Research Information System
PAKISTAN JOURNAL OF STATISTICS AND OPERATION RESEARCH, vol.7, no.2, pp.473-481, 2011 (Scopus)
This study investigates an application of Friedman test statistic as a model selection methodology on post estimation data. Although there are various model selection criteria, their main focus is to fit the model to the estimation data. Some of these criteria are appropriate for nested model selection while the rest is suitable for non-nested model selection. The suggested model selection methodology is indifferent to the distinction between nested or non-nested model selections. In the suggested methodology, the dataset is subdivided into two parts: First part is used to estimate all the competing models while the rest are used for performance comparison of competing models. In this respect, all competing models are ranked according to their forecast performance based on proximity between observed and estimated values of the dependent variable in post-estimation data. Then, we showed that the suitability of Friedman test statistic in order to evaluate the prediction performance of competing models.