9.1.

1 Using the APT:

The Arbitrage Pricing Theory (APT) emphasizes the importance of identifying risk factors in security
valuation. In this example, we consider two factors: unexpected inflation changes and unanticipated
GDP changes. The risk premiums for these factors are 2% per 1% change in inflation and 3% per 1%
change in GDP, with a zero-beta return of 4%.

Two assets are analyzed:

Asset X:

Sensitivity to inflation: 0.50

Sensitivity to GDP: 1.50

Expected return: E(R_x) = 0.04 + (0.02)(0.5) + (0.03)(1.5) = 9.5%

Asset Y:

Sensitivity to inflation: 2.00

Sensitivity to GDP: 1.75

Expected return: E(R_y) = 0.04 + (0.02)(2) + (0.03)(1.75) = 13.25%

If the market prices do not reflect these expected returns, arbitrage opportunities will arise as investors
buy underpriced assets and sell overpriced ones, moving toward price corrections.

9.1.3 Empirical Tests of the APT:

The Arbitrage Pricing Theory (APT) has been the subject of extensive empirical testing, particularly
highlighted in the Roll and Ross (1980) study. Their two-step methodology involved:

1. Estimating expected returns and factor coefficients from time-series data on individual asset returns.

2. Testing whether the expected returns for assets aligned with the common factors identified in step
one.

They tested the pricing relationship, using factor analysis to identify significant factors from a database
of daily returns from 1962 to 1972, grouping 1,260 stocks into 42 portfolios. The initial analysis
suggested a maximum of five factors, with findings indicating at least three meaningful factors, but only
two consistently significant when estimating the risk-free rate.

Roll and Ross also examined cross-sectional consistency between groups, finding no significant
differences in intercept terms across portfolios. They concluded that while their evidence generally
supported the APT, the results were not conclusive.
Extensions of the Roll-Ross Tests:
Cho, Elton, and Gruber (1984)

They examined the APT and concluded that even with a two-factor model, two or three factors are
needed to explain returns, supporting APT's flexibility over CAPM.

Dhrymes, Friend, and Gultekin (1984)

They found no consistent relationship in factor loadings across different stock groups. The number of
required factors varied with group size, ranging from two to nine factors.

Roll and Ross (1984)

They acknowledged the variability in the number of risk factors with different sample sizes but
emphasized that consistent estimates are more important than the exact number of factors in validating
the APT.

Dhrymes, Friend, Gultekin, and Gultekin (1985)

They found that larger security groups resulted in more factors, although most were not statistically
significant. They noted that total standard deviation was as effective as factor loadings in predicting
returns, indicating instability in relationships.

Connor and Korajczyk (1993)

They critiqued traditional tests for priced risk factors, proposing a model that allows for correlated
unsystematic risk, identifying between one and six priced factors.

Harding (2008)

He explored the relationship between systematic and unsystematic risk factors, contributing to the
understanding of risk in asset pricing models.

The APT and Stock Market Anomalies

An alternative set of tests examines how well the APT explains pricing anomalies not accounted for by
the CAPM, notably the small-firm effect and the January effect.

APT Tests of the Small-Firm Effect:

Reinganum (1981) tested the APT's ability to explain the average return differences between small and
large firms, sorting stocks into 10 portfolios by market capitalization. He found that the average excess
returns were not zero for any of the tested factor models. Small-firm portfolios had positive excess
returns, while large-firm portfolios had negative returns, indicating a clear inverse relationship with size.
In contrast, Chen (1983) compared the APT to the CAPM using 180 stocks and five factors. He argued
that if the CAPM fails to capture all relevant information, the APT could explain the residual returns.
Chen concluded that the CAPM was misspecified and that the APT better explained the anomalies.

APT Tests of the January Effect:

Gultekin and Gultekin (1987) examined the January effect, where returns in January are notably higher
than in other months, and found that while risk premia were significant in January, they were rarely
priced in other months. They concluded that the APT does not explain this anomaly better than the
CAPM.

Burmeister and McElroy (1988) also found a significant January effect not captured by either model,
though they favored the APT over the CAPM when looking beyond January. Kramer (1994)
demonstrated that an empirical form of the APT could account for the January seasonal effect in stock
returns, unlike the CAPM.

Is the APT Even Testable?

Shanken (1982) questioned the empirical testability of the APT, similar to Roll's critique of the CAPM. He
noted that if stock returns do not conform to a K-factor model, it doesn't reject the model, while
conformity is seen as support. Additionally, different factor structures could yield varying implications
for expected returns, complicating the identification of factors explaining differential returns.

In response, Dybvig and Ross (1985) argued that the APT could be tested as an equality rather than
through Shanken's "empirical APT." Shanken (1985a) acknowledged that while equilibrium APT pricing
models are testable, the original arbitrage-based models are not.

9.2 Multifactor Models and Risk Estimation

The CAPM benefits from a clearly defined risk factor—excess return on the market portfolio—making
parameter estimation straightforward, although identifying the correct market proxy is challenging. In
contrast, the APT lacks specific risk factors, requiring ad hoc specifications, which complicates its
practical application. Early studies using statistical methods identified three to four significant factors,
but inconsistencies across samples limit the APT's usefulness. Recent research has attempted to refine
this approach, while an alternative method allows investors to directly specify the number and identity
of risk factors in multifactor models.

9.2.1 Multifactor model in practice:

Macroeconomic-Based Risk Factor Models:

Chen, Roll, and Ross (1986) proposed a model linking security returns to broad economic factors,
including industrial production growth, inflation changes, and shifts in bond credit spreads. Their
analysis of monthly returns from 1958 to 1984 revealed that the significance of these risk factors varied
over time, with inflation factors being relevant only between 1968 and 1977. Notably, the market
portfolio proxy was not significant. Burmeister, Roll, and Ross (1994) expanded this by identifying five
macroeconomic risk exposures—confidence risk, time horizon risk, inflation risk, business cycle risk, and
market-timing risk—using data through early 1992 to estimate risk premia for these factors.

Microeconomic-Based Risk Factor Models:

Microeconomic-based models specify risk using proxy variables that focus on characteristics of the
underlying securities. Fama and French (1993) developed a multifactor model with the following
equation:

(Ra - RFR) = a + b1(RM - RFR) + b2(SMB) + b3(HML) + e

Here, SMB (small minus big) captures the risk associated with firm size, while HML (high minus low)
distinguishes between value and growth stocks based on book-to-market ratios. Their analysis from July
1963 to December 1991 showed that the multifactor model significantly reduced the differences in beta
between low and high P/E stocks, suggesting that the market portfolio in a single-factor model captures
some but not all risk dimensions. Carhart (1997) extended this model by adding a momentum factor
(MOM) that accounts for the tendency of stocks with positive past returns to continue performing well,
further refining the understanding of risk factors in stock returns.

Extensions of Characteristic-Based Risk Factor Models:

This approach uses index portfolios, like the S&P 500, as proxies for systematic risk exposures. For
example, the Russell 1000 Growth Index targets large-cap stocks with low book-to-market ratios. Elton,
Gruber, and Blake (1996) employed several indexes in their factor model, while Ferson and Schadt
(1996) included public information variables like yield curves. MSCI Barra uses characteristic-based
variables and numerous industry indexes to analyze U.S. equities, helping to identify investment
exposures relative to benchmarks, revealing significant differences in factors such as firm size, leverage,
and momentum.

9.2.2 Estimating Risk in a Multifactor Setting: Examples:

To estimate expected returns for individual stocks: using a multifactor model, four steps are
required: (1) identify common risk factors, (2) estimate the risk premia for these factors, (3)
determine the stock's sensitivities to each factor, and (4) calculate the expected returns. For
example, using the Fama-French model, the common risk factors include the market excess
return (RMRFR), SMB (small vs. large stocks), HML (high vs. low book-to-market), and MOM
(momentum). Historical data can be used to estimate average risk premia over different time
frames, confirming trends like small and value stocks outperforming larger and growth stocks.
Factor sensitivities for Microsoft, CSX, and Xerox showed that both Microsoft and CSX behaved
like large-cap stocks with positive market correlations, while Xerox resembled a small-cap stock.
Additionally, Microsoft and CSX were identified as growth stocks with positive momentum,
unlike Xerox, which had negative momentum exposure.
Comparing Mutual Fund Risk Exposures:

This section analyzes the risk exposures of two mutual funds: Fidelity's Contrafund (FCNTX) and
T. Rowe Price's Mid-Cap Value Fund (TRMCX). FCNTX is classified as "large-cap growth," while
TRMCX is in the "mid-cap value" category, indicating different sensitivities to risk factors like
SMB (small vs. big) and HML (high vs. low book-to-market). Analysis of monthly returns from
July 2007 to June 2010 revealed that TRMCX has a higher beta, indicating greater market risk.
The multifactor model showed that FCNTX's market beta increased from 0.84 to 0.95 when
accounting for SMB and HML, highlighting how size and value factors impact risk. FCNTX has a
negative HML sensitivity, reflecting its growth orientation, while TRMCX's positive HML
sensitivity aligns with its value focus. FCNTX's negative SMB sensitivity indicates larger-cap
stocks, whereas TRMCX's positive SMB sensitivity suggests a tilt toward smaller firms,
confirming the appropriateness of their classifications.