Lecture 7

SIM and factor models

 Learning outcomes
• By the end of this lecture you should:
– Be able to exploit the mispricing of assets
– Understand the idea behind factor models,
especial the Fama-French-Carhart four-factor
model

2
 Introduction
• Last lecture we studied the CAPM, and saw
that under CAPM, all assets should be priced
by the formula:
E (r ) = r + β [E (r ) − r ]
i f i M f

• Today we’ll deal with cases where some assets

are mispriced according to CAPM
• We’ll also introduce some critiques that have
been raised against the CAPM
• We will introduce factor models in general

3
 Are all assets correctly priced by the CAPM?
• CAPM relies on several strong assumptions (see the discussion
in last lecture) that may not hold in reality. As such, CAPM is
not necessarily the correct asset pricing model.
• However, even if CAPM is the correct model for pricing all
assets, some assets could be mispriced
– In previous lectures, we assumed that all investors were rational,
which is surely at question.
– Some investors may subject to behavioral bias, and thus make
irrational investment decisions, which could affect asset prices.
– And there could be limits to arbitrage such as short sale constraints,
which allow for temporary deviation of an asset’s price from its
fundamental value.
– We will discuss both behavioral biases and limits to arbitrage in length
in the next lecture.

Mispricing 4
 The Stiglitz paradox
• Asset prices should reflect fundamentals to some extent
– But for that to happen, costly analysis is required to map information
about a business into valuation
• According to the CAPM, everybody should buy the market
portfolio, so nobody has any incentive to do the analysis
– If everybody is a passive CAPM investor, there is no channel for
information about the real economy to enter the capital market
• Therefore, we usually think that prices may deviate enough
from their fundamental values for an analyst to achieve
enough excess returns to reward them for their work
– So mispricing is absolute, just by how much
• The CAPM still gives guidance on how to exploit such
mispricing

Mispricing 5
 The Single Index Model
• Theoretically, the market portfolio should include all risky
assets, for example, individuals’ personal human capital.
• Since we cannot observe the true market portfolio, we try to
pick something that seems kind of close when using CAPM in
practice
– A common choice is a broad value-weighted market index, like the
S&P 500 for the U.S. or ASX 200 for Australia
• To emphasize that we are not dealing with the true market
portfolio, we sometimes refer to the resulting empirical model
as the Single Index Model, or the SIM

Mispricing 6
 The Single Index Model
• We saw last time how it was convenient to think of CAPM in a
regression model:
[ ]
rit = rft + β i rMt − rft + ε it
– We are interested in variations in both the cross section and over
time, which is why we have added time indices

• To evaluate an asset’s returns based on CAPM, we typically

regress its excess returns against the market excess returns:
[ ]
rit − rft = α i + β i rMt − rft + ε it

Mispricing 7
 Jensen’s α
rit − rft = αi + β i [rMt − rft ] + ε it
• According to the CAPM the αi should be zero
• If it is not, asset i is mispriced according to the CAPM and αi is
a measure of this mispricing

• We refer to the α as Jensen’s alpha or simply as the alpha

– In practice, these α vary predictably across assets
– Often a non-zero α indicates that there’s something wrong with our
model rather than with the market price
– But for our discussion of mispricing in this lecture, let’s assume that
the model is correct and a non-zero α of an asset indeed indicates
mispricing of that asset

Mispricing 8
 Mispriced assets in E(r)-β space

• If an asset is correctly priced by CAPM, it should be plotted on

the Security Market Line (SML)
– SML plots all combinations of E(r) and β with E(r) – rf = β*(E(rM) – rf )
– In the graph below, assets A and C are mispriced, i.e. have a non-zero
α

C
E (ri ) αC > 0
SML
M
B
αA < 0

rf A

β
Mispricing 9
 Exploiting one mispriced asset
• If we identify assets that are mispriced, like those in the
graph, we want to deviate from the market weights to exploit
the mispricing, α
– This means selecting an optimal risky portfolio, P*, that is not equal to
the market portfolio, M
– By doing so, we can increase expected return of our portfolio without
similar increase in volatility of portfolio returns
– As such, we can improve our utility.

Exploiting mispricing – one asset 10

 Exploiting one mispriced asset
• More specifically, we try to find an optimal risky portfolio (of
the market portfolio and the mispriced asset) to have the
highest Sharpe ratio
– Recall the process of finding optimal risky portfolio in Lecture 5
– There, the optimal risky portfolio with the highest Sharpe ratio turns
out to be the market portfolio, when all assets are correctly priced
– This is no longer the case when there are mispriced assets
• In formal terms, we combine some active position in the
mispriced asset, wA, with a passive position in the market
portfolio, (1 – wA), to form our new optimal risky portfolio
with the highest Sharpe ratio:
E [rP (wA )] − rf
max S P =
wA σ P (wA )
,where rP = wA rA + (1 − wA )rM

Exploiting mispricing – one asset 11

 Exploiting one mispriced asset
• When we take an active position we exploit the α, but we also
incur some unsystematic and hence unpriced risk (sometimes
also referred as residual risk)
[ ]
rAt − rft = αA + β A rMt − rft + ε At
• When solving the maximization problem we weigh these
effects against each other
• We can think of this as if we were “buying” the benefit of
mispricing at the cost of taking on unsystematic risk
– The price at which we can “buy mispricing” is αA/σ2ϵA
– This is in the same spirit of “buying” excess return by taking on risk
when selecting the optimal risky portfolio (reward-to-risk ratio)
• This ratio turns out to be very important when deciding how
much to tilt our portfolio towards the mispriced asset

Exploiting mispricing – one asset 12

 Exploiting one mispriced asset
• It turns out that the optimal weight in the mispriced asset is
w0A
w =
*

1 + w0A (1 − β A )
A

αA E (rM ) − rf
where, w0A =
σ ε2A σ M2

• The active weight depends essentially on w0

– Positive active weight, i.e., buying the mispriced asset, when α is
positive, and sell if α is negative
– The bigger the magnitude of w0, the more extra return per (residual)
risk unit we get from our active asset (relative to the return per risk
unit we get in the market portfolio) – so we should increase the weight
in the active asset
• β of the active asset also plays a role here, but I won’t require
you to discuss its relationship with the optimal weight.

Exploiting mispricing – one asset 13

 Exploiting one mispriced asset

• In effect, the mispriced asset allows us to construct a “new

efficient frontier” of risky assets, with a new associated
optimal risky portfolio and a new associated CAL

CALP*
E (ri )
P* CALM
A
M

rf
σ
Exploiting mispricing – one asset 14
 Numerical example
• Suppose you (or your research department)
has estimated the following data:
E ( rM ) = 10% β A = 1.2
σ M = 25% E ( rA ) = 13%
rf = 2% σ A = 40%

• Is asset A mispriced, and if so, how do we

exploit it?

Exploiting mispricing – one asset 15

 Numerical example
• Step 1: Calculate the α (“the benefit”)
E ( rA ) − rf = α A + β  E ( rM ) − rf  ⇔ α A = E ( rA ) − rf − β  E ( rM ) − rf 
α A = 0.13 − 0.02 − 1.2 ( 0.1 − 0.02 ) = 1.4%

– Since α is not zero, the asset is mispriced

(according to the CAPM)
• Step 2: Calculate the unsytematic risk (“the
cost”)
σ A2 = β 2σ M2 + σ ε2 ⇔ σ ε2 = σ A2 − β 2σ M2
σ ε2 =−
0.42 1.22 0.252 =
0.07

Exploiting mispricing – one asset 16

 Numerical example
• Step 3: Get the optimal weight of the active portfolio
αA E ( rM ) − rf 0.014 0.1 − 0.02
=wA0 = = 0.15625
σ ε2( A) σ M2 0.07 0.252
wA0 0.15625
=w = *
≈ 0.16129
1 + wA0 (1 − β A ) 1 + 0.15625 (1 − 1.2 )
A

• Invest 16.13% of your risky investment in the mispriced

asset and the remaining 83.87% in the “market
portfolio”
• Step 4: Calculate the amount to invest in that risky
portfolio and how much to invest in the risk-free asset
– This depends on the reward to risk ratio of the new
optimal risky portfolio and the risk aversion coefficient

Exploiting mispricing – one asset 17

 Gains from exploiting one mispriced asset
• It turns out that there is a clear relationship between an
asset’s information ratio, IR, (or appraisal ratio) and the
resulting improvement in our Sharpe ratio. The IR is
α
IR =
σε
• The squared Sharpe ratio of the optimal risky portfolio, after
we have incorporated the mispriced assets, is
2
α 
S = S + 
2 2

σε
P M


Exploiting mispricing – one asset 18

 Exploiting multiple mispriced assets
• What if there are multiple mispriced assets?
– First we combine them to form an active portfolio (AP)
– Then we treat the active portfolio as a single mispriced asset, and
follow the same procedure discussed previously to form optimal risky
portfolio comprising of the market portfolio and this active portfolio

• Forming the (optimal) active portfolio of mispriced assets

– Each mispriced asset is given an active portfolio weight proportional to
its alpha to unsystematic risk ratio α σ ε2
– So the weight of the ith mispriced asset in the active portfolio would
be
α i σ ε2i
wi = N

∑α
j =1
j σ ε2 j

Exploiting mispricing – multiple assets 19

 Exploiting multiple mispriced assets
• If we form the active portfolio above, it turns out that the
squared information ratio of the active portfolio equals to the
sum of squared information ratio of individual active assets:
– IR2AP = ∑jIR2j

• And the squared Sharpe ratio of the optimal risky portfolio of

the market portfolio and the active portfolio is:
– S2P = S2M + IR2AP = S2M + ∑jIR2j

Exploiting mispricing – multiple assets 20

 Should assets be priced by the CAPM?
• CAPM relies on several strong assumptions, which we
discussed in the last lecture. Among them, probably most
critical are:
– Assumption 1: Investors are rational mean-variance optimizers - All
investors maximize a utility function and do so correctly
– Assumption 5: All investors can trade all assets (which is not the case
in the reality, for example, human capital is typically viewed as
untradeable)
• These assumptions easily fail in reality, and so does CAPM
– Earlier empirical studies in 1970s supported CAPM
– More recent ones did not
– Messing up the debate even more, practically we are not able to test
CAPM at all, as we don’t have a valid proxy for the market portfolio of
all risky assets (including those not traded)

Factor models 21
 Should assets be priced by the CAPM?
• The exact theoretical prediction of CAPM may be wrong
– An asset’s expected return may not be determined only by its beta
which is in turn determined by the covariance of its return with
market returns

• But the basic principles of optimal portfolio theory and CAPM

should hold
– Investors should hold diversified portfolios to reduce the risk of their
investments
– As such, the relevant risks to investors are those systematic ones that
are not diversifiable
– And assets’ expected returns should be determined by their
systematic risks

Factor models 22
 Factor models
• The CAPM expresses the expected return of any asset as a
function of its covariance with the market portfolio
• We sometimes refer to the market portfolio as a risk factor,
and to the CAPM β as a loading on that factor
• In principle, we could imagine other risk factors with which
assets covary (or “load”) and gain other required return as a
result
• The theoretical arguments underpinning this interpretation
differs from our CAPM arguments, but the interpretations of
the resulting models are very similar
– Most empirical factor models are not derived formally with clearly
specified assumptions and investors’ objective functions
– The theoretical foundation for such models, if any, is typically
Arbitrage Pricing Theory proposed by Stephen Ross

Factor models 23
 Factor models
• A general factor model expresses the return of an asset as
follows:
~ ~
ri = λ0 + β i ,1 f1 + ... + β i ,k f k + ε~i
~

where, ~
fj is jth systematic factor, and β i, j is the loading of asset i on the jth
factor

• And the asset’s expected return should equal to:

E (ri ) = λ0 + β i ,1λ1 + ... + β i ,k λk

where, λ j is the risk premium of the jth factor

Factor models 24
 Fama-French-Carhart four-factor model
• We typically express each factor as some appropriate portfolio
• The factor model most commonly used in practice is Fama-
French-Carhart four-factor model that include four portfolio
returns as factors: market portfolio, small-minus-big (size)
portfolio (SMB), high-minus-low (book-to-market ratio)
portfolio (HML), momentum (winner minus loser) portfolio
(MMT or UMD)
[ ]
E (ri ) = r f + β iM E (rM ) − rf + β iSMB E (rSMB ) + β iHML E (rHML ) + β iMOM E (rMOM )

– Fama and French first developed the three-factor model

– Carhart expanded with the fourth factor, MMT

Factor models 25
 Fama-French-Carhart four-factor model
• Each new factor portfolio is self financing, i.e. it takes long
(buy) and short (sell) positions that offset each other so that
the net portfolio cost is zero
– The Small-Minus-Big portfolio, SMB, consists of a long position in the
market’s 50 % smallest firms by market cap (small) and a short
position in the market’s 50 % largest firms (big)
– The High-Minus-Low portfolio, HML, consists of a long position in the
30 % of the market with the highest Book-to-Market ratios (value
stocks) and a short position in the 30 % of the market with the lowest
Book-to-Market ratios (growth stocks)
– The Momentum portfolio, MOM, takes a long position in the 30 % of
the market that had the highest returns the preceding year (winners)
and a short position in the 30 % of the market that had the lowest
returns the preceding year (losers)
– The loading of any particular asset to each factor depends on its
covariance with the factor

Factor models 26
 Factor models in practice
• It turns out that an asset’s loadings on these factors tend to
predict expected returns, in violation of the CAPM
– The four-factor model is able to explain a big variation of asset returns
in the cross section
• This is an empirical result - we don’t know why these
portfolios should require higher expected returns
– We often hesitate to call these factors risk factors, as it is unclear what
risks are associated with them
– Despite its theoretical shortcomings, the four-factor model is often
used in practice instead of the CAPM due to its ability to summarize
expected returns of assets
– There is nothing to stop a new popular factor model to emerge, as
long as it is able to predict expected returns of assets well.
– In fact, Fama and French recently proposed a five-factor model

Factor models 27