Nonparametric Estimation for Nonstationary Time Series Models

Abstract

This paper aimed to use some nonparametric methods in estimating nonstationary time series through the application of the cointegration regression methodology. The research employed both descriptive and econometric methods to construct the standard ECM (Error Correction Model) for monthly time series data for the period 2010-2015. The results relied on the Phillips-Perron unit root test to ascertain the stationarity of the time series and the Engle and Granger cointegration test to examine the existence of a long-run relationship. The application focused on the use of two nonparametric methods, in order to compare and identify the best method for estimating time series models in the light of the cointegration regression methodology. The results proved the superiority of the Lowess method over the cubic Spline method, as it achieved the shortest period and the highest adjustment ratio for disturbances occurring in the short run, with the aim of returning to the long run