Solutions Methods in DSGE (Open) models

Similar documents
Equilibrium Conditions (symmetric across all differentiated goods)

Approximation around the risky steady state

Optimal Simple And Implementable Monetary and Fiscal Rules

Dynare Class on Heathcote-Perri JME 2002

1. Constant-elasticity-of-substitution (CES) or Dixit-Stiglitz aggregators. Consider the following function J: J(x) = a(j)x(j) ρ dj

Bayesian Estimation of DSGE Models: Lessons from Second-order Approximations

Economics 701 Advanced Macroeconomics I Project 1 Professor Sanjay Chugh Fall 2011

Learning and Global Dynamics

Stochastic simulations with DYNARE. A practical guide.

Lecture 3, November 30: The Basic New Keynesian Model (Galí, Chapter 3)

A Model with Collateral Constraints

1 Bewley Economies with Aggregate Uncertainty

The Small-Open-Economy Real Business Cycle Model

Lecture 15. Dynamic Stochastic General Equilibrium Model. Randall Romero Aguilar, PhD I Semestre 2017 Last updated: July 3, 2017

1 The Basic RBC Model

Graduate Macro Theory II: Notes on Quantitative Analysis in DSGE Models

First order approximation of stochastic models

Applications for solving DSGE models. October 25th, 2011

Comprehensive Exam. Macro Spring 2014 Retake. August 22, 2014

High-dimensional Problems in Finance and Economics. Thomas M. Mertens

DYNARE SUMMER SCHOOL

Small Open Economy RBC Model Uribe, Chapter 4

A Discussion of Arouba, Cuba-Borda and Schorfheide: Macroeconomic Dynamics Near the ZLB: A Tale of Two Countries"

Solving a Dynamic (Stochastic) General Equilibrium Model under the Discrete Time Framework

Indeterminacy and Sunspots in Macroeconomics

4- Current Method of Explaining Business Cycles: DSGE Models. Basic Economic Models

Chapter 6. Maximum Likelihood Analysis of Dynamic Stochastic General Equilibrium (DSGE) Models

Introduction to Numerical Methods

Closed economy macro dynamics: AD-AS model and RBC model.

Research Division Federal Reserve Bank of St. Louis Working Paper Series

Expectations, Learning and Macroeconomic Policy

Demand Shocks with Dispersed Information

Computing first and second order approximations of DSGE models with DYNARE

Getting to page 31 in Galí (2008)

Taylor Rules and Technology Shocks

Incomplete Markets, Heterogeneity and Macroeconomic Dynamics

Lecture 6, January 7 and 15: Sticky Wages and Prices (Galí, Chapter 6)

Neoclassical Business Cycle Model

Second-order approximation of dynamic models without the use of tensors

New Notes on the Solow Growth Model

Indeterminacy and Sunspots in Macroeconomics

Monetary Economics Notes

Macroeconomics Theory II

Optimal Inflation Stabilization in a Medium-Scale Macroeconomic Model

Time-varying Consumption Tax, Productive Government Spending, and Aggregate Instability.

Master 2 Macro I. Lecture notes #12 : Solving Dynamic Rational Expectations Models

General Examination in Macroeconomic Theory SPRING 2013

DSGE-Models. Calibration and Introduction to Dynare. Institute of Econometrics and Economic Statistics

problem. max Both k (0) and h (0) are given at time 0. (a) Write down the Hamilton-Jacobi-Bellman (HJB) Equation in the dynamic programming

New Keynesian Macroeconomics

Global Value Chain Participation and Current Account Imbalances

Macroeconomics II. Dynamic AD-AS model

Dynamic AD-AS model vs. AD-AS model Notes. Dynamic AD-AS model in a few words Notes. Notation to incorporate time-dimension Notes

MA Advanced Macroeconomics: 7. The Real Business Cycle Model

The Basic New Keynesian Model. Jordi Galí. November 2010

Graduate Macroeconomics - Econ 551

The welfare cost of energy insecurity

Chapter 11 The Stochastic Growth Model and Aggregate Fluctuations

International Trade Lecture 16: Gravity Models (Theory)

Foundation of (virtually) all DSGE models (e.g., RBC model) is Solow growth model

Dynamic stochastic general equilibrium models. December 4, 2007

(a) Write down the Hamilton-Jacobi-Bellman (HJB) Equation in the dynamic programming

The Basic New Keynesian Model. Jordi Galí. June 2008

optimal simple nonlinear rules for monetary policy in a new-keynesian model

A Modern Equilibrium Model. Jesús Fernández-Villaverde University of Pennsylvania

Monetary Economics. Lecture 15: unemployment in the new Keynesian model, part one. Chris Edmond. 2nd Semester 2014

ECON 5118 Macroeconomic Theory

Real Business Cycle Model (RBC)

Research Division Federal Reserve Bank of St. Louis Working Paper Series

Solving Nonlinear Rational Expectations Models by Approximating the Stochastic Equilibrium System

Deep Habits, Nominal Rigidities and Interest Rate Rules

Imperfect Information and Optimal Monetary Policy

Monetary Economics: Problem Set #4 Solutions

Inference. Jesús Fernández-Villaverde University of Pennsylvania

Demand Shocks, Monetary Policy, and the Optimal Use of Dispersed Information

Simple New Keynesian Model without Capital

Perturbation Methods II: General Case

Simultaneous equation model (structural form)

Macroeconomics Theory II

Public Economics The Macroeconomic Perspective Chapter 2: The Ramsey Model. Burkhard Heer University of Augsburg, Germany

Modelling Czech and Slovak labour markets: A DSGE model with labour frictions

Monetary Policy with Heterogeneous Agents: Insights from Tank Models

Online Appendix for Slow Information Diffusion and the Inertial Behavior of Durable Consumption

Equilibrium Conditions and Algorithm for Numerical Solution of Kaplan, Moll and Violante (2017) HANK Model.

A simple macro dynamic model with endogenous saving rate: the representative agent model

Introduction to second order approximation. SGZ Macro Week 3, Day 4 Lecture 1

Government The government faces an exogenous sequence {g t } t=0

Monetary Policy and Exchange Rate Volatility in a Small Open Economy. Jordi Galí and Tommaso Monacelli. March 2005

CEMMAP Masterclass: Empirical Models of Comparative Advantage and the Gains from Trade 1 Lecture 3: Gravity Models

Stagnation Traps. Gianluca Benigno and Luca Fornaro

Perturbation techniques

A Dynamic Model of Aggregate Demand and Aggregate Supply

Signaling Effects of Monetary Policy

Uncertainty aversion and heterogeneous beliefs in linear models

Macroeconomics Theory II

Lecture 7. The Dynamics of Market Equilibrium. ECON 5118 Macroeconomic Theory Winter Kam Yu Department of Economics Lakehead University

The Neo Fisher Effect and Exiting a Liquidity Trap

The Ramsey Model. (Lecture Note, Advanced Macroeconomics, Thomas Steger, SS 2013)

Decentralised economies I

Nominal Rigidities, Government Spending, and Long-Run Policy Trade-Off

Transcription:

Solutions Methods in DSGE (Open) models 1. With few exceptions, there are not exact analytical solutions for DSGE models, either open or closed economy. This is due to a combination of nonlinearity and the fact that the systems are stochastic. Almost all dynamic optimization problems give rise to non-linear equations (exceptions - a small open endownment economy with deterministic interest and income stream). But this is compounded by the fact that the stochastic features of the systems. Solutions involve taking expectations of future endogenous variables.. The standard solution methods, used in RBC and IRBC models, as well as most recent generation new open macro models, involve linear approximation around a deterministic steady state. This was discussed in the previous notes. By linear approximation, we mean the first order terms in a Taylor series approximation around a steady state. Let us quickly review the idea of orders of approximation: Say that we have a function y = g(x) where in the notation to come, y will be a non-predetermined variable in a dynamic model, x is a predetermined variable. Say that there is a deterministic steady state for x given by x,sothat: y = g(x) In general, the function g is non-linear. Using a Taylor series approximation, we can approximation this function around {x} by y g(x)+g x (x)(x x)+ 1 g xx(x)(x x) + O 3 (1) This contains 3 types of terms: zero order, first order, and second order. The notation at the end signifies that the error in the approximation is of third or higher order. When evaluated up to a first order, the expected value of y is equal to its steady state value value y. But when evaluated up to a higher order, the expectedvaluedependsonthecurvatureofthefunctionandthevarianceofx. 3. Most DSGE models just analyze the first order properties of models. For characterizing the impulse response and second moments of variables, that is fine. For instance, say we were interested in finding the variance of y, whichis a second order accurate measure. Then we could use (1) to calculate var(y) =E(y Ey) = g xvar(x) Why do we not use the higher order terms? Becuase they are second order, so if we included them, we would have terms in third or higher order in calculating the variance. But the variance is only a second order accurate measure. 4. But for other purposes, first order approximation may be inaccurate. Say that we have a welfare measure for this economy, given by EU(y),andwewishto 1

evaluate expected utility of the agents based on this welfare measure and based on the distribution of x. Then we could take a second order approximation of U(y) given by U(y) U(y)+U y (y)(y y)+ 1 U yy(y)(y y) + O 3 () It is then tempting to think of evaluating expected utility using the first order approximation of y from (1) so that () gives: EU(y) U(y)+U y (y)e(y y)+ 1 U yy(y)e(y y) + O 3 (3) = U(y)+ 1 U yy(y)g xvar(x)+o 3 (4) Except this is wrong. What we are doing here is taking into account the presence of one set of second order terms in evaluating utility, but leaving out another set of second order terms. What do we mean by this? Well, note from (1) that, evaluated up to a second order, wehave Ey y + 1 g xx(x)var(x)+o 3 (5) So, when we are evaluating up to a second order, the second line of (3) should be EU(y) U(y)+U y (y) 1 g xx(x)var(x)+ 1 U yy(y)g xvar(x)+o 3 So that the correct evaluation of welfare is given by: EU(y) U(y)+ U y (y) 1 g xx(x)+ 1 U yy(y)gx var(x)+o 3 (6) Basically, what we are saying is that when when we are evaluating expected utility, we have to take a second order approximation of utility, and a second order approximation of the economic model. If we just take a second order approximation of utility and a first order approximation of the economic model, we are leaving out some relevant terms in the evaluation of utility. These terms are second order, but they are relevant, because utility is evaluated up to second order. 5. This has very wide applicability. Woodford s book documents how to use approximations to evaluate utility in a range of sticky price models. In some cases, Woodford shows that U y (y) =0, so that we can actually validly take a first order approximation of the economic model, but in general this won t be correct. 6. Lets look at two concrete examples to see how we could be led astray. The first is from Kim and Kim, JIE 004. It is based on evaluating welfare in a country endowment economy under incomplete markets, and comparing this to an economy with complete markets. Say country i utility, i =1,, is given by: U i = C1 σ i 1 σ

and endowments are stochastic equal to Y i. Assume that Y i are log-normal and E(y) =0,wherey =log(y ) var(y) =Under autarky, we have C i = Y i. Under complete markets, we have C i = Y1+Y. Now take a log approximation to utility around a steady state c to get U i C1 σ 1 σ + C1 σ bc + 1 σ C 1 σ bc where bc =(c c). If we took a linear approximation of the model around the steady state under autarky we would get bc A i = by i = y i, and under complete markets we would get bc C i =1/(y 1 + y ). So by a linear approximation we would have Ebc A i =0and Ebc C i =0. Under autarky, we have var(c) =var(y), while under complete markets, we have var(c) =1/var(y). So we might be led to the conclusion that utility is higher under autarky when σ<1. But this is based on the assumption that E(bc) =0, which is not true. Why? Because note that C i = Y 1 + Y So that exp(c i )= exp(y 1)+exp(y ) exp(ec + 1 var (c)) = exp(1 var(y)) so that Ec = 1 var(y) 1 var(c) = 1 4 var(y), sothate(c ca ) > 0. Hence, again, we need to take into account the higher order terms which affect the means of consumption. 6. Take another example. Say we have a monetary economy with sticky prices, where each agent is a yeoman farmer, producing a differentiated output and charging a monopoly price for her output based on a constant elasticity of demand. This is a very standard model so we won t get into the details. Say that there is a utility function given by: U = C1 σ i 1 σ ηl i (7) R 1 where C = C 1 1/λ j dj 1/(1 1/λ), and production of each good is one for one o with labor supply of the yeoman farmer. The budget constraint of the household R 1 is PC = P i L i,wherep = Pj dj 1/(1 λ) 1 λ is the price level, and P i is the price o charged by the farmer. Demand for the farmer s good is given by L i = µ λ Pi C P 3

where C is aggregate consumption. Say the farmer has to set her price in advance. Then her optimal price will satisfy i " µ 1 λ E C σ (1 λ) Pi i C + ηλ 1 µ 1 λ Pi C# =0 P i P P i P Imposing symmetry and market clearing, we have the condition EC 1 σ = ηλ EC (8) λ 1 Now assume that the price level is determined by the quantity theory relationship M = PC We can evaluate utility by noting from (7) and (8) that EU = 1 EC 1 σ λ 1 σ If M is log normal then we have log normal utility. This gives us an expression EU = 1 exp((1 σ)(ec + 1 (1 σ)var(c)) λ 1 σ If we took a linear approximation of the quantity theory relationship, we might guess that E(c) =0. But again this would be incorrect, because we have to take into account the necessary condition (8), which gives us Ec = φ + 1 (σ )var(c) We can show that monetary volatility increases Ec is σ>. But it still reduces welfare. 7. So we now know that we must in general take nd order approximation of DSGE models in order to do welfare evaluation. How do we go about taking second order approximations of dynamic forward looking models? The example above was simple, since it was a static function. But when we have dynamic models with stochastic endogenous variables it is not obvious how to do this. Two papers are important. Schmitt Grohe and Uribe, and Lombardo and Sutherland. The first paper gives a general approach based on a kind of undetermined coefficients solution method. The second gives a more direct approach based on standard saddle path solution methods. Both give equivalent results for any given model however. We describe the SGU approach. 8. Take a dynamic general equilibrium rational expectations model given by: E t f(y t+1,y t,x t+1,x t )=0 (9) where f is n by 1. Here y are non-predetermined variables, and y t is an n y 1 vector, x t is an n x 1 vector, where n = n x + n y. Assume that x t = 4

[x 1 t,x t ], where x 1 are endogenous predetermined variables, and x are exogenous predetermined variables (shocks), which are described by: x t+1 = Λx t + eησε t+1 Assumptions: bounded support, stationary shocks, ε is n ε 1. Given the example of the neo-classical growth model. The general solution to this model is: y t = g(x t,σ) (10) x t+1 = h(x t,σ)+ησε t+1 (11) where η = 0 η is nx n ε, g is an n y 1, andh is n x 1. The question is, how do we construct the solutions g(x t,σ) and h(x t,σ)? Clearly, we cannot in general obtain the exact solutions for these functions. Rather, we can derive approximations around a non-stochastic steady state. This is done by the method of perturbation. First, define a non-stochastic steady state by the condition: f(y,y,x, x) =0 To get the first order approximation, we may expand (10) and (11) around the steady state as follows: y t g (x, 0) + g x (x x)+g σ σ x t h (x, 0) + h x (x x)+h σ σ Then, up to a first order, we need to obtain the coefficients g x,h x,g σ,h σ. These are obtained by noting that by the definition of f,itmustbethecasethatforany initial x and σ, wehavef (x, σ) =E t (g(h(x, σ)+ησε 0 ),g(x),h(x)+ησε 0,x)=0, so that all derivatives F x i σj(x, σ) =0. Results: a) Setting F x =0,wecanderivetheexpressionsg x and h x by solving for the roots of the first order dynamic system. b) Setting F σ =0,wefind that this gives a homogeneous equation in g σ and h σ. For all x and σ, the only solution must be that g σ = h σ =0. Simple examples: Take the small economy model described by: where β (C t )=(1+C t C) ψ. C 1 t E t β(c t )C 1 t+1 = 0 B t+1 (1 + r)b t C t + Y t = 0 Y t+1 Y ρ t exp(ε t+1 ) = 0 5

In a steady state, we have: 1 1 = 0 rb C + Y = 0 Y 1 = 0 We define C t = g(b t,y t ) and B t+1 = h (B t,y t ). Then expanding the above system around the steady state, we may solve for the coefficients of g 1,g, h 1, h, etc. For instance, in the case of g b and h b, we have: (1 ψc)g b g b h b = 0 h b (1 + r) g b = 0 It is easy to see that the solution will be g b = r + ψc and h b =1 ψc. More generally, we may use the SGU code [gx,hx] = gx_hx(nfy,nfx,nfyp,nfxp) to solve for the policy functions from the first order system. c) Second order approximation: Take a -0 Taylor series approximation of g() and h(). Then substitute this into the conditions F xx =0, F xσ =0,and F σσ =0, to determine the coefficients g xx,g xσ,g σσ, h xx,h xσ,h σσ. The algebra is quite cumbersome - see the SGU paper. However, one key feature is that the resulting system is linear in the g xx, h xx coefficients. That is, there is no need to solve an eigenvalue problem at this stage. In addition, again it turns out that g xσ = h xσ =0. So, up to the second order, the coefficients of the optimal policy rules do not depend on the volatility parameter. 9. An alternative solution method for nd order approximations of DSGE models is done in Lombardo and Sutherland. Their method is more like traditional approaches to solving RE models using a foreword solution method. The key simplification that they use is that the derivation of the second order aspects of the system may be done using the first order approximation alone. Toseethisusetheexample: E t y t+1 = ay t + byt + ce t yt+1 + v t + O 3 (1) where v t+1 = ρv t + ε t+1, ρ<1, anda>1. To solve this, we may compute yt and E t yt+1 by solving the system E t y t+1 = ay t + v t + O alone, and then substituting in the solutions to (1). This is because (1) is computed up to second order accuracy only. That means a first order solution of the model is enough to compute yt and E t yt+1. 6