Portfolio Theory

Kevin Kaufhold
2006:11 (a), Latest Revision March, 2012

Preamble and Sources

This document started as an outline of portfolio analysis and management issues, initially
styled as working paper 2006:11. Over the course of time, the outline became so large that it
has now been separated into several documents: Portfolio Theory, 2006:11 (a); Portfolio
Management, 2006:11 (b); and Performance Evaluation:, 2006:12. These papers outline
standard textbook portfolio analysis. Since the standard texts typically do not go into great
detail as to the liability streams nor utility theory, separate working papers have been
developed as Liability Models, 2006:4, and General Utility Concepts Applicable to
Investment Choices, 2007:1. Professional-level Asset Allocation concepts and models were
also developed across several working papers, as well. It is hoped that these papers,
collectively, can provide a well-rounded knowledge of portfolio issues.
Additionally, portfolio texts typically will not provide much analysis on multi-period modeling
of investments. To remedy that situation, a number of working papers, presentations, and a
published article have now been developed on utilizing time horizons in investment analysis.
Please refer to those documents separately, for more information of the applicability and
usefulness of adopting a time-oriented perspective to investment and consumption choices.
Several working papers have also delved into fundamental analysis, asset valuation, and
financial statement analysis. Please refer to these documents for details concerning asset
selection.
The following material is based on several texts, college class lecture notes and summaries of
class outlines: Essentials of Investments, Bodie, Kane, & Marcus, 5th ed. 2004; Investment
Analysis and Portfolio Management, Frank K. Reilly, and Keith C. Brown, The Dryden Press,
6th Ed., 2000; Investments: Analysis and Management, Charles P. Jones, Wiley & Sons, 9th ed.
2004; Investment Valuation, Aswath Damodaran, John Wiley & Sons, 1996; Modern Portfolio
Theory and Investment Analysis, Elton, Gruber, Brown, & Goetzmann, John Wiley & Sons,
Inc, 6th ed. 2002; notes of lectures and course outlines of Prof. Robert Phillips and Rekesh
Bharti, given at Southern Illinois University at Edwardsville, circa 2004-2008; and notes taken
from course readings in CFA I and II, Portfolio Management sections, circa 2012.

Table of Contents
Portfolio Theory ........................................................................................................................... 1
Preamble and Sources .............................................................................................................. 1
Table of Contents ..................................................................................................................... 2
The Investment Background .................................................................................................... 3
A Short Review of Financial Instruments ................................................................................ 5
A Short Review of Statistics .................................................................................................... 9
On Risk and Return................................................................................................................ 10
Historical Views of Risk ........................................................................................................ 16
The Beginnings of Economic Markets .............................................................................. 16
Early Development of Expected Utility............................................................................. 17
From Utility to the Beginnings of Portfolio Analysis ........................................................ 19
Modern Portfolio Theory ....................................................................................................... 21
The Start of Portfolio Theory............................................................................................. 21
The Development of MVO and the Efficient Frontier....................................................... 22
The Optimal Portfolio ........................................................................................................ 29
Single Factor Asset Market Model .................................................................................... 32
Critique of MPT ................................................................................................................. 34
Other Portfolio Techniques ................................................................................................ 35
Mathematical Portfolio Relationships / Constraints / Assumptions .................................. 41
Asset Pricing Models The CAPM & APT .......................................................................... 47
Introduction ........................................................................................................................ 47
Assumptions....................................................................................................................... 47
Details of the Model........................................................................................................... 48
The CAPM and the Single-Factor Index Model ................................................................ 53
Tests and Critiques of the CAPM ...................................................................................... 61
The Arbitrage Pricing Theory (APT) ................................................................................. 63
Other Models ..................................................................................................................... 67
The Efficient Market Hypothesis ........................................................................................... 75
Historical Development of Efficiency and Behavior ......................................................... 75
Are Markets Efficient? ....................................................................................................... 77
Interpreting the Evidence ................................................................................................... 91
Conclusion ......................................................................................................................... 92
Behavioral Finance ................................................................................................................ 94
Appendix I - Portfolio Equation Review ............................................................................... 97

The Investment Background

The essential nature of investment involves a deferring current consumption for use at a future
date. Investment is the current commitment of money or other resources in the expectation of
reaping future benefits.
Assets are classified as either Real in nature (those assets used to produce goods and services)
or Financial (assets involving claims on real assets or the income generate by them). Note that
this is slightly different that legal definitions of Real Assets which mostly includes real estate
and other hard physical assets, while things not physically attached to the land are considered
personal property, such as equipment and other physical assets. Real assets generate net
income to the economy, while financial assets define the allocation of wealth between
investors.
Financial assets are commonly divided into various groups: fixed income securities / debt
securities (money markets, CDs, government and corporate bonds, etc), equities (i.e. common
or preferred stock in a business), and derivative securities (options and future contracts).
In the financial markets, issues involve the timing of the consumption at the individual level
(consume now or consume later), the allocation of risk, and the separation of ownership and
management. This in turn presents an agency problem between the managers of a business and
its owners. The Investment Process involves asset allocation, security selection, the risk-return
trade-off, market efficiency, and active versus passive management, in roughly that order.
Capital Markets are normally thought to be competitive, with the risk-return trade-off playing a
key role. Assets with higher expected returns will carry greater risk.
The competitive nature of the markets is a key finding of the Efficient Market Hypothesis
(EMH). Efficiency theory has a central paradox to it: it takes active management to ensure the
correct market pricing and overall market efficiency occurs that makes passive management
superior in many instance. Thus, we may observe near market efficiency, or operational
efficiency, as active managers attempt to take advantage of the inefficiencies of the
marketplace, thereby driving the pricing back to an efficient level.
There are numerous participants in the financial arena. Businesses are net borrowers of assets,
as they raise capital to produce income. Households are net savers, purchasing the securities
issued by firms needed to raise funds, and governments can be both net borrowers and lenders.
Financial intermediaries have developed over the years to bring lenders of financial assets
(businesses trying to sell bonds, for instance) together with the lenders, purchasers and
investors (banks, individuals, and businesses buying securities). Intermediaries include brokers,
investment companies, and credit unions. Investment bankers specialize in advising businesses
and governments on the handling, pricing, and marketing of securities issued to the public.

Market structures include the direct search market, where buyers and sellers seek each other
directly. Brokered markets can involve both the primary sale of securities (IPOs) and the
secondary sale of securities (the stock markets). Dealer markets involve those transactions in
which brokers will buy and sell from their own accounts from a brokered market typically.
Dealers can thus save on search costs, but can also profit in such a transaction. Auction
markets are places where buyers and sellers gather to trade assets in an auction-bidding format
(NYSE is the classic style of an auction market, while NASDAQ is an electronic form of the
auction market).
Recent trends. Globalization of the financial markets has been occurring, with American
deposit Receipts (ADRs) and World Equity Benchmark Shares (WEBS) trading world-wide.
ADRs are domestically traded securities that represent claims on shares of foreign stock.
WEBS are portfolios of foreign stocks in selected countries. Globalization allows domestic
firms to compete in a global marketplace. Globalization is a secular trend to world-wide
investment environment resulting from the emergence of capitalism as the dominant way to
organize a socio-economic political system. Securitization of assets allow for the pooling of
real assets into homogeneous securities that can then be traded in financial markets. Mortgage
backed pass-through securities of the Government National Mortgage Assn (GNMA) were in
1970s, whereby numerous individual real estate mortgages were bundled together and then resold in the financial markets as a security. Thus, individual home mortgages started to be
traded just like any other security issue. Securitization expands the choice of investors. Many
types of loans can be packaged and sold as a financial security (student loans, car loans, home
equity loans, etc). Brady bonds and the US government requiring credit enhancements have
been used by banks packaging weak bonds and debts to other investors through the use of US
treasury notes acting as collateral in the event of default. Financial engineering is also
occurring the bundling or unbundling of assets into and away from securities. This allows
for a complex mix and assignment of returns and risks, sorted out by various classes with an
individual security. Computer networks and internet points are also developing that allows for
the cheaper and faster trading of securities.
In all likelihood, globalization will continue to occur with increasing frequency; securitization
will continue to develop; there will be additional developments in derivatives and exotics;
and there will be an integration of investments with corporate finance concepts.

A Short Review of Financial Instruments

The major classes of financial instruments include debt (money market, CDs, and bonds);
equities (common and preferred); and derivatives.
As to debt instruments. Treasury Bills (T bills) are a form of borrowing by the US
government. T Bills are sold to brokers at a discount, and then the holder of the bills receive
face value at maturity. The difference between the purchase price and face value is the rate of
return. Sales of T Bills are conducted weekly with 28, 91, and 182 day maturities. The bidding
can be competitive or non-competitive in nature. Noncompetitive bids are an unconditional
offer to buy at the average successful bid price. Individuals can purchase the T-bills directly at
the auction or from a secondary market.
T bills are regarded as the RFRR, both due to their being backed by the US government and
also selling at a discount to face value at maturity. This discount then allows the calculation of
a dollar value certain in advance, thereby generating a known rate of return. This is not true of
bond rates of return, where the value of a bond will float, depending upon prevailing interest
rates. The 30 day T Bill is typically sued as the monthly riskless rate of return, although the 91
day T-Bill is also commonly used, too.
Certificate of Deposits are time deposits with a bank. They may not be withdrawn on demand,
as interest and principal are paid only at the end of the deposit period. They are FDIC insured
since they are considered as cash deposits at the bank level.
Commercial Paper is debt of companies, although it can also accrue at the individual level.
Sometimes, a bank line of credit or other collateral backs CP. Maturities range up to 270 days.
Longer maturities require registration with the SEC. CP trades in secondary markets, so it is
very liquid. Most CP carries ratings by Monodys and others.
Bankers Acceptance is a note by a person or company payable to a bank upon certain
conditions at a certain maturity date. It becomes secured paper at the time of acceptance, and
thus can be traded in the secondary markets as a secured instrument. This type of paper will
normally carry the banks creditworthiness as collateral.
Eurodollars are dollars denominated deposits at foreign banks of local domestic branches of
foreign banks. The regulatory powers of the Federal Reserve Board do not exist since it
involves foreign banking. There is also the Eurodollar CD, with the CD being drafted on the
creditworthiness of a foreign bank.
Repurchase Agreements (repos) and reverse repos are used for overnight borrowing. The
dealer sells secured paper to an investor on an overnight basis, with an agreement to buy back
the security the next day at a slightly higher price, with the price differential being the
overnight interest. The securities serve as collateral. A term repo carries a term of 30 days or
more. These forms of repos are very safe since they are backed by government securities. A

reverse repo is the mirror image of repo the dealer finds an investor holding government
securities and buys them with an agreement to resell item at a specified price on a future date.
Brokers Calls are made against individuals who buy stocks on margin from a brokers
account. The broker may in turn borrow from a bank to cover the margin, agreeing to repay
the bank on call if the bank requests it.
As to Federal Funds, banks are required to maintain a certain level of deposit or reserve with
the Federal Reserve Bank. Funds in the banks required reserve account at the Fed are called
Federal Reserve Funds, or Federal funds. Banks with excess federal reverse funds can borrow
from banks with an excess of federal funds above the minimum required reserve amount in
what is called the Federal Reserve fund market. Interest is at the federal fund rate, set by the
Fed. This rate is used as a yardstick for the interest given in the money markets.
The LIBOR Market is the London Inter-bank Offer Rate that large banks in London are willing
to lend money at between each other. This is the often-quoted ST interest rate in the European
money market. It is used as a base rate for setting loans in the US, and is usually quoted in US
dollars. LT commercial rates are often pegged to the T bill or the LIBOR, with the interest rate
on the loan being a fixed amount plus either the T bill or LIBOR.
Most money transactions are not risk free, as they carry the risk of default. Collateral and other
forms of security may reduce that risk. As such, securities in the money markets will carry a
yield higher than the 91 day T-bill rate. There is typically a spread between 90 day CDs and
91 day T Bills of between 0.5% and 3.0% during extreme times of financial distress (OPEC oil
crisis, etc).
Treasury Notes & bonds are obligations of the government with less than 1 year maturity. T
notes range from 1 year to 10 years. Treasury bonds are between 10 to 30 years in duration.
Both forms of Treasury securities make semi-annual payments called coupon payments. Some
of the older long bonds are callable by the Treasury in the last five years of the bond, for par
value, whereas notes are generally not callable. Yield to maturity is the yield if held until
maturity. It is calculated by determining the semi-annual yield and doubling it (rather than
compounding it in two half year periods). Thus, the quote is an average annual yield (APR)
instead of the more accurate effective annual yield.
Federal Agency debt are debt instruments issued by FNMA, GNMA, Federal Home Mortgage
Corp (FHLMC, or Freddie Mac), and Federal Home Loan Bank (FHLB). These bonds are not
backed by the full faith and credit clause, and thus carry some risk, although there is an
assumption in the private markets that the government will back them if necessary. The yield
spread over Treasuries is usually quite small, as a result.
International Bonds exist, and a capital bond market is centered in London, where 70 countries
have offices. Many countries will issue government bonds to trade at London and other
locations. A Eurobond is now offered that is denominated in a currency other than its own
governments currency (for example, a dollar denominated British bond is a Eurobond). This
gets confusing with a eurodollar, which is the new currency of the EU.

Municipal Bonds are federally tax-exempt bonds issued by state and local governments, and
are exempt from state and local taxes in the issuing state (as to interest income only. The
capital gains are taxable). General obligation bonds, or GOBs, are issued on the general power
of the entity to tax. A revenue bond is backed by the revenues of a specific project. Maturities
vary widely. Because of the tax-exempt nature of the bonds, the yields are lower. The investor
should compare after-tax returns when comparing tax-exempt bonds with other forms of debt.
The equivalent taxable yield is used here, and is calculated as: r = r m / (1-t). By rearranging
the equation, the tax bracket that investors are indifferent between taxable and non-taxable
bonds is: t = 1 (rm / r). rm / r is referred to as the yield ratio.
Corporate Bonds have default risk as a key component. Unsecured bonds are called debentures.
Many bonds carry options, such as callability of a bond by the issuer, or convertibility by the
bond holder to common stocks. Corporate debt is often unsecured. Most corporate bonds
issue fixed coupon rates with semi-annual payments based on face value. Bond valuations will
fluctuate, depending upon interest rates and risk / default factors. TVM concepts apply.
Mortgage backed bonds were developed in the 1970s as a way to develop a secondary market
at the national level for what was then only a local lending activity. Mortgages are bundled
together and sold on a bond market with the mortgages acting as security. These bonds are
also called pass-through bonds. GNMA bonds carry a guarantee from the US government as to
timely payment of interest and principal. There has been explosive growth in mortgage back
securities over the years.
As to Equities. Common stock is the ownership interest in a corporation. One vote per share
is the normal rule with common ownership. The text reviews a typical format of board of
directors and managers. Ownership allows only a residual claim to the assets of a business at
liquidation, but also provides for limited liability.
The key features of common stock include: 1) one share-one vote (normally straight voting,
but can also be cumulative); 2) dividends are not deductible by the corporation; 3) owners have
only residual claims; 4) and owners have limited liability.
Preferred stock normally does not carry voting rights in a corporation, but usually carries a
fixed stream of dividends which is sometimes cumulative in nature. Preferred is considered
equity for purposes of accounting balancing statements, but is also viewed as a hybrid type of
security, with aspects of both debt and equity. Preferred is often viewed as a fixed income
security. Dividends are not tax-deductible for the corporation, unlike interest payments on debt.
But, corporations owning preferred stock in another company can exclude up to 70% of
dividends from domestic corporations in their computation of taxable income. Many preferred
stock carry convertibility options for the stock-holder, while others are redeemable by the
issuing firm. Adjustable rate preferred stock also is developing. Preferred stock is valued as a
perpetuity, in most cases.
There are various types of stock market indexes. The DJIA is a price weighted average of 30
industrials on the Dow, adjusted for stock splits. This results in the need for a divisor, to

account for all of the splits over the years. The S&P indexes are market-weighted. This
eliminates the worry over splits, but results in a large cap type of emphasis in the Indexes,
whereby a few of the largest businesses (GE, MSFT, etc) dwarf make up a very large
percentage of the total weight of an index. The Wilshire 5000, the Russell 2000 and 3000
indexes, etc, also exist, and these indexes bring in more of the small caps, but still marketweight. There are also equally weighted indexes (Value-Line arithmetic and geometric).
International indexes are very much in evidence, such as the FTSE (published by the Financial
Times of London), the Nikkei 225 (price weighted) and Nikkei 300 (value weighted), and the
DAX German stock index. Morgan Stanley has an international index called the MSCI, as
well.
There are also bond market Indexes. The investment houses have developed many bond
indexes over the years. The Lehman bond aggregate is an index of many forms of bonds.
Indexes can be price weighted (as is the DJIA), market value weighted (S&P 500, NASDAQ),
or equal weighted (Value Line).
Derivatives. These are securities issued on an underlying asset. A call option carries the right
but not the obligation to buy an asset at a specified price and date. A put option carries the
right but not the obligation to sell at a specified price and by a particular date. Future contracts
call for delivery of an asset at a specific date and price. Futures carry the obligation to sell or
buy a long position. A call option on a future carries the right but not the obligation to buy.
Call options are thus purchased, whereas a future contract is merely a contract with no money
exchanging hands until maturity.

A Short Review of Statistics

As to Raw Data. The mean of the population is x = x / N = (x1 + x2 + xn) / N. This is
just the sum of all data divided by the number of data. The mean of the sample is the same
calculation, but with a different notation. It is: x bar = x / n.
The population variance is 2x = (x )2 / N. The variance can also be seen as the mean
squared error. The sample variance is: S2x = (x x bar)2 / (n 1). The standard deviation for
either population or sample variances is merely the square root of the variance. A short-cut for
variance is: ( x2 ( x)2 / N ) * (1 / N). The short-cut for a sample variance is the same,
except that it uses the n-1 divisor and not N.
The covariance is the central feature of correlation statistics. A covariance can be negatively
or positively correlated. The population covariance is: xy = (1 / N) [ (x x ) (y y) ], and
the sample covariance uses the n-1 divisor instead of N. The short-cut is: (1 / N) (xy
((x)(Ey) / N) ]. The correlation coefficient is driven by the covariance.
As to Grouped Data. The outcomes of grouped data are associated with the probabilities of
the data. The mean becomes the expected value of a probability distribution. E (x) = xi * p
(xn).
The variance becomes: E [(x ) 2 ], which is: (x )2 p (xi). The divisor N drops out since
the data is already normalized with the data always equaling 1.00. The variance short-cut is
the following: E (x2) u2.
The co-variance is: xy = E [ (x x ) (y y), and this = p (x) (x x) (y uy). The shortcut is: E (x y ) (ux uy).
Almost all equities will have a positive correlation with the broad market indexes. The slope
of the regression line of an equity and the market is beta. This is the covariance / the product of
the standard deviations of the two, or more, equities. In math terms: = xy / (x y ).
On probability functions. With a normal distribution, the probability is bell-shaped and
symmetrical about the mean. Standard deviation can be measured and visually observed on
each side of the mean. With skewed distributions, the curve is no longer symmetrical, and the
median can be either higher or lower than the mean.
Characteristics of Probability Distributions include: first, the mean, or most likely value;
second, the variance and standard deviation; third, skewness; and fourth, kurtosis. Skew is the
sideways movement of the distribution away from a bell-shaped normal. Kurtosis if the
vertical rise or reduction from a normal. If a distribution is normal, then skewness = 0 and
excess kurtosis = 0, and the distribution becomes related to the mean and the variance.

On Risk and Return

Return Calculations. All asset returns have two components: income and capital pricing
changes. The Holding Period Return (HPR), also referred to as total return, is the rate of return
at a given moment in time, for each period. HPR = (EV BV + Dividend) / BV.
Total return is the % measure of all cash flows of a security compared to its purchase price. It
is: TR = (dividend or interest + ending price beginning price) / beginning price.
Several other forms of return can be calculated. A return relative is the total return for any
given time period stated as a basis of 1.00. This is used for cumulative wealth indexes or
geometric returns. The RR = TR (in decimal form) + 1.00. So, a total return of 10%, or 0.10
in decimal form, is 1.10.
A cumulative wealth index compares initial wealth to a series of returns over time. And will
display the historical geometric returns. The wealth index is composed of both capital gains
and dividend yields. The cumulative wealth index is quite useful in showing the power of
compounded returns. For example, stocks grew by a factor of 5000 in the modern period since
1919, whereas, bonds grew by only a factor of 100, on the cumulative wealth index. It is
calculated as: CWI = WI (1 + TR1) (1+TR2)(1+TRn), where WI is initial wealth. Then
TRn = (CWIn / CWIn-1) 1, and TR can be calculated if we only know CW in two different
periods.
Cumulative wealth indexes of the various asset classes show the historical geometric returns.
The wealth index can be seen as being composed of capital gains and dividend yields. The
cumulative wealth index is quite useful to show the power of compounded returns. For
example, stocks grew by a factor of 5000 in the modern period since 1919, whereas, bonds
grew by only a factor of 100, on the cumulative wealth index.
The arithmetic mean = X bar = X / n. This is the mean average that is normally used as the
average of typical calculations.
The geometric mean is the return giving the same cumulative figure as the sequence of actual
returns. It is also called the time weighted average return. The geometric mean is typically
used for % changes over time, as it is considered to be the compounded rate of return over
time. In the following equation, note the similarity with the cumulative wealth index,
discussed below.
G = (1+TR1)(1+TR2).(1+TRn) (1/n) 1
The geometric rate of return will be lower than the arithmetic mean because of compounding.
It takes less of a return to compound to the same net wealth over time, since the returns are
compounded, with interest essentially becoming principal in the next period. The geometric
mean is better to use for a time series, whereas the arithmetic mean is appropriate for use in
single periods.

10

If we know the arithmetic mean of a series of asset returns and the standard deviation, we can
approximate the geometric means. As increases, and the arithmetic mean stays constant, the
geometric mean decreases. As the spread increases between the GM and the AM, the
variability of returns must increase [in fact, a higher generates the higher spread between the
two types of means]. The formula is: (1+geo. Mean) 2 = (1+AM)2 2.
With money coming in and out of a fund, the dollar weighted return is used. This is the IRR of
an investment, and is the interest rate that sets the PV of the CF to the initial cost of the
portfolio. The IRR = c1 / (1+r) + c2 / (1+r2)2 + cn / (1+r)n. The Internal Rate of Return is
the discount rate that produces a present value of the future cash flows that is equal to the
investment amount. The IRR considers changes in investment. The initial investment is an
outflow, and the ending value is considered as an inflow. Additional investments are negative
flows, and reduced investments are positive flow.
The APR is the annualized percentage return and = per period rate * # of periods / year. Or,
APR can be defined as: APR = (periods in year) * (rate for period). The effective annual rate
(EAR) can calculate the rate of return for continuous compounding. EAR = effective annual
Periods/yr
rate (1+ rate for period)
-1
The expected return is the sum of the probabilities of each period * the HPR in each period. It
is the mean value of the probability distribution of the HPR. The equation is:
E (r) = p (s) * r (s)
where p is the probability of period s, and r is the HPR of period s.
Currency risk is the risk that a change in the value of the dollar and the foreign currency will
negatively impact the investor. If an investment denominated in a foreign currency gains in
value relative to the investors domestic currency, the investment has experienced a gain from
the currency movement. The total return (TR) after currency adjustment is:
TR in $ = (RR * (EV of foreign currency / BV of foreign currency)) 1.0.
One has to calculate the TR of the asset first, convert that to the RR, and then adjust for the
currency fluctuation. The subtraction of the 1.0 at the end of the equation is to take the RR
back to a TR in $ values.
Inflation adjusted returns are considered to be real returns as opposed to nominal or money
returns. The CPI is normally used for the rate of inflation. The approximation of real return is:
TRIA = (1+TR) / (1+rate of inflation) 1.
The required rate of return is that minimum rate that an investor will accept to compensate for
deferring compensation. The required rate of return is made up of the following components:

11

1) TVM during the investment period; 2) the expected rate of inflation during the periods; 3)
the risk involved.
A more involved formulation for the required rate of return is that it is composed of 1) the
economys RRFR (which is influenced by long run real growth rate); variances affecting the
NRFR, including ease or tightness in capital markets and the rate of inflation; and 3) the risk
premium of the investment.
The nominal RFRR is the default free rate of return on a nominal basis. The RFRR is
considered to be the 91 day Treasury Bill, since it is backed by the US government and is thus
default free as well as having a definite sum certain paid in a short time frame (It is sold at a
discount to face value, and then matures to the stated face value in 30 months). The real RFRR
adjusts for inflation. The equations are:
NRFR = (1 + RFRR) * (1 + Expected inflation rate) 1
RRFR = [(1 + NRFR) / (1 + Expected rate of inflation)] 1
On notions of risk, the probability that the actual outcome is different than the expected
outcome is considered to be risk. Various types of risk exist. Interest rate risk is the uncertainty
that the lending rate will change. Inflation risk is the possibility that inflation will increase.
Business risk is the uncertainty of income flowing from a business. Financial risk is the
uncertainty from financial leverage. Liquidity risk is the inability to find buyers or sellers at
any one time.
Risk aversion is where, given the choice between two assets with equal rates of return,
investors will choose the asset with the lowest risk or variance of expected returns. Evidence
of risk aversion exists with insurance purchases, bond and credit risks, etc. Risk aversion is
assumed in portfolio decisions of risk and return (thus implying a positive relationship between
risk and return).
Most investors require higher rates of return on equities beyond the RFRR due to the above
noted uncertainties. This is known as the risk premium, and it acts as a composite for all of the
extra risk above that of the RFRR. The risk premium is the additional compensation for
assuming risk. Most investors will require higher rates of return to compensate for any
perceived uncertainty regarding the expected rate of return. The amount over the NRFRR that
is required by investors to compensate for the uncertainty is called the risk premium. The
equity risk premium is the difference between the return on stocks and the risk free rate of
return.
Equity Risk premium = ((1+TRstocks) / (1+RFRR)) 1
In terms of portfolio theory, the risk premium is normally considered to be the excess return
above the RFRR, and is: E (rp) rf. The excess return, of course, is the return in excess of the
T bill rate. An investor is normally considered to be risk averse. Note that the actual return
may be different than the expected return, since the expected return involves probabilities
distributions.

12

Historical rates are calculated in geometric terms. The risk premium noted above is the actual
rate of return based on historical returns, and differs from the expected risk premium, which is
the required rate of return of finance theory.
All forms of risk are thought to be included in the risk premium, but several fundamental
sources of risk are of special concern to investors. These include:
-

business risk (the uncertainty of income flows from a business);

financial risk (the uncertainty of financing methods of a business);
Liquidity risk (uncertainty regarding the lack of a secondary market for the investment);
Exchange rate risk (the uncertainty of returns due to currency fluctuations).
Country risk (the political uncertainty of a country).

Thus the risk premium = (total of all fundamental risk).

Markowitz and Sharpe both believed that investors should use external measures of risk, and
that individual asset co-movement with a market portfolio is the assets systematic or market
level risk. Systematic risk can be thought of as that portion of the assets variability of pricing
that was attributable to the pricing variance of the entire market portfolio. The risk premium =
(systematic risk). Several studies have shown a close relationship between the market
measures of risk and the fundamental sources or measures of risk. Thus, these two measures of
risk can be considered complimentary in nature. In a well functioning capital market, external
functions of risk should reflect and incorporate all fundamental notions of risk.
Non-systematic risk (also known as asset-specific risk) is that risk which can be diversified
away. Investors should not expect any additional return for bearing non-systematic risk, since
it can be completely diversified away by simply taking the market portfolio.
With modern finance theories, all types of risk are assumed to affect the pricing of assets.
Ultimately, the risks are incorporated or priced into the asset pricing structure. Thus, the
basic form of risk at the theoretical level is considered to the variability of pricing returns, and
more specifically, the dispersion or range of a probability distribution of expected returns, as
measured by standard deviation. Variance is the expected value of the squared deviations from
the mean value, while the standard deviation is the square root of the variance.
Var (r) = 2 = p (b) [ r( s) E ( r) ] 2
Standard deviation measures the dispersion of the returns against its expected returns, and is
the basic measure of pricing related risk.
2 = ( X X bar) 2 / (n - 1)
=2
The Risk free rate of return is the rate of return that can be earned with certainty. The 91 day T
Bill rate is normally used, since this rate is known at the time of issuance of the note, with a

13

stated discount to face value making the rate of return both certain and risk free in that it is a
debt of the US government that carries the full faith and credit of the government (some debts
do not carry full faith and credit of the US government, only an agency such as the TVA or
Fannie Mae).
Historical rates are calculated in geometric terms. Historically, risky assets have provided
higher average returns. The historical risk premiums are considerable and commensurate with
their greater risk. Markets with the highest average returns are also the most volatile as to
pricing deviations. The probability distribution of variability assumes a normal distribution.
The inflation rate is the rate of rising prices as measured by the CPI. The nominal interest rate
is r, not adjusted for inflation. The real interest rate is adjusted for inflation. Approximately, r
= R i , where r is the real inflation rate, R is the nominal rate, and i is the inflation rate. More
specifically, r = (R i) / (1 + i). The Fisher equation where R = r + E ( i ).
Portfolio Expected Risk and Return. The 91 day T-Bill is considered to be the RFRR
because the rate of return is known in advance due to the T bill be sold at a discount. y = % of
assets in a risky assets. 1 y = % of assets in RF assets. Rate of return = rp = E (rp).
Standard deviation is = p. rf = rate of return on RF asset. E(rc) = y E(rp) + (1 - y) rf, or in lay
terms, the expected rate of return on the portfolio is the expected rate of return on the risky
assets times the % of assets in the risky asset plus the RFRR times the % of assets in the RF
asset.
The Capital Market Line (CML) can be sketched with the expected rate of return, E(Rp) being
on the y axis, standard deviation being on the x axis, and the RFRR being the y intercept. The
risk premium = E (rp) Erf. The slope of the CML, s = (E (rp) rf) / p. The Capital Market
Line is the plot of risk / return combinations available by varying the portfolio allocations
between the RF asset and a risky portfolio.
Note that when we are dealing with expected returns, we are discussing a probability
distribution. So:
rp = wa ra + wb rb + wn rn.
E (rp) = wa E (ra) + wb E (rb) + wa E (rn),
where generally, E (x) = weighting * return. Then beta also becomes weighted, whereby =
wa a + wb b wn n.
The variance equation for a two asset portfolio is:
2p = w2a 2a + w2b 2b + 2wa wb ab.
This equals:
2p = w2a 2a + w2b 2b + 2wa wb ab a b,

14

where is described as rho (from the greek alphabet) and is the correlation coefficient. The
coefficient is then defined as ab = ab / ( a b), or in plain language, as the covariance of the
two assets / the product of the standard deviation of the two assets, separately. Also note that if
the assets are statistically independent, then the coefficient = 0, and the variance of the
portfolio simply drops to the weighted average of the variance of the two assets, or 2p = w2a
2a + w2b 2b. Still assuming that the returns are statistically independent of each other, the
standard deviation of a portfolio can conceptually be reduced to: p = i / n.
Also note that the variance of a T Bill is zero, so beta of a T bill will also be zero. The
covariance of T bill with any other asset = 0, because T bills do not vary.
The Treynor Index is the reward to variability ratio. This is the ratio of the risk premium to the
standard deviation. This = (Erp rf) / p, which is nothing more than the slope of the CML.
The CML assumes private lending at the RFRR, but if commercial lending is greater than the
RFRR (it always is), then the slope of the CML rotates downward, leaving the CML still intact.
Interestingly, a 1996 WSJ article by Jonathan Clements in which it is recommended that for
stability and risk aversion purposes, a portfolio should hold stocks to T bills and not stocks to
bonds. This is due to bond variability versus certainty with T Bills. The articles states that
over the last 25 years, a 75-25 mix of stocks to T-Bills performed as well as a 60 40 split
between stocks to bonds.
The CML uses the market index portfolio as the risky asset, and thus uses a passive strategy.
Also note that leverage is contemplated with the CML. Borrowing activity will occur above
the 100% of assets of a portfolio held in a risky asset. This would be through margin activity.
Below the 100% amount, an investor can be considered to be lending money.
Greater levels of risk aversion lead to larger proportions of the risk free rate. Lower levels of
risk aversion lead to larger proportions of the portfolio of risky assets. The willingness to
accept high levels of risk for high levels of returns would result in leveraged combinations.
Texts provide equations for the measurement of risk aversion.

15

Historical Views of Risk

The Beginnings of Economic Markets
The 19th Century was replete with examples of dubious stock activities, manipulations, and
outright frauds. Even when no suspicious activities were evident, private markets regularly
became embroiled with speculative undertakings and the madness of crowds. 1 The capital
arena was therefore seen as a very uncertain place. By the turn of the 20th Century, stock
investing was generally limited to equities displaying a stability of dividends and earnings. A
basic similarity of desires existed on the part of both stock and bond investors: everyone
wanted dividend yields and safety of investment.2 Investing in common stocks was very
similar to that of bonds. A stable business with sufficient earnings abilities was highly
desirable, just as a stable dividend was highly priced with bonds. From the end of WWI to the
events leading up to the 1929 stock market crash however, the focus of investing shifted
towards the future earnings trend of the stock at hand. Buying stocks with future earnings, but
without regard to current market pricing levels, was seen as a major causative agent for the
1929 pricing bubble.3
During the Depression, Benjamin Graham and others concentrated on stability of past company
finances, the reasonableness of market pricing compared to a fundamental level of company
value, and net asset value of businesses. Risk was generally seen as a possibility of a loss, but
there was not a lot of effort devoted to a formal risk definition. More discussion occurred
regarding the differences between investments and speculations. The emphasis was upon
intrinsic business valuation, interest coverage on bonds, and a general level of safety for stock
investments. Equities trading above book values were considered risky and unsafe as
investments. Most of the analysis revolved around performance standards and quality of
business operations.4 Remindful of the Depression, many investors had very cautious views
towards stocks and corporate bonds. Such attitudes continued through the WWII years and
beyond. The fundamental analysis method of asset valuation currently in use among financial
analysts developed directly from Grahams thoughts.
As we shall see in other working papers, many long-term investors will still today primarily
rely upon and use fundamental notions of risk in their business endeavors. There is a very good
reason for this reliance: in the long-term, fundamental notions of risk may have more
usefulness than pricing connotations. But with the arrival of utility theory, the disciplines of
Economics and Finance moved into more mathematical arenas.

See for example, an excellent rendition of tulipmania, Charles MacKay (1852).

For a nice historical view of what Depression era scholars thought of pre WWI capital markets, See Grahams
and Dodds 1934 edition of Security Analysis, at 303-316.
3
Graham and Dodd, at 310-313. Graham felt that the new-era theory of investment advocated by Edgar
Lawrence Smith in Common Stocks as Long-Term Investments (1924), led to stock price acceleration not founded
upon a fundamental level of value. Id, at 312.
4
See Grahams and Dodds proposed canons of stock market investment at Id, p.317-338.
2

16

Early Development of Expected Utility

While other working papers will go into much greater detail discussing the general case of
utility, a very brief review of the historical aspects of utility provides a good backdrop to
modern portfolio theory.5
Even while Grahams thoughts of fundamental analysis were taking hold in the investment
world, a more mathematical description of risk was being simultaneously developed in several
theoretical treatises. Utility theorists referred to the writings of famous mathematicians and
philosophers from as far back as the 17th century. Pascal and Fermet believed that the value of
a lottery should be equal to its mathematical expectations, and therefore identical for all
people, independent of their risk attitude. Another famous mathematician, Daniel Bernoulli,
wrote a paper in 1738 that contradicted this thinking. Considered radical for the time, Bernoulli
felt that the mathematical expectation of monetary outcomes of a lottery is not an adequate
measure of its value. Instead, a lottery should be valued by the expected utility that it provides.
Bernoulli was the first person to suggest a nonlinear relationship between wealth and the utility
of consuming that wealth. The real question becomes how much happiness is to be derived
from the monetary outcome, rather than possessing the monetary outcome itself. The
relationship between satisfaction and the monetary outcome, can be described as a utility
function which is attained by the agent as a result of possessing the wealth. The monetary
outcome is objective, and is noted as: x. The satisfaction of that wealth is subjective in nature,
depending upon the preferences of each agent. It is noted as: u (x). Utility becomes an indirect
measure of wealth, and is the highest attainable level of satisfaction for a given set of goods
affordable with the individuals income. Mathematically, utility of wealth is some function of
wealth, or u (x) ~ (x).
Francis Edgeworth introduced the generalized utility function U (x, y, z, ...) into the economics
profession and drew the first indifference curve in 1881. Edgeworth's original two axis
depiction was then developed into the now familiar box diagram by Vilfredo Pareto in his
Manual of Political Economy (1906). Pareto effectively began the field of modern
microeconomics when he developed the notion of Pareto-optimality. Society enjoys maximum
optimality when no one can be made better off without making someone else worse off.
Arthur Lyon Bowley popularized the Edgeworth box in The Mathematical Groundwork of
Economics (1924), so much so that it is commonly known as the Edgeworth-Bowley box.
Many years later, von Neuman and Morgenstern (1948) refined Bernoullis thoughts into the
modern-day expected utility (EU) model.6 Several assumptions were made regarding economic
behavior. Investors were assumed to make comparisons between certain outcomes. They are
consistent in the ranking of their outcomes. Investors / agents are indifferent between two
equal probabilities. There is a value at which investors are indifferent between a gamble and a
certain equivalent. These assumptions led to the development of indifference curves, which
5

Many of the concepts expressed in this section are very brief summaries of a chapter on utility by Elton, Gruber,
et al, (2002), and a utility text by Eeckhoudt, Gollier, and Schlesinger (2005).
6
Theory of Games and Economic Behavior, J. von Neuman and O. Morgenstern, Princeton University Press,
1948.

17

have become a standard way of describing economic choices under uncertainty. Based on these
assumptions, an investors gamble can be mathematically expressed as the probability of the
utility of wealth, with notation of: P1 u (w1), which is essentially a simple mathematical
definition of expected utility. Arrow and Pratt then developed expected utility into a market
equilibrium theory, and the stage was set for quantitative scrutiny of risk and return.
On the calculation of a specific equation that makes up a utility function, further assumptions
are needed. A more-to-less preference is almost universally assumed, as it is consistent with
empirical evidence.7 Risk aversion is another factor to consider in developing a utility function.
An investor could be risk averse, risk neutral, or risk taking with regard to a fair gamble.
Assuming a one period model, a risk averse agent will always prefer the utility of the lottery
with certainty rather than the lottery itself. This means that any risk averse agent will have a
utility function that is concave, where u (x) is sketched out on the x axis of a graph and wealth
is on the y axis. Utility of wealth increases as wealth increases, but at a decreasing rate, thereby
producing the concave curve. This is referred to as diminishing marginal utility, and is basic,
economic concept that has been applied to a wide variety of micro and macro economic
settings.
In the following graph, the simple linear utility function is displayed in pink. A square root
utility function is in dark blue. Another square root function is in light blue. Both of the square
root utility functions display a concave pattern, indicating risk averse preferences. The linear
function would be representative of a risk neutral investor. A convex function (not shown)
would be the utility of a risk taker. The concave function mathematically suggests that
marginal utility of wealth decreases as wealth increases.

Utility Functions
12

Utility of Wealth

10
8
6
4
3*sqrt(x)

2*sqrt(x)

0
0

10

Wealth

The most important point in this discussion is not the exact calculation of utility, but lies in the
realization that utility of wealth will itself vary, depending upon which mathematical function
7

See, Elton and Gruber, et al (2003).

18

is chosen for the calculation. It is almost universally accepted that the central goal of all
economic activity, including that of investment, is to maximize utility at the investor level. The
exact utility function used to maximize that utility can be very much in dispute, however.
There may be no clear and absolutely correct mathematical equation in which to use. Indeed,
several different utility equations have been experimented over the years in an effort to
appropriately model investor behavior. Varying utility functions will produce varying results
of maximum utility of wealth, depending upon the forms of the equations and assumptions
used.

From Utility to the Beginnings of Portfolio Analysis

In 1952, a graduate student at the University of Chicago, Harry Markowitz, carried many
utility assumptions into his thoughts and used a quadratic utility function to balance, or tradeoff, risk and return. He wrote a small, 14-page paper on the general subject, believing that a
higher return was only possible by taking a higher risk.8 The world of Finance was forever
changed with that one paper.
Markowitz was probably the first person to note the trade-off between risk and return. A few
months after Markowitzs paper was published, A.D. Roy (1952) independently penned an
almost identical treatise, only giving emphasis to a specific risk-return point in which a
threshold or target rate of return was achieved. 9 Roy described this as the safety-first criterion.
Roys work was the earliest known effort on the liability side of investing. But Markowitz
published first, and was a regular contributor to the field of Finance from that point on. Roy
was a British public servant, with expertise during the war years as a gunnery specialist. The
world focused on Markowitz, while Roy became a footnote in history.
Before we turn to the specifics of portfolio theory, it is important to note the utility aspects of
Markowitzs work. Markowitz was clearly attempting to maximize utility when he adopted the
assumptions developed in prior utility writings and used a quadratic utility function, the
general form of which is: u (x) = aw bw2. The quadratic has a unique ability among utility
functions to be stated in direct pricing terms. All other utility functions are solved in utils,
which becomes rather esoteric to everyone concerned, investors and professionals alike. The
quadratic contains both the mean average pricing return and pricing variance in its framework,
in the equation of: 2w = E(W E(W))2, where 2w is the variance of wealth, and E (W) is the
expected wealth. Even more simply, the function reduces to: u (x) = E(r) - 2, with being
the risk-aversion coefficient. Both expected returns (i.e. E(r)) and variance are expressly made
a part of the basic utility function. Thus, the means-variance designation. This wonderful
ability to state utility in pricing terms may have been a chief reason why Markowitz initially
used a quadratic in his calculations instead of other relying on other utility functions.
A normal probability distribution assumption is required with the quadratic, since only the first
two moments of the distribution that of mean and variance exist in means-variance space.
8

Portfolio Selection, Harry Markowitz, Journal of Finance 7, no.1 (March 1952): 77-91.
Safety-First and the Holding of Assets, A.D. Roy, Econometrica, 20 (1952): 431-439. Aside from the liability
aspects of the paper, Roy considered expected return to be the dependent variable and risk to be the independent
term, a nomenclature that was then adopted by the finance profession.
9

19

With a normal distribution, there is no skew and excess kurtosis, so the third and fourth
moments of the distributions are not relevant. Academic research and data collected for these
working papers conclude that annualized pricing returns are distributed approximately normal.
Other types of distributions, such as options, cumulative wealth, and liability streams, are not
normal in their distributions, but are typically lognormal. Utility theorists have criticized the
use of the quadratic for portfolio optimization on these grounds. As Eeckhoudt (2005) states (at
21): in this case, the EU simplifies to a means-variance approach to decision making under
uncertainty. However. it is very hard to believe that preferences among different lotteries be
determined only by the means and variance of these lotteries.
There are other difficulties with a quadratic, too. Most importantly, a quadratic inherently
generates increasing absolute risk aversion (IARA) even though it is commonly felt that
investors are decreasing in the absolute risk aversion (DARA). Utility texts have therefore felt
that the Markowitz framework is a special and somewhat unrealistic case of utility
maximization, limited to those situations in which normality can be assumed and where a clear
preference exists for increasing absolute risk aversion. In the more usual situation where
returns may not be approximately normal, and especially where investor-level absolute or
relative risk is decreasing, other types of utility functions should be used instead. In recent
years, the power series of equations, having the general form of u(w) = w (1 - ) / (1- ), for w
>0, with >= 0, have become the utility functions of choice for many researchers. The power
functions prove very useful, as varying relative risk preferences and varying conditions of
predictability can both be modeled simultaneously.
In spite of all of this, Markowitzs views of utility maximization have become the dominant
method in which to analyze portfolio issues. It is therefore important to thoroughly understand
the current state of portfolio analysis.
A few years after Markowitzs initial paper, James Tobin considered how an investor could
divide funds between safe assets and risky holdings. He suggested that the portfolio problem be
broken down into two stages: the allocation first among, and then, within asset categories.10
The portfolio that should be held by the investor could combine cash with non-cash assets,
resulting in an optimal portfolio that should be independent of individual risk aversion. This
became known as the Separation Theorem. The investment decision is made first to invest in
the market portfolio, and then a financing decision is made that expresses risk tolerances. The
CAPM was then developed in 1964, which established the optimal portfolio as the tangency of
the Capital Market Line and the efficient frontier. With these theories, the foundation of
investment and portfolio analysis was firmly established within the financial world.

10

Tobin (1958), at 85.

20

Modern Portfolio Theory

The Start of Portfolio Theory
Markowtzs efforts in 1952 and 1959 were initially based upon expected utility concepts
developed some years before. Today, means variance optimization is considered to be a special
case of utility maximization, whereby the optimal portfolio comprises the risk and return
parameters that maximize marginal utility. A review of some simple utility concepts and
indifference curves that state the general case of expected utility is therefore in order.11
An investors opportunity set is initially assumed to involve a two period consumption pattern.
The consumption in period 2 equals the income of period 2 plus the amounts saved in period 1
multiplied by the rate of interest on the savings. Or, C2 = I2 + S1*(1+r). This equation is
considered to be the investors budget constraint, and the full opportunity set is sketched out a
straight line between consumption in period 1 on the x axis and consumption in period 2 on the
y axis. The investor can chooses any point along the straight line / opportunity set. The
specific point along the line that is ultimately chosen is dependent upon an investors
indifference curves. The investor is equally happy or indifferent to the choices at all points
along the indifference curves. Each higher indifference curve farther out from the origin point
on the x - y axis is superior to lower curves, as the individuals choices are greater (i.e. happier)
at all points on the higher curves.
The optimal consumption pattern (i.e. optimal marginal utility) occurs at the point of tangency
between the budget constraint / opportunity set and the indifference curves. Greater points of
consumption on higher indifference curves are preferred but not possible due to the budget
constraint of the opportunity set. Lower points of consumption between the two periods are
considered as inferior choices since one has the available budget to always improve
consumption. Extending this solution to practical applications, an investor has the opportunity
to consume in time 1 (i.e. the present), or defer consumption to time period 2 (i.e. the
retirement years), at least up to the budget constraint. The optimal point of consumption is that
combination of current consumption / future consumption that lies on the highest utility curve
within available reach of the investor by way of available resources (i.e. the budget constraint
or opportunity set).
One can now begin to understand how and why MPT is a special case of utility maximization.
At the point of tangency between the Capital Market Line (which is a regressive line
representing the possible risk returns combinations of all capital assets) and the efficient
frontier (a backwards bending curve representing the risk return trade-offs of specific assets),
an investor optimizes his or her risk and return choices with the appropriate allocation of
assets. No other mix of allocations can better the investors choices than this optimal level,
which as we shall see, is the market portfolio available to all investors. The optimal allocation
of assets with specific levels of risk and return generates the maximum amount of happiness or
11

The following introduction on the economic theory of choice is largely taken from Elton, Gruber, et al, (2002).

21

utility for the investor. Even though we may speak in terms of risk and return optimization,
indifference curves intersect the CML (which is the capital markets version of a budget
constraint) and ultimately produce maximum utility at the point of tangency with the CML.
This point along the CML is the tangency with the efficient frontier. Utility is maximized at
that tangency between the CML and the frontier.

The Development of MVO and the Efficient Frontier

At the heart of modern portfolio theory and the special case of utility maximization lies the
ability to balance the returns of assets with the risks of those assets. The technique is referred
to as Means Variance Optimization, abbreviated as MVO. Continuing with the attempt to
maximize utility, a certain optimal combination of return and risk will produce the greatest
possible utility, given the various utility assumptions involved. Lets examine this more
closely. 12
Assumptions of MPT. Modern Portfolio Theory (also known as Markowitz Portfolio Theory)
rests on several assumptions, some of which are summarized as follows: 13 As with all
economic models, the following assumptions are viewed as simplifying in nature. The model
can then be reviewed as various assumptions are relaxed or become more complex.
1. For any given risk level, investors prefer higher returns to lower returns and less risk to
more risk. This is a risk averse assumption, whereby all investors are assumed to be risk
averse to a large degree.
2. Each investment has a probability distribution of expected returns over some holding
period. Investors know the expected or future values of returns, variances, and covariances.
3. The risk of a portfolio can be estimated by the variability of expected returns, expressed as
standard deviation of expected returns.
4. Decisions are based solely on the expected return and risk, with risk being stated in terms
of variance and covariance.
5. A quadratic utility function is used to maximize utility, which can be stated in terms of
expected return and standard deviation.
6. Investors maximize a one-period expected utility.
7. Returns are Identically and Independently Distributed (IID).

12

The following sections on MVO, MPT, and CAPM are not meant to be exhaustive, but rather illustrative of
basic risk and return concepts. The sections are brief summaries of much more extensive texts including Reilly
and Brown (2000); Bodie, Kane, and Marcus (2004); and Elton, Gruber, et al (2003).
13
Many of these assumptions can be found at Reilly & Brown, at 260.

22

8. Returns are normally distributed, or approximately so, and thus, skewness and kurtosis can
be ignored.
9. Investors can borrow and lend at the risk free rate of return.
10. Investors face no taxes or transaction costs.
Historical Data. Most of the theoretical work on portfolio analysis is based upon a statistical
review of historical data. The focus is on average returns and the deviation from the mean
value of the returns. When calculating covariances and other variables in portfolio equations,
ex ante data from historical time periods is used to essentially project into the future. Thus, an
implied assumption of the process is that the past covariance relationship will continue into the
future.
Means Variance Optimization. The following equations are and are based on Portfolio Theory
initially developed fifty years ago by Harry Markowitz (1952 and 1959). Important procedures,
or rules, when developing a two-asset portfolio include:
1. The expected rate of return on the portfolio is a weighted average of the expected returns of
the component assets. E ( rp ) = wb E ( rb ) + ws E ( rs ).
2. Both variances of the assets and covariances are included in the standard quadratic equation
used to calculate portfolio risk. The variance of the portfolio is not simply the sum of the
weighted average variances of the individual assets, however, as is the case with rates of
return. The variance of the rate of return of a two asset port is the sum of the contributions
of the asset variances plus a term involving the correlation coefficient. This principle stems
from the covariance of all assets of a portfolio, expressed as: ab = (Rai ERa) (RB1
ERB) pr1.
2p = w2a 2a + w2b 2b + 2wa wb ab.
and, ab = ab a b.
This equals:
2p = w2a 2a + w2b 2b + 2wa wb ab a b,,
and, ab = ab / ( a b);
where ab is described as rho and is the correlation coefficient, while and ab is the covariance
of the two assets. Portfolio risk thus depends upon variance of the assets, covariance between
assets, and the weight of the assets in the port. As assets are added into a port, the importance
of any one assets risk decreases since the weight of that asset decreases, and the importance of
the covariance between assets increases. When a new asset is added into the port, what matters
most is its average covariance with the other assets in the port.

23

The 1st derivative of the above equation generates the minimum variance on the portfolio,
which is: w* = (22 12) / (21 + 22 2 12). Another minimum variance equation is: w* = +( (b2 4ac)) / 2a.
The Efficient Frontier. The above equations allow for both risk and return to be calculated for
a two-asset portfolio. The entire investment opportunity set is the set of all available portfolio
risk-return combinations. This generates the efficient frontier.

Optimal Frontier - 1 Year Historical Data

0.14

Ave Annl Rtn

0.12
0.10
0.08
0.06
0.04
0.02
0.00

0.02

0.04

0.06

0.08

0.10

0.12

0.14

0.16

0.18

0.20

0.22

Standard Deviation

A central premise of Markowitzs portfolio analysis is the mean-variance criterion, or as

Markowitz referred to it as, the EV rule. This simply states that an investor will prefer higher
returns and lower variance. This is essentially a combination of the risk-averse and more-toless assumptions used in utility analysis, so it makes sense that the criterion would play such a
key role in optimization procedures. The preference for higher returns and lower risk drives the
investor to points along the efficient frontier.
A series of indifference curves could be drawn to show that regardless of where the investor
starts, the highest attainable return for a given risk would be on part of the frontier above the
minimum variance point. For instance, assuming cash was a riskless asset, Tobin used
quadratic utility and indifference curves to generate a tangency with an opportunity locus
line in means-variance space.14 When a set of all-risky assets was assumed, the line then
developed into the now-familiar efficient frontier.15
The frontier can be sketched out as the tangency of the minimum variance point for an infinite
number of optimized portfolios. The global minimum variance point is point established
farthest to the left of the efficient frontier. Thus, all points along the frontier represent the
14

See, Tobin (1958), at 73-80. Tobin concentrated on the implications for macro economic theory which could be
derived from Markowtizs assumption that investors follow rational behavior rules. See Tobin, at 85, n.1.
15
Markowitz first solved the all-risky asset portfolio problem in his paper, Portfolio Selection (1952) and then
further developed his thoughts in: Portfolio Selection: Efficient Diversification of Investment (1959).

24

maximum utility attainable by an investor possessing the two assets in the portfolio. Anything
to the right of the curve is possible to attain, but is inferior to all points along the frontier, as
return can be enhanced without incurring any extra risk (or risk can be lowered without
decreasing return). All points to the left of the frontier are unattainable by the investor, as they
lie beyond the combinations of possible portfolio risk and return for the given asset mix. All
points below the minimum variance point along the frontier are inferior to all points above the
minimum variance, due to return being able to increase without any corresponding increase in
risk. The efficient frontier contains the highest possible return for any given risk level.
Covariance and Correlation. As we know from the review of financial statistics, covariance is
the extent to two or more assets vary in relation to each other. The variation of stocks and
bonds can be seen in the following graph.

Broad-Based Equity and Bond Indices

0.7
0.5

Return

0.3
0.1

-0.1

1935

1945

1955

1965

1975

1985

1995

2005

-0.3
-0.5
Spliced Dom Equity

Spliced Dom Bond

Since covariance produces a number that is difficult to compare to covariances of other assets,
the statistic is translated to correlation, which can then be used to directly compare with other
assets. The mathematical relationship between covariance and correlation is: ab = ab a b.
The correlation coefficient ranges from 1.0 to 1.0.
Two different portfolios with different correlation coefficients will have a different bowing or
curve to the frontier. The smaller the correlation, the greater the risk reduction / diversification
potential (because of negative correlation). If ab = 1.00, then no risk reduction is possible
through diversification, and the frontier will be a straight line between the two point. The
frontier becomes more bowed as correlations decrease, because portfolio standard deviation
decreases for the same level of portfolio return. This bowing of the frontier is sometimes
viewed as a visual portrayal of the effects of diversification. If ab = -1.0, then the frontier will
be so bowed as to terminate at or close to zero standard deviation and the y intercept, with
almost no risk in the port. If asset weights are optimized, then a -1.00 correlation would
generate a zero standard deviation solution and the frontier would bow out to the y intercept. If
the weights are not optimized, then there would be a minimum variance point on the frontier

25

which would approach zero, but not touch the y axis. Varying levels of correlation is shown in
the following graph.

Varying Correlations
0.14

Ave Annl Rtn

0.12
0.10
0.08
0.06
0.04
0.02
0.00

-1.0 Corr.

0.02

0.04

0.06

0.08

0.10

0.12

0.0 Corr.

0.14

0.16

+1.0 Corr.

0.18

0.20

0.22

Standard Deviation

As will be more thoroughly detailed in the section on diversification chapter (paper 2006:11
(b)), pricing risk can be reduced by simply combining assets, especially assets with low
correlations to each other. To illustrate the point, in the stock vs bond graph from 1926 to 2005
(in the above section on MVO), great volatility of equities exists compared to bonds. There
appears to be some general relationship in the pricing movements of both series, however. The
relationship has become more evident in recent years, leading many researchers to believe that
the equity to bond co-movement changes over time. When a correlation equation is used on
equity and bond data for the entire 80-year period, the correlation between the two asset classes
is calculated as .0695. This indicates a mildly positive relationship in the co-movements of the
equities and bonds. When limited to the last 35 years from 1970 to the present, the equity to
bond correlation has increased to .2387. This shows that in recent time periods, equities and
bonds have been moving together more than in historical time periods. A higher correlation in
recent time frames would have the effect of bowing the efficient frontier inward. This degrades
the return to risk ratio, but just as importantly, reduces the diversification effect of combining
assets into one portfolio.
The n Asset Case. The two-asset portfolio equation can be expanded to include three or more
assets in a portfolio. The general equation for n assets of a portfolio is as follows:
2p = Nj=1 w2j 2j + Ni=1 Nj=1 wi wj ij
In a portfolio composed of a large number of securities, the formula reduces to the sum of the
weighted covariances.16 The important factor to consider when adding investments into a
portfolio with numerous assets is not the assets own variance, but the average covariance with
all other assets in the portfolio. For a portfolio of many assets then, the variance reduces to:
16

Reilly & Brown (2000), at 268.

26

2p = Ni=1 Nj=1 wi wj ij
To more fully develop optimality with n number of assets, please see the below
mathematical analysis. LaGrangian equations are typically used for significant amount of
assets in a portfolio.
Equally-Weighted Portfolio Risk. The formula for the variance of an n asset portfolio is
simplified dramatically for equally-weighted portfolios (each w = 1/n):
2p = (1 / n) 2i + ((n - 1) / n) 2xn
where: 2i = average variance of all portfolio assets; and 2xn = average covariance of all
portfolio assets. Each part of the equation is affected by the size of the portfolio. As n
increases, (1 / n) 2i approaches 0 because 1 / n becomes very small. Also, as n increases, ((n 1) / n) 2xn approaches the average covariance because (n 1) / n becomes very small. The
overall effect of this is that as n increases, the variance of an equally-weighted portfolio
approaches the averaged covariance of the portfolio.
With substitution, portfolio variance can be stated as a function of average correlation: 2p =
2i (((1 ) / n) + )). The maximum amount of risk reduction occurs when the number of
stocks is very large, with 2p 2i . This approximation is very convenient, as portfolio risk
is now stated in terms of correlation. If the average correlation = 0.3, for instance, the
maximum risk reduction that is possible is 30% of the individual asset variance. Also note that
as correlation goes down, diversification benefits increase, but it takes more assets in the
portfolio to realize the potential risk reduction. This is because of the ((1 ) / n) + variable.
Two portfolios with the same average correlation will have different portfolio variances,
depending upon the number of portfolio assets.
Thus, adding assets into a portfolio increases the risk / return potential through a reduction in
variance to systematic levels; Further, the efficient frontier (and all the minimum variance
frontiers) depends on expected returns, variances, correlation / covariance of returns, and the
number of portfolio assets.
Thoughts on the CAL / CML. Investors allocate their wealth across both risky and risk-free
assets. A risk-free asset is the security that has a return known ahead of time, so the variance of
the return is zero (in reality, there is some small amount of pricing variance). By adding the
risk-free asset to the investment mix, a very important property emerges: the shape of the
efficient frontier changes from a curve to a line after the inclusion of the risk-free asset. Since
the RFRR variance = 0, portfolio risk merely becomes the 2p = w2risky + 2risky; which is also:
p = wrisky + risky; The resultant risk-return of the portfolio is visually portrayed as the
traveling along the CAL from the frontier towards the RFRR, ending up at the risk-return
parameters corresponding to the weight of the risky asset.
The Capital Asset Line (CAL) is a linear line from the return of the RFRR at the y axis
outward towards the risk-return combinations of various risky assets. The investor will choose
27

the combination of the RFRR and the risky portfolio along the CAL which best suits his or her
own individual risk tolerance levels. The tangency of the CAL and the efficient frontier will be
the highest return-risk ratio for that investor (but without the RFRR then factored in for risk
tolerances). This ratio is essentially the Sharpe ratio, which is also the slope of the CAL. S =
(ERp RFRR) / p. No other portfolio provides a higher expected reward-to-risk ratio than the
tangency portfolio. This portfolio thus becomes the optimal portfolio for the investor, since
portfolio (and the slope of the CAL at the tangency) represents the best possible risk-return
tradeoff attainable given the investor's expectations. Note that each investor will have a unique
CAL, since each investor has different expectations regarding risk, return, variance,
covariances. A CAL derived by a highly optimistic investor, for instance, will have a steeper
slope than a CAL than a more conservative investor.
All points along the CAL which are between the RFRR and the tangency portfolio are
considered lending portfolios, as some money is invested in the RFRR and is lent to the
government treasury. All points beyond the tangency portfolio along the CAL are borrowing
portfolios, since the investor is borrowing money / engaging in leverage to invest in more of
the tangency portfolio. If borrowing at the RFRR is not available, then the CAL ends at the
tangency portfolio, although borrowing at a higher rate than the RFRR (say, commercial paper
rate) would generate a kink in the CAL, with a lower slope beyond the tangency portfolio.
The CAL equation is based essentially on a regression best-fit line of all risky assets and the
RFRR, with the general equation of y = a + bx. More specifically, ERc = RFRR + c. Beta is
the slope of the CAL, which can be calculated through the above Sharpe equation. Bc = (ER T
Rf / T). Then, ERc = RF + (ERT Rf / T) * c With this equation, any variable can be
deduced if other variables are know, For example,
-

the standard deviation associated with a targeted return rate can also be calculated, as can
the return associated with a change in portfolio risk.

The allocation percentage can be deduced from any given RFRR and risky portfolio riskreturn combination. This is done by using a target return or risk level, and then adjusting
the RFRR weighting or allocation until the target is achieved.

The CML. If homogenous investor expectations are assumed, there will then only be one
CAL, effectively making the CAL into a capital market line for all risky assets and of all
investor expectations. CAL becomes the CML. The tangency portfolio of the CAL becomes
the global optimal portfolio, also known as the market portfolio. The slope of the CML is still
beta or the Sharpe ratio, but it is usually referred to as the market price of risk, since the entire
capital market place is now pricing risky assets. B = (ERm RFRR) / m. This equation in
particular, and the concept of beta in general, can be seen as the active price of risk. With
substitution of beta, the general equation is: ERp = RFRR + ((ERm RFRR) / m) * p.
The CML is thus a special case of the CAL, useful when all investor expectations are
considered homogenous. While there are an unlimited number of CALs, one for each
investor, there will only be one CML.

28

The Optimal Portfolio

The best combination of risky and safe assets form a complete portfolio. The optimal portfolio
will be contained at the intersection line of the CML and the efficient frontier, since no other
combination of assets will provide any better combinations of returns and risks to the investor.
The slope of the CML can be calculated at this point. The slope is referred to as the reward to
variability ratio, or Treynor ratio. At the intersection point, one can then calculate the portfolio
optimal risk and return, along with the asset mix of the portfolio.
The CML at the efficient frontier tangency dominates all other lines in that it has the best
risk/return or the largest slope. At this point, the slope = (E(R) - Rf) / s, and [ E(RP) - Rf) / s P
] > [E(RA) - Rf) / sA]. Regardless of risk preferences, combinations of the equilibrium point
P & the RFRR will dominate.
The combination of the two risky assets plus the RFRR will thus generate a dominant portfolio.
Above the tangency between the CML and frontier, the investor is assumed to be borrowing on
margin, since there will now be more than 100% asset investment. Below the tangency, the
investor is said to be lending to the government at the RFRR. Also, the model assumes lending
/ borrowing at the RFRR. For a higher and more realistic borrowing rate, the CML bends to a
lower slope past the point of tangency.
An investor can lower his portfolio pricing risk, but that might not be the optimal level of risk,
since that could be below the optimal point. It can be shown that the most efficient portfolio is
the market portfolio, so a combination of RFRR and the market indexes puts you right on the
CML. All points on the CML are efficient combinations of the tangency point, M, and the
RFRR. The CML equation is:
E (rp) = rf + bx, or expanded:
E (rp) = rf + [ ( E (rm) rf) / m ] p.
Or, the equation can be put into a three asset / port mix, as such:
E (rp) = rf + [ ( E (rp) rf) / xy ] p.
The above expected return equations take the form of y = a + bx, with the RFRR being the y
intercept, b is beta, and x is the is standard deviation of the portfolio. The CMLs slope at the
optimal tangency point can be computed by the Treynor ratio, or the reward to variability ratio
(also known as beta), as being the rise to run, or more precisely as:
= S = (rp rf) / p
The Security Market Line is similar to the CML, only for specific securities. The security
market line (SML) is the relationship between expected return and risk of the investment. Beta
is now measured on the x axis, and the required rate of return is on the y axis. The CAPM
actually depicts the SML, and not the CML, due to the CAPM equation involving the Beta of a

29

single asset. The required rate of return (on the y axis of the SML) is the minimum expected
rate of return necessary to induce an investor to purchase an asset. It is composed of the RFRR
and the risk premium, as is:
ki = rf + B (E(Rm) Rf)
The line can be sketched out in a linear upward fashion, and is actually generated through a
linear regression analysis of all risk / return combinations via the ordinary least squares method
(OLS).
The slope of the SML is the assets beta, and is the required rate of return per unit of risk, or
more mathematically as the change in the expected rate of return for a change in the risk level
of the asset (.i.e y / x), and is also: (ERm RFRR) / . An assets beta estimates the
sensitivity of the asset's rate of return to changes in the broad market's returns. The y intercept
of the line is the RFRR, typically on a nominal basis. The SML can be seen as Rpi = E(Ri)
NRFRR, where Rpi is the Risk premium of asset i and E(Ri) is the expected return of asset i.
In equilibrium, all assets should be on the SML. If the expected rate of return of the market
going forward is not equal to the required rate of return of a security after factoring in beta,
then an asset can be said to be over or under valued relative to the SML. An asset below the
SML is over valued relative to the markets ER, while an asset above the SML is undervalued
with respect to the expected returns of the market.
Financial Market Equilibrium exists when demand and supply forces cause market prices to
stay at levels corresponding to expected/required returns. The CAPM is an example of an
equilibrium model - its predictions result from market forces acting whenever prices get out of
equilibrium to force them back into equilibrium. Because the CAPM assumes investors hold
the market portfolio, unsystematic risk is irrelevant. The risk related to movements in the
market portfolio is all that really matters, so investors pay for only systematic risk. That's why
the independent variable in the SML is beta, not standard deviation.
The SML for all capital assets can shift as a result of change in attitudes of investors over time,
or from a change in the RRFR or nominal risk free rate. Movements along the line would
indicate that the investment itself is undergoing a change in the business characteristics of the
individual investment (such as a change in financial risk, business risk, etc). A change in the
risk premium resulting from a change in economic conditions (such as a changing yield curve)
would change the slope of the SML. It should be noted that the risk premium is not static in
nature, and will change over time, affecting the risk premium given to all investments. A
change in the expected real growth rate of the macro economy, a change in the expected levels
of inflation, or a change in the capital market structure will cause the SML to shift outward or
inward.
Diversification with Many Assets. The above procedures can be used to develop a fully
diversified portfolio.

30

First, the best possible risk-return combinations of risky assets is identified (i.e. the
efficient frontier).

Second, the optimal port of risky assets is determined by finding the port that supports the
steepest CML (i.e. the optimal portfolio).

Third, the port asset mix is chosen based upon risk aversion by mixing the RFRR with the
optimal risky port (i.e. traveling up or down the CML from the point of optimality).

The ultimate aim is to develop the most northwesterly port in terms of risk and return. The
efficient frontier is the set of portfolios that maximizes the excepted returns at each level of
portfolio risk. Expected return risk trade-offs for any individual asset will end up inside the
frontier, and thus, will be inefficient compared with the frontier itself. We can discard any port
below the minimum variance point on the frontier, since those points will decrease returns
while actually increasing risk. There may be constraints in choosing a mix of assets above the
minimum variance point, however (a prohibition on short sales or margin buys, as two
examples).
In addition to diversifying across the various asset classes, it is also important to diversify
inside each asset class. Most MPT studies assume a market index type of investment for each
asset class, so as to be fully diversified among all assets of any one class. By diversifying
among several investments in each asset class, the risk of any single investment (the
diversifiable risk) will be reduced to market or systematic levels. Older studies suggested that
15 to 20 stocks are all that is needed for diversification. Newer studies indicate that individual
stock volatility has risen sharply, but that stock market volatility overall has not increased.
Fifty or more stocks currently may be needed for proper diversification, based upon recent
studies. This diversification can be extended beyond the borders of any one nation.
International levels of diversification is important in that historically there has been a low
correlation between price movements of assets in different countries. This would have the
effect of reducing standard deviation below systematic levels of any one nation, and towards a
world level of systematic risk.
Elton, Gruber, et al (2002) provides the formula for estimation portfolio variance with N assets.
This equation generates the typical diversification graphs with standard deviation on the y axis
and the number of portfolio assets on the x axis:
2p = (1 / N) (2j ~ j k ~) + j k
where, 2j ~ is the average variance of return on individual securities, and j k ~ is the average
covariance of portfolio assets. If there is only one asset in a portfolio, the variance of the
portfolio, of course, will simply be the variance of the asset. As the number of assets (N) in a
portfolio grows, the contribution of the variance of individual securities is steadily reduced
until it approaches zero at systematic risk levels. The contribution of the covariance to the
portfolio increases as N increases until the portfolio variance approaches that of the average
covariance. Thus, the risks of individual securities can be diversified away, but the covariance
of all assets cannot be diversified away, as it essentially is the systematic, market risk itself.

31

Jones, Investments, uses the simplified IID equation of p = i / N

The separation property implies that port choice can be separated into two independent tasks:
first, determination of the optimal risky portfolio; and second, a personal choice of the best mix
of the risky port and the RF asset. Tobin developed the essential properties of the Separation
Theorem in 1958, when he was the first to note that a line could be drawn from the RFRR to a
point of tangency with the efficient frontier.
During the development of portfolio theory, it was ascertained that the data points of the
inefficient ports below the efficient frontier had diversifiable risk in them, and that is why they
were below the frontier. Studies then discovered that there was a limit to how far one could
minimize the standard deviation, and that limit was the market risk. This is shown on the
typical diversification curve between standard deviation on the y-axis, and the number of assets
on the x-axis. The reason why the curve flattens out beyond 20 to 30 or so assets is because
the only type of variance left is market risk. Bill Sharpe wondered why the markets should pay
for diversifiable risks. This questioning led to the development of beta and the CAPM.
With the development of the CML / CAPM line, a startling conclusion was reached that
everyone should hold the same market portfolio. Each person would end up on the CML no
matter where they began. An investor makes financing decisions either to borrow or to lend,
and eventually the investor attains their own preferred risk / return point lying along the CML.
Those preference points would, of course, be separate and distinct from each other, but all of
the preferences would ultimately lie somewhere upon the CML. In theory then (and assuming
efficiency, a timeless risk / return trade-off, and risk and return defined in pricing terms), the
CML cannot be beat as a combination between the RFRR and risky assets.
Once the optimal portfolio is established, individual risk preferences can be identified. The
investor maintains the proportion of stocks to bonds so as to obtain the optimal portfolio, but
then merely adds in the RFRR to the port for a more conservative risk position, or borrows /
margins for a more aggressive portfolio. Essentially, the investor travels up or down the CML
from the point of tangency with the frontier, adding in the RFRR or borrowing funds while still
maintaining the same proportion of stocks to bonds.
On an interesting note, the covariances typically have been easier to calculate and predict than
future macro economic or firm specific events. Thus, diversification between two or more
assets will generate a better indication of future returns and risks than a prediction of future
returns based on macro factors.

Single Factor Asset Market Model

This is referred to as the market model or the Single Index Model. More extensive treatment of
this model is contained in the section on Asset Pricing Models, as the single factor method is
essentially the CAPM. The single index model was developed by William Sharpe, and amounts
to a short-cut of the Markowitz method of calculating a multitude of covariances of stocks. The
basic equation developing the characteristic line is:

32

Ri = ai + i Rm + ei
Where, Ri = return on asset i; ai = return unrelated to to the asset; i beta of asset i; Rm =
return of the market; Ei = residual. Note that ei and Rm are random variables, each with their
own probability distributions having a mean and standard deviation.
On Beta. Much of the single index model rests on the beta estimates. Beta can be estimated by
regression analysis, with the actual stock return forming a scatter around the regression line.
The y axis is the return of the security, Ri, and the x axis is Rp. The greater the variance of ei,
the greater the scatter.
With Beta estimated, alpha can be calculated as: i = Rit Bi Rmt. With Beta also = covariance
of the stock with the market / variance of market, or B = i m / 2 m, it is now a simple matter
to find the covariance. With substitution, Bi = im i m / 2m; and then, Bi = im i / m.
Alphas and betas are estimates of the true values of the stock, so there will be error in the
calculations. Further, both aloha and beta are not stationary over time, so even in the absence
of error, the estimations may not hold in forward time frames.
Studies have shown that past betas of an entire market, or involving numerous equities, will be
more highly correlated to future betas than are past betas of individual equities being correlated
to future betas. In other words, historical market level betas are better predictors of future betas
than are historical betas of individual equities.
Further, more recent historical betas are better at forward predictions than longer historical
periods. The leads to an adjusted beta, whereby there is a continued extrapolation of past betas
to forward estimates. One could also adjust towards an average beta. Adjusted betas do tend to
predict forward betas better than non-adjusted betas. In a review of the reasons for unadjusted
betas to produce more of a forecast error, it was found that error came almost totally from
overestimating high betas and underestimates low betas (Gruber and Elton, at 146).
A first-order autoregressive process can be used to formulate the beta forecast: i, t = 0 + 1
i, t-1 + i, t . i, t is the beta forecast while i, t-1 is the historical beta. The error term has an
expected value = 0. If 1 = 1, then changes in beta are random over time. Beta can be adjusted
to account for its tendency to gravitate to a value of 1 over time. The adjusted beta equation is:
i, t = 0 + 1 i, t-1, with the sum of 0 + 1 = 1. The most popular settings are: 0 = .33; 1 =
0.66. The mean reverting value for any time series variable equals the intercept divided by 1
minus the slope, or 0 / (1 - 1). By assuming 0 = (1 - 1), then 0 / (1 - 1) = (1 - 1) / (1 1) = 1. Thus, the mean reverting level of beta = 1. If the historical beta > 1, then the adjusted
beta will be less than the historical beta and closer to 1. If the historical beta < 1, then the
adjusted beta will be greater than the historical beta and closer to 1. The adjusted beta forecast
will move toward 1 more quickly for larger values of 0, where it is approaching 1.
Betas have also been used to forecast correlation estimates. This is quite useful, since analysts
maybe able to guess at beta, but have a much more difficult time of estimating at correlations.
ij = ij / i j = Bi Bj 2m / i j . Elton and Gruber, at 147 found that historical correlations
33

of individual equities provided the poorest of fits to forward correlations. Gruber suggests that
the correlations of single equities, not captured by the market model, produce noise in
forecasting, with the single-index model (which was developed to mostly to simplify the input
estimations) generating a better forecast fit.
There have also been attempts at determining betas based on firm level fundamentals. There
have been further attempts at combining market level historical beta estimations with
fundamental beta calculations. This may lead to better forecasts of betas, but results in
numerous inputs needed for the calculations.
Beta Instability. The risk-return characteristics of the minimum-variance frontier are
determined by the expected returns, variances, and covariances among individual assets. The
shape and location of the minimum-variance frontier (in risk-return space) will therefore
changes as the risk and return attributes of individual assets change over time. More optimistic
expected return forecasts will cause the curve to shift up, and lowered variance and covariance
forecasts will cause the curve to shift to the left (also with larger curvature or bulge to the left).
Small changes in risk-return attributes of individual assets can cause large changes in the
composition of the portfolios that form the minimum-variance frontier.
The instability in the minimum-variance frontier and, therefore, the efficient frontier, is a
concern for a number of reasons:
-

The statistical inputs (means, variances, covariances} are unknown, and must be forecast;
greater uncertainty in the inputs leads to less reliability in the efficient frontier.

Statistical input forecasts derived from historical sample estimates often change over time,
causing the estimated efficient frontier to change over time (this is called time instability).

Small changes in the statistical inputs can cause large changes in the efficient frontier
(called the "overfitting" problem), resulting in unreasonably large short positions, and
overly frequent rebalancing.

To address the instability problem, the analyst might consider constraining the portfolio
weights (e.g., prohibiting short sales so that all portfolio weights are positive), employing
forecasting models that provide better forecasts than historical estimates, and avoiding
rebalancing until significant changes occur in the efficient frontier.

Critique of MPT
Now that the basic means-variance procedures have been detailed, we have a better ability to
review its strengths and weaknesses.
Markowitz optimization has some great advantages, the largest of which is the theoretical and
mathematical development of optimal asset allocations. Another tremendous benefit of MVO
analysis comes with the realization that broadly diversified market positions should be held at

34

optimality. Index funds have become popular as a result. Risk tolerances at the individual and
institutional levels can even be incorporated into the optimization equations, with conservative
or aggressive positions being easily produced from a mixture of an optimal asset allocation and
the risk free rate of return (or, its reverse, borrowing cash at a RFRR). Means variance
programming thus provides an ideal starting point for portfolio management and allocation.
MVO has limitations, however. Optimal portfolios that are generated using theoretically
correct efficient frontiers may not be overly practical in nature. Heavy shorting and margin
activities will normally occur on some assets, while other asset classes are virtually ignored by
the equations. In many instances, the process should be constrained to arrive at realistic asset
mixes.
Estimation error looms large with optimization methods. Rates of return, standard deviations,
and correlations that are ultimately experienced will inevitably be different (and sometimes,
very different) from either historical data or forward-looking estimated returns and assumed
risk levels. Major estimation errors are likely to occur as the number of asset groups to be
optimized increases. For instance, with a minimum of three separate inputs being made for
each asset of the model, 30 estimations would be needed for a 10-asset portfolio, for each time
frame studied. Multiple holding periods would require hundreds of estimations. The
Appendixes contain the numerous input assumptions used to generate the probability envelopes
contained in later chapters of this book. Any single estimation that is off the mark could
produce statistically significant errors in the final output. This is a major concern with MVO
analysis. Historical data often makes a poor fit for future projections. Forecasting forward
inputs suffers from qualitative beliefs and guesses as to how asset classes should behave in the
future, instead of how they actually will behave. The estimation problem is severely aggravated
in the alternative asset area. Due to a lack of historical data generally before 1990, very little
information even exists to make realistic long-run projections.
Further, MVO can produce unstable solutions, with small changes in inputs leading to overly
large changes in portfolio allocations. This is especially the case where two highly correlated
assets exist within the same portfolio.17 Sensitivity analysis can reduce this possibility, but that
results in even more complicated modeling.
Optimization also assumes a single period model with risk being exclusively defined as
standard deviation of the rate of return. Long-term investors operate across multiple time
frames and normally consider pricing volatility to be somewhat beside the point. Investors
with a long-term view of the world typically feel that risk is either the likelihood of a
fundamental, business type of loss (Buffett has this belief) or the probability of an eventual
funding shortfall (pension funds, institutionals, and retired individuals are acutely concerned
about this), or both.

Other Portfolio Techniques

17

Taming Your Optimizer: A Guide Through the Pitfalls of Means - Variance Optimization, Scott Lummer,
Mark Riepe, and Laurence Siegel, 1994.

35

While means-variance optimization remains a dominant method of portfolio analysis, variants

of the MVO technique or other methods entirely have been developed over the years. 18
Multi-Factor Models.19 The market model is essentially a single factor model, because it
assumes asset returns are explained by a single factor, being the return on the market portfolio.
There may be other activities affecting pricing, however.
A multi-factor model assumes asset returns are driven by more than one factor. The model can
be stated as: Ri = ai + b1I1 + b2 I2 . + bL IL + ci,. For more extensive considerations of multifactor models, see the following section on Asset Pricing Models.
There are three general classifications of multifactor models: (1) macroeconomic factor
models, (2) fundamental factor models, and (3) statistical factor models. Most investment firms
employ macroeconomic factor or fundamental factor models, or mixtures of the two.
Macroeconomic factor models assume that asset returns are explained by surprises or shocks in
macroeconomic risk factors, such as GDP, interest rates, inflation. Factor surprises are defined
as the difference between the realized value of the factor and its consensus predicted value.
The variable for the model are not the value of the macro variable itself, but the unexpected
part of the variable, with the assumption being that the expected part has already been
incorporated in the asset pricing.
The main features of the macroeconomic factor model include the factor sensitivities and the
systematic "priced risk'' factors themselves. The equation is:
Ri = ERi + bii F1 + bi2 Fi2 + i.
The coefficient of each variable tested are referred to as a factor sensitivity, bi. They operate
much the same as betas in the single-factor model. Asset returns are a function of unexpected
surprises to systematic factors, and different assets have different factor sensitivities. Retail
stocks are very sensitive to GDP growth and, hence, have a large sensitivity to the GDP factor.
Other stocks are less sensitive to GDP and have smaller GDP factor sensitivities. The factor
sensitivities of the model can be estimated by regressing historical asset returns on the
corresponding historical macroeconomic factors.
The variables themselves are the priced risk factors, or Fix. A risk that does not affect many
assets (i.e., an firm-level risk), can usually be diversified away in a portfolio and will not be
priced by the market. Investors cannot expect to be rewarded for being exposed to that type of
risk. The factors in macro factor models are systematic in nature, and therefore can affect even
well-diversified portfolios. Since they cannot be avoided, systematic factors represent priced
risk, for which investors can expect compensation. Remember that the variables are the
different between the expected macro data point and the actual data point. Investors are being
compensated for absorbing the surprise of the factor, since that was not priced into the asset
until news was released of the surprise.
18
19

Material for the section if taken from Elton and Gruber, et al, at 161-175
The section on multi-factor models is largely based on CFA Level II readings, circa 2012.

36

Fundamental factor models assume asset returns are explained by the returns from multiple
firm-specific factors (e.g., P/E ratio, market cap, leverage ratio). The same general equation is
used here: Ri = ai + bii F1 + bi2 Fi2 + i.
The sensitivities in most fundamental factor models are not regression slopes, as is the case
with a single factor model. Instead, the fundamental factor sensitivities are standardized
attributes (similar to z-statistics from the standard normal distribution). For example, the
standardized PE sensitivity in a fundamental factor model is calculated as: bi1 = (PEi - PE) /
PE. Also note that by standardizing the factor sensitivity, we measure the number of standard
deviations each sensitivity is from the average. A factor sensitivity of -1.5 is 1 standard
deviations below the mean. This standardization process allows us to use different fundamental
factors measured in different units in the same factor model, for instance, allowing PE ratios to
be used in the same equation as dividend yield percentages.
The factor returns are rates of return associated with each factor (e.g., rate of return difference
between low and high P/E stocks}. The return difference between low and high P /E stocks is
commonly referred to as the return on a "factor mimicking portfolio." In practice, the values of
the fundamental factors are estimated as slopes of cross-sectional regressions in which the
dependent variable is the set of returns for all stocks and the independent variables are the
standardized sensitivities. In fundamental factor models, the factors are not return "surprises."
Hence, the expected factor values are not zero, and the intercept term is no longer interpreted
as the expected return. The y intercept of ai becomes merely the expected amount when x axis
= 0.
Statistical factor models use statistical methods to explain asset returns. Two primary types of
statistical factor models are used: factor analysis and principal component models. In factor
analysis, factors are portfolios that explain covariance in asset returns. In principal component
models, factors are portfolios that explain the variance in asset returns. The major weakness is
that the statistical factors do not lend themselves well to economic interpretation. Therefore,
statistical factors are often seen as mystery factors, with a statistical explanation for asset
returns without full knowledge of the factors affecting the returns.
Comparisons between models. The key differences between the macroeconomic factor model
and the fundamental factor model are:
-

On the sensitivities, the standardized sensitivities in the fundamental factor model are
calculated directly from the attribute (e.g., PE) data, as they are not estimated. This
contrasts with the macroeconomic factor model, in which the sensitivities are regression
slope estimates.

On the factors, the macroeconomic factors (e.g. GDP) are surprises in the macroeconomic
variables. In contrast, the fundamental factors (P/E) are rates of return associated with each
factor and are estimated using multiple regression.

37

On the number of factors, macroeconomic factors represent systematic risk factors, and are
usually small in number. Fundamental factors often are large in number, providing a more
cumbersome, yet more detailed, model of the risk-return relationship for assets.

On the intercept term, the intercept in the macroeconomic factor model equals the stock's
expected return (based on market consensus expectations of the macro factors) from an
equilibrium pricing model like the APT. In contrast, the intercept of a fundamental factor
model with standardized sensitivities has no economic interpretation; it is simply the
regression intercept necessary to make the unsystematic risk of the asset equal to zero.

MVO Assumption Violations. MVO holds for 1) EU maximizers; 2) preference for more-toless; risk averse; normal distributions; and quadratic utility functions. MVO may also hold
when non-quadratic utility functions are used, so long as the quadratic functions are good
approximations of non-quadratic functions. This can occur with stochastic dominance
procedures (see below). Non-normal distributions may also be tolerable where the distribution
size is sufficiently large that the central limit theorem applies.
Excess Return. This is the difference between expected return of a stock or asset and the
RFRR. (Ri Rf) / B. This ratio can then be used to rank all potential assets for inclusion in a
portfolio. A cut-off for asset inclusion in a portfolio would be: C*. This produces a ranking of
assets. All assets greater than C* are included, and all assets below C* are excluded. Ci can in
theory be calculated as: Ci = Bip (Rp Rf) / Bi, where Bip is the expected change in the rate of
return on stock I associated with a 1% change in the return of the optimal portfolio. An
extended equation for Ci is given at Elton, Gruber, at 186.
Once all Cis are ascertained, then the optimal portfolio can be constructed by:
Xi = Zi / Zj,
Where Xi is the percentage of each security. Zi is:
Zi = (Bi / 2ei)((Ri Rf / Bi) C*)
The above method produces the same percentage asset allocation as does quadratic
programming, but has done so with simplified procedures, as the entire frontier and
equilibrium with the CML is not needed in the calculations.
Short sales can be added to the mix. Positive Zi would be a long position while a negative Zi
would be a short. The cut-off, C*, changes in perspective. Everything above C* would be
held long, and everything below C* would be shorted.
If an index is used as a portfolio asset, then Zi simplifies to: Zi = i / 2ei where i = Ri [Rf
+Bi(Rm Rf)]. If the asset has a higher return than the passive index, it should be held long.
Otherwise, it should be shorted.

38

If a constant correlation is accepted as the co-movements between securities, then all securities
can be ranked by excess return. The securitys rank is determined by: (Ri Rf) / i. The
ranking of assets will now use pricing volatility of each asset and not the beta of each asset.
Normally, for single and multi index models, the rankings are done by excess return to beta.
For multi-groups, the cut-off is done by standard deviation.
Geometric mean return. This method maximizes the geo mean return, without regard to utility
or probability distributions of returns. An example would be highest expected value of
terminal wealth many years away. Proponents of this system believe that a geo mean has the
highest probability of reaching objectives in the shortest time; and has the highest probability
of exceeding any given wealth level over any period of time. Critics note that maximizing the
expected value of terminal wealth is not the same as maximizing utility of terminal wealth. The
geo mean equation can be written as: Rg = Ni=1 (1+Rij)Pij 1.0, where Pij is the probability of
the ith outcome of portfolio j; Rij is the ith return of portfolio j; and Ni=1 is the product of the
ith to N outcome.
The geo mean will usually lead to a diversified portfolio, as it penalizes extreme values.
However, it will not necessarily be mean-variant efficient, with the following two exceptions:
Maxing the geo mean will also max the value of a log utility function, or max E ln w. If
returns are normally distributed, a mean-variant portfolio will maximize expected utility for a
log wealth investor. Further, if returns are log-normal, the portfolio in a mean-variant set will
be preferred. Elton and Gruber notes that mean-variance procedures will develop a portfolio
that has the highest geo mean return, and thus feel that this alternative procedure does not
really add much beyond a EU and MVO type of analysis.
Safety First. The emphasis is on limiting downside risk. Roy (1952) put forth the first such
principle of minimizing the probability that Rp < Rl, which is a threshold level. Min P (Rp <
Rl). If returns are normally distributed, the optimal portfolio would be the one where Rl is the
max number of standard deviations away from the mean. The Roy criterion can then be stated
as: min (RL Rp ) / p. This could be reformulated as: max (Rp - RL) / p, which becomes
very similar to Rp Rf / p. All equally desirable portfolio under Roy would have the same
ratio, which could be expressed as a constant: K = (Rp - RL) / p. All points of equal
desirability would plot on a straight line, with K p.= Rp = RL + K p. RL ends up serving the
role as the riskless rate, rf. The line with the highest slope is the most preferred.
Another safety first criterion was developed by Kataoka. Max RL, s.t. P(Rp <Rl) <= , where
is some predetermined probability value (such as .005). If returns are normal, we can analyze
in mean variance space, just as with Roy. With .0005 = , Rl <= Rp 1.65 p. Changing to a
straight line equation produces, Rp = Rl + 1.65 p. The goal is to maximize RL on a line as
high as possible, with Rp on the y-axis and standard deviation on the x-axis. The tangency of
the lines with the frontiers selects the optimal portfolio. Again, very similar to Markowitz.
Another safety-first concept was developed by Telser. He suggested to maximize expected
return. Or, max Rp, s.t. P(Rp <Rl) <= . The equations can again be rearranged. The
constraint could now develop a sub-optimal portfolio, or generate a line above the highest

39

tangential point with the frontiers. By adjusting the constraint, the optimal portfolio can be
achieved as the tangency with the frontier.
The text mentions that while safety-first analysis assumes a normal, a similar result holds for
non-normal distribution, through the use of the Tchebychevs inequality. This allows the
probability to be determined of an outcome less than some value, without regard to a particular
assumed distribution. It is a general statement applicable to all distributions. If we know the
type of distribution, then a more specific probability could result.
Note that if riskless lending and borrowing really exists, Roys criteria leads to infinite
borrowing or only investment in the riskless asset, depending upon RL and Rf. The shortfall
ratios should therefore be constrained, or other lending and borrowing assumptions used.
Stochastic Dominance. This method does not start off with any assumptions of a probability
distribution, and does not assume particular utility functions. General propositions are initially
developed concerning utility preferences. First order stochastic dominance assumes that
investor prefer more to less. First order and MVO both assume more-to-less. Assuming
normality and no short sales, First order leads to a set of portfolios that lie on the frontier above
the minimum variance point, as well as all portfolios that have the highest return possible for
each level of risk (the portfolio set now looks like sort of a semi-circle). When no short sales
and no leveraging are allowed, First order produces the efficient set from minimum variance to
100% equities. Second order leads to only the efficient set, assuming normality. Thus, using
general concepts of utility produces the efficient frontier of MVO, assuming normality.
Second order dominance assumes risk averse investors in addtion to the more-to-less
assumption. First order will only generate a maximum amount of outcome, without regard to
probabilities of outcome. Second order effectively adds the probabilities of the outcomes to the
analysis. Since investors are now risk averse, they must be compensated for the risk.
When the distribution is normal, 1st and 2nd order dominance thus leads to results consistent
with MVO. When returns are non-normal, or where there is no utility function specified,
stochastic dominance can still be used to develop optimal portfolios. This is a definite
advantage over MVO. For any well behaved two-parameter distribution, stochastic dominance
can be used to generate optimal portfolio selection.
Third order dominance then adds DARA to the assumptions, with the 3rd derivative of utility
being positive. This is necessary for DARA to exist. Third order dominance assumes DARA,
which is inconsistent with the quadratic utility function and IARA. DARA utility functions
have positive 3rd derivatives. In cases where two alternative sets of portfolio do not dominate
one another at the 2nd order dominance, then one can utilize 3rd order to ascertain which set is
preferable. Where one set dominates the other in 2nd order dominance, however, using the
third order is not necessary, since it would produce the same result as the superior set with the
2nd order dominance.
Skewness and Portfolio Analysis. A number of articles have added the third moment of the
distribution into the process. Many researchers believe that investors should prefer positive

40

skews, i.e. lognormals, where the portfolios have a larger probability of very large payoffs.
Three-dimensional space would be needed to solve this problem, with mean, variance, and
skew on separate axis. The efficient set would be on the outer shell of the feasible set with
maximum mean, variance, and skew. Skew depends upon the joint movements of the
component assets and thus is not the weighted average of the skew. Portfolio analysis using all
three moments has not been accomplished in the literature.
Downside Risk and VaR. Measuring the downside risk is another way at looking at risk.
Semivariance has been used at times for this. Using lower partial movements as measures of
downside risk is consistent with stochastic dominance. The text then analyzers VaR in terms
of stochastic dominance. Ordering by VaR will be the same as ordering by the 2 nd parameter
of a two-parameter distribution. This could be ideal when skew or asymmetry is present, since
ordering by VaR could be done in cases where variance is not an appropriate measure of risk.
A delta-normal method of VaR uses normality to infer an entire portfolio of derivatives. A fat
tails data set will underestimate the result, however, since a non-normal distribution is in
existence in such an instance.

Mathematical Portfolio Relationships / Constraints / Assumptions

Basic items. The covariance of an asset, ii , on itself is its own variance, 2i. The covariance
of i against j, ij , is also the covariance of j against i, ji . Also, the variance and covariance of
RF assets = 0, because RF assets do not have variance (This is by assumption, since the RFRR
has no default risk and has a stated sum certain in 91 days. In practice, even the RFRR has a
small amount of pricing volatility).
For two-asset portfolios, the variance of a two-asset portfolio (as stated above) is:
2p = w2a 2a + w2b 2b + 2wa wb ab
2p = w2a 2a + w2b 2b + 2wa wb ab a b
Portfolio Returns are:

E[rp] = wi E [ri)

The portfolio beta is:

p = wi i

For a two-asset portfolio, optimality is calculated by:

wb* = [(E (rb) rf) 2s (E (rs) rf ) s b sb] /
[(E (rb) rf ) 2s + (E (rs) rf ) 2b (E (rb) rf + E (rs) rf ) s b sb]
ws* = 1 - wb*
rp* = rs ws* + rB wb*
2p* = w2s* 2s + w2b* 2b + 2 ws* wb* s b sb
Rounding out the remaining pertinent equations, the Capital Market Line slope at optimality is:
rise / run = (rp* rf) / (p* - 0)

41

While the minimum variance point estimate is established by:

wsmv = (2b (sb s b)) / (2s + 2b (2 sb s b))
rpmv = wsmv rs + (1 - wsmv) rb
2pmv = w2smv 2s + (1-wsmv)2 2b + 2 wsmv (1-wsmv) sb s b
The portfolio variance for K assets is:
2p = wi22i + wi wj ij,.
Reilly & Brown (2000) at 268, states that in a portfolio of large number of securities, it can be
shown that the formula reduces to the sum of the weighted covariances. The important factor
to consider when adding investments into a portfolio with numerous assets is not the assets
own variance, but its average covariance with all other assets in the portfolio. For a portfolio
of many assets then, the variance reduces to:
2p = wi wj ij,.
Expanding this:
2p = w1 w111 + w1 w212 + w1 wk1k +
w2 w121 + w2 w222 + w2 wk2k +

wk w1k1 + wk w2k2 + wk wkkk

By folding the equation along the diagonal and multiplying the repeating parts of the equation
by 2, a shortcut is arrived at:
2p = w1 w111 + 2*w1 w212 + 2*w1 wk1k +
+ w2 w222 + 2*w2 wk2k +

+ wk wkkk
Introducing constraints / examining input assumptions. Some of the most common constraints
of MVO are to:
Limit the sum of the weights of the assets to 1.0; limit the weight of each asset to 1.0; and
considering Erp as a constraint, so that we can then solve for variance. This is done by:
ws + wb = 1.0
E (rp) = ws E (rs) + wb E (rb)
1.0 <= ws => 0
1.0 <= wb => 0
More generally, the constraints are:
Ni=1 wi = 1.00,
42

E (rp) = Ni=1 E[ri] wi

1.0 <= wn => 0
Limiting the weights has the effect of creating a portfolio from 100% long in one position to
100% long in another position. Thus, the parts of the frontier represented by long only
positions would be sketched out, not the full frontier.
Borrowing / lending at other than RFRR. Reviewing from above, traveling above the CML and
the point of tangency with the efficient frontier produces margin or borrowing activity at the
RFRR, and then adding that money to the optimal portfolio asset allocations. Traveling down
the CML effectively constitutes lending at the RFRR, with the investor mixing in the RFRR
with the optimal portfolio.
If we change the assumption of borrowing at RFRR to a commercial lending rate, the upper
portion of the CML would kink downward. The procedure would call for raising the RFRR to
the commercial lending rate, and then projecting a line to create a tangency with the frontier.
Thus we would generate one optimal portfolio allocation with the RFRR lending rate, and
another optimal position with the commercial borrowing rate, since the point of tangency
would intersect at slightly different points along the CML. If the lending rate would also be
higher than the RFRR, that would also cause the CML to shift, as the lending rate on the y axis
would now increase. Elton, Gruber, et al, at 106 states that solutions can be found via quadratic
programming. (KCK note: This can also be solved without quadratic programming in a simple
two-asset port, by setting the RFRR at a higher rate, and solving for the tangency of the
frontier using the above equations. See, excel files on data for the working papers).
Not being able to borrow at the RFRR would kink the CML, with the line above the market
portfolio being flatter. Indeed, the CML may become curve, with the upper part of the CML
beyond the market portfolio assuming the shape of the frontier. Thus, the relationship between
ER and may not be linear along the upper part of the CML.

Other constraints. Any constraints that can be established as a linear function can be solved
via substitution (or LaGrangian). The percentage allocated to stocks (max or min); the % of
the portfolio in an industry; and adding in a minimum dividend yield to the portfolio all can be
programmed linearly or through quadratics. (KCK note: Elton & Gruber has a computer
package in conjunction with their 2002 book, available from John Wiley & Sons, that does
much of this math)
Details of correlations (from Elton, Gruber, et al, 2002). If the assets are perfectly correlated,
with a rho of 1.0, the optimal risk and return combinations simply are a linear, straight line
function of the assets. 100% bonds will be connected by 100% stocks, for instance by a
straight line between the two assets on a scatter graph of return (y) and risk (x). There will be
no benefit of diversification, as there will be no reduction in risk from having both assets in the
portfolio.

43

With perfectly negative correlations, we end up with a corner portfolio whereby zero risk is
achieved with some combination of the two or more assets. This shows that the combination of
two risky assets could and will generate less risk than holding either of eth two assets
separately. This is due to the covariances between the assets. With ab = -1.0, pricing swings
of one asset are completely offset by pricing of the other asset, producing absolutely zero
pricing variability when the two assets are combined.
These two extremes of correlations will establish the absolute boundaries of possible frontiers.
Intermediate correlations will generate frontiers somewhere between a straight line, and a
corner portfolio extending to the y axis. For instance, with ab = 0, the frontier will bow out
half way between the corner position of ab = -1.0, and the straight line of ab = +1.0. A
positive correlation between 0 and 1.0 will bow out the frontier between the straight line of
ab = +1.0 and the curve of ab = 0.00. A negative correlation will produce a more pronounced
bow between ab = 0 and a corner portfolio.
Inflation adjustment. Investors with LT time horizons will offer adjust their return and risk
inputs to be stated in real terms, after inflation is accounted for from the nominal historical
returns. Further, the US I bond (available since 1997) may become increasingly used as the
RFRR, stated in real terms (with return data of other asset series then being converted to real
terms for optimization purposes) as well as its own asset class for optimization purposes.
Input estimation uncertainty. Returns and correlations will shift over time. Thus, averaged
annual data using different time periods will generate different results. To resolve these
problems, analysts will typically start with historical data, and then modify the inputs for their
belief on how current and forward events will differ from historical periods. This may correct
for past events not being similar to forward events, but by itself may add to estimation error, as
the projected forward input may be farther off than exclusively relying upon historical data.
Assuming that returns are uncorrelated over time, the standard error of the mean decreases with
the square root of time. The standard error estimate of the mean of a sequence of independent
variables is 2 / T, and the predicted risk is expressed as: 2pred = 2 + 2 / T, where the
equation is solved for the predicted variance, 2 is the monthly returns, and T is the number of
time periods. In lay terms, the standard deviation of average returns falls as the square root of
the length of the holding period. The first part of the equation is the inherent risk of the
investment, while the 2nd part is the predictive variability. Note that the predicted risk will
always be greater than the historical risk, due to uncertainty about the future. Baysian math
describes this as the predictive distribution of returns.
Time Horizon inputs. Many assets are auto-correlated, meaning that their returns are correlated
over time, with returns in one year being a good predictor of returns for the next year. The
returns of T bills are highly auto-correlated, for example. In fact, T bills are mean averting in
nature. Siegel (1998, at 32-33) also notes that equities are mean reverting, but that LT
government bonds are mean averting from predicted levels. Elton, Gruber, et al (2002), at 92,
provides tables for correction of returns for auto-correlations for a 1 year and ten year horizons.
The mean returns are not greatly affected for the correction, while volatility of assets are
slightly affected for bonds, stocks, and T Bills, and in somewhat different directions.

44

Mathematical systems with constraints. Various ways exist to optimize, subject to constraints.
One way is to use LaGrangian equations, as shown below (from Robert Phillips notes; and
2006:1; 2006:2). To develop efficient frontiers subject to the above constraints, LaGrangian
variables can be inserted into the equations.
1 such that p 2 / 1 = w1 + w2 + wi = 1; then,
1 such that p 2 / 1 = w1 + w2 + wi 1 = 0;
2 such that p 2 / 2 = w1 + w2 + wi = rp; then,
2 such that p 2 / 2 = w1 + w2 + wi - rp= 0
3 such that p 2 / 3 = w1 = c1; then, w1 - c1 = 0;

n such that p 2 / n = wn = cn; then, wn - cn = 0;

Expanding the standard deviation equation now yields:
2p = w1 w1 11 +
w2 w1 21 +
+
wi w1 i1 +
1(w1
+
2(r1 w1 +
3 (w1) +
+
n (wn)

w1 w2 12 + w1 wj 1j +
w2 w2 22 + w2 wj 2 j +
wi w2 i2 + wi wj ij +
w2
+ wi 1) +
r 2 w2
+ ri wi rp) +

Partial derivatives are now set to zero and the equations are solved.
p 2 / w1 = 2 w1 11 + 2 w2 12 + 2 wj 1j + 1 + r 1 1 + 3 +n = 0
p 2 / w2 = 2 w2 12 + 2 w2 22 +2 w2 2j + 2 + r 2 2 + 3 +n = 0

p 2 / wi = 2 wi i 1 + 2 w2 i 2 + 2 wi i j + 1 + ri 2 + 3 +n = 0
p 2 / 1 = w1
+ w2
+ wi
=1
p 2 / 2 = r1 w1
+ r2 w2
+ ri wi
= rp
2
p / 3 = w1
+ 0
+ 0
= c1;

p 2 / n = 0
+ 0
+ wn
= cn;
Putting the partial derivative equations into matrix form:
{2 1 1 + 2 1 2 + 2 1 j + 1 + r1 + 1 + 0}
{2 2 1 + 2 2 2 + 2 2 j + 1 + r2 + 0 + 0}

45

{w1}
{w2}

{0 }
{0 }

{.
{2 i 1 + 2 i
{1
+1
{r1
+ r2
{1
+0
{.
{0
+0

}
2 + 2 i j + 1 + ri + 0 + 1}
+1
+ 0 + 0 + 0 +0}
+ ri
+ 0 + 0 + 0 + 0}
+0
+ 0 + 0 + 0 + 0}
}
+1
+ 0 + 0 + 0 + 0}

{}

{}

{wj}
{0 }
{1}
{1 }
*
{2}
= {rp}
{3}
{c1}
{}
{}
{n}
{cn}

The equations are now in the matrix format of: M * X = K. M is inverted to find the solution
to the weighting vector, X. Thus, X = M-1 K. With weights now known, standard deviation of
the portfolio for any given rp can be obtained by: 2p = Ni=1 Nj=1 wi wj ij; and p = 2p.
Another way to solve is by substitution, and then solving as by an unconstrained equation.
Elton, Gruber, et al, 2002, at 100 +, uses this method. (KCK note: LaGrangian as well as
substitution are both analyzed extensively in Klein math econ texts). The constraint (say, the
asset weighting constraint) is substituted into the RFRR equation, and that is substituted into
the equation for the CML slope. Then, partial derivatives are derived from that equation for all
variables. (Elton, Gruber, at 101). From there, optimal allocations can be derived from solving
the simultaneous equations. Once that is known, portfolio return and variance can be
calculated. Short sales can also be part of the equations (Gruber, at 104-111). A general
solution for deriving the efficient frontiers can be found at Gruber, p. 117-122. Quadratic
programming using the Kuhn-Tucker conditions of ascertaining minimas and maximas is at
122-125

46

Asset Pricing Models The CAPM & APT

Introduction
The Capital Asset Pricing Model (CAPM) predicts the relationship between risk and expected
return of an asset. The expected return becomes a benchmark rate of return for evaluating
potential investments. The model can also be used to evaluate equities and assets not yet
coming to the market that are similar to equities and assets with known parameters. Capital
market theory would seemingly rule out arbitrage, since equilibrium prices should not provide
for a riskless return. Thus, most CAPM models will have a no arbitrage assumption. The
Arbitrage Pricing Theory (APT) developed by Stephen Ross delves into this assumption.
In the 1960s, Bill Sharpe realized that returns and risk in the upper left part of the risk return
curve was not possible, due to non-diversifiable risk. Sharpe, Lintner, & Mossin independently
worked towards the development of the CAPM, and all published on the subject between 1963
and 1966. 20 Black also engaged in early theoretical works along similar lines. Today, their
thoughts collectively form the basis of the standard CAPM theory. The CAPM equation,
described below, is a locus of opportunities on the CML line. Beta was developed as a measure
of nondiversifiable risk. At equilibrium, the return to reward ratio, or beta of the market, should
be the same for ALL assets. As E(Rp) increases from some event, it becomes undervalued as to
the CML, demand for the equity increases, the stock price increases, thereby lowering the
E(Rp) back to the CML equilibrium line and the same beta as all other capital assets. Thats
why the CAPM is seen as an equilibrium model.

Assumptions
This model relates the required rate of return of an asset to its risk as measured by beta. There
are a number of simplifying assumptions that ignore real world complexities, and are rather
controversial to say the least. While not practicable, the assumptions lead to insights into the
investment process. Many studies have shown that relaxations in the assumptions do not
greatly affect conclusions of the model. The following items contain the major assumptions of
the CAPM. 21
1. Investors are price takers in the capital markets, without being able to affect the pricing of
assets by their trades.
2. Capital markets are in equilibrium.
20

The articles that initially developed the CAPM were: Capital Assets: A Theory of Market Equilibrium under
Conditions of Risk, William F. Sharpe, Journal of Finance 19, no. 3 (September 1964): 425-442; Security
Prices, Risk and Material Gains from Diversification, John Lintner, Journal of Finance 20, no. 4 (December
1965): 587-615; and Equilibrium in a Capital Asset Market, Jan Mossin, Econometrica 34, no. 4 (October
1966): 768-783.
21
Generally from Dreman, Contrarian Investment Strategies, at 401-402.

47

3. All investors plan for one identical holding period (This assumption ignores time horizons).
4. All investments are infinitely divisible.
5. There is no inflation, or changes in interest rates are fully anticipated.
6. Investors have access to unlimited risk free borrowing or lending at the RFRR.
7. Investors form portfolios from a universe of publicly traded financial assets.
8. Investors do not pay taxes on returns.
9. Investors do not transaction costs.
10. Information is costless and is available to all investors.
11. All investors attempt to construct efficient frontier portfolios, such that they are mean
variance optimizers.
12. All investors analyze assets in the same manner and have the same economic view of the
world. Thus, they end with identical views on the probabilities of future cash flows and
end up with the same opinion on variance, expected returns, correlations, etc. This is called
homogeneous expectations. (Behavioral finance explores heterogeneous expectations).
13. Investments are limited to publicly traded financial assets (This ignores private capital, and
effectively makes the CML a publicly traded capital line, instead of a line having all capital
opportunities in it).
14. Information is costless and is available to all investors.
15. Risk and return is measured in pricing related terms.
16. Another assumption (that is actually part of the efficient frontier assumptions): the market
portfolio is inclusive of all capital in all forms and places, and the S&P 500 or other
broadly diversified equity cap-weighted market index acts as a proxy for all capital.

Details of the Model

With the above assumptions in place, all capital assets carry a risk to reward relationship,
which is generally thought to be linear in nature. As the risk of an asset increases, the return
must be correspondingly higher so as to induce investors to purchase the asset. A regression
line could be drawn through the risk and return combinations of the various assets or asset
classes. The regression equation has the general form of: y = a + b x, with a being the y
intercept, and b being the slope of the line.

48

The regression line is shown in the following graph. Notice that the goodness of fit between
the various asset classes and a linear regression line is 66%. Data is once again from 1926
through 2005, and a one-year holding period is still assumed.

CML - 1 Year Historical Data

0.20

Return

0.15

0.10

0.05

0.00
0.00

y = 0.546x + 0.0187
R2 = 0.6608
0.05

0.10

0.15

0.20

Assets

0.25

0.30

0.35

CML

The regression equation that is illustrated above came to be known as the Capital Market Line
(CML). Stating the CML as an economic model:
Ri = i + i Rm + ei.
Expanding it to an econometric model produces:
E (rp) = rf + [ ( E (rm) rf) / m ] p
The equation can be put into a multi-asset mix, as such:
E (rp) = rf + [ ( E (rp) rf) / xy ] p
The slope of the CML is simply:
rise / run = (rp rf) / (p - 0)
The slope is also considered to be Beta, and is derived from the Sharpe ratio as:
S = (rp rf) / p

49

The CML equation and the slope of the CML can now be placed into optimization equations as
additional constraints to produce a tangency between the CML and the efficient frontier.22 This
is shown in the following graph. The frontier is now shown along with the CML (the dashed
red line). The red dot is the tangency between the CML and the efficient frontier.
Optimal Frontier - 1 Year Historical Data
0.14

Ave Annl Rtn

0.12
0.10
0.08
0.06
0.04
0.02
0.00

0.02

0.04

0.06

0.08

0.10

0.12

0.14

0.16

0.18

0.20

0.22

Standard Deviation

The tangency between the CML and the efficient frontier now becomes the best combination
of risky assets in a portfolio. No other combinations of assets will provide any better return and
risk to the investor. At the tangency point, optimal allocations between assets in the portfolio
can be calculated. For a two-asset portfolio, optimality is calculated by:
wb* = [(E (rb) rf) 2s (E (rs) rf ) s b sb] /
[(E (rb) rf ) 2s + (E (rs) rf ) 2b (E (rb) rf + E (rs) rf ) s b sb]
ws* = 1 - wb*
rp* = rs ws* + rB wb*
2 *
2* 2
p = w s s + w2b* 2b + 2 ws* wb* s b sb
The optimal solution to a single risky asset portfolio can also be stated in terms of aversion to
variance, with:
wx = (Et Rt+1 Rf, t+1) / k 2
where, k represents risk aversion. With multiple risky assets and no risk free asset, the solution
expands to:23
t = (1 / k ) -1t (Et Rt+1 R0, t+1) - -1t 0t
22

The graphs in the text were optimized using LaGrangian variables and multipliers.
The statements of optimality in terms of risk aversion are taken from Campbell and Viceira, Strategic Asset
Allocation (2002), at 20-22.
23

50

In cases where the risky assets have low levels of risk, as with short-term government bonds,
there will be little difference between portfolios composed of all-risky assets and the prior
equation involving one risky asset and a risk free asset.

CML - 1 Year Historical Data

0.20

Return

0.15

0.10

0.05

0.00
0.00

y = 0.546x + 0.0187
R2 = 0.6608
0.05

0.10

0.15

0.20

Assets

0.25

0.30

0.35

CML

The Security Market Line is similar to the CML, only for specific securities. The security
market line (SML) is the relationship between expected return and risk of the investment. Beta
is now measured on the x axis, and the required rate of return is on the y axis. The CAPM
actually depicts the SML, and not the CML, due to the CAPM equation involving the Beta of a
single asset. The required rate of return (on the y axis of the SML) is the minimum expected
rate of return necessary to induce an investor to purchase an asset. It is composed of the RFRR
and the risk premium, as is:
ki = rf + B (E(Rm) Rf)
The line can be sketched out in a linear upward fashion, and is actually generated through a
linear regression analysis of all risk / return combinations via the ordinary least squares method
(OLS).
The slope of the SML is the assets beta, and is the required rate of return per unit of risk, or
more mathematically as the change in the expected rate of return for a change in the risk level
of the asset (.i.e y / x). The y intercept of the line is the RFRR, typically on a nominal
basis. The SML can be seen as Rpi = E(Ri) NRFRR, where Rpi is the Risk premium of asset
i and E(Ri) is the expected return of asset i.
In equilibrium, all assets should be on the SML. If the expected rate of return of the market
going forward is not equal to the required rate of return of a security after factoring in beta,
then an asset can be said to be over or under-valued relative to the SML. An asset below the

51

SML is over-valued relative to the markets ER, while an asset above the SML is undervalued
with respect to the expected returns of the market.
The SML for all capital assets can shift as a result of change in attitudes of investors over time,
or from a change in the RRFR or nominal risk free rate. Movements along the line would
indicate that the investment itself is undergoing a change in the business characteristics of the
individual investment (such as a change in financial risk, business risk, etc). A change in the
risk premium resulting from a change in economic conditions (such as a changing yield curve)
would change the slope of the SML. It should be noted that the risk premium is not static in
nature, and will change over time, affecting the risk premium given to all investments. A
change in the expected real growth rate of the macro economy, a change in the expected levels
of inflation, or a change in the capital market structure will cause the SML to shift outward or
inward.
Important Observations of the Model. With the development of the CML / CAPM, a startling
conclusion was reached: everyone should hold the same market portfolio, instead of different
portfolios along the frontier based upon risk preferences. Each person should end up at the
tangency point between the CML and the frontier, no matter where they began. Maximum
utility exists at only one point along the frontier, at the tangency with the CML. This revelation
not only gave diversification a large theoretical boost, it also impliedly critiqued the
fundamental analytical method pioneered by Graham. The highest investor utility is achieved
by everyone holding the same, identical mix of holdings. Investors are not only rational, as
suggested by utility assumptions, they also are homogenous in holding the same market
portfolio.
The separation property implies that port choice can be separated into two independent tasks:
first, determination of the optimal risky portfolio; and second, a personal choice of the best mix
of the risky port and the RF asset. Tobin developed the essential properties of the Separation
Theorem as early as in 1958. During the development of portfolio theory, it was ascertained
that the data points of the inefficient ports below the efficient frontier had diversifiable risk in
them, and that is why they were below the frontier. Studies then discovered that there was a
limit to how far one could minimize the standard deviation, and that limit was the market risk.
This is shown on the typical diversification curve between standard deviation on the y-axis,
and the number of assets on the x-axis. The reason why the curve flattens out beyond 20 to 30
or so assets is because the only type of variance left is market risk. Bill Sharpe wondered why
the markets should pay for diversifiable risks. This questioning led to the development of beta
and the CAPM.
Once the optimal portfolio is established, individual risk preferences can be identified. The
investor maintains the proportion of stocks to bonds so as to obtain the optimal portfolio, but
then merely adds in the RFRR to the port for a more conservative risk position, or borrows /
margins for a more aggressive portfolio. Essentially, the investor travels up or down the CML
from the point of tangency with the frontier, adding in the RFRR or borrowing funds while still
maintaining the same proportion of stocks to bonds

52

Individual risk preferences can be dealt with by adjusting the portfolio for the risk free asset.
Once the optimal portfolio is established, the investor maintains the proportion of stocks to
bonds so as to obtain the optimal portfolio, but then merely adds the RFRR to the portfolio for
a more conservative risk position, or borrows / margins for a more aggressive portfolio.
Essentially, the investor travels up or down the CML from the point of tangency with the
frontier, adding in the RFRR or borrowing funds while still maintaining the same proportion of
stocks to bonds. The CML cannot be beat by any other combination of the RFRR and risky
assets. The CML is said to dominate all other combinations of risky and risk free assets.
To sort out the various graphs that come into play in the model:
-

The Capital Market Line (CML) will have expected returns on the y axis and the standard
deviation on the x axis. The RFRR will be the y intercept, and will use standard deviation
on the x axis. The CML will intersect the frontiers on the capital market line graph. The
purpose of the CML is to identify optimal portfolios to investors.

The Security Market Line (SML) plots the expected returns on the y axis versus beta on the
x axis. The y intercept will be the RFRR. The purpose of the SML is to describe how
assets are priced by efficient markets at equilibrium. The slope of the SML will be the
assets beta.

The security characteristic line has the excess return of an asset on the y axis and the excess
return of the market on the x axis. The regression line should run through the 0,0 data
point, as abnormal returns, alpha, should be zero. A higher than zero y intercept would
indicate a positive alpha.

The CAPM and the Single-Factor Index Model 24

Two limitations of the CAPM are: it relies on a theoretical capital asset market portfolio that
includes all assets (such as real estate, foreign investments); and it deals with expected returns
as opposed to actual returns. The process can be simplified considerably through the use of the
market model, or single index model. Initially developed by William Sharpe, the single factor
or market model relates returns to a market index, thereby eliminating Markowitzs complex
method of calculating all covariances of all assets in the portfolio.25
The advantages of a single factor model include the reduction in the number of inputs needed
for diversification and the relative ease of identification of individual equities to concentrate
upon. The single index model was an important advance in a pre-computer era of slide-rules
and long hand math, as it made the calculations of the frontiers practical. Some studies show
that the single factor model is close to the Markowitz method in terms of developing efficient
portfolios, and at a greatly reduced level of complexity.

24

Much of this section on single index models summarize important findings in Bodie, Kane, and Marcus (2004),
at 228-239.
25
Sharpes dissertation A Simplified Model of Portfolio Analysis, Bill Sharpe, 1963, first suggested a base factor
for stock prices, which came to be known as beta.

53

The market model is basically a statistical model measuring the firm specific risk versus
systematic risk. This can be plotted as asset return (y axis) versus market return (x axis).
Values of ai, Bi, and 2ei are often obtained through time-series regression analysis, with asset
return forming a scatter around the regression line. The y-axis is the return of the excess
security and the x axis is the excess return of the market. The greater the variance of the
residual term, the greater the scatter. This develops into the security characteristic line.
An index model can be used with realized returns, and not expected returns. Index models use
actual index returns, such as S&P 500 returns. This makes the index model measurable. But
the CAPM uses a theoretical, all asset portfolio, which does not exist in a real world sense. The
index model can at least test out the CAPMs validity through realized gains and then compare
to the theoretical returns that should be generated by the CAPM. The central proposition of the
CAPM is that it is mean variance efficient. The basic equation is: Ri = ai + i Rm + ei. A
realistic form of the CAPM that can be measured is:
Ri = ri rf = + (rm rf) + ei
Where ri is the HPR on the assets realized excess return, and beta compares the realized
excess return of both the asset and the index. The equation sketches onto the security
characteristic line - note that this is different from the SML. ei is the error term, and measures
the firm specific deviation of the realized HPR from the forecasted regression line. The index
model can be developed as a regression line on a scatter gram with excess returns of the asset
on the y axis and excess returns of the market index on the x axis. The security characteristic
line is the regression line of the index market model. Beta of the asset is the slope of the
regression line.
Residuals are differences between actual and predicted stock returns. The residual = actual
predicted return, and is represented by ei. The residual is also referred to as the error term,
since it is the deviations off of the trend-line in the individually plotted points of a scatter
graph. These residuals are estimates of the monthly firm specific component of an assets
return, since they are different from the expected excess returns. The variance of the residual
is 2ei = e2i .
A key assumption of this model is that covariances of stocks is 0, and that there is no influence
or correlations on stocks beyond that of the single factor, which is the market itself. Thus, ei is
uncorrelated to Rm. Mathematically, E (ei, ej) = 0, with the regression being sketched out on a
graph using expected return on the Y axis and Beta on the x axis. Rather than calculating the
covariances between all assets, the asset would only have to be compared with the dominant
base factor in the single factor model. Beta is the variability of the asset as compared with the
variability of the entire market index. Beta is essentially the slope of the slope of the CML,
which we have seen above. Thus, the variance of an asset is dependent upon the uncertainty
common to the entire market (2 2m) as well as the firm specific factors (2 ei). Covariances
are also stated in terms of beta, by: ij = Bi Bj 2m.

54

By assuming one common factor (i.e., the market is responsible for all the asset pricing comovements in the portfolio), the covariances are reduced considerably. With Markowitzs
methods, 100 assets would force a calculation of 4,950 correlation estimates, but with Sharpes
market beta, only 100 estimates of stock to beta correlations would be needed. Thus, the single
factor model was essentially a theoretical short-cut in determining optimality. Beta is the
sensitivity or slope of the securitys returns to the market factor. Beta becomes an estimate of
market (non-diversifiable) risk. Beta is: = xy / 2m
In the single index model, total risk of a single asset has two components: market risk and asset
specific risk. i2 = Bi2 m2 + ei2. All securities should be priced relative to the risk of the
market portfolio. Because residual risk is uncorrelated with the market return, we can diversify
away the residual risk in a portfolio. Hence, for a well-diversified portfolio, the risk of the
portfolio is equal to: 2 = 2m. This implies that the only risk that matters (in a CAPM world)
is market risk.
Firm specific risk is considered to be the error term, since it shows the total deviation off of the
mean. It also represents the diversifiable risk of the asset. The variance of an asset is dependent
upon the uncertainty common to the entire market as well as the firm specific factors. The
return equation can be seen as a regression of a firms return against the market excess return,
which is: Ri = ri + rf . This can also be seen as:
Ri = E(Ri) + M + ei
and then: ri rf = i + i (rm rf) + ei
The portfolio risk can be decomposed by the following. The market risk of (Rm- rf) and firm
specific risk (ei) constitutes total portfolio risk of 2i. ei is also considered to be the error term,
since it shows the total deviations off of the mean. It is also the diversifiable risk of the asset.
Alpha is the y intercept. The regression is of expected returns, not actual ones. So, the return
equations also include an error term, ei, to include the residual or firm specific surprise. The
average security will have a slope of 1.00, since that is the beta of the market. Equities can
have a negative beta, meaning that there is a negative correlation with the stock market returns
as the market goes up, the equity tends to go down.
r2 is the measure of nondiversifiable risk a percentage of total risk. It is the percentage of one
variable being explained by another variable. The lower the residuals, the lower the r2 values.
The r2 is the percentage of the port that is non diversifiable. In terms of the math, r2 = (2 2m)
/ 2i, which is the amount of nondiversifiable risk (2 2m) as a percentage of total variance or
total risk (2i). r2 is also equal to the square of rho, or 2. r2 = 2.
Diversification of a portfolio will eliminate the firm specific risk, while diversification can
only serve to manage or manipulate the average beta of the portfolio that relates to the market
level risk. One can reduce market risk by choosing low beta stocks for the port, but such risk
cannot be eliminated, only managed to a large extent. The term 2 ei can also be stated as Ni=1
2 ei / N. As N increases, portfolio risk approaches zero. Thus, the diversifiable risk is
eliminated from the portfolio, leaving only systematic risk levels of the entire market.

55

An excess return scattergram will typically intersect the y axis at 0,0. Most researchers will
just use a market model based on historical returns without subtracting the RFRR, so that the
scatter plot derives total return with a y intercept typically at the RFRR. Alpha is the amount of
abnormal returns above that of the predicted expected returns, or excess returns. On a scatter
graph showing excess returns, alpha is the amount of the y intercept above zero. With a more
typical plot, alpha would be the amount above the RFRR and the SML. In notation, = ra
E(ra). With beta estimated, alpha can be calculated as: i = Rit Bi Rmt. The Jensen risk
measure considers alpha to be: ri rf = + (rm rf).
Alpha is expected to be zero, because this is a measure of abnormal returns above the risk
premium. If alpha is positive, then abnormal returns exist. If alpha is positive in practice, it
may be because of a particular index not acting as a good proxy, or it could be because the
CAPM is incorrect. The research has shown however that most positive alphas are limited in
nature, revolving around small versus large stocks; companies that announced earnings
surprises; stocks with high book to market values; and stocks experiencing recent and sudden
price declines. Future alphas are almost impossible to predict from past values.
Since the difference between the required and estimated return is alpha, the CAPM allows an
investor to generate a valuation estimate by comparing the expected rate of return with the
required rate of return to determine value. The estimated return is based on EPS and PE
projections (or PDV of FCF), and price projections can be made through fundamental analysis.
Then estimate the rate of return by comparing the current price and the dividend yield. At
equilibrium, the required return = expected return.
In developing an index model, the parameters can be estimated by using historical returns.
Actual returns however will differ from expected returns, thus producing statistical noise and
sampling error.
Historical returns will at least generate excess returns of an index, excess returns of assets, and
residuals on each asset. The Excess returns and residuals can be generated by using the CAPM
equation. A regression can then be done on the excess returns for each stock against the excess
returns from an index, thereby producing the security characteristic line as well as an SML.
Over or under pricing form predicted norms can be visually identified by looking for a positive
(underpriced) or negative (overpriced) alpha on the characteristic line. A CML and efficient
frontier can also be generated from the data.
Systematic risk can be approximated as well by the regression line and beta estimates, while
non-systematic risk can be viewed by the residual variance of the regression. Beta is estimated
using past historical data in many index models. Betas may exhibit RTM, with a high beta
estimate in one period being followed by a low beta estimate in another period. This is so well
known that researchers will use an adjusted beta that includes reversion to the mean. They
do so by weighting the sample estimate of 1.0 by .33, and then weighting the calculated beta by
.66. Beta may change over time, so we cannot even do a time series adjustment with a longer
time period of data. Auto regression models may be able to partially predict stock changes
from their past levels, but the techniques are too new to know if they will prove to be
successful.

56

To sort out the various graphs that come into play:

-

The security characteristic line has the excess return of an asset on the y axis and the excess
return of the market on the x axis. The regression line should run through the 0,0 data
point, as abnormal returns, alpha, should be zero. A higher than zero y intercept would
indicate a positive alpha.

The Security Market Line plots the expected returns on the y axis versus beta on the x axis.
The y intercept will be the RFRR. The purpose of the SML is to describe how assets are
priced by efficient markets at equilibrium. The slope of the SML will be the assets beta.

The Capital Market Line will have expected returns on the y axis and the standard
deviation on the x axis. The RFRR will be the y intercept, and will use standard deviation
on the x axis. The CML will intersect the frontiers on the capital market line graph. The
purpose of the CML is to identify optimal portfolios to investors.

Relaxing the assumptions of the CAPM does not materially affect the overall usefulness of the
CAPM, in many instances.26 Changing the borrowing or lending rate from the risk free rate of
return only has the effect of shifting or partially rotating the Capital Market Line, resulting in a
different point of optimality with the efficient frontier, but otherwise leaving theory intact. Not
using the risk free asset in the calculation of the risk premium will produce a beta of zero, as
long as the market portfolio is mean-variant efficient.27 Adding transaction costs would
increase the range of mispriced assets, because of added costs to remove the market
inefficiency. This would generate a range or band for the Security Market Line (SML), instead
of having a definite line. Heterogeneous expectations of investors would also create a band of
probable CMLs and SMLs, due to the varying expectations. The introduction of taxes into the
model would lower the slope of the frontier and the market line by lowering the reward to risk
ratio.
Other assumptions can also be relaxed without inflicting much damage onto pricing models.28
Short sales and borrowing can be provided for or restricted in the model. Introducing nonmarketable assets into the model would lower the return-risk trade-off, but would otherwise
produce the same basic equilibrium relationship as with a model having only marketable
assets. Providing for non-price taking activities (i.e., collusion, cartels, monopolies) would
decrease the amount of the risk free asset held by the price-affector, thereby lowering the
overall price of risk. Inflation can also be placed into the model, which would increase the
market price of risk. This also suggests that multi-factor models have relevance if the risk of an
asset is related to multiple factors (such as market pricing and inflation).

26

See, Reilly and Brown, at 314-317, for more specific citations to studies that analyzed the relaxation of CAPM
assumptions.
27
One of the more important papers to discuss relaxations in the CAOM assumptions was Fischer Black, Capital
Market Equilibrium with Restricted Borrowing, Journal of Business 45(3), (July, 1972): 444-455.
28
Elton, Gruber, et al, at 310-338, has extensive references to the various items noted in the text.

57

On a cautionary note, if more than one assumption is relaxed at a time, the ability of the CAPM
to generate an equilibrium relationship may be adversely affected.29 Thus, researchers should
be cautious of concluding that the CAPM correctly models economic behavior in a dynamic
setting.
Additionally, relaxing assumptions by restricting leverage of short selling activities may have
practical consequences. If there are restrictions on risk-free borrowing and / or short selling,
investors will hold different portfolios of risky assets, since risk aversion and the response to
risk aversion, would be affected.
-

More risk-averse investors will hold a diversified portfolio consisting of a large number of
securities, but the portfolio will be over-weighted in low-risk/low expected return securities
relative to the allocation in the market portfolio. This is because the ban on short selling
will push the frontier inward, forcing the new market portfolio to lie to the right of the
original efficient frontier.

Less risk-averse investors will hold a portfolio consisting of a small number of high-risk /
high expected return securities along the upper bounds of the frontier (this is from the lack
of RFRR lending, but could also be from investors seeking out higher returns through
higher allocations in response to a short sale ban).

One common use of the CAPM is to evaluate the performance of securities on a riskadjusted basis. If there are restrictions on risk-free borrowing and/or short selling, these risk
adjustments might not be accurate.

The CAPM and Alpha. An excess return scatter gram will typically intersect the y axis at 0,0.
Most researchers will just use a market model based on historical returns without subtracting
the RFRR, so that the scatter plot derives total return with a y intercept typically at the RFRR.
Alpha is the amount of abnormal returns above that of the predicted expected returns, or excess
returns. On a scatter graph showing excess returns, alpha is the amount of the y intercept above
zero. With a more typical plot, alpha would be the amount above the RFRR and the SML. In
notation, = ra E(ra). The Jensen risk measure considers alpha to be:
ri rf = + (rm rf)
Alpha is expected to be zero, because this is a measure of abnormal returns above the risk
premium. If alpha is positive, then abnormal returns exist. If alpha is positive in practice, it
may be because of a particular index not acting as a good proxy, or it could be because the
CAPM is incorrect. The research has shown however that most positive alphas are limited in
nature, revolving around small versus large stocks; companies that announced earnings
surprises; stocks with high book to market values; and stocks experiencing recent and sudden
price declines. Future alphas are almost impossible to predict from past values.

29

Return, Risk, and Arbitrage, Stephen Ross, Friend and Bickster (eds), Risk and Return in Finance
(Cambrdige, Mass: Ballinger), 1977.

58

Since the difference between the required and estimated return is alpha, the CAPM allows an
investor to generates a valuation estimate by comparing the expected rate of return with the
required rate of return to determine value. The estimated return is based on EPS and PE
projections (or PDV of FCF), and price projections can be made through fundamental analysis.
Then estimate the rate of return by comparing the current
Note on International pricing models30. The CAPM has been extended to the international
environment in a model called the extended CAPM. The justification for the extended CAPM
is to provide a common risk pricing framework that corresponds to the domestic CAPM. In the
extended CAPM, the risk-free rate is the investors domestic RFRR, and the market portfolio is
the market capitalization-weighted portfolio of all risky assets in the world. Unfortunately, the
extended CAPM can only be justified with the addition of two additional assumptions, both of
which are rather unrealistic.
-

Investors throughout the world have identical consumption baskets.

Purchasing power parity holds exactly at any point in time.

In a world where these assumptions hold, exchange rate changes would mirror inflation
differences between any two countries, and real exchange rate risk would not exist. Several
empirical studies of international CAPM models have not found much supporting evidence.
For instance, Engel and Rodrigues (NBER Working Paper, 1987) experimented with an
international type of model, finding that asset pricing based on international currency
considerations was not strongly supported as additional conditions or restrictions of the CAPM.
Thomas and Wickens (1993) engaged in a more refined statistical effort by developing ARCH
efforts in a conditional covariance matrix of errors, but still concluded that an internatuional
model yielded unlikely estimates of relative risk aversion with little explanatory power for
explaining rates of return. In spite of the lack of empirical findings, the International CAPM
has been studied along with the above-noted extended CAPM in efforts to explain returns of
international-based assets.
The international CAPM was developed under the assumption that investors are concerned
about returns in their domestic currency. While a U.S. Treasury bill (T-hill) is riskless to a U.S.
investor, it is not riskless to a Swiss investor because of exchange rate risk. The ICAPM
requires that investors consider the real exchange rate: risk associated with investing in foreign
currency assets. In addition to the normal sensitivity of asset returns to changes in the market),
an ICAPM investor must also assess the foreign currency risk premium associated with each
foreign currency. The foreign currency risk premium is the expected exchange rate movement
minus the risk free interest differential between the domestic currency and the foreign
currency: foreign currency risk premium.
For interest rate parity to exist, the forward premium or discount should be equal to the interest
rate differential. If this holds, the hedged domestic currency expected return on an investment
denominated in the foreign currency is equal to the unhedged expected return. There is no risk
premium for the remaining unhedged. If the forward risk currency premium > 0, then the
30

This section is drawn from CFA Level 2 readings, circa 2012.

59

expected unhedged return is greater than the expected hedged return and there is a positive risk
premium for remaining unhedged (and visa versa).
The international version of the CAPM asserts that all investors will hold a combination of: 1)
The risk-free asset of their own country; and 2) The world market portfolio optimally hedged
against currency risk. The world market portfolio (market-capitalization weighted) will be the
same for all investors. Unfortunately, the ICAPM does not provide operational guidelines for
what constitutes an optimal currency hedge.
The ICAPM differs from the domestic CAPM in two ways. First, the relevant risk is world
market risk Second, additional risk premiums are added for the asset's sensitivity to changes in
each foreign currency. While the risk of the world market is conceptually similar to the
domestic CAPM, the sensitivities to the currency risk premiums are unique to the ICAPM. In
the long run, the currency sensitivities would tend towards zero, reducing the ICAPM to a
world market risk. However, in the ST, investors would very much still be exposed to currency
risk.
Major Findings / Implications of the CAPM:
1. At equilibrium, all investors will choose to hold the market portfolio. The market portfolio
contains all securities and the proportion of each security is its market value as a percentage
of total market value.
2. The market portfolio will be on the efficient frontier, and will be at the tangency between
the CML and the frontier. The assumptions lead to all investors having identical efficient
frontiers with the same tangency of the CML. This aggregates to the market portfolio being
the optimal risky portfolio. A passive index type of strategy is efficient, as that will put the
investor on the CML tangency.
3. The risk premium of the market portfolio will be proportional to both the variance of the
market port and average investors risk aversion. The market risk premium is: E (rm) rf .
Beta of the CAPM is now defined as: = i,m / 2m.
4. Beta can also be viewed in terms of risk aversion, with beta being the covariance of the
asset and the market as a percentage of the systematic risk as measured by the variance of
the market. This is expressed as: E(rm) rf = A * 2m. A is a scalar factor representing
the risk aversion of the investor. The risk premium of the market depends on the average
risk aversion of all market participants.
5. The risk premium on individual assets as compared with the market will be equal to the
beta coefficient. This implies a single factor market index model, as described in a prior
chapter. The expected return / beta relationship postulates that the risk premium will be
proportional to beta. The risk premium on an individual security is a function of its
covariance with the market. The essential CAPM equations become:
E (rp) = rf + [ (E (rm) rf) / m ] p, which is:

60

E (rp) = rf + [ E (rm) rf ]
The rate of return of an asset exceeds the RFRR by a risk premium equal to the assets
systematic risk (beta) times the risk premium of the benchmark market portfolio. The
CAPM predicts rates of returns on investments in securities of a firm, and not the firms
rate of return on its plant and equipment.
6. The CML sketches out the risk premiums of portfolios, with expected returns of the market
on the y axis and standard deviation of the market on the x axis.
7. When beta is sketched out on the x axis, instead of standard deviation, and E (r) is on the y
axis, the expected return / beta relationship is a line described as the security market line
(SML). The SML sketches out individual asset risk premiums. The risk premium on
individual securities is a function of the individual securitys contribution to the risk of the
market portfolio, and the individual risk premium is also a function of the covariance of
returns with the assets that make up the market portfolio.
8. At a market equilibrium, fairly priced assets plots exactly on the SML. Under-priced
assets plot above the SML, and overpriced assets plot below the SML. The difference
between the expected returns and the fair price expected returns is known as alpha. This is
the abnormal rate of return is a security in excess of what is predicted by the CAPM. This
under-pricing of an asset with a higher rate of return for its level of pricing risk is viewed
as a disequilibrium in the marketplace, since it is not predicted by the CAPM.

Tests and Critiques of the CAPM

Calculation methodologies are important. The next period beta should be used, but because this
is not available, historical beta is calculated by ordinary least squares estimation (OLS). Past
betas are not really perfect predictors of future betas, as decisions will change over time.
Various studies have shown that as a portfolio, Historical beta is a good predictor for future
beta. There is an 82% correlation between past and future betas with a 5 stock portfolio, and a
91% correlation with a 10 stock portfolio. Several studies show that there is some correlation
between Beta in the next period and financial ratios. The time interval used to calculate beta
will affect the estimation and calculation of Beta. Value Line derives characteristic lines using
rates of return for the recent five-year period. Others have used longer historical data to
generate historical expected systematic risk. Studies indicate that the time interval used in
historical data to develop beta estimates does make a difference in the computation of Beta.
The expected return estimate can vary, depending upon the historical data used (monthly
versus yearly, etc). Most analysts will use the S&P 500 composite index for the expected
return rate, but the theoretically correct market proxy would use all of the asset classes, and not
just equities.
One researcher showed that the errors on the use of the S&P market index as a proxy for
market portfolios was very serious, and generated a benchmark error (Roll study). Roll was

61

critical of the assumption that the market index was a proxy for all risky asset, and felt that the
CAPM could not be tested because a true market portfolio can never be assembled. 31 A more
recent study concluded that simply changing market proxies from a US large cap index to a
global, all equity index would affect the beta risk measure as well as the position and slope of
the SML.32
Additionally, empirical studies show a different y intercept and slope than predicted. Several
studies have even shown that beta over 1 year time periods for individual stocks is not stable.
Several studies have shown a varying relationship between risk and return, instead of a direct
relationship as is assumed under the CAPM. The intercept is higher than zero (zero is predicted
in the CAPM), and linear studies show that the slope of the CML will vary by time period.
Joint effects may also be involved. Most cross-section studies only involve single-period
CAPM models, which can be unrealistic depictions of actual capital markets. When multifactor
pricing models with several risk factors were used, differences in size-related returns can be
explained by more complete measures of risk. 33
Also, not accounting for information costs may be biasing the results. The direction of the bias
from ignoring such information processing is consistent with evidence on anomalies. Prices
can only be expected to adjust within limits defined by the cost of trading. 34 Further, the price
quote spread may be causing some of the seasonal abnormalities, such as the January effect. 35
Valuation Studies. Aside from the calculation methods used, there is a growing body of
empirical literature contradicting CAPM and the usefulness of beta. Researchers have
examined the issue for many years before the 1992 French and Fama study, with several
studies showing that price to value indicators and financial leverage ratios generated higher
than anticipated risk-adjusted rates of return. Some of the valuation studies were conducted as
early as the 1950s, showing that value strategies could work and that under-valuation of assets
persisted over time. Many of these studies were dismissed or relatively ignored by efficiency
advocates.
Important studies by Black, Jensen, and Scholes in 1972 as well as Fama and Macbeth in 1973
only partially supported the CAPM. The reward for beta risk was less than predicted. Both
size and price to value studies have shown market values to be far different than what is
anticipated by Beta. Then, the 1992 French and Fama showed little difference between returns
31

See, A Critique of the Asset Pricing Theorys Tests, Richard Roll, Journal of Financial Economics 4, no. 4
(March 1977), 129-176.
32
The Benchmark Error Problem with Global Capital Markets, Frank K. Reilly and Rashid A. Akhtar, Journal
of Portfolio Management 22, no. 1 (Fall 1995): 33-52.
33
See, Marc Reinganum, Abnormal Returns in Small Firm portfolios, Financial Analysts Journal, 37, no. 2
(March-April, 1981): 52-57; and K.C. Chan, Nai-fu Chen, and David Hsich, An Exploratory Investigation of the
Firm Size Effect, Journal of Financial Economics, 14, no. 3 (September 1985): 451-471.
34
See, Ray Ball, The Theory of Stock Market Efficiency: Accomplishments and Limitations, contained in The
Revolution in Corporate Finance, Joel M. Stern, and David H. Chew, Jr, Blackwell Publ. 2003; M. Jensen, Some
Anomalous Evidence Regarding Market Efficiency, Journal of Financial Economics 6 (1978): 96.
35
See D.B. Keim, Trading Patterns, Bid-ask Spreads and Estimated Security Returns: The Case of Calendar
Stocks at Calendar Turning Points, Journal of Financial Economics 25 (1989): 75-98.

62

in various portfolio having varying amounts of beta. Instead, the F&F study showed that
returns are correlated to size of the company, the firms earnings yield (E/P), and book to
market value indicators. The 1992 study concluded that alternative measures of risk would be
more closely aligned with returns than beta. 36
Price to valuation indicators have been attributed to higher risk, 37 although to value investors
the argument simply stretches credulity.38 If the value strategy was fundamentally riskier
with pricing volatility and beta not capturing all of the risk factors in a stock, then the equity
should underperform during undesirable periods when marginal utility is high. Data collected
in two separate studies have indicated otherwise, however. 39 Some theoreticians argue that the
size effect is related to liquidity and higher returns are necessary for the small size; that the
Book to market effect is due to high risk; and French and Fama in a 1995 study showed that
these effects are related to risk and not market inefficiency. Other studies using a different
methodology than French and Fama have showed that beta is still predictive. Further, a
portfolio of stocks as well as individual stocks over longer time frames may have a stable beta,
however, according to some research. Some studies have also shown that the error introduced
by using the equity markets as a proxy for an unobserved market portfolio may be small, in
contradiction of Rolls claims.
Uses for the CAPM. Most researchers acknowledge that the CAPM does not work perfectly.
While the CAPM has failed rigorous tests on its validity, the results generally do show that
returns are a linear function of non-diversifiable risk, as suggested by the CAPM. Most
observers regard the CAPM as the best effort to date on how prices are priced at equilibrium,
and that the CAPM still provides a useful framework for thinking through issues of risk and
return. Despite problems in the CAPM, it is still the most heavily used analytical model for the
pricing of capital assets. Its uses included capital budgeting, for calculation of the cost of
equity; valuation models, with a projection for the discount rate; and as a benchmark, to
estimate the expected returns so as to measure alpha and abnormal returns.

The Arbitrage Pricing Theory (APT)

Stephen Ross first developed the APT in 1976. Ross calculated relations between return that
would rule out riskless profits. The measure of nondiversifiable risk comes from sensitivity
towards multiple factors instead of the CAPMs use of a one-factor market portfolio. The
premise is that investors take advantage of arbitrage opportunities and will take a port with
36

The Cross-Section of Expected Stock Returns, Eugene Fama and Kenneth French, Journal of Finance 67,
1992, p.427-465.
37
French and Fama in Multifactor Explanations of Asset Pricing Anomalies, Journal of Finance 51, no. 1
(March 1996): 55-84 believed that value outperformed because of higher business risk factors inherent in
companies exhibiting price to value indicators. Bodie, Kane, and Marcus (at 272-273) argued that reversion to the
mean tendencies in asset classes and price to value factors might only be evidence of risk premiums changing over
time.
38
Value and Growth Investing: Review and Update, Louis K.C. Chan & Josef Lakonishok, Financial Analysts
Journal, Jan / Feb. 2004, pp. 71-84.
39
See, Chan and Lakonishok (2004); and Josef Lakonishok, Andre Shleifler, & Robert Vishny, Contrarian
Investment, Extrapolation, and Risk, Journal of Finance 49, no.5 (December, 1994): 1541-1578.

63

higher returns for the same risk as another port. This then pushes expected returns back to
equilibrium. Unlike the CAPM, the APT does not require that one of the risk factors is the
market portfolio. This is a major advantage of the arbitrage pricing model.
Variance in returns comes from firm-specific and market wide unanticipated changes in a
number of economic variables. APT thus allows for multiple sources of market-wide risk.
Returns are decomposed using various economic factors. Return is the weighted average of
anticipated returns and the weighted average of the multi-factor betas. Investors will take a
higher return for any given risk until prices of assets move in both ports to the same return for
the same risk.
CAPM is sometimes considered a more specialized case of the APT in that only one
underlying risk factor exists with the CAPM, that of the market index factor, Bi, whereas any
number of factors may exist with the APT. Far fewer assumptions exist with the APT,
including:
1) Markets are perfectly competitive.
2) Investors always prefer more wealth to less wealth with certainty.
3) The asset return process can be represented by a k factor model.
The APT uses statistical modeling that produces an expected return beta relationship
identical to the CAPM. The single factor APT is the same as the CAPM:
Ri = i + i Rm + ei
Rm is treated as the single factor. A well diversified portfolio will have zero firm specific risk,
and thus we can eliminate the residuals, ei.
Rp = p + p Rm
The portfolio is still risky because the excess return on the index, Rm, is random. In the above
equations, if beta is zero, then Rp = p, implying a excess return of p. This also implies a
zero p, since otherwise an immediate arbitrage situation would be set up, leading to profit
taking and the resulting reduction in market pricing of the portfolio, with a lowering of the y
intercept to zero. Thus, the alpha beyond the excess return should be zero in well functioning
markets. We can even state that alpha should be zero even where beta is not zero. This is
because two non-zero beta ports could be combined into one zero beta port, producing an
immediate arbitrage situation with the combined port, which in turn, would generate huge
trading volume in the stocks of the port until prices adjust wherein alpha once again return to
zero. Therefore, the only value of alpha that rules out arbitrage is zero. The same return to
beta relationship as the CAPM is thus derived without using the elusive market portfolio of the
CAPM.

64

The APT applies to only well diversified portfolios. There may still be arbitrage situations in
individual assets. However, if too many equities were mispriced, then a diversified port could
take advantage of that, ultimately resulting in the repricing of assets to an equilibrium. Even
though the APT applies only to diversified ports, the theory can be used to produce equilibrium
pricing in individual assets. The APT serves the same functions as the CAPM: it provides a
benchmark for fair rates of return and it highlights the differences between systematic and
nonsystematic risk in the form of a risk premium. Since most empirical studies use a
diversified portfolio and test for relationships between pricing and variables other than beta
(such as price to value indicators), researchers may be effectively testing per APT assumptions
and not through the single-factor CAPM procedures.
The APT assumes there are no market imperfections preventing investors from exploiting
arbitrage opportunities. As a result, extreme long and short positions are permitted and
mispricing will disappear immediately. Existing asset portfolios can be combined to produce
the same beta as another portfolio, but having lower expected return. Arbitrage would then
exist between a long position on the combined portfolio and a short on the other portfolio.
This would represent a riskless gain, as the same beta with a higher return would generate an
excess return.
A two factor APT model is:
Ri = i + i1 Rm1 + i2 Rm2 + ei
Where, two systematic risk factors are present. The equation can be expanded for an infinite
number of factors, with n factors being contemplated. Multiple sources of risk will still
generate a market equilibrium whereby arbitrage situations are eventually ruled out. The result
will be a security market equation and multidimensional SML that is identical to a multifactored APT. The n factor equation is: Ri = i + Bi1 f1 + Bi2 f2 + Bn fn + ei
The expected value of each factor, f, is zero. Thus, the fs are measuring the deviation of each
factor to its expected value. The APT is not critically dependent upon a market portfolio. The
APT is based on the law of one price, which holds that two identical assets cannot sell at two
prices in a stable, informationally efficient market place. APT believes that equilibrium market
prices will eliminate any arbitrage of pricing that will yield a risk free profit (or arbitrage).
In the above equations, if beta is zero, then Rp = p, implying a excess return of p. If beta is
zero, then the portfolio return will be equal to the excess return. This also implies a zero excess
return of p, since otherwise an immediate arbitrage situation would be set up, leading to profit
taking and the resulting reduction in market pricing of the portfolio. Thus, the alpha beyond the
excess return should be zero in well-functioning markets. We can even state that alpha should
be zero where beta is not zero. This is because two non-zero beta portfolios could be combined
into one zero beta portfolio, producing an immediate arbitrage situation with the combined
portfolio. This in turn would generate huge trading volume in the stocks of the portfolio until
prices adjust and alpha once again return to zero. Therefore, the only value of alpha that rules
out arbitrage is zero. The same return to beta relationship as the CAPM is ultimately derived
without using the elusive market portfolio of the CAPM.

65

The APT applies to only well-diversified portfolios. There may still be arbitrage situations in
individual assets. If too many equities were mispriced however, then a diversified portfolio
could take advantage of that, resulting in the repricing of assets to an equilibrium. Even though
the APT applies only to diversified ports, the theory can be used to produce equilibrium pricing
in individual assets. The APT serves the same functions as the CAPM: it provides a
benchmark for fair rates of return and it highlights the differences between systematic and
nonsystematic risk in the form of a risk premium.
The chief difficulty with the APT is that the factors are not well developed, since the APM
uses multiple factors of a generic and unidentified nature. Roll identified five factors, including
changes to expected levels of the market index, inflation, industrial production, the RFRR, and
interest rates.40 The APT identifies and analyzes the effect of a few key factors most affecting
the appreciation of assets over a long-time period. Some studies have supported APT because
it is able to explain different rates of return in some cases with results superior to the CAPM.41
Other studies suggest that APT is not testable, due to the factors not being more identified. 42
Many multi-index models take on the trappings of the APT. In practice, many CAPM-types of
modeling exercises utilizes multiple factors, to at least anticipate the impacts of macro events
upon asset pricing as well as to more fully understand and account for the residual error term,
of the CAPM regression equations. Since most empirical studies use a diversified portfolio and
test for relationships between pricing and variables other than beta (such as price to value
indicators), researchers may be effectively testing per APT assumptions and not through the
single-factor CAPM procedures.
The APT and CAPM Compared. CAPM is sometimes considered a more specialized case of
the APT because only one underlying risk factor exists with the CAPM, that of the market
index factor, beta, whereas any number of factors may exist with the APT.
Both APT and CAPM are models describing equilibrium expected returns for assets. The
CAPM suggests that all investors should hold some combination of the market portfolio and
the risk-free asset. To adjust risk, less risk-averse investors simply hold more of the market
portfolio, and less of the risk-free asset. The risk of the investor's portfolio is determined solely
by the resulting portfolio beta (market risk), so the CAPM assumes there is only one source of
systematic ("priced") risk. In contrast, the APT gives no special role to the market portfolio
(i.e., the APT does not require the existence of the "market portfolio"}, and is, therefore, a far
more flexible model than the CAPM.

40

In addition to Rosss writings on the APT, Richard Roll developed extensive thoughts regarding the APT. The
first of several of Rolls articles was: A Critique of the Asset Pricing Theorys Tests, Richard Roll, Journal of
Financial Economics 4, no. 4 (March 1977), 129-176.
41
See, An Empirical Investigation of the Arbitrage Pricing Theory, Richard Roll and Stephen Ross, The
Journal of Finance 35 (Dec. 1980): 1073-1103; and Some Empirical Tests of Theory of Arbitrage Pricing, Naifu-Chen, Journal of Finance 18, no. 5 (December 1983): 1393-1414.
42
The Arbitrage Pricing Theory: Is It Testable? Jay Shanken, Journal of Finance 37, no. 5 (December 1982):
1129-1140.

66

Also, in contrast to the CAPM, the APT states that asset returns follow a multifactor process,
thereby allowing investors to manage several risk factors, rather than just one. This more
targeted approach to portfolio management may help improve portfolio performance. The
investor can identify and earn returns or hedge risks associated with factors such as recession
risk, interest rate risk, or inflation risk. Thus, the investor's unique circumstances (different
from the average investor) may drive the investor to hold portfolios tilted away from the
market portfolio in order to hedge or speculate on multiple risk factors.
The result of the behavior of the typical investor is that the business cycle, or other macro
factor, can become a priced risk factor, and a substantial risk premium could be earned by
exposure to cyclical stocks, for instance, in the belief that the business cycle is improving. (The
text phrases this in terms of independently wealthy investors).
The APT applies to well diversified portfolios and not necessarily to individual stocks. With
the APT, it is possible for some individual stocks to be mispriced, and not lie on the SML. The
APT is more general in that it develops an expected return and beta relationship without the
assumption of the market portfolio. The APT can be extended to multifactor models, as well.
Both the CAPM and the APT conclude with a theoretical explanation of market equilibrium by
different avenues.
With the CAPM, small incremental changes in a portfolio by large numbers of people will
generate market pricing equilibrium, whereas with the APT, even a few investors can serve to
produce market equilibrium through massive buys and sells of arbitrage situations.

Other Models
In addition to the CAPM and the APT, still other models exist to explain the behavior of
market pricing mechanisms. Such models are not per se pricing related, but are based on more
fundamental notions of risk. Studies show a strong correlation between accounting beta and
market beta, so an emphasis on fundamental models in deriving an anticipated future price for
an asset may be appropriate. Research shows there are various types of risk, including the risk
of financial leverage, the probability of bankruptcy and / or default, and firm specific business
risk. Ultimately, these forms of risk are priced back into a market pricing structure, so by
entertaining notions of fundamental risk, arguably some of the reasons for pricing variability
can be more readily ascertained.
Heterogenousity and Multi-period models.43 Researchers have developed equilibrium models
with non-homogenous assumptions (Lintner, Sharpe, Fama, etc). returns, variances, and
covariances are now complex weighted averages of estimates of different individuals. The
difficulty with the approach lies in the fact that the marginal rate of substitution must be
calculated at the individual level to go further in the model. In turn, the marginal rate was itself
based on equilibrium pricing. Assuming a utility function such that the MRS is not a part of
wealth, Lintner was able to demonstrate that the CAPM holds, with all expected values,
variances, and covariances being complicated averages of the probability risks and beliefs of
43

From Elton, Gruber, et al, at 324.

67

individuals. The utility function used exhibited CARA, with a negative exponential function of
u(w) = e-a w. The measure of risk aversion is a. Restrictions on testable assumptions can also be
placed on the model. Gonedes assumed that a set of basic economic activities exists, and that
heterogenousity occurs because of a disagreement of about the exact combination of the basic
economic conditions. With this assumption, the market portfolio is still used by all investors.
Multi-period models have been attempted. The investor is assumed to maximize his utility over
his lifetime consumption. The multiperiod investment decision can still be reduced to the
maximization of a one period model assuming: 1) consumers tastes for goods and services are
independent of future events; 2) consumers act as if the prices and consumption opportunities
of the goods are known at the beginning of the period; 3) the consumer acts as if the
distribution of one-period returns are known at the beginning of the period. Fama also showed
that if the multiperiod utility function exhibits more-to-less as well as risk aversion, then the
derived one-period utility has the same properties as that periods consumption. Essentially, if
use the normal CAPM assumptions, then investors with a multiperiod horizon would still use
the standard CAPM. The zero-period CAPM may also be appropriate for multi=period
investors. The particular single-period model that results depends on the additional
assumptions being made.
Three models are noted in Elton and Gruber. The consumption-oriented CAPM uses a lifetime
consumption maximization assumption. Returns on assets should be linearly related to the
growth rate in aggregate consumption if the parameters of the linear relationship are constant
over time. This is analogous to the simple form of CAPM, with the per capita growth rate in
consumption replacing the rate of return on the market portfolio as the influence on the time
series of returns and therefore, equilibrium returns. The model is:
Rit = i + i Ci + eit
Where, C is the growth rate in per capita consumption over time, and is the market price of
consumption risk. E(eit) = 0 by assumption, as is the covariance of eit and Ct = 0. Beta is:
Bi = Cov (Rit, Ct) / Var (Ct)
The equilibrium return for any security is:
Ri = Rz +1 Bi
These equations are similar to the zero beta CAPM, with return of the market portfolio
replaced by the rate of growth in consumption between two points in time. Testing of the
model has the same problems as with the zero beta CAPM, namely that the variable driving
return (in this case per capita consumption) is difficult, due to sampling problems.
Consumption estimates will contain sampling error; statistics are on expenditures, not
consumption; expenditure are reported over a period of time, and not at one point in time; and
monthly expenditures are available only after 1958. Elton, Gruber, (at 354) reviews possible
solutions to these statistical problems. Breeden shows that the CAPM holds when growth in
per capita consumption is replaced with the rate of return on a portfolio of assets that has

68

maximum correlation with the appropriate consumption series. Referred to as a consumption

portfolio, it contains returns from 13 industries, the T-Bill, LT government bonds, LT
corporate bonds, and a junk bond premium. For the period 1929-1982, the correlation between
a consumption portfolio and the CRSP value weighted index was 0.67. Tests with the
consumption portfolio produced mixed results. Average return was linearly related to beta, and
the intercept supported a riskless asset rather than a zero beta CAPM. The market price of risk,
1, was positive and statistically different from zero. Tests of efficiency however of both the
consumption CAPM and the CRSP value weighted index were rejected however.
Another model involves inflation being added to the standard CAPM. Across time, equilibrium
still exists, assuming constant risk aversion and a positive relation between inflation and
market return. The market price of risk is higher than in the standard CAPM, however. Risk
of any asset is not only the covariance with the market, but with the rate of inflation, as well.
If the asset rate of return is positively correlated with the rate of inflation, the standard CAPM
overstates the asset risk.
Merton (1973) developed a generalized inter-temporal CAPM. Referred to as the multi-beta
CAPM, Merton maximizes lifetime consumption when faced with multiple sources of
uncertainty over future prices, labor income, investment opportunities, etc. Investors will form
portfolios to hedge away these risks. Future risk will affect expected returns on assets. The
inflation model described above is a simple form of the multi-period CAPM. The multi-beta
CAPM states that the expected return on any asset is related to the assets sensitivity to a set of
influences.
Ri Rf = Bim (Rm Rf) + Bi1(Ri1 Rf) + Bi2 (Ri2 Rf) .
The model does not expressly state what these additional influences should be, or exactly how
to hedge the risks that are represented in the model. Elton and Gruber states that the general
concept of Mertons model is similar to APT, with its multi-factor sensitivities.
Multi-Factor Models. An emerging body of literature suggests that asset-pricing models using
multiple factors can improve upon the statistical results of the CAPM. Such models take on
trappings of the factor analysis of the APT, as well as developing thoughts consistent with the
inter-temporal pricing model (ICAPM) pioneered by Merton in 1969. The literature began
when Fama and French added size and value factors for independent variables as well as the
market risk premium to calculate the individual asset risk premium of the CAPM.44 Thus, a
single-factor market model evolved into a multi-factor model that incorporated value indicators
into its theoretical construct. This was a stunning development in finance circles, as Fama
seemingly reversed course from his efficiency papers, using the very same theoretical model
that had been used since the 1960s to support passive investment styles. The Fama and
French model is stated as:
E(Rit) Rft = i + bi (E(Rmt) Rft) + si E(SMBt) + hi E(HMLt) + it

44

Fama and French (1992; 1993).

69

The first part of the equation is taken directly form the CAPM. SMB is the expected returns of
the size variable, while HML is the expected returns of the value variable. The model was run
as a time series regression and then corrected for serial correlation. The time series equation
was:
Rit, t-1 Rft, t-1 = 0i + 1i (Rm, t-1 Rft, t-1) + 2i SMBt-1 + 3i HMLt-1 + it, t -1
In tests conducted on the three-factor model, the explanatory power in explaining asset pricing
variation increased over the single-factor model. The y intercept is considered alpha, the
returns in excess of the risk premium. In a means-variant efficient portfolio, alpha should be
zero. When prior return portfolios were used for the dependent asset portfolios, the longer-term
reversionary tendencies of equities were explained by the three-factor model, although shortterm momentum suffered in F tests of portfolio efficiency.45 The initial data was for a period
of 366 months between 1963-1993. Using updated data of 522 months (1963 to 2006), initial
results are noted in the following table.46

Depend.
Size / BKP
Size / BKP

Sort
5x5
5x5

Weighting
Explan.
Value
Mktprem
Value
Mktprem
SML

HML

KCK 63-06
ave abs a ave a2
Ave R2
0.320 0.1387
0.72
0.096
0.0194
0.86

ave abs a
0.286
0.093

FF 63-93
ave a2
0.1140
0.0164

Ave R2
0.77
0.93

Note the vastly improved R2 values and closeness of the intercept, a, to zero. When various
dependent asset portfolios were used for the y dependent variable, the model was superior to
the single-factor CAPM in explanatory power and intercept have little returns in excess of the
factor risk premiums.

45

Results of the three-factor test were made over a series of articles, the more important ones being: Fama and
French (1992, 1993, 1996).
46
The author used the same statistical processes with more extensive data from 1963 to 2006, producing the
summary results listed. See, Kaufhold, 2008:1; 2008:2.

70

Dependent
CFP
CFP
DP
DP
EP
EP
BE/ME
BE/ME
Size (ME)
Size (ME)

Sort
Decile
Decile
Decile
Decile
Decile
Decile
Decile
Decile
Decile
Decile

Weighting
Equal
Equal
Equal
Equal
Equal
Equal
Equal
Equal
Value
Value

Explan.
Mktprem
Mktprem
Mktprem
Mktprem
Mktprem
Mktprem
Mktprem
Mktprem
Mktprem
Mktprem

SMB

HML

SMB

HML

SMB

HML

SMB

HML

SMB

HML

ave abs a
0.4664
0.1902
0.4552
0.1443
0.4211
0.1577
0.4876
0.2471
0.1511
0.0349

KCK 63-06
ave a2
Ave R2
0.2764
0.71
0.0473
0.91
0.2194
0.70
0.0248
0.89
0.2367
0.72
0.0358
0.91
0.3166
0.68
0.0832
0.88
0.0281
0.82
0.0017
0.92

FF 63-93
ave abs a ave a2
0.268
0.1007
0.062
0.0068

0.260
0.051

0.1059
0.0039

Ave R2
0.80
0.93

0.83
0.93

Carhart improved upon the three-factor model by adding a fourth factor for short-term
momentum.47 This corrected the problem with the three-factor model not being able to account
for momentum. The asset pricing equation became:
Rit, t-1Rft, t-1 = i + b1i (Rm, t-1 Rft, t-1) + b2i SMBt-1 + b3i MOMt-1 + it, t -1
Where, MOM now represented the momentum variables. Other independent factors have also
been developed for short and long-term reversion. When prior return asset portfolios are used
as the dependent portfolios, the following results were then generated (from Kaufhold, 2008:2).
Note the improvement in result when prior return portfolios where added to generate a four
factor model based on market premium, size, value, and prior returns.

Depend.
pr1-1
pr1-1
pr1-1
pr12-2
pr12-2
pr12-2
pr60-13
pr60-13
pr60-13

Sort
Decile
Decile
Decile
Decile
Decile
Decile
Decile
Decile
Decile

Weight
Equal
Equal
Equal
Equal
Equal
Equal
Equal
Equal
Equal

Explan. Explan.
Mktprem
Mktprem SMB
Mktprem SMB
Mktprem
Mktprem SMB
Mktprem SMB
Mktprem
Mktprem SMB
Mktprem SMB

Explan.
HML
HML
HML
HML
HML
HML

Explan. ave abs a

0.4263
0.2816
ST-Rev 0.1781
0.4545
0.2986
Mom
0.1801
0.4298
0.1478
LT-Rev 0.1370

KCK 63-06
ave a2 ave R2 ave abs a
0.2983
0.69
0.2229
0.87
0.1046
0.89
0.2750
0.67
0.3370
0.1439
0.85
0.3310
0.0376
0.89
0.2241
0.73
0.2680
0.0313
0.89
0.2680
0.0258
0.90

FF 63-93
ave a2

ave R2

0.1647
0.2097

0.79
0.90

0.0899
0.0114

0.80
0.92

So, it appears that the asset-pricing model may be evolving from a single-factor process
favoring passive investments to a multi-factor function impliedly supporting value styles, with
the market risk premium not representing all influences upon asset pricing. The following
graph depicts a multi-factor investment universe. 48

47

Carhart (1997).
The graph is taken from John H. Cochrane, Portfolio Advice for a Multi-factor World, Economic
Perspectives, Federal Reserve Board of Chicago, 1999.
48

71

Fama and French (1995) believe that such models still represents rationality and efficiency, in
that the value factor is conveying a higher return for a higher risk. Specifically, the risk of
financial distress faced by value-oriented firms may be priced through the value factor.
Arguably, the market risk premium is being supplemented by other economic risks facing
investors. This supposition has evoked a strong response from the value writers. Notably, Chan
and Lakonishok (2004) showed that value assets had lower downside risk or semi-variance in
recessions and market downturns. Many other commentators are not persuaded by the distress
argument either, producing evidence showing that value assets generally have lower pricing
volatility. Additionally, there is no real theoretical reason to believe that somehow the value
factor is not priced into the market risk premium but that all other forms of risk are ultimately
priced into the market. Size and value factors are more likely to be direct evidence of market
inefficiencies, according to this line of reasoning. The debate over efficiency versus behavior
thus continues (see below). Instead of the discussion swirling around the existence of value, the
focus has now shifted to what accounts for the value.
Other criticisms of the multi-factor model have surfaced. The size and value factors are
admittedly ad hoc in nature, with no known fundamental or state variables to explain the
enhanced performance of the multiple factors. Many macro-economic variables have been
recently tested. While still a very tentative conclusion, all three independent factors may be
conveying varying information about macro-level events.
The size variable may also be somewhat illusionary, with research showing that most of the
returns from SMB are attributable to micro-cap equities which are generally too illiquid to

72

invest in.49 Most of the assorted independent variables used with multi-factor models have also
failed rigorous stress testing at the extremes (Durham, 2001). On a post-cost, post-tax basis, the
returns attributable to momentum may not be exploitable, either (Carhart, 1997). Presumably,
the size and value factors may have similar problems of out-performing after costs and taxes
are accounted for. Further, the usefulness of the size and value variables has been declining in
recent tests. The above summary tables show the problem as compared to the 366 month time
period. Other studies also have the same problem. This could be from a number of sources,
including the 2000 era tech bubble pricing now being in the data, as well as the size and value
factors currently being in wide-spread use.
Lastly, from a fundamental perspective, theoretical pricing models using market-level risk
premiums to explain asset pricing may not show much of any significance. Business owners
will typically purchase a firm for its probable future stream of free cash flows. Nothing else
much matters, including market-level risk factors. Indeed, there is no market risk premium
contained within the normal DCV earnings or dividend model. Many studies have shown that
dividend yield or earnings-to-price bears only a small statistical relationship to asset pricing in
short holding periods. Hagstrom (1999), for instance, indicated only a 0.25 correlation between
EPS and asset pricing in one year holding periods. Since many fundamental analysts buy or
sell an asset based upon the PDV of the free cash flow streams, the obvious question becomes:
what accounts for the other 75% of asset pricing variation? Inflationary pressures and other
macro concerns, as well as investor-level exuberance, have been proposed. Hagstrom also
notes that it is only as holding periods lengthen that correlations will increase between earnings
and price. The literature has not focused on the issue in this time-referenced manner however,
instead returning to the CAPM and its extensions.
Overall, multi-factor models are emerging as a supplement to the single-factor CAPM, and
some day may become more dominant in theoretically explaining asset pricing.
Comparisons between APT and Factor Models. 50Multi-factor models draw on the APT for
their theoretical origins. As such, it is interesting to compare APT, which was essentially the
first effort at an alternative to the CAPM with the more recent work on multi-factor models:
-

The APT is a cross-sectional equilibrium pricing model that explains the variation across
asset expected returns during a single time period. The multifactor model is a time-series
regression that explains the variation in returns over time for one asset.

The APT is an equilibrium-pricing model that assumes no arbitrage opportunities. The

macroeconomic multifactor models are ad hoc (i.e., rather than being derived directly from
an equilibrium theory, the factors are identified empirically by looking for macroeconomic
variables that best fit the data).

The intercept term in a macroeconomic factor model is the asset's expected return. The
APT intercept is the risk-free rate.

49
50

Fama and French, Migration, Working Paper, 2007.

This section is drawn from CFA readings, Level II, circa 2012.

73

Accounting Based Models. In to the CAPM and the APT, still other models exist to explain the
behavior of market pricing mechanisms. Such models are not per se pricing related, but are
based on more fundamental notions of risk. Studies show a strong correlation between
accounting beta and market beta, so an emphasis on fundamental models in deriving an
anticipated future price for an asset may be appropriate. Research shows there are various types
of risk, including the risk of financial leverage, the probability of bankruptcy and / or default,
and firm specific business risk. Ultimately, these forms of risk are priced back into a market
pricing structure, so by entertaining notions of fundamental risk, arguably some of the reasons
for pricing variability can be more readily ascertained.
Leverage. Starting with the sales variability of the firm, the operating and financial leverage
are initially estimated, as well as variability to the earnings stream. Earnings variance is
primarily related to the variance of the firms sales. The systematic component of earnings and
sales variance is referred to as the accounting beta. This is a measure of the firms operating
results compared to the general economy. Non-systematic beta is the variability uniquely
attributable to the firm itself. Operating leverage and financial leverage measures the firms
variance of income as a result of variation in sales.
Bankruptcy / Default. There are several models used to predict bankruptcy. These models date
to the 1960s. CF / total liabilities proves to be the best predictor of bankruptcy. Univariate
models show how any one factor would predict bankruptcy, while multivariate models
combine several ratios together to generate the prediction. Altmans Z score is the best known
of the models and gives weighting to various factors (1968 study by Altman). Two revised
models developed for proprietary usage are known as the ZETA model (Altman 1977 and
1981). ZETA is more accurate than the Z model in years 2 through 5 preceding the bankruptcy.
Others have also devised the probability of bankruptcy. CF variables are important with
bankruptcy predictors. One probably should not focus on the bankruptcy event, as that is a
legal condition occurring far after the economic distress of the firm is in place. Instead, an
investor should concentrate on healthy versus sick economic status of the firm, and rate the
overall financial abilities of the business. CFO is a good indicator of financial distress.

74

The Efficient Market Hypothesis

Historical Development of Efficiency and Behavior
An efficient market is said to exist where pricing rapidly adjusts to the arrival of new
information.51 The research began in earnest after Maurice Kendall found no predictable
pattern in stock prices.52 This finding initially upset theorists in that a lack of predictability was
rather unsettling. Gradually, researchers came to believe that this showed that the markets were
behaving efficiently. New information is unpredictable, so price changes should be random in
nature. This developed into thoughts of market efficiency, whereby prices fully reflect all
available information on the equity. Competitive pressures to find value in stocks leads to
efficiency.
An efficient market is said to exist where pricing rapidly adjusts to the arrival of new
information. Paul Samuelson and Eugene Fama wrote interesting articles in the early to mid
1960s that were largely aimed at refuting Technical Analysis. Both authors felt that past
pricing could not be used to project future pricing. Today, the early works of Fama are viewed
as the origin of the weak form of the Efficient Market Hypothesis (EMH). 53 Paul Samuelsons
article on the randomness of pricing returns is considered a classic in the field.54 An old paper
by Louis Bachelier dating back to 1900 that was found at the University of Chicago also lent
support for efficiency arguments, with the mathematical expectation of speculation being
found to be zero. 55
The initial writings concentrated on a random walk whereby the pricing of an asset fully
reflects all available information at any point in time. It was not possible to derive trading
systems or investment strategies beyond the expected basis of the assets risk. The expected
pricing should reflect the risks of the asset in an efficient market. It is therefore impossible to
consistently out-perform the market through abnormal returns due to the markets correctly
incorporating all available past information, so went the argument.
Fama then reviewed the developing research in a 1970 article, classifying the literature into
three sub-sets: the weak form, the semi-string, and the strong. 56 With the weak form, pricing
fully reflects all past and current market level info, including historical data. There is an
implication that the past rate of return should have no bearing on future return rates, and the
51

Much of the material in the section on Market Efficiency is drawn from Reilly and Brown (2000); Bodie, Kane,
and Marcus (2003); and Damodaran (1996).
52
The Analysis of Economic Time Series, Part I. Prices, Maurice Kendall, Journal of Royal Statistical Society
96 (1953), pp. 11-25.
53
The Behavior of Stock Market Prices, Eugene Fama, Journal of Business, January, 1963; and Random
Walks in Stock Market Prices, Eugene Fama, Institutional Investor, 1965.
54
Proof that Properly Anticipated Prices Fluctuate Randomly, Paul Samuelson, Industrial Management Review
6 (Spring 1965).
55
The Theory of Speculation, Louis Bachelier, 1900.
56
Efficient Capital Markets: A Review of Theory and Empirical Work, Eugene F. Fama, Journal of Finance 25,
no. 2 (1970): 383-417.

75

return rate of each asset should be independent of every other stock. With the semi-strong
form, pricing adjusts rapidly to all public information and fully reflects all of that information.
Investors should therefore not be able to derive above average profit from transactions. While
the weak form studies were aimed at discrediting Technical Analysis, the semi-strong group of
studies impliedly refuted Fundamental Analysis. The strong form states that pricing fully
reflects all information from private sources as well as public sources, and suggests that even
insider information is quickly incorporated into an assets pricing structure.
Famas assumptions for an informationally efficient market included: 1) the existence of a
large number of competing buyers and sellers, all attempting to maximize their profits; 2) New
information comes to the market in a random fashion; 3) Investors attempt to adjust pricing to
reflect this new information; 4) All forms of information, even private info, is cost free and
available to everyone at the same time.
By the late 1970s, efficiency was under attack in several directions. Mutual fund managers
were never particularly fond of the EMH to begin with. Many valuation studies had been
conducted over the years, that showed value strategies could work, and that under-valuation of
assets was persistent over time. And, certain market anomalies seemed incapable of
reconciliation through either the CAPM or EMH. Behavioral studies were increasingly
showing non-rational behavior in the everyday lives of individual and professionals alike, and
value oriented studies had been consistently demonstrating that price to value indicators did
exist and were persistent through many decades. David Dreman forcefully advanced the case
of investor psychology in the late 1970s with the publication of book on the topic. An article
in 1979 by two psychologists, Daniel Kahneman and Amos Tversky, was a watershed event,
developing psychological and emotional underpinnings to investor behavior. Predictability of
returns across time had also been demonstrated in several papers, in contradiction of the
identical and independently distributed (IID) assumption of portfolio analysis. Other scholars
were also questioning efficiency arguments. For instance, Robert Shiller provided evidence
that the equity markets were excessively volatile.57 If true, such a finding would be a direct
refutation of the EMH.
Then in 1987, the US domestic equity markets dropped precipitously without any major, abrupt
news announcements, much to the disbelief of efficiency proponents. Indeed, the Brady Task
Force that reported on the 1987 crisis felt that computer trading programs and portfolio
insurance products favored by many theoreticians were causative factors for the market drop.
By the early 1990s, there was mounting disbelief in the Efficient Market Hypothesis (EMH),
especially among financial practitioners. While index funds were continuing to develop as
popular investment vehicles, many professionals believed that the indexes were difficult to
beat, not because of theoretical market efficiency, but because of the more practical realities of
low expense ratios.

57

Do Stock Prices Move Too Much to be Justified by Subsequent Changes in Dividends? Robert J. Shiller,
American Economic Review 71 (1981): 421-436.

76

Then in 1992, French and Fama penned an article that rocked the finance world.58 They
produced statistical evidence showing the validity of style, size, and country characteristics
(see the above discussion on multi-factor models). Price-to-book value indicators consistently
were more powerful in explaining returns than other factors, including beta. In one famous
passage, Fama said that: beta as the sole variable in explaining returns on stocks is dead.
A multifactor risk calculation should be explored, instead.
French and Fama went on to develop a three-factor valuation model in 1993.59 As noted in the
above discussion on the CAPM, Fama initially argued that risk levels of under-valued assets
were inherently higher. In later articles, Fama and French argued that financial distress of
under-valued firms was responsible for positive findings of value indicators. But valueoriented professionals quickly and severely criticized these approaches, noting that both
pricing risk and fundamental notions of risk were lower with most value-styled equities and
assets. Other researchers examined the role of the market model itself, believing that current
modeling efforts have not adequately captured all relevant risks. Most value styled writers and
behavioral thinkers are of the opinion that market inefficiency can and does regularly occur.

Are Markets Efficient?

There are several general thoughts to keep in mind. The magnitude issue a very tiny
inefficiency in the market may result in a large gain to a money manager that can take
advantage of the mispricing, if the fund is large enough. For example, a $5 Million profit
results from a 1% inefficiency for a $5 Billion fund that can buy or sell on the inefficiency.
The selection bias issue the people with knowledge of market beating methods keep them
secret, while those who do not know how to beat the market will publish on any number of
topics. This results in a bias in the literature towards an efficiency argument. The lucky event
issue A superior manager may be nothing more than one who has been lucky in their
investments, and yet they will invariably feel they have found a superior system, at least until
the law of averages kicks in.
Weak Form Tests of Technical Analysis. Various trading rules and systems have been devised
over the years in an attempt to outrun the markets. Some of these attempts go back many
decades. Graham (1934) even noted activities of Chartists before the Depression. A
widespread consensus exists in the literature that technical methods resting solely on past data
and information cannot beat a buy and hold policy.
As to studies involving the weak form of the EMH, tests of trading rules require subjective
interpretation of data, and an almost infinite number of potential trading rules exist. As a
result, only the better-known rules can be examined. Filter rule studies (sell at a 5% loss, etc),
have generally found that excess returns cannot be obtained, especially on a risk-adjusted basis

58

The Cross-Section of Expected Stock Returns, Eugene Fama and Kenneth French, Journal of Finance 67,
1992, p.427-465.
59
Common Risks Factors in the Return of Stocks and Bonds, Eugene F. Fama and Kenneth French, Journal of
Financial Economics 33, no. 1 (February 1993): 3-56.

77

and after trading expenses are considered.60 For example, large filters did not yield returns
above that of the buy and hold strategy.61 A study on the Value-Line timeliness ranking system
showed that portfolios of best performing stocks in the recent past appear to outperform
predictably enough to offer profit possibilities, although other studies suggest otherwise. With
the filter rules, studies have shown that small filters would yield above average profit before
transaction costs, but that after the costs are considered, all trading profit turned to losses.
Large filters did not yield returns above that of the buy and hold strategy. Most of the evidence
on the trading rules shows that trading rules cannot beat the buy and hold strategy.
The biggest risk of market timing may be that investors will not be in the market at critical
periods (Jones, 2004). There is some evidence (OShaughnessy, 1998, at 234-235) that oneyear pricing performance (also known as relative strength) outperforms, but the volatility risk
is extraordinarily high. There is also some recent evidence that a 200 day moving average,
when used with a 1% filter rule, may possibly be used to improve returns (Siegel, 1988, at 250252; Brock, et al, 1992). Jegadeesh and Titman (1993) found that stocks exhibit momentum of
3 to 12 month duration, with good or bad pricing performance continuing in the short-term.
This finding might actually be supportive of behavioral studies suggesting initial overreactions
to news. Several studies have investigated the Value Line Investment Survey Ranking System,
with evidence that it does provide useful information, but that the markets adjust quickly to the
rankings (Jones, 2004, at 334). Carhart (1997) found validity to a momentum factor placed
into Fama and Frenchs multi-factor analysis.
It is doubtful however whether short-term trading patterns can outperform a long-term buy and
hold strategy on a post-tax, post-cost, post risk-adjustment basis. For example, Ready (1997)
has shown that a moving average rule would not produce profits in practice due to trading
expenses and movement in pricing by the time the trader could respond to a technical signal.
Jones (2004) felt the evidence against technical analysis was so overwhelming that the burden
must be on the proponents of technical rules to prove outperformance in properly designed
testing procedures.
Time-Series Studies. Serial correlations studies have generated some very interesting results.
Returns over short horizons of 3 to 6 months show that stocks tend to relate back to past trends.
This is a momentum type of finding, whereby positive returns tend to be followed by further
positive returns. The correlation coefficients of some studies tended to be rather small
however, and unlikely to have enough correlation to generate excess returns. Thus, trading
opportunities in real time may not be so apparent, since the gains may be too small to exploit.
While a consensus exists that short-term pricing movements cannot be predicted, longer period
pricing becomes remarkably predictable. The overestimation of growth possibilities and the
underestimation of value-oriented equities noted in behavioral studies are consistent with
reversion to the mean principles. Shiller (1981) indicated that market prices exhibit excessive
60

Ray Ball, S.P. Kothari, and Charles Wesley, Can We Implement Research on Stock Trading Rules? Journal
of Portfolio Management 21, no. 2 (winter 1995): 54-63; and Hendrick Bessembinder and Kalok Chan, Market
Efficiency and Returns to Technical Analysis, Financial Management 27, no. 2 (summer 1998): 5-17.
61
Eugene Fama and Marshall Blume, Filter Rules and Stock Market Trading Profits, Journal of Business 39,
no. 1 (January 1966 Suppl.): 226-241.

78

volatility compared to intrinsic valuation, implying initial overreaction to news and

information. DeBondt and Thaler (1985, 1987) demonstrated that reversion to the mean
concepts was being experienced at the individual stock level, and not just in the aggregate
markets. Fama and French (1988) found significant serial correlation with five-year return data
(but little correlation in a one-year period), and the effect was more pronounced with the small
cap firms. Poterba and Summers (1988) also found negative long-term serial correlation.
Chopra, Lakonishok, and Ritter (1992) discovered that poorly performing stocks exhibited
strong reversals in future time frames, leading to significant outperformance. Reichenstein &
Dorsett (1995) showed that bad periods of market behavior were predictably followed by good
periods, and visa versa. This suggests that initial overreaction to information is followed by
mean reversion when the market eventually recognizes the overreaction (Haugen, 1995).
Substantial negative correlation exists in longer-term price behavior, in apparent contradiction
to market efficiency (Damodaran, 1996, at 164). Carhart (1997) found that almost all mutual
funds revert to the mean within 5 years of initial year performance measurement, with some
evidence that a few funds will continue to be persistently good or bad in their performance.
Studies indicate that stock predictability exists across time horizons (Barberis 2000). Siegel
(1998, at 12, 33) attributed the stability of real returns on stocks over extraordinarily long
periods (of 195 years) to mean reversion tendencies in equities.
The degree of serial correlation in equity returns is so strong that profit opportunities may be
possible, while other anomalies (such as the calendar effect) are not sufficient in their
correlations to consistently allow abnormal profits. (Bodie, Kane, and Marcus, 2004, at 656).
Thus, initial overreaction to information noted in behavioral studies may be sufficiently large
to generate excess returns from value-oriented buys, with a gradual mean reversion then
occurring until market equilibrium is established between market pricing and intrinsic
valuation levels. This line of reasoning would be consistent with not only mean reversion
studies, but behavioral finance research and the value style, generally.
The findings as to serial correlation are consistent with market overreaction in ST and RTM
correlation in the LT. This suggests that stock prices might overreact to news. Initial
momentum in ST time frames to news will generate a longer-term correction. This tendency to
overshoot followed by a correction may be evidence of pricing revolving around a fair value
with large amounts of volatility.
Some authors (Bodie, Kane, and Marcus, 2004) feel this is actually evidence of efficiency,
with a rational response in market prices to changes in discount rates. One study by French and
Fama (1989) suggests that the predictability in returns is owing from a risk premium rather
than evidence of market inefficiency. These beliefs are highly controversial, and others,
including Lakonishok (1994; 2004), Siegel (1998), and Shiller (2000) strongly believe that
return predictability constitutes direct evidence of market inefficiency, and implied evidence
supporting behavioral issues. Dreman (1999) believe RTM in long-times supports a behavioral
argument, with overreaction in the ST being driven by emotional behavior, and then rational
behavior occurring into the LT, generating an eventual mean reversion.

79

Semi-Strong Form Tests Cross Sectional. Cross section return studies have been very
interesting. All assets should have equal risk adjusted returns in an efficient market.
Low P/E stocks have been noted in the literature for some time. Graham (1934) argued for a
reasonable multiple of average earnings. Schneider (1951) reviewed the period 1917 to 1950,
and concluded that low price to earnings did not outperform until 1933, but from that time until
1950, the strategy worked very well. Paul Miller, Jr (1966) studied firms from 1948 to 1964
having over $150 million in sales. He found that low PE stocks had higher returns than
equities with high PE ratios. Francis Nicholson (1960, 1968) initially found outsized returns in
the chemical industry from low PE firms, and then described a low PE effect from all types of
firms as producing superior returns. McWilliams (1966) found better performance from low PE
stocks for the period of 1953 to 1964. Breen (1968) also studied low PE ratios after
eliminating all firms with low growth of under 10%. Graham (1974) commented upon an inhouse study done by Drexel Firestone, covering the periods of 1933 to 1969 on the NYSE.
Drexel discovered that equities with low price to earnings outperformed the high PE equities
and the DJIA in 25 of 28 years.
In more recent time frames, Basu wrote three papers in 1977, 1978, and 1983 that looked into
PE ratios. In the earliest study, he found that the lowest price to earnings stocks had greater
returns than high price to earnings stocks going forward over a 14 year study period. Basu felt
that low PE stocks were not associated with higher systematic risks, contrary to conventional
market theory. In the 1983 paper, the NYSE was ranked into deciles by market cap and price to
earnings, and showed that the lowest price to earnings stocks outperformed stocks for every
market size. Basu also found that small size outperformed large cap stocks at most price to
earnings deciles, suggesting both a size and style effect. Goodman and Peavy (1983) examined
firm size, industry effects, and infrequent trading, and found that risk-adjusted returns for
stocks in the lowest P/E quartiles were superior than those in the highest quartiles. Goodman
and Peavy (1985) then examined industry performance between 1962 and 1980 and found that
the lowest price to earnings firms in each industry had higher annual returns and cumulative
returns, across each industry. Ibbotson (1986) ranked the NYSE stocks into deciles and found
that the compound return for the lowest price to earnings ratios significantly outperformed the
NYSE return.
Dreman Value Management (1989) studied the Compustat data base for a 202 year period
ending 1989, and produced similar findings to that of Basu, namely that small decile stocks
outperformed large stocks for the lowest three price to earnings deciles (but then large caps
outperformed small caps for the two highest price to earnings deciles). Dreman also showed
that the low price to earnings did considerably better for each market size decile, again
implying both a size and value effect. Lakonishok, Vishny & Shleifer (1994) used NYSE data
between 1968 and 1990, and concluded that low price to earnings stocks performed better than
the highest decile price to earnings firms on both the average five return and the cumulative
five year returns. OShaughnessy (1998) used the Compustat data base, and also determined
that stocks with low PE ratios do much better than those with high PE ratios. Dreman (1999, at
194-195, 201), in conjunction with Eric Lufkin, examined both PE and price to dividend ratios
within industries and discovered that low industry-relative PEs and price to dividends
outperformed the broader markets and the higher price to value ratios in each industry. Just as

80

importantly, four separate industry-relative price to value indicators (P/D, PCF, PBV, & PE)
performed better than the markets. Tweedy, Brown (2003) was of the belief that companies
with low price to earnings often had above average cash dividend yields, and retained earnings
were often reinvested in the business for the benefit of the shareholders.
Levis (1989) studied UK companies between 1961 and 1985 and concluded that annual and
cumulative return of the lowest decile price to earnings performed better than the highest delice
price to earnings equities. Chisholm (1991) examined price to earnings effects for four
European countries covering the period 1974 through 1989 and found that the lowest price to
earnings quintiles outperformed the highest price to earnings quintiles for all four countries.
While aggregate level information highly suggests the usefulness of a low PE ratio, significant
problems may be encountered when the methodology is applied at the firm level. Damodaran
(1996, at 307) notes that PE ratios are not overly helpful when they are negative; that the PE
will change across periods for the same firm, across industries, and across national boundaries;
that the discount rate should decrease across time; and that the earnings growth rate, interest
rates, and firm volatility risk all dramatically affect a firms PE ratio. For instance, studies
generally show a direct relationship between the PE ratio and earnings growth, and an inverse
relationship between PE ratios and interest rates as well as PE ratios and volatility risk.
Investors attempting to utilize a low PE (or low PEG or even PCF ratio) must keep these
variables in mind, lest they go astray.
Several PE to Growth ratio studies have been conducted recently. Peters (1991) discovered an
inverse relationship between the between PEG and average rates of return in the period 19821989. Reilly and Brown (2000, at 228) cited to several additional studies also showing the
existence of an inverse relationship. The significance of such a finding is a refutation to the
EMH, similar to that of a low PE ratio generating future period higher returns. But, when
Reilly and Marshall (1999) used Value Line data and exempted low beta and low growth rate
stocks, they found that no inverse relationship existed between the PEG and forward returns.
Instead, the study supported both a small cap effect and a P/E effect.
Employing Grahams original valuation technique of selecting stocks with low price to Net
Cash Asset Values, Oppenheimer (1986) studied calculated NCAV types of stocks from 1970
to 1983, and concluded that such stocks outperformed the general markets. Tweedy & Brown
research (2003) has suggested that a NCAV strategy not only produced superior stock
performance, but firm pricing was often below liquidation value for the entire corporation.
Stattman (1980) felt that price to book ratios were even more significant than PE ratios in
predicting future stock returns. DeBont & Thaler (1982) demonstrated that the lowest 20% of
stocks in terms of price to book outperformed the market indexes by 8.9%. The paper pointed
out that pricing and earnings were mean reverting over time. Rosenberg, Reid, & Lanstein
(1985) found that a low price to book strategy did better than firms having high price to book
as well as the overall market. Ibbotson (1986) showed that low price to book ratios had
significantly better returns of an 18 year test period did than high price to book equities.
Leverage (i.e. debt to equity ratios) explained the cross section of returns after both beta and
size were considered (Bhandari, 1988). French and Fama (1992) showed that for every market

81

cap category, the best returns were produced by stocks with low price to book, and this method
was consistently more powerful for explaining returns than beta, market cap, price to equity,
leverage, price to book percentage. Lakonishok, Vishny & Shleifer (1994) showed that the
average five year returns and average cumulative five year returns were much higher for stocks
with the lowest price to book value. Low price to book outperformed high price to book
equities in 16 out of 22 years, or 73% of the time. For five-year holds, low price to book
outperformed in all periods. OShaughnessy (1998) found that low PBV performed better than
an all-stock universe, but with much greater pricing volatility. Risk was moderated when large
stocks with low PBVs were selected. Chan, Lakonishok, Sougiannis (2001) incorporated
tangible assets in book value and discovered that this improved the results. Tweedy, Brown
(2003) also confirmed out-performance with a low price to book philosophy. Chan &
Lakonishok (2004) showed that book to market value outperformed, even after size was
controlled.
Internationally, several studies have also shown a price to book effect. Morgan Stanley (1991)
studied an internal data-base with 80% of the companies being outside of the US. Low price to
book significantly outperformed. Chisholm (1991) studied firms in UK, JA, France, Germany.
In all four countries, low price to book outperformed high price to book equities. Chan,
Hamao, and Lakonishok (1991) found positive returns for BV / MV ratios in Japan. Capaul,
Rowley, and Sharpe (1993) examined several countries and found that value stocks
outperformed growth stocks in each country studied during the test period, both absolutely and
after risk adjustment.
The price to book ratio is dependent upon several accounting items, namely the Return on
Equity, the required rate of return, and the dividend growth rate. PBV will generally be higher
for high firms with high ROE (Damodaran, 1996). An undervalued portfolio of low PBV
and high ROEs was found to substantially outperform an overvalued portfolio of high PBV
and low ROEs (Damodaran, at 334). Thus, a low PBV may not matter so much as a low PBV
compared with a high ROE. Other researchers have generated lists of undervalued firms from
Tobinss Q (Lang, Stulz, and Walkling, 1989), and from a consolidation of ROE, growth, and
PBV ratios, referred to as the Estep T Score (Estep, 1985, 1987). In general, mismatches
between a low PBV and high ROE may present value opportunities.
A study on the Book value / market value ratio showed a significantly positive relationship
between the historical BV/MV and returns. In fact, the strongest evidence on this came from
French and Famas 1992 study showing a positive relationship between beta and returns before
1969 but no relationship between 1963 and 1990. F&F also found a significant positive
relationship between BV/MV and return that persists even when other variables are included.
Both size and BV/MV are significant when included together. So, F&F 1992 casts doubt on
the CAPM and use of beta as well as lending support to the BV/MV as a predictor of return
rates. Another study suggests that P/BK related to ROE in the future, while PE was related to
future growth in earnings. Another study confirmed that optimal combination was a portfolio
of small firms with a high BV / MV. Generally, studies show a negative and inverse
relationship with returns and P/BK. There is a positive correlation between Book value and
Market value. Low P/BK yields excess returns. The same is true in international markets, too,
with 1% in the US and 3% internationally gained by a P/BK tactic.

82

Keppler (1991) found that low price to cash flows outperformed high price to cash flow
equities among stocks listed on 18 national indexes of Morgan Stanley. Lakonishok, Vishny &
Shleifer (1994) determined that low price to cash flow produced far greater returns than did
equities having high price to cash flow. This was true over each year of a five-year hold, the
average return for a five-year hold, and the cumulative return for a five-year hold.
OShaughnessy (1998) also determined that low price to cash flows outperform on an absolute
basis, but at a much higher risk level. Selection of large caps with low PCF could reduce the
volatility significantly.
Litzenberger and Ramaswamy (1979) established a correlation between dividend yield and
future returns. Williamson (1985) came to the conclusion that excess returns could be obtained
through a portfolio of undervalued stocks, as calculated by the Dividend Discount Model.
Fama and French (1988) showed that aggregate stock market returns were higher when the
dividend yield was high, and Campbell and Shiller felt that the yield could actually predict
forward returns. Levis (1989), using over 4,000 companies on the London exchange,
demonstrated that high dividend yield equities produced far greater cumulative value than did
stocks with lower yields. Lenhoff (1990) found an inverse relationship between price to net
asset value (as well as price to earnings ratio), and the dividend yield. Lenhoff believed that
high yield companies were often takeover targets. Keppler (1991), using 18 Morgan Stanley
International equity indexes and equally weighting them, discovered that investments in the
highest yielding stocks provided far greater total returns than low dividend yield stocks. When
using a Dividend Discount Model, Haugen (1993) showed that 250 undervalued large-cap
equities produced excess returns on a risk-adjusted basis. James OShaughnessy (1998)
showed that the 50 highest dividend yielding stocks had 1.7% higher annualized return than an
all-stock universe for deciles of medium to large size companies. But then an all-size, high
dividend screen failed to beat the benchmark, implying that a high yield strategy should be
confined to mid-size to larger firms. Siegel (1998) indicated that a high dividend yield
selection of 10 stocks on the Dow Jones (the strategy is often referred to as Dow 10 or Dogs
of the Dow) generated higher returns and at lower risk levels than either the DJIA or S&P
500. Miller (1999) strongly believed in dividend growth investing preferences, maintaining
that companies with dividend growth exhibited strong cash flows, superior quality, and usually,
reasonable to low valuation levels.
There is some evidence to suggest that high dividend-yielding stocks may only be a proxy for
low PE ratios, however (Jacobs and Levy 1988), so it is unclear whether a dividend strategy
would generate different results from a tactic based on a low PE effect.
Ken Fisher (1984) believed in PSRs as a measure of value, and felt that a stocks price to sales
ratio was almost a perfect measure of popularity. Senchack and Martin (1987) found that
low PSR equities out-performed high PSRs, but not a low PE strategy. There was also a
determination that low PEs provided more consistent returns than a low PSR, and that low PS
ratios were biased towards smaller-caps. Jacobs and Levy (1988) used an industry analysis,
and believed that low PSRs generated excess returns. Studies suggest that a direct relationship
exists between the PSR and profit margins. Damodaran, (1996), suggested that under
valuation (or over valuation) may exist with equities having low PSR and high profit margins

83

(or high PSRs and low profit margins). This is due to PSRs being greatly affected by profit
margins. OShaugnessy (1998) indicated that a low PSR strategy had outstanding results,
better than any other value ratio. The results were very consistent, as well, with 90% of all
rolling ten-year periods outperforming an all-stock portfolio. This was true for both small and
large caps having low PSRs.
DeBondt and Thaler (1985) studies the NYSE between 1932 and 1977, and showed that the
worst performing stocks over a past five year period produced an average cumulative of 18.2%
return in excess of the market index in 17 months after portfolio formation. The best
performing stocks in the preceding five year period generated a cumulative average of 6% less
than the market index. Poterba and Summers (1988) studied large price declines
internationally. They discovered that stocks tended to revert to the mean average over periods
of time in excess of one year. Power & Lonie (1988 ?) studied 200 UK companies, and found
the same effect.
Firms that have had recent difficulties with negative earnings, low or negative book value and
equity, and having no dividends ironically may develop the greatest returns into the future. The
effect was long ago noted by Kenneth Van Strum (1925). Assuming the firm survives the
distress and ultimately resumes a profitable future course, it would seem that excess returns are
in store for an investor. More recent evidence shows that the ability to generate excess returns
from businesses in distress still exists today (Siegel, 1998).
Size Effects. In addition to French and Fama (1992) and others cited above, numerous articles
have examined whether size matters in the returns of equities. Ibbotson (1980) computed
returns for all stocks other than large companies in the S&P 500 from 1958 to 1979 at twice the
total return as that of the entire S&P. Rolf Banz (1981) found that small stocks systematically
outperformed large cap equities, even after adjusting for risk. Reinganum (1983) studied the
NYSE and American Stock exchange between 1963 and 1980. Smaller market cap equities
considerably outperformed large caps. (Chisholm (1991) also found a small size effect in his
four country study. Levis (1988) looked at only the UK market, and also determined that a size
effect exists. Small cap stocks have also been found to outperform in Canada (Berges, 1984);
Australia (Brown, 1983); and Japan (Nakamura and Terada, 1984).
Some studies suggest that the small size effect is actually being generated by low price to
valuations, and therefore may ultimately be part of the value style (Basu, 1983). In fact,
BARRA ranked firms in the S&P 500 and discovered that value stocks tended to be actually
smaller in size than most growth stocks, implying that any small cap study would, by the vary
nature of data sampling, pick up value styled stocks within the study group (cited in Siegel, at
103). Dreman (1999, at 319-330) critiqued the work of Banz, Reinganum, and Ibbotson, and
concluded that except for 1931 to 1935 and 1941 to 1945, small stocks did historically
outperform the markets, but they had a far greater risk of insolvency. Dremans statistical runs
showed that excess returns were being generated by small firms having low PEs.
OShaughnessy (1998, at 42-43) believed that almost all of the small cap effect was due to a
boosting of returns from micro-caps with under 25 million in market capitalization. Once
micro-caps are removed from a statistical sort, the return rate of the small caps is greatly

84

lowered to levels more closely aligned with large cap issues. Even worse, small-caps without
the micro-sized companies underperformed larger firms on a risk adjusted basis (Id., at 47).
Siegel (1998, at 254) showed that the entire small cap gain came almost exclusively as part of a
January Effect. Once the returns for January were deleted from the data, small caps showed no
more promise than any other stock. In fact, small caps under-performed large caps and the S&P
for all months except January. Siegel also demonstrated that small caps have only
outperformed since 1975. Before that, small caps returns were quite close to that of large-caps
(Id., at 94). Some studies have also concluded that the small size effect is not overly stable,
with small size firms going through irregular periods of higher or lower rates of returns.
(Brown, Kleidon, and Marsh, 1983).
Other writings (contained above) note that both size and value effects are going on
simultaneously, while still other researchers are of the opinion that a basic interrelationship
among several styles may be involved with size and value criteria (contained below).
The lack of information and limited institutional interest in certain equities, especially smallcap issues, may generate out-sized returns for those investors that can identify under-analyzed
firms. Arbel and Strebel (1983) found a small sized effect but also an effect for those
businesses not extensively followed by investment professionals. Arbel (1985) showed that the
January effect was the greatest for under-researched firms. Another study however found no
evidence of a neglected firm effect after controlling for size (Beard and Sias, 1997). Barry and
Brown (1984) suggested that firms with less information require higher returns in order to
account for the higher levels of uncertainty of forecasting. Damodaran (1996, at 185) noted
that return rates decreases as the number of analysts following an equity increases.
Concentrating on businesses having a dominant position in their respective industries is
favored by some value investors, including Warren Buffett. The Standard & Poors Company
has maintained an earnings and dividend ranking of stocks since 1956. The rankings with a
high quality rating, based on superior earnings and dividends, have outperformed on both an
absolute and risk adjusted basis, and with lower amounts of variance and semi-variance or
downside risk (S&P Report on Earnings & Dividend Ranking System, 2003). Statistical studies
have also shown that quality matters. OShaughnessy (1998, at 47) found that market leading
businesses outperformed the S&P 500, while having much less risk. They provided excellent
performance over a variety of market conditions.
Macro Effects. All types of investors, whether value-oriented or otherwise, should be aware of
certain macro-economic factors which influences investment returns. Several studies have
shown that most changes in a firms earnings can be traced to changes in aggregate corporate
earnings and changes in a firms industry. Studies consistently demonstrate that the economic
environment has a significant effect on firm earnings. (Reilly and Brown, 2000, at 442). A
relationship exists between stock prices and economic expansions and contractions (Siegel,
1991). Most of the changes in rates of return for individual businesses can be explained by
changes in the rates of return in the aggregate stock market and a firms industry (Meyers,
1973), and interest rates are considered the single most important macroeconomic factors for
investment analysis purposes (Bodie, Kane, and Marcus, 2004). The direction and magnitude
of business cycles are also important factors to analyze, and stock pricing normally leads

85

economic activity (Jones, 2004, at 347). Further, evidence exists of an inverse relationship
between the rate inflation and price to value multiples (Id., at 434).
Accordingly, a three-step process in the valuation and selection of assets is often advocated:
first, analyze the country-level economic environment; second, review industry factors; and
third, identify firm characteristics. (Reilly & Brown, 2000). While many value investors prefer
a bottoms-up investing approach due to their focus on the individual selection of businesses,
looking to various macro-economic factors is quite consistent with blending value indicators
together to form an overall appraisal of business prospects.
Blending Various Value Indicators. The literature suggests that incorporating several value
indicators into an overall stock selection strategy reinforces valuation characteristics in the
assets chosen by the strategy. Indeed, value and size indicators may be interrelated, and show
the same or similar undervaluation of firms. For example, Keim (1988) reviewed pricing data
on the NYSE between 1964 and 1982, and found that lower price to book ratios were
associated with lower price to earnings ratios, smaller size, and lower stock prices. In another
study, Keim (1988), showed that price to earnings were associated with smaller market
capitalizations over a period of 35 years. And, many of the value factors may overlap as well,
with the same stocks possessing several low price to valuation indicators (OShaughnessy,
1998).
Chan, Lakonishok, Sougiannis (2001) found that blending various approaches might allow
investors to have larger returns than just with value versus growth distinctions. Chan &
Lakonishok (2004), in a review of past valuation articles, noted that French and Famas book
value to market value method was not necessarily an ideal measure in that strategies based on
several value signals might enhance portfolio performance. And, mixing various valuation
ratios, such as PE and P/BK with size, monetary environments, neglected firm analysis, and the
anticipation of earnings surprises has also been recommended (Reilly and Brown, 2000, at
249). OShaughnessy (1998) mixed various value factors together to increase returns. In one
screen, market-leading firms with high dividend yields substantially outperformed a large stock
universe. Another screen composed of both value and growth factors outperformed an all-stock
holding of stocks by 4% per year, and with very similar standard deviation. Brown, Tweedy
(2003), in an extensive review of the literature, believed that there are recurring and often
interrelated value-oriented patterns existing over very long periods of time.
After penning their article in 1992, French and Fama went on to propose a three-factor model
incorporating the style, size, and country distinctions (1993). Back-testing of the three factor
model has produced some interesting results. In one study (Ibbotson and Sinquefield, 1989), a
large value strategy outperformed both the S&P 500 Index and a large growth grouping of
stocks, while small value also generated superior returns over small growth stocks. Further,
small caps generally outran the large companies. Very importantly, the value and size effects
occurred with no increase in the standard deviation of variance. Size and BV/ MV appear to
provide good combinations, but only during periods of expansive monetary policy (Jensen,
Johnson, and Mercer, 1996). In another writing (Tanous, 1997), a mix of styles, sizes, and
country passive funds (and with a zero weighting for ST or LT bonds) outperformed the S&P
500 by over 5%, and with slightly less deviation risk. Even when the mix of passive equity

86

funds were only 60% of the total allocation, the rate of return still exceeded the S&P Index by
1.5%. Importantly, the standard deviation was then markedly reduced to 8.5% versus the S&P
500s 13.65% deviation. Similar statistics have been generated for active management, so long
as there is broad diversification with sub-grouping for the three factors of size, style, and
country weighting. For a ten-year back-testing period, active funds sorted by the three factors
produced the same to higher rate of return versus the S&P (and by as much as 3%), and with
very similar deviation risk compared to the S&P (Tanous, 1997).
Some studies have examined the make-up of the price to value ratios themselves, and have
uncovered complex accounting relationships. Fairfield (1994) showed that the Price to Book
ratio depends upon the expected levels of future ROE, while the P/E ratio is dependent upon
expected future growth in earnings. French and Fama (1995) ascertained that equities with low
price to book had low ROE prior to the test period, but increases in ROE after portfolio
formation and the test period began. Damodaran (1996, at 292) indicated that the PE ratio was
based upon the pay-out ratio, the dividend growth rate, and the required rate of return, while
Price to Book Value turned on ROE, the growth rate, and the required rate of return. Price to
Sales was composed of profit margins, payout ratios, the growth rate, and the required rate of
return. These ratios could vary across nations, industries, business cycles, and various time
periods of a firms development.
Contra Indications on Cross Sectional Studies. While many studies have shown validity to the
value style, some other works believe value investing to be a dubious endeavor. Bogle (1999)
showed that value and growth styles produce almost identical results over very long time
frames of 60 years duration. And, high-grade stocks versus low-priced stocks also had very
similar returns. Siegel (1998) found that excluding 1975-1983, large growth stocks actually
beat larger value stocks, although smaller value equities still had better performance than small
growth stocks. Additionally, a strategy shown to be superior on a back-testing basis may not
out-perform when projected into the future. For example, regression analysis has been quite
useful in explaining past PE ratios, but has not been particularly helpful in predicting future
performance (Damodaran, 1996, at 305-307).
Semi-Strong Form Tests Specific News Events. As to event studies, numerous types of
events have been analyzed. There is a general threshold question with these studies though, as
to whether the reaction is instantaneous and unbiased. The three general types of reactions
being sought out are a quick pricing reaction (efficient); a slow learning pricing with gradually
price rises following an info release (evidence of undervalue and excess returns in the price
drift); and a high initial reaction followed by a gradually declining price (overreaction with
short selling generating excess returns).
Studies on stock splits, IPOs, world events, economic news, accounting changes, and a variety
of corporate financial events have shown that the market reacts quickly to such events (Reilly
& Brown, 2000; Jones, 2004). Evidence indicates that valuable info is conveyed to the
financial markets, and there are very quick pricing reactions following the announcements.
Investment announcements normally have positive reactions by the market this is evidence of
a rational and efficient LT reaction to R&D announcements, even though one would expect an
initial ST reaction, due to a ST loss of cash flow being devoted to capital needs.

87

Some evidence also shows a market reaction on the day before the announcement. This could
show insider trading but could also be from release of the info to the news media the day
before at market closing followed by market reaction the next day. Many studies show that a
drift in prices occurs in the week before an announcement, implying insider trading, and then a
large price impact on the day of or the day before. A return drift following the announcement
appears in some papers, as well. 62 Further, there is consistent evidence that a delayed
announcement is much more likely to be negative, as well as announcements made on Friday.
Late announcements more than 6 days also will tend to convey bad info.
The literature suggests that the markets do not adjust to quarterly earnings surprises as quickly
as anticipated by the EMH, and that market pricing begins adjusting even before the
announcement, implying insider activity (Jones, 2004, citing to other studies). Favorable
information on the reports is not immediately reflected in the stock pricing and a significant
relationship exists between the size of the earnings surprise and the post announcement price
change. Another study showed abnormal returns even after all adjustments have been made
under the CAPM. On the standardized unexpected earnings (SUE) studies, the difference
between the actual and expected earnings is normalized by regression. Various studies
generally showed large SUEs being accounted for in stock pricing.63 Other papers studies
demonstrate that pricing is not adjusted to reflect the release of quarterly earnings surprises.
To profit from such surprises however, one must do a superior job at predicting the earnings
surprises, which is very difficult to do on a consistent basis.
But the information may not be instantaneous in nature. There is also a drift in prices
occurring in the week prior to an announcement, and then a large price impact on the day of or
the day before the announcement. This implies insider trading, or at least leakage. While the
earnings studies show less than perfect efficiency, some commentators feel that reactions to
earnings announcements and surprises generally support efficiency concepts.
Semi-Strong Form Tests Anomalies. There are several studies that are inconsistent with
risk/return models. Some of the problems may be that the wrong models are being used
instead of the behavior being irrational.
The calendar studies show a January effect, with downward pressure in pricing in December
and then upward pricing pressure in January due to tax selling. The effect is in every market
internationally, regardless of tax year consequences (inconsistent with the CAPM).
There has also been some other calendar effects noted. There is a weekend effect, with Monday
returns being quite negative. The effect still exists in Japan with weekend trading, and does not
62

An early paper on earnings announcement was: O. Maurice Joy, Robert H. Litzenberger, and Richard
McEnally, The Adjustment of Stock Prices to Announcements of Unexpected Changes in Quarterly Earnings,
Journal of Accounting Research 15, no. 2, (Fall 1977): 207-225.
63
See for example, Henry A. Latane, and Charles P. Jones, Standardized Unexpected Earnings A Progress
Report, Journal of Finance 32, no. 5 (December 1977): 1457-1465; and Richard J. Randleman, Jr, Charles P.
Jones, and Henry A. Latane, Empirical Anomalies Based on Unexpected Earnings and the Importance of Risk
Adjustments, Journal of Financial Economics 10, no. 3 (November 1982): 269-287.

88

exist on Tuesdays following a Monday holiday (refuting the argument that the effect is due to
earnings announcements being released on Friday).
The evidence also suggests that small firms can generate higher returns than larger firms of
smaller beta. Size studies show that small firms consistently experienced significantly larger
risk adjusted returns than the larger firms. The thought is that less info is available on small
firms, and that transaction costs are higher. But the cost differential is unlikely to account for
the size of the small firm effect. CAPM may not be the right model here, since beta may be
underestimated for small firms. Additional risk other than beta may exist with small firms.
There is also evidence of the effect existing outside of the US.
Summary on the Semi-Strong form. As to the cross sectional studies, substantial evidence
exists against the semi-strong form of EMH. Perhaps Reilly and Brown (2000, at 233) summed
up the issue in the clearest terms when they wrote: the tests of publicly available ratios that
can be used to predict the cross section of expected returns for stocks have provided substantial
evidence in conflict with the semi-strong form of EMH.
The evidence is more mixed on reaction to news events, while many anomalies are in apparent
contradiction to the EMH. Thus, the tests of the semi-strong form present greater challenges to
the EMH. One interpretation is that the returns are not properly adjusted for risk, leading the
CAPM to calculate a higher than normal return. Others argue that the price to value studies are
evidence of inefficiency, with supply and demand failing to be fully efficient until the RTM
occur in the LT.
There is still the issue of operational efficiency: profit making activities cannot be reliably
engaged in and inefficiencies taken advantage of, once trading expenses, taxes, and expense
ratios are factored into the process. While some highly skilled professionals can and do
regularly outperform the broad markets (i.e. Buffett, Templeton, Bill Miller, etc), on the
average, professional managers have underperformed the relevant benchmark indexes by at
least the costs of research and trading, let along tax impacts. Many, if not most, professional
and individual investors would be better off by simply using long-term passive strategies that
emulate the benchmarks.
Strong Form Tests. As to inside information, several studies have shown abnormal profits
from insider transactions. Once the insider trades are announced by the SEC, however, any
further abnormal profits are insufficient to overcome trading costs. Specialists enjoy monopoly
power over the limit orders info, and thus also have abnormal returns. Analysts
recommendations are mixed, and the mutual fund industry under performs, after costs are
factored in. All of this supports the strong form. On analyst recommendations of the VL
timeliness, recent studies show that the pricing change is not large enough to offset costs. One
study shows that most of the VL impact occurs when a stock moves from a #2 to a #1 timely.
Another study shows that most of the impact is really due to a quarterly earnings surprise
impact (previously noted in the semi-strong). Also, the VL centurion fund has underperformed over the last decade. Studies on investment managers and other professionals
potentially possessing some information show that most managers do not consistently
outperform a buy and hold philosophy on a risk adjusted basis.

89

Note on international efficiency.64 In an international context, the notion of efficiency requires

not only that individual national markets be efficient but also that the individual markets be
properly priced relative to the world index. International market efficiency would preclude
fund managers from using active asset allocation between countries to consistently beat the
world index.
The fundamental question of international efficiency involves the issue of international market
integration versus segmentation. Integrated world markets allow capital to move freely across
borders, allowing investors to seek the most efficient portfolios. If capital flows freely,
securities with similar risk / return characteristics will be priced similarly in all markets relative
to the world index.
However, if world markets are segmented, then impediments to the flow of capital prevent
investors from taking advantage of relative mispricing between countries. In this case, pricing
of similar securities will not be equivalent in all markets, and superior performance by some
investors may be possible. There are several impediments to the international flow of capital:
-

Psychological barriers. Unfamiliarity with international markets (e.g., language and

information sources) may cause investors to limit foreign investment.

Legal restrictions. Institutional investors may be prohibited by domestic regulation from

investing internationally. Local governments may impose restrictions on foreign
investment flowing into or out of the country.

Transaction costs are often higher for international investing.

Discriminatory taxation. Foreign investment may be taxed at a higher rate than locally
generated investment.

Political risks. Local government actions can cause adverse changes in the value of
invested funds.

Foreign currency risk. Foreign investors bear the risk of changes in the local market value
of invested funds, as well as the risk of changes in the exchange rate between the local
foreign currency and investors' home currency.

These impediments are most commonly observed in emerging markets, but could be present, at
least to some degree, in any foreign market. International market integration requires only that
a sufficient number of investors be able to move capital between markets. Several factors have
caused international capital mobility to increase dramatically over the past twenty years. There
are many private and institutional investors who are internationally active. All major
corporations have multinational operations. Corporations and governments borrow and lend on
64

This section is drawn from CFA Level II reading, 2012.

90

an international scale. Given these developments, international markets are likely heading
towards integration, with segmentation occurring where barriers remain high.

Interpreting the Evidence

Are the anomalies simply evidence of risk premiums or are they inefficiencies? Fama and
French believe that their three-factor model may merely be evidence of a market premium,
while Lakonishok (1995) argues that it is evidence of systematic errors in analysts forecast, and
thus is representative of inefficiencies. There may also be data mining issues at work, with
back-testing techniques showing good patterns that cannot then be repeated going forward in
time.
The classical view is that prices quickly react to information; that the flow of information is
random, and therefore, price changes are random. Stock prices fully and accurately reflect
publicly available information. Once information becomes available, market participants
analyze it. Competition assures that prices reflect information. Even with the markets being
efficient, a role exists for portfolio management to ascertain and develop appropriate portfolio
investment policies at both individual and institutional levels.
Prices may not exhibit a pure random walk, but many observers believe that markets are
operationally efficient. The question is not whether the markets are efficient in theoretical
terms, but to what extent are the markets efficient. Markets are generally viewed as constantly
moving towards efficiency, with a great tendency towards centering over the long-term.
The behavioral view is that prices do not always quickly adjust to information; when
adjustments do occur, they are not necessarily accurate in their reflection of intrinsic valuation
levels; that human behavior may be a major cause of inefficiency through overreaction to news
and a slew of cognitive biases and errors; and that both ST and LT time-series studies support a
behavioral view, with overreaction in the ST, followed by RTM in the LT, and ultimately, a LT
market equilibrium that can said to be rational.
Implications. Some of the implications of EMH include:
-

Past patterns and past trends of prices are unreliable indicators of future prices (TA does
not work).

If markets are efficient, investors can only expect to make a fair, but not abnormal return,
commensurate with the assets risk (Fundamental Analysis is also of little use).

Abnormal returns are randomly dispersed and are therefore difficult to repeat.

The NPV of all investments is zero, because the rate of return will be the discount rate of
future benefits, equaling zero NPV.

Markets are efficient because people believe they are not, and thus try to earn abnormal
returns, thereby forcing the markets back to efficiency.

91

Publicly known strategies will not generate abnormal returns, since everyone takes
advantage of the strategy.

Out-performance cannot be reliably repeated.

Past performance is no guarantee of future performance, and may even be a poor guess at
future performance.

Financial markets are more efficient than physical markets that have substantial
informational and transactional frictions.

Other important implications include:

EMH does not mean that investors will not make a profit; that prices are perfectly random; nor
that we do not need to do our homework prior to investing. In fact, LT investing in solid
companies and index funds makes a great deal of sense, and will eventually lead the markets to
efficiency. Even with the EMH, stock pricing can still deviate from true value, it is just that
the deviations will be random in nature. Further, investors can beat the market and EMH
would still hold. Indeed one-half of all investors should do better and one-half should do
worse than the market average in any given time period. The laws of probability would still
allow for a fairly large number of investors to consistently beat the market over the LT.
The probability of finding inefficiencies decrease as liquidity increases and as trading costs and
info costs decrease. Investors who can exact a cost advantage will be more able to exploit small
inefficiencies. The speed of inefficiency resolution is directly related to how easy the method is
executed for exploiting the inefficiencies.

Conclusion
Prices may not exhibit a pure random walk, but many academic-oriented authors believe that
the EMH is still the best model that describes price behavior. Financial markets are not
perfectly efficient, but are operationally efficient. The question is not whether the markets
are efficient in theoretical terms, but to what extent are the markets efficient? Markets are
generally viewed as constantly moving towards efficiency, with a great tendency towards
centering especially over the LT.
Behavioralists argue that the markets center because of reversion to the mean tendencies, and
can be inefficient in terms of intrinsic valuation not equaling the market price at any one time.
Glaring inefficiencies between PV and market price can continue until such time that a catalyst
occurs, with pricing then moving to back intrinsic levels, sometimes within very short time
frames. The question may thus be seen in time related terms. It is not whether the markets are
generally efficient. The central question becomes: in what time horizons do the markets
achieve efficiency?

92

KCK Note: I have additional thoughts on efficiency and behavioral matters on the Value
working paper, 2005:3.

93

Behavioral Finance
Anomalies have long been noted in the literature, and abnormal returns derived by using price
to value indicators also exist. These items suggest problems with rational behavior
assumptions. Behavioral Finance seeks to explain the inconsistencies in how the markets are
supposed to behave as opposed to how they actually do behave. MPT and CAPM oversimplify
or ignore many behavioral aspects of finance. Rational behavior is an important starting point
to the analysis, but then behavioral issues may force a reevaluation of the theories.
Numerous articles have been written on the impact from investor psychology. As far back as
1852, Charles Mackay noted emotional behavior in the financial arena in Extraordinary
Popular Delusions and the Madness of Crowds. This may have been the first writing to detail a
mass psychological force at work. More recently (circa 1960s and 1970s), a few studies
discussed psychological and emotional behavior. Ignoring prior probabilities in decisions
interfere with intuitive predictions in a wide variety of fields, including financial analysis
(Slovic, et al, 1977). Sherif and Sherif (1969) demonstrated that group opinion can be shaped
and manipulated quiet easily, and the judgment of test subjects can be shifted by as much as
80%. Ignoring prior probabilities in decisions interfere with intuitive predictions in a wide
variety of fields, including financial analysis (Slovic, et al, 1977). David Dreman, an
investment manager, wrote two very good books in the late 1970s, one on Psychology in the
Capital Markets, and then a second book that more formally introduced his investment
practices styled as Contrarian Investing.
In 1979, Kahneman and Tversky moved the discussion into high gear when they penned an
article involving what they referred to as Prospect Theory. 65 Through testing procedures
normally associated with psychological and behavioral disciplines, these two researchers
discovered that individuals did not always make rational decisions under conditions of
uncertainty. This one paper written by two psychologists had a huge impact on the financial
literature. A wide array of thoughts quickly developed from there.
DeBondt and Thaler (1985) observed an overreaction bias that occurred in response to
unexpected and dramatic news events. Chopra, Lakonishok, and Ritter (1992) found that this
bias existed even after adjustments were made for beta and seize effects. Investors tend to be
overconfident in their own abilities to make wise investment choices, exhibit overreaction bias,
have undue amounts of aversion, and engage in mental accounting (Hagstrom, 1999, at 148151, citing to Kahneman and Thaler). A study of investment analysts by Tversky (1995)
showed that investment analysts were markedly overconfident in their own projections, even
though forecasting was known to be error-prone. Another study by Fischhoff (1982) showed
that cognitive illusions exist among professionals, as well: even when experts were warned of
an overconfidence bias in forecasts, they still were unable to adjust for it. Not only does
overconfidence occur on a regular basis, business people are also greatly overoptimistic
(Taylor and Brown, 1988). Dreman (1999, at 76-77, 107) believed that most investment
65

Prospect Theory: An Analysis of Decision under Risk, Daniel J. Kahneman and Anis Tversky, Econometrica,
Vol. 47, No.2 (1979): 263-291.

94

professionals process information in linear ways, even though the analysis of assets requires
high-level configural processing abilities. Informational processing short-cuts, such as
anchoring, hindsight biases, the tendency to draw analogies when none exist, and not following
probabilities in conditions of uncertainty, are additional examples of cognitive biases that can
interfere with decision making in a financial setting. Regret avoidance issues (DeBondt and
Thaler, 1987) and mental accounting (Statman, 1997) may also be involved.
Expectational errors may be part of the reason for superior returns in value stocks. Mispricing
patterns can persist over long periods of time, due to the limits of arbitrage. (Shleifer and
Vishny, 1997). Dreman (1999, at 250-251) felt that pricing changes overreact to changes in
fundamentals. Fuller, Huberts, and Levinson (1993) demonstrated that investors continually
overprice favorites while discounting unpopular firms. Shiller (1981) found that stock price
volatility was too high to be attributed to just new information hitting the market. Rich pricing
of stocks may not be reflective of their fundamental value, but is indicative instead of
investors overly optimistic sentiment towards future growth and a firms ability to sustain
growth (Chan, Karceski, Lakonishok, 2000). Investors may extrapolate past performance into
the future, without realizing that growth rates may not be persistent (Chan, Karceski,
Lakonishok, 2003). Chan and Lakonishok (2004) believed that investors have exaggerated
hopes about growth stocks, but then end up being disappointed when future expectations fall
short. They are also unduly pessimistic about value stocks and wind up being pleasantly
surprised. Chan & Laknoishok (2004) have postulated that low price to value indicators may be
due to investors consistently overestimating growth, thereby paying too much for high-growth
firms and too little for stable firms.
Another area of discussion swirls around the massive market pricing drop in October, 1987 as
well as the more recent 2000 stock bubble. Dreman (1999) believed that the 1987 market
fiasco was proof of market inefficiency, while Shiller (2000) was of the opinion that excessive
pricing of equities leading up to the market peak in early 2000 was evidence of overreaction
and undue euphoria exhibited on a mass scale. Many observers have even accepted the belief
that a bubble did occur in 2000 which was not based on rational behavior (Jones, 2004). While
some theorists, such as Malkiel (6th ed. of Random Walk), argues that the 1987 event was still
consistent with market efficiency, the general line of reasoning is rather strained. It is more
likely that the build-up of stock valuations in the late 1990s and the market drops in both 1987
and 2000 had little or nothing to do with new information hitting the markets. Instead, these
events appear to be in accord with periodic and precipitous market drops that occurred in
historical time periods in the 1700s and 1800s, as well as more recently in 1929. Shiller
(2000) makes a compelling case for irrational behavior existing at the aggregate level.
Critique of Behavioralism. A very powerful criticism of Behavioral issues is blunt and to the
point: It amounts to subjective thinking in search of a theory. Among financial statisticians
used to quantitative calculations and extensive modeling, Behavioral Finance has no central,
guiding theory putting together all the studies. Experts with quantitative backgrounds generally
recoil from anything that does not contain a precise, mathematical explanation. Without a
generalized theoretical basis to it, behavioralism ends up being an ad hoc collection of
qualitative beliefs.

95

Another oft-cited response to the behavioral studies is that the anomalies noted have not
generated an ability to outperform the markets. If the markets are so inefficient, then why has
not anyone been able to use the developing research to their own economic advantage? Why
cant the mutual funds, for instance, take the behavioral research, and then out-perform the
benchmark indices? This leads back to the argument that markets are operationally efficient, in
spite of apparent inefficiencies.
Still another critique of the growing body of work is that it essentially involves data mining.
Using many qualitative and individualized surveys, behavioral studies have emphasized
individual abnormalities that then cannot be used in the future in any systematic way at the
aggregate level.
A related argument postulates that individual-level abnormalities and heterogeneous
preferences, when aggregated, can still sum up to an approximately efficient market operation.
In essence, individual level irrationalities do not necessarily sum to a market-level
irrationality. They may instead average out into a market level response that is still quite
consistent with rational expectations of investors and overall market efficiency. This argument
however is countered with observations of anchoring and herd-like behavior at the aggregate
level. The modern era of equity investing has seen at least three such episodes in the US:
October, 1929; October, 1987; and the events leading up to the 2000 pricing bubble. While
aggregate behavior may smooth out individual abnormalities in some instances, investor-level
irrational behavior can actually be accelerated at the aggregate level upon occasion.

96

Appendix I - Portfolio Equation Review

Raw Data. mean of the population is x = x / N = (x1 + x2 + xn) / N.
mean of the sample is x bar = x / n.
population variance = 2x = (x )2 / N.
sample variance is: S2x = (x x bar)2 / (n 1).
standard deviation for either pop or sample is the square root of the variance.
A short-cut for pop. variance is: ( x2 ( x)2 / N ) * (1 / N).
The short-cut for a sample variance is the same, except that it uses the n-1 divisor.
The covariance for population is: xy = (1 / N) [ (x x ) (y y) ].
The sample covariance uses the n-1 divisor instead of N.
Covariance short-cut is: (1 / N) (xy ((x)(Ey) / N) ].
Grouped data. Outcomes of grouped data associated with the probabilities of the data.
The mean becomes the expected value of a probability distribution. E (x) = xi * p (xn).
The variance becomes: E [(x ) 2 ], which is: (x )2 p (xi). The divisor N drops out since the
data is already normalized with the data always equaling 1.00.
The variance short-cut is the following: E (x2) u2.
The co-variance is: xy = E [ (x x ) (y y), and this = p (x) (x x) (y uy).
The covariance short-cut is: E (x y ) (ux uy).
Return Calculations. Holding Period Return (HPR) = (end. beginning + div.) / beg.
Arithmetic mean = sum or returns in each period / # of periods.
Geometric mean = [ (1 + r1) (1 + r2) . (1 + rn) ] 1/n . aka time weighted return.
Dollar weighted return, or IRR = c1 / (1+r) + c2 / (1+r2)2 + cn / (1+r)n
(APR) annualized percentage return = = (periods in year) * (rate for period).
Effective Annual Rate (EAR) = eAPR 1, or APR = Ln (1+ EAR).
As to inflation: r = real interest rate; R = nominal interest rate, and i = inflation rate.
Approximately, r = R i; More specifically, r = (R i) / (1 + i). Fisher: R = r + E(i).
As to valuation, a one stage DDM is: Po = D1 / r g; or Di / r, with a perpetuity.
Portfolio Expected Return. The Return of a portfolio is: rp = wa ra + wb rb + wn rn.
Expected return: Erp = wi E(ri); expand to: E(rp) = wa E(ra)+wb E(rb) + wn E (rn).
Full equations: E (rp) = rf + bx, or expanded for CAPM: E (rp) = rf + [(E(rm) rf)/m] p.
The expected return is E (r) = p (s) * r (s), where r(s) is the HPR of period s.
Portfolio Risk. risk premium is the excess return above the RFRR, and is: E (rp) rf.
Variance is: Var (r) = 2 = p (b) [ r( s) E ( r) ] 2. Cannot weight var, unlike or Er.
eta of a portfolio = p = wi i; expanded to: wa a + wb b wn n.
Variance in a Two asset portfolio is: 2p = w2a 2a + w2b 2b + 2wa wb ab.
And, 2p = w2a 2a + w2b 2b + 2wa wb ab a b.
The Correlation coefficient, rho, ab = ab / ( a b); then, ab = ab a b.
Portfolio minimum variance is: w* = (22 12) / (21 + 22 2 12);
Variance in 3 to K asset Portfolio = wi wj ij, which expands to:
2p = w1 w111 + w1 w212 + w1 wk1k +
97

w2 w121 + w2 w222 + w2 wk2k + +

wk w1k1 + wk w2k2 + wk wkkk
Note: ii = 2i; Note, ij = ji ; On RFRR, note: 2rf and rf & any = 0.
Further note: Markowitz has the K asset portfolio derived as: 2p = wi22i + wi wj ij,. Reilly &
Brown (2000) at 268, states that in a portfolio of large number of securities, formula reduces to the
sum of the weighted covariances, as stated by: 2p = wi wj ij.
Single Factor Market Model.
Excess return is: Ri = ri + rf. Also, Ri =E(Ri) + M + ei.
Asset Return is: ri = i + i rm + ei; and then, as: ri rf = i + i (rm rf) + ei
Total portfolio risk is: 2i = i (rm rf) + ei.
Market risk is: (Rm- rf); Firm specific risk is ei.
Total risk, variance, is then: 2i = 2i 2m + 2 ei
Nondiversifiable risk is (2i 2m). r2 is the nondiversifiable risk as a % of total risk, and is: r2 = (2i
2m) / 2i . From lecture notes, r2 is also = 2. 2 is also the coeff. of determ.
Diversifiable risk is: 1 Nondivers. Risk; or 2 ei / 2i.
Capital Asset Pricing Theory.
General from is: y = a + bx, which is a linear equation with y intercept = RFRR & b = .
SML / CAPM is calculated by: E (ri) = rf + i [ E (rm) rf ] ;
CML / port is calculated by: E (rp) = rf + [ (E (rm) rf) / m ] p; or E(rp) = rf + cml p
3 Asset / port mix is: E (rp) = rf + [ (E (rp) rf) / xy ] p
Beta is: = im / 2m. Beta of the CML = Slope = (E (rp) rf) / p, or = (E(rm) rf) /m
Excess Return is: Ri = ri rf = + (rm rf) + ei
Risk Aversion is: E(rm) rf = A*2m; The variance of the residual is 2ei = e2i .
Arbitrage Pricing Theory. Basic equation: Ri = i + i Rm + ei
Single factor: Rp = p + p Rm. For multi factors: Ri = i + i1 Rm1 + i2 Rm2 + ei
Bond formulas
PV = c [((1+r) t 1) / (r ( 1+r)t) ], at infinity, reducing to PV = cf / r, for a perpetuity.
For the entire I and P stream: PV = c [((1+r) t 1) / (r ( 1+r)t) ] + P / (1+r) t
Book: PV = coupon / (1+r)t + Par value / (1+r) t
On semi-annual payments, divide the coupon premium or rate and discount rate by 2; and multiply the
time period t by 2.
Yield to Maturity (YTM) is the discount rate making PV of bond payments = its price.
EAR = [ 1 + rq / m ]m 1; m = # of payments per time period.
Equity formulas
Similar to bonds: PV = Di / (1+r)t. Terminal PV = Di / (1+r)i + Pi / (1+r)I
Note: Po = Dt / (1+r) t + Dt / ( r ) * (1+r) t (similar to Philips P&I bond stream equation).
Preferred is generally valued at PV = D / r.
DDM: Po = D1 / (r-g); Po = Do (1+g) / (r-g); for dividends in future yrs only: initially, Pi = D/(r-g);
then Po = Pi / (1+r) i -1
On Discount rate, derive r from Er of CAPM; and r = (D1 / Po) + g, or r = D% + g.

98

Performance Evaluation Measures

Sharpe = (rp rf) / p. Treynor = (rp rf ) / p. Jensen = ri rf = + (rm rf).

99