Property Mass Valuation on Small Markets

The main bases for land taxation are its area or value. In many countries, especially in Eastern Europe, reforms of property taxation, including land taxation, are being carried out or planned, introducing property value as a tax base. Practice and research in this area indicate that such a change in the tax system leads to large changes in land use and reallocation. The taxation of land value requires construction of mass valuation system. Different methodological solutions can serve this purpose. However, mass land valuation requires a large amount of information on property transactions. Such data are not available in every case. The main objective of the paper is to evaluate the possibility of applying selected algorithms of machine learning and a multiple regression model in property mass valuation on small, underdeveloped markets, where a scarce number of transactions takes place or those transactions demonstrate little volatility in terms of real property attributes. A hypothesis is verified according to which machine learning methods result in more accurate appraisals than multiple regression models do, considering the size of training datasets. Three types of models were employed in the study: a multiple regression model, k nearest neighbor regression algorithm and XGBoost regression algorithm. Training sets were drawn from a larger dataset 1000 times in order to draw conclusions for averaged results. Thanks to the application of KNN and XGBoost algorithms, it was possible to obtain models much more resistant to a low number of observations, a substantial number of explanatory variables in relation to the number of observations, a low property attributes variability in the training datasets as well as collinearity of explanatory variables. This study showed that algorithms designed for large datasets can provide accurate results in the presence of a limited amount of data. This is a significant observation given that small or underdeveloped real estate markets are not uncommon.