Automated forecasting procedure (Q1286307)

We have described some methods of a new direction in forecasting system engineering as represented by the Forecasting ToolPak automated forecasting package. The key issues in automated system development include automated identification and training of extrapolation models, analysis of very large databases, and intelligent evaluation and comparison of models. An algorithm is proposed for local step-by-step model identification and training designed for very large databases. The model quality can be measured by a weighted combination of white noise criteria, information criterion, and an independence statistic.