14 Quick summary Basics: 1. Value? To an investor. ∙ 0 ¸ u (ct+1 ) xt+1 pt = Et β 0 u (ct ) p = E(mx) u0 (ct+1 ) m = β 0 u (ct ) 2. Alternative expressions and special cases basic p = E(mx) riskfree Rf = 1/E(m) E(x) risk corrections, price discount p = + cov(m, x) Rf expected returns E(R) = Rf − Rf cov(m, R) beta models = Rf + βλ; λ = −Rf var(m) µ ¶ dΛ f continuous time Et (dRt ) = rt dt − Et dRt Λ 3. Economics rf = 1 − γ (γ + 1) σ 2 | 2 {z } intertemporal substitution. precautionary saving µ ¶ dct e E(R ) = γcovt dRt , ct δ + γμ | {z }c Risk (measured by covariance), risk aversion 4. Equity premium/risk free rate puzzle. kEt (dRte )k = γσ t σ t (dRte ) µ dct ct ¶ ρ (a) SR = 0.5 needs huge (50) risk aversion (b) Worse (250) if ρ (∆c, R) = 0.2. (c) High γ means rf high (γμc dominates initially), or unbelievable balance around γ = 200. (d) High γ means rf is very sensitive to ∆c. (e) At a minimum, need high risk aversion, Epstein-Zin or habits to separate risk aversion from intertemporal substitution. 98 Representation 1. States s; Contingent claims, risk neutral probabilities. Introduction to state space geometry. 2. LOOP and Arbitrage (a) Loop (linearity) ⇔ ∃ a unique x∗ ∈ X s.t. p = E(x∗ x)∀x ∈ X. (b) Loop + No arbitrage ⇔ ∃m > 0 s.t. p = E(mx). (c) Formulas for x∗ x∗ = p0 E(xx0 )−1 x ¤ −1 1 1 £ f x∗ = − E(R) − R Σ [R − E(R)] Rf Rf £ ¤ −1 dΛ∗ f f = −r dt − E(dR) − r dt Σ dR (σdz) Λ∗ 3. Mean variance frontier (a) m is minimum second moment return (b) mvf implied by p = E(mx). kE(Rte )k σ(Rte ) < σ(m) E(m) Cone. (c) Hansen-Richard. i. R∗ = x∗ /p(x∗ ); Re∗ = proj(1|Re ) ii. all returns R = R∗ + wRe∗ + n iii. mvf: Rmv = R∗ + wRe∗ 4. MVF ⇔ p = E(mx) ⇔ E(Re ) = βλ. All three representations are equivalent. (a) m, x∗ , R∗ , Rmv = R∗ + wRe∗ all are reference variables for ER − β. (b) x∗ = proj(m|X), R∗ = E(x∗ )/E(x∗2 ) on mvf (c) Rmv on mvf → can construct m (m = a + bRmv ) (d) f : E(Re ) = β f λ then m = b0 f (e) f = Rmv satisfies beta representation 5. Conditioning information (a) Condition down: pt = Et (mt+1 xt+1 ) ⇒ E(p) = E(mx) (b) Add instruments=managed portfolios.; In principle this is sufficient for all conditioning information pt = Et (mt+1 xt+1 ) ⇔ E(zt pt ) = E [mt+1 (xt+1 zt )] (c) Models with time t parameters do not necessarily condition down. R mv mt+1 = a − bft+1 ⇒ mt+1 = at − bt ft+1 on unconditional mvf ⇒ Rmv on conditional mvf Not ⇐. Don’t forget scaled portfolios in the definition of unconditional mvf. 99 (d) Scaled factors model mt+1 = a(zt ) − b(zt )ft+1 ⇔ mt+1 = a ⊗ zt − b (ft+1 ⊗ zt ) Factor Pricing Models 1. Basic idea: β u0 (ct+1 ) ≈ a + b0 ft+1 u0 (ct ) f: Market. (CAPM). News (ICAPM) Macro variables. Mimicking portfolios. APT: other big portfolios. 2. CAPM: 2 period quadratic, quadratic iid, exponential normal, log utility... 3. ICAPM: Other state variables 4. APT: use f ∗ to price portfolios ”near” f ? Works with Sharpe ratio limits. Then E(Re ) = α + βf + ε, extra sharpe ratio = α0 Σα, so small ε ⇒ small α. Estimation and testing 1. General: procedure for estimating free parameters (α, β, b, γ,etc.), standard errors, test for 0 pricing errors α̂0 V −1 α̂ 2. GMM for consumption-based models and in general. (a) agT (b) = 0 d S gT (b) = 0 0 −1 formulas for σ(b̂), cov(gT ). (Don’t have to memorize!) (b) Linear combinations from minimization min gT (b)0 W gT (b) ∂gT W gT (b) = 0 ∂b0 [d0 W ] gT (b) = 0 (Solve by numerical minimization. a = d0 W for distribution theory) (c) Two step procedure for consumption model (d) Interpretation: gT = E(mx − p) = Rf α = pricing errors. J = α0 V −1 a = test 3. GMM corrections for OLS standard errors gT (β) = E(xt (yt − βxt )) = 0 4. Time series regressions for E(Re ) = βλ 100 (a) Idea: Rtei = αi + β i ft + εit t = 1, 2, ...T ∀i., OLS estimates for α, β. λ̂ = ET (ft ) (b) OLS and GMM distributions. (c) GRS statistic for α0 V −1 α and its GMM counterpart. 5. Cross sectional regressions for E(Re ) = βλ (a) Idea. TS for βs. OLS and GLS E(Rei ) = (γ) + β i λ + αi i = 1, 2, ...N (b) Difference between CS and TS (c) Fixed β standard errors (d) GMM standard errors (e) Shanken as a special case of GMM, correct for β̂ (f) Note: If errors are not iid, (β 0 Σ−1 β)−1 β 0 Σ−1 ET (Re ) is not d0 S −1 gT (β, λ). Efficient GMM is no longer GLS cross sectional regression. 6. Fama MacBeth (a) Idea, and procedure. (b) Point: correct OLS std. errors for cross sectional correlation (c) Equivalence to pooled, XS when betas are fixed 7. No need to assume iid homoskedastic any more! Do α0 V −1 α even if FMB 8. GMM for linear factor models m = a − b0 ft ; gT = E(mRe ) = 0 or E(mR) = 1. (a) E(p) on E(fx0 ) (b) E(Re ) on E(Re f 0 )(a = 1) ³ ´ e e 0 (c) E(R ) on cov(R f ) (E f˜ = 0); but harder s.e. since Ef is estimated. 9. All work about the same 10. λ vs. b. λ is not a test for “do we need this factor to price other assets.” Options and bonds 1. Options without intermediate trading, p = E(mx) treatment (a) Call, put definitions and payoff diagrams (b) LOOP, ∃m : put-call parity (c) NA, m > 0. Arbitrage bounds. Payoff approach (A > B ⇒ P (A) > P (B)). m approach and linear program (d) ( σ 2 (m) or other restrictions can narrow arbitrage bounds. Like APT) min(max){m} C = E(mxc ) s.t.S = E(mST ); 1 = E(mRf ), m > 0, σ 2 (m) < A,... 2. Options with dynamic trading, Black Scholes 101 (a) Use Λ∗ to price xc . Λ∗ (b) Solve Λ∗ forward, C = E( ΛT0 xcT ) (c) Guess C(S, t), use E(dR) − rf = −E 3. Bonds ¡ dΛ Λ ¢ dR , to pde for C. Solve PDE (a) y,p,f, definitions. Expectations hypothesis (b) Discrete time term structure model. Example, mt+1 + δ = ρ (mt + δ) + εt+1 (n) pt = ln Et (emt,t+n ) i. Solve m forward ii. Solve p backward (c) Character of result: one factor model with constant risk premium (d) Continuous time term structure models. Form dΛ = −r(xt )dt − σ Λ (xt )dz Λ dxt = μx (xt )dt + σ x (xt )dz (e) Bond PDE derivation (f) Solution for CIR, Vasicek cases. (g) Example: setting up differential equation for a term structure option. 102 4. Big picture for APT, options, bonds, relative pricing X, prices you know m>0, arbitrage bounds Hedge portfolio b’f Small or 0 error Payoff you want to price f*, discount factor for f m with Sharpe ratio, other limts (a) If the payoff xc is in X (Black Scholes), you can price it exactly (given prices in X) by using f ∗ (Λ∗ ), p = E(f ∗ xc ) (b) If the payoff is near X — small error — it’s often priced with f ∗ anyway, assuming market price of extra risk is zero (Term structure models.) (c) Other m give other answers. For options m > 0 gives arbitrage bounds. For APT, σ(m) < h gives Sharpe ratio bounds. Small ε can lead to usefully tight bounds. 103 15 Important things you’re missing 1. Ch 20, 21 empirical work! (a) Size, B/M factors, momentum, the latest alphas, (Fama). (b) Predictability of stocks (D/P) bonds (long-short yields) and foreign exchange (if oreign − idomestic ). (c) Present P value formulas and volatility. (Campbell Shiller approximation pt − dt = Et ρj+1 (∆dt+j − rt+j )) 2. Economic models that capture Equity premium, predictability, size and B/M. (a) Habits (b) Nonrecursive Epstein-Zin etc. utility (c) Heterogeneity and incomplete markets. (Heaton!) 3. Portfolio theory! (a) Classic mean-variance (b) Merton, ICAPM hedge portfolios (c) Dynamic and multifactor portfolio theory — How to incorporate stock, bond predictability; how to incorporate size, value, momentum, etc. (d) Bayesian portfolio theory — how parameter uncertainty affects portfolios (e) Portfolio facts — the crazy things people hold. 4. Much more fixed income, risk management, options, futures, swaps, etc. 5. New things: Liquidity, short sales constraints, etc. (My 35150 reading list) 104