Structural Labor Supply Models when Budget Constraints are Nonlinear

Despite the advantages of structural labor supply methods, many recent papers cite the arguments of MaCurdy et al. as a reason for avoiding structural methods, and instead use simple estimation methods such as differences in differences. This paper examines the role of economic assumptions in structural labor supply methods and how some of the assumptions may be relaxed. We first show the sources of inconsistency in the local linearization method. We then examine the standard approach generally attributed to Hausman, and show that this approach relies on the convexity of preferences in the construction of the likelihood function. We show that the criticisms of MaCurdy et al. can be reinterpreted as showing where in the estimation method the assumption of maximization of convex preferences is enforced. We provide a formal argument that if observed preferences are nonconvex, but the estimation method does not allow for nonconvexity, then estimated parameters may not satisfy the Slutsky restrictions, as has often been found in empirical work. Our simulations show that this event occurs for a substantial range of parameter values. Other Monte Carlo experiments show the sensitivity of estimates to functional form assumptions, the stochastic specification, and the hours endowment. We discuss how other generalizations of the Hausman method dispense with utility maximization altogether. Finally, we show that the standard methods in the literature do not permit estimation of parameters consistent with nonconvex preferences, and describe methods that allow for less restrictive assumptions.