Macroeconomics, ECON6023 Instructor: Serhiy Stepanchuk Homework 1 Due Monday, 23 November 2020 (by 11:59 PM) Weights: Q1=10%, Q2=90% 1. Consider the following matrix: P = 0.95 0.05 0.000.15 0.75 0.10 0.10 0.00 0.90 (a) Can this matrix be used as a transition matrix of a Markov chain? Please explain. (b) Suppose that pi′0 = [0.1, 0.8, 0.1]. Compute pi ′ 1. (c) Can you prove that the Markov chain with a transition matrix P has a unique stationary distribution and that the process is asymp- totically stationary? (d) Compute the stationary distribution by iterating on the transition matrix. (e) Compute the stationary distribution directly, as a fixed point of pi′ = pi′P . Is your answer here the same as in part (d)? Is this to be expected? 2. Consider the stochastic version of the consumption-savings problem. The dynamic optimization problem that the consumer solves has the following sequential formulation: max {ct(zt),at+1(zt)}∞t=0 E0 ∞∑ t=0 βt ct(z t)1−γ 1− γ s.t.: ∀zt ct(z t) + at+1(z t) ≤ at(zt)(1 + r) + wt(zt) ct(z t), at+1(z t) ≥ 0 log(wt) = ρ log(wt−1) + εt, εt ∼ N(0, σ2) where u(c) = c 1−γ 1−γ is the per-period utility function, r is the interest rate on savings (which we assume to be constant), and wt is the wage income which we assume is random and follows an AR(1) process. 1 (a) Provide the recursive formulation of the above dynamic optimiza- tion problem. What is/are the state variable(s)? (b) First, assume that γ = 1 so that the per-period utility function becomes u(c) = log(c). Furthermore, assume that wt = 0, which means that the consumer does not have any wage income, and the problem becomes deterministic. Solve this problem using the method of undetermined coefficients (Hint: guess that the value function has the following form: V (a) = b + d log(a), where b and d are the unknown coefficients). (c) Next, assume that γ = 2, β = 0.95, r = 0.02, ρ = 0.8 and σ = 0.12. Solve the dynamic optimization problem numerically, discretizing the state space and using the ‘value function iterations’ method. Some suggestions: – Use the Tauchen method to approximate wt with a Markov chain with n = 5. – Discretize the state space for a using Ad = [0, A¯]. Try A¯ = 40. Use Na = 300 equally spaced grid points over Ad. Make sure that you get a′ = g(ai, w) < ai for all ai ∈ Ad and all w during each iteration of the algorithm. Otherwise, increase A¯. (d) Plot your solutions for the value function and policy functions for c and a′. (e) What is the value of E(a) predicted by the model? Some suggestions: – Generate a random draw of the realizations of w of length 100000. – Use the generated path for wt and the policy function for a ′ to generate the path for at+1. You can use any a0 ∈ Ad. I would suggest choosing a0 close to E(w), but you can experiment and see how robust your results are with respect to your choice of a0. – Discard the first 10% of the generated at+1. Use the remaining observations to compute E(a). (f) How does the predicted E(a) change under the following scenarios: (1) β increases from 0.95 to 0.965, (2) β remains equal to 0.95, but the interest increases to r = 0.03, (2) β remains equal to 0.95, r remains equal to 0.02, but the stan- dard deviation of the wage shocks increases to σ = 0.18, Can you provide intuition for your results in each of the three cases above? 2
欢迎咨询51作业君