Heterogeneity in Macroeconomics

Abstract

In order to study aggregate issues, macroeconomists typically use representative agent models, in which aggregates are the choice of a fictitious average agent in the economy. By construction, these models ignore the tremendous heterogeneity across individual agents documented in microeconomic data. My dissertation addresses this gap by developing tools to incorporate micro heterogeneity into macro models, and using these tools to study an important aggregate issue.

In the first chapter, I develop a new method to efficiently solve models with micro heterogeneity. The main challenge is that individual agents base their decisions on the entire distribution of agents, an infinite-dimensional object. I approximate this distribution using a finite-dimensional parametric family. I show how to easily implement the method using Dynare, a publicly available software package commonly used to solve representative agent models.

In the second chapter, I use this method to incorporate realistic firm-level investment behavior into a model of aggregate investment. I find that including the extensive margin of whether a firm invests or not has important implications for business cycles and countercyclical stimulus policy. First, aggregate investment is less responsive to productivity shocks in recessions than in expansions, because in recessions fewer firms are likely to make an extensive margin investment. Second, the policy multiplier also falls in recessions. Third, a simple size-dependent policy, which targets extensive margin investment, is five times more cost effective than existing size-independent policies.

In the third chapter, I develop a new method to solve heterogeneous agent models in continuous time. Continuous time models are promising in the heterogeneous agent context because they easily handle nonconvexities, like borrowing constraints or fixed costs, which are important for describing micro data. Following Achdou et al. (2015), I use a finite difference method, which approximates individual decisions and the distribution over a finite grid. I solve for the dynamics of these functions using local perturbation methods. I find that the method is very efficient at solving a continuous time version of Krusell & Smith (1998), suggesting that it is able to solve more complicated models which are intractable in discrete time.