DYNAMIC STRUCTURAL MODELS OF RETIREMENT AND DISABILITY John Rust, University of Maryland and NBER Moshe Buchinsky, UCLA and NBER Hugo Benítez-Silva, SUNY--Stony Brook February 2003 SUMMARY We propose to use all available waves of the Health and Retirement Survey (HRS) and AHEAD Survey to estimate a comprehensive dynamic programming (DP) model of behavior at the end of the life cycle that provides a detailed treatment of the Social Security Administration's (SSA) Old Age and Survivors (OASI), Supplemental Security Income (SSI) and Disability Insurance (DI) programs. Major changes to these programs are being contemplated. Yet, we currently lack a unified model of social insurance at the end of the life cycle that can help us evaluate the behavioral and distributional impacts of these policies. Particular attention is paid to developing, estimating, and testing a multi-stage dynamic programming (DP) model of the SSI and DI application, appeal, and award process, for (possibly) heterogeneous agents. We are developing a tractable empirical model that captures an individual's decisions regarding (1) labor supply and retirement (2) application for OA, DI and SSI benefits, and (3) consumption and savings. The resulting model will allow us to derive predictions of the behavioral and welfare implications of policy changes. While there is a large literature using reduced-form and static structural models that has investigated some of these issues, it suffers from two major shortcomings. First, reduced-form models cannot be used for welfare analysis or to predict behavior responses to policy changes. Second, static structural models do not accurately reflect the level of complexity and uncertainty facing individual decision makers, nor do they capture the important dynamic elements of the decision processes. The DP model we are developing will circumvent these shortcomings, providing a tractable framework for analyzing individual behavior and well-being, and forecasting their response to a wide range of policy changes. Our model could provide new insights into a number of puzzling aspects about disability in the U.S. One puzzle is to determine the factors responsible for the pronounced swings in DI incidence rates in recent years. Another puzzle is to determine why the fraction of Americans receiving SSI and DI benefits continues to increase despite overwhelming epidemiological evidence of steady improvements in various objective indicators of health status. The SSA is currently contemplating significant changes to the disability award process, in order to reduce delays, and reduce large unexplained state-level differences in award rates. We will use detailed health and functional status indicators from the HRS to evaluate whether or not there are alternative screening rules that can reduce the level of classification errors in the DI award process. Our estimated DP model will produce detailed predictions of the behavioral and welfare effects of changes in benefit levels, delays, the probability of being awarded benefits, and the probability that a DI beneficiary will be audited. This framework will allow us to develop methodologies for characterizing efficient policies, i.e., those that minimize the expected discounted cost of providing a stream of social insurance benefits subject to the constraint that individuals' expected discounted utilities are at least as high as under the status quo.