Serially Correlated Wages in a Dynamic, Discrete Choice Model of Teacher Attrition

Todd R. Stinebrickner, University of Western Ontario


This paper suggests and implements a method for dealing with the problems which are encountered during the estimation of dynamic, discrete choice models with serially correlated unobservables. The method takes advantage of Gaussian quadrature integral approximation techniques and is based on a new, non-parametric value function approximation algorithm which allows the econometrician to avoid potentially problematic functional form specifications. A desirable property of the approach is that the econometrician has complete control over the factors which ensure that parameter estimates from an approximate solution to a model with serial correlation can be made arbitrarily close to the true parameter estimates. This property potentially allows the econometrician to gauge how close the approximate solution to a model with serially correlated unobservables is to the true solution. The method is illustrated using an example of the occupational choice decisions of certified elementary and high school teachers. Tests of the approximation quality suggest that the estimation of dynamic, discrete choice models with serial correlation can often be achieved with little approximation bias, even without incurring large amounts of computational costs.