A Conditioned Forecasting Model: A-priori Screening Validation Testing

Context The forecasting literature over the last three decades documents that judgmental conditions on the performance of the forecasting model are often used to rationalize the acceptance of the intel from a forecasting model that will be used in creating an action-plan. However, rarely are these judgmental-conditioning protocols recorded as they should be to intelligently process the interactions of the conditioning protocols with possible adjustments made in the forecasts. Focus In this research report, we will offer, four judgmental conditioning aspects that are not infrequently used by managers of forecasting divisions. Specifically, the acceptance contingencies of a forecasting model under evaluation scrutiny are- (i) The desired magnitude of the Median benchmarked Precision is in evidence, (ii) The Holdback is in the (1-FPE) Confidence Interval, (iii) The Pearson Product Moment Correlation-Null of the residuals is not rejected, and (iv) The Autocorrelation-Null of the residuals is not rejected. Each of these four conditioning aspects will be evaluated for two standard models typically in the panoply of forecasters- The Two-Parameter [Intercept & Slope] Linear OLS-Regression & the ARIMA(0, 2, 2)/Holt models. The measure of interest for ALL of the selected inferential analyses is- How often do selections among these conditioning aspects result in the forecasting model being rejected as informing the decision-making process? Results Surprisingly, the range of Failures for the conditions tested ranged grosso modo in the interval-{40% to 80%}depending on the nature of the Conditions. These implications are discussed.