Implementing the Single Bootstrap: Some Computational Considerations

In applied econometrics, the researcher typically has two recourses for conducting inference: assuming normal errors or relying on asymptotic theory. In economic models, the assumption of normal errors is rarely justified and, for moderate sample sizes, the applicability of a central limit theorem is questionable. Researchers now have a third alternative: the bootstrap. Central to the bootstrap methodology is the idea that computational force can substitute for theoretical analysis. This article explains the bootstrap method, shows how a simple transformation can improve the reliability of inference, gives an algorithm for bootstrapping a regression equation, and discusses some computational pitfalls.