Title: RE-PURPOSING THE RBF-MODEL: KEY Y-VARIATE MEASURES- & GRAPHICAL-TRIAGE
Authors:
Moncef Belhadjali & Edward J. Lusk*
Abstract:
Context The Rule Based Forecasting Model [RBF] of Collopy & Armstrong (1992) [C&A] has been the progenitor of numerous versions that have offered refinements of their RBF [99-Rules] Model. Most of these progenies offered Rule Reduction Refinements or Forecasting Model Consolidation Options. The pistons that drive the RBF Model are: The 99-Rules that offer conditions that suggest and rationalize:(i) The Modification of the Y-Variate Panel Values, or (ii) The Recalibration of the Forecasting Model Weightings given certain Conditions of the Y-Variates under examination. Simply The Decision Analyst [DA] is encouraged to modify :{ The Y-Variates or The FM-Weightings} so as to be better aligned with the FMs engaged to inform the Planning Committee. Method In the spirit of continuing to offer tested options to the ever-growing RBF-corpus, we suggest that the DA examine: (i)Three Y-Variate Measures that relate to the following Features suggested by C&A:{Significant Trends}&{Domain Intel}&{Uncertainty and Stability, as well as (ii) A Chart of the Y-Variate under evaluation. With this binary profile, the DA decides if that Y-Variate should be transferred to:
{Dossier [I]: Likely to Create Viable Forecasting Intel}, or
{Dossier [II]: Evaluation for Possible Anomalies}
Our measures for C&A’s features are: {The Pearson Product Moment Correlation [PPMC]}, {The Pearson Measure of Skewness [P_Sk^y ̂ ]}&{The Coefficient of Variation [CoV]}.We will use:(i) The Chart of the Y-Variate, and (ii) Calibrations of these measures to inform the DA who will decide on the appropriate Dossier to populate for each of the Y-Variates evaluated. We will inferentially test the logic of the DA’s assignment using the Absolute Percentage Forecasting Error [APE] of these two Dossiers. Results As expected, the APE-Profile of the Y-Variates in Dossier [I] tested to have superior forecasting acuity relative to the Y-Variates in Dossier [II]. In addition, as a vetting-confirmation, the measures {PPMC &P_Sk^y ̂ & CoV} were inferentially different in the expected direction as between :{ Dossier [I] v. Dossiers [II]}. We offer a VBA-Computational program, at no cost as a download, for creating all of the information to test our protocol: called the Triage Alignment Protocol [TAP].
Keywords: Forecasting Acuity, APE, ETS.
PDF Download