Regression Analysis in Real Estate Valuation - Using Purchase Price Data with Gaps

Hedonic models in real estate context are already frequently used to derive data for market description and appraising e.g. with automated valuation models (AVM). However, the number of purchase cases is sometimes not sufficient. This is caused by a general small number of transactions in specific areas and lacks in meta data collection. These missing meta data is sometimes used as independent variables in the hedonic model. Many purchase cases are useless for investigations because information of one main explanatory variable is missing. Also data exist but is not discoverable e.g. due to data privacy policy. investigations fail when the general small number of transactions is coinciding with the lack in data accessibility. Sometimes, some information is available but not for all purchases. Thus, some influencing factors are unusable because of these lacks even if we know that its explanatory power on the dependent variable would be very high. Question raises, if the remaining information can be used. In this paper, we present investigations on different approaches to fill gaps. In addition to approaches from social science (e.g. pairwise exclusion and case exlusion) also mathematical approaches are analysed. As objects of investigation, we compare classical approches (elimination of data with gaps), imputation-based approaches (use e.g. the mean to close the data gap) and model based approaches (based on Maximum-Likelihood and Gauß-Helmert-Modell). Maximum likelihoodThe benefit of a Gauß-Helmert-Model is the additional information about the accuracy of the estimated observations. The investigation is done with simulated data. We compare the different approaches with a reference dataset (no data missing) in a closed-loop simulation. As results show, a higher accuracy especially in small data sets can be determined with the non-classical approaches. A larger number of usable data also increase the potential of detecting outliers in the dataset (inner reliability).