The evaluation of economic forecasts
The evaluation of economic forecasts is a substantial and important aspect of economic research, and a considerable part of such evaluation is performed by comparing competing forecasts. This thesis focuses on the development of statistical procedures in order that reliable comparison of contending forecasts can be made. The study considers three issues in particular. The first two issues are closely related and concern testing the companion null hypotheses of equal forecast accuracy and forecast encompassing. The established equal accuracy and encompassing tests are found to display problematic behaviour in certain situations, and new modified tests are proposed to overcome these shortcomings. Analysis of the tests results in a recommendation for employing one of the newly proposed tests for each of the respective hypotheses. The recommended tests follow parallel formulations and have a number of attractive features, notably robustness to likely forecast error properties of contemporaneous correlation, autocorrelation, non-normality and autoregressive conditional heteroscedasticity, reliable behaviour in finite samples, and good power performance. The third issue examines the ranking of rival forecasts according to a pre-determined evaluation criterion. A recently proposed summary criterion for multi-step-ahead forecasts, comprising a single measure for all model representations and all forecast horizons of interest, is analysed, and a more reliable alternative proposed. This summary criterion approach is compared to the more conventional method of ranking forecasts at a specific horizon for a particular model representation, and the related issue of forecast encompassing for linear combinations of forecasts is discussed. This thesis therefore develops robust well-behaved tests for equal forecast accuracy and forecast encompassing, and advances techniques for ranking competing multi-step forecasts, providing improved, more reliable procedures for conducting economic forecast evaluation.