Market value predictions for residential properties are important for investment decisions and the risk management of households, banks and real estate developers. The increased access to market data has spurred the development and application of Automated Valuation Models (AVMs), which can provide appraisals at low cost. We discuss the stages involved when developing an AVM. By reflecting on our experience with md*immo, an AVM from Berlin, Germany, our paper contributes to an area that has not received much attention in the academic literature. In addition to discussing the main stages of AVM development, we examine empirically the statistical model development and validation step. We find that automated outlier removal is important and that a log model performs best, but only if it accounts for the retransformation problem and heteroscedasticity.
|Number of pages||23|
|Journal||Journal of Property Research|
|Early online date||31 Oct 2013|
|Publication status||Published - 2014|
- hedonic regression
- log transformation
- predictive performance