An exploration of evolutionary methods in macroeconomics.

摘要

The dissertation introduces two evolutionary methods that can be used to solve heterogeneous and homogeneous agent stochastic dynamic programming problems. A third solution method, used for homogeneous agent models and based on viral evolution, produces forward looking and discounting behavior that closely mirrors experimental evidence without assuming that agents are forward looking.; The first evolutionary method attempts to replicate a known result to establish the viability of evolutionary methods in solving representative agent problems. Two test problems, Hansen's indivisible labor model and the cash-in-advance extension of the indivisible model, are solved by both a simple evolutionary method and linearized optimality conditions. The results are nearly identical, thus establishing the usability of evolutionary methods.; The second demonstration of evolutionary methods is applied to a heterogeneous agent problem, specifically a version of the indivisible labor model without unemployment insurance. This more complex demonstration shows the economic results that without unemployment insurance agents will save more, and that the predicted variability of macroeconomic aggregates is more similar to the US economy than the original model. It also shows the viability of using evolutionary methods in more complex environments. One of the most intriguing results is that two species of agent evolved, "capitalists" and "workers", and the composition of these two groups can be explained as an evolutionarily stable equilibrium.; The final demonstration takes the solution method out of the background and brings it forward as a model of learning. Learning is modeled as an infection, which jumps from person to person. The rate of infection mimics individual discount rates and induces savings behavior on its own. It is shown that the apparent discount rate, the combination of the agents' true discount rate and the infection rate, decreases over time and approaches the agents' true discount rate. This decrease, known as hyperbolic discounting, is consistent with what is observed in psychology studies, while the limiting case is consistent with market level observations. This model closes the gap between individual and market level observations of discounting behavior without explicitly assuming the two kinds of discounting nor relying on commitment mechanisms.