Jesus Fernandez-Villaverde University of Pennsylvania/NBER
This project develops new tools for the estimation of Dynamic Stochastic General Equilibrium (DSGE) models using a Bayesian approach. DSGE models are a concise and simplified representation of the economy of a country like the United States. The models start by specifying the behavior of households, firms, and the government, with a special emphasis in the dynamic aspects of the decisions of families and firms. Then, the models carefully sum up all these decisions into aggregate variables and study how the economy as a whole reacts to different shocks and to changes in fiscal and monetary policies.
DSGE models are a popular research tool to understand the U.S. economy. Moreover, an increasing number of policy-making institutions, both in the United States (the Federal Reserve Board and several of the regional Federal Reserve Banks, the International Monetary Fund) and abroad (the European Central Bank, the Bank of England, the Bundesbank, the Central Banks of Austria, Canada, Italy, Spain, and Sweden, to name a few) employ DSGE models to help in the formulation of better economic policies. Finally, economists are accumulating evidence of the good forecasting performance of DSGE models, even when compared with judgmental predictions from staff economists at the Federal Reserve System.
All these three type of exercises (research to understand the U.S. economy, model specification to help formulate economic policy, and forecasting) require the estimation of the model, i.e., to use real data to make the model "fit" the real world as well as possible. Bayesian methods are especially suitable for this task since they efficiently summarize the sample information and mix it with the prior information in a flexible way. Moreover, recent advances in computation make the implementation of the Bayesian approach straightforward, robust, and direct.
However, this estimation of DSGE models is a challenging task. DSGE models are complex structures. In addition, their statistical properties are not fully understood and economists have been forced to make simplifying assumptions that limit the applicability of the methodology.
This project develops new tools to estimate DSGE models. The unifying view of the research agenda is simple: making the estimation of DSGE models more flexible. Economists want to capture richer dynamics and relax some of the tight assumptions that they currently impose to estimate DSGE models. The project has three parts. First, it finds how to perform Bayesian estimation of Markov-switching DSGE models. Second, it shows how to undertake semiparametric Bayesian estimation of DSGE models. Third, it studies the estimation of dynamic games in macroeconomics with a semi-parametric Bayesian approach.
The newer and better tools that this proposal outlines are designed explicitly for the purpose of helping the Federal Reserve Board and other policy-making institutions develop more flexible models that will contribute to the implementation of an effective monetary policy in the United States. Finally, many of the tools outlined in the proposal have potential applications in other fields of economics (such as international economics, industrial organization, or labor economics), and other social sciences where researchers want to estimate dynamic models using flexible, yet powerful tools.