The Perils of Underestimation of Standard Errors in a Random-coefficients Model and the Bootstrap

Abstract

Simulation experiments provide an important extension to two standard components of econometrics, namely, mathematical statistics and data analysis. For over a decade Professor Henri Theil and his associates, among others, have made extensive use of simulation experiments to obtain considerable new insight into the problems of statistical inference in large systems of demand equations. These experiments can be particularly relevant in small sample situations where theoretical results generally correspond to larger sample approximations. The use and power of simulation experiments is clearly evidenced in Theil’s recent survey of simulation results of the econometrics of demand systems by his team of researchers (see Theil, 1987). Indeed, the use of simulation experiments in econometrics has a long and established tradition for statistical inference (for a comprehensive survey and related references, see Hendry, 1984).