Heterogeneity in economics

• 1. Introduction

  ›

  Heterogeneity in economics begins with a simple observation: people are different. Economists have long known this, yet for decades many of the most influential models in the field proceeded as if every consumer, every worker, every firm were essentially the same person.

  That shortcut has a name: the representative agent. It is a powerful simplification, and in many situations it works well. But the question at the heart of this documentary is when it breaks down, and what happens when researchers try to build models that treat people as genuinely distinct from one another.

  Why does heterogeneity matter enough to require its own set of tools? What goes wrong when you ignore it? And how have economists learned to handle a world where no two agents are alike?

• 2. The Representative Agent Problem

  ›

  Individual demand can be added up to market demand, but only under a specific condition. Economists have established that this aggregation works if and only if individual preferences take what is called the Gorman polar form, or equivalently satisfy linear and parallel Engel curves.

  When that condition holds, it does not matter that consumers differ from one another. You can sum their individual demands and treat the result as if it came from a single aggregate agent. The diversity of the population collapses into a clean, manageable fiction.

  The trouble is that many important economic questions cannot be addressed with that fiction intact. Some phenomena only emerge when the model actually tracks differences across individuals. For those questions, researchers need a heterogeneous agent model, and building one introduces a set of challenges that economists are still working to solve.

• 3. Unobserved Heterogeneity

  ›

  Statistical inference in economics runs into a particular hazard when the units being studied differ from one another in ways that researchers cannot measure. In econometrics, if variables that are unobserved but relevant happen to be correlated with the variables that are observed, the statistical conclusions drawn from that data can be wrong.

  The problem is not ignorance alone. It is specifically the correlation between unobserved and observed variables that causes the inferences to go astray. A researcher who ignores this risks drawing conclusions that look solid on paper but do not reflect the true relationships in the data.

  Economists have developed several methods to handle this. The instrumental variables method is one approach. Multilevel models, which include both fixed effects and random effects specifications, offer another path. A third tool is the Heckman correction, designed specifically to account for selection bias, the pattern that arises when who or what ends up in a dataset is not random.

• 4. Expectations and Model Types

  ›

  How a heterogeneous agent model is solved depends heavily on what the model assumes about how agents form their expectations about the future. This single design choice divides the landscape of these models into two broad camps.

  If agents in the model are assumed to have adaptive expectations, adjusting their beliefs based on experience rather than optimizing over all possible futures, the model falls into the category of agent-based computational economics, sometimes called ACE. Artificial financial markets are one example of this approach in practice.

  If agents are instead assumed to have rational expectations, forming beliefs that are consistent with the model itself, the model becomes a dynamic stochastic general equilibrium model, known as a DSGE. DSGE models with heterogeneous agents are especially difficult to solve. For much of the history of DSGE research, the field focused instead on representative agent versions precisely because they were tractable. Only recently has the heterogeneous-agent variant become a widespread topic of research.

• 5. Solving the Hard Problem

  ›

  Heathcote, Storesletten, and Violante, writing in the American Economic Journal: Macroeconomics in 2009, found a way to preserve an analytical solution for the general equilibrium while still allowing certain dimensions of heterogeneity. Their approach relies on making convenient functional form assumptions that keep the mathematics manageable.

  Krusell and Smith, publishing in the Journal of Political Economy in 1998, took a different route. They allowed the distribution of wealth across agents to be essentially arbitrary, a major concession to realism, but then showed that prices and other equilibrium variables behave approximately as functions of just the mean, or of a small number of other statistics of that distribution.

  Algan, Allais, and den Haan, also writing in 2009, addressed the distribution differently still. Rather than summarizing it with a few statistics, they approximated it at every point in time using a parameterized distributional form. Reiter, in the Journal of Economic Dynamics and Control in 2009, and Mertens and Judd in a working paper from 2011, pushed further, developing perturbation methods capable of approximating how a distribution evolves over time under arbitrary distributional forms. The progression from Krusell and Smith to these later methods reflects how much the technical frontier has shifted since 1998.