Because learning changes everything.

Lecture 2
Arbitrage Pricing Theory, Multifactor
Models, and Empirical Evidence

© McGraw Hill LLC. All rights reserved. No reproduction or distribution without the prior written consent of McGraw Hill LLC.
 Overview: APT and Multi-Factor Models

Arbitrage is the exploitation of security mispricing in such a

way that risk-free profits can be earned.
• Most basic principle of capital market theory is that well-
functioning security markets rule out arbitrage
opportunities.
Generalization of the security market line of the CAPM to
gain richer insight into the risk–return relationship.
• Arbitrage pricing theory (APT).

© McGraw Hill 2
 Factor Models of Security Returns 1

Single-factor model of excess returns

Ri = E ( Ri ) + βi F + ei

E ( Ri ) = expected excess return on stock i

βi = sensitivity of firm i

F = deviation of the common factor from its expected value

ei = nonsystematic components of returns

Betas are sometimes called factor loadings or factor

betas.

© McGraw Hill 3
 Arbitrage Pricing Theory
Arbitrage pricing theory (APT) was developed by Stephen
Ross.
Predicts a SML linking expected returns to risk, but the path
is takes to the SML is quite different.
APT relies on three key assumptions:
1. Security returns can be described by a factor model.
2. There are sufficient securities to diversify away
idiosyncratic risk.
3. Well-functioning security markets do not allow for the
persistence of arbitrage opportunities.

© McGraw Hill 4
 Arbitrage, Risk Arbitrage, and Equilibrium

Arbitrage opportunity exists when an investor can earn

riskless profits without making a net investment.
• For example, shares of a stock sell for different prices on
two different exchanges.
Law of One Price.
• Enforced by arbitrageurs; If they observe a violation, they
will engage in arbitrage activity.
• This bids up (down) the price where it is low (high) until the
arbitrage opportunity is eliminated.

© McGraw Hill 5
 Well-Diversified Portfolios
In a well-diversified portfolio, firm-specific (nonsystematic) risk becomes
negligible, so that only factor risk remains.
• When n is large, nonsystematic variance → zero
• Consider an equally weighted portfolio of n securities:

1 n σ ( ei ) 1 −2
2 2
1 2
σ ( e p ) =  w σ ( ei ) =    σ ( ei ) = 
n n
2 2
i
2
= σ ( ei )
i =1 i =1  n  n i =1 n n

where:

σ 2 ( e p ) = Nonsystematic Variance.

σ −2 ( ei ) = Average nonsystematic variance

A well-diversified portfolio is one with each weight small enough that for
practical purposes the nonsystematic variance is negligible.

© McGraw Hill 6
 Figure 10.2 Excess Returns as a Function of
the Systematic Factor

Figure 10.2 Excess returns as a function of the systematic factor: Panel A, Well-
diversified portfolio A; Panel B, Single stock (S).

Access the text alternative for slide images.

© McGraw Hill 7
 Figure 10.3 Returns as a Function of the
Systematic Factor: An Arbitrage Opportunity

Access the text alternative for slide images.

© McGraw Hill 8
 Figure 10.4 An Arbitrage Opportunity

Access the text alternative for slide images.

© McGraw Hill 9
 The SML of the APT
All well-diversified portfolio with the same beta must have the
same expected return (Figure 10.3.
And if two well-diversified portfolios have different betas, their
risk premia must be proportional to beta (Figure 10.4):
𝐸 𝑅𝑃 = 𝛽𝑃,𝐹 × 𝑘
for some constant k.

Assume some market index portfolio is well-diversified, then

𝐸 𝑅𝑀 = 𝛽𝑀,𝐹 × 𝑘
Combining the two equations, we get an SML for any well-
diversified portfolio P:
𝐸 𝑅𝑃 = 𝛽𝑃,𝑀 𝐸 𝑅𝑀 , where 𝛽𝑃,𝑀 = 𝛽𝑃,𝐹 /𝛽𝑀,𝐹 .

© McGraw Hill 10
 APT and CAPM
APT CAPM
Built on the foundation of Model is based on an
well-diversified portfolios. inherently unobservable
• Cannot rule out a violation “market” portfolio.
of the expected return–beta Provides unequivocal
relationship for any statement on the expected
particular asset. return–beta relationship for
all securities.
Does not assume investors
are mean–variance
optimizers.
Uses an observable market
index.
© McGraw Hill 11
 A Multifactor APT 1

To this point, we have examined the APT in a one-factor

world.
• However, there are several sources of systematic risk and
exposure to these factors will affect a stock’s appropriate
expected return.
• APT can be generalized to accommodate these multiple
sources of risk.
• As an example, we could consider a two-factor model:
Ri = E ( Ri ) + βi1 F1 + βi 2 F2 + ei

© McGraw Hill 12
 A Multifactor APT 2

Benchmark portfolios in the APT are factor portfolios.

β = 1 for one of the factors and β = 0 for any other factors
Factor portfolios track a particular source of macroeconomic
risk, but are uncorrelated with other sources of risk
• Referred to as a “tracking portfolio”
Multifactor SML predicts that a well-diversified portfolio’s risk
premium is equal to:

𝐸 𝑅𝑃 = 𝛽𝐹1 𝐸 𝑅𝐹1 + 𝛽𝐹2 𝐸 𝑅𝐹2

© McGraw Hill 13
 Fama-French (FF) Three-Factor Model

Rit = αi + βiM RMt + βiSMB SMBt + βiHML HMLt + eit

SMB = Small minus big (the return of a portfolio of small

stocks in excess of the return on a portfolio of large
stocks)
HML = High minus low (the return of a portfolio of stocks with
a high book-to-market ratio in excess of the return on
a portfolio of stocks with a low book-to-market ratio)

© McGraw Hill 14
 Estimating and Implementing a Three-Factor
SML
Begin by estimating U.S. Steel’s beta on each of the FF
factors.

U.S. Steel’s size is in the 50th to 60th percentile.

But its book-to-market ratio, about 1.5, is quite high.

© McGraw Hill 15
 Table 10.1 Estimates of Single-Index and
Fama-French Regressions
Single-index model Three-factor model

Regression Regression
Coefficient t-Statistic Coefficient t-Statistic
Intercept (alpha,
In the following %)
table, read −2.034 −0.991 −0.745 −0.363
'rm - rt' as r sub m minus r
rm − rt
sub t
2.030 4.785 1.771 4.074

S MB 0.369 0.487

H ML 1.249 2.303

R-square 0.283 0.352

Residual std dev (%) 15.171 14.675

Table 10.1
Estimates of single-index and three-factor Fama-French regressions for U.S.
Steel, monthly data, 5 years ending in 2021.
Source: Authors' calculations.

© McGraw Hill 16
 Testing Factor Models

© McGraw Hill 17
 The Index Model and the Single-Factor SML

Expected Return–Beta Relationship.

E ( ri ) = rf + βi  E ( rM ) − rf 

Cov ( ri , rM )
with βi =
σ 2M

Estimating the security characteristic line (SCL)

rit − rft = αi + bi ( rMt − rft ) + eit

© McGraw Hill 18
 Estimating the SCL
Tests of the expected return beta relationship.
First-pass regression
• Time series regression to estimate the betas and average
returns of each security or market portfolio.

Second-pass regression
• Cross-sectional regression of average returns on
estimated betas (we denote with b): 𝑟𝑖 − 𝑟𝑓 = 𝛾0 + 𝛾1 𝑏𝑖
• If CAPM holds, then 𝛾0 should be 0, and 𝛾1 should be
𝑟𝑀 − 𝑟𝑓

© McGraw Hill 19
 Tests of the CAPM
Early tests performed by Lintner and later replicated by Miller
and Scholes.
Results are inconsistent with the CAPM
• SML is “too flat” and intercept is “too large”

Difficulties with approach employed.

• Hard to measure entire market (Roll’s critique)
• Betas are estimated with error

© McGraw Hill 20
 The Market Index: Roll’s Critique
1. Single testable hypothesis associated with C APM: market
portfolio is mean–variance efficient.
2. All other implications of the model (such as SML) are not
independently testable.
3. Theory is not testable unless all individual assets are
included in sample.
4. Market proxy (also called a benchmark), such as S&P
500, may be mean-variance efficient while market may
not be. (This problem sometimes called benchmark
error).

© McGraw Hill 21
 Measurement Error in Beta
Problem – If beta is measured with error and appears as a
right-hand-side variable in the second-pass regression:
• Slope coefficient of the regression equation will be biased
downward.
• Intercept biased upward.

Solution – Construct portfolios rather than securities, gives

more precise beta estimates.
• Fama McBeth: Rank securities by beta, then group in
portfolios of different average beta.
𝑟𝑖 − 𝑟𝑓 = 𝛾0 + 𝛾1 𝑏𝑖 + 𝛾2 𝑏𝑖 2 + 𝛾3 𝜎(𝑒𝑖 )

© McGraw Hill 22
 Table 13.1 Summary of Fama–MacBeth Study

Table 13.1
Risk and return in the Fama–MacBeth study. All rates are in basis points per
month. Sample period was 1935 to June 1968.

Average  0 Average  1 Average  2 Average  3

(t-static for (t-static for (t-static for (t-static for
test of  0 = 0) test of  1 = rM − rf ) test of  2 = 0) test of  3 = 0)

20 114 −226 516

(0.55) (1.85) ( − 0.86) (1.11)

Source: Fama, E. F., & MacBeth, J. D. (1973, May-June). Risk return, and
equilibrium: Empirical tests. Journal of Political Economy, 81, 607–636.

© McGraw Hill 23
 Summary of CAPM Tests
Fama-MacBeth verify the following:
1. Observed relationship between average excess returns
and beta is indeed linear.
2. Nonsystematic risk does not explain average excess
returns.

But more recent tests have been less kind to the CAPM:
• Fama and French (1992) find the coefficient on beta is too
small and not even statistically significant.

© McGraw Hill 24
 Fama-French-Type Factor Models
High book to market firms experience higher returns (HML)
Smaller firms experience higher returns (SMB)
Size and value are priced risk factors, consistent with APT

E ( ri ) = rf + ai + bi  E ( rM ) − rf  + si E  RSMB  + hi E  RHML 

where bi , si , hi are betas (loadings)

© McGraw Hill 25
 Figure 13.1 CAPM Versus Fama and French
Model

Figure 13.1 CAPM versus the Fama and French model. The figure plots
the average actual returns versus returns predicted by CAPM and the FF
model for 25 size and book-to-market double-sorted portfolios.

Access the text alternative for slide images.

© McGraw Hill Source: Amit Goyal, “Empirical Cross Sectional Asset Pricing: A Survey,” Financial Markets and Portfolio Management 26 (2012), pp. 3–38. 26
 Risk-based interpretations: SMB

• Fama and French argue that small firms may be more

likely to collapse during a downturn
• as if their beta on the market were enhanced in bad times.
• Risk premium on SMB compensates investors for this
enhanced risk.

© McGraw Hill 27
 Risk-based interpretations: HML
Petkova and Zhang
Provide evidence of a countercyclical value beta
• Beta of value stocks (high BtM) is greater than that of
growth stocks (low BtM) during recession, and vice versa

Zhang
Explains this with irreversible investment
• notes that value firms (high BtM ratios) on average have
more tangible capital, which is hard to sell off in a
downturn. Beta of value stocks is greater than that of
growth stocks during recession.
• Growth firms (low BtM), can instead simply defer
investment plans, so are not as harmed in bad times.
© McGraw Hill 28
 Behavioral Explanations
Glamour firms
• Recent good performance.
• High prices.
• Lower book-to-market ratios.

Overreaction
• High past growth is extrapolated and then impounded in
price.

Extrapolation error
• Market ignores evidence that past growth cannot be
extrapolated far into the future.

© McGraw Hill 29
 Figure 13.3 Book-to-Market Reflects Past
Growth

Figure 13.3 The book-to-market ratio reflects past growth, but not future growth prospects.
B/M tends to fall with income growth experienced at the end of a five-year period; but
actually increases slightly with future income growth rates.
Source: Chan, L. K. C., Karceski, J., & Lakonishok, J. (2003, April). The level and persistence of growth
rates. Journal of Finance, 58, 643–684.
Access the text alternative for slide images.

© McGraw Hill 30
 Momentum: A Fourth Factor
Fourth factor has come to be added to the standard controls
for stock return behavior since original FF model.
• Momentum factor evaluates abnormal performance of a
stock portfolio.

Winners minus losers (WML) – winners/losers based on past

returns.
Seems to work well, but hard to interpret economically.

© McGraw Hill 31
 Table 13.4 The Cross Section of Expected
Stock Returns
Slope t-Statistic
Size: Stock market capitalization −.15 −5.01
Book-to-market ratio .35 6.18
Momentum: Return in past year 96 6.86
Stock issues: Growth in shares outstanding −.35 −3.52
Accruals: Change in net working capital −1.38 −5.69
Profitability-Return on assets 1.43 3.57
Asset growth: Growth in total assets −.54 −4.49
Dividend yield −.46 −.27
Beta: Single index model .33 3.05
Volatility: Standard deviation of stock returns −1.45 −3.48
Turnover: Shares traded as a fraction of outstanding shares −4.49 −3.68
Sales: Total sales as multiple of market capitalization .04 3.10

Source: Jonathan Lewellen, J. (2015). The cross-section of expected stock returns. Critical Finance
Review, 4, 1–44.

© McGraw Hill 32
© Imperial College Business School

Several paper have questioned the

economic and statistical significance
of the zoo of factors
• Harvey, Liu, and Zhu (2016): 296 anomalies, 27% to 53% are likely to
be false discoveries

• Hou, Xue, and Zhang (2018): 452 anomalies, 82% turn insignificant upon
excluding microcaps + value-weighting

• Anomalies attenuate, and often disappear, in recent years (Chordia,

Subrahmanyam, and Tong, 2014)

33

© McGraw Hill
 Famous paper by Harvey et al (2016): t-
statistics need to be adjusted upwards due
to data mining

Source: … and the Cross-Section of Expected Returns by Campbell R. Harvey, Yan Liu, Heqing Zhu
The Review of Financial Studies, Volume 29, Issue 1, January 2016, Pages 5–68,
https://doi.org/10.1093/rfs/hhv059
34
© Imperial College Business School
© McGraw Hill
 Factor Zoo and Machine Learning

Another response to the questionable factor zoo has been to apply

new methods to the same data to increase statistical significance by
• addressing the high-dimension of noisy and correlated predictors
• utilizing flexible, possibly non-linear, functional forms
• implementing model selection
• mitigating overfitting biases through regularization

Machine Learning:
• automated detection of complex patterns in data
• combine multiple, possibly weak, sources of information into a
meaningful composite signal
• See for example, Gu, Kelly and Xiu (2020, RFS)

© McGraw Hill
 GKX( 2020): Performance of ML
portfolios

Source: Empirical Asset Pricing Via Machine Learning, Gu, Kelly and Xiu (RFS, 2020); available
here https://dachxiu.chicagobooth.edu/download/ML.pdf
© McGraw Hill