Sie sind auf Seite 1von 30

European Financial Management, Vol. 16, No.

1, 2010, 43–71
doi: 10.1111/j.1468-036X.2009.00530.x

The CAPM is Alive and Well: A Review


and Synthesis
Haim Levy
The Hebrew University and The Center of Law and Business
E-mail: mshlevy@mscc.huji.ac.il

Abstract
Mean-Variance (M-V) analysis and the CAPM are derived in the expected utility
framework. Behavioural Economists and Psychologists (BE&P) advocate that
expected utility is invalid, suggesting Prospect Theory as a substitute paradigm.
Moreover, they show that the M-V rule, which is the foundation of the CAPM,
is not always consistent with peoples’ choices. Thus, BE&P cast doubt on the
validity of expected utility paradigm and of the M-V rule, hence the CAPM is
theoretically questionable. In addition, there is very little empirical support to
the CAPM. We show in this study that the CAPM is theoretically valid even
when one accepts the BE&P framework and even when expected utility is invalid.
Moreover, within the BE&P framework there is a strong experimental support for
the CAPM.

Keywords: CAPM, M-V , expected utility, Prospect Theory


JEL classification: D81, C91

1. Introduction

Since the emergence of modern finance, only a few papers, which are highly cited and
considered to be the pillars of the profession, have been published. There is no doubt that
Markowitz’s (1952a) Mean-Variance (M-V) efficiency analysis and Sharpe (1964) and
Lintner’s (1965) Capital Asset Pricing Model (CAPM) are two such pillars. The CAPM
is the base model on which hundreds of academic studies rely. The impact of the CAPM
is not confined to academic research. Practitioners also frequently use the Sharpe Ratio,
which relies on the M-V model and use beta, which is derived from the CAPM as the
risk index. The list of studies which rely on the M-V analysis and the CAPM is too
long to mention here. However, it is sufficient to open any finance textbook to see the
great impact of the M-V rule and the CAPM on the profession. Indeed, in recognition
of these important contributions to science, Markowitz and Sharpe won the Nobel Prize
in economics in 1990.

The author acknowledges the helpful comments of Moshe Levy, Frank Fabozzi and Harry
Markowitz. This study was partially financed by of the Krueger Center of Finance.

C 2009 Blackwell Publishing Ltd.
44 Haim Levy

Yet, the CAPM is under severe and ongoing theoretical and empirical academic
attacks. Most of the early attacks emerge from numerous empirical studies in finance
and economics, which reveal that the CAPM does not fit empirical asset pricing well
(for an excellent survey of the empirical results, see Fama and French, 2004) However,
the more severe attacks come from an unexpected angle: behavioural economists and
psychologists cast doubt on the validity of von Neumann and Morgenstern’s (1953)
Expected Utility Theory (EUT); thereby also casting indirect doubt on the validity
of the M-V rule and the CAPM, which are derived within the EUT framework (this
criticism, however, does not apply to the APT derivation of the CAPM by Ross, (1976),
which does not rely on expected utility).
In a breakthrough series of studies, Kahneman and Tversky (K&T) and Tversky
and Kahneman (T&K) (see for example, 1979, 1992) show that the typical investor is
not always a rational and ‘efficient machine’, as assumed by EUT economists. This
implies that the foundations of EUT are not valid, further implying that all models -
including the M-V efficiency analysis and the CAPM, derived in the EUT framework -
are questionable. Kahneman and Tversky first suggested Prospect Theory (PT) and, later
on, a modified version of PT, called Cumulative Prospect Theory (CPT), as substitutes
for EUT. The influence of PT and CPT on economics and finance has been enormous.
Indeed, in recognition of these contributions to academic research, Kahneman won the
Nobel Prize in economics in 2002. Thus, we have two conflicting paradigms - EUT and
CPT - and if CPT is correct and EUT is rejected, the M-V efficiency analysis and the
CAPM have no theoretical justification. As a result, these models seem to lose ground.
Nevertheless, despite these criticisms, in the current paper we show that M-V analysis
and the CAPM are, in fact, alive and doing well. We do not claim that the other alternate
theories are wrong. Rather, we attempt to show that a modified version of M-V analysis
and the classic CAPM can also be justified and safely used in the CPT framework,
despite the fact that under CPT, EUT is rejected. In this paper we focus on the theoretical
criticisms of the M-V and the CAPM, but we also briefly discuss the implications of
the empirical and experimental results to the validity of the CAPM. We show that the
CAPM cannot be rejected as was previously thought, as long as ex ante rather than
ex post parameters are employed in the CAPM tests, or when ex ante experimental
parameters are employed.
The structure of this paper is as follows: In Section 2, we discuss the common
assumptions needed to justify the M-V rule and the CAPM within the EUT framework,
and provide a brief review of the CAPM and M-V criticisms. In Section 3, we provide
the detailed theoretical criticisms of the CAPM, and show that the CAPM easily
withstands the barrage of theoretical criticisms; it is valid even in the non-expected
utility framework. Namely, EUT is criticised (and rightfully so); however, the CAPM
remains intact - quite a surprising result. In Section 4, we briefly discuss the empirical
criticisms of M-V and the CAPM and show that the CAPM cannot be empirically
rejected. Moreover, we show that experimental studies which use ex ante parameters
strongly support the CAPM. Section 5 concludes the paper.

2. The Main Criticisms of M-V Analysis and the CAPM

To understand the criticisms of the CAPM and M-V efficiency analysis in a transparent
way, let us first be more specific regarding the role of Normality and the shape of
preference required to justify the employment of these models. To derive the CAPM, it

C 2009 Blackwell Publishing Ltd
The CAPM is Alive and Well: A Review and Synthesis 45

is generally common to require that investors maximise expected utility with a quadratic
preference, or alternatively, to require that the distributions of rates of return are Normal
and risk aversion prevails. 1 With some additional assumptions (e.g. homogeneous
expectations, no transaction costs, unrestricted borrowing, etc.), one can derive the
linear risk-return relationship, as implied by the CAPM. As the quadratic preference is
too specific and since it also suffers from other severe drawbacks (see Arrow, 1965), we
focus here on the case of Normality and risk aversion. After defining the assumptions
needed to justify the CAPM, let us now turn to the criticisms of this model.

2.1 Theoretical criticisms of EUT and CAPM


The main theoretical criticisms are as follows:

• Allais (1953) criticises EUT. He shows that using EUT in making choices between
pairs of alternatives, particularly when small probabilities are involved, may lead to
some paradoxes within EUT theory. Thus, it casts doubt on the validity of EUT, which
is the foundation of both the M-V rule and CAPM. This paradox motivated the idea of
using decision weights instead of objective probabilities (for more details, see below).
• Roy (1952) also criticises EUT. He asserts that,

‘A man who seeks advice about his actions will not be grateful for the suggestion
that he maximises expected utility’ (see Roy, 1952, p. 433).

He suggests that people should rely on the Safety First (SF) rule, rather than on EUT.
If one accept Roy’s claim, EUT is generally invalid; hence, the M-V and the CAPM
are also invalid. 2
• Even if EUT is valid, some fundamental papers question the validity of the risk-
aversion assumption. Just to mention a few of these studies, Friedman and Savage
(1948), Markowitz (1952b), Swalm (1966), Levy (1969) and Kahneman and Tversky
(1979) all claim that the typical preference must include risk-averse, as well as risk-
seeking, segments. Thus, the variance cannot be an index for risk, which casts doubt
on the validity of the CAPM.
• Kahneman and Tversky’s (1979) Prospect Theory (PT) and its modified version -
Tversky and Kahneman’s (1992) Cumulative Prospect Theory (CPT) - show that
subjects behave in contradiction to what is predicted by EUT; hence, they reject EUT
which, once again, indirectly casts doubt on the validity of the M-V analysis and the
CAPM. It is worth noting that the CPT’s criticism of EUT is quiet general and has
various dimensions, beyond the criticism of the shape of the preference mentioned in
Point 3, above.

1
One can justify the M-V rule with the more general family of distributions, the Elliptic
distributions, which include - apart from the Normal distribution - other distributions, e.g.,
the Logistic distribution, see Chamberlain (1983) and Berk (1997). However, we focus here
on the Normal distribution, and the results can be extended to the other distributions included
in the Elliptic family.
2
Though Roy explicitly rejects EUT, there is one specific degenerate utility function which
conforms to the SF rule. However, this preference has a severe drawback and is thus
considered as unacceptable (see Figure 2).

C 2009 Blackwell Publishing Ltd
46 Haim Levy

• Providing some simple examples, Baumol (1963) and Leshno and Levy (2002) show
that there are cases where there is no M-V dominance of one prospect over the other,
yet allegedly all subjects would prefer one of these prospect. Therefore they claim
that the M-V rule is a sufficient but not a necessary investment decision rule; thus, it
is not an optimal rule, leading to an elimination of a portion, or portions, of the M-V
efficient frontier from the efficient set. Therefore, the market portfolio may also be
eliminated from the efficient set, which has an ambiguous effect on the CAPM.

2.2 Empirical criticisms of the M-V and the CAPM


The main empirical criticisms are as follows:

• The M-V criterion and the CAPM rely on the Normal distribution assumption.
Numerous studies test the goodness of fit of actual rates of return distributions to
the Normal distribution. In almost all cases, the null hypothesis asserting that the
distribution of rates of return is Normal is strongly rejected; hence, one of the main
justifications of the M-V analysis and the CAPM loses ground.
• Testing the CAPM directly reveals only minimal support for the expected linear risk-
return relationship; in some cases it reveals a strong rejection of the CAPM, when
beta reveals almost no explanatory power of the variation in mean returns.
• Deriving the M-V efficient set, it is generally found that some of the investment
weights of the tangency portfolio are negative. Moreover, as the number of assets
increases, it is empirically shown that the percentage of assets corresponding to the
tangency portfolio with negative weights approaches 50%. 3 These findings contradict
the CAPM because, in order to guarantee equilibrium, the investment weights of the
tangency portfolio must all be positive. 4

Thus, the M-V criterion and the CAPM are theoretically and empirically attacked.
One may argue that some or all of these criticisms are invalid. We choose a different
route: we accept all these criticisms and assume that they are valid, and examine their
effect on the CAPM. Taking this approach, one would suspect that the M-V analysis and
the CAPM would not survive the above list of harsh criticisms. Surprisingly, however,
this is not the case.
In this study, we show that the M-V analysis and the CAPM survive all these criticisms.
Indeed, the M-V efficiency analysis has to be slightly modified to incorporate some of
the above theoretical criticisms, but the CAPM is found to be alive and doing well, even
after these modifications. We conclude that unless we observe in the future new and
more convincing evidence refuting the M-V framework and the CAPM, we can safely
continue to include these two fundamental models in our teaching curriculum, in future
academic research, as well as in practice.

3
For studies which investigate the mathematical conditions needed to obtain positive weights
portfolios, and for the empirical results showing that negative weights always exist, see for
example, Roll (1977), Rudd (1977), Levy (1983), Green (1986) and Best and Grauer (1992).
4
According to the CAPM, it is assumed that unrestricted short selling is allowed, which is
an unrealistic assumption. However, the end result of the CAPM is that all investors hold
the market portfolio; hence, short selling is not optimal.

C 2009 Blackwell Publishing Ltd
The CAPM is Alive and Well: A Review and Synthesis 47

3. Overcoming the Theoretical Criticisms of the CAPM

Let us examine each criticism separately:

3.1 The Allais paradox


The Allais paradox may be solved once one employs decision weights (DW), rather than
objective probabilities (see Kahneman and Tversky, 1979). As can be seen later on in
the paper, when Normal distributions are assumed and monotonic DW are employed,
as suggested by Quiggin (1982) and Tversky and Kahneman (1992), all investors select
their portfolios from those portfolios located on the Capital Market Line (CML); hence,
the CAPM is intact. Thus, on the one hand, one can use DW to solve the Allais paradox,
while on the other hand, with DW as suggested by CPT, the CAPM is valid (for more
details, see Points 3 and 4 discussed below). However, it is important to emphasise that
the original Allais paradox is presented with discrete distributions, and hence, is not
directly relevant to our study, as we assume Normality. Yet, if a paradox like the one
presented by Allais can also be constructed with Normal distributions, then we suggest
that monotonic DW may solve the paradox without violating the CAPM, even when
EUT is rejected (see discussion at the end of Point 4 below).

3.2 Roy’s Safety First Rule


Roy advocates that the risk of disaster is the main factor one should consider in making
an investment choice; therefore, expected utility, in its classic form, should not be taken
seriously. In his 1952 model, Roy goes to the extreme and sets the goal of the investor
solely as the minimisation of the probability of disaster. Thus, according to this extreme
Safety First (SF) rule, one should select the diversification strategy, which minimises
the probability of disaster. Namely, the goal is:
Min Pr (x < d) (1)
p

where d stands for a disaster level, x stands for the rate of return, and p denotes a
vector of optimal investment proportions. 5 Though Roy does not provide a theoretical
justification for this rule, he is aware of the M-V efficiency analysis; in his paper there
is a full mathematical derivation of the M-V efficient frontier, which is very similar to
the derivation developed by Merton (1972). Moreover, Roy does not provide a recipe for
how to determine the disaster level, d, which is subjective. For example, for an executive
or a mutual fund manager, d can be a level of profit such that he or she will be fired
if the achieved performance is below this profit level. However, if the riskless asset
prevails - and if investors experience a feeling of failure whenever the rate of return is
below the riskless interest rate - then the disaster level d is equal to the interest rate,
a case where the CAPM is valid (with one additional and reasonable assumption, see
discussion below), even under the Safety First rule. Getting less than the safe rate of
return may be considered a failure, by individual and institutional investors alike; hence,
choosing d = r seems to be appropriate for many, albeit not all, investors. Therefore,
special attention is given below to the case where d = r.

5
There are several experimental studies supporting the idea that a disaster level (or aspiration
level) plays an important role in decision making. See, for example, Payne et al. (1980, 1981).

C 2009 Blackwell Publishing Ltd
48 Haim Levy

We would now like to discuss the claim that various versions of the CAPM are also
intact in Roy’s framework, regardless of whether the riskless asset prevails. To show
this, we consider various scenarios. As we shall see, all these interpretations, except for
one (which is not accepted by economists, as it implies that risky assets do not exist in
the market), are consistent with the classic M-V analysis and with various forms of the
CAPM.
Let us first take Roy’s rule strictly as a given by Eq. (1). Though Roy does not discuss
the allocation of the investment between risky and riskless assets, we first assume (this
assumption will be relaxed later on in the paper) that the investor decides to invest some
portion of her wealth in risky assets; the only question is how to diversify the allocated
amount to risky assets according to Roy’s rule among the available individual risky
assets.
Using Chebysheff’s inequality, the following holds,
Pr {|x − μ| > kσ } ≤ 1/k2 (2)
Of course, this inequality is meaningful only if k is larger than 1.
Choosing k = (μ − d)/σ , Roy shows that Eq. (2) can be rewritten as,
Pr {|x − μ| > (μ − d)} ≤ σ 2 /(μ − d)2 (3)
This implies a fortiori that,
Pr {(μ − x) > (μ − d)} ≤ σ 2 /(μ − d)2
Hence, the upper bound of the probability of disaster is given by,
Pr (x < d) ≤ σ 2 /(μ − d)2 (4)
Thus, according to Roy the goal is to minimise the ratio, σ/(μ − d), or alternatively
to maximise the ratio,
(μ − d)/σ (4a)
If the riskless asset does not prevail an investor with a disaster level d 1 will invest
in the risky portfolio m 1 , while an investor with a disaster level d 2 will invest in the
risky portfolio m 2, and so on (see Figure 1). Since investing in an interior portfolio will
increase the probability of a disaster (see Eq. (4a)), all investors regardless of d i , will
invest in M-V efficient portfolios; hence, the Zero Beta version (see Black, 1972) of
the CAPM is intact.
Let us now assume that the riskless asset exists and analyse various relationships
between r and d. Choosing first the reasonable value d = r, we see that the expression in
eq. (4a) becomes the well known Sharpe ratio and therefore, Roy’s rule coincides with
the maximisation of the Sharpe ratio (see Sharpe, 1966), i.e. with the CAPM. Figure 1
demonstrates that by choosing d = r one invests, according to Eq. (1), in a combination
of the tangency portfolio and the riskless asset (see Point p); hence, the CAPM holds 6
(for a discussion of this point, see Levy and Sarnat, 1971). However, this result holds true
only when some proportion of the assets must be invested in the risky asset, otherwise
according to Roy’s rule 100% of the assets should be invested in the riskless asset.

6
Roy’s SF rule inspired a series of academic studies, see, for example, Levy and Sarnat
(1972), Arzac and Bawa (1977), Pyle and Turnovsky (1970), Goetzmann and Broadie (1992)
and Milevsky (1999).

C 2009 Blackwell Publishing Ltd
The CAPM is Alive and Well: A Review and Synthesis 49

Mean

m
p
p1 * m1
p1
r m2

d1

d2

Standard Deviation

Fig. 1. The optimum portfolios when the riskless asset does not prevail are denoted by m, m 1
and m 2 , corresponding to disaster levels d = r, d 1 and d 2 , respectively. When the riskless
asset prevails, and there is a constraint that some proportion of the wealth must be invested in
the risky assets, the optimal portfolios, according to Roy’s rule, are given by Portfolio P
corresponding to m, and Portfolio P 1 corresponding to m 1 (the portfolio corresponding to m 2
is omitted to provide a transparent picture). However, when the riskless asset prevails, having
a disaster level d 1, the investor would benefit from shifting from m 1 to m; hence, shifting
from p 1 to, say, p1∗ , the probability of disaster corresponding to d 1 decreases. Therefore, with
the riskless asset, all investors will choose some combination of m and r, regardless of the
disaster level d i .

Suppose now, as before, that some portion of the assets must be invested in risky
assets, that the riskless asset prevails, and that the disaster level, d i , is not necessarily
equal to r. Moreover, d i may vary across investors, but all d i are smaller than r (see the
values d 1 and d 2 in Figure 1). 7 In this case, we claim that the classic Separation Theorem
holds true as all investors will select the same portfolio of risky assets, Portfolio m, i.e,
the market portfolio, as advocated by the CAPM. To better understand this, first recall
that Portfolio m 1 minimises the probability of disaster when d 1 is the disaster level and
the riskless asset does not prevail. However, with the riskless asset, for any combination
of the riskless asset and Portfolio m 1 located on line rm 1 , there is a combination of
Portfolio m and the riskless asset located on line rm, which dominates it in accordance
with Roy’s rule. For example, suppose that by combining m 1 with r, Portfolio p 1 is
selected (see Figure 1). Then, by combining m with r, Portfolio p1∗ can be achieved and
this portfolio dominates Portfolio p 1, as it has the same mean return and a lower risk of
a disaster, when d 1 is the disaster level. The reason for this is that the slope of line d1 p1∗
is higher than the slope of line d1 p1 . To sum up, when the investor’s goal is to minimise
the risk of disaster, as suggested by Roy, one of the versions of the CAPM holds true; the
selected version depends on the relationship between r and d, and whether the riskless
asset prevails.

7
The case where d is greater than r is not considered in this paper, as according to Roy’s
rule, it leads to an infinite borrowing, which contradicts equilibrium. To better understand
this claim, recall that with d > r, moving to the right on the CML, the slope (μ − d)/σ
increases.

C 2009 Blackwell Publishing Ltd
50 Haim Levy

U(x)

d x

Fig. 2. The only preference which is consistent with Roy’s strict Safety-First rule.

How can Roy’s severe criticism of EUT be reconciled with the fact that the CAPM,
which is derived within the EUT framework, remains intact? To answer this question
one must analyse the preference implied by Roy’s rule. It can be shown that the only
utility function which is consistent with Roy’s strict SF rule is presented in Figure 2.
Thus, if r is greater than d, and if the investor is not constrained by having to invest
some proportion of the wealth in the risky asset (and in addition, if the risky asset does
not dominate the safe asset by First degree Stochastic Dominance (FSD), as occurs
with Normal distributions), then it is easy to see that with the preference presented in
Figure 2, risky assets vanish from the market, since investing 100% in the riskless asset
is optimal, resulting in a zero probability of disaster. This result is intact for any value
d, as long as it is smaller than r. It is clear from the above analysis and from Roy’s paper
that Roy did not mean to employ this extreme strategy, as in such a case risky assets
vanish; and hence, all the mathematical M-V analyses conducted by him are redundant.
The above interpretation of the SF rule is, of course, unacceptable. As explained
above, in order to receive acceptable results, one can alternatively assume that there
is no riskless asset, or that, as explained previously, the riskless asset exists, but there
is a constraint: some portion of the money must be allocated to the risky assets. In
this realistic scenario, Black’s Zero Beta CAPM is intact. However, with the riskless
asset and with the above imposed constraint, the optimal investment policy will not be
consistent with a maximisation of expected utility, as dictated by Figure 2.
Alternatively, one can relax the assumption that some portion of the assets must be
invested in the risky assets and assume another utility function, which is still in line
with Roy’s approach, but also takes monotonicity into account. To be more specific,
the preference presented in Figure 2 is unacceptable because it ignores the magnitude
of the outcomes. Namely, according to the strict SF rule, a utility of $10 or one million
dollars is the same, as long as these two values are greater than d. Indeed, in a later
paper even Roy (1956), himself, suggests an extension of his SF rule to a more realistic
framework. He suggests that the disaster level d may be uncertain implying that the
utility function is continuous.
How can one keep the main component of Roy’s SF rule in an EU framework without
the absurd results asserting that there is no difference for the investor if he receives $10

C 2009 Blackwell Publishing Ltd
The CAPM is Alive and Well: A Review and Synthesis 51

U(x)

d x

Fig. 3. The preference which is consistent with the modified Safety–First rule: There is a
penalty when the outcomes are below the disaster level d i , but the preference is monotonic.

or a million dollars? Indeed, in a recent paper, Levy and Levy (2008) suggest a synthesis
of Roy’s rule with EUT; however, this time they suggest a preference which, on the one
hand, in the spirit of Roy, penalises the potential investment if the return is below some
disaster level, d, and on the other hand, does not imply indifference to all outcomes
above d, and the larger the outcome the better off the investor is. This preference is
presented in Figure 3.
The suggested preference given in Figure 3 is neither continuous nor completely
concave. Yet, as long as the distribution of returns is Normal, in order to obtain the
CAPM there is no need to assume risk aversion; rather, it is sufficient to assume a
non-decreasing preference (see Point 3 below). In addition, continuity of the preference
is not needed (see Kroll and Levy, 1982). Thus, with this interpretation of the SF rule,
this rule as well as the CAPM is valid in an expected utility framework, even when the
constraint that some portion of the wealth must be invested in the risky assets is relaxed
(see Point 3 below).
To summarise Roy’s criticism then, we put forward the following possible cases:
Case A: In this case, it is assumed that the riskless asset prevails and that for all investors
d i is smaller than r. Furthermore, it is assumed that no risky asset dominates the
riskless asset by First degree Stochastic Dominance (FSD) and that there is no
requirement to invest some portion of the wealth in the risky assets. This case
is unacceptable because, according to Roy’s rule, in equilibrium all risky assets
vanish from the market in contradiction to the observed facts. As such, we rule
out this case, also because Roy does not claim that in equilibrium there will be
no risky assets.
Case B: In this case, the riskless asset exists and the CAPM is intact, as long as d i < r
and there is a constraint that some portion of the investment must be allocated to
the risky assets.
Case C: In this case, the riskless asset is not available; in which case, the Zero Beta
CAPM holds true.
Case D: In our opinion, this is the most relevant case. The riskless asset may exist; if
it exists it may be smaller or greater than (or equal) to d i . There is no constraint

C 2009 Blackwell Publishing Ltd
52 Haim Levy

on investing in either the risky or riskless asset. The suggested preference is


presented in Figure 3, when both SF and monotonicity are accounted for. Thus,
in this scenario, Figure 3, rather than Figure 2, is relevant for Roy’s analysis.

In the next section, we will show that with Normal distribution the CAPM is intact
with the preference presented in Figure 3, despite the fact that this preference is neither
concave nor continuous. Thus, according to Roy’s modified version of the SF rule, on
the one hand, EUT is not violated, while on the other hand, the CAPM is intact. If the
riskless asset does not exist with this preference, the Black version of the CAPM is
intact.

3.3 Overcoming the risk-aversion assumption


We next show that with Normal distributions one does not need to assume risk aversion
and that the CAPM also holds true with risk seeking preference. This is crucial, as it
allows the CAPM to survive most of the above theoretical criticisms. To be more specific,
we advocate that with Normal distributions, all investors - including risk-seekers - will
choose their portfolio from those located on the CML. To prove our claim, we employ
First degree Stochastic Dominance (FSD) asserting that prospect F dominates prospect
G, for all investors with non-decreasing utility functions, if for all values x we have
F(x) ≤ G(x), when F and G are the relevant cumulative distributions (for a proof, see
Hadar and Russell (1969) and Hanoch and Levy (1969)).
With Normal distributions, the FSD condition is represented by Theorem 1 below. As
this case is heavily employed in this paper, let us write down the precise conditions for
dominance.
Theorem 1. Let F and G denote two Normal distributions. Then, F dominates G by
FSD if,
a) E F (x) > E G (x) (5)
and
b) σ F (x) = σG (x)
where E and σ stand for the mean and standard deviation of the return. For a proof
and discussion of the equivalence of Theorem 1 in the Normal case to the general FSD
condition, see Hanoch and Levy (1969) and Levy (2006).
Armed with the FSD criterion, corresponding to the Normal case, let us now
show that the CAPM is also intact for risk-seeking investors. Suppose the investor
considers investing in either Portfolio F or Portfolio G (see Figure 4a). For any possible
interior portfolio G, there is a portfolio denoted by F located vertically above it. As
both portfolios have the same variance, the corresponding two cumulative Normal
distributions do not intersect, and the one with the highest mean dominates the other by
First degree Stochastic Dominance (FSD). For the relative location of the two cumulative
distributions, see Figure 4b. Therefore, according to Theorem 1, for risk averters, risk-
seekers, as well as any other non-decreasing preference, Portfolio F dominates Portfolio
G. Therefore, for any portfolio located below the CML, there is a portfolio on the CML
which dominates it by FSD.
As can be seen, to justify the CAPM, we rely on the FSD criterion, which is derived
from the EUT framework. Are we allowed to use the FSD rule, which itself relies on

C 2009 Blackwell Publishing Ltd
The CAPM is Alive and Well: A Review and Synthesis 53

r'
Mean F

m G

Standard Deviation

Fig. 4a. The Mean-Variance efficient frontier: Portfolio F dominates Portfolio G by the
Mean-Variance rule, as well as by the First degree Stochastic Dominance rule.

Cumulative
Distributions
G

Return

Fig. 4b. The cumulative distributions of F and G, where F dominates G by First degree
Stochastic Dominance rule.

EUT, to refute most of the CAPM’s criticisms asserting that the EUT is not intact? Most
competing decision-making paradigms, even though they criticise EUT, accept FSD as a
foundation that should not be contradicted. For example, Kahneman and Tversky realise
that the original PT (1979) may violate FSD, an unacceptable characteristic; hence,
suggesting the modified theory, CPT, which does not contradict FSD. Similarly, other
generalisations or extensions of EUT do not violate FSD (see for example Quiggin’s,
1982 and 1993). 8 Thus, FSD is a stronger foundation, even more general than EUT, to
build on.

8
For an excellent review on non-expected utility theory, see Machina (1994).

C 2009 Blackwell Publishing Ltd
54 Haim Levy

Finally, with the above analysis, we see that risk seekers and risk averters alike will
choose their optimal portfolios from portfolios located on the CML (see Figure 4a);
hence, we claim that risk aversion is not necessary to justify the CAPM equilibrium.
However, as regards risk-seeking, one must impose a restriction on borrowing in order to
guarantee equilibrium. 9 Otherwise, we may have a risk seeker who wishes to borrow an
infinite amount of money which, of course, contradicts equilibrium. Such a restriction
seems reasonable since in practice such restrictions exist regarding the amount borrowed.
It can be shown that if the restrictions on borrowing are ineffective, then the Sharpe-
Lintner CAPM is intact. However, if the restriction is effective, it is possible that not
all investors will hold the tangency portfolio - a case where the Black version of the
CAPM is intact (see Kroll et al., 1988).

3.4 Cumulative prospect theory


As Kahneman and Tversky modified their Prospect Theory and suggested the Cumula-
tive Prospect Theory (CPT), we focus here only on CPT. According to CPT, the investor
maximises a value function of the form,

xα if x ≥ 0
V(x) = β
(6)
−λ(−x ) if x < 0
where 0 < α < 1, 0 < β < 1, λ > 1 and x is the change of wealth, rather than the total
wealth. This value (utility) function is not completely concave. 10 In addition, according
to CPT, investors employ decision weights estimated by the formula,

w+ (P) = γ
[P + (1 − P)γ ]1/γ
(7)

w− (P) = δ
[P + (1 − P)δ ]1/δ
where P stands for cumulative probability, and w− and w+ denote the transformed
cumulative probability in the negative and positive domain, respectively. 11
According to CPT, it seems that the CAPM does not hold true because EUT is
contradicted for the following two reasons: because CPT employs change of wealth,
rather than total wealth, and because CPT employs decision weights (DW), rather than
objective probabilities. The M-V rule may also be contradicted because even if the
distribution of rates of return is Normal, the transformed distribution (see Eq. (7)) is
not; hence, the M-V is not an optimal decision rule. Thus, it seems that with DW, the
M-V rule is invalid, which casts doubt on the validity of the CAPM.
We claim that despite these differences between CPT and EUT, the M-V rule is optimal
and the CAPM is intact, even under CPT. First, note that the M-V efficient set is not
affected by the initial wealth because F(w + x) dominates G(w + x) if and only if F(x)

9
The constraint can be stated in terms of the initial wealth, e.g. such that it will not be more
than, say, 10 or 100 times the initial wealth.
10
Tversky and Kahneman (1992) provide experimental estimates of the various parameters
revealing loss aversion, namely, a dollar loss weighs more than a dollar gain. They provide
the following estimates: α = β = 0.88 and λ = 2.25.
11
Tversky and Kahneman provide the following experimental estimates: γ = 0.61 and δ =
0.69.

C 2009 Blackwell Publishing Ltd
The CAPM is Alive and Well: A Review and Synthesis 55

dominates G(x),where F and G denote two prospects under consideration, w denotes the
initial wealth and x denotes the change of wealth. Thus, the claim of CPT that investors
make investment decisions based on change of wealth, rather than total wealth, has no
impact on the M-V efficient frontier. It is worth mentioning that the FSD efficient set
is also not affected by the initial wealth (for more details see Levy, 2006).
Now, let us turn to the more challenging difference between CPT and the CAPM.
Suppose that investors examine various portfolios, e.g. mutual funds or ETF’s. Denote
by F and G the cumulative distribution of two portfolios under consideration and by
F∗ and G∗ the transformed distributions, as suggested by the DW function represented
by Eq. (7). It is well known that if F and G are Normal, F∗ and G∗ are no longer
Normal; hence, with DW one cannot employ the M-V rule, as it is no longer an optimal
investment decision rule. Therefore, it is not clear whether the CAPM is intact.
Surprisingly, the competing paradigm of the M-V rule – the stochastic dominance
paradigm – ‘comes to the rescue’ of the M-V rule and the CAPM. To better perceive this,
recall that CPT’s DW are constructed such that they do not violate FSD. Suppose that F
and G are as presented in Figure 4a. As by assumption F and G are Normal, F dominates
G according to the M-V rule. Moreover, as the variances of F and G are identical, F also
dominates G according to FSD; therefore, such dominance is also intact for the value
function suggested by CPT (see Theorem 1 and Figure 4b). However, F∗ also dominates
G∗ according to CPT, since CPT does not violate FSD. 12 By the same token, for any
portfolio below the CML, there is a portfolio on the CML which dominates it by EUT
as well as CPT. Thus; all CPT investors will choose a portfolio located on the CML,
implying that the Separation Theorem and the CAPM are intact. The same results also
hold true as regards to Quiggin’s Rank Dependent Expected Utility, since according to
his suggested DW system, FSD is not violated.
Finally, note that while the compositions of the efficient M-V set, as well as the
SD efficient sets, are not affected by the investor’s initial wealth, the selection of the
optimal portfolio from the CML is generally affected by the initial wealth. However, this
selection does not affect the CAPM, as all selections are made from the M-V efficient
frontier, i.e., from the CML. Therefore, all investors hold a combination of the tangency
portfolio and the riskless asset.
To conclude, under both the CAPM and CPT (and the RDEU), all investors choose
their optimal portfolio from the CML. Although the optimal selected portfolio under
M-V and CPT are not necessarily identical, the Separation Theorem is intact; therefore,
the CAPM is intact - quite a surprising result.
Having noted this reconciliation between the CAPM and CPT, we can now return
to the Allais paradox and to Roy’s SF rule discussed above. Suppose that one can also
establish a paradox - similar to the Allais paradox - regarding Normal distributions (if
such a paradox does not hold true with Normal distributions, then Allais’s criticism of
EUT is irrelevant to the CAPM). The paradox is generally resolved once one employs
DW, when the weight of a very small probability dramatically increases. Suppose that
CPT’s decision weights solve the paradox. Then, one can use these weights which, of
course, contradict EUT. However, we have seen above that although EUT is contradicted
by CPT, the CAPM is still intact. Thus, using CPT’s decision weights may, on the one
hand, solve the paradox, while on the other hand it is consistent with the generalised

12
It is interesting to note that although F is located vertically above G, F∗ is not necessarily
located above G∗ . However, we proved that for any portfolio like G∗ (corresponding to
portfolio G), there is a portfolio F∗ on the CML, which dominates it by FSD.

C 2009 Blackwell Publishing Ltd
56 Haim Levy

CAPM framework, when convex as well as concave preferences together with DW are
allowed.
Going back to Roy’s SF rule: In the relevant case, where the preference presented
in Figure 3 represents the SF’s rule, we obtain that with Normal distributions the
M-V and the CAPM are intact, as they also hold true for non-concave functions, which
include - as a specific case - the preference presented in Figure 3. Thus, the SF rule and
the CAPM coexist with no need to impose the requirement for a positive investment in
the risky assets.

3.5 Overcoming Baumol and Leshno and Levy’s criticisms


As Baumol and Leshno and Levy raise a criticism which is similar in nature, we relate to
both criticisms in this section. Suppose that the following ideal situation exists favoring
the M-V and the CAPM: Distributions are Normal and investors maximise expected
utility. In this case, the M-V criterion is optimal; hence, under the additional standard
assumptions the CAPM is also valid. Even in this theoretically ideal case for the CAPM,
academics point to a drawback of the M-V analysis which, if valid, casts doubt on the
M-V efficiency analysis, hence casting doubt on the validity of the CAPM.
The first to show that the M-V rule may lead to a paradox is Baumol (1963). Baumol
gives an example of a choice of one of two prospects, such as:

Prospect G Prospect F
Mean, μ 5 50
Standard Deviation, σ 1 2

It is obvious that the M-V rule cannot determine dominance between these two
prospects; yet, Baumol claims that all rational investors will choose prospect F, because
even with the very low return observed with investment F, and the very large positive
return observed in G (say, 10 standard deviations to the left with Prospect F and ten
standard deviations to the right with Prospect G, an event with close to zero probability),
the investor will end up being better off by choosing F. Of course, there is no need for
such an extreme example to show that the M-V efficient set may also contain portfolios
which are, for all practical purposes, inefficient.
Thus, Baumol claims that some of the M-V efficient portfolios should be relegated
to the inefficient set. However, if one adjusts the M-V rule to eliminate a portion or
several portions of the M-V efficient set, there is a risk that the tangency portfolio will
be eliminated, which results in some ambiguous implications about the validity of the
CAPM.
While Baumol’s rule is based solely on intuition, in a recent study, Leshno and Levy
(2002) suggest new rules, called Almost Stochastic Dominance (ASD) and Almost M-V
(AMV), denoted by SD∗ and M-V∗ rules, respectively. These rules avoid paradoxes like
the one described above. Baumol published his paper before the CAPM was published;
hence, he analysed the implications relevant to the M-V efficient set. Leshno and Levy,
who published their paper in 2002, also analysed the implications relevant to the CAPM.
Moreover, regarding Normal distributions, they show that there are cases where neither
F nor G dominates the other; yet, they suspect that in practice such dominance exists.

C 2009 Blackwell Publishing Ltd
The CAPM is Alive and Well: A Review and Synthesis 57

Cumulative
Distribution
G

Return

Fig. 5a. Neither F nor G dominates the other by First degree Stochastic Dominance rule or
by Mean–Variance rule; yet, F dominates G by Almost First degree Stochastic Dominance
(FSD∗ ) rule and by Almost Mean Variance (M-V∗ ) rule.

Mean
U2

U1 F

Standard Deviation

Fig. 5b. Portfolios F and G in Figure 5a are shown in the M-V space. Utility U 1 is M-V
relevant, while it is M-V∗ irrelevant.

To understand this more clearly, consider the following two Normal distributions:
G ∼ N (μ, σ ) = N (1, 1)
F ∼ N (μ, σ ) = N (10, 2)
Though neither FSD nor M-V dominance exists, Leshno and Levy claim that after
removing economically irrelevant preferences, FSD∗ and M-V∗ dominance of F over G
exists.
As Figure 5 plainly shows, there is no dominance between the two distributions F
and G; yet, presumably all investors will select Prospect F. Leshno and Levy claim that
the SD rules are established for all possible mathematical preferences, while in practice
many of these preferences do not fit any realistic investor; hence, they are economically
irrelevant. This concept introduces a contradiction between the mathematical decision

C 2009 Blackwell Publishing Ltd
58 Haim Levy

rule (e.g. FSD) and investors’ realistic choices. To overcome this difficulty of the SD
rules, as well as the M-V rule, Leshno and Levy developed new rules which eliminated
irrelevant preferences. These new rules reveal a preference for option F, in the above
example, over option G, and this preference holds true for 100% of the investors.
Leshno and Levy analyse the FSD rule, while Baumol focuses on the M-V rule.
However, if the distributions are Normal, then the M-V and FSD rules coincide; 13 hence,
Leshno and Levy, like Baumol, eliminate a segment from the M-V efficient frontier,
which may affect the validity of the CAPM, especially if the market portfolio is located
on the relegated segment. In terms of Figure 5b, this implies that while, according to M-
V both F and G are efficient, according to M-V∗ , G is inefficient because the preference
that tangents to Point G, in practice, does not actually exist.
We show below that although Baumol and Leshno and Levy relegate some segment of
the M-V efficient set to the inefficient set, and despite the fact that the market portfolio
may be eliminated from the efficient set, when the riskless asset prevails the CAPM
is, once again, intact. Thus, the M-V efficiency analysis may be affected by Baumol
and Leshno and Levy’s reduction in the efficient set, while the CAPM is not. Let us
elaborate:
Baumol suggests the following investment rule instead of the M-V rule: F dominates
G if,
a) EF (x) ≥ EG (x) (8)
and
b) LF (x) = EF (x) − kσF ≥ LG (x) = EG (x) − kσG
where k is larger than 1.
As can be seen, the risk index is the lower floor L, rather than the standard deviation.
Of course, the higher the value L, the lower the risk involved with the investment under
consideration. By taking the first derivative of L with respect to E, and treating it as
a function of E, it can easily be shown that the lower portion of the M-V efficient set
is relegated to the inefficient set according to Baumol’s criterion. Namely, ∂ L/∂ E < 0
should hold true on the M-V frontier, which after some algebraic manipulations reduces
to ∂ E/∂σ < k. Thus, the lower the value k, the larger the segment of the frontier that
is relegated to the inefficient set will be. From this analysis it is clear that it is possible
that the market portfolio, which according to the CAPM is located on the efficient
frontier, will be inefficient by Baumol’s rule. Therefore, as per Baumol’s suggestion,
the Markowitz efficient set is divided into two segments: segment bc which is Baumol’s
efficient set and segment ab which is Baumol’s inefficient set. The market Portfolio m,
theoretically, can be located on either bc or ab (see Figure 6).
In Figure 6a, Portfolio m is located on what remains from the efficient set, while in
Figure 6b, it is not. Let us now add the riskless asset. If the tangency Portfolio m is
located on segment bc, the CAPM trivially also holds true with Baumol’s analysis; the
relegated segment is irrelevant because Portfolio m remains on the efficient segment
(See Figure 6a).
Now let us turn to the less trivial case, presented in Figure 6b. In this case, the tangency
Portfolio m is located on the inefficient frontier; hence, it is inefficient. We shall show

13
Actually, SSD and M-V coincide in the Normal case. However, FSD and M-V coincide
only in the case of two prospects with equal variance. As this is all that we need here (see
Theorem 1), we will use this equivalence in the rest of the paper.

C 2009 Blackwell Publishing Ltd
The CAPM is Alive and Well: A Review and Synthesis 59

Mean
r'

b
r

Standard Deviation

Fig. 6a. Segment ab of the M-V frontier is inefficient according to Baumol’s rule, where
Portfolio m remains on the Baumol efficient set given by segment bc.

Mean
r'

b
m

r
a

Standard Deviation

Fig. 6b. Segment ab is inefficient according to Baumol’s rule, where Portfolio m is located
on the inefficient segment bc.

below that with the riskless asset, Portfolio m is always efficient, regardless of whether
it is relegated to the inefficient asset or not in the case where the riskless asset does not
prevail. First, note that the inefficient set, according to Baumol, contains the segment
where the derivative ∂ E/∂σ > k. Therefore, the lower part of the efficient frontier may
be inefficient. Suppose that segment ab in Figure 6b is inefficient and segment bc is
efficient; hence, when the riskless asset does not prevail Portfolio m is inefficient.
We turn now to the case with a riskless asset. The M-V efficient set becomes line rr
of Figure 6b represented by the CML formula,
μm − r
μp = r + σp (9)
σm
where p stands for portfolios located on the CML represented by line rr .

C 2009 Blackwell Publishing Ltd
60 Haim Levy

According to Baumol, the efficient set is still defined by the derivative ∂ E/∂σ < k.
However, on the CML, the derivative is constant and represented by,
μm − r
= constant (10)
σm
There are two potential possibilities: 1) k is selected, such that on the CML, the
derivative ∂ L/∂ E > k (which also implies that ∂ E/∂σ > k). Namely, as we move to the
right on the CML, both E and L increase; therefore, each point on the CML dominates
all points with a lower mean, a case wherein infinite borrowing is optimal. 2) The
derivative on the CML fulfills the condition ∂ E/∂σ < k. Hence, all points on the CML
are also efficient according to Baumol’s rule. Of course, only Case 2 is consistent
with equilibrium. Indeed, Case 2, which does not contradict equilibrium, holds true
in practice. To understand this more clearly, recall that we have approximately the
following empirical estimates of the CML parameters: the mean annual return on the
market portfolio is about 12% with a standard deviation of about 20%. The long-run
riskless interest rate is about 4% (see Ibbotson, 2007); hence, the slope of the CML is
about 0.40. Recalling that Baumol’s k must be greater than 1.0, it is obvious that the
derivative requirement for Bamuol’s efficiency holds true on the CML (∂ E/∂σ < k);
hence, all portfolios located on the CML are efficient according to Markowitz-Sharpe
and Baumol, alike. 14
Let us now turn to Leshno and Levy’s claim that, once again, the classic M-V efficient
set includes too many portfolios, which may relegate the tangency portfolio to the
inefficient set. This is similar to Baumol’s claim; however, the analysis and the economic
setting are different. Leshno and Levy suggest eliminating irrelevant preferences; hence,
the market portfolio may be inefficient. To be more specific, they define a new set of
preferences which is a subset of the preferences assumed by the M-V analysis. For
example, preference U 1 (see Figure 5b) is considered irrelevant by Leshno and Levy,
while it is a legitimate preference according to M-V analysis. Leshno and Levy may
eliminate preference U 1 because Portfolio F dominates G according to the M-V∗ rule,
though no such dominance exists according to the M-V rule.
To better understand this claim, refer back to Figure 5, where neither F nor G dominates
the other according to M-V; although F may dominate G according to MV∗ (for more
details, see Leshno and Levy, 2002). As in the case of Baumol, Leshno and Levy also
suggest the possibility that the market portfolio is relegated to the inefficient set. While
this is the case when the riskless asset is not available, once again, its existence ‘rescues’
the CAPM, even in Leshno and Levy’s framework (see Figure 7).
The segment below point b may be relegated by Leshno and Levy to the inefficient
set according to the M-V∗ rule. Hence, Portfolio m may be inefficient. Yet, when one
adds the riskless asset, Portfolio m or some combination of m with the riskless asset
becomes efficient. This is because for any portfolio located on a line below rr , i.e.
located on rr , there is a portfolio on rr with the same variance and a higher mean;
hence, it is dominated by SD, as well as by MV (recall that Normal distributions are
assumed). Furthermore, as SD implies SD∗ (and M-V implies M-V∗ , see Leshno and
Levy 2002) the dominance also holds true according to Leshno and Levy’s rules. Thus,
all portfolios located below rr are MV∗ inefficient.

14
It is interesting to note that without the riskless asset portfolio m may be theoretically
inefficient. However, as at this point the slope is equal to the CML’s slope, with the existing
empirical estimates of the various parameters we find that portfolio m is efficient.

C 2009 Blackwell Publishing Ltd
The CAPM is Alive and Well: A Review and Synthesis 61

r' r''
Mean

m'

m b

Standard Deviation

Fig. 7. The M-V and M-V∗ efficient frontiers: Left to point b is the M-V∗ inefficient
segment, where m is located on this segment.

Unlike the case of Baumol, in the case of Leshno and Levy not all portfolios located
on rr are necessarily efficient. It is possible that Portfolio m is inefficient because
Portfolio m dominates it by MV∗ . However, as m is a linear combination of m and
the riskless asset, all investors will end up holding a combination of m and the riskless
asset. As a result, the CAPM is also valid when investors employ either the SD∗ or
M-V∗ rule. Finally, even if the riskless asset does not exist, according to both Baumol
and Leshno and Levy, the optimal portfolios will be selected from the reduced M-
V frontier, and in equilibrium Black’s (1972) Zero Beta CAPM holds true. This also
occurs because according to Baumol’s criterion and M-V∗ , all interior portfolios are
always inefficient.

4. Overcoming the Empirical Criticism of the CAPM

Let us examine each criticism separately:

4.1 The normality assumption


It is well known that Normality (or an Elliptic distribution) is very crucial to the
derivation of the CAPM. We also assume Normality by showing the validity of the
CAPM in various scenarios. Numerous studies examine the Normality hypothesis with a
clear cut result: the Normality of the return distributions is statistically strongly rejected.
The primary findings of these studies show that the empirical distributions are skewed
and that they have fatter tails, relative to the Normal distribution. Prominent studies on
this topic include Mandelbrot (1963), Fama (1965), Officer (1972), Clark (1973), Gray
and French (1990), Zhou (1993), Mantegna and Stanley (1995), Focardi and Fabozzi
(2003), and Levy and Duchin (2004).
As regards to finding the distribution which best fits best the empirical data, Harvey
et al. (2002) found that the skewed Normal provides the best fit, while Levy and Duchin
(2004) discovered that the best fit depends on the investment horizon, with the Logistic

C 2009 Blackwell Publishing Ltd
62 Haim Levy

distribution providing the best fit when monthly data is employed. However, they also
found that for longer investment horizons, other distributions best fit the data with the
Lognormal distribution, as expected, providing the best fit for the longest investment
horizon studied in their paper (four years). 15 As the CAPM strongly relies on the
Normality assumption, we next analyse the effect of the departure from Normality
on the CAPM validity.
The M-V rule is justified as an approximation to expected utility, even when
distributions are not Normal on two grounds:

• Levy and Markowitz (1979) have shown that with empirical return distributions which
are not normal, the M-V rule is an excellent approximation for expected utility (see
also Kroll et al., 1984 and Markowitz, 1991). Kroll et al. and Markowitz show that
virtually all relevant risk-averse utility functions can be approximated locally by a
quadratic function; hence, a direct and precise expected utility maximisation yields
almost the same expected utility obtained by selecting the best portfolio only from
the set of efficient portfolios located on the M-V efficient frontier. This conclusion is
valid as long as the range of returns is not too wide. Thus, Levy and Markowitz claim
that in most cases one can safely use the M-V rule, despite the statistical rejection of
Normality. The utility loss induced by adopting this approach has been empirically
found to be negligible. Other studies directly estimate the financial loss, rather than the
utility loss, due to the assumption of Normality, when the empirical rates of return are
not really distributed Normally. The procedure employed is as follows: First, maximise
the expected utility directly and find the certainty equivalent of the selected portfolio
investment. Then, once again, maximise expected utility, but this time by assuming
Normality. Hence, by construction, a lower certainty equivalent is obtained because
in practice distributions are not Normal. As a result, one deviates from the optimal
investment diversification. The difference between the two certainty equivalent sums
measures the financial loss induced by the deviation of the empirical distribution from
Normality. For these types of studies, which make different assumptions regarding
the investor’s preference, see Simaan (1993), Tew et al. (1991), and Duchin and Levy
(2008).

All these studies reveal that the financial loss is negligible, despite the strong
statistical rejection of Normality. For example, Duchin and Levy, who employ the myopic
preference, find that the loss per $10,000 investment is merely $2-$6, depending on the
degree of the relative risk-aversion parameter. To put things in perspective, suppose that
the planned investment horizon is one year. The mean rate of return on risky assets is
about 12% (see Ibbotson, 2007). Then, a loss of $2-$6 per $10,000 investment implies
that the mean rate of return drops, on average, from 12% to 11.94%-11.98%, a negligible
loss. 16

15
With a very long horizon, the distributions become positively skewed and approaching the
Lognormal distribution. For the role of skewness in efficiency analysis, see Arditti and Levy
(1975) and Harvey and Siddique (2000).
16
One may claim that even this small loss will induce the investor to select her portfolio
by expected utility maximisation; hence, an interior portfolios is selected, a case where the
CAPM does not hold true. However, for this scenario to hold true, one must assume the
existence of very sophisticated investors who select their portfolio according to a complex
method, rather than the simple M-V method.

C 2009 Blackwell Publishing Ltd
The CAPM is Alive and Well: A Review and Synthesis 63

Finally, recall that the direct optimal diversification - without the Normality assump-
tion - is the correct one and any optimisation with Multivariate-Normality assumption (as
was done by Duchin and Levy, 2008) induces a loss, since the distribution of empirical
distribution is, in practice, not Multivariate-Normal. However, the two calculations yield
identical results: i.e. a zero loss is involved, when the preference is quadratic, or when
the empirical distributions are indeed Multivariate-Normal. The surprising result is
that although the Multivariate-Normal distribution is strongly statistically rejected, the
financial loss induced by making the normality assumption is minimal, and one can
safely assume Normality and employ the M-V criterion for investment selection - a very
encouraging result from the point of view of the CAPM.

4.2 The CAPM – empirical and experimental tests


Classical economists employ empirical test to examine the validity of a given theory.
Behavioural economists and psychologists rely on laboratory experiment. We show
in this section that with ex ante parameters the CAPM can not be rejected. Even
more important, employing behavioural economics approach the CAPM has a strong
experimental support.
Shortly after the theoretical model of the CAPM was published, testing the model
empirically was in fashion and was a subject researched by top researchers in finance.
From these empirical studies, one is tempted to conclude that the CAPM is empirically,
questionable, which allegedly drastically reduces its value.
We claim below that the empirical results are inconclusive: Although the CAPM
is rejected with ex–post parameters, it cannot be rejected with ex ante parameters,
in particular with ex ante beta. Recall that the CAPM is defined by Sharpe and
Lintner in terms of the ex ante, rather than ex post, parameters. Unfortunately, in
the empirical studies ex post parameters are employed, simply because the ex ante
parameters are unknown. Obviously, there are differences between the ex post and the
ex ante parameters, and it can be shown that with small changes in the parameters -
which take into account these possible differences - the CAPM cannot be rejected.
In the early CAPM tests, the first-pass and second-pass regressions were defined as
follows:
First-Pass : Rit = αi + βi Rmt + eit (11)

Second Pass : R̄i = γ0 + γ1 β̂i + δi (12)


where: R it and R mt stand for the rate of return on the ith asset and the market portfolio
during Period t, respectively, while β̂i is the estimate of the ith asset risk, as estimated
in the first-pass regression. 17
The CAPM relates to the expected rate of return of each security i and to its risk as
follows:
μi = r + (μm − r )βi (12a)

17
The above descrbed empirical test of te CAPM is the old static test. Of course,with decades
of empirical tests there are more sophisticated tests, like the conditional CAPM test where
betas are allowed to change overtime. For extension of the CAPM test see for example,
Cochrane (2001), Roll and Ross (1994) Kandel and Stambaugh (1995), Levhari and Levy
(1977), Ammann and Verhofen (2007) and Schrimpf et al. (2007).

C 2009 Blackwell Publishing Ltd
64 Haim Levy

where r is the risk-free interest rate and μm is the expected rate of return on the market
portfolio. If the CAPM is intact, we expect that γ̂0 will not be significantly different from
the risk free interest rate and that γ̂1 will not be significantly different from μm − r . 18
Virtually all empirical tests reveal that this is not the case; therefore, the CAPM is
rejected. Moreover, the R2 is generally low - about 4% with monthly data and about
20% with annual data - as long as individual assets, rather than portfolios, are considered,
as required by the CAPM tests (see Levy, 1978). 19
We now turn to the difference between ex post and ex ante parameters. There are
several approaches for incorporating these differences. Generally, it can be shown that
accounting for even small possible differences between ex post and ex ante parameters
the CAPM cannot be rejected. Let us elaborate:

4.2.1. Ex ante beta. In testing the CAPM (see Eq. (12)), it is assumed that beta,
estimated by Eq. (11), is the correct ex ante beta. We claim that taking into account
possible differences between the ex post and ex ante beta, the CAPM cannot be rejected.
Indeed, Levy (1983) tested the CAPM when such differences were taken into account. He
showed that regardless of the possible measurement errors involved in the measurement
of rates of returns, the CAPM cannot be rejected on an ex ante basis. This is clear if
we simply recall that the beta used in the second-pass regression is an estimate of the
unknown true ex ante beta. Using joint confidence intervals for all betas, and taking into
account this possible difference between ex ante and ex post betas, he showed that the
CAPM cannot be rejected, and that an almost perfect second-pass regression is obtained
with a possible ex ante vector of betas. 20
Finally, even with the ex post data, the CAPM empirical results are not always negative.
Actually, the results depend on the sub-period selected for such testing. For example, for
the ten portfolios ordered by their book to market ratio (B/M), Fama and French (2004)
report a negative relationship between average return and beta for the period 1963–2003
(see Figure 8a). However, using the same data, Levy (2008b) reports a strong positive
relationship between these two variables for the sub-period 1927-1962 (see Figure 8b),
and even a stronger relationship (with 84% explanation power) for the whole period
1927–2007 (see Figure 8c). Thus, even with ex post data, the empirical results remain
ambiguous.

4.2.2. The efficiency of the market portfolio. In a breakthrough article, Roll (1977)
showed that with the above procedure of testing the CAPM, the only relevant CAPM
test is to examine whether the employed market portfolio is M-V efficient. If it is,
then in the second-pass regression the CAPM would reveal a perfect fit, and this is
a technical result. Moreover, the fact that in the empirical tests a less than perfect fit
is obtained merely indicates that the market portfolio is not M-V efficient; hence, it
is commonly concluded that the CAPM is invalid. Taking into account the possible
difference between ex post and ex ante parameters, in a recent paper, Levy and Roll
(2008) show that when only small changes in the sample means and standard deviations
18
When excess returns are employed γ0 is expected not to be significantly different from
zero.
19
However, to avoid some statistical errors, some studies employ the rates of return on
portfolios rather than individual assets.
20
The standard CAPM regressions take into account possible differences between sample
and population means, but the sample beta is assumed to be the correct beta.

C 2009 Blackwell Publishing Ltd
The CAPM is Alive and Well: A Review and Synthesis 65

Fig. 8. The regression results for ten portfolios ordered by book to market ratio (B/M), using
K. French data base. (For more details, see http://mba.tuck.dartmouth.edu/pages/faculty/
ken.french/data library.html)


C 2009 Blackwell Publishing Ltd
66 Haim Levy

are applied, the observed market portfolio is M-V efficient, which according to Roll
(1977) implies that the linear CAPM is intact. They employ a novel ‘reverse engineering’
approach. To be more specific, they take a given market proxy portfolio and investigate
the minimal necessary changes that must be applied to the parameters, such that this
portfolio will be located on the M-V efficient frontier. Surprisingly, they find that only
small changes are required, well within the allowed differences between sample means
and true ex ante means. Therefore, they conclude that one cannot reject the market
portfolio efficiency; hence, one cannot reject the CAPM. In their words:

‘Surprisingly, slight variations of the sample parameters, well within the estimation
error bounds, suffice to make the proxy efficient. Thus, many conventional market
proxies could be perfectly consistent with the CAPM’.

4.2.3. Negative investment weigh. Using historical rates of return to derive the M-V
efficient set, it is generally found that some of the weights are negative; of course, with a
small number of assets, e.g. 3–5 assets, it is possible to find that all weights are positive.
However, in the relevant case for testing the CAPM, where hundreds if not thousands
of assets need to be incorporated, negative investment weights always exists. Moreover,
the percentage of negative weights becomes close to 50% of the assets included in the
study, when the number of assets increases. Furthermore, one does not need to have an
astronomically large number of assets to obtain this result: Levy (1983) shows that even
with 15 assets the percentage of negative weights is about 50%. Thus, having negative
weights on the efficient frontier - and in particular about 50% negative weights, when
the market portfolio by definition is composed of only positive weights - is disturbing
because it indicates that the market portfolio is interior to the M-V efficient frontier.
Once again, according to Roll (1977), this is evidence against the validity of the
CAPM. 21 This empirical evidence attacks the CAPM in a similar way to the previous
point (see Point 3 above), but the attack comes from a slightly different angle. Therefore,
it is not surprising that the remedy, i.e. ‘saving’ the CAPM from this attack, is very
similar to the previous point, as suggested by Levy and Roll (2008). Indeed, in a recent
paper, Levy (2008a) conducts a technique to examine whether with ex ante data it is
possible that all weights will be positive, even though with ex post data, we find negative
weights corresponding to portfolios located on the M-V frontier. He shows that with
small changes in the parameters to account for the possible difference between ex post
and ex ante parameters, a tangency portfolio with only positive weights is obtained.
Levy shows that the fact that empirically negative weights are almost everywhere does
not constitute any evidence against the CAPM. Moreover, this result is rather expected
in the CAPM framework. To cite from his paper,

‘We show that the probability of obtaining a positive tangency portfolio based
on sample parameters converge to zero exponentially with the number of assets.
However, at the same time, very small adjustments in the return parameters, well

21
Levy (2008) has shown that when there is a reasonable heterogeneity in the end of period
distributions of the individual risky assets then positive prices imply that the risk premium
must be close to zero, an unacceptable result. He suggests that in such a case the generalised
segmented CAPM of Levy (1978) and Merton (1987) may serve as a substitute to the CAPM.
Once again, some generalised form of the CAPM is intact.

C 2009 Blackwell Publishing Ltd
The CAPM is Alive and Well: A Review and Synthesis 67

within the estimation error, yield a positive tangency portfolio. Hence, looking for
positive portfolios in parameter space is somewhat like looking for rational numbers
on the number line: if a point in the parameter space is chosen at random it almost
surely corresponds to non-positive portfolio (an irrational number); however, one
can find very close points in parameter space corresponding to positive portfolios
(rational numbers)’.

This finding sheds light on the empirical results, which consistently show negative
weights corresponding to M-V efficient portfolios: an infinite number of portfolios with
positive weights exist; however, there is zero chance to discover them - quite an amazing
result. Thus, the empirically obtained negative weights cannot prove that the CAPM is
invalid. The positive tangency portfolios indeed exist - in large numbers - but we cannot
find them.

4.2.4. Experimental studies of the CAPM. Finally, in the above analysis an effort has
been made to account for the possible difference between ex post and ex ante parameters.
Can we test the CAPM with ex- ante parameters? Empirically, we cannot, as the ex ante
parameters are not available. However, one can set up an experiment wherein the subjects
simultaneously determine the mean rate of return, beta and equilibrium prices; hence,
the CAPM can be tested experimentally with ex ante parameters. For example, suppose
that the end of a period return is given by $100(1+R), where the return, R, is a random
variable whose distribution is given to the subjects. What price is one willing to pay
for this asset today? At what price is one willing to sell such an asset today? Having
many assets similar to this one - and many subjects - one can look at the aggregate
demand and aggregate supply and find the equilibrium prices which clear the market.
Note that by determining the equilibrium prices one also simultaneously determines the
mean return and beta of each asset, and these parameters are ex ante parameters. Using
this technique, Levy (1997) found strong support for the CAPM with more than 70%
explanatory power. Levy concludes,

‘. . . mean return and risk are strongly positively related when these parameters are
determined on an ex ante basis, as claimed by the Sharpe-Lintner model’ (see p.145).

Bossaerts and Plott (2002), also found experimental support for the CAPM. In their
words,

‘when interpreted as the equilibrium to which a complex financial market system


has a tendency to move, the CAPM received support in the experiments reported
here’ (see p. 1110).

To sum up, empirically we have no evidence that rejects the CAPM, and experimen-
tally, with ex ante parameters, we strongly support it.

5. Concluding Remarks

The M-V efficiency analysis of Markowitz (1952) and the Sharpe-Lintner CAPM are
both derived from the expected utility framework, which was long ago criticised by both
Allais (1953) and Roy (1952). Even more severe criticisms of expected utility come from

C 2009 Blackwell Publishing Ltd
68 Haim Levy

behavioural economists and psychologists who claim that investors are not rational
‘efficient machines’ and systematically deviate from expected utility maximisation.
Thus, if one accepts the above criticisms - and especially if the Cumulative Prospect
Theory of Tversky and Kahneman (1992) is intact - then the Expected Utility Theory
(EUT) is invalid. Hence, the M-V and CAPM which are derived from the expected
utility framework lose ground.
Based on reasonable subjects’ choices and employing behavioural economic argu-
ments, Baumol (1963), and Leshno and Levy (2002) reveal some drawbacks of the M-V
criterion. They show that it cannot distinguish between two prospects, even when 100%
of the subjects have a clear cut choice. This criticism implies that a segment of the
M-V efficient frontier is relegated to the inefficient set, and if the market portfolio is
located on the relegated segment, the CAPM may be invalid. Thus, based on inventors’
behaviour the M-V is questionable, hence the CAPM is questionable.
In this study, we show that the CAPM is surprisingly valid in behaviour economics and
psychologists paradigms even though expected utility is invalid. We first show that, due
to the correct criticisms of Baumol (1963) and Leshno and Levy (2002), the M-V rule
is modified; hence, indeed some portions of the M-V efficient set become inefficient.
However, even with these modifications the CAPM is theoretically intact. Assuming
Normality (but not risk aversion) and using First degree Stochastic Dominance (FSD),
we theoretically show that the CAPM is also valid in Roy’s framework, as well as in
CPT’s framework, i.e. even when a non-concave preference, which is a function of
change of wealth and decision weights, are allowed.
The CAPM can not be rejected also on empirical ground when ex ante rather than ex
post parameters are employed. These ex ante parameters can be obtained experimentally
but not empirically. Thus, not only that the CAPM is not refuted in behavioural economic
framework, it gets its strong support within this framework.
It is interesting to discover that the M-V investment rule and the CAPM are robust
under wide possible frameworks, even when the empirical distributions of rates of
return are not Normal, let alone when Normality prevails. These are very encouraging
results: We therefore conclude that one can safely continue to use the CAPM in
academic research and in practice, as it cannot be refuted, and is even strongly supported
experimentally with ex ante parameters. The difficult issue is, of course, how to estimate
the ex ante parameters; however, this cannot be counted as a disadvantage of the CAPM,
because virtually all theoretical models encounter this issue.

References

Allais, M., ‘Le comportement de l’homme rationnel devant le risque: critique de postulats et axioms
de l’ecole Americaine’, Econometrica, Vol. 21, 1953, pp. 503–46.
Ammann, M. and Verhofen, M., ‘Testing conditional asset pricing models using a Markov chain Monte
Carlo approach’, European Financial Management, Vol. 14, 2007, pp. 391–418.
Arditti, F. and Levy, H., ‘Portfolio efficiency analysis in three moments: the multi-period case’, Journal
of Finance, Vol. 30, 1975, pp. 137–50.
Arrow, K.J., Aspects of the Theory of Risk-Bearing (Helsinki: Yrjö Hahnsson Foundation,
1965).
Arzac, E.R. and Bawa, V.S., ‘Portfolio choice and equilibrium in capital markets with safety-first
investors’, Journal of Financial Economics, Vol. 4, 1977, pp. 277–88.
Baumol, W.J., ‘An expected gain-confidence limit criterion for portfolio selection’, Management
Science, Vol. 10, 1963, pp. 174–82.


C 2009 Blackwell Publishing Ltd
The CAPM is Alive and Well: A Review and Synthesis 69

Bawa, V.S., ‘Safety-first, stochastic dominance, and optimal portfolio choice’, Journal of Financial
and Quantitative Analysis, Vol. 13, 1978, pp. 255–71.
Berk, J.B., ‘Necessary conditions for the CAPM’, Journal of Economic Theory, Vol. 73, 1997,
pp. 245–57.
Best, M.J. and Grauer, R., ‘Positively weighted minimum-variance portfolios and the structure of asset
expected returns’, Journal of Financial and Quantitative Analysis, Vol. 27, 1992, p. 4.
Black, F., ‘Capital market equilibrium with restricted borrowing’, Journal of Business, Vol. 45, 1972,
pp. 444–55.
Bossaerts, P. and Plott, C., ‘The CAPM in thin experimental financial markets’, Journal of Economic
Dynamics & Control, Vol. 22, 2002, pp. 1093–1112.
Chamberlain, G., ‘A characterization of the distributions that imply mean-variance utility functions’,
Journal of Economic Theory, Vol. 29, 1983, pp. 185–201.
Clark, P., ‘A subordinated stochastic process model with finite variance for speculative prices’,
Econometrica, Vol. 41, 1973, pp. 135–55.
Cochrane, J., Asset Pricing (Princeton, NJ: Princeton University Press, 2001).
Duchin, R. and Levy, H., ‘Yes, you can assume normality’, Working Paper (Hebrew University, 2008).
Fama, E.F., ‘The behavior of stock market price’, Journal of Business, Vol. 38, 1965, pp. 34–105.
Fama, E.F. and French, K.R., ‘The capital asset pricing model: theory and evidence’, Journal of
Economic Perspectives, Vol. 18, 2004, pp. 25–46.
Focardi, S.M. and Fabozzi, F.J., ‘Fat tails, scaling, and stable laws: a critical look at modeling external
events in financial phenomena’, Journal of Risk Finance, Vol. 5, 2003, pp. 1–22.
Friedman, M. and Savage, L.J., ‘The utility analysis of choices involving risk’, Journal of Political
Economy, Vol. 56, 1948, pp. 279–304.
Goetzmann, W.N. and Broadie, M., ‘Safety first portfolio choice’, Working Paper (Columbia First
Boston Series in Money, Economics and Finance, 1992).
Gray, B. and French, D., ‘Empirical comparisons of distributional models for stock index returns’,
Journal of Business, Finance and Accounting, Vol. 17, 1990, pp. 451–9.
Green, R.C., ‘Positively weighted portfolios on the minimum-variance frontier’, Journal of Finance,
Vol. 41, 1986, pp. 1051–68.
Hadar, J. and Russell, W., ‘Rules for ordering uncertain prospects’, American Economic Review,
Vol. 59, 1969, pp. 25–34.
Hanoch, G. and Levy, H., ‘The efficiency analysis of choices involving risk’, Review of Economic
Studies, Vol. 36, 1969, pp. 335–46.
Harvey, C.R., Liechty, J.C., Liechty, M.W. and Müller, P., ‘Portfolio selection with higher moments’,
Working Paper (Duke University, 2002).
Harvey, C.R. and Siddique, A., ‘Conditional skewness in asset pricing tests’, Journal of Finance,
Vol. 55, 2000, pp. 1263–95.
Ibbotson Associates, Stocks, Bonds, Bills and Inflation 2007 Yearbook (Chicago, Illinois: Morningstar,
2007).
Kahneman, D. and Tversky, A., ‘Prospect theory: an analysis of decision under risk’, Econometrica,
Vol. 47, 1979, pp. 263–91.
Kandel, S. and Stambaugh, R., ‘Portfolio inefficiency and the cross-section of expected returns’,
Journal of Finance, Vol. 50, 1995, pp. 157–84.
Kroll, Y. and Levy, H., ‘Stochastic dominance: a note’, Journal of Finance, Vol. 37, 1982, pp. 871–
75.
Kroll, Y., Levy, H., and Markowitz, H., ‘Mean-variance versus direct utility maximization’, Journal
of Finance, Vol. 39, 1984, pp. 47–61.
Kroll, Y., Levy, H. and Rapoport, A., ‘Experimental tests of the separation theorem and the capital
asset pricing model’, American Economic Review, Vol. 78, 1988, pp. 500–19.
Leshno, M. and Levy, H., ‘Preferred by all and preferred by most decision makers: almost stochastic
dominance’, Management Science, Vol. 48, 2002, pp. 1074–85.
Levhari, D., and Levy, H., ‘The capital asset pricing model and the investment horizon’, Review of
Economics and Statistics, Vol. 59, 1977, pp. 92–104.


C 2009 Blackwell Publishing Ltd
70 Haim Levy

Levy, H., ‘A utility function depending on the first three moments’, Journal of Finance, Vol. 24, 1969,
pp. 715–19.
Levy, H., ‘A test of the CAPM via a confidence level approach’, Journal of Portfolio Management,
Vol. 9, 1982, pp. 56–61.
Levy, H., ‘The capital asset pricing model: theory and empiricism’, Economic Journal, Vol. 93, 1983,
pp. 145–65.
Levy, H., ‘Risk and rerun: An experimental analysis’, International Economic Review, Vol. 38, 1997,
pp. 119–49.
Levy, H., Stochastic Dominance: Investment Decision Making under Uncertainty (New York: Springer,
2006).
Levy, H. and Duchin, R., ‘Asset returns’ distributions and the investment horizon’, Journal of Portfolio
Management, Vol. 30, 2004, pp. 44–62.
Levy, H. and Markowitz, H.M., ‘Approximating expected utility by a function of mean and variance’,
American Economic Review, Vol. 68, 1979, pp. 643–58.
Levy, H. and Sarnat, M., ‘The mean variance criterion and the efficient frontier’, Western Economic
Journal, Vol. 9, 1971, pp. 46–51.
Levy, H. and Sarnat, M., ‘Safety first – an expected utility principle’, Journal of Financial and
Quantitative Analysis, Vol. 7, 1972, pp. 1829–34.
Levy, H. and Levy, M., ‘The safety first expected utility model: experimental evidence and economic
implications’, Journal of Banking and Finance, Vol. 33, 2009, pp. 1494–1506.
Levy, M., ‘Conditions for a CAPM equilibrium with positive prices’, Journal of Economic Theory,
Vol. 137, 2007, pp. 404–15.
Levy, M., ‘Positive portfolios are all around’, Working Paper (Hebrew University, 2008a).
Levy, M., ‘CAPM risk-return tests and the length of the sampling period’, Working Paper (Hebrew
University, 2008b).
Levy, M. and Roll, R., ‘The market portfolio may be efficient after all’, Working Paper (UCLA, 2008).
Lintner, J., ‘Security prices, risk, and the maximal gains from diversification’, Journal of Finance,
Vol. 20, 1965, pp. 587–615.
Machina, M.J., ‘Review of generalized expected utility theory: the rank dependent model’, Journal of
Economic Literature, Vol. 32, 1994, pp. 1237–38.
Mandelbrot, B., ‘The variation of certain speculative prices’, Journal of Business, Vol. 36, 1963,
p. 394.
Mantegna, R.N. and Stanley, H.E., ‘Scaling behavior in the dynamics of an economic index’, Nature,
Vol. 376, 1995, pp. 46–49.
Markowitz, H.M., ‘Portfolio selection’, Journal of Finance, Vol. 7, 1952a, pp. 77–91.
Markowitz, H.M., ‘The utility of wealth’, Journal of Political Economy, Vol. 60, 1952b, pp. 151–56.
Markowitz, H.M., ‘Foundations of portfolio theory’, Journal of Finance, Vol. 46, 1991, pp. 469–77.
Merton, R.C., ‘An analytic derivation of the efficient portfolio frontier’, Journal of Financial and
Quantitative Analysis, Vol. 7, 1972, pp. 1851–72.
Milevsky, M.A., ‘Time diversification, safety-first and risk’, Review of Quantitative Finance and
Accounting, Vol. 12, 1999, pp. 271–81.
Officer, R.R. ‘The distribution of stock returns’, Journal of the American Statistical Association,
Vol. 67, 1972, pp. 817–21.
Payne, J.W., Laughhunn, D.J. and Crum, R., ‘Translation of gambles and aspiration level effects in
risky choice behavior’, Management Science, Vol. 26, 1980, pp. 1039–60.
Payne, J.W., Laughhunn, D.J. and Crum, R., ‘Further tests of aspiration level effects in risky choice
behavior’, Management Science, Vol. 27, 1981, pp. 953–8.
Pyle, D.H. and Turnovsky, S.J., ‘Safety-first and expected utility maximization in mean-standard
deviation portfolio analysis’, Review of Economics and Statistics, Vol. 52, 1970, pp. 75–81.
Quiggin, J., ‘A theory of anticipated utility’, Journal of Economic Behavior and Organization,
Vol. 3, 1982, pp. 323–43.
Quiggin, J., Generalized Expected Utility Theory: the Rank Dependent Model (Boston: Kluwer
Academic Publishers, 1993).


C 2009 Blackwell Publishing Ltd
The CAPM is Alive and Well: A Review and Synthesis 71

Roll, R., ‘A critique of the asset pricing theory’s test, part I: on past and potential testability of theory’,
Journal of Financial Economics, Vol. 4, 1977, pp. 129–76.
Roll, R. and Ross, S., ‘On the cross-sectional relation between expected returns and betas’, Journal of
Finance, Vol. 49, 1994, pp. 101–21.
Ross, S.A., ‘The arbitrage theory of capital asset pricing’, Journal of Economic Theory, Vol. 13, 1976,
pp. 341–60.
Rothschild, M. and Stiglitz, J.E., ‘Increasing risk. I. A definition’, Journal of Economic Theory,
Vol. 2, 1970, pp. 225–43.
Roy, A.D., ‘Safety-first and the holding of assets’, Econometrica, Vol. 20, 1952, pp. 431–49.
Roy, A.D., ‘Risk and Rank or Safety First Generalized’, Economica, Vol. 23, 1956, pp. 214–28.
Rudd, A., ‘A note on qualitative results for investment proportions’, Journal of Financial Economics,
Vol. 4, 1977, pp. 277–88.
Schrimpf, A., Schroder, M. and Stehle, R., ‘Cross-sectional tests of conditional asset: Evidence from
the German stock market’, European Financial Management, Vol. 13, 2007, pp. 880–907.
Sharpe, W.F., ‘Capital asset prices: a theory of market equilibrium’, Journal of Finance, Vol. 5, 1964,
pp. 259–63.
Sharpe, W.F., ‘Mutual fund performance’, Journal of Business, Vol. 39, 1966, pp. 119–38.
Simaan, Y., ‘The opportunity cost of mean-variance investment strategies’, Management Science,
Vol. 39, 1993, pp. 578–87.
Swalm, R.O., ‘Utility theory – insights into risk taking’, Harvard Business Review, Vol. 44, 1966,
pp. 123–36.
Tew, B., Reid, D. and Witt, C., ‘The opportunity cost of mean-variance efficient choice’, Financial
Review, Vol. 26, 1991, pp. 31–43.
Tversky, A. and Kahneman, D., ‘Advances in prospect theory: cumulative representation of uncer-
tainty’, Journal of Risk and Uncertainty, Vol. 5, 1992, pp. 297–323.
von Neumann, J. and Morgenstern, O., Theory of Games and Economic Behavior, 3rd ed. (Princeton,
N.J.: Princeton University Press, 1953).
Zhou, G., ‘Asset-pricing tests under alternative distributions’, Journal of Finance, Vol. 48, 1993,
pp. 1927–42.


C 2009 Blackwell Publishing Ltd
Copyright of European Financial Management is the property of Blackwell Publishing Limited and its content
may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express
written permission. However, users may print, download, or email articles for individual use.

Das könnte Ihnen auch gefallen