Factor Models

• Idea: asset returns are driven by common economic factors as well as firm-specific
information.
• Examples for common factors: business cycle, interest rates, costs of resources.
• Unexpected changes in common factors affect all or at least many firms
simultaneously.

• Factor models decompose the return of an asset into:

• Its expected return.
• A component due to unexpected changes in common factors.
• A component due to unexpected firm-specific information.

3 Investments
Single-Factor Model

• In a single-factor model, macroeconomic uncertainty is captured by a single factor F:

ri = E (ri ) + β i F + ei

• F reflects new/unanticipated information about the economy.

• βi captures the sensitivity of firm i with respect to unexpected changes in F.
• F and ei are uncorrelated and have mean zero.
• ei captures firm−specific uncertainty and are uncorrelated across firms.

σ i2 = β i2σ F2 + σ 2 (ei )
• Variance of ri:
Systematic Idiosyncratic
risk risk

• Covariance between two assets i and j:

Cov (ri , r j ) = Cov (E (ri ) + β i F + ei , E (r j ) + β j F + e j ) (E(ri) and E(rj) are constant)

= Cov ( β i F , β j F ) + Cov ( β i F , e j ) + Cov (ei , β j F ) + Cov (ei , e j )

= β i β j σ F2 =0 =0 =0

4 Investments
Multifactor Models

• Multifactor models allow a more detailed description of returns and allow different stocks
to have different sensitivities to numerous common factors Fj:
k
ri = E (ri ) + ∑ β ij Fj + ei
j =1

• Fj reflect new/unanticipated information about the economy.

• βij captures the sensitivity of firm i with respect to unexpected changes in Fj (factor
loadings or factor betas).
• Fj and ei are uncorrelated and have mean zero.
• ei captures firm−specific uncertainty and are uncorrelated across firms.
k k
• Variance of ri: σ = ∑∑ β ij β il Cov (Fj , Fl ) + σ 2 (ei )
i
2

l =1 j =1

5 Investments
Multifactor Models: Example

• Suppose the two most important macroeconomic sources of risk are uncertainties about
• the state of the business cycle (measured by unanticipated growth in GDP), and
• changes in interest rates.
The return on any stock will respond to both sources of macroeconomic risk as well
as to its own firm-specific influences.

• We can write a two-factor model describing the excess return on stock i in some time
period as follows:
ri = E (ri ) + β i ,GDPGDP + β i ,IR IR + ei

• GDP measures any unexpected changes in GDP growth.

• IR measures any unexpected changes in interest rates.

• Consider two firms:

• Firm A: a regulated residential electric-power utility provider.
• Firm B: an airline.

6 Investments
Multifactor Models: Example (cont.)

• Sensitivities to GDP growth:

• Residential demand for electricity is not very sensitive to the business cycle.
The utility provider has a low beta on unexpected changes in GDP.
• The performance of the airline is very sensitive to economic activity.
The airline has a high beta on unexpected changes in GDP.
• Sensitivities to interest rate changes:
• Cash flow generated by the utility is relatively stable, wherefore its present value
behaves much like that of a bond.
The utility provider has a high beta on unexpected changes in interest rates.
• The performance of the airline is less sensitive to interest rates.
The airline has a low beta on unexpected changes in interest rates.
• Suppose that on a particular day, a news item suggests that the economy will expand.
GDP is expected to increase, but so are interest rates.
• For the utility provider, this is bad news as its dominant sensitivity is to interest rates.
• For the airline, which responds more to GDP, this is good news.
A one-factor or single-index model cannot capture such differential responses to
varying sources of macroeconomic uncertainty.

7 Investments
Application to Portfolio Optimization (1/2)

• Mean-Variance portfolio optimization: minimization of the variance of a portfolio for a

given expected portfolio return
n n n
min σ = ∑∑ w i w j Cov (ri , r j )
2
p E (rp ) = ∑ w i E (ri )
i =1 j =1 i =1

• Problem of traditional (“brute force”) portfolio optimization:

• Large number of covariance estimates required.
• Estimation errors lead to suboptimal portfolio weights.
• Example: in case of n = 100 assets, n = 100 variance estimates and (n*n-n)/2 =
4,950 covariance estimates are required.

8 Investments
Application to Portfolio Optimization (2/2)

• Factor Models: simplify the estimation of covariances and thereby facilitate the selection
of optimal portfolios
Variance of a portfolio with n assets with weights 𝑤𝑤𝑖𝑖 in terms of k factors:
k k n n
σ = ∑∑ β pj β pl Cov (Fj , Fl ) + ∑ w σ (ei )
2
p
2
i
2
β pj = ∑ w i β ij
l =1 j =1 i =1 i =1

Only the assets’ sensitivities to the common factors, the covariances of the
common factors, and the assets’ residual variances are needed to estimate the
portfolio’s risk.
Dimensionality reduction leads to less noisy and more robust measurement of
portfolio risk.
Example: in case of n = 100 assets and k = 10 factors, n = 100 residual variance
estimates, n*k = 1,000 sensitivity estimates, (k*k-k)/2 = 45 factor covariance
estimates, and k = 10 factor variance estimates are required.

9 Investments
The Barra US Equity Model

• Rosenberg (1974) was the first to develop multi-factor risk models in order to estimate
the asset covariance matrix.

• Rosenberg founded Barra, which made widespread use of multi-factor risk models and
dedicated itself to helping practitioners implement the theoretical insights of Markowitz.

• The first multi-factor risk model for the US market, dubbed the Barra USE1 Model, was
released in 1975 and subsequently refined by the USE2 Model in 1985 and the USE3
Model in 1997.

• The most recent Barra equity risk model is the USE4 Model from 2011. It contains
• one country factor (essentially the market portfolio of the respective country)
• 60 industry factors
• 12 style factors

• MSCI applies the Barra model for the construction of their minimum volatility indexes.

10 Investments
The Arbitrage Pricing Theory (APT)

• So far, multifactor models offer just a description of the factors that drive returns.
• The question how assets’ expected returns are determined is not yet addressed.

• The APT links, like the CAPM, expected returns to systematic risk.

• The APT relies on three key propositions:

1. Security returns can be described by a factor model.
2. There is a sufficient number of securities to diversify away idiosyncratic risk.
3. Security markets do not allow for the persistence of arbitrage opportunities.

• Arbitrage opportunity:
• Investor can earn riskless profits
• Without making a net investment

• Arbitrage opportunities are eliminated by arbitrageurs (they ensure the law of one price).

12 Investments
Well-Diversified Portfolios

• Consider the return of a stock portfolio P in a multifactor model:

k
rp = E (rp ) + ∑ β pj Fj + ep
j =1

• The idiosyncratic risk of the portfolio is given by:

n
σ (ep ) = ∑ w i2σ 2 (ei )
2

i =1
As the number of stocks in the portfolio grows, the weights wi of the individual
stocks shrink, and the idiosyncratic variance goes to zero (diversification).
Thus, the firm-specific risk becomes negligible and only systematic factor risk
remains.

• Since E ep = 0 and σ2(ep) → 0, any realized value of ep will be virtually zero. Therefore,
the return of a well-diversified portfolio approximately satisfies:
k
rp = E (rp ) + ∑ β pj Fj
j =1

13 Investments
Arbitrage (1/2)

• Suppose that the impacts of the common factors that drive returns can be summarized
by the market factor (i.e. by the return on the market portfolio).

• Thus, the return on the well-diversified portfolio P can be described by the single-index
model:
rp = E ( rp ) + β p m E (rp ) = α p + rf + β p (E (rm ) − rf )

m is the demeaned excess return on the market.

• Since neither the market portfolio nor portfolio P have residual risk, the only risk to the
returns of the two portfolios is the systematic risk due to their exposure to the common
factor (the market factor).

• Consequently, it is possible form a riskless portfolio Z (i.e., a portfolio with a beta of

zero) as a long-short combination of the market portfolio and portfolio P:
β z = w p β p + (1 − w p )β m = 0
1
Since βm = 1, we get wp =
1−βp

14 Investments
Arbitrage (2/2)

• The expected return on the zero-beta portfolio Z is:

1
E (rz ) = w pE (rp ) + (1 − w p )E (rm ) = rf + αp
1− βp
• As Z is a riskless portfolio, its expected return has to be equal to the risk-free rate rf in
equilibrium (i.e. αp = 0)
• If E(rz) > rf: one could make a riskless profit by borrowing at the risk-free rate at
investing in portfolio Z.
• If E(rz) < rf: one could make a riskless profit by going short portfolio Z and investing
the proceeds at the risk-free.
• Investors will engage in this strategy until the price of portfolio Z (resp. of portfolio
P) adjusts such that this arbitrage opportunity disappears.

• Thus, in market equilibrium, αp = 0 and the expected return on the well-diversified

portfolio P is
E (rp ) = rf + β p (E (rm ) − rf )

When the market factor is the only factor, the APT implies the same expected
return-beta relationship as the CAPM.

15 Investments
Generalization to Multiple Factors

• Describing returns by a single-factor model is likely to be a oversimplification:

• Intuitively, returns might be driven by multiple common factors.
• Thus, the return on a well-diversified portfolio P might be described by a multifactor
model and the portfolio’s systematic risk is determined by its exposures to these
common factors

• Applying the concept of arbitrage analogously to a multifactor model yields the following
expected return-beta relationship on the well-diversified portfolio P:
k
E (rp ) = rf + ∑ β pj γ j
j =1

In equilibrium, the expected return of the well-diversified portfolio depends on its

exposures βpj to the k common risk factors.
For the exposure to these common factors, investors are compensated by the risk
premia γj that are associated with the common factors.
The APT implies a multifactor version of the security market line in which expected
returns depend on exposure to multiple risk sources, each with their own risk
premium.

16 Investments
Factor Risk Premia

• Contrary to the special case of the market factor, for which the risk premium is simply
the excess return on the market portfolio over the risk-free rate, the risk premia
associated with general common factors are not directly observable.

• Factor Portfolios: well-diversified portfolios constructed to have a beta of 1 on one of the

common factors and betas of zero on all other common factors.
• The returns on factor portfolios track the evolution of the particular source of risk but
are uncorrelated with other sources of risk.
• Thus, the returns on the factor portfolios represent the risk premia that are
associated with the common factors.

17 Investments
Non-well-diversified Portfolios

• The APT holds exactly only for well-diversified portfolios (i.e., portfolios with (almost)
zero idiosyncratic risk).

• In practice, most portfolios and even entire market indexes are not completely well-
diversified.

• Nevertheless, the APT holds approximately also for non-well-diversified portfolios, i.e.,
the APT approximately predicts also the expected returns on these portfolios.
• Deviations should on average be zero.
• The predicted expected returns should be unbiased.
• However, the precision of the prediction deteriorates with increasing idiosyncratic
risk.

18 Investments
APT vs. CAPM

• The CAPM applies to all assets whereas the APT applies only to (well-) diversified
portfolios.

• The CAPM relies on an unobservable market portfolio that includes all risky assets
whereas arbitrary observable portfolios are sufficient for the APT.

• The market portfolio resp. the market factor plays no special role in the APT.

• The APT allows assets’ expected returns to depend on multiple risk factors whereas the
CAPM justifies only the existence of the market factor.

• The CAPM assumes all investors to be mean-variance optimizers whereas the APT
requires investors only to be non-satuated such that they eliminate arbitrage
opportunities.

19 Investments
Fama-French Three-Factor Model

• The Fama-French three-factor model (Fama and French, 1993, 1996) includes a size and
a value factor in addition to the market factor motivated by the CAPM:
ri ,t − rf ,t = α i + β im (rm,t − rf ,t ) + β iSMBSMBt + β iHMLHMLt + ei ,t

where
SMB = Small-Minus-Big, i.e., the return on a portfolio of small stocks in excess
of the return on a portfolio of large stocks.
HML = High-Minus-Low, i.e., the return on a portfolio of stocks with high
book-to-market ratios in excess of the return on a portfolio of stocks
with low book-to-market ratios.

• The market factor aims to capture systematic risk originating from macroeconomic factors.

• The two firm-characteristics are chosen because of the empirical observations that market
capitalization (size) and book-to-market ratio predict deviations of average stock returns
from levels consistent with the CAPM.
• They are not themselves obvious candidates for relevant risk factors.
• But they may proxy for yet-unknown systematic risks.

21 Investments
Carhart Four-Factor Model

• Carhart (1997) extended the Fama-French three-factor model by a momentum factor:

ri ,t − rf ,t = α i + β im (rm,t − rf ,t ) + β iSMBSMBt + β iHMLHMLt + β iUMDUMDt + ei ,t

where
UMD = Up-Minus-Down, i.e., the return on a portfolio of past winner stocks in
excess of the return on a portfolio of past loser stocks.

• The momentum factor is constructed analogously to the value factor, only that
• the second sort is according to the stocks’ performance from month t-12 to month t-
1 rather than their book-to-market ratio, and
• the portfolios are rebalanced monthly rather than annually.

22 Investments
Fama-French Five-Factor Model

• Fama and French (2015) extended their three-factor model by a profitability and an
investment factor:

ri ,t − rf ,t = α i + β im (rm,t − rf ,t ) + β iSMBSMBt + β iHMLHMLt + β iRMW RMW + β iCMACMAt + ei ,t

where
RMW = Robust-Minus-Weak, i.e., the return on a portfolio of stocks with high
profitability in excess of the return on a portfolio of stocks with low
profitability.
CMA = Conservative-Minus-Aggressive, i.e., the return on a portfolio of stocks
with low investment in excess of the return on a portfolio of stocks with
high investment.

• The profitability and the investment factor are constructed analogously to the value
factor, only that the second sort is according to the firms’ operating profitability
respectively the firms’ asset growth rather than their book-to-market ratio.

23 Investments
Other Factors

• Liquidity Factor: the risk-adjusted average return on stocks with high sensitivities to
aggregate liquidity exceeds that for stocks with low sensitivities to aggregate liquidity.

• Betting-Against-Beta Factor:
• Goes long leveraged low-beta assets and short high-beta assets
• Produces significantly positive risk-adjusted returns
• High (low) beta assets exhibit negative (positive) alphas

• Quality-Minus-Junk:
• Goes long high-quality stocks and shorts low-quality stocks
• Earns significantly positive risk-adjusted returns.

24 Investments
Limitations of Factor Models

• Potential interpretations of empirical factor models:

• Multifactor ICAPM where the factors reflect extra hedging demands of investors.
• Multifactor APT where factors proxy for common risk factors.
• Yet-unexplained anomalies, i.e., firm characteristics are correlated with CAPM
alphas.

• Problems:
• None of the factors in the proposed models can be clearly identified as hedging a
significant source of uncertainty.
• Factors could so far not be clearly linked to some common risk factors (although
some vague interpretations exist).
• Factors may be due to data snooping, i.e., the patterns underlying these factors
may be due to pure chance.

25 Investments
Factor Investing

• Factor Investing: investment strategy in which securities are selected based on certain
stock/firm characteristics (“factors”).

• Factors have historically offered favorable risk and return patterns.

• Factors can but do not have to rely on factors of empirical multifactor models.

• Factor investing has gained much popularity in the investment management industry in
recent years.

• But:
• Since many factors that earn return premia are likely to be associated with higher
risk, the higher returns are not a “free lunch”.
• Factors that are due to data snooping might not earn return premia in the future.
• Wide-spread factor investing might lead to the disappearance of the factors’ return
premia.

27 Investments
Relevant Factors in Practice

Source: Invesco Global Factor Investing Study (2021)

28 Investments
Reasons for Factor Investing

Source: Invesco Global Factor Investing Study (2021)

29 Investments
Implementation Choices

• Factor: the factor whose premium the investor wants to harvest (e.g. value factor).

• Long vs. Long-Short: decision whether the factor investing strategy is implemented as a
long-only strategy (e.g. investing only in value stocks) or a long-short strategy (e.g.
going long value stocks and going short growth stocks).

• Investment Universe: definition of the universe of stocks from which the investor selects
stocks for his portfolio (e.g. all stocks that are in the S&P500).

• Portfolio Selection: rule how the investor selects stocks from the defined investment
universe.

• Portfolio Weighting: determination how the selected stocks are weighted in the portfolio
(e.g. value-weighting, equal-weighting)

• Rebalancing Frequency: frequency with which the portfolio is newly formed (e.g.
monthly, quarterly, annually).

30 Investments
Implementation Steps

• Step 1: Choose the factor and whether to implement a long-only or a long-short strategy.

• Step 2: Define your investment universe.

• Step 3: Obtain fundamental data on the stocks in your investment universe.

• Step 4: Define a rule that selects stocks from your investment universe for your long
portfolio (and your short portfolio).

• Step 5: Define a rule how you weight the stocks in your long (and short) portfolio.

• Step 6: Buy the stocks.

• Step 7: Repeat Steps 1 to 6 at your next rebalancing date.

31 Investments
 Example Value Investing: Implementation Choices

Example 2: Market Portfolio Example 3: Value-Minus-

Example 1: Value Portfolio
with Value Tilt Growth Portfolio

Factor Value (Book-to-Market Ratio) Value (Book-to-Market Ratio) Value (Book-to-Market Ratio)

Long vs. Long-Short Long-only Long-only Long-Short

All stocks listed on NYSE, All stocks listed on NYSE, All stocks listed on NYSE,
Investment Universe
AMEX, or NASDAQ AMEX, or NASDAQ AMEX, or NASDAQ

Long: Top 20% Value Stocks

Portfolio Selection Top 20% Value Stocks All Stocks Short: Bottom 20% Value
Stocks

Market Value-Weighted: Based on B/M Rank: Equal-Weighted:

1
Portfolio Weighting 𝑀𝑀𝑀𝑀𝑖𝑖 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑖𝑖 𝑤𝑤𝑖𝑖 = 𝑖𝑖 = 1, … , 𝑁𝑁 ∈ 𝑉𝑉𝑅𝑅𝑉𝑉𝑉𝑉𝑉𝑉
𝑤𝑤𝑖𝑖 = 𝑤𝑤𝑖𝑖 = 𝑁𝑁
∑𝑖𝑖 ∈ 𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉𝑉 𝑀𝑀𝑀𝑀𝑖𝑖 ∑𝑖𝑖 ∈ 𝑀𝑀𝑉𝑉𝑀𝑀𝑀𝑀𝑉𝑉𝑀𝑀 𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑅𝑖𝑖 1
𝑤𝑤𝑗𝑗 = − 𝑗𝑗 = 1, … , 𝐾𝐾 ∈ 𝐺𝐺𝐺𝐺𝐺𝐺𝑤𝑤𝐺𝐺𝐺
𝐾𝐾

Rebalancing
Annually (each July) Annually (each July) Annually (each July)
Frequency

32 Investments