Low CPU Cost Semiparametric Estimation

with Steven Stern and Nese Yildiz

Slides and draft available upon request

Abstract

This paper introduces a Low CPU Cost Semiparametric (LCS) estimator for linear single index models. The LCS estimator significantly reduces estimation time when compared to the standard semiparametric estimator in Ichimura (1993). It does so by more than 90% in medium sample sizes. Moreover, it makes estimation feasible in a regular PC when the sample size exceeds 10,000 observations. We provide conditions for the consistency of the LCS estimator based on spline function theory.  In our empirical application, we study determinants of expenditures in vocational rehabilitation (VR) programs using the RSA-911 data, containing information on more than 900,000 workers with disabilities. We find that minorities such as African Americans, Hispanic or females have lower expenditures in VR programs. On the other hand, expenditure is greater for more educated workers.

Keywords:

Low CPU cost; Semiparametric estimation; Single index models; Vocational rehabilitation.

JEL Codes:

C14; C55; C51; J14.