The Rise of Computationally Enabled Economics: Introduction to the Special Issue of the Eastern Economic Journal on Agent-Based Modeling

Abstract

The first digital computers were primarily used by scientists and engineers to solve mathematical equations numerically, that is, to approximate analytical solutions, most commonly for difficult-to-solve differential equations. The economics profession was also an early adopter of digital computing, and many of the first uses of computation by economists involved numerical solution of economic equations that were hard or impossible to solve analytically. In this, the computer served as a powerful calculation engine while the problems being solved were conceptually similar to those that had come before, that is, constrained optimization. Much of this work had a heavy normative flavor, as companies had great financial incentive to learn how to better manage their inventories, optimize assembly lines, forecast demand for their products, and so on. Indeed, Herbert Simon, one of the early pioneers in this area, was fond of recalling that his work on dynamic programming in the 1950s taught companies how to better manage their production processes. It was very clear to that generation of economists that firm behavior was far from optimal, a fact that, Simon lamented, later generations seemed to have forgotten. In his Machine Dreams: How Economics Became a Cyborg Science, Philip Mirowski (2001) masterfully charts the early adoption of digital computer technology by economists and closely affiliated researchers (in game theory and operations research primarily, including many refugees from physics). Indeed, as operations researchers learned to solve wide classes of optimization problems computationally, through so-called mathematical programming methods (e.g., linear programming, integer programming, dynamic programming), economists were early users of such techniques for practical and policy problems. Monte Carlo methods were systematized early-on in the digital era, primarily for their use in evaluating difficult to solve integrals. This led to the development of reasonably sophisticated pseudo-random number generation algorithms and the rise of modern simulation techniques. Already in the 1950s, the so-called microsimulation was being used by economists and policy makers as a way to better forecast and understand alternative economic policies (Orcutt et al. 1961). This use of computation was somewhat different from the mathematical programming that had come before. While still involving optimization, such calculations were done at