Parametric versus non-parametric time series forecasting methods: A review

Abstract

The non-parametric methods have been proposed in the research literature as an alternative to parametric methods for time series forecasting. However, scarce evidence is available about the relative performance and computational ability of both parametric and non-parametric methods. This paper reviews the comparative studies conducted for evaluating the accuracy of parametric and non-parametric methods, especially ma-chine learning methods. For this, we briefly review widely applicable the parametric and non-parametric methods. Moreover, an empirical study has been carried out on real time series datasets to evaluate the comparative performance of parametric methods over machine learning methods. Additionally, the limitations of the machine learning methods are highlighted which leads to the selection of parametric methods over non-parametric methods by the researchers in recent years. Further, some recommendations for future research are presented.