Optimal forecasting model selection and data characteristics

• Autores: Robert Fildes, Gary Madden, Joachim Tan
• Localización: Applied financial economics, ISSN 0960-3107, Vol. 17, Nº. 13-15, 2007, págs. 1251-1264
• Idioma: inglés
• Texto completo no disponible (Saber más ...)
• Resumen

  • Selection protocols such as Box-Jenkins, variance analysis, method switching and rules-based forecasting measure data characteristics and incorporate them in models to generate best forecasts. These protocol selection methods are judgemental in application and often select a single (aggregate) model to forecast a collection of series. An alternative is to apply individually selected models for to series. A multinomial logit (MNL) approach is developed and tested on Information and communication technology share price data. The results suggest the MNL model has the potential to predict the best forecast method based on measurable data characteristics.