Week 5 – Factor Models

We have examined MVP and the related CAPM

- Minimise variance for a required rate of return/maximise utility subject to the budget
constraint

Both MVP and CAPM assume prior knowledge of asset expected returns and covariance
- This involves a very large number

The idea behind factor models is that there are a relatively small number of factors that are
the main determinants of risk and return
- The CAPM states that return is earned only for bearing risk relative to the market

It is not difficult to imagine firms returns may be influenced by other underlying factors
- Mining firms may be influenced by the iron ore price
- Airliners may be influenced by fuel prices

Factor Models

A factor model is a statistical specification of how a dependent variable (asset returns) is

linked to independent variables (factors) which are usually identified with the return on an
index (or multiple indices)

Assumptions:
- Idiosyncratic risks are independent (covariance between error terms is 0)
- Assume each factor is a unique source of systematic risk (Covariance between factor
terms is 0)
- Assume that there is no covariance between the factors and the error terms
(otherwise – is the error term really idiosyncratic)?
SM & Diversification

Visualising Total Risk with the total number of assets

Treynor Black Model

- Wa – proportion of wealth in active portfolio

- Wm – proportion of wealth in the market portfolio
- What’s the economic intuition of the TB model

SIM Estimates
- When there is only one factor (M), the expected return/volatility for portfolios/assets reduces
to

- How many parameter estimates are needed now?

-
Advantages
- It allows analysts from different sector to combine their information in a simple way
to construct portfolios
- All the information required are the respective stocks α ' s and β ' s

Disadvantages
- It assumes that there is only one common source of risk (captured by the market
index) and this omit risk for a particular industry
- E.g. a carbon tax would likely affect some industries (miners, energy, etc) more than
others (banks/financials). This will not be explicitly captured in a model of this form
- It ignores correlations among firm specific risks and thus may not provide the best
risk minimises weights in portfolio theory
Estimation Issues

Do Factor Models Work?

- The efficacy of factor models is difficult to ascertain as it comes down to how well
one can estimate α and β

Factor Models in Practice

Factor Models in Practice

Application: Market Tracking

-

Week 6 – APT Theory

Arbitrage Pricing Theory
 - Construct a portfolio with no probability of loss and strictly a positive probability of
making a positive return
- Earning money without bearing any risk or investing any of my own money
- Results from securities being mispriced
- Exploiting the arbitrage returns prices back to arbitrage free values via strong
buying/selling
- Happens quickly due to strong buy/sell pressures

Arbitrage
- Pure Arbitrage: truly risk free investment has a positive probability of return.
- Statistical Arbitrage: Assets are mispriced on average and a net zero investment
which exploits this can profit in expectation. Casino’s used this (and HFT)
- Merger Arbitrage: Betting on the outcome of a merger. (Not really arbitrage)

Asymptotic Arbitrage
- This occurs when an arbitrage opportunity may be reached by building portfolios of a
large number of asset (lim n  ∞ ) which can be used to diversify away all risks
- If a portfolio can still earn positive returns it is called

APT – Law of One Price

- Previous example is a violation of the law of one price

- APT 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 permit the prolonged presence of arbitrage
(pure)
- Note that we have made no assumptions regarding the way investors set their
preference between risk and return

- Assume the expected value of the factor

- Now the expected return is just alpha

Portfolios with same beta and risk factor, different alphas

- Can take a long short position in these two portfolios to generate a return (5% return
on this trade)
- Selling this investment requires no net position (purchasing A is funded by
selling B) and the return is guaranteed, this is an arbitrage opportunity
- A rational investor would trade this position as heavily as possible
 - Two assets with the same betas (systematic risk) therefore, should have the same
expected return (otherwise there is an arbitrage opportunity)
- The exploitation of this arbitrage opportunity will eliminate this arbitrage
opportunity

- Portfolio with different β ' s

- We can show that these portfolios have a return which is proportional to their
factor β
- Consider a portfolio C which has a beta of 0.5 and E { r c } =0.06 which currently
plots below the line

Multiple Factors

- E(fj) = 0
 - The ideas behind the multifactor model are identical to the single factor model,
though we require the concept of a factor portfolio as a benchmark
- This is a portfolio which is constructed to have a β of 1 with respect to one
factor, and a β of 0 with respect to all other factors

APT and Risk Premia

- APT can generally be written as:

- This allows us to define whether a factor is relevant or not

- A factor that has a risk premium of 0 has no systematic impact on expected
returns
- We now examine a statistical test to determine if risk premia ( λ ¿ are
statistically different from 0

Fama MacBeth Regressions

- Fama and MacBeth developed a method to test risk premia associated with factors
- Their original applications was the CAPM, but the procedure works equally
well for multifactor models
- Idea is to perform two regressions
1. A time series regression to compute β ' s
2. A cross sectional regression to compute risk premia
 3. Test the estimate risk premia for statistical significance

Fama MacBeth Regressions

- Like all statistical tests, Fama-MacBeth regressions are imperfect
- They suffer from the error in-variables problem because the betas used in the second
pass regression must be estimated
- The tests assume that the time series residuals ε
- These issues can be corrected for, but this is outside the scope of the course

APT-A Single Factor

- Imagine a well-diversified portfolio consisting of market cap weights as the only
relevant factor. Consider this the excess returns on this portfolio as the relevant
factor
- By APT, the expected return relationship is E { r f }=r f + β M (E { r M }−r f )
 APT-A Single Factor
- Thus, via APT, we have derived an equation for portfolio expected returns which is
identical to that derived by the CAPM but without having to resort to the many
assumptions of the CAPM
- Also, the market portfolio in APT need not be some “true” market portfolio as
in the CAPM, it simply needs to be a well-diversified portfolio which lies along
the SML

- At this stage, it is pertinent to query what is the scenario for a single asset
- The theory developed thus far applies to well diversified portfolios, so how
does a single asset fit into the structure?

APT – Portfolios vs Stocks

Appropriate Factors
- A weakness in APT theory is that it provides no guidance on what factors are
appropriate
- Practical use requires knowing all the relevant factors
- We thus empirically test a range of factors for suitability

Appropriate Factors
 - Market return: Excess return on a market cap weighted portfolio
- Small – Big: Return of a portfolio of small stocks in excess of returns of a
portfolio of big stocks (size premium). Small stocks are fundamentally riskier
than large ones, so a premium is required to invest in them.
- High – Low: Return of stocks with a high book to market ratio minus return on
portfolio of stocks with low book to market ratio (value premium). Low BM
implies undervalued stock or poorly performing. This factor proxies for this
systematic risk.
- Conservative – Aggressive Factor
- Robust-weak (RMW)

Week 7 – QEPM

QEPM aims to bring a more scientific approach to investing

- Eliminate bias and leverage technology
QEPM helps to eliminate bias

7 Tenets of QEPM
1. Markets are mostly efficient
2. Pure arbitrage opportunities do not exist
- Risks eliminate pure arbitrage opportunity (liquidity risks, default risk)
3. Quantitative analysis can create statistical arbitrage opportunities
4. Quantitative analysis combines available information an efficient way
5. Quantitative models should be based on sound economies theories
6. Quantitative models should reflect persistent and stable patterns
7. Deviations from a benchmark are justified only if the uncertainty in the view that
leads to deviation is small enough

Stock Screening
- Quantitative method that ranks stock according to a criteria (which reflect an
investment philosophy)
- This is achieved by allocating scores to the stock’s attributes
- There are two kinds of screening:
1. Sequential screening: A different rule is applied in order, and as each rule is applied,
the selection of stocks decreases
2. Simultaneous screening: Rather than having to prioritise each attribute, we simply
select the stock which attains the highest score

Sequential Screen
- Rank stocks by P/B ratio – exclude firms which don’t hit P/B ratio

Why Screen
 - If sequential screens are based on philosophy, rather than statistically refined
methods, what’s the point of screening?
- Stock screens can be automated so that even stocks which may not be
covered by an analyst are still considered for inclusion in the portfolio
-

Week 8 – Anomalies

Anomaly – deviation from underlying theory

In finance – anomalies refer to deviations from the CAPM or efficient markets hypothesis
(EMH)
- This topic will focus on CAPM anomalies

Anomalies are typically identified via a self funded portfolio (a long leg and an offsetting
short leg) that is constructed by sorting on a particular characteristic
- Factor replication

If such a portfolio earns a significant risk adjusted returns, relative to the CAPM, we may
interpret this as an anomaly

Classes of Anomalies
Cross Sectional Anomalies – found by comparing the returns on assets at the same point
in time. Cross sectional studies suggest that on average, firms with certain characteristics
are mispriced relative to each other. This can be interpreted as a missing risk factor.
Time Series anomalies – compare a pattern of returns through time

Smart Beta Portfolios

Smart Beta Portfolios – smart beta are a set of investment strategies that are designed
to sit in between active and passive management

- Actively managed portfolios are frequently altered to take advantage of newly

arriving information
- Passively managed portfolios use market capitalisation weights and do not deviate
from this allocation
- When costs are accounted for, there is little evidence that managed funds
outperform funds on average
- Smart beta funds aim to use known anomalies to provide investors with access to
additional sources of expected returns
- Smart beta funds achieve this through a transparent, rules based allocation

- The CAPM implies passive management is the best

- Some costs to smart beta portfolios – but less than true active management

- Classic Anomalies
- Size
- Value
- Momentum & reversal
- Volatility: Idiosyncratic, systematic, total

- Size
- Banz (1981) was the first article to identify the size anomaly
- It was identified by constructing portfolios that are composed of a long
position in small stocks and a shot position in large stocks measured by
market capitalisation
- Banz (1981) finds that portfolios constructed based on firm size achieved
significant returns relative to what is suggested by the CAPM
- A premium for size makes intuitive sense from multiple perspectives:
 Small firms are inherently riskier than large ones and require
additional premium to hold them
 Small firms are likely to perform better when market conditions are
favourable
- Size is usually measured by SMB factor
 Stocks sorted by market cap and portfolio is long small and short large
stocks
- Fama French finds that SMB no longer appears to be significant (returns have
been competed away)

- Value
- Value stocks are those that appear to be undervalued by the market
- We typically identify this use book to market ration

Common Shareholder Equity

Book ¿ market =
Market Cap

- A high book to market value means that accounting value of the firm is large
relative to the market value
 Firms with low book to market values are called growth stocks
- Build HML factor
- Economic intuition
 Value stocks are riskier as they represent value or distress
 The HML factor may proxy for an economic condition that drives firm
performance

- Momentum & Reversals

- Momentum refers to the phenomenon that stocks have had good (bad)
performance (winners/losers) in the past typically continue to perform well
(poorly)
- Momentum portfolios earn positive alphas over a 3-12 month horizon
 - De Bondt and Thaler showed that over 3-5 years, the opposite result tends to
hold (long previous winners and short previous losers) \

- In the short term – investors overact to information

 Good news pushes price too high and bad too low
- In the longer term (3-5 years) investors learn of their overreaction and prices
revert to more appropriate levels

- Volatility Anomalies
- According to economic theory – assets which are risky should provide higher
rates of return
- Since we measure risk as volatility – higher volatility assets should have
higher returns
- However – empirically – this is not always the case
- Portfolios that hold a short position in high volatility assets and a long
position in low volatility assets tend to provide higher returns on average
than suggested by the CAPM

- Volatility managed portfolios

r VM ,t +1=
c
r
2 M ,t +1
σt ( )c
+ 1− 2 r f , t+ 1
σt

- Betting against Beta Approach

1. Identify a set of low beta stock and high beta stocks (sort and split into top
and bottom 20%)
2. Build a fully invested portfolio of high (H) and low (L) beta stocks (usually
value weighted)
3. Create a levered portfolio by borrowing (1-x) and combine x with Beta-L to
form a new portfolio such that β IL =β H
4. Go short β H and long β IL

n1
β H =∑ wi β i
i=1

n2
β L = ∑ w i βi
i=1

l
β L =xβ L +(1−x) β rf

l
β L =xβ L (beta of rf is 0)
 - We lever up the low beta stocks until they equal the beta of the high beta stocks
- We then take a long position in the low beta portfolio and a short position in the high
beta portfolio (this portfolio has a beta of 0)

Why does BAB strategy earn a premium

- There must be a risk source driving these returns
- It is funding liquidity risk
- During strong economic times, institutional investors bid up the price of high beta
assets which reduces their average returns
- However – when funding liquidity dries up, investors are forced to hold lower beta
(lower risk) assets and the strategy becomes less successful
- Alternatively, Bali argues that lottery preferences (the desire for assets that have a
small chance of large returns) explains BAB returns

Smart Beta Strategies

- Smart beta is an investment strategy that lies between active and passive
management
- Passive: Use market capitalisation weights. Low cost weights
- Active: Weights selected by fund manager. High cost, little transparency
- Smart beta strategies don’t use market cap weights (hence differing from passive)
but weight are set according to a formula and hence are transparent

- Costs are higher than passive (rebalance costs) but lower than active information
costs

Factor Tilt

x OL=θ x M +(1−θ) x RP , HML

- This will tilt a portfolio towards earning a premium associated with a risk factor (in
this case, HML)
- Xm is the

Idiosyncratic Vol Anomaly

 Ang et al was the first to identify the idiosyncratic volatility anomaly
 These authors have showed that portfolios were constructed to have assets that are long
high idiosyncratic vol stocks and short low idiosyncratic vol stocks performed poorly
 This result suggests that idiosyncratic volatility carries a negative price of risk
 (This differs to what we expect from the CAPM)
 Empirical results of Ang et al. (2006) have led academic to search for a theoretical
explanation
 However, recent work by Park, Wei and Zhang (2020) attempt to replicate the findings of Fu
(2009) and find look ahead bias
 These authors conclude that there is no idiosyncratic volatility anomaly
  However, all studies share the same flaw that idiosyncratic volatility is measured with
respect to standard models (CAPM, Fama French)

The Total Volatility Anomaly

 Clarle et al (2011) show that the performance of the GMVP is superior to what would be
expected from mean-variance portfolio theory
 These authors show that GMVPs have a risk adjusted performance (Sharpe ratio) that
exceeds the market portfolio
 However, to simplify their approach, these authors employ the SIM to construct their
covariance matrix
 This suggests that the result may be due to the low beta anomaly
 Study restricted to the US

 A more recent study (that has gained a lot of attention in both academic and practitioner
circles) are the volatility managed portfolios of Moreira and Muir (2017)
 These authors show that a portfolio that takes less risk when volatility is high and more risk
when volatility is low produces significant alpha with respect to a variety of models
 Financial theory says that you should ‘buy the dip’ and go long when the risk is high
 But Moreira and Muir showed that taking more risk when volatility

Low Beta Anomaly

 Black et al. found that low beta assets had higher expected returns than predicted by the
CAPM
 Low Beta assets were producing positive alphas and high beta assets produce negative
alphas
 Frazinni and Pedersen (2014) investigate the low beta anomaly via a strategy called “betting-
against-beta”
 This strategy is a major investment approach at AQR Capital (where Frazzini and Pedersen
are Principals)

Smart Beta Strategies

 One of the earliest smart beta strategies was inspired by the work of Arnott
 Another way to think about it is that by altering the weighting scheme, investors can gain
exposure to other sources of risk premia, such as those examined in the first half of the
lecture
 This is the interpretation employed by most smart beta fund managers and has led to them
employing other names
 Smart beta portfolios tilt towards assets with exposure to risk factors

Week 8 – Anomalies
 - In this

Conventional Theory
- Simplest way to determine a portfolios performance is to see what types of return it
makes on average

Time Weighted Returns

- Geometric average represented realised returns
- Worse for predicting future returns

1
1+r g=[ ( 1+r 1) ( 1+r 2) ( 1+r 3 ) … ( 1+ r 20) ] 20

IRR
- Discount rate at which the NPV of an investment is equal to 0
- These methods are ways to compute average returns – however – they do not
account for risk
- Question Becomes
- What type of adjustment should be made for risk?
- Do we account for total risk?

r p−r f
1. Sharpe’s Ratio
σP
- We always compute a Sharpe ratio – but it is not very interpretable

- Might look at using an M 2 measure to overcome these limitations

- Imagine that a portfolio P is mixed with T-bills. We may then select the appropriate
amount of T-bills to use such that the total risk (volatility) of the adjusted portfolio
¿
P matches that of the market

2
M =r P −r M
¿

σ P =x σ P
¿

σ M =x σ P

r P =x r P + ( 1−x ) r f
¿
 r p−r f
2. Treynor’s measure
βP
- Might look at Treynor’s measure when
we care about the total systematic risk in
the portfolio – e.g. Fund of Funds

3. Jensen’s measure α =r p −[r f + β P ( r M −r f ) ]

α
4. Information ratio
σP
- Measures abnormal return per unit of unsystematic risk

Downside Risk
- All performance measures considered thus far have used a standard measure of risk
- σ,β
- This is consistent with the standard theories of risk/reward we have covered
in this course
- However, not all risks are the same
- Example: Which is riskier?
 A: guaranteed loss of 5%
 B: gain of 5% or gain of 10% with equal probability
- If we use return variance as the measure of risk, then B is considered riskier than A
despite B being a clearly superior investment

- What really concerns investors is uncertainty associated with losses rather than
uncertainty in general
- This has led to the development of downside risk measures

DR= √ ∫ T ( TR−r ) p ( r ) dr
−¿ ¿
r
2

DR=√ E ¿ ¿

- DR is just the expected value of squared deviations from the target return
conditional on those returns being below the target
- In practice, we can replace the random return r with a sequence of observations
r t ,t=1 , 2 ,… . , T .
- We can also replace the expectation with a simple average
Sortino Ratio
- Downside risk is used to produce the performance metric called the Sortino ratio

E [ r ] −TR
SR=
DR
- SR measures the expected excess return (with respect to the target) per unit of
downside risk
- Though the end user can select the value of TR, it is common to set it equal to
rf

- There has been recent innovations in the use of semi-variances to split traditional
risk measures into upside/downside components

- We can relate the Treynor and Sharpe ratios to alpha as this table below shows

Alpha and Performance Measures

- The most important takeaway is that all performance measures are positively related
to alpha
- This means that all other things equal – an increase in alpha will increase the
measure of portfolio performance
- In some cases, alpha, is the best measure of a fund’s performance
- Hedge funds typically don’t create well diversified portfolios and usually
trade in illiquid and exotic products or try to use arbitrage to purchase
mispriced securities

Hedge Fund Performance

- For these reasons, the Sharpe ratio is a very bad measure for fund performance
- Hedge funds can be thought of as being providers of alpha
- However – this alpha can often come at the cost of very high specific risks
- In this case the appropriate measure of excess return is the alpha and the
appropriate measure of risk is the firm’s specific risk
- Thus, combing these gives the information ratio as the appropriate measure of
performance for hedge funds
 αP
I R P=
σP
- This is similar to when we examined the SIM and found that the information ratio
was important in determining portfolio weights when we combine active with
passive management

Hedge Fund Performance

- There are issues in evaluating hedge fund performance which must also be
considered;
1. The risk profile of hedge funds can change rapidly as they have freedom to
change their asset allocations easily and without the necessity of permission
from investors
2. Hedge funds tend to focus on illiquid assets. It is difficult to ascertain the
required parameters for assets which trade infrequently
3. Some hedge funds purposely suffer severe short-term losses to gain large
long-term profits. Thus, to evaluate their performance often requires a long
period of time
4. When evaluated as a group, survivorship bias is a real problem. There is a
high degree of turnover in hedge funds which mutual funds are not subject to

Changing Portfolio Composition

- As we have seen – many measures of fund performance require statistical estimates

of means and variances
- However – if a fund changes its composition – these means and variances can
become problematic
- By changing composition, we are creating a new distribution for returns
- This new distribution must be disentangled from the old one to properly
evaluate the fund
- This is a very difficult thing to do and if it cannot be done properly it often
leads to the appearance of increased volatility which would bias most
- common epr

Market Timing
- Involves shifting money from T-bills into the market when it is expected to perform
and vice versa
- How do we determine if a manager has timing ability?
- Construct a situation where a manager has a position in T-bills and the
remaining wealth in the market
 - This manager will try to shift a greater proportion of his wealth into the
market when he believes that the market is going to outperform

r p =x r M + ( 1−x ) r f

β P =x β M =x

Market Timing
- If a manager has good timing ability, they they should have an upwards
sloping characteristic line
- As the market performs better, the slope of the line (beta) increases as the
manager places greater weight in the market fund

r P−r f =a+b ( r M −r f ) + c ( r M −r f ) + ε P

- Using the regression equations on the previous slides, it has been found on
average that fund managers do not possess superior market timing
 There are clearly exceptions, however, in a market which is close to
efficient, it would be expected that there are only a few of these
- There have been several attempts to establish the value of market timin
 One study determined the improved returns assuming that an
investor

Market Timing
- Clearly, market timing is a very valuable thing to possess
- Another way to look at market timing is the possession of a call option
 Since the lowest return possible is the risk free rate, the payoff
becomes that associated with call options
- The value of this call option should be the value of the market timing
- Merton showed that according to Black-Scholes options theory this market
value is

MV ( Perfect Timing )=2 N ( σ M √ T )−1

- Using the same data as before, this was found to be a 16.05% return
- The difference between this measure and the perfect time measure used
previously is due to distributional assumptions in the BS model
P=P1 + P2−1
 - P1 – proportion of correct bull market you predict
- P2 – proportion of correct bear market you predict
- Note : If an investor always predict bull markets – they will have a forecast
predictive power of 0

MV ( Imperfect Timing )=( P1 + P2−1 ) [ 2 N ( σ M √ T ) −1 ]

- Another important aspect of portfolio performance is based on the “style” of

investing that the fund employs
- Explains 97% of returns

Style Analysis

- Rather than comparing a managed fund to a single broad market index, it can often
be better to regress excess fund returns on indices which represent specific asset
classes
- Small cap, medium cap, large cap, T-bills, High P/E, … etc.

Performance Attribution
- The idea is to explain the difference between a managed portfolio P and some bogey
portfolio B
- Suppose there is a universe of n asset classes such as equities, bonds, bills
etc.
- For each class, a representative portfolio is selected
- The bogey portfolio is set to have fixed weighted in each class, and thus, its
return is given by

n
r B=∑ wBi r Bi
i=1

- The portfolio manager chooses weights for his asset classes and thus his
return is given by

n
r P=∑ wBi r Bi
i=1

- The difference in these portfolio’s returns is given by:

n
r P−r B=∑ (w P i r P i¿−w Bi r Bi)¿
i=1
 ( wPi−w B i ) r Bi +w P i ( r Pi−r Bi )=w Pi r Pi−w Bi r Bi

Contribution Contribution
from asset from security
allocation selection

Conclusions