Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd Logo
Upload

Welcome to Scribd!

  • 22 views

    EC744 Lecture Note 8 Applications of Stochastic DP: Prof. Jianjun Miao

    Uploaded by

    binicle
    1. The document summarizes stochastic dynamic programming applications to consumption-savings problems and investment problems with adjustment costs. 2. For consumption-savings with exponential utility, the optimal saving rule depends on human wealth and financial wealth. For investment with adjustment costs, the marginal Q satisfies an asset pricing equation and investment is determined by convex or non-convex adjustment costs. 3. Recursive utility models are discussed, including expected utility, Kreps-Porteus preferences, and multiple priors models. These can be used in asset pricing applications like the Lucas model.

    Copyright:

    Attribution Non-Commercial (BY-NC)
    Download as PDF, TXT or read online from Scribd

    EC744 Lecture Note 8 Applications of Stochastic DP: Prof. Jianjun Miao

    Uploaded by

    binicle
    22 views31 pages
    1. The document summarizes stochastic dynamic programming applications to consumption-savings problems and investment problems with adjustment costs. 2. For consumption-savings with exponential utility, the optimal saving rule depends on human wealth and financial wealth. For investment with adjustment costs, the marginal Q satisfies an asset pricing equation and investment is determined by convex or non-convex adjustment costs. 3. Recursive utility models are discussed, including expected utility, Kreps-Porteus preferences, and multiple priors models. These can be used in asset pricing applications like the Lucas model.

    Original Title

    ec744note08

    Copyright

    © Attribution Non-Commercial (BY-NC)

    Available Formats

    PDF, TXT or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    1. The document summarizes stochastic dynamic programming applications to consumption-savings problems and investment problems with adjustment costs. 2. For consumption-savings with exponential utility, the optimal saving rule depends on human wealth and financial wealth. For investment with adjustment costs, the marginal Q satisfies an asset pricing equation and investment is determined by convex or non-convex adjustment costs. 3. Recursive utility models are discussed, including expected utility, Kreps-Porteus preferences, and multiple priors models. These can be used in asset pricing applications like the Lucas model.

    Copyright:

    Attribution Non-Commercial (BY-NC)
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    22 views31 pages

    EC744 Lecture Note 8 Applications of Stochastic DP: Prof. Jianjun Miao

    Uploaded by

    binicle
    1. The document summarizes stochastic dynamic programming applications to consumption-savings problems and investment problems with adjustment costs. 2. For consumption-savings with exponential utility, the optimal saving rule depends on human wealth and financial wealth. For investment with adjustment costs, the marginal Q satisfies an asset pricing equation and investment is determined by convex or non-convex adjustment costs. 3. Recursive utility models are discussed, including expected utility, Kreps-Porteus preferences, and multiple priors models. These can be used in asset pricing applications like the Lucas model.

    Copyright:

    Attribution Non-Commercial (BY-NC)
    Download as PDF, TXT or read online from Scribd
    Download as pdf or txt
    of 31

    EC744 Lecture Note 8 Applications of Stochastic DP

    Prof. Jianjun Miao

    Consumption-Savings with Exponential Utility

    ec . The income follows an AR(1) process Let U (c) =

    yt = yt1 + y + t, where t is iid with a standard normal distribution. The budget constraint is given by at+1 = (1 + r) at + yt ct, a0 given, The Bellman equation V (a, y) = max U (c) + E V subject to
    h i 0, y 0 |y a

    a0 = (1 + r) a + y c.

    Conjecture 1 V (a, y) = exp (r (a + Ay + b)) , r where A and b are constants to be determined. By the envelope condition 1 V1 (a, y) . 1+r Thus, the consumption rule is given by U 0 (c) =

    1 c (a, y) = r a + Ay + b + log (1 + r) . r Thus, a0 = (1 + r) a + y r 1 a + Ay + b + log (1 + r) r 1 = a + (1 rA) y rb log (1 + r)


    !

    Substitute them back into the Bellman equation to derive

    1 exp (r (a + Ay + b)) r !! 1 1 = exp r a + Ay + b + log (1 + r) r " !! 1 1 +E exp r a + (1 rA) y + (1 r) b log (1 + r) + Ay 0 r Note that y 0 = y + y + . Matching coecients yields A = 1 , 1+r " 2# (rA) A y 1 . b = 2 r log (1 + r) + log ( (1 + r)) + r r 2

    Dene human wealth

    and nancial wealth w = (1 + r) a. Then

    Ht = Et

    yt+s (1 + r)s s=0

    c (w, y) = r (w + H) + constant. Friedman (1957) Permanent Income Hypothesis Optimal saving rule s (a, y) = ra + y c ! 1 y r 2 log ( (1 + r)) = y + + 1+r 1 r 2 (1 + r )2 Can be more general income process. See Wang (2004, AER).

    Investment with Adjustment Costs

    Cooper and Haltiwanger (2005) V (A, K) = max (A, K) C (I, A, K) p (I) I + E V


    I

    subject to K 0 = (1 ) K + I

    A0, K 0

    where (A, K) = AK and A follows a Markov process

    comes from static choice 0 < 1, monopoly power or returns to scale


    0 0 = p Neoclassical theory (no adjustment cost), E V2 A , K

    Convex adjustment costs (q theory) C (I, A, K) = (I/K)2 K and p (I) = p. 2 Let Q be the Lagrange multiplier for the capital accumulation equation. FOC for I : K I= (Q p) FOC for K 0 :
    h

    Q = E V2

    A0, K 0

    Envelope condition: V2 (A, K) = 2 (A, K) C3 (I, A, K) + (1 ) Q Q satises an asset pricing equation:


    i 0, K 0 C I 0, A0, K 0 + (1 ) Q0 Q = E 2 A 3 h

    If = 1, V2 (A, K) = V (A, K) /K =constant. Marginal Q vs average Q

    Nonconvex adjustment costs where

    o i (A, K) , V a (A, K) V (A, K) = max V

    h i a (A, K) = max (A, K) F K pI + E V A0, K 0 V I

    h i i (A, K) = (A, K) + E V A0, K 0 V

    Without shocks, there are two triggers of capital. When the current capital stock is too high, then disinvest to bring capital to a certain level. When the current capital stock is too low, then invest to bring capital to that level. When the current capital is within this region, no investment occurs and capital depreciates. With shocks, the two triggers are state dependent.

    Transactions Costs p (I) = pb if I > 0, p (I) = ps if I < 0, pb > ps.


    o i (A, K) , V b (A, K) , V s (A, K) V (A, K) = max V n

    where

    h i s (A, K) = max (A, K) p I + E V A0, K (1 ) + I V s I<0 h i i (A, K) = (A, K) + E V A0, K (1 ) V

    h i b (A, K) = max (A, K) p I + E V A0, K (1 ) + I V b I>0

    Recursive Utility: Backus, et al (2004)

    3.1

    Risk (Static): Machina (1987)

    Independence Axiom: P P P + (1 ) Q P + (1 ) Q The Allais Paradox: a1 : ($1M, 1) versus a2 : ($5M, 0.10; $1M, 0.89; $0, 0.01) and a3 : ($1M, 0.11; $0, 0.89) versus a4 : ($5M, 0.10; $0, 0.90) EU: a1 () a2 a3 () a4 Experiments: a1 a2 but a4 a3

    Common consequence eect b1 : x + (1 ) P versus b2 : P + (1 ) P and b3 : x + (1 ) P versus b4 : P + (1 ) P

    where P involves outcomes both greater and less than x and P stochastically dominates P . Experiments: b1 b2 but b4 b3 Allais Paradox: = 0.11, x = $1M, P = ($5M, 10/11; $0, 1/11) , P = ($0, 1) , P = ($1M, 1) .

    Common ratio eect c1 : ($X, p; $0, 1 p) versus c2 : ($Y, q; $0, 1 q) and c3 : ($X, rp; $0, 1 rp) versus c4 : ($Y, rq; $0, 1 q) where p > q, X < Y, rp (0, 1) . EU: c1 () c2 c3 () c4 Experiments: c1 c2 but c4 c3

    Disappointment Aversion

    c1 : ($200, 1; $0, 0) versus c2 : ($300, 0.8; $0, 0.2) and c3 : ($200, 0.5; $0, 0.5) versus c4 : ($300, 0.4; $0, 0.6) Experiments: c1 c2 but c4 c3 Mixture: 0.5 (200, 1; 0, 0) + 0.5 0 versus 0.5 (300, 0.8; 0, 0.2) + 0.5 0

    Ambiguity: the Ellsberg Paradox An urn contains 30 red balls and 60 white and black balls R B W 10 0 0 0 10 0 10 0 10 0 10 10

    f g f0 g0 EU: f g f 0 g 0 Experiments: f g but g 0 f 0

    Chew-Dekel class: The certainty equivalent function for a set of payos and probailities {c (z) , p (z)} is dened implicitly by a risk aggregator M satisfying: = z p (z) M (c (z) , ) . M satises the following properties: a. M (m, m) = m. b. M (, ) is increasing. c. M (, ) is concave. d. M (kc, km) = kM (c, m) .

    Expected utility. Let M (c, m) = cm1/ + m (1 1/) , 1 Then = (z p (z) c (z))1/ Weighted utility (Chew (1983)). Let M (c, m) = (c/m) cm1/ + m (1 (c/m) /) where either (i) 0 < < 1 and + < 0 or (ii) < 0 and 0 < + < 1. Then
    P + z p (z) c (z) = P p (z) c (z) z

    Disappointment aversion (Gul (1991)). Let cm1/ + m (1 1/) 1 m) M (c, m) = cm1 + m 1 1 + (c m =


    X

    for c m for c < m

    0. Then where

    p (z) c (z) ,
    !

    p (z) =

    1 + I [c (z) < ] P 1 + x p (x) I [c (x) < ]

    Indierence Curves and Certainty Equivalents Two states c (1) = 1 , c (2) = 1 + . Taylor expansion: (EU ) ' 1 (1 ) 2/2 (W U ) ' 1 (1 2) 2/2 4 + 4 (1 ) 2 (DA) = 1 2 + 4 + 4 + 2 2 First-order risk aversion Homework: Derive these certainty equivalents

    3.2

    Time and Risk

    General recursive utility (Epstein and Zin (1989)): Vt (c) = W (ct, t (Vt+1 (c))) where t () is the generalized certainty equivalent

    Separation between risk aversion and intertemporal substitution.


    i 1/ , Vt (c) = (1 ) ct + t (Vt+1 (c)) h

    where 1/ (1 ) is EIS

    Expected utility CE t (x) = u1 (Et [u (x)]) ,

    Kreps-Porteus preferences: t (x) = (Et [x])1/ , where 1 is CRRA Expected utility: = . Weil (1993, ReStud): t (x) = log (Et [exp (x)]) /. where is CARA

    Recursive multiple-priors utility t (x) = 1

    QQ

    inf Et [ (x)]

    Let (x) = x/ and = . Then, we have (Epstein and Wang (1994)): Vt (c) = inf
    QQ

    i ] 1/ + E Q [V (1 ) ct t+1 (c) t

    Recursive smooth ambiguity model (Klibano, Marinacci and Mukerjee (2006), Ju and Miao (2007)) Suppose there is an unknown parameter .
    n o 1 E t() u1 E P [u (x)] t (x) = . t

    u describes risk aversion, describes ambiguity aversion. If u1 is linear, then it reduces to the KP-EZ model

    Divergence CE (Maccheroni, Marinacci and Rustichini (2006))

    Divergence index: : R++ R strictly concave, (1) = 0 (1) = 0, 00 (1) = 1,


    x

    lim 0 (x) =

    For (u, , ) ,

    Q 1 E Qu (x) + E t+1 |F , Q = dQ | t (x) = min u t t F t t Q Q dP t t

    Entropic CE (Hansen and Sargent (2008)) (x) = x log x x + 1 Reduce to expected utility CE t (x) = ( u)1 Et [ u (x)] where u (u) = 1 exp

    The entropic CE is the only divergence CE that is also an expected utility CE.

    3.3

    Asset Pricing Application

    Lucas (1978) model with recursive utility.


    i + J a0, z 0 1/ J (a, z) = max (1 ) c c,w h

    subject to

    a0 = (a c) Let rp =
    Pn i=1 wiri z, z h
    0

    i=1

    n X

    wiri

    z, z 0

    be the portfolio return.

    Conjecture J (a, z) = aL (z) . Then we have


    i + (1 b) L z 0 r z, z 0 1/ L (z) = max (1 ) b p b,w

    where b = c/a.

    The portfolio FOC:


    X
    z0

    h i h i 0|z M L z 0 r z, z 0 , L z 0 r z, z 0 r z, z 0 = 0 p z p 1 i j

    (1)

    for any two assets i and j. Consumption FOC:


    1 = (1 b)1 L z 0 r z, z 0 . (1 ) b p

    (2)

    Solve for and substitute into the scaled Bellman equation, we obtain
    0 r z, z 0 L z p = [(1 ) /]1/ [b/ (1 b)](1)/ , L (z) = (1 )1/ b(1)/.

    (3) (4)

    Using the preceding two equations and the homogeneity of , we have


    0r z, z 0 h p =1

    where we dene

    L z0 0 = h (L (z 0) rp (z, z 0)) L z0 = [(1 ) /]1/ [b/ (1 b)](1)/ = [(1 ) /]1/ [b/ (1 b)](1)/
    (1 )1/ b0 (1)/

    Using the fact that c0 1b 0 b0a0 = = b0 rp, c ba b whe have


    ! 1/ 1 0 1 0 = c 0 0 h rp rp

    So we obtain:

    !1 1/ c0 0 h0rp z, z 0 = rp = 1.

    (5)

    Similarly, using portfolio FOC


    i 0r 0 , 1 h0 r 0 r 0 |z = 0. E M1 h p i j h

    (6)

    The above two equations are the Euler equations.

    Let aggregate output be a Markov process


    0 = y 0/y = g z 0 g

    In equilibrium, optimal portfolio is a claim to the stream of output. Let the price be q. Then
    0 0 0 0 0 0 0 = q + y = Q y + y = g0 Q + 1 . rp

    Qy

    Now we solve Q (z) using


    h i1/ Q0 + 1 = Q1/. g0

    We can then compute returns of any other assets.

    IID case. Using (2) and (4), we have


    1 = r 0 . (1 b) p

    If rp is iid, and b are constants. Thus, we cannot identify and for a given risk parameter. Homework: Derive asset pricing implications for the KP-EZ specications. What is the bond return?

    You might also like