Empirical Study of Fama-French Three-Factor Model and Carhart Four-Factor Model in Indonesia
Empirical Study of Fama-French Three-Factor Model and Carhart Four-Factor Model in Indonesia
Empirical Study of Fama-French Three-Factor Model and Carhart Four-Factor Model in Indonesia
ABSTRACT
This research compares the performances of the Fama-French three-factor and Carhart four-
factor models on the Indonesian stock market. It employs ordinary least square (OLS) with
monthly time-series data from July 2005 to June 2015. The results document that the Carhart
four-factor model performs better than Fama-French three-factor model in explaining the
portfolio excess returns in Indonesia. The momentum factor displays a weak effect on the
portfolio excess returns. The results are robust to the equally-weighted method.
Keywords: asset pricing, three-factor model, size, book-to-market, momentum
INTRODUCTION
The Capital Asset Pricing Model (CAPM) introduced by Sharpe (1964), Lintner (1965),
Mossin (1966), and Black (1972) contribute significantly to the understanding of risk
relationships with returns for academia and practitioners. The return of an asset in the CAPM
model is determined only by systematic risk, i.e., beta. The expected return on risk assets is
predicted to be positively related to beta. The main purpose of CAPM is to determine the
required rate of return on an investment. The market equilibrium by Markowitz (1952)
confirms two things, the positive relationship between expected return and beta, and beta as
the only measure of risk.
Initially, empirical tests generally support the argument that beta is the only predictor of
cross sectional differences in stock portfolio returns (Fama and MacBeth, 1973). However,
later empirical findings suggest that not only the beta can explain stock returns, but there are
other factors that can explain stock returns variation and eventually develop other asset pricing
models. A number of studies have found that firm characteristics may be a significant
explanatory factor on average returns, such as firm size (Banz, 1981; Reinganum, 1981),
earnings to price ratio (Basu, 1983), leverage (Bhandari, 1988 ), or book-to-market equity ratio
(Stattman, 1980; Rosenberg et al., 1985; Chan et al., 1991).
Encouraged by the above findings, an article that has a major impact on systematic risk
validity as a measure of stock risk is Fama and French (1992) who use size, leverage, earnings
to price ratio, book-to-market equity ratio and beta. They generate two main results. Firstly,
The three-factor model of Fama-French (1993) states that the excess return portfolio is
explained by three asset pricing factors, i.e., market, size (SMB), and book-to-market (HML).
Market factors represent the difference between market return and risk-free rate. Small Minus
Big (SMB), represents the difference of return on the portfolio of small cap stock shares with
large capitalized portfolio return.
The three-factor model of Fama-French (1993) can be expressed in the following model:
where 𝑅𝑖𝑡 is return of securities or portfolio i for period t, 𝑅𝑓𝑡 is risk free rate for period
t, 𝑅𝑚𝑡 is market portfolio return for period t, 𝑆𝑀𝐵𝑡 is size factor (Small Minus Big) for period
t, 𝐻𝑀𝐿𝑡 is book-to-market factor (High Minus Low) for period t, and 𝑒𝑖𝑡 is error term of
securities or portfolio i of period t.
The Fama-French three-factor model is basically the development of CAPM, by adding
size (SMB) and book-to-market (HML) factors into the CAPM model. SMB is a risk measure
of the company, where the stock of small companies is expected to be more sensitive. On the
other hand, HML represents a higher risk exposure for stocks of firms with high book value-
to-market ratios (value stocks) and lower for firm stocks with low book-to-market ratios
(growth stocks).
Jegadeesh and Titman (1993) document that the momentum strategy generates an
average yield of 1% per month over the next 3-12 months. The momentum strategy is a system
by purchasing stocks or other financial securities that have high returns during the previous 3-
12 months, and sell those stocks or other securities that have low returns for the same period.
In an attempt to capture the momentum yield, Carhart (1997) adds a factor to the Fama-French
(1993) three-factor model.
The four-factor model of Carhart can be expressed in the following equation:
where 𝑅𝑖𝑡 is return of securities or portfolio i for period t, 𝑅𝑓𝑡 is risk free rate for period t, 𝑅𝑚𝑡
is market portfolio return for period t, 𝑆𝑀𝐵𝑡 is size factor (Small Minus Big) for period t, 𝐻𝑀𝐿𝑡
is book-to-market factor (High Minus Low) for period t, 𝑊𝑀𝐿𝑡 is momentum factor (Winner
Minus Loser) for period t, and 𝑒𝑖𝑡 is error term of securities or portfolio i of period t.
Subsequent studies were conducted based on the results of Carhart’s study (1997).
Wermers (1997) and Daniel et al. (2001) find evidence that the fourth factor of Carhart (1997)
can investigate strategies that spur persistence in mutual fund performance. Brav et al. (2000)
document that the four-factor model has the ability to explain underperformance in returns on
the samples of companies conducting an initial public offering (IPO) and seasoned equity
offering (SEO). Kim and Kim (2003) report that a four-factor model can explain the abnormal
pattern of returns after earnings announcements, which are sorted with standardized
unexpected earnings (SUE).
Awwaliyah and Husodo (2011) test the validity of three-factor Fama-French and four
factor Carhart models using US data from January 1963 to December 2010. They find that the
four-factor model of Carhart was better than the Fama-French three-factor model in explaining
the variation of excess return of stock portfolio in the US. Fama and French (2012) test the
effect of size, value, and momentum in 23 developed countries that are divided into 4 regions,
namely North America, Japan, Asia Pacific, and Europe. Fama and French examine the data
RESEARCH METHOD
Data
This study uses data obtained from Datastream databases. The use of the Datastream
database helps with survivorship bias because Datastream samples include active and non-
active companies (Cakici et al., 2013). The research period is July 2005-June 2015. This study
uses monthly data. The data used in this study include the stock closing price, the number of
shares outstanding, the Indonesian capital market Composite Stock Price Index, the book value
of equity, and the risk-free interest rate using monthly data of 90 days of Bank Indonesia
Certificates.
The criteria for determining the sample follow Fama and French (1992, 1993, 2012). This
study excludes stocks that fall within the financial sector. This study excludes stocks in the
financial sector because high leverage is common to financial firms, but that does not apply to
non-financial companies, where high leverage would indicate financial difficulties. This study
also drops stocks with negative equity because negative equity may also indicate that a
company is experiencing financial difficulties.
Research variables
The independent variables in this research are asset pricing factors that include market,
size (SMB), book-to-market (HML), and momentum (WML) factors. The calculation of return
of each asset pricing factor is determined using the value-weighted method. Explanation of
each factor is described as follows.
a. Market Factor (Rm – Rf)
Market factor represents the difference of each market's return on risk-free interest rate.
b. Size factor (SMB)
SMB (Small Minus Big) is intended to illustrate the risk factors associated with firm size.
SMB represents the difference of each month of average returns on three portfolios over
small share capitalization with average returns on three portfolios with large share
capitalization. SMB calculation is formulated as follows.
Empirical model
The empirical models used in this study are the three-factor model Fama-French (1993)
and the four-factor model Carhart (1997), as described in equations 1 and 2 (previous section).
This research uses ordinary least square estimation method (OLS) which adjusted to
Newey and West (1987). Through this standard error adjustment, the OLS method becomes
more robust on issues of heteroscedasticity and autocorrelation. This study uses several criteria
in comparing the Fama-French three-factor model and the four factors of Carhart in Indonesia.
Following Merton (1973), a well-estimated asset pricing model produces an insignificant
intercept. This study tested this by computing the F-statistic of the GRS test (Gibbons, et al.,
1989). The formula for calculating the GRS test is as follows.
𝑇 𝑇−𝑁−𝐿 𝛼′ ∑−1 𝛼
𝐺𝑅𝑆 = (𝑁) ( 𝑇−𝐿−1 ) [1+ 𝜇′ Ώ−1 𝜇] ~ 𝐹(𝑁, 𝑇 − 𝑁 − 𝐿) (6)
where T is the number of observations, N is the number of described portfolios, L is the number
of explanatory factors, α is the vector of the intercept of regression, Σ is the covariance matrix
of the residual of regression, μ is the vector of the mean of explanatory, and Ώ is the covariance
matrix of Explanatory factors. The null hypothesis states that all the regression intercepts are
zero, GRS test statistics has the F distribution with degrees of freedom N and T-N-L.
Following the recommendation of Lewellen et al. (2010), this study also calculates
Sharpe Ratio (SR (α). The formula for calculating Sharpe Ratio is as follows.
where α is the column vector of 25 regression intercepts produced by each model when
applying the Size-B/M portfolio, and S is the covariance matrix of the residual regression.
Smaller Sharpe Ratio values indicate that an asset pricing model is better. This study also
compares the average adjusted R2, the mean absolute value of the intercept, and the average
standard error of the intercept to indicate which model is better.
Descriptive statistics
The statistical summary for each independent variable (asset pricing factor) during the
period of July 2005 - June 2015 (120 observations) is presented in Table 2. The monthly
average of market factor (Rm - Rf) is 0.69%, the average of the size factor (SMB) is 0.06% per
month, the average of the book-to-market (HML) factor per month is 0.01%, and the
momentum factor has an average value per month of 0.03%. All factors have a positive average
value. This indicates that the market, size, book-to-market, and momentum factors have a
premium to compensate for risk.
The correlation between factors is shown in Table 3. The book-to-market and momentum
factors are positively correlated with market factors, while the size factor is negatively
correlated with market factors. The book-to-market and momentum factors are negatively
correlated with the size factor. The momentum factor is positively correlated with the book-to-
market factor.
Table 4 shows the average monthly excess return (the excess return of the Bank Indonesia
Certificates monthly interest rate) for a 25 value-weighted portfolio of independently sorted of
stocks into 5 Size groups and 5 Book-to-Market groups (B/M). In each of the B/M columns of
Table 4, the average excess return generally decreases from the stocks in the small market
capitalization group to the stocks in the large market capitalization group. This shows the size
effect. The average relationship of return with B/M, called the value effect, is only indicated
Stationarity Test
The result of test of stationarity with Augmented Dickey-Fuller (ADF) method at data
level is presented in Table 5. The test result on each research variable shows that all research
variables have t-stat value smaller than critical value. That is, all research variables have been
stationary (do not have root units) on the data level.
10
As shown in Table 8, the GRS statistics on both models are significant at the 1% level,
but the GRS value statistic for the Carhart four-factor model is smaller than the Fama-French
three-factor model. The average adjusted R2 for the Carhart model is greater than the Fama-
French model. The Sharpe ratio on the Carhart model is smaller than the Fama-French model.
The average absolute value of the intercept on the Carhart model is smaller than the Fama-
French model. The average standard errors of intercept for both models generate the same
magnitude of 0.53. Thus, robust test results on the equally-weighted method show that the
Carhart four-factor model is superior in explaining the excess return of the stock portfolios in
Indonesia than the Fama-French three-factor model.
CONCLUSION
This study compares the performance of Fama-French three-factor model and Carhart
four-factor model in explaining the excess return of stock portfolio in Indonesia during the
period of July 2005 to June 2015. To do so, this study compares GRS statistic, Sharpe ratio,
the average adjusted R2, the mean absolute value of the intercept, and the average standard
error of intercept on both models. This study uses time-series regression with monthly data
frequency.
The findings show that the Carhart-four factor model is better in explaining the excess
return of stock portfolios in Indonesia than the Fama-French three-factor model. This finding
supports Awwaliyah and Husodo (2011) using US stock market data. This result is robust when
tested using equally method in calculating asset pricing factor and excess return portfolio.
Although the four-factor model of Carhart is superior to the Fama-French three-factor model,
the results suggest that there are other factors to consider in determining asset pricing models
that better capture stock return variations in the Indonesian stock market.
Future studies may use other proxies of risk-free interest rates, such as government bond
yields with the tenure of less than one year. Further research can also compute J-statistics to
see the significance of the regression intercept with generalized method of moments (GMM)
technique, given the assumptions in GRS statistics that cannot be held identical in the
practitioner's world, independent, and normal distributed asset returns (Cochrane, 2005:234).
11
12
13
14