Stimation of Expected Return - CAPM vs. Fama and French

International Review of Financial Analysis 14 (2005) 407 – 427
Estimation of expected return:

CAPM vs. Fama and French
Jan Bartholdy*, Paula Peare
Aarhus School of Business, Fuglesangs Allé 4, 8210 Aarhus V, Denmark
Accepted 22 October 2004

Available online 24 November 2004
Abstract
Most practitioners favour a one-factor model (CAPM) when estimating expected return for an
individual stock. For estimation of portfolio returns, academics recommend the Fama and French
three-factor model. The main objective of this paper is to compare the performance of these two models
for individual stocks. First, estimates for individual stock returns based on CAPM are obtained using
different time frames, data frequencies, and indexes. It is found that 5 years of monthly data and an
equal-weighted index, as opposed to the commonly recommended value-weighted index, provide the
best estimate. However, performance of the model is very poor; it explains on average 3% of
differences in returns. Then, estimates for individual stock returns are obtained based on the Fama and
French model using 5 years of monthly data. This model, however, does not do much better;
independent of the index used, it explains on average 5% of differences in returns. These results
therefore bring into question the use of either model for estimation of individual expected stock returns.
D 2004 Elsevier Inc. All rights reserved.
JEL classification: G11; G12; G31

Keywords: Capital Asset Pricing Model; Fama and French three-factor model; Beta estimation
A commercial provider of betas once told the authors that his firm, and others, did
not know what the right period was, but they all decided to use five years in order to
reduce the apparent differences between various services’ betas, because large
differences reduced everyone’s credibility! (Brigham & Gapenski, 1997, p. 354,
footnote 9.)
* Corresponding author.
E-mail addresses: jby@asb.dk (J. Bartholdy)8 ppe@asb.dk (P. Peare).
1057-5219/$ - see front matter D 2004 Elsevier Inc. All rights reserved.
doi:10.1016/j.irfa.2004.10.009
408 J. Bartholdy, P. Peare / International Review of Financial Analysis 14 (2005) 407–427
1. Introduction
Estimation of expected return or cost of equity for individual stocks is central to many
financial decisions such as those relating to portfolio management, capital budgeting, and
performance evaluation. The two main alternatives available for this purpose are a single-
factor model (or Capital Asset Pricing Model [CAPM]) and the three-factor model
suggested by Fama and French (1992, for example).1 Despite a large body of evidence in
the academic literature in favour of the Fama and French model, for estimation of portfolio
returns, practitioners seem to prefer CAPM for estimating cost of equity (see, for example,
Bruner, Eades, Harris, & Higgins, 1998; Graham & Harvey, 2001). The main objective of
this paper is therefore to compare the performance of the Fama French model with that of
CAPM for individual stocks.
The view taken in this paper, therefore, is that of a firm estimating its cost of equity. It is
assumed that if estimation is based on CAPM, then an estimate for beta is obtained using a
simple OLS regression, and this estimate is multiplied by an estimate for the risk premium
on the market to obtain an estimate for excess return on equity.2 If estimation is based on
Fama French, then an estimate for the beta for each factor is obtained, also using a simple
OLS regression, and these estimates are multiplied by the risk premium for the relevant
factor to obtain an estimate for cost of equity. That is, for both CAPM and Fama French, it
is assumed that an estimate for cost of equity is obtained using a simple estimation
technique, in particular, in relation to the amount of data required for estimation. For the
method described here, the only data requirements are the return on a market index and the
return on the stock, over the estimation period, if CAPM is used. If Fama French is used,
then data for the additional two factors are also required. There are a variety of different
methods available to improve the estimates of beta and for implementing the two models;
however, all of these methods require returns on a large sample of stocks, to form
portfolios for example. This amount of data is only available at significant expense. This
paper therefore compares the performance of the CAPM with the Fama French model
under the assumption that a simple estimation technique, in particular in relation to the
amount of data required, is used. That is, comparison is made from the point of view of the
practitioner.3
1
We are aware that it is not possible to estimate CAPM since the world market portfolio is not observable; that
what is in fact estimated is a single-factor model. This is, however, referred to as bestimation based on CAPMQ
throughout the paper as this is consistent with common usage.
2
From Bartholdy and Peare (2003), this procedure results in a biased estimate for cost of equity. This, however,
is not the focus of this paper. The focus here is on the bqualityQ of the beta estimate(s) obtained from CAPM and
Fama French, for estimation of cost of equity, when a very simple estimation technique is used. Of course, in this
paper, the unbiased method recommended in Bartholdy and Peare (2003) is used for evaluation of beta estimates.
Further, this technique, however, does not involve additional costs in terms of econometric sophistication or data
required.
3
The problems associated with this approach, such as measurement errors in the various explanatory variables,
are well documented. We know that this simple technique is not the most efficient from an econometric point of
view, although it may be from an economic point. The results in this paper should be interpreted as a test of
CAPM and Fama and French, given the simple estimation techniques available to companies, and not as a formal
econometric test of CAPM against the Fama French model.
J. Bartholdy, P. Peare / International Review of Financial Analysis 14 (2005) 407–427 409
To compare the performance of CAPM and the Fama and French model in this context,
we need to obtain estimates for expected return based on each of these models. To make a
fair comparison, we need the bbestQ possible estimate for expected return in each case.
What do we mean by bbestQ? Since the view taken here is that of the practitioner who
needs an estimate for cost of equity, a financial manager making a capital budgeting
decision for example, the objective here is to find the model and data that provide the
bbestQ estimate for next year’s return, using a simple estimation approach. Now, the R 2
from a cross-section regression using individual 1 year stock returns as the dependent
variable and estimated factor(s) based on past returns as the explanatory variable(s)
measures how much of the differences in individual stock returns is explained by the
estimation procedure. By bbest,Q therefore, we mean the model and data that result in the
highest R 2 when a simple estimation procedure is used.
Despite the existence of a large academic literature which discusses implementation of
CAPM, in particular in relation to estimation of the key parameter beta, there is no
consensus in relation to how a best estimate should be obtained. There is no consensus
with respect to the index, time frame, and data frequency that should be used for
estimation. Previous research has focused on reasons for differences in estimated betas
between periods and the ability of historical betas to predict future betas. See for example,
Blume (1975), Carleton and Lakonishok (1985), Klemkosky and Martin (1975), and
Reilly and Wright (1988). As illustrated by the quote at the beginning of this paper, the
situation is no better among professional beta providers.4 This lack of consensus manifests
itself in different beta estimates for the same company by different beta providers. Bruner
et al. (1998), for example, found the average beta for a small sample of stocks to be 1.03
using betas provided by Bloomberg, whereas using Value Line betas, it was 1.24. A
difference of this magnitude results in significantly different expected returns (costs of
equity) for individual companies leading potentially to conflicting financial decisions, in
capital budgeting for example. The second objective of this paper is therefore to find the
best index, time frame, and data frequency for estimating beta, and therefore expected
return, based on CAPM. Two other issues are also examined; whether or not dividends
should be included in the returns and whether raw returns or excess returns should be used
when estimating beta.
The results obtained suggest that, for estimation of beta, 5 years of monthly data are in
fact the appropriate time period and data frequency. However, it is also found that an equal-
weighted index, as opposed to the commonly recommended value-weighted index,
provides a better estimate. It does not appear to matter whether dividends are included in the
index or not or whether raw returns or excess returns are used in the regression equation.
Richard Roll pointed out in his presidential address to the American Finance Association
(Roll, 1988) that the general performance of bbetaQ in explaining portfolio returns is not
great. As discussed above for many practical applications, it is individual returns that are
relevant, so it is pertinent to ask how well (or poorly!) CAPM explains returns on individual
stocks. From the results obtained here, the answer is once again not great. The bbestQ
4
See Reilly and Wright (1988) for a discussion of Merill Lynch’s betas. A discussion of Ibbotson Associates’
Beta Book can be found at the following address: www.ibbotson.com.
estimates, namely, those obtained using 5 years of monthly data and an equal-weighted
index are, on average, only able to explain about 3% of differences in returns on individual
stocks. It is therefore surprising that CAPM is used at all by practitioners.
The main alternative to CAPM and the one academics recommend, at least for
estimation of portfolio returns, is the three-factor model suggested by Fama and French
(1992, 1993). In this model, size and book to market factors are included, in addition to a
market index, as explanatory variables. As discussed above, this model is not popular
among practitioners. The question is, why? In an attempt to answer this question, the
performance of the three-factor model is compared with that of CAPM. Using 5 years of
monthly data, it is found that the Fama French model is at best able to explain, on average,
5% of differences in returns on individual stocks, independent of the index used. Such a
small gain in explanatory power probably does not justify the extra work involved in
including two more factors.
The remainder of the paper is organised as follows. In Section 2, various issues
associated with estimation of expected return based on CAPM are discussed. The Fama
and French three-factor model is presented in Section 3. A description of the data used for
analysis is provided in Section 4. The results obtained from estimation based on CAPM
are presented in Section 5 and those from estimation based on Fama and French in Section
6. Section 7 concludes the paper.
2. Estimation issues for CAPM
In this section, a brief discussion of issues relating to estimation of expected return

based on CAPM is provided. According to CAPM, the expected excess return (expected
return minus the risk-free rate) on an asset is i given by:

E ri rf ¼ bM
i ð E ½ rm rf Þ ð1Þ
where is r i the return on asset i, r f is the risk-free return, r m is the return on the world
market portfolio, bM 2
i =Cov(r i ,r f)/r M is the systematic risk of asset i relative to the world
2
market portfolio (beta), and r M is the variance of the return on the world market portfolio.
Now, the world market portfolio, which consists of all assets in the world, is not
observable, so it is necessary to use a proxy. An estimate for bM i therefore is typically
obtained by running the following time series regression:
rit ¼ ai þ bIi rIt þ eit t ¼ 1; N ; t0 : ð2aÞ
where r It is the return (at time t) on the index I used as a proxy for the world market
portfolio, bIi =Cov(r i ,r I )/r 2I is the systematic risk of asset i relative to the index, r 2I is the
variance of the index, and e it is a white noise error term. As discussed in the Introduction,
little research has been done concerning what index, data frequency, and time frame
provide the bbestQ estimate, so one of the objectives of this paper is to address this
shortcoming. In addition, the issues of whether or not dividends should be included in the
index used and whether raw or excess returns should be used are addressed. In summary,
therefore, the following issues in relation to estimation of expected return for individual
stocks based on CAPM are addressed in this paper:
! Do dividend adjustments in the index matter?

! Should excess returns or raw returns be used?
! What data frequency and time period should be used?
! What index should be used?
Before discussing each of these issues in more detail, two important estimation issues
not dealt with in this paper are clarified: thin trading and mean reversion in beta. Before
obtaining best estimates for thinly traded stocks, we need to find the best way of
estimating beta for frequently traded stocks. This is because we need to find the best
estimation procedure for the most basic situation before we can deal with complexities.
Thus, only frequently traded stocks are included in the sample analysed in this paper. With
respect to mean reversion in beta (the tendency for a high beta to be followed by smaller
beta), the method suggested by Blume (1975) for dealing with this is, in fact, implicit in
the procedure used here.5 None of the other methods designed to deal with mean reversion
are considered. Again, this is based on the argument that it is necessary to find the
benchmark first before starting to compare alternative improvements.
2.1. Dividends
CAPM assumes dividends are included in the returns on the world market portfolio.
Given the number of indexes constructed without dividends, it is relevant to ask if it
matters, in relation to obtaining a bbestQ estimate, whether or not dividends are included in
the index used. Therefore, where possible, indexes with and without dividends are
considered here.
2.2. Excess returns or raw returns
As indicated above, the time series regression Eq. (2a) is commonly used to obtain an
estimate for beta. A more general formulation for Eq. (2a) is given by:
rit rf t ¼ ai þ bIi ðrIt rf t Þ þ eit t ¼ 1; N ; t0 : ð2bÞ
Eq. (2a) can be obtained from Eq. (2b) by assuming the risk-free rate is constant over
the estimation period. Whether it is more appropriate to use Eq. (2a) or Eq. (2b) for
estimation is an empirical question. Therefore, both formulations are examined here.
2.3. Data frequency and time period
In general, for estimation, the more observations, the better. This suggests using as long
a time period as possible. With a long estimation period for beta, however, it is likely that
5
See Footnote 12.
the value of the true beta changes over the period. The resulting estimate for beta will
therefore be biased. This pulls us in the direction of shortening the period. One way of
obtaining more observations, over a shorter time period, is to increase the sampling
frequency. However, moving from monthly to daily returns, for example, results in an
increase in the amount of noise in the data, which reduces the efficiency of the estimates.
Thus, there is a trade off between the length of the time period and sampling frequency. In
this paper, therefore, the performance of monthly data for 5 years (the standard data
frequency and time period used), weekly data for 2 years, and daily data for 1 year are
examined.6
2.4. Index
The underlying theory for CAPM is quite specific in its recommendation of index;
it specifies that a value-weighted index consisting of all the assets in the world should
be used. Since only a small fraction of all assets in the world trade on stock
exchanges, it is impossible to construct such an index, so a proxy must be used
instead. The most commonly used proxy is the value-weighted Standard and Poor’s
Composite Index.7 One can also use an equal-weighted index. The question is, which
gives the best estimate? In this paper, six standard indexes are considered. In addition, a
seventh index, an Economy Index, is developed in an attempt to construct an alternative
proxy that is more closely related to the world market index than the standard indexes
used.
In relation to construction of a proxy that is closely related to the world market
index, consider for a moment what a world market index would look like in a simple
world where all assets are traded in a frictionless market. The weight for an asset in
the index would be the value of the asset divided by the total value of all assets in
the world. These weights are, of course, not observable, but they can be proxied
using the National Income Accounts. From these accounts, it is possible to obtain the
value of goods and services produced in each sector of the economy. As long as the
value of the goods is correlated with the underlying value of the assets, it is possible
to use each sector’s share of Gross Domestic Product (GDP) as a proxy for the
weights in the world market portfolio.8 The return on assets within each sector can be
proxied by the return on an equal-weighted index of the stocks for each sector trading on
the stock exchange. Such an Economy Index can differ from a standard value-weighted
6
A related issue discussed in the literature is the bInterval EffectQ. Hawawini (1983), for example, suggests that
betas are affected by the interval used to calculate the returns due to thin trading (proxied by size of the firm).
Reilly and Wright (1988) found a relationship between beta and size of the firm. The sample used in this paper is
restricted to stocks that traded on more than 95% of the days in the relevant subperiod (see Section 3 for further
discussion), thus avoiding this type of bInterval EffectQ.
7
Merill Lynch uses the Standard and Poor 500 Composite Index and Value Line uses the NYSE Composite
Index. Reilly and Wright (1988) found no difference in estimated betas based on these two indexes. Therefore, the
NYSE Composite Index is not included in the analysis here.
8
In theory it is possible to obtain the GDP accounts for all countries in the world and corresponding weights.
For this paper, however, only US data is used due to resource restrictions.
stock index. For example, firms with severe asymmetric information problems are funded
by means other than the stock market, e.g., banks, whereas firms with fewer asymmetric
problems can raise funds on a stock exchange. These problems are more profound in some
industries than in others. This results in some industries having a smaller share of the
capitalisation on the stock exchange than their share in the economy. As a result, these
industries are under represented in a value-weighted stock index relative to their
representation in the economy. Also, the companies listed on a stock exchange tend to be
the larger firms in the economy. This results in under representation of smaller firms in a
value-weighted stock index relative to their representation in the economy. Construction of
an Economy Index, such as the one developed here, is an attempt to deal with these
problems.9
Whatever index is used it is a proxy for the world market portfolio. The question is,
which proxy provides the best estimate for beta, and therefore expected return? In this
paper, the following seven proxies are considered: the Economy Index (described above),
the Standard and Poor’s Composite Index (value-weighted) since it is the most commonly
used, CRSP equal and value-weighted indexes, with and without dividends, representing
broad based domestic indexes, and the Morgan Stanley Capital World Index (value-
weighted) since this represents an alternative attempt to construct a proxy for the world
market portfolio.
2.5. Evaluation
Finally, it is necessary to decide how bbestQ should be measured. Fama and MacBeth
(1974) propose the following estimation procedure. First, for each stock in the sample, an
ˆ
estimate for beta, bIit , is obtained based on either Eqs. (2a) or (2b) (using 5 years of
monthly data, for example). Then, for all stocks in the sample, an estimate for the risk
ˆ
premium, c 1t+1, in the subsequent year is obtained using bIit as the explanatory variable in
the following cross-section regression:
ˆ
ritþ1 rf tþ1 ¼ c0tþ1 þ c1tþ1 bIit þ eitþ1 i ¼ 1; N ; N ð3Þ
Regression Eq. (3) measures how much of the variation in the next year’s stock returns
beta is able to explain.10 The data frequency for the left hand side is annual data,
independent of the data frequency used for estimation of beta, as this reflects the general
application of the model by practitioners where historical data are used to obtain an
9
The single-factor version of the Arbitrage Pricing Theory (APT; Ross, 1976) provides a possible justification
for the Economy index. One interpretation of APT is that the single factor included in the model is an economic
state variable. Construction of an index that represents the economy allows this to be captured. The consumption
based CAPM also provides a possible justification. If consumption is related to the state of the economy then an
index representing the economy can be used as a proxy for consumption and estimation of beta can be based on
this index.
10
For a discussion of the general problems when using the two-stage procedure to test for Asset Pricing Models,
see for example Shanken (1992) and Kan and Zhang (1999).
estimate of expected annual return for the next year.11 The process is then repeated to
obtain estimates for expected returns in subsequent years, for the entire sample period.
This procedure has also been used in previous academic studies. The objective of these
studies, however, is typically to test different factor models for the existence of anomalies
in stock returns. For this purpose, portfolios are used for estimation of Eq. (3) to avoid
measurement error problems. The use of portfolios, however, is not appropriate here since
the objective is to obtain an estimate for expected return or cost of equity for an individual
company.12
Two evaluation criteria result from the procedure described above. The first P criterion
is the significance of the average estimated value obtained for c1 ; c1 ¼ T1 Ttˆ c1tþ1 , the
average estimated risk premium for the entire sample period. If this value is not
significant and positive, then beta is not able to explain the excess return on the left-
hand side. Thus, significance of this value is a necessary condition for the model to be
of any use. The second criterion is the average R 2 from Eq. (3). This is a direct
measure of the ability of the beta estimate to explain differences in returns on
individual stocks in the period following estimation. As discussed in the Introduction,
previous research in this area has focused on reasons for differences in estimated betas
between periods and the ability of historical betas to predict future betas. What is
important for both practitioners and academics, however, is not so much an estimated
beta’s ability to predict next period’s beta, but its ability to explain next period’s returns
and how much of the differences in stock returns can be explained by differences in their
betas. Also, from an academic point of view, it is important that the beta used in tests of
CAPM and various market anomalies explain as much as possible of expected return.
Otherwise, CAPM may be incorrectly rejected in favour of other models or anomalies, or
beta may even be incorrectly declared dead. Thus, the R 2 from Eq. (3) is used here to rank
the different estimates obtained. Using R 2 as a yardstick for CAPM follows Roll (1988).
The significance of c1 is also noted as this provides a necessary condition for the model to
be of any use.
In summary, to obtain estimates for expected return based on CAPM, different beta
estimates are obtained using 5 years of monthly data (the standard frequency used), 2
years of weekly data, and 1 year of daily data. Seven different indexes are considered as
potential proxies for the world market portfolio. In addition, the issues of whether or not
dividends should be included in the index, and whether or not excess or raw returns
11
The results obtained using monthly returns as the dependent variable are similar and so are not included
here.
12
This approach implicitly makes Blume’s adjustment for mean reversion in betas. The first step in Blume’s
method is to regress the current period’s beta on previous period’s beta across all stocks in the sample. This step
provides for a common model for all stocks to adjust for mean reversion in betas. In the second step, a standard
time series regression is used to estimate beta which is then inserted into the model estimated in step one to obtain
the predicted beta for next period, which in this paper is used for the cross-section regressions (3). The predicted
or adjusted beta used in the cross-section regression is therefore a linear transformation of the estimated beta from
the time series. Therefore, when running the cross-section regression, the slope coefficient in Eq. (3), c 1,
embodies an estimate of the parameters from Blume’s original regression—this is equivalent to changing an
explanatory variable in a regression from being measured in dollars to being measured in cents, the consequence
is to change the size of the estimated parameter by a factor 100.
should be used when estimating beta, are examined. Finally, bbestQ is measured by the
average R 2 value obtained from Eq. (3). Significance of the average estimated value for
c 1 is also noted since a necessary condition for the model to be of any use is that this value
is significant.
3. The Fama and French model
The three-factor model suggested by Fama and French (1992, for example) provides an
alternative to CAPM for estimation of expected return. In this model, two additional
factors are included to explain excess return; size and the book to market ratio. Thus, for
each stock, i, to estimate excess return, first beta estimates for each of the factors are
obtained from the following time series regression:
rit rf t ¼ ai þ b1i ðrI t rf t Þ þ b2i SMBt þ b3i HMLt þ eit t ¼ 1; N ; t0 : ð4Þ
where SMB is the return on a portfolio of small stocks minus the return on a portfolio of
large stocks, and HML is the return on a portfolio of stocks with high book to market
values minus the return on a portfolio of stocks with low book to market values. The
time period and data frequency found to provide the best estimate based on CAPM are
used, and estimates for the betas obtained, based on different indexes, are examined.
Then, for each stock in the sample, an i estimate for the risk premium, for each factor, in
the subsequent year is obtained using the beta estimates from Eq. (4) ˆ b1it ;ˆ
b2it ;ˆ
b3it as
the explanatory variables in the following cross-section regression:
ritþ1 rf tþ1 ¼ c0;tþ1 þ c1tþ1ˆ

b1it þ c2tþ1ˆ
b2it þ c3tþ1ˆ
b3it þ eitþ1 i ¼ 1; N ; N : ð5Þ
As for CAPM, the process is rolled forward, the resulting average R 2 is used to
evaluate the estimates
P obtained, and significance of the average estimated value for the
coefficients,
cj ¼ T1 Tt ˆ
cjtþ1 ; j ¼ 1;2;3; is noted. The use of monthly data may bias the
results against the Fama and French model. However, since monthly data are only
marginally better than daily and weekly data in the case of CAPM and the market beta
is also in the Fama and French model, it is unlikely that using monthly data will
significantly bias the results against the Fama and French model. Further monthly data
is commonly used in the literature.
4. Data
Daily adjusted prices were extracted from the CRSP tapes from 1970 to 1996. Daily
returns were calculated as simple holding period rates of return between days. Weekly
returns were calculated from Wednesday to Wednesday to avoid any contaminating effects
from weekends and Mondays, and end of month prices were used for calculating monthly
returns. Daily, weekly, and monthly yields on 3 month T-bills were used for the risk-free rate
for the time series regression, and the yield on 12 month T-bills for the dependent variable in
the cross-section regressions. T-bill yields were calculated at the beginning of the period; for
example, for the risk-free rate on Thursday, the closing price/yield on the T-bill on
Wednesday was used.
As discussed in Section 2, the estimation procedure based on CAPM involves first
obtaining an estimate for beta using either Eq. (2a) or (2b) and then using the resulting
estimate to obtain an estimate for expected return using Eq. (3). Thus, if 5 years of
monthly data is used, then for each stock, first an estimate for beta is obtained using the
years 1970 to 1974. Then, the resulting estimate is used in Eq. (3) to obtain an estimate for
expected return in 1975. If estimation is based on 2 years of weekly data, then 1973 and
1974 are used to obtain an estimate for expected return in 1975, and if 1 year of daily data
is used, then 1974 is used for beta. Thus, the sample of stocks used for monthly, weekly,
and daily data is the same. The process is then repeated to obtain estimates of expected
return for subsequent years (1976 to 1997). The same procedure is used to obtain estimates
for expected return based on the Fama and French model, using Eqs. (4) and (5) in place of
Eqs. (2a) or (2b) and (3), respectively.
Also, as discussed in Section 2, to avoid problems which result from thin trading, and
to keep the analysis manageable, only stocks which traded on more than 95% of the
days in the estimation period are included. From the above description of the estimation
procedure, it follows that, for a stock to be included in the sample, it must have traded
on more than 95% of the days over a 6 year period. For example, for an estimate of
expected return in 1975, to be included in the sample, the stock must have traded on
more than 95% of the days over the period 1970 to 1975. The number of stocks in the
resulting sample ranges from a low of 780 in 1975 to a high of 1308 in 1994.
Finally, from Section 2, seven indexes are considered as potential proxies for the
world market portfolio: the four CRSP indexes (value and equal-weighted, with and
without dividends), Standard and Poor’s Composite Index which is value-weighted and
does not include dividends, the Morgan Stanley Capital World Index which is also
value-weighted and does not include dividends, and an Economy Index which does
include dividends. The CRSP indexes were obtained from the CRSP tapes. Standard
and Poor’s Composite Index and Morgan Stanley Capital World Index were retrieved
from Datastream. The Morgan Stanley Index is only available on a monthly basis for
the whole period. The Economy Index was constructed in the following manner. The
SIC code was obtained for each stock in the sample from the CRSP tape, and an
equal-weighted index was calculated for each two digit SIC code. The GDP by
industry in current dollars for the period 1970 to 1996 was obtained from the Bureau
of Economic Analysis. The industries in the GDP Product Table were matched with
the SIC portfolios generated from the CRSP sample. The weight assigned to each SIC
portfolio is total GDP for the associated industry divided by total private GDP.13
Data for SMB and HML in regression Eq. (4) were obtained from K. French’s
homepage.14
13
GDP has a private and a government component. To ensure that the weights add up to one only, the private
component of the GDP is included.
14
mba.tuck.dartmouth.edu/pages/faculty/ken.french/.
Table 1
Summary statistics
Index Number of Mean Standard Minimum Maximum
observations deviation
Economy Index 323 1.659 6.1902 26.306 32.3537
Standard and Poor 323 0.7626 4.3578 21.763 16.3047
Morgan Stanley 323 0.7493 4.0825 17.1242 14.2656
Value-weighted including dividends 323 1.0619 4.496 22.5006 16.5798
Value-weighted excluding dividends 323 0.7499 4.4798 22.6758 16.2516
Equal-weighted including dividends 323 1.845 5.7023 25.0866 30.2876
Equal-weighted excluding dividends 323 1.653 5.6975 25.1587 30.0282
Summary statistics are for monthly returns on the indexes expressed as percentages. Economy Index refers to the
index constructed in this paper. Standard and Poor refers to the Standard and Poor’s Composite Index, and
Morgan Stanley refers to the Morgan Stanley World Market Capital Index. The equal and value-weighted indexes
refer to the indexes included on the CRSP tape, with and without dividends.
Summary statistics for the monthly returns, for each of the indexes, are provided in
Table 1. The CRSP equal-weighted index, including dividends, has the largest average
monthly return (1.845%). This is followed by the Economy Index (1.659%), and the
CRSP equal-weighted index, excluding dividends, is not all that different (1.653%).
The returns on the other indexes, which are all value-weighted, are significantly lower.
This difference in returns is probably due to the higher weight given to large firms in
value-weighted indexes and the exclusion of dividends.15
5. Empirical analysis—CAPM
In this section, it is first determined whether or not dividend adjustments to the index
used matter and whether raw or excess returns should be used for estimation. It is then
determined which data frequency, time period, and index provides the best estimate for
expected return.
5.1. Dividends
Table 2 provides the correlation coefficients between the different indexes. For both
the equal-weighted and the value-weighted CRSP index, the correlation between the
indexes with and without dividends is 0.999; the standard deviations for these two
indexes are also very close (see Table 1). This suggests that, for estimation of beta, it is
irrelevant whether or not dividend adjusted indexes are used. Therefore, in the interest of
saving space, for the remainder of the analysis in this paper, only the CRSP indexes with
dividends are used.16
15
See for example Banz (1981) for a discussion of the small firm effect; that large firms, in general, have lower
returns than smaller firms.
16
Note if the models are used for performance evaluation, then dividends should be included in the index since
the intercept term is affected by inclusion/exclusion of dividends.
Table 2
Pearson correlation coefficients for the indexes
Economy S&P MSCI CRSP equal-weighted CRSP value-weighted
Including Excluding Including Excluding
dividend dividend dividend dividend
Economy Index 1 0.8244 0.7031 0.8845 0.8844 0.9732 0.9721
Standard and Poor 1 0.826 0.9883 0.9897 0.785 0.7851
Morgan Stanley 1 0.8216 0.8236 0.6853 0.686
Value-weighted 1 0.9993 0.8485 0.848
including dividends
Value-weighted 1 0.8506 0.8504
excluding dividends
Equal-weighted 1 0.9999
including dividends
Equal-weighted 1
excluding dividends
The Pearson correlation coefficients are calculated based on monthly returns for each of the indexes. Economy
Index refers to the index constructed in this paper. Standard and Poor (and S&P) refers to the Standard and Poor’s
Composite Index and Morgan Stanley (and MSCI) refers to the Morgan Stanley World Market Capital Index. The
equal and value-weighted indexes refer to the indexes included on the CRSP tape, with and without dividends.
5.2. Excess returns or raw returns
From Section 2, to address the issue of whether excess returns or raw returns should be
used for estimation, for each stock in the sample, the following two regression equations
are examined:
rit ¼ ai þ bIi rIt þ eit t ¼ 1; N ; t0 ð2aÞ
rit rf t ¼ ai þ bIi ðrI t rf t Þ þ eit t ¼ 1; N ; t0 ð2bÞ
With the exception of the Morgan Stanley Index for which only monthly data is
available, this was done for each of the indexes examined, using 1 year of daily data, 2
years of weekly data, and 5 years of monthly data, over the entire sample period. Thus, for
each stock in the sample for 1975, for each of the indexes examined, three beta estimates
were obtained; one based on daily data from 1974, one based on weekly data from 1973
and 1974, and one based on monthly data from 1970 to 1974. The process was then
repeated for each year from 1976 to 1997. For the Morgan Stanley Index, only beta
estimates based on 5 years of monthly data were obtained, as this was the only data
available. This process resulted in 22,905 beta estimates for each index, time period, and
data frequency. The only difference between the above two equations is that, with Eq. (2a),
raw returns on the asset and the index are used for estimation, whereas with Eq. (2b),
excess returns are used. The correlation between the betas resulting from estimation of (2a)
and (2b) is 0.999, and the actual beta estimates only differ at the third decimal point.17 This
17
To save space, these results are not reported here but are available upon request.
Table 3
Summary statistics for beta estimates
Index Number Average beta Standard deviation Minimum Maximum
Daily data
Standard and Poor 22.905 0.769 0.4443 4.4319 3.1315
Economy 22.905 0.9515 0.4868 3.9077 7.0235
CRSP value-weighted 22.905 0.894 0.4905 5.1519 3.4304
CRSP equal-weighted 22.905 1.1758 0.609 4.5076 6.2057
Weekly data
Economy 22.905 0.9299 0.4619 1.0759 6.6588
Monthly data
Morgan Stanley 22.905 0.9765 0.5133 1.5106 3.8294
Economy 22.905 0.8594 0.4239 0.556 5.8326
For each stock in the sample, beta is estimated using the following time series regression:
rit rf t ¼ ai þ bIi ðrIt rf t Þ þ eit t ¼ 1; N ; t0 : ð2bÞ
With the exception of the Morgan Stanley Index, this is done for each of the indexes using 1 year of daily data,
2 years of weekly data, and 5 years of monthly data for each year of the sample. The result is 22,905 beta
estimates for each index, time period, and data frequency. Only beta estimates based on 5 years of monthly data
were obtained using the Morgan Stanley Index as this was the only data available.
suggests that either raw or excess returns can be used when estimating beta. Since the
theoretical model (1) lends itself to expression in terms of excess returns, again in the
interest of saving space, only excess returns are used in the subsequent analysis.
5.3. Data frequency, time period, and index
From Table 3, estimation using 1 year of daily data, 2 years of weekly data, and 5 years
of monthly data, yields a wide range of average betas, from a low of 0.77 to a high of 1.18.
Since a total of 22,905 betas were estimated for each index, time period, and data
frequency, this is a significant difference. Using Standard and Poor’s Composite Index, the
average beta for daily returns is 0.77 increasing to 1.10 when monthly returns are used.
The situation is similar for the CRSP value-weighted index; 0.89 for daily data and 1.12
for monthly data. For the CRSP equal-weighted index, however, the largest average beta is
obtained using daily returns (1.18) and the smallest using monthly returns (0.92). Similarly
for the Economy Index, daily data give 0.95 and monthly data give 0.86.18 Analysis for the
18
This suggests that small and large firms betas react differently to differences in data frequency since the
difference between an equal weighted index and a value-weighted index is the relatively large weight given to
smaller firms in the equal-weighted index. That is, it suggests that, for small firms, the beta estimates from daily
returns are larger than those from monthly returns, whereas for large firms, it is the opposite.
Morgan Stanley Index is not possible since only monthly returns are available for the
whole period.
Table 4 contains the correlation coefficients for the betas estimates obtained using
Eq. (2b) for different indexes and different data frequencies. Within each data frequency,
the correlation between betas estimated using the CRSP value-weighted index and the
Standard and Poor’s Composite Index is very high (0.99 approximately). The
correlation between the betas estimated using the CRSP equal-weighted index and
the Economy Index is also high within each data frequency (between 0.96 and 0.99).
For all indexes, the correlation across data frequencies is relatively low. For example,
for the Standard and Poor’s Composite Index, the correlation between betas estimated
using daily data and monthly data is only 0.52 and 0.69 between daily and weekly
returns. This correlation drops even further for betas estimated using different indexes
and frequencies. For example, the correlation between betas estimated using monthly
data for the CRSP equal-weighted index and daily data for the Standard and Poor’s
Composite Index is only 0.34. Thus, there are large differences in estimated betas
depending on the index, data frequency, and time period used. It is therefore relevant
to ask which of these estimates provide the best estimate for estimation of expected
return.
5.4. Evaluation
From Section 2, the average R 2 value from the cross-section regression (3) is used to
rank the beta estimates
P obtained from Eq. (2b) and significance of the average estimated
value for c1 ;
c1 ¼ T1 Tt ˆ
c1tþ1 , is noted because significance of this value is a necessary
condition for the model to be of any use.
The values obtained for c1 , from Eq. (3), are provided in Table 5.19 For the Economy

Index, c1 is found to be significant if either weekly data or monthly data are used. For the
CRSP equal-weighted, c1 is only significant if monthly data are used. The highest
significant estimate is obtained if monthly data and the Economy Index are used (0.0786)
followed by monthly data and the CRSP equal-weighted (0.0703); weekly data and the
Economy Index gives 0.0679. For the other indexes, c1 is not significant. Thus, based on
significance of the average estimated value for c 1, beta can only explain excess return if 5
years of monthly data are used together with the equal-weighted CRSP Index or the
Economy Index or if 2 years of weekly data is used together with the Economy Index.
R 2 values obtained from estimating Eq. (3) using annual data for the dependent variable
are reported in Table 6. For most indexes, moving from daily to monthly data tends to
result in an increase in R 2. For all data frequencies, the Economy Index and the CRSP
equal-weighted index give the highest average R 2 values. For both of these indexes, the
highest value is obtained if 5 years of monthly data is used; 0.0271 for the Economy Index
and 0.0296 for the CRSP equal-weighted index. From the R 2 values therefore, 5 years of
PT
19
The average estimated values obtained for c0 c0 ¼ T1 ˆ are also reported in this table for
t c0tþ1
completeness. With one exception they are all insignificant; which is consistent with CAPM.
Table 4
Correlation of beta estimates across time period, data frequency, and indexes
J. Bartholdy, P. Peare / International Review of Financial Analysis 14 (2005) 407–427

Daily returns Weekly returns Monthly returns
S&P Economy Value Equal S&P Economy Value Equal S&P Economy Value Equal Morgan
weight weight weight weight weight weight Stanley
Daily returns
Standard and Poor 1 0.8348 0.9875 0.7867 0.6923 0.5781 0.6816 0.555 0.5172 0.3845 0.5052 0.3448 0.4206
Economy 1 0.8929 0.9608 0.6954 0.7415 0.7194 0.7181 0.5403 0.5672 0.5552 0.5521 0.4428
Value-weighted 1 0.8544 0.7192 0.6352 0.719 0.6133 0.5332 0.4373 0.5306 0.4169 0.4245
Equal-weighted 1 0.6738 0.7341 0.7074 0.7388 0.5398 0.5831 0.5581 0.5784 0.4674
Weekly returns
Standard and Poor 1 0.8843 0.9916 0.8473 0.6074 0.5485 0.6175 0.5234 0.4545
Economy 1 0.9247 0.979 0.6105 0.6995 0.6452 0.685 0.4758
Value-weighted 1 0.8961 0.6187 0.5902 0.6359 0.5677 0.4671
Equal-weighted 1 0.6075 0.7132 0.6468 0.7128 0.4859
Monthly returns
Standard and Poor 1 0.8094 0.987 0.7695 0.8219
Economy 1 0.8689 0.9859 0.647
Value-weighted 1 0.8323 0.7829
Equal-weighted 1 0.6429
Morgan Stanley 1
The beta estimates were obtained using the following time series regression:
rit rf t ¼ ai þ bIi ðrIt rf t Þ þ eit t ¼ 1; N ; t0 ð2bÞ
For each index, except the Morgan Stanley Index, this regression was estimated using daily returns for 1 year, weekly returns for 2 years, and monthly returns for 5 years
over the entire sample period. For the Morgan Stanley Index, only estimates based on 5 years of monthly data were obtained as this was the only data available. For each
data frequency, time period, and index, this resulted in 22,905 estimated betas. The correlation coefficients were calculated as the simple correlation coefficients between
the betas estimated for different data frequencies, time periods, and indexes.
421
Table 5
Average estimates for intercept and slope coefficients from CAPM
γ0 = 1
T
T
γ0t+1
∑t ˆ γ1 = 1
T
T
γ1t+1
∑t ˆ
Average Standard Average Standard
estimate deviation estimate deviation
Daily data
Standard and Poor 0.0727 0.1676 0.0148 0.1116
Economy 0.0455 0.1413 0.0456 0.1271
CRSP value-weighted 0.0691 0.1659 0.0169 0.1055
CRSP equal-weighted 0.0429 0.1409 0.0361 0.1052
Weekly data
Economy 0.0251 0.1314 0.0679 0.1497
Monthly data
Economy 0.0205 0.1189 0.0786 0.1654
Morgan Stanley 0.0443 0.137 0.0401 0.1164
For each year, for each stock in the sample, the coefficients are obtained from the following cross-section
regression:
ˆ
ritþ1 rf tþ1 ¼ c0tþ1 þ c1tþ1 bIit þ eitþ1 i ¼ 1; N ; N : ð3Þ
ˆ
where bIit is obtained from estimation of Eq. (2b) for the previous period; the previous year of daily data, 2 years
of weekly data, and 5 years of monthly data. This process was repeated for each index; except the Morgan Stanley
Index for which only monthly data is available. Shaded areas indicate a significant coefficient at the 5% level. The
standard deviation is the sample standard deviation of the individual annual estimates.
monthly data and either the Economy Index or the CRSP equal-weighted index provide
the best estimates. The significance of the average estimated values for c 1 further
support this conclusion. The average R 2 values for all data frequencies and indexes are,
however, very low; they indicate that beta explains at best 3% of excess return, on
average. For some individual years, the R 2 value is a little higher; the highest value
obtained is 11.73%. There are also some very low values, most indexes have a minimum
R 2 value of 0.01%.
In summary therefore, the general recommendation of using 5 years of monthly data for
estimation of expected return based on CAPM appears to be reasonable. However, from
the results obtained here, the Economy Index and the equal-weighted CRSP index perform
better than the value-weighted indexes examined (see Tables 5 and 6). Given the high
correlation between the Economy Index and the CRSP equal-weighted index (see Table 4),
it is probably irrelevant which one is used and the CRSP equal-weighted index is easier to
obtain. It is, however, worth noting that, in relation to construction of the Economy Index,
there were a number of problems associated with the GDP figures for the service
Table 6
Average R 2 values from CAPM
Index Average R 2 Standard deviation Minimum Maximum
Daily data
Economy 0.0211 0.0322 0.0001 0.1081
Weekly data
Economy 0.0247 0.0305 0.0001 0.1044
Monthly data
Economy 0.0271 0.0266 0.0001 0.0909
Morgan Stanley 0.0168 0.0185 0.0001 0.0752
For each year, for each stock in the sample, R 2 is obtained from the following regression:
ˆ
ritþ1 rf tþ1 ¼ c0tþ1 þ c1tþ1 bIit þ eitþ1 i ¼ 1; N ; N : ð3Þ
ˆ
where bIit is obtained from estimation of the time series regression (2b) for the previous period; the previous year
of daily data, 2 years of weekly data, and 5 years of monthly data. This was done for each index; except the
Morgan Stanley Index for which only monthly data are available.
industries, etc. Solving these problems could lead to an even better estimate using the
Economy Index. A topic for future research.
6. Empirical analysis—the Fama and French model
For estimation based on the three-factor model, the time frame and data frequency that
provide the best estimates, based on CAPM, are used; i.e., 5 years of monthly excess
returns. From Section 3 therefore, the first step of the analysis involves estimation of Eq.
(4) using 5 years of monthly data to obtain estimates for the individual betas. Then, these
estimates are used in Eq. (5) to obtain an estimate for the subsequent year’s expected
return. As for CAPM, the estimation is rolled forward, R 2 is used to determine the best
estimate, and significance of the average estimates for the coefficients in Eq. (5) is noted.
Table 7 presents the average estimates for the coefficients in Eq. (5). The results
obtained are similar to those for CAPM; only the CRSP equal-weighted index and the
Economy Index have significant coefficients for the market index ðcP1Þ. The sizes of the
expected risk premia for the market portfolio in the Fama and French model, ðcP1Þ, are very
similar to the estimates from CAPM, around 7%. For the additional factors, only the size
factor (SMB) is found to be significant. There also appears to be significant differences in
between the periods 1975–1985 and 1986–1996. During the first period, the risk premium
Table 7
Average estimates of intercept and slope coefficients from the Fama and French model
Index Average coefficient
1
PT 1
PT 1
PT 1
PT
c0 ¼ T t c1 ¼ T t c2 ¼ T t c3 ¼ T t
cˆ
0tþ1 cˆ
1tþ1 cˆ
2tþ1 cˆ
3tþ1
1975–1996
Morgan Stanley 0.0340 0.0129 0.0429 0.0017
Value-weighted 0.0279 0.0311 0.0416 0.0009
Equal-weighted 0.0293 0.0649 0.0417 0.0002
Economy 0.0286 0.0737 0.0414 0.0018
1975–1985
Morgan Stanley 0.0258 0.0221 0.0794 0.0055
Value-weighted 0.0182 0.0447 0.0778 0.0026
Equal-weighted 0.0264 0.1046 0.0795 0.0007
Economy 0.0220 0.1205 0.0787 0.0039
1986–1996
Morgan Stanley 0.0423 0.0036 0.0064 0.0021
Value-weighted 0.0375 0.0174 0.0054 0.0009
Equal-weighted 0.0323 0.0524 0.0039 0.0004
Economy 0.0352 0.0269 0.0042 0.0004
For the Fama and French model, first, for each stock in the sample, the time series regression:
rit rf t ¼ ai þ b1i ðrIt rf t Þ þ b2i SMBt þ b3i HMLt þ eit t ¼ 1; N ; t0 ð4Þ
ˆˆ ˆ
is estimated using 5 years of monthly data to obtain estimates for the betas b1it ; b2it ; and b3it . Then, an estimate
for expected return in the subsequent year, for each stock in the sample, is obtained by using the estimated betas
as explanatory variables using the following cross-section regression:
ˆ ˆ ˆ
ritþ1 rf tþ1 ¼ c0tþ1 þ c1tþ1b1it þ c2tþ1b2it þ c3tþ1b3it þ eitþ1 i ¼ 1; N ; N ð5Þ
This was done for each of the five indexes examined. The shaded areas indicate a significant coefficient at the 5%
level.
on the market index and the size factor are significant, whereas they are not significant for
the last subperiod.
The resulting R 2 values are shown in Table 8. It can be seen that the Fama and French
model does not perform much better than CAPM under the simple estimation technique
used in this paper; an average R 2 of 5% is all the model is able to produce. Thus, at least
for the simple technique used here, it is not clear that the Fama and French model offers
any advantages over CAPM in terms of significance of the factors and explanatory power.
As for the risk premia, the performance is better for the subperiod 1975–1985 than for the
period 1986–1996. The R 2 is around 7% for the first period but drops to below 3% for the
second period for all the indexes. In contrast to CAPM, the choice of index does not seem
to matter in terms of R 2. This suggests that the use of an equal-weighted index for
estimation based on CAPM captures part of the size effect found significant in the three-
Table 8
Average R 2 values from the Fama and French model
Index Average R 2 values
Mean Minimum Maximum
1975–1996
Standard and Poor 0.048 0.004 0.145
Morgan Stanley 0.047 0.005 0.144
Value-weighted 0.049 0.004 0.144
Equal-weighted 0.049 0.004 0.145
Economy 0.048 0.005 0.145
1975–1985
Economy 0.068 0.025 0.145
1986–1996
Economy 0.028 0.005 0.083
For each year, the R 2 value from the Fama and French model is obtained from estimation of
ritþ1 rf tþ1 ¼ c0tþ1 þ c1tþ1ˆ
b1it þ c2tþ1ˆ
b2it þ c3tþ1ˆ
b3it þ eitþ1 i ¼ 1; N ; N ð5Þ
where the beta estimates are from estimation of Eq. (4) for the previous period. This was done for each index
examined.
factor model. In summary therefore, from the results obtained here, it is not clear that
inclusion of two additional factors for estimation of expected return provides any
significant benefits. The extra work involved in collecting data for these two factors is
probably not justified.
7. Conclusion
Given the importance of cost of equity, and more generally expected return, in
financial decision making and academic studies, there is surprisingly little research
relating to the properties of the estimates obtained from the various models and methods
used for estimation. The general recommendation for estimation of expected return based
on CAPM is to use 5 years of monthly data and a value-weighted index. From the
analysis done in this paper, while 5 years of monthly data appears to be appropriate, the
Economy Index constructed here and the CRSP equal-weighted index provide better
estimates than the value-weighted indexes examined. Further, use of the Economy Index
can be justified by a single-factor APT model or the consumption-based capital asset
pricing model. The Economy Index is nontrivial to construct, but since it is also
established here that it is highly correlated with the CRSP equal-weighted index, it can be
concluded that the use of 5 years of monthly data and the CRSP equal-weighted index
provides the best estimate for beta, and therefore expected return, based on CAPM.
However, it is also found that the ability of beta to explain differences in returns in
subsequent periods ranges from a low of 0.01% to a high of 11.73% across years and is
at best 3% on average. Given that the analysis was done for NYSE stocks that trade
more than 95% of the time, it is of concern that so little of the differences in returns
between stocks are explained, under such favourable circumstances. Further, the Fama
and French three-factor model does not do much better; although the size factor is
found to be significant, the R 2 is only around 5%. The low explanatory power of both
the CAPM and the Fama French model suggests that neither model is useful for
estimation of cost of equity, at least for the simple estimation techniques used here.
What is the alternative? One possibility is the use of more sophisticated estimation
techniques to deal with problems such as errors in variables which arise when a simple
technique is used. Such techniques are, however, probably cost prohibitive for
individual firms, in particular, in relation to the amount of data required. This is not
the case for professional beta providers. This suggests that individual firms should use
professional beta providers for obtaining their beta estimates instead of estimating them
themselves, and that professional beta providers should use more complex techniques
than used in this paper. This is of course only the case if the betas obtained from beta
providers are in fact superior to those obtained using simple estimation techniques.
From the quote at the beginning of this paper, this cannot be taken for granted.
Therefore, we need to compare the betas from beta providers with those obtained using
simple estimation techniques. A topic for future research.
References
Banz, R. (1981). The relationship between return and market value of common stock. Journal of Financial
Economics, 9, 3 – 18.
Bartholdy, J., & Peare, P. (2003). Unbiased estimation of expected return using CAPM. International Review of
Financial Analysis, 12, 69 – 81.
Blume, M. (1975). Betas and their regression tendencies. Journal of Finance, 30(3), 785 – 795.
Brigham, E., & Gapenski, L. (1997). Financial Management (8th ed.). The Dryden Press.
Bruner, R. F., Eades, K., Harris, R., & Higgins, R. (1998). Best practices in estimating the cost of capital: Survey
and synthesis. Financial Practice and Education, 8(1), 13 – 28.
Carleton, W., & Lakonishok, J. (1985, January–February). Risk and return on equity: The use and misuse of
historical estimates. Financial Analyst Journal, 41, 38 – 47.
Fama, E., & French, K. (1992, June). The cross-section of expected stock returns. Journal of Finance, 47,
427 – 465.
Fama, E., & French, K. (1993). Common risk factors in the returns on stocks and bonds. Journal of Financial
Economics, 33(1), 3 – 56.
Fama, E., & MacBeth, J. (1974). Tests of the multiperiod two-parameter model. Journal of Financial Economics,
1, 43 – 66.
Graham, J. R., & Harvey, C. R. (2001). The theory and practice of corporate finance: Evidence from the field.
Journal of Financial Economics, 60(1–2), 187 – 243.
Kan, R., & Zhang, C. (1999). Two-pass tests of asset pricing models with useless factors. Journal of Finance,
54(1), 203 – 235.
Klemkosky, R., & Martin, J. (1975). The adjustment of beta forecasts. Journal of Finance, 30(4), 1123 – 1128.
Hawawini, G. A. (1983 May–June). Why beta shifts as the return interval changes. Financial Analyst Journal, 39,
73 – 77.
Reilly, F., & Wright, D. (1988, Spring). A comparison of published betas. Journal of Portfolio Management,
64 – 69.
Roll, R. (1988). R2. Journal of Finance, 43(2), 541 – 566.
Ross, S. (1976). The arbitrage pricing theory of capital asset pricing. Journal of Economic Theory, 13, 341 – 360.
Shanken, J. (1992). On the estimation of beta-pricing models. Review of Financial Studies, 5, 1 – 33.

Stimation of Expected Return - CAPM vs. Fama and French

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Stimation of Expected Return - CAPM vs. Fama and French

Uploaded by

Copyright:

Available Formats

International Review of Financial Analysis 14 (2005) 407 – 427

Estimation of expected return:

Accepted 22 October 2004

JEL classification: G11; G12; G31

2. Estimation issues for CAPM

In this section, a brief discussion of issues relating to estimation of expected return

rit ¼ ai þ bIi rIt þ eit t ¼ 1; N ; t0 : ð2aÞ

! Do dividend adjustments in the index matter?

2.2. Excess returns or raw returns

2.3. Data frequency and time period

3. The Fama and French model

rit rf t ¼ ai þ b1i ðrI t rf t Þ þ b2i SMBt þ b3i HMLt þ eit t ¼ 1; N ; t0 : ð4Þ

ritþ1 rf tþ1 ¼ c0;tþ1 þ c1tþ1ˆ

5.2. Excess returns or raw returns

rit rf t ¼ ai þ bIi ðrI t rf t Þ þ eit t ¼ 1; N ; t0 ð2bÞ

rit rf t ¼ ai þ bIi ðrIt rf t Þ þ eit t ¼ 1; N ; t0 : ð2bÞ

5.3. Data frequency, time period, and index

J. Bartholdy, P. Peare / International Review of Financial Analysis 14 (2005) 407–427

rit rf t ¼ ai þ bIi ðrIt rf t Þ þ eit t ¼ 1; N ; t0 ð2bÞ

6. Empirical analysis—the Fama and French model

You might also like