MIT18 S096F13 Lecnote14

Portfolio Theory
Lecture 14: Portfolio Theory
MIT 18.S096
Dr. Kempthorne
Fall 2013
MIT 18.S096 Portfolio Theory 1

Markowitz Mean-Variance Optimization
Mean-Variance Optimization with Risk-Free Asset
Von Neumann-Morgenstern Utility Theory
Portfolio Theory
Portfolio Optimization Constraints
Estimating Return Expectations and Covariance
Alternative Risk Measures
Outline
1 Portfolio Theory

Portfolio Theory
Markowitz Mean-Variance Analysis (MVA)

Single-Period Analyisis
m risky assets: i = 1, 2, . . . , m
Single-Period Returns: m−variate random vector
R = [R1 , R2 , . . . , Rm ]0
Mean and Variance/Covariance of Returns:
α1 Σ1,1 ··· Σ1,m
   
 . 
E [R] = α =  ..  , Cov [R] = Σ =  ... .. .
..
 
. 
αm Σm,1 ··· Σm,m
Portfolio: m−vector of weights indicating the fraction of
portfolio wealth held in each P asset
m
w = (w1 , . . . , wm ) : i=1 wi = 1.
m
Portfolio Return: Rw = w0 R = i=1
P
wi Ri a r.v. with
αw = E [Rw ] = wα 0
2
σw = var [Rw ] = w0 Σw
Portfolio Theory
Markowitz Mean Variance Analysis
Evaluate different portfolios w using the mean-variance pair of the

2 ) with preferences for
portfolio: (αw , σw
Higher expected returns αw
Lower variance varw
Problem I: Risk Minimization: For a given choice of target mean
return α0 , choose the portfolio w to
1 0
Minimize: 2 w Σw
Subject to: w0 α = α0
w 0 1m = 1
Solution: Apply the method of Lagrange multipliers to the convex
optimization (minimization) problem subject to linear constraints:

Portfolio Theory
Risk Minimization Problem

Define the Lagrangian
L(w, λ1 , λ2 ) = 21 w0 Σw +λ1 (α0 −w0 α)+λ2 (1−w0 1m )
Derive the first-order conditions
∂L
∂w = 0m = Σw − λ1 α − λ2 1m
∂L
∂λ1 = 0 = α0 − w0 α
∂L
∂λ2 = 0 = 1 − w 0 1m
Solve for w in terms of λ1 , λ2 :
w0 = λ1 Σ−1 α + λ2 Σ−1 1m
Solve for λ1 , λ2 by substituting for w:
α0 = w00 α = λ1 (α0 Σ−1 α) + λ2 (α0 Σ−1 1m )
1 = w00 1m = λ1 (α0 Σ−1 1m ) + λ2 (10m Σ−1 1m )

α0 a b λ1
=⇒ = with
1 b c λ2
a = (α0 Σ−1 α), b = (α0 Σ−1 1m ), and c = (10m Σ−1 1m )
−1
Portfolio Theory
Risk Minimization Problem

Variance of Optimal Portfolio with Return α0
With the given values of λ1 and λ2 , the solution portfolio
w0 = λ1 Σ−1 α + λ2 Σ−1 1m
has minimum variance equal to
σ02 = w00 Σw0
= λ21 (α0 Σ−1 α) + 2λ1 λ2 (α0 Σ−1 1m ) + λ22 (10m Σ−1 1m )
0
λ1 a b λ1
=
λ2 b c λ2
−1
λ1 a b α0
Substituting = gives
λ2 b c 1
0 −1
2 α0 a b α0 1 2

σ0 = = ac−b 2 cα0 − 2bα0 + a
1 b c 1
Optimal portfolio has variance σ02 : parabolic in the mean
return α . MIT 18.S096 Portfolio Theory 6
Portfolio Theory
Equivalent Optimization Problems

Problem II: Expected Return Maximization: For a given choice
of target return variance σ02 , choose the portfolio w to
Maximize: E (Rw ) = w0 α
Subject to: w0 Σw = σ02
w 0 1m = 1
Problem III: Risk Aversion Optimization: Let λ ≥ 0 denote the
Arrow -Pratt risk aversion index gauging the trade-ff between risk
and return. Choose theportfolio w to
E (Rw ) − 21 λvar (Rw ) = w0 α − 12 λw0 Σw

Maximize:
Subject to: w0 1m = 1
N.B
Problems I,II, and III solved by equivalent Lagrangians
Efficient Frontier:{(α0 , σ02 ) = (E (Rw0 ), var (Rw0 ))|w0 optimal}
Efficient Frontier: traces of α0 (I), σ02 (II), or λ (III)
Portfolio Theory
Outline
1 Portfolio Theory

Portfolio Theory

Risk-Free Asset: In addition to the risky assets (i = 1, . . . , m)
assume there is a risk-free asset (i = 0) for which
R0 ≡ r0 , i.e., E (R0 ) = r0 , and var (R0 ) = 0.
Portfolio With Investment in Risk-Free Asset
Suppose the investor can invest in the m risky investment as
well as in the risk-free
Pm asset.
0
w 1m = i=1 wi is invested in risky assets and
1 − w1m is invested in the risk-free asset.
If borrowing allowed, (1 − w1m ) can be negative.
Portfolio: Rw = w0 R + (1 − w0 1m )R0 , where
R = (R1 , . . . , Rm ), has expected return and variance:
αw = w0 α + (1 − w0 1m )r0
σw 2 = w0 Σw
Note: R0 has zero variance and is uncorrelated with R
Portfolio Theory

Problem I’: Risk Minimization with Risk-Free Asset
For a given choice of target mean return α0 , choose the portfolio
w to
1 0
Minimize: 2 w Σw
Subject to: w0 α + (1 − w0 1m )r0 = α0
Solution: Apply the method of Lagrange multipliers to the convex
optimization (minimization):
Define the Lagrangian
L(w, λ1 ) = 12 w0 Σw + λ1 [(α0 − r0 ) − w0 (α − 1m r0 )]
Derive the first-order conditions
∂L
∂w = 0m = Σw − λ1 [α − 1m r0 ]
∂L
∂λ1 = 0 = (α0 − r0 ) − w0 (α − 1m r0 )
Solve for w in terms of λ1 : w0 = λ1 Σ−1 [α − 1m r0 ]
and λ1 = (α0 − r0 )/[(α − 1m r0 )0 Σ−1 (α − 1m r0 )]
Portfolio Theory

Available Assets for Investment:
Risky Assets (i = 1, . . . , m) with returns: R = (R1 , . . . , Rm )
with
E [R] = α and Cov [R] = Σ
Risk-Free Asset with return R0 : R0 ≡ r0 , a constant.
Optimal Portfolio P: Target Return = α0
Invests in risky assets according to fractional weights vector:
w0 = λ1 Σ−1 [α − 1m r0 ], where
(α0 −r0 )
λ1 = λ1 (P) =
(α−1m r0 )0 Σ−1 (α−1m r0 )
Invests in the risk-free asset with weight (1 − w00 1m )
Portfolio return: RP = w00 R + (1 − w00 1m )r0
Portfolio Theory
Portfolio return: RP = w00 R + (1 − w00 1m )r0

Portfolio variance:
Var (RP ) = Var (w00 R + (1 − w00 1m )r0 ) = Var (w00 R)
= w00 Σw0 = (α0 − r0 )2 /[(α − 1m r0 )0 Σ−1 (α − 1m r0 )]

Portfolio Theory
Market Portfolio M
The fully-invested optimal portfolio with
wM : w0M 1m = 1.
I.e.
wM = λ1 Σ−1 [α − 1m r0 ], where
− 1
λ1 = λ1 (M) = 10m Σ−1 [α − 1m r0 ]
Market Portfolio Return: RM = w0M R + 0 · R0
(α0 Σ−1 [α−1m r0 ])
E (RM ) = E (w0M R) = w0M α =
(10m Σ−1 [α−1m r0 ])
0 −1
[α−1m r0 ] Σ [α−1m r0 ])
= r0 +
(10m Σ−1 [α−1m r0 ])
Var (RM ) = w0M ΣwM
(E (RM )−r0 )2 [α−1m r0 ]0 Σ−1 [α−1m r0 ])
= =
[(α−1m r0 )0 Σ−1 (α−1m r0 )] (10m Σ−1 [α−1m r0 ])2

Portfolio Theory
Tobin’s Separation Theorem: Every optimal portfolio invests in

a combination of the risk-free asset and the Market Portfolio.
Let P be the optimal portfolio for target expected return α0 with

risky-investment weights wP , as specified above.
P invests in the same risky assets as the Market Portfolio and
in the same proportions! The only difference is the total
weight, wM = w0P 1m :
λ (P ) (α0 −r0 )/[(α−1m r0 )Σ−1 (α−1m r0 )]
wM = λ 1(M) = −1
1 (10m Σ−1 [α−1m r0 ])
(10 Σ−1 [α−1m r0 ])
= (α0 − r0 ) [(α−1m r )Σ−1 (α−1 r
m 0 m 0 )]
= (α0 − r0 )/(E (RM ) − r0 )
RP = (1 − wM )r0 + wM RM
σP2 = var (RP ) = var (wM RM ) = wM
2 Var (R ) = w 2 σ 2 .
M M M
E (RP ) = r0 + wM (E (RM ) − r0 )
Portfolio Theory
Mean Variance Optimization with Risk-Free Asset

Capital Market Line (CML): The efficient frontier of optimal
portfolios as represented on the (σP , µP )-plane of return
expectation (µP ) vs standard-deviation (σP ) for all portfolios.
CML = {(σP , E (RP )) : P optimal with wM = w0P 1m > 0}
= {(σP , µP ) = (σP , r0 + wM (µM − r0 )), wM ≥ 0}
Risk Premium/Market Price of Risk
E (RP ) = r0 + wM [E (RM ) − r0 ]

σP
= r0 +
σhM [E (RM ) − r0 ]
i
E (RM )−r0
= r0 + σP
σM
h i
E (RM )−r0
σM is the ‘Market Price of Risk’
Portfolio P’s expected return increases linearly with risk (σP ).
Portfolio Theory
Mean Variance Optimization

Key Papers
Markowitz, H. (1952), “Portfolio Selection”, Journal of
Finance, 7 (1): 77-91.
Tobin, J. (1958) “ Liquidity Preference as a Behavior Towards
Risk,”, Review of Economic Studies, 67: 65-86.
Sharpe, W.F. (1964), “Capital Asset Prices: A Theory of
Market Equilibrium Under Conditions of Risk,” Journal of
Finance, 19: 425-442.
Lintner, J. (1965), “The Valuation of Risk Assets and the
Selection of Risky Investments in Stock Portfolios and Capital
Budgets,” Review of Economics and Statistics, 47: 13-37.
Fama, E.F. (1970), “Efficient Capital Markets: A Review of
Theory and Empirical Work,” Journal of Finance,” 25:
383-417.
Portfolio Theory
Outline
1 Portfolio Theory

Portfolio Theory
Rational portfolio choice must apply preferences based on

Expected Utility
The optimal portfolio solves the
Expected Utility Maximization Problem
Investor: Initial wealth W0
Action: Portfolio choice P (investment weights-vector wP )
Outcome: Wealth after one period W = W0 [1 + RP ].
Utility Function: u(W ) : [0, ∞) −→ <
Quantitative measure of outcome value to investor.
Expected Utility: E [u(W )] = E [u(W0 [1 + Rp ])]

Portfolio Theory
Utility Theory
Utility Functions
Basic properties:
u 0 (W ) > 0: increasing (always)
u 00 (W ) < 0: decreasing marginal utility (typically)
Defnitions of risk aversion:
00
Absolute Risk Aversion: λA (W ) = − uu0 (W
(W )
)
00
Relative Risk Aversion: λR (W ) = − Wu (W )
u 0 (W )
If u(W ) is smooth (bounded derivatives of sufficient order),
u(W ) ≈ u(w∗ ) + u 0 (w∗ )(W − w∗ ) + 12 u 00 (w∗ )(W − w∗ )2 + · · ·
= (constants) + u 0 (w∗ )[W − 12 λA (w∗ )(W − w∗ )2 ] + · · ·
Taking expectations
E [u(W )] ∝ E [W − 21 λ(W − w∗ )2 ] ≈ E [W ] − 21 λVar [W ]
(setting w∗ = E [W ])
Portfolio Theory
Utility Functions
Linear Utility: u(W ) = a + bW , b>0
Quadratic Utility: u(W ) = W − 21 λW 2 , λ > 0,

( and W < λ−1 )
Exponential Utility: u(W ) = 1 − e −λW , λ > 0
Constant Absolute Risk Aversion (CARA)
Power Utility: u(W ) = W (1−λ) , 0 < λ < 1
Constant Relative Risk Aversion (CRRA)
Logarithmic Utility: u(W ) = ln(W )

Portfolio Theory
Outline
1 Portfolio Theory

Portfolio Theory
Long Only:
w : wj ≥ 0, ∀j
Holding Constraints:
Li ≤ wi ≤ Ui
where U = (U1 , . . . , Um ) and L = (L1 , . . . , Lm ) are upper and
lower bounds for the m holdings.
Turnover Constraints:
∆w = (∆w1 , . . . , ∆wm )
The change vector of portfolio holdings satisfies
|P
∆wj | ≤ Ui , for individual asset limits U
m
i=1 |∆wj | ≤ U∗ , for portfolio limit U∗

Portfolio Theory

Benchmark Exposure Constraints:
wB the fractional weights of a Benchmark portfolio
RB = wB R, return of Benchmark portfolio
(e.g., S&P 500 Index,PNASDAQ 100, Russell 1000/2000)
|w − wB | = m i=1 |[w − wB ]i | < UB
Tracking Error Constraints:
For a given Benchmark portfolio B with fractional weights wB ,
compute the variance of the Tracking Error
TEP = (RP − RB ) = [w − wB ]R
var (TEP ) = var ([w − wB ]R)
= [w − wB ]0 Cov (R)[w − wB ]
= [w − wB ]0 Σ[w − wB ]
Apply the constraint:
var (TEP ) = [w − wB ]0 Σ[w − wB ] ≤ σ̄TE
2 .
Portfolio Theory
Risk Factor Constraints:
For Factor Model P

Ri,t = αi + K
k=1 βi,k fj,t + i,t
Constrain P
Exposure to Factor k
| mi=1 βi,k wi | < Uk ,
NeutralizePexposure to all risk factors:
| mi=1 βi,k wi | = 0, k = 1, . . . , K
Other constraints:
Minimum Transaction Size
Minimum Holding Size
Integer Constraints
Portfolio Theory
General Linear and Quadratic Constraints

For
w : target portfolio
x = w − w0 : transactions given current portfolio w0
wB : benchmark portfolio
Linear Constraints: Specify m-column matrices Aw , Ax , AB
and m-vectors uw , ux , uB and constrain
Aw w ≤ uw
Ax x ≤ ux
AB (w − wB ) ≤ uB
Quadratic Constraints: Specify m × m-matrices Qw , Qx , QB
and m-vectors qw , qx , qB and constrain
w0 Qw w ≤ qw
x0 Qx x ≤ qx
0
(w − wB ) QB (w − wB ) ≤ qB
Portfolio Theory
Outline
1 Portfolio Theory

Portfolio Theory

Sample Means and Covariance
Motivation
Least squares estimates
Unbiased estimates
Maximum likelihood estimates under certain Gaussian
assumptions
Issues:
Choice of estimation period
Impact of estimation error (!!)
Alternatives
Apply exponential moving averages
Apply dynamic factor models
Conduct optimization with alternative simple models
Single-Index Factor Model (Sharpe)
Constant correlation model
Portfolio Theory
Outline
1 Portfolio Theory

Portfolio Theory
When specifying a portfolio P by wP , such that

RP = w0P R, with asset returns R ∼ (α, Σ).
consider optimization problems replacing the portfolio variance
with alternatives
Mean Absolute Deviation:
0
MAD(RP ) = E (|w P(R p − α)|)
m
= E (| i=1 wi (Ri − αi )|)
Linear programming with linear/quadratic constraints
Semi-Variance:
SemiVar (Rp ) = E min(Rp − E [Rp ], 0)2

Down-side variance (probability-weighted)

Portfolio Theory
Value-at-Risk(VAR): RiskMetrics methodology developed by JP

Morgan. VaR is the magnitude of the percentile loss which
occurs rarely, i.e., with probability (= 0.05, 0.01, or 0.001)
VaR1− (Rp ) = min{r : Pr (Rp ≤ −r ) ≤ }
Tracking and reporting of risk exposures in trading portfolios
VaR is not convex, or sub-additive, i.e,
VaR(RP1 + RP2 ) ≤ VaR(RP1 ) + VaR(RP2 )
may not hold (VaR does not improve with diversification).
Conditional Value-at-Risk (CVar): Expected shortfall, expected tail
loss, tail VaR given by
CVaR1− (Rp ) = E [−RP | −RP ≥ VaR1− (Rp )]
See Rockafellar and Uryasev (2000) for optimization of CVaR

Portfolio Theory
Coherent Risk Measures A risk measure s(·) for portfolio return

distributions is coherent if it has the following properties:
Monotonicity: If RP ≤ RP 0 , w .p.1, then s(RP ) ≥ s(RP 0 )
Subadditivity: s(RP + RP 0 ) ≤ s(RP ) + s(RP 0 )
Positive homogeneity: s(cRP ) = cs(RP ) for any real c > 0
Translational invariance: s(RP + a) ≤ s(RP ) − a, for any real a.
N.B.
Var (Rp ) is not coherent ( not monotonic)
VAR is not coherent (not subadditive)
CVaR is coherent.

Portfolio Theory
Risk Measures with Skewness/Kurtosis

Consider the Taylor Series expansion of the u(W ) about
w∗ = E (W ), where W = W0 (1 + RP ) is the wealth after one
period when initial wealth W0 is invested in portfolio P.
u(W ) = u(w∗ ) + u 0 (w∗ )(W − w∗ ) + 21 u 00 (w∗ )(W − w∗ )2
1 (3) 1 (4)
+ 3! u (w∗ )(W − w∗ )3 + 4! u (w∗ )(W − w∗ )4
+O[(W − w∗ )5 ]
Taking expectations
E [u(W )] = u(w∗ ) + 0 + 12 u 00 (w∗ )var (W )
1 (3) 1 (4)
+ 3! u (w∗ )Skew (W ) + 4! u (w∗ )Kurtosis(W )
+O[(W − w∗ ) ]5
Portfolio Optimization with Higher Moments

Max: E (RP ) − λ1 Var (RP ) + λ2 Skew (RP ) − λ3 Kurtosis(RP )
Subject to: w0 1m = 1, where RP = w0 RP
Portfolio Theory
Portfolio Optimization with Higher Moments
Notes:
Higher positive Skew is preferred.
Lower even moments may be preferred (less dispersion)
Estimation of Skew and Kurtosis complex: outlier sensitivity;
requires large sample sizes.
Optimization approaches
Multi-objective optimization methods.
Polynomial Goal Programming (PGP).

MIT OpenCourseWare
http://ocw.mit.edu
18.S096 Topics in Mathematics with Applications in Finance

Fall 2013
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.

MIT18 S096F13 Lecnote14

Uploaded by

Copyright:

Available Formats

MIT18 S096F13 Lecnote14

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

MIT18 S096F13 Lecnote14

Uploaded by

Copyright:

Available Formats

Portfolio Theory

Lecture 14: Portfolio Theory

MIT 18.S096 Portfolio Theory 1

MIT 18.S096 Portfolio Theory 2

Markowitz Mean-Variance Analysis (MVA)

Markowitz Mean Variance Analysis

Evaluate different portfolios w using the mean-variance pair of the

MIT 18.S096 Portfolio Theory 4

Risk Minimization Problem

Risk Minimization Problem

Equivalent Optimization Problems

MIT 18.S096 Portfolio Theory 8

Mean-Variance Optimization with Risk-Free Asset

Mean-Variance Optimization with Risk-Free Asset

Mean-Variance Optimization with Risk-Free Asset

Mean-Variance Optimization with Risk-Free Asset

Portfolio return: RP = w00 R + (1 − w00 1m )r0

MIT 18.S096 Portfolio Theory 12

MIT 18.S096 Portfolio Theory 13

Tobin’s Separation Theorem: Every optimal portfolio invests in

Let P be the optimal portfolio for target expected return α0 with

Mean Variance Optimization with Risk-Free Asset

Mean Variance Optimization

MIT 18.S096 Portfolio Theory 17

Von Neumann-Morgenstern Utility Theory

Rational portfolio choice must apply preferences based on

MIT 18.S096 Portfolio Theory 18

Linear Utility: u(W ) = a + bW , b>0

Quadratic Utility: u(W ) = W − 21 λW 2 , λ > 0,

MIT 18.S096 Portfolio Theory 20

MIT 18.S096 Portfolio Theory 21

Portfolio Optimization Constraints

MIT 18.S096 Portfolio Theory 22

Portfolio Optimization Constraints

Portfolio Optimization Constraints

Risk Factor Constraints:

For Factor Model P

General Linear and Quadratic Constraints

MIT 18.S096 Portfolio Theory 26

Estimating Return Expectations and Covariance

MIT 18.S096 Portfolio Theory 28

Alternative Risk Measures

When specifying a portfolio P by wP , such that

Down-side variance (probability-weighted)

MIT 18.S096 Portfolio Theory 29

Alternative Risk Measures

Value-at-Risk(VAR): RiskMetrics methodology developed by JP

MIT 18.S096 Portfolio Theory 30

Alternative Risk Measures

Coherent Risk Measures A risk measure s(·) for portfolio return

MIT 18.S096 Portfolio Theory 31

Risk Measures with Skewness/Kurtosis

Portfolio Optimization with Higher Moments

Portfolio Optimization with Higher Moments

MIT 18.S096 Portfolio Theory 33

18.S096 Topics in Mathematics with Applications in Finance

You might also like