The purpose of this research is to develop a finite sample algorithm for testing general hypotheses regarding a vector- valued time series. Such hypotheses might concern the moments of functions of the moments of the series; or they might concern parameter values and OLS estimation biases in a VAR representation of the series. Since the algorithm is an elaboration of the bootstrap, it is computationally intensive but requires very little in the way of maintained assumptions. The major goals of this project include (1) developing the algorithm and associated software so as to make it both accurate for use in small, serially correlated samples and also reasonably convenient to apply and (2) demonstrating the feasibility and desirability of post sample statistical inference by applying the algorithm to Granger causation testing and to the assessment of the relative and absolute worth of recent macroeconomic forecasts. This research is important because it will provide economists interested in empirical research with a new and powerful tool for hypothesis and model testing using time series data.
