Prelude to Chapter 7

GENERATING PATTERNS IN
THE TIMING OF RETIREMENT

 

 

WHAT IS THE connection between individual rationality and aggregate efficiency? And what is the role of local interactions and social networks in determining that connection? Regarding the first question, the opening Generative chapter argues that individual rationality is neither necessary nor sufficient for the attainment of macroscopic efficiency. The two are logically independent; neither implies the other. The retirement model furnishes the necessity half of the independence proof: a society of autonomous agents arrives at the economically optimal retirement behavior even though the overwhelming majority do not optimize individually. More prosaically, the invisible hand does not require rational fingers. In my own mind, the other half of the independence proof—individual rationality is not sufficient for macro efficiency—is given in the trade chapter of Growing Artificial Societies, where agents do maximize utility in the orthodox fashion, under evolving preferences. Equilibrium is not attained despite orthodox optimization of utility functions that are themselves orthodox at all times (Cobb-Douglas algebraically).

Turning to its specifics, the retirement model exhibits many of the core themes of the Generative chapter. Here, agents are heterogeneous by age, by social network, and by retirement status. Social interactions are local with most agents playing a coordination game (retire vs. work) with others in their network. In answer to the second question posed initially, local interactions in networks is the mechanism whereby overall optimality is attained in our population of predominantly nonoptimizing individuals. One novel feature of this model, however, is that these networks change over time—they are transient. Bounded rationality is evident in that most agents simply imitate within their dynamic network, or play a random strategy, rather than optimizing in any economic sense. That few agents optimize is, of course, consistent with a wealth of data from psychology and experimental economics. The model thus aims to provide a more plausible microfoundation for an important macroeconomic phenomenon than the optimizing representative agent picture.

The research was motivated by an empirical puzzle brought to our attention by Brookings colleagues Henry Aaron and Gary Burtless. In truth, neither Rob Axtell nor I was thinking about retirement economics at all. But Henry and Gary quickly convinced us that there was an empirical challenge here and, more intriguing, a promising area for agent-based modeling. Although the model concerns stylized facts, and is “quasi-empirical,” I dare say that, with reference to the U.S. data, it is stronger on the observed dynamics of retirement norms than the neoclassical efforts to date.