Forecasting Provincial Business Indicator Variables and Forecast Evaluation

Abstract

This paper evaluates a number of univariate and multivariate time-series forecasting models of selected indicator variables for three Canadian provinces: Alberta, British Columbia and Manitoba. The out of sample forecasts from these models are compared not only with themselves but with common indicators from the quarterly provincial forecast model of the Conference Board of Canada. The concepts of directional accuracy, the conditional efficiency, and the robust regression are used in evaluating the forecasts. In most cases, the strategy of combining forecasts produced superior results to those given by the Conference Board of Canada.