Analýza účinnosti vybraných statistických metod zaměřených na detekci heteroskedasticity

Abstract

The theoretical part of the thesis summarizes regression analysis, where we learn how to formulate a regression model and what method can be used to obtain estimates of the parameters of the regression line. Since most of the thesis deals with statistical tests, a large part of the theoretical part is devoted to them, where the Breusch-Pagan, White and Goldfeld-Quandt tests are described in particular detail. The problem of heteroskedasticity, which is another of the central themes of this paper, is also discussed in detail. By using the mentioned statistical tests we are able to determine the requirements of the assignment depending on the selected factors. The testing section is divided into two parts, the first one describing general statistical tests and the second one naming the specific tests mentioned above. The experimental part of this paper focuses on the analysis of selected statistical methods that are aimed at detecting heteroskedasticity. All the simulations included in the thesis are created in MS Excel, where different factors affecting these tests are successively tested and this influence is further investigated. Using selected statistical methods, both the strength of the test and the significance of heteroskedasticity were analyzed depending on the choice of the mentioned factors. Three types of models generating heteroskedasticity were analyzed, namely linear, quadratic and exponential models. For these models, the influence of these parameters was further evaluated, which are further described in the text of the experimental part of the paper.