>> Decomposition of Stochastic Discount Factor and their Volatility Bounds 2012年11月21日 >> Framework • Motivation • Decomposition of SDF • Permanent and Transitory Bounds • Comparisons with Alvarez & Jermann (2005) • Eigenfunction and Eigenvalue Method • Asset Pricing Models Representation • Empirical Application to Asset Pricing Models 2012/12/17 Asset Pricing 1 >> Motivation • Economic Intuitions: • Explanation Inability of Equilibrium Asset Pricing Model - Various Puzzles (Return, Volatility) - Frequency Mismatch (Daniel & Marshall,1997) - Features of Investor Preference: Local Durability, Habit Persistence or Long Run Risk • Unit Root Contributions of Macroeconomic Variables • Econometric Similarity: - Beveridge-Nelson Decomposition 2012/12/17 Asset Pricing 2 >> Decomposition of SDF • No Arbitrage Opportunities in Frictionless Market if and only if a strictly positive Pricing Kernel{M }exists: t Vt ( Dt + k ) = M t+ 1 Mt Et (M t + k Dt + k ) Mt • So SDF for any gross return on a generic portfolio held from t to t + 1 æ ö M ÷ 1 = Et ççç t + 1 ×Rt + 1 ÷ ÷ ÷ M è t ø • DefineRt + 1,k as the gross return from holding from time t to t + 1 a claim to one unit of the numeraire to be delivered at time t + k 2012/12/17 Asset Pricing 3 >> Decomposition of SDF Rt + 1,k Vt + 1 ( I t + k ) = Vt ( I t + k ) • So risk-free return: Vt + 1 ( It + 1 ) 1 Rt + 1,1 = = Vt ( It + k ) Vt ( It + k ) • Long term bond return: Rt + 1,¥ 2012/12/17 Vt + 1 ( It + 1 ) = lim = lim Rt + 1,k k® ¥ V (I k® ¥ t t+ k ) Asset Pricing 4 >> Decomposition of SDF • Assumptions: - SDF and Return Jointly Stationary and Ergodic - There is a number such b that 0 < lim k® ¥ - For each Vt ( I t + k ) < ¥ , for all t k b t+ 1 there is a random variable M t + 1 Vt + 1 ( I t + 1+ k ) £ xt + 1 t+ 1 k b b with 2012/12/17 Et finite xt + 1 for all xt + 1 that such a.s. k Asset Pricing 5 >> Decomposition of SDF • Unique Decomposition (Alvarez & Jermann,2005) Mt = MtP MtT and: Et M t + k M = lim k® ¥ b t+ k P t b t+ k M = lim k® ¥ V (I t t+ k ) T t with: Et M 2012/12/17 P t+ 1 =M P t Asset Pricing Rt + 1,¥ M tT = T M t+ 1 6 >> Decomposition of SDF Proof : Rt + 1,¥ Et + 1M t + k Et + 1M t + k / b t + k V (I ) M t+ 1 M t+ 1 = lim t + 1 t + k = lim = lim t+ k k® ¥ V (I k® ¥ k® ¥ E M E M / b ) t t + k t t+ k t t+ k Mt Mt Et + 1M t + k / b t + k M tP+ 1 lim k® ¥ M t+ 1 M t+ 1 M tT = = = t+ k P Et M t + k / b Mt M tT+ 1 lim k® ¥ Mt Mt • How to link transitory component to Long term bond? • No cash flow uncertainty 2012/12/17 Asset Pricing 7 >> Permanent and Transitory Bounds 2012/12/17 Asset Pricing 8 >> Permanent and Transitory Bounds 2012/12/17 Asset Pricing 9 >> Permanent and Transitory Bounds • Inequality (6) bounds the variance of the permanent component of the SDF, useful for understanding what time-series assumptions are necessary to achieve consistent risk pricing across multiple asset markets 2 • pc is receptive to an interpretation as in Hansen & Jagannathan (1991) bound: R t 1 log(R t 1 ) log( Rt 1, ) Rt 1, • So pc2 can be interpreted as the maximum Sharpe ratio, but relative to the long-term bond 2012/12/17 Asset Pricing 10 >> Permanent and Transitory Bounds 2012/12/17 Asset Pricing 11 >> Permanent and Transitory Bounds 2012/12/17 Asset Pricing 12 >> Permanent and Transitory Bounds • The transitory component equals the inverse of the gross return of an infinite-maturity discount bond and governs the behavior of interest rates • The quantity on the right-hand side of equation (9) is tractable and computable from the return data. And the T bound in (9) is a parabola in E M t T1 , tc2 space. Mt 2 • tc is positively associated with the square of the Sharpe ratio of the long-term bound. • (9) to assess the bound market implications of asset pricing models. 2012/12/17 Asset Pricing 13 >> Permanent and Transitory Bounds 2012/12/17 Asset Pricing 14 >> Permanent and Transitory Bounds 2012/12/17 Asset Pricing 15 >> Comparisons with Alvarez & Jermann (2005) • In Alvarez & Jermann, L-measure (entropy) a random variable u as a measure of volatility: L[u] log[ E (u)] E (log[u]) • One-to-one correspondence exists between variance and L-measure when u is log-normally distributed 1 L[u ] Var[log(u )] 2 • Such discrepancies between the two measures can get magnified under departures from log-normality. 2012/12/17 Asset Pricing 16 >> Comparisons with Alvarez & Jermann (2005) 2012/12/17 Asset Pricing 17 >> Comparisons with Alvarez & Jermann (2005) 2012/12/17 Asset Pricing 18 >> Comparisons with Alvarez & Jermann (2005) 2012/12/17 Asset Pricing 19 >> Comparisons with Alvarez & Jermann (2005) 2012/12/17 Asset Pricing 20 >> Comparisons with Alvarez & Jermann (2005) 2012/12/17 Asset Pricing 21 >> Comparisons with Alvarez & Jermann (2005) 2012/12/17 Asset Pricing 22 >> Eigenfunction and Eigenvalue Method • Continuous Time Version (Luttmer,2003): • Consider State-Price Process: • Suppose: dLt = - rt dt + s L ,t dWt Lt L tP = lim Et [L T ] T® ¥ • For Any t > 0 , and Et + t [L T ] is bounded for all T , the dominated convergence theorem implies that P Et [L tP+ t ] = Et éêlim Et + t [L T ]ù = lim E [ L ] = L t ú ëT ® ¥ û T® ¥ t T 2012/12/17 Asset Pricing 23 >> Eigenfunction and Eigenvalue Method P L • The process t is referred to as the permanent component of SDF • Define L Tt to be the residual, So: L t = L t L t P • And suppose: T d L tP = L tPs P,t dWt dLTt = LTt [(- rt - s P,t (s L ,t - s P,t )dt + (s L ,t - s P,t )dWt ] • As we all know, it also can be decomposed: éM t + k ù - rk µt ( D ) ê Vt ( Dt + k ) = Et Dt + k ú= e E t+ k ê Mt ú ë û M t+ k dQ Þ dP = e- rk dQ Þ M t = e- rt Mt dP 2012/12/17 Asset Pricing 24 >> Eigenfunction and Eigenvalue Method • So How to Decompose? What’s b ? • Hansen & Scheinkman (2009, Econometrica) • Let L be a Banach space, and let {M : t ³ 0} be a family of operators on L . If: 1, M0 = I, Mt + s = Mt Ms for all s, t ³ 0 2, Positive if for any t ³ 0, Mt y ³ 0 whenever y ³ 0 3, For eachy Î L , E(Mt y ( X t ) | X 0 = x) Î L Then {M : t ³ 0} is a semi-group. 2012/12/17 Asset Pricing 25 >> Eigenfunction and Eigenvalue Method • Consider General Multiplicative Semi-group: t : ( x) E[Mt ( X t ) | X 0 x] • Extended Generator: a Boral function y belongs to the domain of the extended generator A of the multiplicative function M t if there exists ta Boral function c such that Nt = M t y ( X t ) - y ( X 0 ) - ò0 M s c ( X s )ds is a local martingale wrt. filtration I t . In this case, the extended generator assigns function c to y and write c = Ay • Associates to function y a function c such that Mt c ( X t ) is the expected time derivative of Mt ( X t ) 2012/12/17 Asset Pricing 26 >> Eigenfunction and Eigenvalue Method • A Borel function is an eigenfunction of the extended generator A with eigenvalue if . • Intuitively, So if is an eigenfunction, the expected time derivative of M t ( X t ) is M t ( X t ). Hence, the expected time derivative of exp(t )Mt ( X t ) is zero. • How to get ? 1 d ( M t ( X t )) E dt M t ( X t ) • Expected time derivative is zero 2012/12/17 Asset Pricing Local Martingale 27 >> Eigenfunction and Eigenvalue Method 2012/12/17 Asset Pricing 28 >> Eigenfunction and Eigenvalue Method • 6.1Proof: let Yt M t ( X t ) , so: Nt = M t y ( X t ) - y ( X 0 ) - ò t 0 M s c ( X s )ds Þ dNt = dYt - r Yt dt Þ dYt = dNt + r Yt dt • And: t exp(- r t )Yt - Y0 = - t t ò r exp(- r s)Y ds + ò exp(- r s)dY = ò exp(- r s)dN 0 s 0 s 0 s • Interpretation: - : Growth rate of multiplicative functional - ( ( X 0 )) / ( ( X t )): Transient or Stationary Component ¶: tMartingale Component, Distort the drift M - 2012/12/17 Asset Pricing Mt 29 >> Eigenfunction and Eigenvalue Method • Further more: ( X t ) E[ M t ( X t ) | X 0 x] exp( t ) ( x) E | X 0 x (Xt ) • If we treat exp(t ) ( X t ) as a numeraire, similar to the risk-neutral pricing in finance. • Decomposition Existence and Uniqueness is given in Proposition 7.2 (Hansen & Scheinkman,2009) • Congruity of Bakshi & Chabi-Yo Decomposition M t 1 1 v (X0 ) e M t exp( t ) M t , Et e (Xt ) M t M t 1 M t 1 e let : v exp( ), M t MtT vt Mte , MtP Mt / MtT ( Xt ) 2012/12/17 Asset Pricing 30 >> Eigenfunction and Eigenvalue Method • Example: consider a multiplicative process M exp( A) : dAt ( f X t f o X to )dt X t f f dBt f o dBto • And X f , X o : dX t f f ( x f X t f )dt X t f f dBt f dX to o ( xo X to )dt odBto • Guess an eigenfunction of the form ( X t ) exp(c f X f d (ln M t ) d (ln ( X t )) d (ln( M t ( X t ))) 2012/12/17 Asset Pricing co X o ) 1 d ( M t ( X t )) E dt M t ( X t ) 31 >> Eigenfunction and Eigenvalue Method 2012/12/17 Asset Pricing 32 >> Eigenfunction and Eigenvalue Method ¶ t define a new probability measure, resulting distorted •M drift of X f , X o : Xf : Xo : 2012/12/17 Asset Pricing 33 >> Asset Pricing Models Representation • Consider the modification of the long-run risk model proposed in Kelly (2009). • The distinguishing attribute: the model incorporates heavy-tailed shocks to the evolution of nondurable consumption growth (log), governed by a tail risk state variable t . 2012/12/17 Asset Pricing 34 >> Asset Pricing Models Representation • While the transitory component of SDF is distributed log-normally, the permanent component of SDF and SDF itself are not log-normally distributed. • The non-gaussian shock Wg are meant to amplify the tails of the permanent component of SDF and SDF. 2012/12/17 Asset Pricing 35 >> Asset Pricing Models Representation 2012/12/17 Asset Pricing 36 >> Asset Pricing Models Representation 2012/12/17 Asset Pricing 37 >> Asset Pricing Models Representation 2012/12/17 Asset Pricing 38 >> Asset Pricing Models Representation 2012/12/17 Asset Pricing 39 >> Empirical Application to Asset Pricing Models 2012/12/17 Asset Pricing 40 >> Empirical Application to Asset Pricing Models 2012/12/17 Asset Pricing 41 >> Empirical Application to Asset Pricing Models 2012/12/17 Asset Pricing 42 >> Empirical Application to Asset Pricing Models 2012/12/17 Asset Pricing 43 >> Empirical Application to Asset Pricing Models 2012/12/17 Asset Pricing 44 >> Empirical Application to Asset Pricing Models 2012/12/17 Asset Pricing 45 >> Empirical Application to Asset Pricing Models 2012/12/17 Asset Pricing 46 >> Empirical Application to Asset Pricing Models 2012/12/17 Asset Pricing 47 >> Thank you for listening and Comments are welcome. 2012年11月21日