Estimation and Inference in Econometrics-英文文献.pdf

Bootstrap Inference in Econometrics by James G. MacKinnon Department of Economics Queen’s University Kingston, Ontario, Canada K7L 3N6 jgm@qed.econ.queensu.ca Abstract The astonishing increase in computer performance over the past two decades has made it possible for economists to base many statistical inferences on simulated, or bootstrap, distributions rather than on distributions obtained from asymptotic theory. In this paper, I review some of the basic ideas of bootstrap inference. The paper discusses Monte Carlo tests, several types of bootstrap test, and bootstrap confidence intervals. Although bootstrapping often works well, it does not do so in every case. Revised June, 2002 This paper was presented as the Presidential Address at the 2002 Annual Meeting of the Canadian Economics Association. This research was supported, in part, by two grants from the Social Sciences and Humanities Research Council of Canada. I am grateful to Russell Davidson, Dwayne Benjamin, and John Galbraith for comments. 1. Introduction One of the most remarkable examples of technological progress has been the massive increase in the speed of digital computers during the past forty years. For scientific computing, a typical personal computer of today is several hundred times as fast as a typical PC of just a decade ago, although it costs less than half as much. Even the PC of a decade ago was faster than multi-million dollar mainframe computers of just two decades ago. Some of this progress is document