| Abstract [eng] |
The aim of the research is to find out which dwelling characteristics are significant to one square meter price of apartments in Vilnius city municipality and prepare a few different forecasting models. For forecast models we have used 2008–2012 years Real Estate Transactions Database records of State Enterprise Centre of Registers of Vilnius city apartments’ purchases transactions, where the buyer is Lithuanian or foreign natural person. During evaluation of one square meter price of apartments, Vilnius city municipality was divided in four zones: center, prestige, living and other districts. Additionally, the value of the apartment, number of objects in the contract, number of rooms, wall type, presence of a basement and whether the building is newly built are also taken into account as a criteria. One square meter price of apartments in Vilnius city municipality is forecasted using linear, quantile regression and neural network models. Quantile regression was made for 0,05; 0,25; 0,5; 0,75 and 0,95 quantiles. All models were compared to each other according to their predictive accuracy for the year 2008–2012 data and year 2013 data, which were not included in the training set for the model. Comparison was made using adjusted determination coefficient and errors – MAPE, MSE, MAE, MPE. One square meter price of apartments in Vilnius city municipality is most accurately predicted by the median quantile and linear regression models. However it was found that the linear regression model does not meet the assumptions of residuals normality and homoscedasticity. |
