Analýza účinnosti vybraných statistických metod pro detekci a odstranění autokorelace v regresních modelech

Abstract

The aim of this thesis is to analyze the power of statistical tests used for the detection of autocorrelation in linear regression models. Particular emphasis is placed on comparing the effectiveness of the Durbin-Watson and Breusch-Godfrey tests in the context of autoregressive models of the first and second order (AR(1) and AR(2)). The research employs simulation-based approaches that allow for the quantification of the influence of various model parameters, such as sample size, the value of the autocorrelation coefficient, and the variance of the random component, on test power. The thesis also addresses methods for eliminating autocorrelation using the classical Ordinary Least Squares method and the Generalized Least Squares method, evaluating their effectiveness in the presence of autoregressive structures.