Multivariate Generated Regressors and Heteroskedasticity in a Cross-Section: an Application to the Value of Neighborhood Schools.

This paper describes the estimation and testing of regression models that include multivariate generated or computed regressors in the presence of heteroskedasticity in the cross-section case. Heteroskedasticity is often a problem in cross-section data and the usual tests for its presence can not be applied when the heteroskedasticity is in some measure due to computed regressors. We investigate the case of multiple computed regressors that are generated from the results of a system of seemingly unrelated regressions and we proposed a method to test and correct the covariance estimates for unknown heteroskedasticity in the errors of the model of interest. In contrast to most time-series applications, we allow for the observation in the first step regression to be different from those for the second stage regression.