Fast Estimation of Structural Economic Models with Fixed Effects

We propose a novel approximate fixed effects (AFE) estimator which makes it computationally feasible to estimate non-linear economic models with fixed effects. We study the asymptotic behavior of the AFE estimator and show it converges to the standard fixed effects (FE) estimator under mild regularity conditions. Monte Carlo results indicate a fast convergence rate. The AFE estimator relies on interpolation of the FE criterion function, which greatly reduces the number of times the underlying economic model needs to be solved. In the case of dynamic programming models this can reduce the estimation time from days to minutes. To showcase our estimator, we estimate a buffer-stock consumption-saving model on high quality Danish register data allowing for unrestricted heterogeneity in the discount factor. We find substantial variation in the discount factor across invidiuals.