New Methods of Evaluating the Forecasts Accuracy: A Case Study for USA Inflation

Abstract

Using the two-year inflation forecasts provided by CBO, Blue Chips and Administration for USA on the forecasting horizon 1982-2011, the accuracy of forecasts was assessed. According to U1 Theil's statistic, the CBO projections are the best, followed by Administration and Blue Chips predictions. A new accuracy measure is proposed to be introduce in literature (ratio of radicals of sum of squared errors) in order to compare the forecasts with the naïve ones. The same hierarchy of institutions, according to accuracy criterion is obtained if all the computed accuracy indicators are taking into account using ranks method. According to the relative distance method with respect to the better institution and other methods (binary logistic regression and some non-parametric tests (Wilcoxon and Kruskall-Wallis tests)) the following rank is gotten: Administration, CBO and Blue Chips. All these methods (multi-criteria ranking, logistic regression and non-parametric tests) were not mentioned before in literature as possible ways of comparing the forecasts accuracy, but most of them gave better results than the classical U1, because these methods take into consideration more aspects regarding the accuracy measurement. Some empirical strategies of improving the forecasts accuracy were applied (combined forecasts, smoothed predicted values based on Holt-Winters technique, Hodrick-Prescott, Baxter King and Christiano-Fitzegerald filters), getting more accurate predictions only for the combined forecasts of Blue Chips and Administration using inverse MSE scheme (the highest improvement) and equally weighted scheme, but also for forecasts based on CBO and Blue Chips using optimal scheme.