POLITECNICO DI TORINO

Corso di Laurea in Engineering and Management

Tesi di Laurea Magistrale

Theoretical analysis and comparison of CAPM and APT and

their roles in risk management

Relatore: Candidata:
Prof.ssa Elisa Ughetto Gunel Taghizade

December 2021

1
 Abstract

Financial markets are constantly changing and becoming more complex, and everyone involved tries to
understand the causes of unpredictable fluctuations. With the volume and variety of financial products on
the market today, it is more difficult than ever to predict the outcome of an investment. Investors are risk-
averse. That is, they are willing accept higher risk only if you are expecting a higher return.
Unfortunately, it is not possible to actually observe the risks and expected returns associated with
financial instruments. This is because financial instruments are constantly affected by the individual
institutions that issue financial instruments and the unlimited number of events that affect their sectors. To
the economy or the whole world. However, while it is possible to estimate them using statistical
methods and historical records of securities performance, these estimates are not always accurate and it
is certain that new and unexpected events will occur at some point.

Throughout the years, many scholars have tried to come up with theories that could reflect the market
fluctuations. The two most focused models of asset pricing are Capital asset pricing model and the
Arbitrage pricing theory. These models try to demonstrate the relationship between risk and return. Many
empirical studies have been conducted on these two models to check if they were truly functional. In
different years, and in different parts of the world, scholars have obtained various results.

At some points, certain methodologies performed in the testing of these models, have lead to biased data,
and they have been evolved and become more effective. The purpose of this study is to discuss the
methodologies used to conduct empirical studies on CAPM and APT and their differences have been
examined.

2
 Theoretical analysis and comparison of CAPM and APT and their roles
in risk management
Contents
Abstract ........................................................................................................................................... 1
1. Literature overview.................................................................................................................. 4
1.1. Modern portfolio theory ................................................................................................... 4
1.2. The Capital Assets Pricing Model .................................................................................... 9
1.3. The arbitrage pricing theory ........................................................................................... 10
2. Analysis of Capital Asset Pricing Model .............................................................................. 12
2.1. Methodologies for empirical analysis in CAPM ............................................................ 14
2.2. Analysis of data and explication of the model through decades .................................... 24
3. Analysis of Arbitrage Pricing Theory ................................................................................... 35
3.1. Methodologies for empirical analysis in APT................................................................ 37
3.2. Analysis of data and explication of the model through decades .................................... 39
4. Comparison of Arbitrage Pricing Theory (APT) and the Capital Asset Pricing Model
(CAPM). Their impact on risk management................................................................................. 45
Conclusion .................................................................................................................................... 49
References ..................................................................................................................................... 51

3
 1. Literature overview

1.1.Modern portfolio theory

Modern portfolio theory (MPT), also known as mean-variance analysis, is the theory that helps
create a portfolio of assets that give the maximum possible expected return is at a certain level of
risk. It is a mainly idea is that diversification - owning different kinds of financial assets is less
risky than owning just one type. The theory states that the effect of portfolio's overall risk and
return should also be considered when an asset's risk and return is assessed. The variance of asset
prices is used to represent the risk.
Modern portfolio theory was developed by Economist Harry Markowitz in 1952 as "Portfolio
Selection" in the Journal of Finance . For his work he was awarded Nobel Memorial Prize in
Economic Sciences.
One of the assumptions of the modern portfolio theory is that investors are risk averse, that is,
investors will prefer the less risky portfolio when they have to choose between two portfolios
having the same expected return. Which means an investor will accept to take higher risk if they
think they are likely to obtain higher expected returns. Or vice versa, they should be ready to
take higher risk in order to get increased expected return.

Expected Return
The expected return is the amount that an investor expects to earn or lose on an investment. It is
calculated by multiplying the possible returns with the probability of each consequence and
summing these together:
Expected Return = Σ (Returni x Probabilityi)
Where i denotes each of the containing assets.
The calculation of expected return is usually done by considering historical data and which
means it may not be true in all cases. But it is usually helpful to form an idea on expectations.
The expected return of the portfolio is found by computing the the weighted average of the
expected returns of the containing assets of the portfolio. Expected return on portfolio of two
securities is calculated as:
Expected return = 𝑅A XA+ 𝑅B XB = 𝑅P

4
Where:
• 𝑅A, 𝑅B are the expected returns of each security
• XA, XB are the corresponding weight of the security

Variance. Standard deviation.

One of the most common ways to measure the volatility of a security’s return is using variance.
Variance is calculated by finding the sum of squared deviations from the expected return of
return.
The formula of variance is written as:
Σ(R − 𝑅)
σ2 =
𝑁

Where:
• R is the actual return
• 𝑅 is the expected return
• N is the

Standard deviation could be found by calculating the square root of the variance:

SD=√σ2

Covariance and Correlation

In statistics these terms measure how the expected returns of two different stocks are related.
Covariance shows the relationship between the returns on two assets in terms of direction. If the
value of covariance is positive, this indicates a positive dependency, which means both of the
return of the two securities are above or below its average at the same time. If the return of the
one security is above its average value, and the other one’s value is below, this shows negative
dependency, and will result in a negative value of covariance. The value of the covariance will
be zero if there is no relationship between the two returns.

The formula of covariance is written as:

5
 σA,B = Cov (RA, RB) = Expected value of [(RA - 𝑅A) x (RB - 𝑅B)]

Where:
• σAB is the covariance of the two securities A and B
• 𝑅A, 𝑅B are the expected returns of each security
• RA, RB are the actual returns of each security

Correlation is an important concept in finance and investment industries and shows how two
securities move in with respect to one another. It also indicates how weak or strong the
relationship is between the returns. Correlation is calculated as the correlation coefficient, and
has a value in the interval of -1.0 and +1.0.

Cov (RA, RB)

ρAB = Corr (RA, RB) =
σA x σB

Where:
• Cov (RA, RB) is the covariance of the two securities A and B
• σA and σB are the variances of each security

The variance of a portfolio.

The formula for the variance of a portfolio that contains two securities can be expressed as:

σ2p = X2A σ2A + 2 XA XB σA,B + X2B σ2B

Where:
• σ2p is the variance of the portfolio containing the securities A and B
• XA, XB are the corresponding weight of the security
• σ2A and σ2B are the variances of the two securities
• σAB is the covariance of the two securities

The formula shows that the variance of a portfolio is depends on the variances of each security in
the portfolio and the covariance of them. The variance of a security indicates the spread of the

6
return of that particular security, and the covariance will show the relationship between them. If
the value of the covariance turns out positive, then this will increase the value of the variance of
a portfolio. If the covariance is negative, then the variance of a portfolio will be decreased. This
means that if the securities move in the opposite directions, they are compensating each other,
which in financial terms is called hedging. Otherwise if the securities are moving in the same
direction, then there is no hedging, and the risk is higher.

The standard deviation of the portfolio’s return

The standard deviation of a could be found by calculating the square root of the variance:

SD=√σ2 p
Diversification effect
Diversification occurs when the correlation of securities falls below 1, which means securities
are not perfectly correlated, then the risk is reduced.

Efficient frontier
The graph of efficient frontier represents the relationship between the return and variance of a
portfolio. The curve above the point “Minimum variance portfolio” plots the maximum possible
return for a certain risk level. The minimum variance portfolio point is where the variance is
lowest, thus the risk is the least.

Figure 1.1. The graph of efficient frontier

Diversification

7
Diversification is the selection of a portfolio containing different classes of assets in order to
reduce risk. Also, securities from the foreign market could be selected, as they are less likely to
have much correlation with domestic securities.

Systematic and unsystematic risk

Systematic risk, also called undiversifiable risk, volatility, market risk or aggregate risk, affects
the whole market or the market segment. It is not possible to foresee or totally avoid this type of
risk. Systematic risk depends on important changes in politics, inflation, recession, change of
interest rates, catastrophes, etc.
The beta of a security can give information about the relationship between that security and the
systematic risk of market. If the value of beta is less than 1, this suggests that the systematic risk
of the security is less than the that of the market. If the value of beta is more than 1, this shows
that that the systematic risk of the security is less than systematic risk of the market.
The formula of beta is expressed as:

Cov (Ri, RM)

βM =
σ2(RM)

Where:
• Cov (Ri, RM) is the covariance of between the return on Asset i and the return on the
market portfolio
• σ2(RM) is the variance of the market

Unsystematic risk, however, corresponds to a particular security and it is possible to mitigate

with diversification. It is associated with internal factors in the activities of enterprises issuing
these securities. Non-systematic risks of two different enterprises are not related to each other
and can occur independently of each other.

8
 1.2.The Capital Assets Pricing Model

The capital assets pricing model (CAPM) is a model that describes the relationship between the
expected return of a portfolio of securities and the degree of its risk. It was proposed by the
financial economist William Sharpe, in his book “Portfolio Theory and Capital Markets”(1970).
The formula to demonstrate the relationship between risk and expected return can be expressed
as:

𝑅 = RF + β ( 𝑅m - RF)
Where:
• 𝑅 is the expected return
• RF is the risk-free rate
• β is Beta of the security
• 𝑅 m - RF is the risk premium

Figure 1.2. The security market line

The security market line is the graphical expression the above equation. The intercept is the point
where beta is zero, the point where there is no systematic risk. It represents the asset that is the
only one which is not correlated to the market fluctuations, and promises a certain return.

9
The CAPM is based on the following assumptions:

• all investors are risk averse, with the same risk level, they choose securities with higher
returns and with the same level of return, they choose securities with lower risk.
• all assets in the capital market can be completely subdivided. All assets are fully liquid,
marketable and decentralized.
• Market participants can borrow and lend unlimited amounts at risk free rate
• all investors have the same thoughts on probability distribution of the returns on assets,
there is only one efficiency boundary in the market. The same expected value for the
expected rate of return, the standard deviation, the covariance between securities are the
same for all
• all investors can get information on market in time and free of charge.
• there is no inflation, discount rate stays same, there is no tax and transaction costs when
buying and selling securities.
• Individual investors cannot affect market prices through buying and selling, they are all
price takers

1.3.The arbitrage pricing theory

The arbitrage pricing theory was proposed by Ross S. in 1976, in the “Journal of Economic
Theory.”
The APT model is based on the fact that the investor seeks to increase the return on his portfolio
without increasing risk whenever the opportunity arises. Investors use the principle of arbitrage,
that is, they strive to increase the profitability of their portfolio without increasing risk every time
there is an opportunity to obtain a risk-free profit by using different prices for the same securities
or other assets.
The arbitrage pricing model, unlike the CAPM, makes it possible to include any number of risk
factors. So the required rate of return can be a function of three or more factors.
The main assumptions of the APT model are that the profitability of each share depends, on the
one hand, on general macroeconomic factors, and on the other, on internal factors (hindrances) in
the company's activities.

10
The profitability of an individual share in the theory of arbitrage pricing is calculated using the
formula:
𝑅i= RF + βn ( 𝑅i - RF)
Where:
• 𝑅i is the expected return of asset i
• RF is the risk-free rate
• βn is sensitivity of asset to the macroeconomic factor n
• 𝑅i - RF is the risk premium

To use the APT model, it is necessary to determine the list of macroeconomic factors, then
estimate the expected risk premium for each of the factors and determine the sensitivity of each
stock to these factors. If the actual value of profitability does not coincide with the calculated
one, the more profitable asset will be bought, and its price will rise; the less profitable asset will
be sold and its price will fall.
Arbitrage pricing theory is also based on a number of assumptions that critically affect its
practical applicability and the possibility of a qualitative forecast of market movements. Here are
the main ones:
• Since the main goal of an investor is to maximize profit, he owns such a number of shares
that the unsystematic risk becomes negligible.
• Markets are frictionless (no barriers due to transaction costs taxes, or lack of access to
short selling).
• There are no arbitrage opportunities and if they are uncovered, they will be quickly
exploited by investors

11
 2. Analysis of Capital Asset Pricing Model

Empirical tests of the model revealed a large number of weaknesses, such as inaccurate estimates
of the beta coefficient. Therefore, some economists began to deviate from the basic
concept. Blume (Blume 1970), Black, Jensen and Scholes (Black, Jensen, Scholes 1972) applied
the model not to individual securities but to a portfolio. This approach turned out to be very
successful.
Jensen (1968) presented a time series test that rejected the classic Sharp-Lintner-Mossin version
of the CAPM. Jensen modified the classical model to include the so-called Jensen's alpha, a
constant. Jensen’s alpha also called “Jensen’s return index” or “Jensen’s coefficient” is used to
evaluate the abnormal return of a security or portfolio of assets in relation to the expected market
returns. The latter is calculated based on the capital asset pricing model (CAPM).
The higher the alpha, the greater the portfolio’s return relative to the expected return.
This indicator was first used by Michael Jensen in 1968. It was originally used to assess the
performance of mutual fund managers, to assess how well managers can obtain above market
returns over time. It turned out that the constant is significant and even more than the risk-free
rate. This direction of testing the CAPM model is one of the main and most popular. Studies by
Miller and Scholes (Miller, Scholes 1972), Blume and Friend (Blume, Friend 1973) and Fama
and French (Fama, French 1992) arrive at similar results as Jensen.
Another area of testing the model was beta testing. It was tested whether this ratio can explain
one or another level of expected returns (Fama, MacBeth 1973).
However, Roll (Roll 1977) argued in his work that all tests of the model are not reliable and
never will be, because the proxies used in the tests are not complete, they cannot be designed to
include absolutely everything. assets. Continuing the criticism of Roll, Fama and French (Fama,
French 2004) consider the impossibility of testing the model and overly simplistic assumptions
as a possible reason for this empirical “failure”. Another problem, according to the authors, is
that some definitions are misinterpreted, such as the market portfolio
However, this did not stop the researchers from continuing to test the CAPM modifications. For
example, Gibbons (1982) rejects both Black's CAPM and the classical version of the
model. However, in the same year, Stambaugh (1982) comes up with opposite conclusions.

12
Subsequently, a large number of studies of conditional CAPM appeared. For the first time, the
ability of conditional CAPM to explain differences in average returns on a large number of
portfolios was tested in (Jagannathan, Wang 1996). They concluded that the conditional CAPM
is quite capable of dealing with market inefficiencies. The same conclusions were reached
(Lettau, Ludvigson 2001; Gomes et al. 2003; Petkova, Zhang 2005). On the other hand, in
(Lewellen, Nagel 2006), the authors conclude that conditional CAPM is unable to explain
specific market anomalies, but more importantly, conditional constants are large and significant,
which rejects the applicability of conditional CAPM.
The Estrada model also received widespread publicity among researchers. So in the articles
(Post, Vliet 2004; Estrada 2006; Estrada 2007) shows that the one-sided approach to estimating
the beta coefficient is more preferable than the classical one. Likewise, (Abbas et al 2011) argues
that DCAPM is the solution to the problem of asset pricing.
Over time, models with higher moments were tested as a measure of risk. In particular, a model
with skewness (Harvey, Siddique 2000) and kurtosis (Dittmar 2002) was tested in the American
stock market. The authors conclude that such models are more suitable for the American market
than the classic one.
Conditional CAPM tests using multivariate GARCH models allow us to take into account the
fact that conditional covariance varies greatly over time, as a result, and the beta coefficient also
changes over time.
The first work devoted to testing CAPMs using the multivariate GARCH model is considered to
be (Bollerslev, Engle, Wooldridge 1988). It is the first to use the multidimensional diagonal VEC
GARCH model. The authors test the CAPM for quarterly yields of 6-month US Treasury bills,
20-year US Treasury bonds, and stocks for the period from Q1 1959 to Q2 1989. US Treasury
bills. They conclude that the conditional covariance model performs better than the
unconditional covariance model. The authors conclude that CAPM is rejected because the lagged
dependent variable added to the model is significant, i.e. the classic CAPM is unable to fully
predict excess returns.
A slightly different methodology was used in (Ng 1991). It tested CAPM not on individual
securities, but on a portfolio. Various portfolio construction techniques were applied. The
portfolios consisted of all securities traded on the New York Stock Exchange from January 1926
to December 1987. Monthly returns were used. Both the classic Sharpe-Lintner-Mossin CAPM

13
and Black's CAPM without a risk-free asset were tested. The author concludes that the testing
results depend on the portfolio construction methodology. Thus, when testing the CAPM on
portfolios sorted by size, it is concluded that the models are rejected. When testing the model on
beta sorted portfolios, the CAPM is not rejected.
Similar studies have become very popular and have been carried out in other developed markets
as well. In particular, the article (Hansson, Hordahl 1998) conducted tests of conditional CAPM
in the Swedish stock market for the period from 1977 to 1990. The authors used a variety of
portfolio construction techniques and monthly returns. They conclude that the conditional CAPM
cannot be rejected in favor of other modifications (for example, the authors used zero-beta
CAPM as an alternative model).
The next round of CAPM testing using multivariate GARCH models was the transition from
developed to developing markets. For example, (Godeiro 2013) tests a conditional CAPM in the
Brazilian stock market using the multivariate VCC GARCH. The daily returns of stocks included
in the Brazilian market index for the period from January 1, 1995 to March 20, 2012 are used.
The author does not use a portfolio, but tests the CAPM for each stock separately. He concludes
that the conditional CAPM is being rejected.

2.1.Methodologies for empirical analysis in CAPM

Most of the empirical test conducted on the capital asset pricing model were by the use of time
series test. After estimating the betas this way, using the cross-sectional regression, the
hypothesis was tested. Below are listed different studies with different methodologies applied to
test the model, and the discussion of the methodology.

Lintner and Douglas studies

In 1965 John Lintner carried out the first empirical test of CAPM. Later Douglas in 1967
replicated the study with results similar to Linter's original test. Lintner applied a two-pass
procedure for individual stocks. The data he used corresponds the decade covering 1954 and
1963 with one measurement for each year. He examined the returns of 301 companies listed on
the New York Stock Exchange and used the predecessor of the S&P 500 Index as proxy for the
market portfolio. To estimate the beta of each security, Lintner used the following regression. In

14
this regression, the annual returns of each security are regressed against the returns of the market
index.
𝑅𝑖𝑡= 𝛼𝑖+𝛽𝑖∗𝑅𝑚𝑡+𝜀𝑖𝑡 (1)

Where:
𝛽𝑖 is the beta of the asset, 𝑅𝑖𝑡 is the return on the asset at time t, 𝑅𝑚𝑡 is the return on the market at
time t, 𝜀𝑖𝑡 is the error term and 𝛼𝑖 is the intercept.
After the estimation of the betas, the second pass cross sectional regression is carried out.

𝑅𝑖=𝑦0+𝑦1𝛽1+𝑦2𝜎2(𝜀𝑖)+𝜀𝑖 (2)

Where 𝑦1 is the slope of the beta of the asset, 𝑅𝑖 is the return on the asset i, 𝑦2 is the slope of the
residual variance and 𝜀𝑖 is the error term and 𝑦0 is the intercept.
If the results were in compliance with CAPM, the coefficients have to be: 𝑦0 = 0, 𝑦1 = 𝑅𝑖 - 𝑅𝑖 and
𝑦2 = 0. But the results do not comply with the CAPM, the coefficient for the intercept, 𝑦0, was
larger than risk-free rate, the coefficient for the residual risk, 𝑦2, was significantly different from
zero and slope of the SML, 𝑦1 was much different from the market risk premium.

Miller & Scholes study

In 1972 Miller and Scholes also performed a study on CAPM using the two-pass method also
coming up with very similar results. It turns out that, the slope was too flat and crucially different
from the market risk premium. The intercept and residual variance were too much different from
zero.
To investigate the statistical problems of the two-pass method, Miller and Scholes arranged
numbers and ran a simulation test. According to the CAPM model, they generated numbers that
exactly corresponded to the expected beta ratio of the rate of return. That is, the simulated returns
are created to exactly match the CAPM. However, the CAPM using the two-pass method was
discarded because the results were similar to those generated with the actual data. This clearly
proves that there are some statistical problems with the two-pass methodology. They have shown
that the model may be valid even if the CAPM is rejected using this method. They further argue
that the methodology used to test the model continues to result in model rejection and is the
result of errors in the beta measurement of the second cross-sectional regression.

15
Due to statistical problems, it is clear why Miller and Scholes conclude that there is a problem
with the two-pass method. Regression equation (1) estimates the coefficients at the same time,
and these estimates are interdependent. When estimating the intercept of a single variable, the
regression relies on the estimation of the slope coefficient.
In summary, if the beta estimates are distorted, so are the section estimates. Miller and Scholes
conducted a CAPM study using the procedure twice and obtained very similar results. However,
the slope is too flat and deviates significantly from the market risk premium. The intercept and
residual variance are not statistically significant.
To test the statistical problems of the two-pass procedure, Miller and Scholes performed a
simulation test by generating random numbers. Following the CAPM model, they identified
numbers that exactly matched the expected return beta relationship. That is, the simulated rate of
return has been set to fully agree with the CAPM. However, the results were almost identical to
those produced with real data, so the CAPM was rejected using the two-pass technique.
Obviously, this demonstrates that there are some statistical problems with the two-pass method.
They demonstrated that even if CAPM is rejected using this approach, the model can still be
valid. They further assert that the methodology used to test the model and result in the model
being rejected is a consequence of the error in the beta measurement for the second cross-
sectional regression.
Due to statistical problems, Miller and Scholes apparently concluded that the two-pass procedure
was problematic. The regression equation estimates the coefficients simultaneously, and the
estimates depend on each other. When estimating the intersection of a single variable, the
regression depends on the estimate of the slope. As a conclusion, if the beta estimate is biased, so
is the intercept estimate.
They show that the statistical problems caused by taking into account the beta from the first
pass regression estimated with significant sampling error must be addressed. Therefore, the
estimated beta is not a proper input for the second pass regression, but provides misleading
results and ultimately does not test the validity of the model. By using the beta estimated by the
measurement error, the regression in the second run causes the slope y1 to be the downward
factor and the y-intercept y0 to be the upward factor. Empirical testing of the above model shows
that this is exactly the case. That is, the intercept is higher than predicted by the CAPM results
and the gradient is much flatter than predicted by the model.

16
Fluctuations in beta estimates are also caused by the fact that other major but unmeasured causes
of common stock volatility are at work. Bias in the model specifications misrepresents beta
estimates because the market model does not take into account other significant systematic
impacts on stock market volatility.

Black, Jensen & Scholes study

The rejection of the CAPM in the initial verification tests above does not necessarily mean that
the model is defective. It may be due to a statistical error in the beta measurement. To avoid
measurement errors that effectively distort SML estimates, Black, Jensen, and Scholes provided
additional insights into the nature and structure of security returns. Instead of testing individual
stocks, they suggested ways to improve the accuracy of the estimated beta by grouping the stocks
into a portfolio.
In their article, in previous tests of the model, the structure of the process, where the data
appears to be generated, leads to a cross-section test of misleading importance and does not
provide a direct test of the effectiveness of the CAPM. Aggregation makes portfolio beta
estimates less susceptible to measurement errors than individual securities beta estimates. This
reduces the statistical error that can occur in beta factor estimates. They examined all New York
Stock Exchange securities from 1926 to 1965 using an evenly weighted portfolio of all New
York Stock Exchange stocks as a substitute for the market index.
The outline of their study began with estimating the beta of each securities by using equation (1)
to regress monthly returns against the market index for the first 60 months of the period. Then,
by sorting the securities according to the calculated beta, 10 portfolios were formed. Portfolio 1
consisted of the highest beta stocks and Portfolio 2 consisted of the next highest beta stocks. By
doing this, they found that the portfolio had a huge expansion in beta.
However, the beta used to select the portfolio is still subject to measurement error, so building a
portfolio of securities based on the estimated beta gives an unbiased estimate of the portfolio
beta. To solve this problem, they built a portfolio for the following year using betas of individual
securities estimated in the previous period. In this way, many of the sample fluctuations in the
estimated beta of individual securities can be eliminated.
The next step is to calculate the monthly returns for each portfolio for the following year. This
step is then repeated once a year for the entire sampling period. They then calculated the average

17
monthly return and estimated beta factor for each of the 10 portfolios. Finally, they used
equation (2) to regress the average portfolio return against the portfolio beta for the entire sample
period and the various sub-periods. Evidence from their empirical analysis led to the rejection
of traditional forms of CAPM because the duration of interception was higher than the risk-free
rate. However, the SML gradient term was significant and was positively linear as predicted by
the model, so the results appeared to be consistent with zero-beta CAPM. In general, the results
were similar for the entire and partial test period.

Sharpe and Cooper study

In 1972 Sharpe and Cooper performed a simple test of CAPM in the form of a simulated
portfolio strategy. From 1931 to 1967, stocks were investigated and tested on the New York
Stock Exchange. The original purpose was to determine if the higher the beta stock, the higher
the return. Portfolios constructed in different betas by first regressing individual asset returns
with market returns based on the last 5 years using equation (i). Once a year, they evaluated all
strains according to beta and divided them into 10 categories. They formed portfolios that were
evenly weighted by category, stocks with the highest beta in one portfolio, stocks with the
second highest in another portfolio, and so on. The investment strategy followed was to hold
only one category of securities for the entire period. The results of this study show that high beta
stocks produce higher returns. They found that 95% of the fluctuations in expected return could
be explained by the difference in beta, so there was a linear relationship between realized
average return and that beta. However, the intercept, or risk-free rate, is 5.54%, which is
significantly higher than the 2% for that period, so it does not match the Sharpe Lintner version
of the model. However, these results support zero beta CAPM.

Fama & MacBeth

Fama and MacBeth have followed a similar methodology to Black to test the CAPM, but add
another explanatory variable, the square of the beta coefficient, to the equation to test that there
is no non-linearity in the relationship between risk and reward. bottom. The importance of linear
conditions was largely overlooked in early empirical testing of the model. Another difference
from the BJS study is that Fama and MacBeth use one period beta to predict returns for later
periods, while the BJS method calculates beta and average returns for the same period. is.

18
To test the effectiveness of CAPM the following model was used:

𝑅𝑝𝑡=𝑦0𝑡+𝑦1𝑡𝐵𝑝+𝑦2𝑡𝐵𝑖2+𝑦3𝑡𝜎2+𝜂𝑝𝑡 (3)

Index p is related to the constructed portfolio, but not to individual stocks. Using equation (3),
Fama and MacBeth tested some hypotheses about CAPM. They tested the expected value of the
risk premium y1t, that is, whether the slope of SML is positive, that is, E (𝑦1𝑡) = E (𝑅mt) - rf)> 0.
To test the linearity, the beta-square variable of the equation, the hypothesis tested under this
condition is E (𝑦2𝑡) = 0. Hypotheses are tested in connection with testing non-beta risks, E (𝑦3𝑡)=0.
We also test that the intercept is equal to the risk-free rate, that is, E (𝑦0𝑡) = rf. The mean of the
disturbance term η is zero and is assumed to be independent of all other variables in the
equation. Like Jensen, Black & Scholes, Fama and MacBeth find that the intercept is larger than
the risk-free rate and reject the Sharpe Lintner CAPM. However, the results of their research
confirmed the important testable implications of zero-beta CAPM. The SML slope is positive
because the coefficient y1 is significantly different from zero, which indicates a positive risk-
return trade-off. The coefficients y2 and y3 were not significantly different from zero. There is no
evidence of non-linearity or residual variance affecting the return. Finally, the observed fair
game characteristics of risk-return regression coefficients and residuals are consistent with an
efficient capital market where prices perfectly reflect all the information available.

Modigliani study
Empirical testing and research focuses primarily on US-listed stocks. In essence, the literature is
not very extensive when it comes to the European market. The reason for this could be the
availability of data and the increased efficiency of the US stock market. In 1972 Franco
Modigliani conducted the first test of the CAPM on the European stock exchange. Prior to his
research, CAPM was not represented in capital markets other than the US market.
The tests were conducted in eight major European markets and reproduced previous tests in the
US market (Black, Jensen & Scholes (1972), Friend & Blume (1970), Jacob (1971)). Modigliani
found that the European market is generally considered less efficient than the US market. If so,

19
that would mean that the risk assessment of European securities is less rational than that of
American securities.
For the test, the data to be used included the daily prices and dividend data for 234 common
shares corresponding to eight major European countries covering the period of March 1966 to
March 1971. Data have been revised for all capital adjustments. Since the data are from eight
countries, the stocks were analyzed individually so that the regression results are displayed
individually for each country. The results of the US market were used for similar periods for
comparison.
The results of the study supported the hypothesis by showing that systematic risk is an important
factor in the pricing of European securities. In addition, there was a positive correlation between
realized returns and risk for 7 of the 8 markets tested. Moreover, there were no signs that the
European stock market was unreasonable or inefficient. Nevertheless, the test period was as short
as 5 years and the samples were limited.

Roll´s critique
Richard Roll criticized previous work on the capital asset pricing model. He suggested that the
methodology was not accurate because the tests were only performed using market portfolio
proxies, not the actual market portfolio.
Roll stated that the only verifiable hypothesis related to CAPM is that the market portfolio has
mean-variance efficiency, and all other suggestions of the model, such as the linear relationship
between returns and beta, are due to the efficiency of the market portfolio. He pointed out that
this is the reason why it cannot be tested independently.
Roll argued that the real estate market portfolio contained all assets, including human capital and
real estate. He also points out that using only one proxy in the market portfolio is not enough to
state or reject whether CAPM works. He further argued that it was impossible to define a real
market portfolio and therefore could not seriously test the CAPM.

Roll & Ross and Kandel & Stambaugh study

Roll, Ross, Kandel, and Stambaugh performed further study on Roll’s critique. They say that the
refusal of the risk-return relationship specified by CAPM does not refute the model relationship
between average returns and beta, but because the market portfolio proxies are mean-variance

20
inefficient. Insisted that there was a possibility. They show that market portfolio proxies, which
are not significantly inefficient and should produce a positive linear risk-return relationship in
large samples, cannot establish a significant relationship. Kandel and Stabaugh explain how
many people don't see the meaning of CAPM separately because one implies the other.
The CAPM suggestions could be seen in two forecasts:
• the market portfolio is efficient
• the security market line (the expected relationship of return and beta) is capable of
demonstrating the risk-return trade-off in an accurate way, in other words, alpha values
equal to zero.
Their article further shows that one implication is close to perfection and the other implication
can fail dramatically. They tested Schwarz's Zero Beta CAPM using a two-pass method with a
market portfolio proxy, as an efficient portfolio is only theoretical.
After performing the first-pass time series regression (1), the following generalized least-squares
the second-pass regression was performed. In this way, they took into account the correlation
between the residuals.
𝑟𝑖−𝑟𝑧=𝑦0+𝑦1∗(𝐸𝑠𝑡𝑖𝑚𝑎𝑡𝑒𝑑 𝛽𝑖)

Where 𝑟𝑖 is return on stock i, 𝑟𝑧 is, 𝑦0 is the intercept, 𝑦1 is the beta and 𝛽𝑖 is the estimated beta.
Their conclusion is that terms proportional to the relative efficiency of the selected proxy that
represents the true market portfolio is the reason why the intercept and slope coefficients are
biased. If the market index used in the regression is perfectly efficient, the test is well specified.
However, if the proxy is inefficient for the market portfolio, the second path regression provides
a bad test for the CAPM. Therefore, without a reasonably efficient market proxy, the model
cannot be meaningfully tested.

David W. Marines
In 1982, David W. Mullins made a complete and detailed evaluation on the effectiveness of
CAPM. He found the model incomplete, but said it was an important tool for investors in
determining the return on total assets needed. In addition, he suggests that even if the model
assumptions are theoretical and unrealistic, it is important to simplify the reality in order to build
a comprehensive model for determining asset prices.

21
Clare, Smith & Thomas study
In 1997, Clare, Smith, and Thomas tested both conditional and unconditional versions of
CAPM on the UK Stock Exchange. Their dissertation creates an important link between formal
asset pricing and evidence of predictability of excess returns. They use both market value and
dividend yield ranked portfolios to find the right risk return spread. The results show that beta
plays an important role in explaining expected return.

Pettengill, Sundaram & Mathur study

In 1995 Pettengill, Sundaram & Mathur found a consistent and very important relationship
between beta and portfolio returns. They make a significant distinction between their tests and
previous empirical tests, and as predicted by the Sharpe-Lintner-Black model, and they realize
that the positive relationship between returns and beta is based on expected returns rather than
realized returns. They consider the impact of using the realized market revenue as a substitute for
the expected market revenue. If the excess market return is negative, there should be an inverse
relationship between beta and portfolio returns. Balancing expectations for negative excess
market returns finds a consistent and very important relationship between beta and portfolio
returns across sample and sub-periods. There is also a positive risk-return trade-off.

Fama & French study (1992)

Fama & French wanted to investigate, in addition to beta, factors that could affect stock returns,
such as book-to-market capital and size. The data they used are from all non-financial companies
in NYSE, AMEX, NASDAQ, and annual industry files containing income statements and
balance sheet data from 1962 to 1989. They had values for required accounting variables, so they
were able to estimate the beta of the portfolio and appropriate that beta to each stock in the
portfolio. This way, its possible to use individual stocks in Fama-MacBeth regression. Prior to
1969, it’s recognized that there was a positive relationship between beta and average returns
(Black, Jensen & Scholes, 1972; Fama & MacBeth, 1973). However, their results from 1969 to
1990 suggest that the relationship between average returns and beta does not exist. In addition,

22
they found a weak beta-return relationship for the 50 years from 1941 to 1990. In conclusion,
their tests did not support the prediction of the basic Sharpe-Lintner-Black model.
Fama & French study suggests that poor beta results may be due to other explanatory variables
that correlate with the true beta. This obscures the relationship between average return and
estimated beta. However, the beta alone does not explain why it seems to have little explanatory
power in each period.
The results of Fama & French also show that the book-to-market equity-to-revenue-to-price ratio
is an inadequate alternative to beta. However, their key findings conclude that the easily
measurable variables size and book-to-market fairness appear to explain the cross-section of
average returns.
The summary of the results are as follows:
• If there is variation in beta that is not related to the size, the relation between beta and
average return is not reliable.
• The opposing roles of market leverage and book leverage in average returns are precisely
described by book-to-market equity.
• The relation between E/P and average return is balanced with the help of the connection
of size and book-to-market equity.

Fama & French study (2004)

Additional research on CAPM includes another article by Fama and French covering the latest
related research on CAPM. They state that the CAPM presented by Sharp and Lintner has never
had empirical success, but the Black version has produced some positive results.
However, as the survey began to include variables such as size, various pricing ratios, and
dynamics, it also helps explain the average revenue that the beta provides. The model results
faced a problem that Fama & French said was sufficient to explain most invalidations. CAPM
application.
The three-factor model presented by Fama & French adds market characteristics and size to the
official book and forms the three factors together with the beta. With this model, you can achieve
up to 90% of diversified portfolio returns. These results are supported since Fama & French. In
this study the focus will be on Pharma & Macbeth's previous approach, which does not take into
account non-beta factors.

23
 2.2.Analysis of data and explication of the model through decades

The Capital Asset Pricing Model (CAPM) test does not directly test the CAPM assumptions, but
the properties of the stock market line are tested. Which means, it tests if there is a positive
correlation between the security or portfolio beta factor and the expected return.
Since the capital asset pricing model briefly describes the relationship between the returns and
risks of different assets in the capital market, the capital asset pricing model is naturally very
important if it can describe the actual situation of the capital market well. For this reason, many
empirical tests have been conducted on asset pricing models.
Below are some empirical tests conducted on the capital asset pricing model and the analysis of
data and results.

The Black, Jensen, and Scholes study

To conduct the study, all stocks on the NYSE covering the interval 1926 – 1965 were examined
and as proxy for the market index, equally weighted portfolio of all stocks were taken.
The study was carried out based on the following steps:

1. Monthly returns of the first 60 month period, 1926-1930 is used for the estimation of
Beta of each stock.
2. 10 portfolios were created by ranking the betas: top 10 portfolios named portfolio no. 1,
the successive portfolios named by increasing order and the last and no. 10 containing the
stocks with the smallest betas.
3. Each portfolios returns for 12 months in 1931 were calculated.
4. Steps 1 to 3 were applied to all 10 portfolios for throughout the interval 1926-1965, each
time taking 5 year interval and estimating the beta and the return of the stock in the
successive year.
5. The mean portfolio returns were regressed against the portfolio betas throughout the
period 1926-1965, to estimate the ex post Security Market Line.

The results of this study rejects the traditional version of the CAPM, because the intercept turns
out to be greater than risk free rate. But it is consistent with zero beta CAPM.

24
 Figure 2.1. Estimating beta with regression

Equilibrium models for the pricing of capital assets has always been focused on by many
scholars throughout the years. The most prominent equilibrium model is the mean variance
model of Sharpe initiated in 1964. This model was in the following years advanced and
explicated by the scholars Lintner, Mossin, Fama, Long. Later, Treynor, Sharpe, Jensen came up
with models which would be used for also portfolio evaluation. These asset pricing models were
built on or extended from the mean variance model of Sharpe. The idea behind the model
demonstrates the connection between the expected value of risk premium and the systematic risk.
This relation could be expressed such:

E (𝑅̃j) = γ1βj (1)

Where:
̃ ̃
𝐸 (𝑃𝑡)−𝑃𝑡−1− 𝐸 (𝐷𝑡)
E (𝑅̃j) = - rFt is expected excess returns on the jth asset
𝑃𝑡−1

γ1 = E (𝑅̃M)
̃ 𝑡 is dividends paid on the jth security at time t
𝐷
rFt is riskless rate of interest
E (𝑅̃M) is expected excess returns on a “market portfolio” consisting of an
investment in every asset outstanding in proportion to its value
βj is systematic risk of the jth asset

25
It can be seen from relation (1) that the expected return of an asset is directly proportionate with
its beta. Expressing αj in a way that:

αj = E (𝑅̃j) - E (𝑅̃M) βj

leads to the result that α of every asset is equal to zero according to relation (1).
If relation (1) proves to be true in practice, it can eventually lead to the simplification of analysis
in many economic matters including portfolio selection, capital budgeting, cost benefit analysis
or many other issues that require the connection of risk and return. The study carried out by
Jensen (1968; 1969) about the connection of expected return and risk return of mutual funds
demonstrates that there could be a rational explanation of the relation between risk and return of
assets. But the studies carried out by Douglas (1969), Lintner (1965), later by Miller and Scholes
(1972) states otherwise, that the model is unable to explain this connection. The study conducted
by Miller and Scholes describes that the α of a security is dependent on the β of that security in a
systematically. In other words securities with higher systematic risk are inclined to show
negative alpha, and on the other hand, securities with lower systematic risk are inclined to show
negative alpha.
In the study of Jensen, Miller, Scholes (1972), the main objective is to perform further tests of
the capital asset pricing model and eliminating the errors in the earlier tests carried out. The tests
that were performed before were based on the cross-sectional procedure, which is the regression
of the average return of the group of securities over a specific time period on the approximation
of the systematic risk of each security. The equation is expressed as:

𝑅j = γ0 + γ1𝛽̂j + 𝑢̃j
Where:
𝑅j is the mean excess return
𝛽̂j is approximation of the systematic risk

This lead to the results showing γ0 is sufficiently different from 0, γ1 is sufficiently different from
𝑅M, which is the slope demonstrated in the model.

26
In the study certain aspects will be discussed including: it is shown that the cross-sectional tests
may result with errors because of the organization of the procedure that creates the data, the
reliability of the model is tested using methods that have been able to eliminate the
complications related to cross-sectional tests. The results obtained demonstrate that the (1)
version of the asset pricing model does not lead to accurate security returns. It is shown that
expected excess returns on assets with higher beta prove to be lower than that proposed by (1)
and expected excess returns on assets with lower beta prove to be higher than that proposed by
(1).
The alternative hypothesis of capital assets pricing arises from the relaxation of one of the
traditional forms of assumptions in the capital goods pricing model. Relaxing the assumption that
risk-free lending and borrowing opportunities exist leads to the formulation of a two-factor
model. In equilibrium, the expected return E (𝑟̃ j) of an asset is given by:

E (𝑅̃j) = E (𝑟̃ z) + [E (𝑟̃ M) - E (𝑟̃ Z) ] βj (2)

Where:
r’s indicate total returns
E (𝑟̃ z) is the expected return on a portfolio with zero covariance (βz=0)
𝑟̃ M is the return on the market portfolio.

In this model, the expected return on a 30-day Treasury bill of return (used as a substitute for a
"risk-free" rate) only represents the return on a particular asset in the system. Therefore, by
subtracting 𝑟̃ F from both sides of (2), (2) can be rewritten as follows in terms of "excess" returns:

E (𝑅̃j)= γ0 + γ1 𝛽j (3)
Where:
γ0 = E (𝑅̃z)
γ1 = E (𝑅̃M) - E (𝑅̃z)
Whereas in the traditional model, γ0 = 0 and γ1 = E (𝑅̃M), the two factor model demonstrates that
γ0 = E (𝑅̃z), and is not equal to zero in all cases, and γ1 = E (𝑅̃M) - E (𝑅̃z).

27
Additionally, several other models have emerged as a result of relaxing some assumptions of
traditional asset pricing models, implying γ0 ≠ 0 and γ1 ≠ E (𝑅̃M),
These models include and briefly account for the problem of RM measurement, the existence of
non-marketable assets, and the existence of differential taxes on capital gains and dividends. Our
main focus has been to test the rigorous traditional forms of asset pricing models. That is, γ0 ≠ 0.
It has not been attempted to provide a direct test for these other alternative hypotheses.
To test the traditional model, all securities listed on the New York Stock Exchange between
1926 and 1966 were used. The problem faced was getting an efficient estimate of the mean beta
and its variance. An alternative hypothesis can be tested by randomly selecting one security,
estimating a beta coefficient based on a time series, and finding out whether the average return
differs significantly from that predicted by traditional forms of capital asset pricing models.
However, this would not be an efficient testing procedure.
For efficiency reasons, securities were grouped into 10 portfolios so that the portfolio had large
spread of β’s. It was known that grouping securities based on estimated β’s would not provide an
unbiased estimate of the "beta" portfolio, because the β’s used to select the portfolio contains
measurement errors. This procedure would lead to selection bias in the tests.
To remove this bias, the previous period's estimated beta was used an instrumental variable, to
select a stock portfolio group for the following year. Using these procedures, ten portfolios
whose β’s estimate is an unbiased estimate of the “Beta” portfolio was constructed.
It was found that much of the sampling variation of the estimated β’s for individual securities is
eliminated using portfolio groups. So the β’s of the built-up portfolio varies from 0.49 to 1.5, and
the β’s portfolio's estimates for the sub-periods shows significant stationarity.

28
Table 2.1

Portfolio number 1 includes the highest-risk securities and portfolio number 10 contains the
lowest-risk securities. The critical intercepts α are shown in the second row of table 2.1. Below
that the student “t” values are placed. The table also includes the correlation of the market and
portfolio returns, r(𝑅̃M, 𝑅̃K) and the autocorrelation of the residuals.
The autocorrelation turns out to be small, but the correlation of the market and portfolio returns
are high as anticipated. In the table for each portfolio, is demonstrated the standard deviation of
the residuals, the average monthly return, the standard deviation of the monthly excess return.
Time series regression of portfolio excess returns against market portfolio excess returns has a
significantly negative intercepts for high beta securities and a significantly positive intercepts for
low beta securities, as opposed to the traditional form of forecasting in the model.
Over the 35 year interval, the predicted amount to be earned by high risk securities was higher
than the actual one, and the low risk securities happened to earn more than predicted. There was
also important evidence that this effect became stronger over time and was most pronounced
around 1947-65.
To check the stationarity of the empirical relations, the 35 year period was divided into 4 equal
subperiods each consisting of 105 months.

29
Table 2.2.
The table 2.2 has demonstrated the regression statistics with the data relating to each of ten
portfolios. It can be seen that apart from portfolio 1 and 10, the risk coefficients β were quite
stationary.
On the other hand, the section of critical intercepts were not stationary throughout the period.
The positive indication of α for high risk portfolios in the subperiod 1 (January, 1931-September,
1939) shows that these securities had earned more than the forecast of the model. The negative α
of the low risk portfolios show that they earned less than the prediction of the model.
The following 3 subperiods have shown opposite results and have deviated from the prediction
of the model even further.
In the cross-sectional test of the model, securities were grouped and then with the time series
method the risk measure was estimated for each portfolio. Then the cross-sectional parameters
were estimated by portfolio mean return over risk coefficients.

30
Figure 2.1.
Figure 2.1. plots the mean monthly returns against systematic risk for 35 year interval 1931-
1965 for each of 10 portfolios and the market portfolio. The points expressed with the sign x
denote average excess return and risk of all the portfolios, whereas the point 􀀀 is average excess
return and risk of the market portfolio.
The traditional version of the capital asset pricing model demonstrates that intercept should be
zero, and the slope is the mean excess return on the market portfolio. Over this interval,
theoretical values of the mean excess return on the market portfolio was calculated to be 0.0142.
The 35-year interval was divided into four subperiods. The plots of average excess return versus
risk of all the portfolios over these subperiods are shown in figure 2.2.

31
Table 2.3.

A cross-section plot of the mean excess return and the estimated β’s of the portfolio shows that
the relationship between the mean excess return and β is linear. However, the slope of the
intersection and the cross-sectional relationship differs depending on the subperiods, which does
not match the traditional form of the capital asset pricing model.

32
Figure 2.2. Figure 2.3.

Figure 2.4. Figure 2.5.

33
In the two sub-periods of 105 months before the war investigated, the slope in the first period
was steeper than predicted in the traditional form of the model and flatter in the second period. In
each of the two periods of 105 months after the war, it was significantly flatter than expected.
Based on evidence from both time series and cross-section execution, the hypothesis that γ0 in
(3) is zero was rejected.
As a result, it was concluded that the traditional form of asset pricing model is inconsistent with
the data.
Explicit time series estimate of beta returns were made to get more efficient estimates of mean
and variance, thereby allowing to directly test that the mean excess return of beta is zero. From
the return data of the portfolio, a minimum-variance, unbiased linear estimator of the returns of
the β factor was derived. It was demonstrated that, when independence of the residuals is given,
the optimum estimator requires that of each portfolio's unobservable residual variances is known,
but that this difficulty could be avoided if they were equal. The time series of returns on the beta
factor was calculated using this assumption of equal residual variances.

Extensions of the CAPM.

The capital asset pricing model has been extended in many ways. The most well-known
extensions include:
• allowing heterogeneous beliefs (Lintner, Merton),
• Elimination of the possibility of risk-free lending and borrowing (Black),
• setting some assets not marketable (Mayers)
• examination of multiple periods and investment options that change from one period to
the next (Merton, Breeden);
• Expansion of international investment (Solnik; Stulz; Adler and Dumas);
• Use of weaker assumptions considering arbitrage pricing (Ross).
For most CAPM extensions, a single portfolio of risky assets is not optimal for all. Rather,
investors distribute their assets differently to several risky portfolios that are aggregated to all
investors to form a market portfolio.

34
 3. Analysis of Arbitrage Pricing Theory

An alternative to CAPM is the Arbitrage Pricing Theory (APT), developed in 1976 Stephen
Ross. APT , unlike CAPM, does not assume that shareholders evaluate decisions using the
mathematical expectation and variance of returns. It assumes that the performance of a stock
depends partly on macroeconomic factors, and partly on events related directly to the company
itself. Instead of defining a stock's return as a function of one factor (the market portfolio return),
it presents that return as a function of several macroeconomic factors on which the market
portfolio depends.
Diversification allows to eliminate the specific risk of a particular security, leaving only
macroeconomic risk as the main determinant of the required return.
The APT model does not specify which factors are the main ones: they can be the stock market
index, the value of the gross national product, oil prices, interest rates, and other
indicators. Typically, some companies are more sensitive to some factors than others.
In theory, it is possible to create a zero-risk portfolio (i.e. a portfolio with zero "beta"), which
will provide returns equal to the interest rate in the absence of risk. If such a portfolio provided a
higher return, then investors could make a risk-free return by borrowing to buy such a portfolio
at a risk-free rate. This process, called arbitrage (i.e., making a profit with zero risk), would
continue until the expected risk premium for a given portfolio was equal to zero.
APT avoids the inherent problem with CAPM of identifying a market portfolio. However, it
poses even more difficult challenges. First, the use of APT requires the identification of
macroeconomic variables. Secondly, it is necessary to solve the problem of assessing risk
premiums for all factors and to determine the sensitivity of the values of stock returns to each of
them.
APT proponents point out that it is not really necessary to isolate the relevant factors. To
determine APT parameters you can use the mathematical apparatus of factor analysis. Initially,
data on hundreds and even thousands of shares are taken into account, then several different
portfolios are formed that do not closely correlate with each other in terms of profitability. Thus,
in this set of portfolios, each of them is more strongly influenced by one of the known
factors. Then the required return of each portfolio is considered as λ for this factor, and the
sensitivity of the return of each individual stock to the return of this portfolio becomes the

35
sensitivity of the factor (β). Unfortunately, the results of factor analysis are not easy to interpret,
since it does not allow deep penetration into the essence of the main economic components of
risk.
Earlier research also suggests that the multivariate APT model explains expected rates of return
better than the univariate CAPM does. On the one hand, its variables are not specified. In
addition, there is no consensus as to which variables should have the greatest impact. On the
other hand, APT is more complicated in that for each company and for each specific period of
time, coefficients must be generated for several factors, and not just for one factor, when they are
going to be used. Finally, the CAPM also takes into account internal risk factors, in contrast to
the APT.
The following factors have been identified and proposed to be used in the APT model by Chen,
Roll and Ross:

• The spread between short-term and long-term interest rates, the yield curve
• Expected vs unexpected inflation
• Industrial production
• The spread between low risk and high risk corporate bond yields

The arbitrage pricing model has a number of significant drawbacks:

• the model does not take into account systemic factors affecting risk and profitability;
• there is no guarantee that all relevant factors were identified and taken into account;
• the factors taken into account can change over time, that is, undergo a temporary
transformation;
• the calculation of the model is quite complicated and requires a more thorough
preparation of information, the collection of more statistical data and their more detailed
analysis.

In addition, this approach is retrospective, since it is based on the tacit assumption that in the
future exchange rates will behave in the same way as in the past.

36
 3.1. Methodologies for empirical analysis in APT

To implement a test on APT one of the three approaches could be chosen for the estimation of
factors:
The first approaches includes the algorithmic analysis of the estimated covariance matrix
corresponding to asset returns. Roll and Ross (1980), Chen (1983) and Lehman and
Modest (1985a, 1985b) use factor analysis. While Chamberlain and Rothschild (1983) and
Connor and Korajczyk (1985, 1986) use principal component analysis.
The second approach is for the researcher to start with an estimated covariance matrix of asset
returns, use his judgment to select factors, and then estimate the matrix.
The third approach is purely judgmental. Researchers primarily use intuition to select factors and
estimate the load of the factors to see if they explain the estimated cross-sectional variation in
expected return. Chan, Chen, and Hsieh (1985) and Chen, Roll, and Ross (1986) choose financial
and macroeconomic variables as factors. These include variables such as stock index returns, the
difference between short-term and long-term interest rates, private sector default premium
indicators, inflation rate, industrial production growth rate, and total consumption. Due to
business intuition, researchers continue to add many new elements.
The first two approaches are implemented to match the underlying factor structure of APT. The
first approach is based on algorithmic design, and the second approach is to ensure that the factor
they are using actually leaves the unexplained portion of the asset's earnings almost uncorrelated.
The third approach is implemented without considering the factor structure.

Gehr
In 1975, Gehr collected monthly securities return data from CRSP tapes for 30 years. He uses
the Principal Component Analysis (PCA) method to extract common factors, which rotate
obliquely. The first step is to regress each industry index of factor values to extract potential
factors and their corresponding loads. In the next step, he performed a cross-sectional OLS
regression to get an estimate of the coefficients of the APT model. His results show that over the
30-year study period, only one factor seemed to have a significant price. Gehr's empirical
evidence has not been tested against the specified alternatives, so it can be seen that there is no
empirical power to separate CAPM from APT. In addition, Gehr's work doesn't make sense for

37
several other reasons. On the other hand, factor solutions are arbitrarily rotated into one of many
acceptable solutions. This creates strong multicollinearity between the explanatory variables in
the second phase of the experiment, ignoring the difficulty of interpreting the statistics into the
explanatory variables. Second, the variable declared in the design of experiments is the average
(mean) return of the industry index, not the return of individual stocks. This implicitly limits the
experimental design of the APT test. Finally, it does not try to deal with the variable error in the
section OLS regression used. Therefore, the estimated risk premium is highly biased.

Roll and Ross

Roll and Ross in 1980 performed the most detailed APT test to date. Although they can
overcome the first and second drawbacks of Gehr’s tests, their cross-section evaluation tests are
still at the expense of variable errors. The Roll and Ross study selected approximately 42 groups
of 30 stocks listed on the New York Stock Exchange in alphabetical order to cover the period
from July 3, 1962 to December 31, 1972.
They perform the following procedure for each group to adjust the daily arithmetic return for all
capital changes and dividend payments:

• Calculate the covariance matrix of the time series sample for each of the groups of 30
stocks that cover the analysis period.
• The maximum likelihood factor analysis (MLFA) method is used to extract the number
of factors and the corresponding factor load.
• At the end, regresses the factor loading matrix of average securities returns to provide an
estimate of the factor risk premium whose significance is assessed.

To test the validity of the systematic risk hypothesis, the standard deviation of returns were
imported into the APT model and the cross-sectional regression was repeated. Therefore, the
explanatory variables are the standard deviations of the factor load estimates and returns. Their
results show that of the four factors evaluated in cross section, three factors have significant
prices. Therefore, their results support APT and empirically distinguish it from CAPM based on
pricing equations.

38
Brown and Weinstein
In 1981 Brown and Weinstein try to replicate some of the empirical results of Roll and Ross
(1980) based on another method. It could be seen that none of the grouped securities has an
obvious five-factor return generation process. In addition, cross-section pricing results show that
the average revenue of five or less factors is inconsistent with the APT framework. Eventually,
they found that the results responded slightly to normality because of the surprising
responsiveness of APT evidence to the test methodology.

3.2.Analysis of data and explication of the model through decades

To have a better understanding of the testing of APT and methodology, it would be a best to
have a detailed look on the study of Roll and Ross.
The empirical test includes two major steps:
In the first step, the expected return and factor factors are estimated from the time series data of
the individual returns of the asset.
The second step uses these data to test the basic conclusions about APT pricing. This procedure
is similar to the known empirical CAPM task of using time series analysis to capture the market
beta and perform a cross-sectional regression of the expected return on the estimated beta over
different time periods. Although this has flaws in several respects, the two-step approach does
not pose a major conceptual problem it shows in CAPM testing. In particular, APT applies to a
subsets of whole of assets.
Data are described in Table 3.1. In selecting them, some arbitrary choices were needed. Although
daily data were available through 1977, the calculations used data only through 1972. The
objective was to make sure to have a calibration sample without losing the benefit of a large
estimation sample, large enough for some statistical reliability even after aggregating the basic
daily returns into monthly returns. The calibration sample is also used for later application and
investigating the problems like non-stationarity.

39
Table 3.1

In the empirical analysis, an estimated covariance matrix of returns was calculated for each
group of assets. Since the calculation of covariance requires simultaneous observations, the start
and end dates are set to exclude very short-lived securities. This resulted in a significant increase
in the time series samples for each group, but there were still some differences in the number of
observations between the groups.
Among the 42 groups, none included all the data for the 2619 trading day. However, the
minimum sample size was still 1445 days, and only 3 groups had less than 2000 days. There
were at least 2400 observations in 36 groups (86%). The size of 30 individual securities had to be
a compromise. For purposes likes estimating the number of return factors present in the
economy, the optimal group size includes all individual assets. But, this would mean a
covariance matrix too large to process.

Estimating factor model

The model analysis consists of the following phases: 1. For individual asset groups, an
exemplary product-moment covariance matrix is calculated from time series returns. 2.
Maximum likelihood factor analysis is conducted on the obtained covariance matrix. This
estimates the number of factors and the matrix loading. 3. The individual asset factor load
estimates found are used to account for the cross-sectional fluctuations in the individual

40
estimated expected returns. The procedure here is similar to the generalized least squares
regression of a cross section. 4) Estimates from the cross-section model are used to measure the
size and statistical significance of the risk premium associated with the estimated factors. This
procedure is similar to estimating the size and importance of the factor score. 5. Steps (1)
through (4) are repeated for all groups and the results are set in table.

Table 3.2

The first panel demonstrates 6% λ per year during the sample period from July 1962 to
December 1972. The first results in Table 3 show the percentage of groups in which more than a
certain number of factors are associated with a statistically significant risk premium. In daily
data, 88.1% of the group had at least one important risk premium factor, 57.1% had two or more
important factors, and one-third of the group had at least three risk premiums. These percentages
are well above what is expected to be ineffective under the null hypothesis by chance alone. The

41
next row in Table 3 shows the expected relevant percentages under this null hypothesis. For
λ=0, the probability of significantly observing at least the specified number of s at the 95% level
is the top of the binomial distribution with a success probability of p = .05.
In fact, if the four factors are really important, the percentage of importance of 4.8 observed in
the five factors is about the same as expected at the 95% level. If three are really important, then
16.7% of the groups where at least four are found to be important exceed 9.75%, which occurs
by chance alone. If there are less than three important factors, the disparity will be much larger.
From this it can be concluded that at least three factors are important in determining pricing, but
it is unlikely that it will exceed four. The second result set still has λ at 6%, but shows the
percentage of groups in which the first, second, and remaining factors generated by factor
analysis have a significant risk premium. As can be seen, all factors are well above the
probability level (5%), with the first two being particularly heavily weighted. The remaining
three are important, and can be due to the order of the elements in the group being mixed.
The second part of the table also shows similar consequences but λ is estimated and not assumed
nevertheless the results comply with the previous data. It can be concluded that three factors are
surely included in expected returns.

Historical development of arbitrage pricing theory

The basic work of the asset pricing model begins with the development of the Capital Asset
Pricing Model (CAPM). CAPM is a one-factor model and its empirical validity depends on the
market portfolio, but the argument that the true market portfolio cannot be inspected severely
limits the acceptance of the model. Ross argues that CAPM has never been tested and will not
be tested because it cannot be identified the market portfolio. Roll expanded his criticism until he
completely rejected the CAPM and became a supporter of Ross's arbitrage pricing theory (APT).
The APT model includes two pricing identifications.
One is called the factor likelihood APT (FLAPT) and the other one is related to the pre-specified
macroeconomic variable APT (PMVAPT). PMVAPT shows the relationship between the
expected rate of return and the number of randomly selected macroeconomic variables.
Conversely, FLAPT provides an intuitive linear relationship between expected return and
asymptotically large potential factors.

42
It is assumed that the asset return rates obtained in APT are determined by "n" independent
factors. These risk factors have different effects on financial assets in different situations and
times. However, it is not possible to give any explanation about the exact number and nature of
these factors. It is assumed that there is a linear relationship between asset return rates and these
risk factors.
It is possible to examine the Arbitrage Pricing Models in the literature under three headings.
These are: Single Risk Factor Arbitrage Pricing Model, Two Risk Factor Arbitrage Pricing
Model and Multi Risk Factor Arbitrage Pricing Model.

Single Factor Arbitrage Pricing Model and Arbitrage Pricing Line.

The simplest case of Arbitrage Pricing Model is accepted as Single Factor Arbitrage Pricing
Model. In this model, it is assumed that only a single systematic risk affects the return on assets.
The relationship between a risk factor and the expected rate of return is represented by a linear
function.

ri = E(ri) + β1,i F1 + ei

Graphically showing this equation, which reveals the relationship between the expected rate of
return and a single risk factor, is called the Arbitrage Pricing Line (APL).

Two Factor Arbitrage Pricing Model and Arbitrage Pricing Plane.

According to the Arbitrage Pricing Model with two risk factors, it is assumed that two different
systematic risk factors will be affected by the return rates of their assets. The relationship
between these two risk factors and the expected rate of return is shown in the following equation
with the help of a linear function.

ri = E(ri) + β1,i F1 + β2,i F2 + ei

The graphical expression of the relationship between the expected return on assets and the two
risk factors seen in the equation is called the Arbitrage Pricing Plane. Arbitrage Pricing Plane is
called an extended form of APL. As with APL, all assets in equilibrium must have a point on the

43
Arbitrage Pricing Plane. According to the Arbitrage Pricing Plane, expected return is on the
vertical axis.

Multi-Factor Arbitrage Pricing Model

According to Arbitrage Pricing Theory, it is assumed that asset return rates are determined by
"k" independent factors. As a result of these risk factors, they have different effects on financial
assets in different situations and time periods.

ri = E(ri) + β1,i F1 + β2,i F2 + … + βn,I Fn + ei

However, it is not possible to make an explanation about the exact number and nature of these
factors. It is assumed that there is a linear relationship between asset return rates and these risk
factors.

44
 4. Comparison of Arbitrage Pricing Theory (APT) and the Capital Asset
Pricing Model (CAPM). Their impact on risk management

When making a comparison between the capital asset pricing model and arbitrage pricing theory,
it could be seen that the capital asset pricing model is based on and Markowitz's optimal
portfolio selection procedure. The arbitrage pricing theory is based on factor models.
As we know, CAPM is a factor model that describes the expected return on a portfolio as a
function of the risk-free rate and excess return on the market portfolio as one factor. APT also
uses factor models, but includes more than one factor. Although the CAPM model states that the
relative risk factor is the market portfolio, the APT model does not say exactly what the risk
factor is.
While the CAPM in practical application uses the market index as the only factor, a
simplification of the original market portfolio model is obtained. The index can best represent
the market portfolio. It is possible that many factors in the APT model give only an approximate
value of the market, while one market index can represent it better in the CAPM model. The
CAPM model requires all investors to act according to the quadratic utility function and have the
same expectations. That is, we can look at CAPM as a special case of APT, with certain
additions and restrictions.
Also, the APT model can be viewed as a sampling model: it can include economic factors that
are best sensitive to a security or portfolio. Sharp fluctuations of one factor, therefore, causes
structural changes in the indicator of the expected return of a security.
The capital asset pricing model and arbitrage pricing theory play a significant role in risk
management. When mentioning risk, it could be understood as both a positive or negative result.
Obviously when taking opportunities with high risk, higher return would be expected, and vice
versa, when taking opportunities with low risk, one would expect lower return. Risk does not
always represent the amount of the loss or gain. In many cases potential loss is big but it could be
forecasted and somehow be given attention to, using risk management techniques. The main
issue is determining the variability of the potential loss. Especially if the loss has the probability
of rising too much.
Risk management contains certain order of procedures, which have the ability to decrease the
size of or completely get rid of the potential loss. In terms of both expected and unexpected

45
losses, the duty of risk management is to be able to predict the and consider how much risk could
be taken in exchange for the potential returns.
The activities of the risk management procedure are formulated such that, as a result they
determine if the expected return is worth the expected risk.
The main steps are the following:

1. Identifying risks
2. Measuring and managing risks
3. Distinguishing the type of risk: expected or unexpected
4. Consider the connection between the risks
5. Design a risk mitigation plan
6. Monitor the mitigation strategy and regulate when necessary

Figure 4.1. The risk management process

46
The identification of the risk can be carried out by brainstorming. The participants are the key
business leaders, who could also involve their subordinates. Industry-level resources could be
used for this purpose such as regulatory standards, etc. One of the most common methods is
scenario analysis, which is analyzing actual loss to determine the amount and rate of different
losses.
Risk management involves taking important decisions including:

• Avoiding the risk

• Retaining the risk, considering the expected return taking into account the frequency and
amount of expected loss
• Mitigating risk, trying to reduce the impact of the risk factor
• Transferring the risk, with the help of derivatives, or purchasing insurance

To understand risk management and its techniques better, firm theoretical foundation is
necessary, including some key theories and models that have an important connection with the
approaches to risk management. These certain key theories and models propose specific
simplifying assumptions. Implementing these theories and models in the real world is not so
simple in all cases, and many factors in real life is difficult to express in models. The duty of
models is to clarify and make less complicated and concentrate on the most significant factors.
As stated by Milton Friedman, models must be assessed according to their ability to forecast.
The assumptions made should not be considered when evaluating. At the same time, a model
does not have to contain all the factors and details in the real world. In other words, the degree to
which it is complicated is not the main requirement from a model.
The main indication that a model is useful is its power of prediction, and the degree to which it
aids in making decisions.
Markowitz’s study in 1952 on the principles of portfolio selection is one of the fundamentals in
modern risk analysis. In this study, Markowitz has demonstrated that an investor acts in a way
such that, when considering different portfolios, the main factors to focus on are the mean, the

47
variance of the rate of return. The assumptions made by Markowitz in this study are the
following: capital markets are perfect, rate of returns are normally distributed.
According to certain theories based on the interest of shareholders, executives of corporations
should not be involved in managing the risk of the corporations.
Franco Modigliani and Merton Miller set forth the idea that the value of a firm does not change
with transactions. The assumption made by Modigliani and Miller is that the capital markets are
perfect: they are highly competitive and the participants are not exposed to transaction costs,
information costs or taxes.
In 1964 William Sharpe established the capital asset pricing model, and in this study Sharpe
states the idea that firms should not focus on risks concerning the firm itself, but rather
concentrate on mutual risks with other firms, because the specific risk can be eliminated by
diversification, without costs, under the assumption of perfect capital markets. As a result firms
do not have to manage risk that individual investor can manage on their own.
The role of theoretical models in this risk management could be expressed such that, the models
assist in determining the risk, and deciding which measures are most suitable in certain cases.
The models also reveal the significance of the difference between specific and systematic risk,
the connection between financial modeling and key parameters.

48
 Conclusion

Certain models such as CAPM and APT have been devised to predict the outcome of an investment in the
financial markets that are constantly changing. The capital assets pricing model (CAPM) is a model
that describes the relationship between the expected return of a portfolio of securities and the
degree of its risk. It was proposed by the financial economist William Sharpe, in his
book “Portfolio Theory and Capital Markets”. An alternative to CAPM is the Arbitrage Pricing
Theory (APT), developed in 1976 Stephen Ross. APT , unlike CAPM, does not assume that
shareholders evaluate decisions using the mathematical expectation and variance of returns. It
assumes that the performance of a stock depends partly on macroeconomic factors, and partly on
events related directly to the company itself. Instead of defining a stock's return as a function of
one factor (the market portfolio return), it presents that return as a function of several
macroeconomic factors on which the market portfolio depends.
The studies on the models goes further, in order to check the validity of them. To test the CAPM,
most of the empirical test conducted on the capital asset pricing model were by the use of time
series test. After estimating the betas this way, using the cross-sectional regression, the
hypothesis was tested. In 1965 John Lintner carried out the first empirical test of CAPM. Later
Douglas conducted the same test with similar results. Lintner applied a two-pass procedure for
individual stocks. But the results do not comply with the CAPM. Miller and Scholes also
performed a study on CAPM using the two-pass method also coming up with very similar
results. To investigate the statistical problems of the two-pass method, Miller and Scholes
arranged numbers and ran a simulation test, coming to a result that the method had statistical
problems. The rejection of the CAPM in the initial verification tests above does not necessarily
mean that the model is defective. To avoid measurement errors that effectively distort SML
estimates, Black, Jensen, and Scholes provided additional insights into the nature and structure of
security returns. Instead of testing individual stocks, they suggested ways to improve the
accuracy of the estimated beta by grouping the stocks into a portfolio. the SML gradient term
was significant and was positively linear as predicted by the model, so the results appeared to be
consistent with zero-beta CAPM. In general, the results were similar for the entire and partial test
period.

49
As for the tests implemented on the APT, In 1975, Gehr collected monthly securities return data
from CRSP tapes for 30 years. He uses the Principal Component Analysis (PCA) method to
extract common factors, performed a cross-sectional OLS regression to get an estimate of the
coefficients of the APT model His results show that over the 30-year study period, only one
factor seemed to be significantly priced, the estimated risk premium was highly biased.
Roll and Ross in 1980 performed the most detailed APT test to date. Their results show that of
the four factors evaluated in cross section, three factors have significant prices. Therefore, their
results support APT. In 1981 Brown and Weinstein try to replicate some of the empirical results
of Roll and Ross (1980) based on another method. It could be seen that none of the grouped
securities has an obvious five-factor return generation process.
When making a comparison between the capital asset pricing model and arbitrage pricing theory,
it could be seen that the capital asset pricing model is based on and Markowitz's optimal
portfolio selection procedure. The arbitrage pricing theory is based on factor models.
As we know, CAPM is a factor model that describes the expected return on a portfolio as a
function of the risk-free rate and excess return on the market portfolio as one factor. APT also
uses factor models, but includes more than one factor. Although the CAPM model states that the
relative risk factor is the market portfolio, the APT model does not say exactly what the risk
factor is.

50
 References

Black, Fischer and Scholes. 1970. "Dividend Yields and Common Stock Returns:
A New Methodology". Cambridge, MA: Massachusetts Institute of Technology.
Sloan School of Management Working Paper #488-70.

Black, F., Jensen, M.C. and Scholes, M. (1972) The capital Asset Pricing Model:
Some empirical tests. Studies in the Theory of Capital Markets. New York:
Praeger.

Bodie, Z., Kane, A., & Marcus, A. J. (2011). Investment and portfolio management. McGraw-
Hill Irwin.

Douglas, G. W. (1967). Risk in the equity markets: an empirical appraisal of market efficiency
(Doctoral dissertation, Yale University).

Fama, Eugene F. 1968b. "Risk, Return, and Equilibrium: Some Clarifying Comments."
Journal of Finance, V. 23: pp 29-40.

Fama, E. F., & French, K. R. (1992). The cross‐section of expected stock returns. the Journal of
Finance, 47(2), 427-465.

Fama, and French (2004) The capital Asset Pricing Model: Theory and
Evidence. The Journal of Economic Perspectives.

Fama, E. F., & MacBeth, J. D. (1973). Risk, return, and equilibrium: Empirical tests. Journal of
political economy, 81(3), 607-636.

Hirschey, M., & Nofsinger, J. R. (2008). Investments: analysis and behaviour (Vol. 281). New
York, USA: McGraw-Hill Irwin.

51
Huberman, Zhenyu Wang Arbitrage Pricing Theory Federal Reserve Bank of New York Staff
Reports, no. 216 August 2005 JEL classification: G12

Lintner, J. (1965a). The valuation of risk assets and the selection of risky investments in stock
portfolios and capital budgets. The review of economics and statistics, 13-37.

Lintner, J. (1965b). Security prices, risk, and maximal gains from diversification. The journal of
finance, 20(4), 587-615.

Miller, M. H., & Scholes, M. (1972). Rates of return in relation to risk: A re-examination of
some recent findings. Studies in the theory of capital markets, 23.

Modigliani and Miller, “The Cost of Capital, Corporation Finance, and the Theory of
Investment,” American Economic Review 48, 1958, pp. 261-297.

Roll R. and Ross S. (1980) " An empirical Investigation of the Arbitrage Pricing
Theory " . Journal of Finance

Sharpe, W.F (1964) Capital Asset Prices: A theory of market Equilibrium under
Condition of Risk. Journal of finance.

Sharpe, “Capital Asset Prices: A Theory of Market Equilibrium under Conditions of Risk,”
Journal of Finance 19, 1964, pp. 425-442.

Sharpe, W. F., & Cooper, G. M. (1972). Risk-return classes of New York stock exchange
common stocks, 1931-1967. Financial Analysts Journal, 46-81.

52