PROBLEM DESCRIPTION
The residential real estate market is an integral part of any nation’s economy. In Norway, a strong tradition of home-ownership exists, with as many as 84% of the population living in a self-owned home. Furthermore, during the past few decades, the price of real estate in Norway’s capital Oslo has experienced strong growth. These factors have established a keen interest for the understanding of real estate prices among private individuals, commercial actors and policymakers. Moreover, the housing market is characterized by substantial complexity, illiquidity and transaction costs, making it a far from frictionless market. In total, this motivates the research of residential real estate prices.
In this study, we aim to develop an automated valuation model (AVM) to estimate the selling price of individual dwellings. Our approach is largely novel in the field of real estate finance and combines concepts from data science communities with techniques from traditional real estate research. Specifically, we leverage the concepts of stacked generalization and comparable market analysis. Stacked generalization, or stacking, is a machine learning technique where a meta-estimator is trained to combine the predictions of underlying sub models. We implement five different sub models; four of these are ensemble learning methods, while the fifth model is based on a repeat sales index. We apply the concept of comparable market analysis by selecting a set of comparable previous transactions tailored for each value estimation.
The AVM produces a value estimate for a dwelling by analysing data describing comparable dwellings’ characteristics and previous transactions. To assess the model, we implement it for the residential real estate market in Oslo and evaluate its performance out of-sample on all transactions in Oslo in the first quarter of 2018.
Proptechauto9
Developing and Evaluating an Automated Valuation Model for Residential Real Estate in Oslo