Speaker:
Doug McKee
Department of Economics
UCLA

Date/time:
March 9th, 2pm

Location:
SDSC auditorium

Title:
Forward Thinking and Family Support: Explaining Retirement Behavior in Indonesia

Abstract:
In developed countries, most workers depend on government-based social security systems and private company-based pensions to support them during retirement. In much of the developing world, the mechanisms are quite different. In the absence of significant formal institutional support, older individuals rely on their own labor income and on family support in the form of transfer payments, coresidence, and participation in family businesses. However, the dramatic gains in life expectancy and declines in family size and coresidence that accompany development suggest that these traditional forms of support may break down. In this paper I build and estimate a structural dynamic model of labor supply for older men in Indonesia that incorporates these mechanisms.

Estimating this model requires advanced computational methods and resources. To fully exploit the distributional assumptions made in the model, I use Simulated Maximum Likelihood estimation. The parameters that best fit the model are those that maximize the likelihood function. Computing the likelihood function for a given set of parameters requires solving a discrete choice dynamic programming problem with ten choices, a large state space, and nine independent identically distributed (iid) shocks in each period. I use backwards recursion to solve the problem and use Monte Carlo integration to take expectations over the shocks. Because computing analytic derivatives of the model's likelihood function is not feasible, and computing accurate numeric derivatives is very expensive, I use a combination of two derivative-free optimization methods. First, I use a parallel simulated annealing algorithm to search the whole parameter space and find the right neighborhood for a solution. Second, I use asynchronous parallel pattern search (APPSPACK) to identify a local maximum. While direct search methods in general require more function evaluations to converge to a solution than derivative-based methods, the ability to run each algorithm in parallel more than makes up for the "loss" in efficiency.

I use the estimated model to simulate the effects of demographic change on labor supply and to evaluate two possible public pension reforms, which may be necessary to address the growing needs of an aging population in a developing setting. The results show that families and health play a key role in labor supply choices in old age. When faced with declining family support, simulations show that older men stay in the labor force at ages when they would otherwise exit. Moreover, when pension benefits are extended to private sector workers, people switch from unemployment and self-employment into the private sector even when wages are taxed to pay for the pension program. These results highlight the complexities of old age labor supply and pension reform in the context of rapidly developing societies.