Chosen statistical methods for the detection of outliers in real estate market analysis

Abstract

The paper contains the comparison of mechanism of two separately constructed statistical methods for the detection of outliers in real estate market analysis. For this purpose, databases with various types of real estate from local markets were created. Then the estimation of parameters of functional models describing dependencies prevailing on the examined markets was carried out. Subsequently, statistical tools called Baarda method and model residue analysis were used to detect outliers in the collected datasets. The last stage was a comparison of the obtained results of the parameters' estimation of the analyzed models and the measures of their quality, before and after the removal of outliers. The obtained results indicate that algorithms of chosen statistical methods, detecting outliers, allow to eliminate a smaller number of them, at the same time obtaining an improvement of the parameters of the functional model and its adjustment to the analyzed dataset. This gives the premise for the development of criteria for the selection of statistical methods that look for gross errors in the analyzed databases, among others, depending on functional model used, type of property and number of properties.