So I hooked up with Kent Smetters and Jan Walliser, who were both at the Congressional Budget Office, to add intragenerational heterogeneity to the model. We decided to follow very closely Fullerton and Rogers’ innovative approach of specifying 12 different human capital ability groups within each cohort. Kent and Jan did the yeoman’s job of reprogramming the A-К Model, after which we started writing papers on social security reform (Kotlikoff, Smetters, and Walliser 1997, 1998a, 1998b, and 1998c). During this period, I was also talking with Da i Altig who, together with Chuck Carlstrom, had done simulation work on a model that featured non differentiable budget constraints where the non differentiabilty arose from the discrete brackets of the federal income tax. I suggested that David team up with Alan, me, Kent, and Jan to work on a simulation study of U.S. tax reform that included non differentiable budget constraints as well as intragenerational heterogeneity. This collaboration resulted in Altig, et. al. (1997).
Before pointing out key lessons learned, illustrating the model’s current capacities, and outlining further improvements needed in the model’s structure, let me say a couple of words about the model’s acceptance by the economics profession. The Reaction to the A-К Model Reaction to the A-К Model has generally been quite favorable. Stiglitz, who discussed on our first paper at the 1981 simulation conference, referred to it as “a tour de force.” And Lucas (1990) gave the work a strong endorsement.
But other economists voiced concerns. Some felt that simulating policies was intrinsically different from using the calculus to sign and size their effects. Some referred to the model as a “black box ” whose results could neither be verified nor fully understood. And some (e.g., Kehoe and Levine 1985) questioned whether our model had a unique equilibrium (a unique transition path) and suggested that our fixing of a year by which the economy reaches a steady state might be forcing the model to pick one out of a potentially very large number of transition paths. I’ve always viewed the concerns as misplaced.
At a conceptual level, it’s hard to understand how finding an exact solution to a set of equations and, therefore, being able to determine the precise response to potentially large policy changes could be worse than using the calculus. Taking and evaluating derivatives is only valid for very small policy changes. Furthermore, the evaluation of derivatives of any economic variable at any point in time with respect to some policy change requires, in general, knowing the initial (pre-policy-change) values of all the economic variables at that point in time.