hhif lecture series: modern portfolio theory and asset pricing
DESCRIPTION
HHIF Lecture Series: Modern Portfolio Theory and Asset PricingAlexander RemorovUniversity of TorontoFebruary 18, 2011Alexander Remorov (University of Toronto)HHIF Lecture Series: Modern Portfolio Theory and Asset Pricing1 / 23Portfolio ManagementLet’s say after performing some analysis you decided to invest in several different securities Your combined investments in the securities form a portfolio Question: What are the best weights in this portfolio?Things to consider:ExpectedTRANSCRIPT
HHIF Lecture Series:Modern Portfolio Theory and Asset Pricing
Alexander Remorov
University of Toronto
February 18, 2011
Alexander Remorov (University of Toronto) HHIF Lecture Series: Modern Portfolio Theory and Asset Pricing 1 / 23
Portfolio Management
Let’s say after performing some analysis you decided to invest inseveral different securities
Your combined investments in the securities form a portfolio
Question: What are the best weights in this portfolio?
Things to consider:I Expected Return of the whole portfolioI Risk of the whole portfolio and how to measure itI What are your targets for portfolio performance?
To form a portfolio that is expected to meet your targets, need to:I Model the expected return and risk of securitiesI Model the risk of the whole portfolioI Use the model to determine optimal weights
Alexander Remorov (University of Toronto) HHIF Lecture Series: Modern Portfolio Theory and Asset Pricing 2 / 23
Modern Portfolio Theory
Assumptions:I Standard Deviation of Return should be used to measure RiskI Asset Returns are jointly normally distributedI Investors are risk-averse: Between two assets with same return,
the one with lower risk is chosen
Benefits of Diversification - portfolio risk can be reduced byholding uncorrelated assets
Given assets in a portfolio, can plot all possible risk-returncombinations of portfolios combined of those assets
I Min. Variance portfolio - lowest level of risk possibleI Efficient Frontier - highest return for given level of risk
Investor Goal is to maximize Sharpe Ratio: S =E (R − Rf )
σ(R − Rf )
I Sometimes assume risk-free rate (Rf ) is constant
Alexander Remorov (University of Toronto) HHIF Lecture Series: Modern Portfolio Theory and Asset Pricing 3 / 23
Modern Portfolio Theory - Optimization
We consider two common optimization problems in portfolio theoryI Consider a portfolio P of n assets A1,A2, . . . ,An
I Asset Ai provides return Ri and has weight wi in portfolioI Let w be the vector of weights, w = (w1,w2, . . . ,wn)I Let Σ be the covariance matrix of r = (R1,R2 . . . ,Rn); Σij = σij
Which weights produce the portfolio with minimum variance?
Minimize: σ2(P) = σ2(w1R1 + w2R2 + . . .+ wnRn)=
∑ni=1 w 2
i σ2(Ri ) +
∑i<j 2wiwjσij = w tΣw
subject to: w1 + w2 + . . .+ wn = w t(1 1 . . . 1) = 1
Which weights produce the portfolio with maximum Sharpe Ratio?I We assume for simplicity Rf is constant
Maximize: S(P) = E(R)−Rf
σ(R) = w t r−Rf√w tΣw
subject to: w1 + w2 + . . .+ wn = w t(1 1 . . . 1) = 1
Alexander Remorov (University of Toronto) HHIF Lecture Series: Modern Portfolio Theory and Asset Pricing 4 / 23
Modern Portfolio Theory - Optimization
These problems can be explicitly solved in terms of r ,Σ
They can also be solved using software (e.g. Excel, Matlab, R, etc.)
We can consider more complicated problems related to portfoliomanagement
I The portfolio is expected to achieve a specific returnI There are constraints on portfolio weightsI Adjusting weights periodicallyI Many others, depending on portfolio objective
What is the big picture here?
Diversification decreases risk (but also usually expected return)
Assumption: investors hold portfolios to maximize Sharpe Ratio
For a given set of assets, the optimal risky portfolio(with maximum Sharpe Ratio) is unique
I Regardless of risk characteristics of the investorI Invest in optimal risky portfolio and risk-free asset depending
on risk aversion to get optimal complete portfolio
Alexander Remorov (University of Toronto) HHIF Lecture Series: Modern Portfolio Theory and Asset Pricing 5 / 23
Modern Portfolio Theory - GraphSeries/Efficient.jpg
Alexander Remorov (University of Toronto) HHIF Lecture Series: Modern Portfolio Theory and Asset Pricing 6 / 23
Capital Asset Pricing Model
Assumptions:I MPT: risk aversion, care about mean and variance onlyI Single period investment horizonI Investors have homogenous expectationsI Perfect Market
F Individual investor cannot affect stock priceF No taxes or transaction costs; everyone is equally informedF Lending and borrowing at risk-free rate
Using these assumptions we can price assets (e.g. stocks)
Market Portfolio: portfolio of all assets on the marketI Exists only in theory, impossible to constructI However, can approximate, e.g. S&P 500 (≈ 75% of US equity)
Total Supply = Total DemandI Net supply of risk-free asset is zeroI Therefore: combine investor portfolios to get market portfolio!I Optimal Risky portfolio for every investor is the same!I Hence optimal risky portfolio is the market portfolio
Alexander Remorov (University of Toronto) HHIF Lecture Series: Modern Portfolio Theory and Asset Pricing 7 / 23
CAPM - Continued
Consider an asset A with expected return Ra, standard deviation σa,correlation with market σam
Adding A to market portfolio should give same risk-adjusted return
Using basic algebraic manipulations, get following result:
E (Ra) = Rf + βa(E (Rm)− Rf ),
where Rm is market return, βa =σam
σ2m
is beta
This can also be written as E (Ra)− Rf = βa(E (Rm)− Rf )I E (Ra)− Rf is risk premiumI β is riskI E (Rm)− Rf is market risk premium, here ”price of risk”
Usually estimate beta by using historical stock and benchmark pricesI Calculate historical returns (daily/monthly/yearly/etc.)I Find σam and σ2
m, then βa = σam
σ2m
I Or run regression: Ra,t − Rf = β0 + βa(Rm,t − Rf ) + εa,t
Alexander Remorov (University of Toronto) HHIF Lecture Series: Modern Portfolio Theory and Asset Pricing 8 / 23
CAPM - Systematic and Unsystematic Risk
The CAPM can be rewritten to include an error term εa:
Ra = Rf + βa(Rm − Rf ) + εa
I Assume E (εa) and error is normal (from MPT)I The error is specific to asset, results in unique riskI The βa(Rm − Rf ) term results in systematic (market) risk
By definition of unique risk Rm and εa uncorrelated, so:
σ2a = β2
aσ2m + Var(εa)
I Total risk can be decomposed into market and unique risk
Consider n assets A1,A2, . . .An in a portfolio PI Each asset has weight wi , beta βi and unique risk εiI We can determine unsystematic and unique risk of P
βp =σpm
σ2m
= Cov(w1R1+...+wnRn,rm)σ2
m=
∑ni=1 wi
Cov(Ri ,rm)σ2
m=
∑ni=1 wiβi
Portfolio beta is weighted sum of asset betas
Alexander Remorov (University of Toronto) HHIF Lecture Series: Modern Portfolio Theory and Asset Pricing 9 / 23
CAPM - Example
Consider a portfolio P of three stocks A,B,CI Assume βA = 0.5,wA = 0.1I βB = 1.1,wB = 0.4I βC = 1.8,wC = 0.4I Remaining part - cash, β = 0
Assume Rf = 1%, market risk premium E (Rm)− Rf = 9%
We can then calculate expected return on whole portfolioI βP = waβA + wbβB + wcβC = 1.21I Then expected return of portfolio is 1 + 1.21 · 9 = 11.89%
Alexander Remorov (University of Toronto) HHIF Lecture Series: Modern Portfolio Theory and Asset Pricing 10 / 23
Systematic and Unique Risk for a Portfolio
Consider a portfolio P of n assets with same notation as before
Systematic risk is not hard to compute:I systematic risk = β2
pσ2m = (w1β1 + . . .+ wnβn)2σ2
mI Holding low beta stocks results in lower riskI Almost always βi > 0 so βp > 0; very hard to eliminate market risk
Let us look at unique risk:I Recall R = w1R1 + w2R2 + . . .+ wnRn
I According to CAPM Ri = Rf + βi (Rm − Rf ) + εiI So R = Rf + βp(Rm − Rf ) + εp and εp = w1ε1 + . . .+ wnεnI We assume unique risks are uncorrelated, so cov(εi , εj) = 0; then:
Var(εp) = w 21 Var(ε1) + w 2
2 Var(ε2) + . . .+ w 2n Var(εn)
Assume wi = 1n ,Var(εi ) = σ2
I Then Var(εp) = 1nσ
2. This decreases as n (number of assets) increasesI For sufficiently large n, Var(εp) (unique risk) becomes very smallI Note we cannot have n→∞, this is real life
Alexander Remorov (University of Toronto) HHIF Lecture Series: Modern Portfolio Theory and Asset Pricing 11 / 23
Recap of CAPM and Advantages
Optimal risky portfolio is market portfolio, therefore:
Ra = Rf + βa(Rm − Rf ) + εa
Portfolio beta is weighted sum of asset betas
Using regression can estimate systematic and unique riskI Market risk usually cannot be eliminatedI Unique risk can be greatly reduced with many assets
Advantages
A simple way to estimate asset and portfolio return and risk
Helps explain the benefits of diversification and decomposition of risk
Allows for measurement of portfolio performanceI For a specific period, find average return R on portfolioI Calculate β̂, Rf , Rm based on prices during periodI Perform regression: R − Rf = α + β̂(Rm − R)I α called Jensen’s alpha; if positive - beat the market
Alexander Remorov (University of Toronto) HHIF Lecture Series: Modern Portfolio Theory and Asset Pricing 12 / 23
Drawbacks of CAPM
CAPM on average explains only 80% of portfolio returnsI Some other models do much better (e.g. 90% for Fama-French)
The market portfolio should include everythingI Stocks, bonds, commodities, real estate, human capital, etc.I Evidence that using a more complete portfolio explains more of returns
Very strong assumptions about distribution of returnsI Stock returns are often not normally distributedI Standard Deviation, Covariance are not constant
Fama and French have shown that small cap value stocks exhibitabnormal positive returns
Not clear that investors have ”homogenous” expectationsI Although many argue that outlying expectations ”cancel out”
Various constraints on portfolio selection and tradingI Institutional Level, e.g. investment grade, margin requirementsI Individual Investor level - risk aversion, irrational behavior?
Alexander Remorov (University of Toronto) HHIF Lecture Series: Modern Portfolio Theory and Asset Pricing 13 / 23
Arbitrage Pricing Theory
Assumptions:I No taxes or transaction costsI Some agents are smart and can form desired portfoliosI Securities have finite expected values and standard deviations
Assumptions allow a smart investor to execute any trading strategy
Arbitrage: ”free lunch”; there exists a strategy which:
1. Starts with portfolio P containing nothing.2. At the end is guaranteed to produce positive money.
(Self-financing condition - cannot invest external funds in portfolio)
Example: two 1-yr zero-coupon bonds with 3% and 4% yields
APT: whenever there is arbitrage, some smart investor(s) willimmediately exploit it ⇒ Therefore there is no arbitrage!
Note: in real life can have arbitrageI Maybe not worth to exploit due to transaction costs, trading rules, etc.
Alexander Remorov (University of Toronto) HHIF Lecture Series: Modern Portfolio Theory and Asset Pricing 14 / 23
Multifactor Model
Asset returns are driven by k factorsI e.g. inflation, unemployment, manufacturing, market, etc.
Return Ri on Asset i is given by:
Ri = E (Ri ) + βi ,1f̃1 + βi ,2f̃2 + · · ·+ βi ,k f̃k + εk
I Assume for 1 ≤ j ≤ k , Factor j has return Fj
I Define f̃j = Fj − E (Fj) so we see E (f̃j) = 0I βi,j is factor loading/beta - how return Ri responds to change in f̃jI εk is unique risk (not explained by factors); assume E (εk) = 0
Consider a stock of a copper mining company with a factor loading of1.5 on a US manufacturing index (e.g. Purchasing Managers Index)and a factor loading of 0.6 on inflation.If the manufacturing index increases by 4% and inflation increases by5%, we expect the return on the stock to increase by:1.5 · 4 + 0.6 · 5 = 9%.
Alexander Remorov (University of Toronto) HHIF Lecture Series: Modern Portfolio Theory and Asset Pricing 15 / 23
APT Model
Assuming the factor model and no arbitrage, we must have:
E (Ri ) = λ0 + βi ,1λ1 + βi ,2λ2 + · · ·+ βi ,kλk (1)
I Here λj is risk premium on factor j ; the extra return you expectto get for an extra unit of exposure to factor j
I Note that risk premium can be negativeI λ0 corresponds to the return on an asset with no factor risks, i.e.
risk-free asset
Where does Arbitrage Pricing Come in?I Two portfolios with the same exposure to each factor should
have the same priceI For assets 1, 2, ..., n in real life, there exist unique numbersλ0, λ1, . . . , λk such that equation (1) holds for every asset
I Otherwise there is arbitrage!
Alexander Remorov (University of Toronto) HHIF Lecture Series: Modern Portfolio Theory and Asset Pricing 16 / 23
APT - Example
Consider stocks of two companies: Ebay and AmazonI Assume each company will buy back all its shares in one yearI Assume you can buy the stock now and cannot sell until the buybackI Neither company pays dividends so capital gains is all you will get
Consider two cases for the online retail market in one yearI Measure ”online retail market index” I ; now I = 100I Massive Expansion: I = 120 (20% return); probability 70%I Minor Expansion: I = 105 (5% return); probability 30%
Alexander Remorov (University of Toronto) HHIF Lecture Series: Modern Portfolio Theory and Asset Pricing 17 / 23
APT ExampleEstimate prices for each stock in one year:
EBAY AMZN
I=120 $50.0 $300
I=105 $31.0 $160
E(Price) $44.0 $258
Price Today $34.7 $187.8
Return 27.7% 37.4%Estimate factor loadings βEBAY ,I , βAMZN,I
I For EBAY stock: REBAY = E (REBAY ) + βEBAY ,I · f̃I + εEBAY
I Recall f̃I = FI − E (FI ), NOT the factor return!
Major Expansion: 44.1 = E (REBAY ) + βEBAY ,I · 4.5Minor Expansion: −10.6 = E (REBAY ) + βEBAY ,I · (−10.5)
I Solving get:
E (REBAY ) = 27.7% (already know this) and βEBAY ,I = 3.65
I Do the same thing for AMZN, get:
E (RAMZN) = 37.4% (already know this) and βAMZN,I = 4.97
Alexander Remorov (University of Toronto) HHIF Lecture Series: Modern Portfolio Theory and Asset Pricing 18 / 23
APT Example
Now that we know factor loadings, can find risk premia
E (REBAY ) = λ0 + βEBAY ,I · λI
For EBAY: 27.7 = λ0 + 3.65λI
For AMZN: 37.4 = λ0 + 4.97λI
Solving, get λI = 7.35%, λ0 = 0.87%
λI - risk premium on the ”Internet Retail” factor
λ0 - return on asset not exposed to factorI If there are no other factors, this is risk-free rate
Alexander Remorov (University of Toronto) HHIF Lecture Series: Modern Portfolio Theory and Asset Pricing 19 / 23
Applying APT
The previous approach can be done for more than 2 factors
Can apply APT to historical data to price assets
Two general approaches:
Principal Component AnalysisI Statistical procedure to extract factors from historical returns;
estimates βijs and λjs togetherI However factors have no real life significance
Come up with factors yourselfI Could use macroeconomic factors (e.g. inflation, GDP,
manufacturing index. etc.)F See Chen, Roll, Ross 1986 paper
I Could use Fama-French factors - based on size and P/B ratioF See Fama-French three-factor model
I Once you have identified factors, get factor returns, and estimate f̃jsI Perform regression of Ri on f̃js to estimate βijsI Perform cross-sectional regression of E (Ri )s on βijs to estimate λjs
Alexander Remorov (University of Toronto) HHIF Lecture Series: Modern Portfolio Theory and Asset Pricing 20 / 23
Concluding Remarks
Asset allocation is crucial when it comes to investing
Finding optimal portfolio weights is a very hard problemI Depends on portfolio objectives and constraints
Sharpe Ratio is often used to measure performance
CAPM: a quick way to estimate portfolio return and riskI Makes a lot of assumptions that don’t hold true in real lifeI But does illustrate the benefits of diversification
APT: also used to estimate return and risk, but harder to doI Identifying the factors is a major challenge
F How many factors do we really need?
I Need to be careful about which statistical tools and which datais used to estimate betas and risk premia
These are just models!I Should be used as guidelines, not just blindly followedI Good idea to think about why a model outputs certain results
Alexander Remorov (University of Toronto) HHIF Lecture Series: Modern Portfolio Theory and Asset Pricing 21 / 23
Any Questions?
Upcoming Events:
HHIF Quarterly MeetingI Friday, March 4, 6:00-8:00 p.m., South Dining Room,
Hart House
Alexander Remorov (University of Toronto) HHIF Lecture Series: Modern Portfolio Theory and Asset Pricing 22 / 23
References
1. prof. Jiang Wang (MIT) - Arbitrage Pricing Theory Lecturehttp://web.mit.edu/15.407/file/Ch12.pdf
2. prof. Giovanna Nicodano (University of Toledo) -Multi-Factor Models and the APT Lecturehttp://web.econ.unito.it/nicodano/handout_bkm1_apt.pdf
3. Investopedia: Financial Conceptshttp://www.investopedia.com/university/concepts/default.asp
Alexander Remorov (University of Toronto) HHIF Lecture Series: Modern Portfolio Theory and Asset Pricing 23 / 23