman, 1997; Daniel, Titman, and Wei, 2001; Lakonishok, Shleifer, and Vishny, 1994; LaPorta,
Lakonishok, Shleifer, and Vishny, 1997; Skinner and Sloan, 2002; Teo and Woo, 2004). Third, that
data-snooping has lead to the success of the model (Black, 1993a,b; Kothari, Shanken, and Sloan,
1995; Lo and Mackinlay, 1990).
2.1 Australian Evidence
In Australia, tests of the CAPM and its implications in Australia have found similar results to that
observed overseas. Earlier studies suggested support for the CAPM (Ball, Brown, and Officer, 1976)
but anomalies similar to overseas evidence are also prevalent in Australia. These regularities have
included seasonalities (Brailsford and Easton, 1991; Officer, 1975), medium term momentum (Demir,
Muthuswamy, and Walter, 2004; Durand, Limkriangkrai, and Smith, 2006b; Gaunt and Gray, 2003;
Hurn and Pavlov, 2003) and the size anomaly (Beedles, Dodd, and Officer, 1988; Brown, Keim,
Kleidon, and Marsh, 1983; Durand, Juricev, and Smith, 2007; Gaunt, Gray, and McIvor, 2000).
This lead to tests of multi-factor models which suggested mixed support for the APT (Faff, 1988).
In contrast, there has been only limited studies on the book to market (Gaunt, 2004; Gharghori,
Chan, and Faff, 2006; Halliwell, Heaney, and Sawicki, 1999), earnings yield (Allen, Lisnawati, and
Clissold, 1998) and other leverage anomalies. The lack of studies in Australia on B/M, earnings
7
yield and leverage has been due to the difficulty in obtaining a comprehensive series of accounting
data. This lack of accounting information has also resulted in only limited studies on the Fama and
French model.
Halliwell, Heaney, and Sawicki (1999) was the first study that replicated the Fama and French (1993)
methodology in Australia using data from an eleven year period (1981 to 1991). The study only
collected, on average, accounting data on 350 companies per year. over the eleven year period there
existed, on average, 1370 companies per year, meaning only 26% of the total number of companies are
covered. As the authors acknowledge, this lack of accounting data means that the sample is heavily
skewed towards larger stocks, potentially influencing the results. Their study reports evidence of
premiums to small firms and high book to market firms, though it is not as strong or as consistent as
overseas studies. Their tests of the three factor model indicate that the SMB factor is significant and
positively related to size. In contrast to overseas studies there is little evidence that the HML factor
is significant in explaining the cross section of returns. Overall their results indicate that the Fama
and French (1993) model provide marginal improvement over the CAPM in Australia in explaining
the cross-section of returns. Further this improvement is solely due to the SMB factor.
Faff (2001, 2004) provides further evidence for the three factor model in Australia using ‘off the
shelf’ style indexes to construct the SMB and HML factors. Faff (2001, 2004) uses four Australian
equity style indexes provided by the Frank Russell Company. These indexes are the the ASX/Russell
Value 100, ASX/Russell Growth 100, ASX/Russell Small Value and the ASX/Russell small Growth
index. Faff (2001) examines the three factor model over the period 1991 to 1999 which provides an
external validity test of the Halliwell, Heaney, and Sawicki (1999) study. This study finds strong
support for the three factor model, but there is a significant negative size premium, rather than the
expected positive premium. Faff (2001) argues that this result is consistent with the recent evidence
of the reversal of the size premium (Dimson and Paul, 1999; Horowitz, Loughran, and Savin, 2000).
In contrast to Halliwell, Heaney, and Sawicki (1999) the study finds stronger support for the HML
factor being a priced factor in Australia. Faff (2004) re-examines the three-factor model over the
the period 1996 to 1999 using daily returns. Similar results are observed to his previous study and
the premium on the SMB factor is again negative.
Gaunt (2004) tests the Fama-French model in Australia over a ten year period spanning 1991 to
2000. The sample has an average of 650 companies, while the average number of companies during
the 10 year period studied was 1310. This is a significant increase in coverage over Halliwell, Heaney,
and Sawicki (1999), but only 50% of the market is considered. The methodology also means that
the sample is skewed towards larger more establish firms. Gaunt (2004) demonstrates the negative
association between size and returns with the smallest quintile of firms being the driver of the results,
with the other size quintiles having similar returns. His study also has the strongest evidence of high
book to market stocks earning a premium over low book to market stocks in Australia. The regression
results using the Fama and French (1993) model are generally consistent with Halliwell, Heaney, and
8
Sawicki (1999), with the market risk premium and SMB being highly significant and positive. In
contrast to Halliwell, Heaney, and Sawicki (1999) Gaunt (2004) results demonstrate that HML plays
a role in explaining the cross-section of returns, although its role is minor when compared to the
market risk premium and SMB. He also demonstrates that the Fama and French (1993) three-factor
model provides increased explanatory power over the CAPM.
Durack, Durand, and Maller (2004) uses a similar time period to Gaunt (2004) to study the per-
formance of the conditional CAPM (Jagannathan and Wang, 1996) against the Fama and French
(1993) three-factor model. The sample used in this study is significantly smaller than Gaunt (2004)
study and has access to only 264 companies book value per year on average. This implies that only
20% of all companies are included in the formation of the HML factor. Results indicate that the
three factor model outperforms the conditional CAPM. The result seems to be primarily driven by
the important role SMB plays in the Australian market. However, the results for the HML factor are
inconclusive and again suggests that HML is not priced. These results leads Durack, Durand, and
Maller (2004) to suggest further research is required on the book to market anomaly in Australia.
Durand, Limkriangkrai, and Smith (2006a) re-analyse the period and data studied by Durack, Du-
rand, and Maller (2004) and tests whether US or Australian factors are better at explaining the
cross-section of assets returns. They find that the SMB factor is highly significant and positive,
but that the HML is generally insignificant. However when the US factors are used more HML
co-efficient’s are found to be significant. This would suggest that HML may play a role in Australia,
although why only the US HML and not the Australian HML is priced has not been studied. Gen-
erally, using the US factors instead of the Australian factors leads to lower explanatory power in the
model and more noisy estimates, suggesting that local factors are important.
Recently Gharghori, Chan, and Faff (2006) analysed whether the Fama and French (1993) factors
proxy for risk or behaviourial bias. The study covers the period 1992 to 2003, but the number of
companies that are covered ranges from 35% to 50% and are skewed towards larger more established
companies. Similar to Gaunt (2004) the results indicate a strong relationship between book to market
and returns. The test of whether the factors are compensation for differences in risk or are capturing
behaviourial bias indicate some support for the risk based argument. Results also indicate that the
HML factor, in particular, is a proxy for underlying risk, while the results from the SMB factor are
inconclusive. Overall Gharghori, Chan, and Faff (2006) suggest that a longer more extensive data
set is required for this research.
The confusing results in testing the Fama and French three factor model on Australia data, in par-
ticular the lack of evidence of why the Australian HML and SMB factor is in-consistently priced,
suggests that more studies of the model are required. The conflicting results, even when using simi-
lar time-periods, may indicate that the short time periods analysed and substantial data limitations
could be influencing the results. In particular, all previous studies have had limited access to ac-
9
counting data meaning that less than half of all firms in the market are included when forming book
to market portfolios. Overall the number of companies covered in the Australian studies has been
around 35% of the total firm population. By hand collecting book values, this study is the first
Australian study to comprehensively analyse the book to market anomaly.
3 Data and Methodology
3.1 Data
Previous studies in Australia on the book to market anomaly has suffered because of a lack of quality
accounting data. To rectify this problem this study hand collects accounting information from annual
reports for the period 1981 to 2005. Company annual reports were catalogued and stored by the
Australian Stock Exchange (ASX) and its forebears until the late 1990s. These reports are stored
in each state library throughout Australia and are the primary source of our annual reports for the
period 1981 to 1997. We have further collected hard and electronic copies of annual reports from
companies to supplement and extend the period of reports beyond 1997. This has been accomplished
through a variety of methods including accessing reports directly from the companies, either through
their website or directly requesting a copy. This allows us to then match the accounting data
to readily available price data sources including the Australian Graduate School of Management
(AGSM) Centre for Research in Finance (CRIF) database. Table 1 records the number of annual
reports collected, the number of companies that did not produce an annual report during the year,
the number of companies that we can not find an annual report for and the number of companies
with price data in CRIF. This indicates that we have been able to collect data from approximately
98% of all companies that produced an annual report during the time period considered. This is a
substantial increase over previous studies in Australia where accounting information coverage was
always less than 50% (Gaunt, 2004) and in most cases less than 25% (Durack, Durand, and Maller,
2004; Durand, Limkriangkrai, and Smith, 2006a; Halliwell, Heaney, and Sawicki, 1999) of companies
listed.
From the annual reports the following information is collected to allow us to calculate book values;
• total value of equity,
• outside equity interests,
• value of preference shares capital,
• future tax benefits.
Following Fama and French (1992, 1993) we define book value as the total value of equity minus
outside equity interests, the value of preference shares capital and future tax benefits. Consistent
10
Table 1
Number of annual reports collected per year
Year Number of annual Number of firms Number of firms Number of unique
reports collected that did not produce that produced an firms in CRIF
an annual report annual report but
can not be found
1981 835 73 111 1019
1982 855 72 61 988
1983 846 81 49 976
1984 892 81 39 1012
1985 976 145 31 1152
1986 1190 167 42 1399
1987 1572 206 49 1827
1988 1624 231 33 1888
1989 1437 369 27 1833
1990 1272 327 15 1614
1991 1135 245 5 1385
1992 1037 185 5 1227
1993 1054 128 2 1184
1994 1155 70 3 1228
1995 1167 64 2 1233
1996 1164 88 3 1258
1997 1155 98 27 1280
1998 1181 83 21 1285
1999 1227 118 17 1362
2000 1337 136 10 1483
2001 1371 118 7 1496
2002 1384 117 7 1508
2003 1410 116 1 1527
2004 1522 128 0 1650
2005 1634 132 0 1766
11
with previous studies companies, with negative book value are dropped from the sample. We then
match the book values to market capitalisation information from the AGSM-CRIF database. To be
consistent with previous literature and to avoid any look ahead bias, because the release of accounting
information is later than the balance date on the annual report, we only use accounting information
that is at least 6 months old. For example, when we are calculating book values for December 1982
only accounting information released prior to and including 30 June 1982 can be used. Book to
market ratio is then calculated as the book value dividend by the market capitalisation following
the methodology in Fama and French (1992, 1993). For example, if a company has a balance date
of 31 December and we are calculating book to market value in December 1982. We use the book
value from the annual report with a balance date of 31 December 1981 and use the stocks market
capitalisation as of 30 June 1982.
Throughout the study price and market capitalisation information from the AGSM-CRIF database
is utilised. The AGSM-CRIF database contains monthly prices, market capitalisation dividends,
adjustments for capitalisation changes and returns for all Australian Stock Exchange (ASX) listed
stocks. This database also contains the monthly 13-week treasury note yield and the value-weighted
monthly market return of all stocks in the database. In this study we remove all property trusts2
and investment funds from the database. Using our accounting information and the AGSM-CRIF
database allows us to to analyse the CAPM and the Fama and French (1993) three factor model
for the period 1982 to 2006. This sample period covers the combined periods studied by Durack,
Durand, and Maller (2004); Durand, Limkriangkrai, and Smith (2006a); Gaunt (2004); Gharghori,
Chan, and Faff (2006); Halliwell, Heaney, and Sawicki (1999) and extends these studies to 2006.
This significantly extends the previous Australian studies to a 25 year period.
3.2 Portfolio Construction
To test the CAPM and Fama and French (1993) model we follow the portfolio formation technique of
Fama and French (1993) and construct 25 portfolios. First, each December all stocks in our sample
are ranked by their book to market value and each stock is assigned to one of five book to market
portfolio, where each portfolio contains an equal number of stocks. The first portfolio (growth)
contains the first 20% of stocks with the lowest book to market value. The next 20% of stocks are
assigned to portfolio 2. The process continues with the last portfolio (value) containing the last 20%
of stocks who on average have the highest book to market value.
Independently, the sample is ranked by market capitalisation at year end and assigned to one of five
size groups, where each portfolio contains an equal number of stocks. Portfolio 1 (big) contains the
first 20% of stocks and contains the stocks with the largest market capitalisation. Portfolio 2 contains
the next 20% of stocks. We continue this process with portfolio 5 (small) containing the last 20% of2These are similar to REIT’s in the USA
12
stocks with the lowest market capitalisation. Thus each stock is assigned to one size portfolio and to
one book to market portfolio, giving a total of 25 size-book to market portfolios. Each portfolio is
then held for the next twelve months and equal-weighted and value-weighted returns are calculated.
At the end of the holding period the procedure is repeated. As a result of the above procedure we
have a series of 300 monthly returns covering the period January 1982 to December 2006 for the 25
size-book to market portfolios.
Table 2
Characteristics of Portfolios
This table presents the average number of companies, mean and median market capitalisation and
mean book to market values of the 25 size-book to market portfolios. The 25 size-book to market
portfolios are formed by first ranking all stocks in the sample by their book to market values and
assigning each stock to one of five book to market portfolio where each portfolio contains an equal
number of stocks. Independently all stocks are ranked by their market capitalisation and assigned to
one of five size portfolio with each portfolio containing an equal number of stocks. The intersection of
the five book to market portfolios and the five size portfolio leads to the creation of our 25 portfolios.
Panel A: Number of Companies
Growth 2 3 4 Value
Big 52 53 43 25 10
2 39 42 41 38 23
3 33 35 39 40 38
4 30 31 33 42 48
Small 29 23 29 39 65
Panel B: Mean Market Capitilisation ($ millions)
Growth 2 3 4 Value
Big 1,894.6 2,095.8 1,516.6 1,055.7 1,358.1
2 69.5 69.3 67.0 69.0 65.5
3 19.7 19.8 19.4 19.3 18.8
4 7.6 7.7 7.5 7.5 7.4
Small 2.6 2.8 2.7 2.7 2.5
Panel C: Median Market Capitilisation ($ millions)
Growth 2 3 4 Value
Big 1,640.2 1,811.0 1,396.5 782.5 713.3
2 67.3 77.7 69.5 74.6 58.9
3 20.4 20.6 19.5 21.5 20.1
13
Table 2
(continued)
4 7.8 7.8 7.8 7.9 7.4
Small 2.4 2.7 2.7 2.7 2.6
Panel D: Mean Book to Market Values
Growth 2 3 4 Value
Big 0.31 0.60 0.88 1.27 2.98
2 0.30 0.60 0.89 1.27 7.26
3 0.28 0.60 0.90 1.31 3.71
4 0.28 0.61 0.89 1.30 2.99
Small 0.26 0.60 0.90 1.31 4.94
Table 2 provides summary information for each of the 25 size-book to market portfolios. Panel A
reports the average number of companies within each of the 25 size-book to market portfolios. These
results clearly demonstrate that growth stocks are over-represented in the big quintile with 28% of
stocks classified as growth also being classified as big. As we move down the growth quintile this
percentage declines with only 16% of growth stocks being classified as small. In contrast value stocks
are under-represented in the big quintile with 6% of stocks classified as value being big, while 35%
of stocks classified as value are also classified as small. This result, which suggests that value stocks
are on average smaller than growth stocks, is confirmed by Panel B and Panel C which report the
mean and median market capitalisation respectively. These Panels indicate that the mean (median)
market capitalisation of growth stock within the big quintile is $1,894.6 million ($1,640.2 million),
while the value stocks have a substantially lower market capitalisation of $1,358.1 million ($713.3
million). On average, value companies are smaller than growth companies which is similar to prior
Australian and overseas evidence (Fama and French, 1993; Gaunt, 2004; Halliwell, Heaney, and
Sawicki, 1999). It is also consistent with the argument that value stocks are under stress and are in
industries that are in decline, causing a fall in their market capitalisation, while growth companies
have been growing rapidly causing market capitalisation to increase.
Panel D of Table 2 reports the average book to market value of each portfolio. The result demonstrate
that the average book to market value is fairly consistent across the five size quintiles within each
book to market quintile, i.e. the growth quintile has an average book to market ratio of 0.29, with
big growth having the highest average of 0.31 and small growth having the lowest at 0.26. The
only exception is the value classification where the average is 4.38, with big-value having the lowest
average of 2.98 and portfolio 2-value having the highest average of 7.26. This higher variability is
being driven by a few companies with extremely large book to market values.3 If these stocks are3All companies with extremely large book to market values are being re-checked to confirm their book and market
14
removed from the calculations the variability is substantially reduced with the average within the
value classification falling to 3.27 and the largest average occurring in small-value with 3.76, and
the lowest being 2-value with 2.90. Overall the portfolio construction procedure has achieved its
objective of controlling for unwanted intra-quintile variability.
We now turn to the performance of the 25 size-book to market portfolios, and five value minus
growth (VMG) portfolios within each size quintile. The VMG portfolios are formed as the difference
between the return of the value portfolio less the return on the growth portfolio within each size
quintile. Table 3 reports the average monthly return of the size-book to market portfolios. Panel
A reports the equal-weighted portfolio returns while Panel B reports the value-weighted portfolio
returns.
The results indicate that the growth portfolio within each size quintile earns the lowest return. As
we move towards the value portfolio returns steadily increase with the value portfolio earning the
highest return. The F -test and Kruskal-Wallis test both indicate that there are significant differences
across the portfolio returns within each size portfolios, with the exception being the small quintile,
where there is evidence of no significant difference between the mean and median monthly returns.
Focusing on the VMG portfolio, the results clearly demonstrate a premium to value firms across each
size quintile with the average monthly return of the five VMG portfolios being 1.3%. This result
is consistent for both equal and value-weighted returns and indicates that size has effectively been
neutralized across the portfolios within each size quintile. The highest value premium is observed
in size quintile three with the lowest occurring in the small quintile. This strong premium to value
stocks is consistent with previous Australian studies of Gaunt (2004); Gharghori, Chan, and Faff
(2006) and is larger than overseas evidence, where the premium is usually around 0.55% in the US
(Fama and French, 1993, 1996) when portfolio and returns are calculated in a similar way.
Table 3 also reports the standard deviation of returns for each of the portfolios. These results
indicate that the five different book to market portfolios have similar volatility in returns within the
big quintile. As we move down the the size quintiles the results start to change with the growth
portfolio having a higher volatility compared to the value portfolio. There is also strong evidence
of higher volatility in returns as size declines. Among the various arguments for this finding is the
observation that small stocks typically have a lower price per share. This implies that small stocks
are more likely to display higher volatility because a small change in price leads to a larger percentage
change.
values.
15
Tab
le3
Ret
urn
sto
25si
ze-b
ook
tom
arke
tpor
tfol
ios
Pre
sent
edar
eth
em
onth
lym
ean,
F-s
tati
stic
,Kru
skal
-Wal
lisst
atis
tic
and
stan
dard
devi
atio
nof
mon
thly
equa
land
valu
ew
eigh
ted
retu
rns
of
each
ofth
e25
size
book
tom
arke
tpo
rtfo
lioan
dfiv
eva
lue
min
usgr
owth
(VM
G)
port
folio
sdu
ring
the
peri
od19
82to
2006
whe
npo
rtfo
lios
are
refo
rmed
inD
ecem
ber
each
year
.T
heV
MG
port
folio
sar
efo
rmed
asth
edi
ffere
nce
betw
een
the
retu
rnof
the
valu
epo
rtfo
liole
ssth
e
retu
rnon
the
grow
thpo
rtfo
liow
ithi
nea
chsi
zequ
inti
le.
**an
d*
deno
tesi
gnifi
canc
eat
the
1%an
d5%
leve
lsre
spec
tive
ly.
Pan
elA
:Equal
-Wei
ghte
dR
eturn
s
Mea
nM
onth
lyRet
urns
(%)
Gro
wth
23
4V
alue
VM
GF
-sta
tist
icK
rusk
al-W
allis
Big
-0.2
340.
644
0.79
90.
777
1.18
11.
415
-4.7
2**
-3.2
5**
2-0
.954
0.09
90.
152
0.26
40.
261
1.21
5-3
.75
**-2
.54
*
3-2
.252
-0.6
78-0
.314
0.00
40.
030
2.28
2-7
.69
**-4
.33
**
4-1
.361
-0.3
44-0
.558
-0.0
53-0
.360
1.00
0-2
.46
*-1
.56
Smal
l0.
290
0.44
01.
015
0.69
30.
946
0.65
6-1
.83
-1.1
4
F-s
tati
stic
3.66
**
Kru
skal
-Wal
lis10
2.71
**
Stan
dard
Dev
iation
Gro
wth
23
4V
alue
VM
G
Big
6.48
55.
138
4.78
45.
217
5.89
15.
194
26.
828
5.20
05.
181
4.84
16.
497
5.61
5
37.
914
6.59
65.
701
5.13
96.
268
5.13
7
410
.341
8.38
27.
375
6.65
97.
246
7.04
4
Smal
l10
.420
10.1
549.
140
8.06
77.
800
6.21
1
16
Tab
le3
(con
tinu
ed)
Pan
elB
:V
alue-
Wei
ghte
dR
eturn
s
Mea
nM
onth
lyRet
urns
(%)
Gro
wth
23
4V
alue
VM
GF
-sta
tist
icK
rusk
al-W
allis
Big
0.22
40.
653
0.83
40.
913
1.32
11.
097
-2.9
3**
-2.4
7*
2-0
.877
0.10
50.
171
0.29
30.
352
1.23
0-3
.94
**-2
.67
**
3-2
.229
-0.6
26-0
.282
0.07
30.
049
2.27
8-7
.74
**-4
.53
**
4-1
.432
-0.3
04-0
.545
-0.1
13-0
.366
1.06
6-2
.63
**-1
.81
Smal
l-0
.228
0.08
80.
723
0.34
50.
569
0.79
7-2
.09
*-1
.38
F-s
tati
stic
3.39
**
Kru
skal
-Wal
lis10
6.71
**
Stan
dard
Dev
iation
Gro
wth
23
4V
alue
VM
G
Big
5.62
75.
535
5.24
76.
102
6.83
06.
495
26.
628
5.22
45.
165
4.77
86.
070
5.40
8
37.
809
6.33
35.
441
4.95
36.
164
5.09
9
410
.103
8.18
57.
210
6.40
57.
231
7.01
3
Smal
l10
.657
10.3
619.
218
8.16
47.
856
6.61
9
17
3.3 Testing Framework
The CAPM can be expressed as:
E(ri) = βi[E(rm)], (1)
while Fama and French (1993) proposes the following three-factor model;
E(ri) = βi[E(rm)] + siE(SMB) + hiE(HML). (2)
Where E(ri) is the expected excess return on asset i, E(rm) is the expected excess return on the
market portfolio, E(SMB) is the expected return on the mimicking portfolio for the size factor and
E(HML) is the expected return on the mimicking portfolio for the book to market factor. These
models can be converted into their empirical counterpart where expected returns are replaced by ex
post versions of the market portfolios and Fama and French factors which takes the form;
rit = αi + βirmt + εit i = 1, · · · , N, (3)
and
rit = αi + βirmt + siSMBt + hiHMLt + eit i = 1, · · · , N. (4)
Where rit is the excess return on asset i in time t. rmt is the excess return of the market. SMBt
is the return on the mimicking size portfolio and HMLt is the return on the mimicking book to
market portfolio. αi, βi, si and hi are regression coefficients, εit and eit are the error terms and N
is the number of test assets. Equation 3 indicates that the CAPM is a restricted three-factor model
where the restrictions are si = hi = 0.
While there are a number of different ways to test the CAPM and the Fama-French three factor
model this study will use the generalised method of moments (GMM) technique. This method has
several advantages; first, it relaxes the assumption that excess returns are independent and identically
distributed (i.i.d.) normal. Second, it allows all asset parameters to be estimated simultaneously.
Third, we can derive a statistic to test that all the pricing errors are jointly equal to zero, which
is equivalent to the Gibbons, Ross, and Shanken (1989) (GRS ) statistic but allows the errors to be
cross-correlated, autocorrelated and heteroskedastic. Fourth, we can utilise the D-statistic of Newey
and West (1987) to test whether the restriction si = hi = 0 is true.4
In the case of the CAPM empirical model (equation 3) there are 2N sample moment conditions.
First, that the mean regression error term is equal to zero,
E[εit] = 0 ∀i = 1, · · · , N.
Second, that the regression error is orthogonal to the market return,
E[εitrmt] = 0 ∀i = 1, · · · , N.
4For further details on the GMM technique and its uses in testing asset pricing models see Cochrane (2005).
18
As the system has 2N unknown parameters the system is just-identified and the estimated parameters
are equivalent to their ordinary least square (OLS) counterparts.
Similarly, in the Fama and French three factor model (equation 4) there are 4N sample moment
conditions. The first 2N are identical to the CAPM moment conditions. The final two moment
conditions are that the SMB and HML factors are orthogonal to the regression error term:
E[eitSMBt] = 0 ∀i = 1, · · · , N
E[eitHMLt] = 0 ∀i = 1, · · · , N.
As the CAPM is a restricted three-factor model with the restriction si = hi = 0, this restriction can
be tested by forming the following D-statistic which has a χ2 distribution;
TgT (φ̂r)′S−1gT (φ̂r)− TgT (φ̂u)′S−1gT (φ̂u) ∼ χ24N−2N . (5)
Where gT (φ̂r) and gT (φ̂u) are the empirical moment condition vectors for the restricted and unre-
stricted model respectively, S−1 is the optimal weighting matrix and T is the number of observations.
A complete model of excess returns should also result in the pricing errors to be equal to zero. This
can be tested using
α̂′var(α̂)−1α̂ ∼ χ2N , (6)
where α̂ is the estimated intercept and var(α̂)−1 is the variance covariance matrix of the estimated
intercept terms.5
4 Construction and Attributes of the Fama and French Factors
An essential component of testing and analysing the Fama and French three-factor model is the
construction of the SMB and HML factors. The factor SMB captures the premium that small stocks
earn over large stocks, while HML captures the premium that value stocks earn over growth stocks.
It is still debatable what these premiums are capturing, with the two major arguments being that
they are proxies for underlying economic risk (Fama and French, 1995, 1996, 2006, 2007), or that they
are capturing a behavioural bias in investors (Daniel and Titman, 1997; Daniel, Titman, and Wei,
2001; Durand, Juricev, and Smith, 2007; Lakonishok, Shleifer, and Vishny, 1994). Whatever they
are capturing, the size and value premium are proxies for economic risk or behavioural bias and the
fluctuations in the premiums allows us to better understand how assets are priced. As an example,
when HML increases investors will demand a higher return on assets with a strong weighting on the
HML factor. As structures, breadth and depth of markets vary around the world, manifestations
of the size and value premium could also occur in different ways. Hence, when creating the factors5For full details of this test statistic see Cochrane (2005) pp 231-235.
19
Table 4
Portfolios used in the formation of SMB and HML in the USA market
This table demonstrates the interaction of the two size portfolios and three book to market portfolio
lead to the formation of six size book to market portfolios
Book to Market portfolio
30% 40% 30%
50% big big big
Size growth middle value
portfolio 50% small small small
growth middle value
we need to take into account these differences and tailor the factors accordingly. We also need to
remember that fluctuations in these factors rather than the absolute value of the factors are the
major determinant in explaining the cross-section of asset returns.
Fama and French (1993) propose that SMB and HML be formed by ranking all stocks on the New
York Stock Exchange (NYSE) by market capitalisation and assigning all stocks (including those listed
on the NASDAQ stock market and American Stock Exchange (AMEX)) with a market capitalisation
higher than the median NYSE market capitalisation into the big portfolio and all other stocks into
a small portfolio. Independently all NYSE stocks are ranked on book to market values. Any stock
containing a negative book to market value are excluded. The first 30% of stocks with the lowest
book to market ratios are then assigned to the growth portfolio. The next 40% of stocks based on
book to market value are assigned to the middle portfolio and finally the last 30% of stocks are
assigned to the value portfolio. This leads to all stocks being assigned to one of two size portfolio
and one of three book to market portfolio giving a total of six portfolios (see Table 4 for details).
These portfolio are held for twelve months and returns are value-weighted. After twelve months the
process is repeated. The SMB factor is then formed by calculating the average return of the three
small portfolios (small growth, small middle and small value) and subtracting the average of the
three big portfolios (big growth, big middle and big value). Similarly the HML factor is formed by
calculating the average return of the two value portfolios (big value and small value) and subtracting
the average return of the two growth portfolios (big growth and small growth). This methodology has
subsequently been used in countless studies including Campbell (1996); Campbell and Vuolteenaho
(2004); Carhart (1997); Chan, Jegadeesh, and Lakonishok (1995, 1996); Conrad and Kaul (1998);
Daniel and Titman (1997); Daniel, Titman, and Wei (2001); Davies (1994); Fama and French (1995,
1996, 1998, 2006, 2007); Jagannathan and Wang (1996); Jegadeesh and Titman (2002); Kim (1995,
1997); LaPorta, Lakonishok, Shleifer, and Vishny (1997); Lettau and Ludvigson (2001) and Zhang
(2005)
20
To determine whether this is an ideal method for capturing the size and value premium in Australia
we need to understand where the size and value premium occurs in Australia. Table 5 reports the
mean return, market capitalisation and book to market value of portfolios formed on size and book
to market values in Australia. The ten size portfolios are formed by ranking all stocks by market
capitalisation at the end of December each year and assigning an equal number of companies into one
of ten size portfolio. Portfolio 1 contains the 10% of stocks with the largest market capitalisation,
portfolio 2 contains the next 10% of stocks (i.e. those ranked between 10% and 20%) as determined
by market capitalistion. This process continues for all the portfolios, with portfolio 10 containing
the 10% of all stocks with the smallest market capitalisation. A similar process is followed for book
to market portfolios. First all stocks are ranked by their book to market value as at December each
year. Then each stock is assigned to one of ten book to market portfolio with the first 10% of stocks
with the lowest book to market value being assigned to portfolio 1, the next 10% of stocks are then
assigned to portfolio 2. This process continues for all the portfolios with portfolio 10 containing the
last 10% of stocks with the highest book to market values. The portfolios are then held for twelve
months and value-weighted logarithmic returns calculated each month. The process is repeated each
year resulting in a series of 300 monthly returns for ten size portfolios and ten book to market
portfolios covering the period January 1982 to December 2006.
Table 5 highlights a number of important issues that we must take into consideration when forming
factors that attempt to capture the premium to small and value companies. First, as we move
from the largest stocks to the smallest stock the returns initially decline, reaching a minimum for
portfolios 5 through 7 and then increases until we reach the smallest decile. If we take the difference
between the small and big portfolio there is a premium of approximately 0.52% per month. This
U-shape pattern in average returns is in contrast to the US market where there is a steady increase
in returns from the largest to smallest portfolio when portfolios are formed on size (Banz, 1981;
Fama and French, 1992; Reinganum, 1981). Second, returns to portfolios formed on book to market
ratios indicate that the returns for growth stocks are lowest and it increases as we move towards
value stocks. The only anomaly from this pattern is portfolio 10 which has a lower return than
portfolio 9. If we take the difference between portfolio 10 and portfolio 1 we find a value premium
of 0.73% per month. This result is more consistent with previous evidence from the US (Barber and
Lyon, 1997; Fama and French, 1992). Third, there is evidence that the size and value premiums may
be correlated with the smallest stocks on average having a higher book to market value. This can
be seen with size portfolio 10 having the highest book to market value of 3.1 and book to market
portfolio 10 which has the lowest market capitalisation of $62.2 million. This is consistent with
overseas evidence that small companies generally have higher book to market values compared to
big companies (Fama and French, 1992).
Results from Table 5 indicate that although small stocks earn a premium, this premium is not
linear. This would indicate that if we formed the factors using the original Fama and French (1993)
21
Table 5
10 size and book to market portfolios
The table reports the monthly mean return, market capitalisation and book to market value of ten
size and ten book to market portfolios in Australia for the period January 1982 to December 2006.
The size portfolios are formed by ranking all stocks by market capitalisation and assigning each
stock to one of ten size portfolio with each portfolio containing an equal number of stocks. Portfolio
1 contains the first 10% of stocks with the biggest market capitalisation, while portfolio 10 contains
the last 10% of stocks with the smallest market capitalisation. Similarly, all stocks are ranked by
their book to market value and assigned to one of ten book to market portfolio. Portfolio 1 contains
the first 10% of stocks with the lowest book to market values, while portfolio 10 contains the last
10% of stocks which have the highest book to market values.
Size Portfolios Book to Market Portfolios
Portfolio Returns Market Book to Returns Market Book to
Capitalisation Market Capitalisation Market
($ millions) Value ($ millions) Value
1 0.738 3,144.1 0.722 -0.305 437.0 0.189
2 0.362 262.0 0.851 0.394 709.7 0.390
3 0.010 88.8 0.961 0.528 718.2 0.534
4 -0.106 42.2 1.843 0.716 530.6 0.669
5 -0.664 23.4 1.460 0.839 463.2 0.812
6 -0.569 14.0 1.375 0.730 314.8 0.968
7 -0.619 8.9 1.279 0.852 207.4 1.161
8 -0.497 5.6 1.510 0.726 145.2 1.428
9 -0.011 3.4 1.634 1.101 88.4 1.870
10 1.256 1.6 3.093 0.424 62.2 6.524
22
methodology we may not capture the size premium. Whether this affects our results will depend on
how the factor fluctuates. If it fluctuates in a similar way to the size premium and the underlying
economic phenomenon then even though the premium is not on average the same, we will still
be able to better understand how the cross-section of assets are priced. Taking into consideration
the underlying regularities in the Australian market three different methods are used to form the
factors. First, we follow the methodology of Fama and French (1993) except that breakpoints for
portfolio allocation is determine by the Australian market and portfolios are formed in December.
The factors formed using this methodology are called SMB 2x3 and HML 2x3. Second, following
previous Australian studies and some overseas studies (Durack, Durand, and Maller, 2004; Durand,
Limkriangkrai, and Smith, 2006a; Liew and Vassalou, 2000) we form SMB and HML in the following
manner. Each December we rank all stock by their market capitalisation and assign them to one of
three size portfolios. The first portfolio (big) contains the first 30% of stocks and contains the biggest
stocks. The middle portfolio contains the next 40% of stocks ranked by market capitalisation. Finally
the small portfolio contains the last 30% of stock and contains the smallest stocks. Independently
all stocks are ranked by their book to market value and assigned to one of three book to market
portfolios. The first 30% of stocks are assigned to the growth portfolio which contains stocks with
the lowest book to market values. The middle portfolio contains the next 40% of stocks ranked
by their book to market value. The last 30% of stocks are assigned to the value portfolio which
contains the stocks with the highest book to market value. This leads to the formation of nine
portfolios as outline in Panel A of Table 6. Each portfolio is then held for 12 months and value-
weighted logarithmic returns are calculated. At the end of 12 months the process is repeated. SMB
3x3 is formed by taking the average of the three small portfolios (small growth, small middle and
small value) and subtracting the average return of the three big portfolios (big growth, big middle
and big value) each month. Similarly HML 3x3 is formed as the average return of the three value
portfolios (large value, middle value and small value) less the average of the three growth portfolios
(large growth, middle growth and small growth). Third, in December we rank all stocks by market
capitalisation and assign each stock to one of five size portfolio. The big portfolio contains the first
20% of stocks with the largest market capitalisation. Portfolio 2 contains the next 20% of stocks by
market capitalisation. This process continues with the last portfolio (small) containing the last 20%
of stocks which have the smallest market capitalisation. As with the previous two methodologies
we independently assign each stock to one of three book to market portfolio. As with the previous
methodologies the 30% of stocks with the lowest book to market value are assigned to the growth
portfolio, the next 40% of stocks are assigned to the middle portfolio and the final 30% of stocks are
assigned to the value portfolio. This leads to each stock being assigned to one of fifteen portfolios
as outlined in Panel B of Table 6. Each portfolio is then held for 12 months and value-weighted
logarithmic returns are calculated. At the end of 12 months the process is repeated. SMB 5x3 is
formed by taking the average of the three small portfolios (small growth, small middle and small
value) and subtracting the average return of the three big portfolios (big growth, big middle and big
23
value) each month. Similarly HML 5x3 is formed as the average return of the five value portfolios
(large value, 2 value, 3 value, 4 value and small value) less the average of the five growth portfolios
(large growth, 2 growth, 3 growth, 4 growth and small growth).
Panel A of Table 7 reports the average monthly returns of the 3 different methods of constructing
the factors in Australia. We also construct one further SMB factor called SMB 10. SMB 10 is
formed by ranking all stocks each December by market capitalisation and assigning each stock into
one of ten size portfolio where each portfolio contains an equal number of stocks. Portfolio 1 (big)
contains the 10% of stocks with the largest market capitalisation and portfolio 10 (small) contains
the 10% of all stocks with the smallest market capitalisation. SMB 10 is then calculated as the
value-weighted logarithmic return of the small portfolio less the value-weighted logarithmic return of
the big portfolio. We also report the average monthly return to the market risk premium in Australia
(MRP) which is calculated as the value-weighted monthly market return less the 13-week treasury
note yield with both extracted from the AGSM-CRIF price relative file, the US market risk premium
(USMRP), the US SMB factor (USSMB) and the US HML factor (USHML). The US market risk
premium and factors are obtained directly from the data library in Ken French’s webpage.6 Panel
B of Table 7 also reports the correlation coefficients between each of the factors.
6The address of the website is http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data library.html.
24
Table 6
Alternative Methods to forming SMB and HML
This table demonstrates the alternative methods of assigning stocks to either one of nine size book
to market portfolios (Panel A) or one of fifteen size book to market portfolios (Panel B). These
assignments are used to form various SMB and HML factors
Panel A: 3x3 portfolio formation
Book to Market portfolio
30% 40% 30%
30% big big big
growth middle value
Size 40% middle middle middle
portfolio growth middle value
30% small small small
growth middle value
Panel B: 5x3 portfolio formation
Book to Market portfolio
30% 40% 30%
20% big big big
growth middle value
20% 2 2 2
growth middle value
Size 20% 3 3 3
portfolio growth middle value
20% 4 4 4
growth middle value
20% small small small
growth middle value
25
Tab
le7
Ret
urn
sto
Fac
tors
Pan
elA
pres
ents
the
mea
n,m
edia
n,st
anda
rdde
viat
ion
t-st
atis
tic
and
z-va
lue
ofm
onth
lyva
lue-
wei
ghte
dre
turn
sof
diffe
rent
HM
Lan
dSM
B
fact
ors
for
the
peri
od19
82to
2006
.T
heco
rrel
atio
nco
effici
ents
betw
een
each
fact
oris
repo
rtin
Pan
elB
.SM
B2x
3an
dH
ML
2x3
was
form
ed
follo
win
gth
em
etho
dof
Fam
aan
dFr
ench
(199
3).
SMB
3x3
and
HM
L3x
3is
form
edby
assi
gnin
gea
chas
set
into
one
ofth
ree
size
port
folio
and
one
ofth
ree
book
tom
arke
tpo
rtfo
lio.
SMB
3x3
isth
enfo
rmed
asth
eav
erag
ere
turn
ofth
eth
ree
smal
lpor
tfol
ios
less
the
aver
age
retu
rn
ofth
eth
ree
big
port
folio
s.H
ML
3x3
isth
enfo
rmed
asth
eav
erag
ere
turn
ofth
eth
ree
valu
epo
rtfo
lios
less
the
aver
age
retu
rnof
the
thre
e
grow
thpo
rtfo
lios.
SMB
5x3
and
HM
L5x
3is
form
edby
assi
gnin
gea
chas
set
into
one
offiv
esi
zepo
rtfo
lioan
don
eof
thre
ebo
okto
mar
ket
port
folio
.SM
B5x
3is
then
form
edas
the
aver
age
retu
rnof
the
thre
esm
allp
ortf
olio
sle
ssth
eav
erag
ere
turn
ofth
eth
ree
big
port
folio
s.H
ML
5x3
isfo
rmed
asth
eav
erag
ere
turn
ofth
efiv
eva
lue
port
folio
sle
ssth
eav
erag
eof
the
five
grow
thpo
rtfo
lios.
SMB
10is
form
edby
rank
ing
alls
tock
sby
mar
ket
capi
talis
atio
nan
das
sign
ing
each
stoc
kto
one
ofte
nsi
zepo
rtfo
lios
whe
reea
chpo
rtfo
lioha
san
equa
lnum
ber
ofst
ocks
.
The
big
port
folio
cont
ains
the
10%
ofst
ocks
wit
hth
ela
rges
tm
arke
tca
pita
lisat
ion,
whi
leth
esm
allp
ortf
olio
cont
ains
the
10%
ofst
ocks
wit
h
the
low
est
mar
ket
capi
talis
atio
n.SM
B10
isfo
rmed
asth
ere
turn
ofth
esm
allpo
rtfo
liole
ssth
ere
turn
ofth
ebi
gpo
rtfo
lio.
We
also
repo
rt
the
aver
age
retu
rnof
the
mar
ket
risk
prem
ium
(MR
P)
inA
ustr
alia
,whi
chis
calc
ulat
edas
the
valu
e-w
eigh
ted
mon
thly
mar
ket
retu
rnle
ssth
e
13-w
eek
trea
sury
note
yiel
dw
ith
both
bein
gex
trac
ted
from
the
AG
SM-C
RIF
pric
ere
lati
vefil
ean
dth
eU
Sm
arke
tri
skpr
emiu
m(U
SMR
P),
US
HM
L(U
SHM
L)
and
US
SMB
(USS
MB
)w
hich
was
sour
ced
from
Ken
Fren
ch’s
web
page
.
**an
d*
deno
tesi
gnifi
canc
eat
the
1%an
d5%
leve
lsre
spec
tive
ly.
Pan
elA
:Ret
urns
Mea
nM
edia
nSt
anda
rdt-
stat
isti
cz-
valu
e
Dev
iati
on
HM
L2x
30.
916
0.85
23.
362
4.72
**-4
.94
**
HM
L3x
30.
867
0.97
53.
203
4.69
**-4
.93
**
HM
L5x
30.
959
0.66
73.
044
5.46
**-5
.41
**
SMB
2x3
-1.2
33-1
.224
5.12
7-4
.16
**-4
.42
**
SMB
3x3
-0.9
10-1
.235
6.67
7-2
.36
*-3
.14
**
SMB
5x3
-0.4
86-1
.066
7.40
3-1
.14
-1.9
8*
26
Tab
le7
(con
tinu
ed)
SMB
100.
518
-0.0
957.
751
1.16
-0.5
3
MR
P0.
392
0.96
64.
907
1.38
-2.9
5**
USM
RP
0.67
81.
045
4.34
32.
70**
-3.4
4**
USS
MB
0.08
6-0
.125
3.27
90.
46-0
.02
USH
ML
0.46
20.
425
3.11
02.
57*
-2.6
9**
Pan
elB:C
orre
lation
coeffi
cien
t
HM
L2x
3H
ML
3x3
HM
L5x
3SM
B2x
3SM
B3x
3SM
B5x
3SM
B10
MR
PU
SMR
PU
SSM
BU
SHM
L
HM
L2x
31.
000
HM
L3x
30.
892*
*1.
000
HM
L5x
30.
862*
*0.
927*
*1.
000
SMB
2x3
-0.3
20**
-0.4
07**
-0.4
62**
1.00
0
SMB
3x3
-0.3
13**
-0.4
31**
-0.4
85**
0.96
5**
1.00
0
SMB
5x3
-0.2
92**
-0.3
95**
-0.4
66**
0.94
3**
0.97
7**
1.00
0
SMB
10-0
.128
*-0
.251
**-0
.326
**0.
885*
*0.
911*
*0.
922*
*1.
000
MR
P-0
.209
**-0
.182
**-0
.266
**-0
.083
-0.0
69-0
.086
-0.1
21*
1.00
0
USM
RP
-0.1
36*
-0.1
46*
-0.1
80**
-0.0
66-0
.053
-0.0
66-0
.069
0.54
0**
1.00
0
USS
MB
-0.0
79-0
.084
-0.1
29*
0.06
80.
058
0.04
30.
058
0.23
7**
0.19
4**
1.00
0
USH
ML
0.17
9**
0.19
2**
0.19
9**
0.02
2-0
.003
0.02
30.
052
-0.1
78**
-0.5
06**
-0.4
28**
1.00
0
27
Starting with the HML factors we see that the three methodologies produce similar average monthly
returns with the highest being HML 5x3 with 0.959% per month and the lowest being HML 3x3 with
0.867% per month. This average monthly return is statistically different from zero with the average
t-statistic being 4.96 and is similar to the return reported in Gharghori, Chan, and Faff (2006)
and citetHalliwelletal:1999:ARJ. The average monthly return for the USHML factor is significantly
lower with a return of 0.462% per month. The extremely large HML factor indicates that the value
premium is not only statistically significant but economically significant in the Australia equities
market. The results also indicate that the standard deviation of returns of the three Australian
HML factors and USHML are very similar with the Australian factors having an average of 3.203%
and USHML having a standard deviation of 3.110%. Turning to the correlation coefficients, the
results demonstrate that the three Australian HML factors are highly correlated with an average
correlation coefficient of 0.894. Surprisingly the correlation between the USHML and the Australian
HML factors are quite low with an average correlation of 0.190 although it is still statistically
significant. This result would indicate that to properly test the Fama and French (1993) model
in Australia that factors formed on Australia information is essential. The results taken together
suggest that the three methods of forming the HML factor in Australia produce very similar results
and our choice of methodology will have little impact on the results. To further examine whether
the time-series behaviour of the different formation procedures are similar we calculate the average
monthly return during each year and the average monthly return for each month of the year for the
three HML factors. These results are reported in Figure 2 and 1 respectively. This will also allow
us to check for market seasonalities in the HML factor.
Figure 1 plots the average return each month and indicate that the three formation methodologies
lead to similar result with all three portfolios following a similar trend. Figure 1 also suggest that
there is no evidence of a monthly seasonality in the HML factor. Figure 2 demonstrates that the
three methodologies lead to similar outcomes with all three portfolios having similar average monthly
returns each year. The results demonstrate one clear anomaly during the 25 year period, 1989. In
1989 there is clear evidence of severe underperformance by value stocks with the HML 2x3 factor
earning an average monthly return of -4.98% during the year. HML 3x3 and HML 5x3 also have
extremely large loses at an average of -3.41% and -2.73% per month respectively. We investigated
these extreme negative returns further and two important features surfaced. First, the average book
to market ratio in the value portfolios were substantially higher at the start of 1989 than during the
rest of the sample. For example, the average book to market value for large value and small value
in the 2x3 portfolio formation procedure in December 1988 is 6.57 against an average of 3.36 during
the full period studied. This result suggests that value stocks were under extreme financial distress
leading into 1989. This is further confirmed by the fact that the percentage of companies delisting
during the year nearly doubles. Second, the extreme negative result is being driven by a few large
value companies suffering extremely large loses during the year. This includes one company suffering
28
-2.0
-1.0
0.0
1.0
2.0
January February March April May June July August September October November December
Month
Av
era
ge
Re
turn
(%
)
HML 2x3
HML 3x3
HML 5x3
Figure 1
Average Monthly Return for various HML factors
monthly returns of -101.2%, -86.0%, -58.0% and -32.5%, which had a value-weight of 5.55%.7 While
another company with a value-weight of 17.74% had monthly returns of -64.19%, -32.85%, -19.63%
and -12.72%. The extreme negative return being driven by a few large companies is confirmed
by calculating equal-weighted logarithmic returns for 1989 and the average return for HML 2x3
increases to -1.91% HML 3x3 to -1.95% HML 5x3 to -1.78%. This result suggests that value stocks
did particularly badly in 1989 and were in severe distress leading into to 1989, but the extremely
negative number is being driven by a few companies. As such to test the robustness of our results we
should include a 1989 dummy variable in the regressions. Overall, the results are supportive of the
conclusion that the three methods produce similar time-series behaviour in the HML factor indicate
the exact formation technique will be unimportant.
Turning to the SMB factors. The first thing that we have found is that the average monthly return
for the three methodologies is negative. The SMB 2x3 factor has the lowest average monthly return
of -1.23%, which is statistically significant, while SMB 5x3 has the highest at -0.486% which is
statistically insignificant. In contrast the SMB 10 factor has a positive average monthly return of
0.518%, although it is statistically insignificant. These results are not surprising given the results
reported in Table 5, which shows a non-linear relationship between size and average returns in the7As could be expected this company was delisted during the year of 1989
29
-5.0
-4.0
-3.0
-2.0
-1.0
0.0
1.0
2.0
3.0
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
Date
Av
era
ge
Mo
nth
ly R
etu
rn (
%)
HML 2x3 HML 3x3
HML 5x3
Figure 2
Average Monthly Return for various HML factors each year
Australian equity market. It is interesting that the USSMB also has a very low average monthly
return of 0.086% which is insignificantly different from zero. These results support recent evidence
that the size premium has been declining in equity markets in recent years (Barber and Lyon,
1997; Dimson and Paul, 1999), with the only evidence of the size premium in Australia being
extremely small stocks that have extremely low liquidity. The SMB factors in Australia also have a
standard deviation that is approximately double the HML factors and 37% larger than the market
risk premium. As expected, the standard deviation of SMB 10 and SMB 5x3 are the highest with
7.75% and 7.4% respectively, while SMB 2x3 has the lowest standard deviation of 5.13%. Among
the various arguments for this finding is the observation that small stocks typically have a lower
price per share. This implies that small stocks are more likely to display higher volatility because
a small change in price leads to a larger percentage change. This argument is supported by our
results which demonstrate that as we increase the number of divisions in the data, higher standard
deviations occur. Turning to the correlation coefficients, the results indicate that the four Australian
SMB factors are highly correlated with each other with an average correlation coefficient of 0.934. As
with the HML factor, the Australian SMB factors have a very low correlation with the USSMB with
an average correlation coefficient of 0.057. This result confirms that factors formed on Australian
information is essential for testing the Fama and French (1993) model.
30
-5.0
-3.0
-1.0
1.0
3.0
5.0
7.0
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
Date
Avera
ge M
on
thly
Retu
rn (
%)
SMB 2x3 SMB 3x3
SMB 5x3 SMB 10
Figure 3
Average Monthly Return for various SMB factors each year
The high correlation coefficients between the four different ways of forming the SMB factor in
Australia suggest that the choice of formation methodology may not influence our results when we
test the Fama and French (1993) model. This is because the fluctuations, rather than the average
monthly return of the factor, is the major determinant in its role in explaining the cross-section
of asset returns. To further analyse this issue and to check for market seasonalities in the SMB
factor we calculate the average monthly return during each year and the average monthly return
for each month of the year for the four SMB factors. These results are reported in Figure 3 and 4
respectively. Figure 3 again demonstrates the high correlation and volatility in returns for the SMB
factor in Australia. Although there are a number of years that have a large negative average return
(including 1988, 1997 and 2001), there are just as many years with very large positive average return
(including 1993, 1996, 1999 and 2003). We therefore conclude there is little evidence of any abnormal
year in the SMB factor. In contrast Figure 4 demonstrates strong seasonalities in the SMB factor
in Australia. The result demonstrates a large negative return in June which is followed by a large
positive return in July. This result is consistent with the tax-loss hypothesis explaining the small
firm effect (Brown, Keim, Kleidon, and Marsh, 1983; Keim, 1983). Unfortunately this can not be the
complete explanation because there is evidence of a strong positive return in January. These results
are consistent with earlier findings on the size premium in Australia (Brown, Keim, Kleidon, and
Marsh, 1983; Durand, Juricev, and Smith, 2007) and suggests that in our empirical test that a June,
31
-7.0
-5.0
-3.0
-1.0
1.0
3.0
5.0
7.0
January February March April May June July August September October November December
Month
Av
era
ge
Re
turn
(%
)
SMB 2x3
SMB 3x3
SMB 5x3
SMB 10
Figure 4
Average Monthly Return for various SMB factors
July and possibly a January dummy variable should be included in robustness checks. Overall, the
time-series behaviour of the four different ways of forming the SMB factor are similar and indicate
that the three methods will have similar explanatory power in explaining the cross-section of asset
returns. We are inclined to support the SMB 5x3 factor because it has the highest average monthly
return and is closest in capturing the size premium in Australia.
The results in Table 7 also indicate that the HML factors and SMB factors have a moderate corre-
lation between them with the average correlation being -0.356. This is similar, although lower, to
the correlation between the USSMB and the USHML which has a correlation of -0.428. This result
suggests that robustness checks that separate out the factors in the regressions may be required. It
also indicates a low correlation between the factors and the MRP in Australia with the correlation
between the Australian HML averaging -0.219 while the SMB and MRP has an average correlation
of -0.090. This suggests that the factors are proxies for economic risk or behaviourial bias which are
independent of the market risk premium.
Results in this section show that the Australian equity markets display different regularities than
their North American cousins. This suggests that the construction of the SMB and HML factor may
need to be tailored to take into account these differences. We propose three different methodologies
to form SMB and HML factors in Australia. The HML factors all demonstrate very similar average
32
monthly returns and are highly correlated with each other thus it would seem that it will not matter
how we form the HML factor in Australia. In contrast, the SMB factors differ substantially in their
average monthly returns, although they are all highly correlated with each other and demonstrate
similar time-series properties. We are inclined to support the SMB 5x3 factor because it has the
highest average monthly return and thus captures the size premium better than the other two
methodolgies. Given this information we believe the most appropriate factors to use in Australia
are the HML 5x3 and SMB 5x3. To test our assumptions we will also use the SMB 2x3 and HML
2x3 in our tests of the Fama and French (1993) model.
5 Results
The results from estimating equation 3 to explain portfolio excess returns are reported in Table
8. For brevity we only report the results using the value-weighted returns.8 As expected, Table 8
indicates that market risk tends to be higher for low book to market portfolios (growth) and decline
as we move toward higher book to market portfolios (value). This result is consistent with Fama and
French (1993) and Gaunt (2004); Halliwell, Heaney, and Sawicki (1999). Contrary to expectation,
the results also indicate that the small portfolios exhibit similar market risk to the big portfolios
and on average the reported β’s are less than one.
If the CAPM explained the returns of the 25 size-book to market portfolios the intercepts should
not be significantly different from zero. Results indicate that the majority of the intercept terms
are statistically different from zero. This result is supported by the GRS statistics being statisti-
cally significant, indicating we should reject the null that all intercepts are jointly equal to zero.
Surprisingly the results suggest little evidence of the size effect with the small portfolios generally
earning an adjusted return less than their equivalent big portfolio, although all the intercept terms
within the small quintile are insignificantly different from zero. This result needs to be considered
with caution because this appears to be a result of the low explanatory power of the model for the
portfolios within the small quintile. This is demonstrated in a number of ways, first, the adjusted R2
of the portfolios within the small quintile averages only 21% compared to the average of 68% for the
five portfolios within the big quintile. Second, the intercept terms are generally larger in absolute
terms for the portfolios within the small quintile compared to the portfolios within big quintile; but
the intercept terms are insignificant for the portfolios within the small quintile but significant for the
portfolios within the big quintile. This result indicates that the variability around the estimates for
the portfolios within the small quintile are substantially larger. It also reinforces the non-linear re-
lationship between size and returns in Australia with the intercept terms of the middle size quintiles
being substantially lower than any of the portfolios within the big or small quintiles.
8The equal-weighted return results are similar and can be obtained from the author upon request.
33
Tab
le8
Reg
ress
ions
resu
lts
from
the
CA
PM
The
tabl
epr
esen
tsth
ere
sult
sfr
omre
gres
sing
the
300
exce
ssm
onth
lyre
turn
sof
each
ofth
e25
size
-boo
kto
mar
ket
port
folio
son
exce
ss
mar
ket
retu
rns.
The
25si
ze-b
ook
tom
arke
tpo
rtfo
lios
are
form
edby
usin
gin
depe
nden
tso
rts
base
don
mar
ket
capi
talis
atio
nan
dbo
okto
mar
ket
valu
esdu
ring
the
peri
od19
82to
2006
.T
hefo
llow
ing
tim
e-se
ries
regr
essi
onis
esti
mat
ed;
r it=
αi+
βir
mt+
ε it
i=
1,···,
N.
Whe
rer i
tis
the
retu
rnon
port
folio
iin
mon
tht
less
the
13w
eek
trea
sury
note
yiel
dan
dr m
tis
the
valu
e-w
eigh
ted
mar
ket
mon
thly
retu
rn
less
the
13w
eek
trea
sury
note
yiel
d.B
oth
the
mar
ket
retu
rnan
dth
e13
wee
ktr
easu
ryno
tyi
eld
are
extr
acte
dfr
omth
eC
RIF
pric
ere
lati
ve
file.
The
syst
emis
esti
mat
edus
ing
the
GM
Mte
chni
que
wit
hth
efo
llow
ing
mom
ent
rest
rict
ions
E[ε
it]=
0,E
[εitr m
t]=
0∀i
=1,···,
N.
The
t-st
atis
tic
for
the
regr
essi
onco
effici
ents
uses
HA
Cst
anda
rder
rors
.T
head
just
edR
2ar
eca
lcul
ated
for
each
equa
tion
inth
esy
stem
.W
e
also
repo
rtth
eG
ibbo
ns,R
oss,
and
Shan
ken
(198
9)(G
RS
)te
stst
atis
tic
and
anad
just
edχ
2st
atis
tic
(see
equa
tion
6)th
atad
just
edth
eG
RS
stat
isti
csfo
rcr
oss-
corr
elat
eder
rors
.
**an
d*
deno
tesi
gnifi
canc
eat
the
1%an
d5%
leve
lsre
spec
tive
ly.
Coeffi
cien
tt-
stat
isti
c
αi
Gro
wth
23
4V
alue
Gro
wth
23
4V
alue
Big
-0.8
620
-0.4
531
-0.2
196
-0.1
612
0.32
23B
ig-5
.19
**-3
.71
**-1
.18
-0.5
90.
79
2-1
.951
0-0
.910
4-0
.841
2-0
.694
8-0
.671
82
-5.5
9**
-3.8
2**
-3.2
3**
-3.1
2**
-1.9
9*
3-3
.325
6-1
.633
8-1
.255
6-0
.871
2-0
.969
63
-7.8
9**
-4.1
7**
-4.1
1**
-2.9
7**
-2.9
1**
4-2
.574
6-1
.359
7-1
.538
5-1
.065
4-1
.409
74
-4.4
1**
-2.7
7**
-3.4
8**
-2.6
4**
-3.3
5**
Smal
l-1
.308
9-0
.959
7-0
.294
3-0
.634
2-0
.441
4Sm
all
-1.9
2-1
.49
-0.5
3-1
.25
-0.9
1
βi
Gro
wth
23
4V
alue
Gro
wth
23
4V
alue
Big
1.01
891.
0699
0.93
730.
9896
0.79
68B
ig25
.00
**42
.34
**14
.66
**17
.74
**9.
98**
20.
9880
0.84
050.
8310
0.76
950.
8611
211
.23
**23
.71
**20
.43
**14
.13
**10
.93
**
31.
0464
0.82
060.
7339
0.65
790.
8466
311
.36
**13
.60
**16
.40
**18
.45
**16
.58
**
41.
1646
0.94
350.
7838
0.67
800.
9125
48.
39**
9.87
**8.
75**
11.2
7**
14.6
0**
34
Tab
le8
(con
tinu
ed)
Smal
l1.
0073
0.92
280.
8441
0.74
690.
8266
Smal
l8.
27**
5.79
**7.
24**
7.53
**12
.57
**
Adj
R2
Gro
wth
23
4V
alue
Big
0.78
840.
8943
0.76
040.
6313
0.32
24
20.
5333
0.61
650.
6149
0.61
040.
4738
30.
4289
0.40
000.
4308
0.41
510.
4445
40.
3164
0.31
620.
2807
0.26
570.
3763
Smal
l0.
2114
0.18
760.
1985
0.19
720.
2625
GR
Sχ
2
11.0
3**
269.
80**
35
Consistent with Fama and French (1993); Gaunt (2004) and Halliwell, Heaney, and Sawicki (1999)
there is evidence of abnormal returns monotonically increasing from the lowest book to market
portfolios (growth) to the highest book to market portfolios (value). This result is consistent across
all five size quintiles and reinforces the evidence of a strong value premium in Australia. Overall the
explanatory power of the CAPM is relatively low with the average R2 of the 25 equations within the
system being 43.9%. The model particularly struggles to explain the returns of the portfolios within
the value quintile, and portfolios within the small quintile.
Table 9 presents the estimates from estimating equation 4 when SMB 2x3 and HML 2x3 are used.
The results now indicate a strong abnormal performance is present for small firms, even after the
effects of SMB and HML are taken into account, with a strong non-linearity still present in the size-
effect. This is demonstrated with the average intercept of the portfolios within the big quintile being
slightly negative. As we move down the size quintile the average intercepts declines in value and
then increases as we reach the portfolios within the small quintile where the intercepts are positive,
though insignificant.
The results also indicate that the value premium has substantially disappeared. This is demonstrated
by similar intercepts for the growth and value portfolios within each size quintile, e.g. small growth
having an intercept 0.46 while small value has an intercept of 0.45. This suggests that the value
premium is now being captured by the HML factor. This is supported by evidence in Table 9 that
the HML factor does possess explanatory power, particularly for the portfolios with high average
book to market values (portfolio 3, 4 and value). Contrary to Halliwell, Heaney, and Sawicki (1999)
and the mixed evidence in Gaunt (2004) this evidence is the strongest, so far in Australia, that HML
possesses explanatory power. As expected, the results indicate a linear relationship between HML
and book to market values, with growth portfolios having a negative or very low loading and as we
move towards value portfolios this loading steadily increases.
Consistent with Fama and French (1993, 1996); Gaunt (2004) and Halliwell, Heaney, and Sawicki
(1999) there is evidence of a monotonic relationship between size and the SMB factor. This is
demonstrated by the steady increase in loading on the SMB factor as we move from the big to small
size quintile. There is also some evidence of an increased loading on SMB as we move from value to
growth portfolios. These results all indicate that SMB is a very important factor in the Australian
market.
36
Tab
le9
Reg
ress
ions
resu
lts
from
the
thre
efa
ctor
model
The
tabl
epr
esen
tsth
ere
sult
sfr
omre
gres
sing
the
300
exce
ssm
onth
lyre
turn
sof
each
ofth
e25
size
-boo
kto
mar
ket
port
folio
son
exce
ssm
arke
tre
turn
s.
The
25si
ze-b
ook
tom
arke
tpo
rtfo
lios
are
form
edby
usin
gin
depe
nden
tso
rts
base
don
mar
ket
capi
talis
atio
nan
dbo
okto
mar
ket
valu
esdu
ring
the
peri
od
1982
to20
06.
The
follo
win
gti
me-
seri
esre
gres
sion
ises
tim
ated
r it=
αi+
βir
mt+
s iS
MB
t+
hiH
ML
t+
e it
Whe
rer i
tis
the
retu
rnon
port
folio
iin
mon
tht
less
the
13w
eek
trea
sury
note
yiel
d,r m
tis
the
valu
e-w
eigh
ted
mar
ket
mon
thly
retu
rnle
ssth
e13
wee
k
trea
sury
note
yiel
d.S
MB
tis
the
retu
rnon
the
mim
icki
ngsi
zepo
rtfo
lioan
dH
ML
tis
the
retu
rnon
the
mim
icki
ngbo
okto
mar
ket
port
folio
.B
oth
mim
icki
ngpo
rtfo
lios
are
form
edus
ing
two
size
and
thre
ebo
okto
mar
ket
port
folio
split
s.T
hesy
stem
ises
tim
ated
usin
gth
eG
MM
tech
niqu
ew
ith
the
follo
win
gm
omen
tco
ndit
ions
;E
[εit]=
0,E
[εitr m
t]=
0,E
[eitS
MB
t]=
0,E
[eitH
ML
t]=
0∀i
=1,···,
N.T
het-
stat
isti
cfo
rth
ere
gres
sion
coeffi
cien
ts
uses
HA
Cst
anda
rder
rors
.T
head
just
edR
2ar
eca
lcul
ated
for
each
equa
tion
inth
esy
stem
.W
eal
sore
port
the
Gib
bons
,R
oss,
and
Shan
ken
(198
9)
(GR
S)
test
stat
isti
c,an
adju
sted
χ2
stat
isti
c(s
eeeq
uati
on6)
that
adju
sted
the
GR
Sst
atis
tics
for
cros
s-co
rrel
ated
erro
rsan
dth
eN
ewey
and
Wes
t(1
987)
D-s
tati
stic
.
**an
d*
deno
tesi
gnifi
canc
eat
the
1%an
d5%
leve
lsre
spec
tive
ly.
Coeffi
cien
tt-
stat
isti
c
αi
Gro
wth
23
4V
alue
Gro
wth
23
4V
alue
Big
-0.5
943
-0.4
557
-0.2
489
-0.6
243
-0.5
119
Big
-3.2
7**
-3.1
4**
-1.0
1-2
.10
*-1
.01
2-1
.248
8-0
.620
8-0
.633
9-0
.631
5-0
.703
12
-3.9
8**
-3.0
2**
-2.6
2**
-2.5
7*
-1.8
7
3-2
.181
1-0
.828
7-0
.732
1-0
.530
0-0
.582
43
-7.3
1**
-3.2
8**
-3.0
9**
-2.4
6*
-2.5
2*
4-0
.917
9-0
.160
7-0
.572
4-0
.363
5-0
.958
64
-2.3
1*
-0.5
0-2
.01
*-1
.89
-3.8
4**
Smal
l0.
4645
0.68
961.
0620
0.43
120.
4528
Smal
l1.
121.
593.
07**
1.34
1.75
βi
Gro
wth
23
4V
alue
Gro
wth
23
4V
alue
Big
0.95
551.
0520
0.94
841.
0709
0.91
81B
ig24
.25
**34
.35
**12
.04
**22
.96
**5.
97**
21.
0281
0.90
790.
8920
0.83
200.
9811
214
.40
**27
.68
**19
.43
**14
.48
**13
.60
**
31.
1094
0.89
880.
8180
0.75
890.
9807
320
.07
**20
.54
**16
.62
**23
.41
**21
.73
**
41.
2653
1.05
080.
9132
0.82
791.
1056
414
.12
**15
.33
**12
.51
**20
.23
**21
.80
**
Smal
l1.
1488
1.06
301.
0114
0.90
431.
0120
Smal
l15
.06
**9.
08**
10.9
2**
11.3
6**
21.9
9**
37
Tab
le9
(con
tinu
ed)
s iG
row
th2
34
Val
ueG
row
th2
34
Val
ue
Big
-0.0
909
-0.0
698
0.02
570.
0499
-0.0
059
Big
-1.9
2-2
.98
**0.
580.
86-0
.06
20.
5464
0.41
870.
3483
0.27
310.
4389
29.
29**
8.01
**6.
28**
5.36
**4.
71**
30.
8812
0.74
900.
6137
0.57
550.
7270
313
.81
**14
.59
**12
.76
**12
.24
**11
.26
**
41.
3120
1.08
041.
0338
0.96
360.
9874
414
.89
**18
.10
**18
.02
**19
.74
**12
.94
**
Smal
l1.
5323
1.45
781.
3966
1.19
611.
2063
Smal
l18
.93
**14
.91
**14
.88
**12
.07
**18
.60
**
hi
Gro
wth
23
4V
alue
Gro
wth
23
4V
alue
Big
-0.3
874
-0.0
834
0.06
190.
5379
0.85
09B
ig-4
.51
**-2
.07
*0.
885.
04**
3.18
**
2-0
.048
50.
2185
0.21
640.
2718
0.57
362
-0.4
73.
25**
2.60
**4.
06**
5.06
**
3-0
.090
50.
0956
0.21
840.
3589
0.49
853
-0.8
81.
323.
01**
5.32
**5.
34**
4-0
.086
00.
0992
0.28
120.
4666
0.75
384
-0.4
70.
982.
46*
6.66
**6.
13**
Smal
l0.
0656
0.10
140.
3273
0.37
930.
5681
Smal
l0.
510.
652.
88**
3.59
**6.
47**
Adj
R2
Gro
wth
23
4V
alue
Big
0.83
260.
8980
0.76
030.
7069
0.48
70
20.
7170
0.76
180.
7169
0.68
930.
6226
30.
7765
0.74
450.
7233
0.71
990.
7613
40.
7685
0.75
190.
7664
0.78
380.
8094
Smal
l0.
7388
0.68
730.
7472
0.69
740.
8027
GR
Sχ
2D
-tes
t
8.99
**14
1.28
**15
639*
*
38
As discussed in Section 4 our preferred methodology in forming the SMB and HML factors in
Australia is to use SMB 5x3 and HML 5x3. Table 10 reports the results from estimating equation
4 when SMB 5x3 and HML 5x3 are used. As expected, the results are similar to those reported in
Table 9 and again indicate that the exact methodology in forming the factors is immaterial. The
results demonstrate that there is a monotonic relationship between size and the SMB factor with
the loading on SMB factor increasing as we move from the portfolio in the big to small quintile.
Results also indicate a monotonic relationship between book to market value and HML, with the
loading on HML increasing as we move from portfolios in the growth quintile to portfolios in the
value portfolios.
The results still demonstrate a non-linear relationship between abnormal returns and size with the
intercept terms declining as we move from the portfolio in the big quintile through quintile 2 and 3
and then increasing as we go from quintile 3 until we reach the portfolios within the small quintile.
In contrast, the relationship between book to market and abnormal returns has substantially been
removed.
The results from Tables 9 and 10 indicate that the Fama and French (1993) three factor model
explains a significantly higher proportion of the variation in returns in Australia when compared
to the CAPM. This can be seen in the average adjusted R2 for each equation in the system, with
the three factor model having an average adjusted R2 of 73% compared to the CAPM average of
44%. In particular is the increased explanatory power of the three factor model in explaining the
returns of portfolios in the small quintile where the average R2 has risen from 21% to 77%. The
D-test also indicates that we should reject the null that the coefficients on SMB and HML are equal
to zero. This substantial increase in explanatory power of the Fama and French (1993) model over
the CAPM in explaining the cross-sectional variation in returns is similar to the evidence presented
in Fama and French (1993), although the US evidence suggests the three factor model explains a
higher proportion of returns. Though the results are supportive of the Fama and French (1993)
three factor model the model still struggles to explain a number of the portfolio returns, particularly
the portfolio big value. The results also indicate that the vast majority of the intercept terms are
significantly different from zero which is confirmed by the GRS statistic. The large and significant
intercept terms generally occur in the middle size quintiles. This is consistent with previous evidence
of a non-linear relationship between size and returns and indicates that some other unknown factor
is driving this under performance. Overall, these results indicate that although the Fama and French
(1993) model is superior to the CAPM in explaining the cross-section of average stock returns in
Australia it is not a complete model and still leaves the non-linear relationship between size and
returns unexplained.
39
Tab
le10
Reg
ress
ions
resu
lts
from
the
thre
efa
ctor
model
The
tabl
epr
esen
tsth
ere
sult
sfr
omre
gres
sing
the
300
exce
ssm
onth
lyre
turn
sof
each
ofth
e25
size
-boo
kto
mar
ket
port
folio
son
exce
ssm
arke
tre
turn
s.
The
25si
ze-b
ook
tom
arke
tpo
rtfo
lios
are
form
edby
usin
gin
depe
nden
tso
rts
base
don
mar
ket
capi
talis
atio
nan
dbo
okto
mar
ket
valu
esdu
ring
the
peri
od
1982
to20
06.
The
follo
win
gti
me-
seri
esre
gres
sion
ises
tim
ated
r it=
αi+
βir
mt+
s iS
MB
t+
hiH
ML
t+
e it
Whe
rer i
tis
the
retu
rnon
port
folio
iin
mon
tht
less
the
13w
eek
trea
sury
note
yiel
d,r m
tis
the
valu
e-w
eigh
ted
mar
ket
mon
thly
retu
rnle
ssth
e13
wee
k
trea
sury
note
yiel
d.S
MB
tis
the
retu
rnon
the
mim
icki
ngsi
zepo
rtfo
lioan
dH
ML
tis
the
retu
rnon
the
mim
icki
ngbo
okto
mar
ket
port
folio
.B
oth
mim
icki
ngpo
rtfo
lios
are
form
edus
ing
five
size
and
thre
ebo
okto
mar
ket
port
folio
split
s.T
hesy
stem
ises
tim
ated
usin
gth
eG
MM
tech
niqu
ew
ith
the
follo
win
gm
omen
tco
ndit
ions
;E
[εit]=
0,E
[εitr m
t]=
0,E
[eitS
MB
t]=
0,E
[eitH
ML
t]=
0∀i
=1,···,
N.T
het-
stat
isti
cfo
rth
ere
gres
sion
coeffi
cien
ts
uses
HA
Cst
anda
rder
rors
.T
head
just
edR
2ar
eca
lcul
ated
for
each
equa
tion
inth
esy
stem
.W
eal
sore
port
the
Gib
bons
,R
oss,
and
Shan
ken
(198
9)
(GR
S)
test
stat
isti
c,an
adju
sted
χ2
stat
isti
c(s
eeeq
uati
on6)
that
adju
sted
the
GR
Sst
atis
tics
for
cros
s-co
rrel
ated
erro
rsan
dth
eN
ewey
and
Wes
t(1
987)
D-s
tati
stic
.
**an
d*
deno
tesi
gnifi
canc
eat
the
1%an
d5%
leve
lsre
spec
tive
ly.
Coeffi
cien
tt-
stat
isti
c
αi
Gro
wth
23
4V
alue
Gro
wth
23
4V
alue
Big
-0.5
315
-0.4
199
-0.2
786
-0.6
701
-0.3
086
Big
-2.4
5*
-2.8
7**
-1.0
8-1
.94
-0.5
3
2-1
.422
0-0
.908
3-0
.885
1-0
.966
0-1
.326
62
-4.2
2**
-3.7
7**
-3.0
7**
-3.6
8**
-3.6
2**
3-2
.704
9-1
.339
1-1
.238
3-1
.123
9-1
.311
13
-7.6
2**
-4.5
8**
-4.4
9**
-4.4
5**
-4.9
9**
4-1
.629
7-0
.962
8-1
.366
2-1
.199
1-1
.999
74
-3.4
7**
-2.6
1**
-3.9
6**
-4.4
5**
-5.9
9**
Smal
l-0
.492
4-0
.433
5-0
.136
7-0
.778
9-0
.783
6Sm
all
-1.3
4-1
.15
-0.4
3-2
.47
*-3
.35
**
beta
iG
row
th2
34
Val
ueG
row
th2
34
Val
ue
Big
0.95
081.
0551
0.95
211.
0819
0.89
37B
ig21
.03
**35
.61
**11
.47
**20
.93
**6.
02**
20.
9588
0.89
140.
8785
0.85
241.
0361
217
.46
**23
.78
**16
.09
**14
.86
**15
.44
**
31.
0455
0.86
110.
8060
0.77
781.
0038
318
.79
**19
.20
**13
.44
**20
.40
**22
.84
**
41.
1613
1.01
310.
8878
0.82
801.
1459
417
.91
**15
.03
**13
.74
**17
.34
**19
.05
**
Smal
l1.
0895
1.05
271.
0188
0.95
581.
0675
Smal
l16
.27
**11
.63
**11
.80
**13
.21
**26
.17
**
40
Tab
le10
(con
tinu
ed)
s iG
row
th2
34
Val
ueG
row
th2
34
Val
ue
Big
-0.0
745
-0.0
475
0.02
650.
0525
-0.0
227
Big
-1.7
4-2
.95
**0.
731.
09-0
.29
20.
2798
0.25
590.
2023
0.19
660.
3486
25.
64**
5.68
**4.
42**
4.48
**4.
09**
30.
4951
0.43
980.
3749
0.39
680.
5115
38.
68**
9.57
**7.
84**
8.65
**8.
68**
40.
7441
0.66
760.
6594
0.64
320.
6928
49.
34**
10.9
7**
13.8
4**
12.4
6**
9.98
**
Smal
l1.
0684
1.07
351.
0008
0.92
880.
9297
Smal
l21
.84
**19
.59
**18
.35
**15
.62
**22
.52
**
hi
Gro
wth
23
4V
alue
Gro
wth
23
4V
alue
Big
-0.3
544
-0.0
526
0.06
880.
5193
0.60
65B
ig-3
.25
**-1
.17
0.70
4.16
**2.
20*
2-0
.397
70.
1068
0.12
890.
3484
0.78
772
-2.8
8**
1.14
1.20
3.81
**5.
88**
3-0
.395
7-0
.100
90.
1424
0.41
540.
5510
3-2
.93
**-0
.87
1.38
4.40
**5.
14**
4-0
.606
5-0
.103
90.
1119
0.40
390.
8706
4-2
.58
*-0
.78
0.76
4.19
**4.
98**
Smal
l-0
.343
3-0
.057
70.
2714
0.53
610.
7293
Smal
l-2
.27
*-0
.37
2.08
*4.
22**
6.95
**
Adj
R2
Gro
wth
23
4V
alue
Big
0.81
260.
8966
0.76
020.
6796
0.39
29
20.
7157
0.72
620.
6795
0.67
880.
6293
30.
7387
0.68
620.
6509
0.67
090.
7189
40.
7378
0.69
990.
7037
0.70
610.
7413
Smal
l0.
8363
0.78
310.
7754
0.76
980.
8543
GR
Sχ
2D
-tes
t
8.77
**11
6.24
**89
410*
*
41
5.1 Sensitivity analysis
Section 4 demonstrated a strong June and July seasonality in the SMB factor and an abnormal result
in 1989 for the HML factor. To test if these seasonalities or abnormal results affect are analysis, we
adjust equation 4 to take these phenomena into account by adding interactive dummy variables to
the equation and evaluate the following model:
rit = αi + βirmt + siSMBt + hiHMLt+
djunDjunSMBt + djulDjulSMBt + d1989D1989HMLt + eit i = 1, · · · , N. (7)
Where rit is the excess return on asset i in time t. rmt is the excess return of the market. SMBt is
the return on the mimicking size portfolio and HMLt is the return on the mimicking book to market
portfolio. αi, βi, si, hi, djun, djul and d1989 are regression coefficients, eit are the error terms, N is
the number of test assets. Djun, Djul and D1989 are dummy variables that takes the value of one
if the month is June (Djun), if the month is July (Djul), if the year is 1989 (D1989) otherwise the
value is zero. An interactive dummy variable approach is used because the seasonality is specific to
the SMB factor, while the anomalies behaviour in 1989 is specific to the HML factor.
The results from estimating equation 7 are reported in Table 11. These results are consistent with
those reported in Table 10. The results indicate that the June and July seasonal regularity is
captured within the SMB factors and separating out the two effects adds little to the model. This
is demonstrated by insignificant co-efficient values on djun and djul indicating a similar loading on
the SMB factor no matter the time of the year. In contrast, the loading on the HML factor is quite
different in 1989 than for the rest of the time period. Nearly all of the portfolios in 1989 have a
significant positive loading on the HML factor and the loadings are larger than the rest of the time
period. The result is particularly dramatic for the portfolios in the growth and value quintiles. The
loadings on the portfolios within the growth quintile generally goes from a significant negative to
a significant positive loading. For the portfolios in the value portfolio the loadings substantially
increases particularly portfolio big-value whose loading goes from 0.34 to 2.92. These results confirm
that the return in 1989 for the HML factor and the resulting responses in the portfolio returns are
unusual and a control for 1989 is required.
A number of other unreported sensitivity test have also been carried out and indicate that the load-
ings on the factors are relatively constant when the sample is split into two sub-periods, although
individually estimates have higher variability, as expected, because of the lower number of observa-
tions available in the estimation. The results also indicate that both the SMB and HML factors are
significant when the system is run with only one of the factors at a time. As expected the results
indicate that the SMB is essential to explain the difference in returns between the big and small
quintiles, while the HML factor is required to explain the difference in returns between the growth
and value quintiles.
42
Tab
le11
Reg
ress
ions
resu
lts
from
the
thre
efa
ctor
model
The
tabl
epr
esen
tsth
ere
sult
sfr
omre
gres
sing
the
300
exce
ssm
onth
lyre
turn
sof
each
ofth
e25
size
-boo
kto
mar
ket
port
folio
son
exce
ssm
arke
tre
turn
s.
The
25si
ze-b
ook
tom
arke
tpo
rtfo
lios
are
form
edby
usin
gin
depe
nden
tso
rts
base
don
mar
ket
capi
talis
atio
nan
dbo
okto
mar
ket
valu
esdu
ring
the
peri
od
1982
to20
06.
The
follo
win
gti
me-
seri
esre
gres
sion
ises
tim
ated
r it=
αi+
βir
mt+
s iS
MB
t+
hiH
ML
t+
dju
nD
ju
nS
MB
t+
dju
lDju
lSM
Bt+
d1989D
1989H
ML
t+
e it
Whe
rer i
tis
the
retu
rnon
port
folio
iin
mon
tht
less
the
13w
eek
trea
sury
note
yiel
dan
dr m
tis
the
valu
e-w
eigh
ted
mar
ket
mon
thly
retu
rnex
trac
ted
from
the
CR
IFpr
ice
rela
tive
file.
SM
Bt
isth
ere
turn
onth
em
imic
king
size
port
folio
and
HM
Lt
isth
ere
turn
onth
em
imic
king
book
tom
arke
tpo
rtfo
lio.
Bot
hm
imic
king
port
folio
sar
efo
rmed
usin
gfiv
esi
zean
dth
ree
book
tom
arke
tpo
rtfo
liosp
lits.
Dju
n,D
ju
lan
dD
1989
are
dum
my
vari
able
sth
ateq
uals
one
ifth
em
onth
isJu
ne(D
ju
n),
orJu
ly(D
ju
l),o
rth
eye
aris
1989
(D1989)
othe
rwis
eth
edu
mm
yva
riab
leeq
uals
zero
.T
hesy
stem
ises
tim
ated
usin
gth
e
GM
Mte
chni
que.
The
t-st
atis
tic
for
the
regr
essi
onco
effici
ents
uses
HA
Cst
anda
rder
rors
.T
head
just
edR
2ar
eca
lcul
ated
for
each
equa
tion
inth
esy
stem
.
We
also
repo
rtth
eG
ibbo
ns,R
oss,
and
Shan
ken
(198
9)(G
RS
)te
stst
atis
tic.
**an
d*
deno
tesi
gnifi
canc
eat
the
1%an
d5%
leve
lsre
spec
tive
ly.
Coeffi
cien
tt-
stat
isti
cp-v
alue
αi
Gro
wth
23
4V
alue
Gro
wth
23
4V
alue
Big
-0.4
177
-0.4
041
-0.2
615
-0.5
122
0.01
12B
ig-1
.94
-2.8
9**
-1.1
1-1
.83
0.03
2-1
.235
8-0
.738
3-0
.730
8-0
.904
7-1
.308
02
-4.4
2**
-3.7
4**
-2.8
9**
-3.8
2**
-3.6
8**
3-2
.644
8-1
.270
9-1
.093
6-0
.986
2-1
.273
43
-8.9
5**
-4.7
1**
-4.6
2**
-4.8
3**
-5.3
6**
4-1
.295
7-0
.829
1-1
.043
3-1
.154
9-1
.651
74
-3.3
7**
-2.5
6*
-4.2
5**
-4.7
7**
-7.8
1**
Smal
l-0
.307
1-0
.434
5-0
.022
6-0
.586
4-0
.695
3Sm
all
-0.9
8-1
.41
-0.0
8-2
.34
*-3
.84
**
βi
Gro
wth
23
4V
alue
Gro
wth
23
4V
alue
Big
0.94
311.
0518
0.94
901.
0660
0.84
80B
ig22
.66
**40
.98
**13
.58
**25
.14
**8.
82**
20.
9386
0.87
450.
8610
0.84
351.
0265
224
.94
**25
.05
**18
.20
**16
.46
**16
.66
**
31.
0233
0.85
380.
7924
0.76
370.
9894
321
.67
**22
.47
**14
.91
**25
.45
**26
.57
**
41.
1130
0.99
080.
8595
0.81
211.
1069
423
.39
**18
.77
**20
.06
**20
.75
**22
.21
**
Smal
l1.
0711
1.03
900.
9980
0.94
331.
0560
Smal
l18
.42
**14
.52
**14
.46
**18
.01
**27
.38
**
43
Tab
le11
(con
tinu
ed)
s iG
row
th2
34
Val
ueG
row
th2
34
Val
ue
Big
-0.0
921
-0.0
490
0.03
090.
0114
-0.0
371
Big
-2.0
6*
-3.0
7**
0.91
0.28
-0.6
6
20.
2719
0.23
500.
1799
0.17
850.
3085
25.
25**
4.94
**3.
97**
4.03
**3.
68**
30.
4511
0.44
830.
3559
0.38
730.
4561
38.
40**
10.0
0**
7.55
**8.
20**
8.76
**
40.
6557
0.62
160.
6302
0.59
860.
6265
411
.63
**11
.30
**14
.41
**12
.61
**10
.10
**
Smal
l1.
0549
1.04
530.
9642
0.93
980.
9199
Smal
l22
.87
**22
.46
**17
.04
**17
.35
**25
.10
**
hi
Gro
wth
23
4V
alue
Gro
wth
23
4V
alue
Big
-0.3
847
-0.0
721
0.04
600.
4582
0.33
74B
ig-3
.67
**-1
.79
0.53
5.40
**2.
45*
2-0
.513
10.
0234
0.04
190.
3087
0.76
032
-4.6
7**
0.31
0.51
3.62
**5.
71**
3-0
.499
0-0
.152
30.
0769
0.33
820.
5064
3-4
.47
**-1
.36
0.88
5.11
**5.
14**
4-0
.829
1-0
.203
0-0
.032
60.
3410
0.69
044
-4.7
3**
-1.9
7*
-0.3
53.
86**
6.21
**
Smal
l-0
.442
5-0
.122
10.
1732
0.45
450.
6670
Smal
l-3
.60
**-0
.95
1.69
4.64
**7.
98**
dju
nG
row
th2
34
Val
ueG
row
th2
34
Val
ue
Big
0.10
86-0
.034
1-0
.069
10.
1680
-0.3
947
Big
1.73
-0.8
0-1
.10
2.36
*-1
.55
2-0
.110
10.
0259
0.00
500.
0272
0.11
702
-1.4
20.
400.
080.
431.
16
3-0
.046
5-0
.118
30.
0435
-0.0
409
0.15
403
-0.4
1-1
.40
0.56
-0.6
02.
30*
40.
0865
0.05
240.
0425
0.05
870.
1287
40.
490.
620.
570.
751.
58
Smal
l-0
.037
9-0
.066
0-0
.007
2-0
.087
2-0
.047
6Sm
all
-0.4
1-0
.40
-0.0
7-0
.93
-0.7
0
dju
lG
row
th2
34
Val
ueG
row
th2
34
Val
ue
Big
-0.0
041
0.01
73-0
.002
50.
0861
0.10
98B
ig-0
.07
0.46
-0.0
51.
280.
88
20.
0032
0.02
620.
0533
0.06
420.
1770
20.
040.
320.
540.
961.
43
30.
2572
-0.0
308
0.01
920.
0056
0.24
203
2.55
-0.3
00.
190.
082.
09*
40.
3201
0.18
38-0
.011
70.
2189
0.15
954
2.20
1.51
-0.1
52.
80**
2.09
*
Smal
l0.
0036
0.20
300.
1661
-0.1
248
0.03
62Sm
all
0.04
2.22
*1.
45-1
.17
0.31
44
Tab
le11
(con
tinu
ed)
d1989
Gro
wth
23
4V
alue
Gro
wth
23
4V
alue
Big
0.25
330.
2116
0.26
960.
5111
2.92
04B
ig1.
922.
51*
1.64
0.80
2.67
**
21.
2330
0.82
910.
8694
0.37
240.
1670
28.
32**
7.42
**5.
84**
3.08
**0.
76
31.
0045
0.59
530.
6388
0.80
690.
3052
35.
97**
3.82
**2.
21*
4.85
**2.
15*
42.
1339
0.93
061.
4545
0.54
841.
7252
47.
88**
2.75
**3.
89**
3.76
**3.
83**
Smal
l1.
0295
0.63
420.
9583
0.91
320.
6499
Smal
l4.
99**
1.61
1.95
3.47
**1.
63
Adj
R2
Gro
wth
23
4V
alue
Big
0.81
380.
8965
0.76
000.
6846
0.49
51
20.
7310
0.73
570.
6903
0.67
940.
6319
30.
7504
0.68
920.
6543
0.68
050.
7279
40.
7623
0.70
580.
7205
0.71
320.
7698
Smal
l0.
8392
0.78
510.
7801
0.77
520.
8564
GR
S
7.15
**
45
6 Conclusion
While the size and book to market effects and the application of the Fama and French (1993) model
have been extensively documented using US data, there have been few studies on these effects in
Australia. This has been a results of an absence of a comprehensive database containing accounting
information. This lack of a comprehensive database has lead to severe data limitations on previous
studies in Australia, potentially affecting the results. Previous studies in Australia have found only
weak evidence of a value effect being present in Australia, although a size effect is well documented.
In contrast, evidence from the US and other markets suggests that a value premium is common.
A lack of evidence for the value premium in Australia could be the result of data limitation and
previous studies only focusing on relatively short periods of time, generally covering the mid to late
1990’s when accounting information in Australia is more readily available. The previous studies on
the value premium in Australia have had, on average, access to less than 35% of all companies that
produced annual reports. This study is the first to rectify this lack of market coverage by hand
collecting accounting information from over 98% of all companies that produced an annual report.
We also extend the previous studies to cover the period 1982 to 2006.
We present evidence on the size and book to market anomalies that indicate that the size effect
is non-linear in Australia while the book to market effect is linear. These results suggest that
forming the SMB and HML factors following the methodology proposed by Fama and French (1993)
in Australia may not capture the size and book to market anomalies correctly. We analyse three
potential methods for forming the two factors in Australia and conclude that the three methodologies
will give factors that behave in a similar manner, although the preferred methodology in Australia
is to use five size and three book to market independent sorts. These results indicate a large HML
premium in Australia, with earnings on average of 0.959% per month. This result is consistent with
previous studies in Australia (Gharghori, Chan, and Faff, 2006; Halliwell, Heaney, and Sawicki, 1999)
and is larger than the US HML factor which averages 0.462% per month. In contrast, the SMB factor
earns an average monthly return of -0.486%, which is insignificantly different from zero. This result
is consistent with the US SMB factor which earns an average monthly return of 0.086% which is
insignificantly different from zero. These results support recent evidence that the size premium has
been declining in equity markets in recent years (Barber and Lyon, 1997; Dimson and Paul, 1999).
We also demonstrate a strong June and July seasonality in the SMB factor which is consistent
with previous research on the size anomaly in Australia (Brown, Keim, Kleidon, and Marsh, 1983;
Durand, Juricev, and Smith, 2007). We also document that in 1989 the HML factor experienced
a large negative return. This result seems to be driven by value firms experiencing severe distress
leading into 1989, with the average book to market value in 1989 being substantially higher than
other years in the sample. We also document that the result is also influenced by a few large firms
experiencing extremely large negative returns of over 50% for several months in 1989.
46
The current study extends prior Australian studies analysing the Fama and French (1993) model
by combining the periods analysed in Halliwell, Heaney, and Sawicki (1999) and Gaunt (2004) and
extending it by 5 years giving a time period of 25 years. We also significantly increase the number
of companies in our sample. Consistent with prior evidence in the US and Australia (Fama and
French, 1993, 1996; Gaunt, 2004; Halliwell, Heaney, and Sawicki, 1999), our results demonstrate a
strong monotonic relationship between size and the SMB factor. In contrast to previous Australian
evidence there is also a strong monotonic relationship between book to market portfolios and the
HML factor, which is consistent with the US evidence (Fama and French, 1993, 1996). Consistent
with a number of Australian studies (Durack, Durand, and Maller, 2004; Durand, Limkriangkrai,
and Smith, 2006a; Gaunt, 2004) we find that the Fama and French (1993) three factor model provides
significant improvement over the CAPM in explaining the cross-section of portfolio returns. Finally,
in contrast to Durack, Durand, and Maller (2004); Durand, Limkriangkrai, and Smith (2006a); Gaunt
(2004); Halliwell, Heaney, and Sawicki (1999) which find that the bulk of the increased explanatory
power is due to the SMB factor, but consistent with Gharghori, Chan, and Faff (2006) we find that
both SMB and HML factors are important in explaining the cross-section of portfolio returns in
Australia.
Overall our study suggests that the Fama and French (1993) three factor model provides a significant
improvement over the CAPM in explaining the cross-section of portfolio returns. The results also
indicate that the three factor model can not explain the returns of the portfolios in the middle
size quintiles, which earn substantial negative abnormal returns. This result confirms the non-
linear relationship between returns and size. This indicates that to fully explain the cross-section
of portfolio returns in Australia more work is required in understanding the non-linear relationship
between returns and size.
47
References
Allen, D.E., H. Lisnawati, and M. Clissold, 1998, Predicting earnings growth using e/p ratios, Australian
Journal of Management 23, 115–128.
Bagella, Michele, Leonardo Becchetti, and Andrea Carpentieri, 2000, “the first shall be last”, size and value
stratedy premia at the london stock exhange, Journal of Banking and Finance 24, 893–919.
Ball, Ray, 1978, Anomalies in relationship between securities yields and yield-surrogates, Journal of Financial
Economics 6, 103–126.
, Philip Brown, and R.R. Officer, 1976, Asset pricing in the australian industrial equity market,
Australian Journal of Management 1, 1–32.
Banz, Rolf W., 1981, The relationship between return and market value of common stocks, Journal of Financial
Economics 9, 3–18.
Barber, Brad M., and John D. Lyon, 1997, Firm size, book-to-market ratio and security returns: A holdout
sample of financial firms, Journal of Finance 52, 875–883.
Basu, Sanjoy, 1977, Investment performance of common stocks in relation to their price-earnings ratios: A
test of the efficient market hypothesis, Journal of Finance 32, 663–682.
, 1983, The relationship between earnings’ yield market value and return for nyse common stocks,
Journal of Financial Economics 12, 129–156.
Beedles, William L., Peter Dodd, and P.R. Officer, 1988, Regularities in australian share returns, Australian
Journal of Management 13, 1–29.
Bhandari, Laxmi Chand, 1988, Debt/equity ratio and expected common stock returns: Empirical evidence,
Journal of Finance 43, 507–528.
Black, Fischer, 1972, Capital market equilibrium with restricted borrowing, Journal of Business 45, 444–455.
, 1993a, Beta and return, Journal of Portfolio Management 20, 8–18.
, 1993b, Estimating expected return, Financial Analysts Journal 49, 36–38.
, Michael C. Jensen, and Myron Scholes, 1972, The capital asset pricing model: Some empirical tests,
in Michael C Jensen, ed.: Studies in the Theory of Capital Markets (Praeger).
Brailsford, Tim, and Stephen Easton, 1991, Seasonality in australian share price indices between 1936 and
1957, Accounting and Finance 31, 69–85.
Brown, Philip, Donald B. Keim, Allan W. Kleidon, and Terry A. Marsh, 1983, Stock return seasonalities and
the tax-loss selling hypothesis, Journal of Financial Economics 12, 105–127.
Campbell, John Y., 1996, Understanding risk and return, Journal of Political Economy 104, 298–345.
, and Tuomo Vuolteenaho, 2004, Bad beta, good beta, The American Economic Review 94, 1249–1275.
48
Carhart, Mark M., 1997, On persistence in mutual fund performance, Journal of Finance 52, 57–82.
Chan, Louis K.C., Yasushi Hamao, and Josef Lakonishok, 1991, Fundamentals and stock returns in japan,
Journal of Finance 46, 1739–1764.
Chan, Louis K.C., Narasimhan Jegadeesh, and Josef Lakonishok, 1995, Evaluating then performance of value
versus glamour stocks. the impact of selection bias, Journal of Financial Economics 38, 269–296.
, 1996, Momentum strategies, Journal of Finance 51, 1681–1713.
Chen, Nai-Fu, Richard Roll, and Stephen A. Ross, 1986, Economic forces and the stock market, Journal of
Business 59, 383–403.
Cochrane, John H., 2005, Asset Pricing (Princeton University Press) revised edn.
Conrad, Jennifer, and Gautam Kaul, 1998, An anatomy of trading strategies, The Review of Financial Studies
11, 489–519.
Daniel, Kent, and Sheridan Titman, 1997, Evidence on the characteristics of cross sectional variation in stock
returns, Journal of Finance 52, 1–33.
, and K.C. John Wei, 2001, Explaining the cross-section of stock returns in japan: Factors or charac-
teristics?, Journal of Finance 56, 743–766.
Davies, James L., 1994, The cross-section of realized stock returns: The pre-compustat evidence, Journal of
Finance 49, 1579–1593.
, Eugene F. Fama, and Kenneth R. French, 2000, Characteristics, covariances and average returns:
1929 to 1997, Journal of Finance 55, 389–406.
De Bondt, Werner F.M., and Richard Thaler, 1985, Does the stock market overreact?, Journal of Finance 40,
793–805.
, 1987, Further evidence on investor overreaction and stock market seasonality, Journal of Finance 42,
557–581.
Demir, Isabelle, Jay Muthuswamy, and Terry Walter, 2004, Momentum returns in australian equities: The
influences of size, risk, liquidity and return computation, Pacific-Basin Finance Journal 12, 143–158.
Dimson, Elroy, and Marsh Paul, 1999, Murphy’s law and market anomalies, Journal of Portfolio Management
25, 53–69.
Durack, Nick, Robert B. Durand, and Ross A Maller, 2004, A best choice among asset pricing models? the
conditional capital asset pricing model in australia, Accounting and Finance 44, 139–162.
Durand, Robert B., Alex Juricev, and Gary W. Smith, 2007, Smb - arousal, disproportionate reaction and
the size-premium, Pacific-Basin Finance Journal 15, 315–328.
Durand, Robert B., Manapon Limkriangkrai, and Gary W. Smith, 2006a, In america’s thrall: the effects
of the us market and us security characteristics on australian stock returns, Accounting and Finance 46,
577–605.
49
, 2006b, Momentum in australia - a note, Australian Journal of Management 31, 355–364.
Faff, Robert, 1988, An empirical test of the arbitrage pricing theory on australian stock returns 1974:85,
Accounting and Finance 31, 23–43.
, 2001, An examination of the fama and french three-factor model using commercial available factors,
Australian Journal of Management 26, 1–17.
, 2004, A simple test of the fama and french model using daily data: Australian evidence, Applied
financial Economics 14, 83–92.
Fama, Eugene F., 1970, Multiperiod consumption-investment decisions, American Economic Review 60, 163–
174.
, and Kenneth R. French, 1992, The cross-section of expected stock returns, Journal of Finance 47,
427–465.
, 1993, Common risk factors in the returns on stocks and bonds, Journal of Financial Economics 33,
3–56.
, 1995, Size and book-to-market factors in earnings and returns, Journal of Finance 50, 131–155.
, 1996, Multifactor explanations of asset pricing anomalies, Journal of Finance 51, 55–84.
, 1998, Value verses growth: the international evidence, Journal of Finance 53, 1975–1999.
, 2006, The value premium and the capm, Journal of Finance 61, 2163–2185.
, 2007, Migration, Financial Analysts Journal 63, 48–58.
Fama, Eugene F., and James D. MacBeth, 1973, Risk, return and equilibrium empirical test, Journal of
Political Economy 81, 607–636.
Gaunt, Clive, 2004, Size and book to market effects and the fama french three factor asset pricing model:
evidence from the australian stockmarket, Accounting and Finance 44, 27–44.
, and Philip Gray, 2003, Short-term autocorrelation in australian equities, Australian Journal of
Management 28, 97–117.
, and Julie McIvor, 2000, The impact of share price on seasonality and size anomalies in australian
equity returns, Accounting and Finance 40, 33–50.
Gharghori, Phil, Howard Chan, and Robert Faff, 2006, Factors or characteristics?: That is the question,
Pacific Accounting Review 18, 21–46.
Gibbons, Michael R., Stephen A. Ross, and Jay Shanken, 1989, A test of the efficiency of a given portfolio,
Econometrica 57, 1121–1152.
Halliwell, Jason, Richard Heaney, and Julia Sawicki, 1999, Size and book to market effects in australian share
markets: A time series analysis, Accounting Research Journal 12, 122–137.
50
Horowitz, J., T. Loughran, and N Savin, 2000, Three analyses of the firm size premium, Journal of Empirical
Finance 7, 143–153.
Hurn, San, and Vlad Pavlov, 2003, Momentum in australian stock returns, Australian Journal of Management
28, 141–156.
Jaffe, Jeffrey, Donald B. Keim, and Randolph Westerfield, 1989, Earnings yield, market values and stock
returns, Journal of Finance 44, 135–148.
Jagannathan, Ravi, and Zhenyu Wang, 1996, The conditional capm and the cross-section of expected returns,
Journal of Finance 51, 3–53.
Jegadeesh, Narasimhan, and Sheridan Titman, 1993, Returns to buying winners and selling losers: Implica-
tions for stock market efficiency, Journal of Finance 48, 65–91.
, 2001, Profitability of momentum strategies: An evaluation of alternative explanations, Journal of
Finance 56, 699–720.
, 2002, Cross-sectional and time-series determinants of momentum returns, The Review of Financial
Studies 15, 143–157.
Keim, Donald B., 1983, Size related anomalies and stock return seasonality, Journal of Financial Economics
25, 75–97.
Kim, Dongcheol, 1995, The errors in the variables problem in the cross-section of expected stock returns,
Journal of Finance 50, 1605–1634.
, 1997, A reexamination of firm size, book-to-market and earnings price in the cross-section of expected
stock returns, Journal of Financial and Quantitative Analysis 32, 463–489.
Kothari, S.P., Jay Shanken, and Richard G. Sloan, 1995, Another look at the cross-section of expected stock
returns, Journal of Finance 50, 185–224.
Lakonishok, Josef, Andrei Shleifer, and Robert W. Vishny, 1994, Contrarian investment, extrapolation, and
risk, Journal of Finance 49, 1541–1578.
LaPorta, R., Josef Lakonishok, Andrei Shleifer, and Robert W. Vishny, 1997, Good news for value stocks:
further evidence on market efficiency, Journal of Finance 52, 859–874.
Lettau, Martin, and Sydney Ludvigson, 2001, Resurrecting the (c)capm: A cross-sectional test when risk
premia are time-varying, Journal of Political Economy 109, 1238–1287.
Liew, Jimmy, and Maria Vassalou, 2000, Can book-to-market, size and momentum be risk factors that predict
economic growth, Journal of Financial Economics 57.
Lintner, John, 1965a, Security prices, risk and maximal gains from diversification, Journal of Finance 20,
587–615.
, 1965b, The valuation of risk assets and the selection of risky investments in stock portfolios and
capital budgets, Review of Economics and Statistics 47, 13–37.
51
Lo, Andrew W., and A. Craig Mackinlay, 1990, Data-snooping biases in test of financial asset pricing models,
The Review of Financial Studies 3, 431–467.
Merton, Robert, 1971, Optimum consumption and portfolio ruiles in a continuous-time model, Journal of
Economic Theory 3, 373–413.
, 1973, Intertemporal capital asset pricing model, Econometrica 41, 867–887.
Mossin, J., 1966, Equilibrium in a capital asset market, Econometrica 34, 768–783.
Newey, Whitney K., and Kenneth D. West, 1987, Hypothesis testing with efficient method of moments,
International Economic Review 28, 777–787.
Officer, R.R., 1975, Seasonality in australian capital markets, Journal of Financial Economics 2, 29–51.
Reinganum, Mark R., 1981, Misspecification of capital asset pricing: Empirical anomalies based on earnings
yields and market values, Journal of Financial Economics 9, 19–46.
Roll, Richard, and Stephen A. Ross, 1980, An empirical investigation of the arbitrage pricing theory, Journal
of Finance 35, 1073–1103.
Rosenberg, B., K. Reid, and R. Lanstein, 1985, Persuasive evidence of market inefficiency, Journal of Portfolio
Management 11, 9–17.
Ross, Stephen A., 1976, The arbitrage theory of capital asset pricing, Journal of Economic Theory 13, 341–360.
Sharpe, William F., 1964, Capital asset prices: A theory of market equilibrium under conditions of risk,
Journal of Finance 19, 425–442.
Skinner, Douglas J., and Richard G. Sloan, 2002, Earnings surprises, growth expectations and stock returns
of don’t let an earnings torpedo sink your portfolio, Review of Accounting Studies 7, 289–312.
Teo, Melvyn, and Sung-Jun Woo, 2004, Style effects in the cross-section of stock returns, Journal of Financial
Economics 74, 367–398.
Zhang, Lu, 2005, The value premium, Journal of Finance 60, 67–103.
52