One-Period Portfolio Choice Prof. Fernando Chague
2016
Definitions and Setup
There are i = [1, ..., n] risky assets Each cost pi at T = 0 but entitles the owner with a random payoff I I
˜i = R
x˜i pi
is the (gross) return ˜ i = x˜i −pi is the rate of return ˜ri = 1 − R pi
There is one risk-free asset with certain return at T = 1: Rf h i ˜ i − Rf Risk premium of a risky asset is E R
One-Period Problem The allocation problem: max
E [u (w ˜ )]
[θ1 ,...,θn ] n
s.t.
∑ θi pi = w0
i=1
n
s.t. w ˜ = y˜ + ∑ θi x˜i i=1
where: I I I
θi is the quantity of shares of asset i purchased w0 is the initial wealth y˜ is a source of random income at T = 1 (i.e. labor income)
Remarks: I I
No short-selling constraints No solvency constraints,w ˜ >0
One-Period Problem The allocation can be expressed in terms of the “volume” (of consumption good) invested φi = θi pi : max
E [u (w ˜ )]
[φ1 ,...,φn ] n
s.t.
∑ φi = w0
i=1
n
˜i s.t. w ˜ = y˜ + ∑ φi R i=1
or in portfolio weights πi = θi pi /w0 = φi /w0 : max
E [u (w ˜ )]
[π1 ,...,πn ] n
s.t.
∑ πi = 1
i=1
n
˜i s.t. w ˜ = y˜ + w0 ∑ πi R i=1
First-order Condition The allocation problem can be solved using the Lagrangean: !# " n # " n max [θ1 ,...,θn ,γ]
L = E u y˜ + ∑ θi x˜i
−γ
i=1
First-order condition: E u 0 (w ˜ ) x˜i − γpi
∑ θi pi − w0
i=1
= 0
n
(abusing notation: w ˜ = y˜ + ∑i=1 θi∗ x˜i ) Which can be rewritten as: 0 u (w ˜) E x˜i = pi γ
First-order Condition
Most asset-pricing formulas are variations on the first-order condition: u 0 (w ˜) E x˜i = pi γ
This can be rewritten as: I I
I
E
h
˜) ˜ u 0 (w γ Ri
i
= 1, if pi > 0 (gross returns) i h ˜i − R ˜ j = 0, if pi > 0 and pj > 0 (long-short portfolio, E u 0 (w ˜) R zero h cost portfolio,iexcess return) ˜ i − Rf ˜) R = 0, if pi > 0 and ∃ Rf E u 0 (w
Stochastic Discount Factor
Definition: A Stochastic Discount Factor (SDF) is any random variable m ˜ such that
E [m ˜ x˜i ] = pi
∀i
The proof of the existence of a SDF is also called the Fundamental Theorem of Asset Pricing
Stochastic Discount Factor If there are only k states of the nature ω1 , ..., ωk , then: k
∑ m˜ (ωj ) x˜i probj = pi
j=1
where probj is the probability of state j. Arrow-Debreu Security: I I I
Payoff: Pays 1 unit of consumption at state ωj and 0 units otherwise Price: qj If a SDF exists, then qj = m ˜ (ωj ) probj and m ˜ (ωj ) = qj /probj
m is the price of a unit of the consumption good in each state per unit of probability Other names for SDF: State Price Density, Pricing Kernel. Statistically, m ˜ is the Radon-Nikodym derivative of the set function that assigns prices to events (set of states) with respect to the probability of events.
Stochastic Discount Factor
A great portion of the asset-pricing literature is dedicated to finding SDFs Intersection with macroeconomics: impose economic structure and find equilibrium ˜ ) /γ We already imposed one such structure: m ˜ = u 0 (w I
From risk-aversion we have ∂ m/∂ ˜ w ˜ < 0: investors discount more (less) states with high (low) wealth.
Let’s solve for γ. For that, consider now the problem including c0 :
Stochastic Discount Factor The allocation problem including c0 : max
E [v (c0 , c˜1 )]
[θ1 ,...,θn ,c0 ]
n
n
s.t. c0 + ∑ θi pi = w0 and c˜1 = y˜ + ∑ θi x˜i i=1
i=1
The FOCs are:
∂ E v (c0 , c˜1 ) = γ ∂ c0 ∂ E v (c0 , c˜1 ) x˜i = γpi ∀i ∂ c1 This implies a SDF: ∂ ˜1 ) ∂ c1 v (c0 , c
m ˜= E
h
∂ ˜1 ) ∂ c0 v (c0 , c
i
Stochastic Discount Factor Assuming time-additive utility v (c0 , c1 ) = u0 (c0 ) + u1 (c1 ) then: E u10 (˜ c1 ) x˜i = pi u00 (c0 ) I
The expected marginal gain must be equal the marginal cost of saving at t = 0
If we simplify further: u0 (c0 ) = u (c0 ) and u1 (c1 ) = δ u (c1 ) then 0 δ u (˜ c1 ) E x˜i = pi u 0 (c0 ) This is the standard Consumption CAPM SDF: m ˜=
δ u 0 (˜ c1 ) u 0 (c0 )
Stochastic Discount Factor and Returns If pi > 0 then:
h i E m ˜ R˜i = 1
Note that by linearity of expectations and portfolio returns ˜ p = ∑ni=1 πi R ˜ i this is also true for portfolios (note πi is not random) R h i E m ˜ R˜P =
n
∑ πi = 1
i=1
Similarly, for zero-cost long-short portfolio it equals zero: i h E m ˜ R˜i − R˜j = 0
Stochastic Discount Factor and Returns We can also relate SDF with risk-premia: h i 1 = E m ˜ R˜i h i = cov m, ˜ R˜i + E [m] ˜ E R˜i Provided that E [m] ˜ > 0: h i E R˜i −
1 E [m] ˜
= −
cov m, ˜ R˜i E [m] ˜
If a risk-free asset exists, then E [mR ˜ f ] = 1 implies E [m] ˜ Rf = 1 and: h i E R˜i − Rf = −Rf cov m, ˜ R˜i
Stochastic Discount Factor and Returns
h i E R˜i − Rf = −Rf cov m, ˜ R˜i Remarks: Risk premia has to do with covariances! I I I
“you discount more assets that pay off when wealth is high” “you dislike assets that covary with your wealth” “you require a higher risk-premium for assets that covary a lot with the SDF”
Likewise: I I I
“you discount less assets that pay off when wealth is low” “you like assets that offer you protection when wealth is lower” ...
Portfolio Allocation (Two Assets) ˜ ∼ µ, σ 2 and Rf . Assume no y˜ . R ˜ + φf Rf Allocation problem reduces to finding [φ , φf ]. Now w ˜ = φR Since w0 = φ + φf , set φf = w0 + φ , and the allocation problem simplifies to: h i ˜ + (w0 − φ ) Rf max E u Rφ φ
FOC:
h i ˜ − Rf = 0 E u 0 (w ˜) R
Remarks (when E [R] − Rf > 0 and risk aversion): I I
h i ˜ − Rf > 0 φf = w0 is never optimal, since u 0 (w ˜ )E R In fact, it can be shown that φ ∗ > 0 (this is only true when n = 1!)