An evaluation of the Effectiveness of Automated Valuation methods (AVMs) versus Traditional Valuation Methods in Residential Real Estate : A Comparative Analysis of Predictive Accuracy and Price reflecting Efficiency in recent property transactions

Abstract

This academic paper evaluates the effectiveness of both traditional valuation approaches and Automated valuation Models in the context of predictive accuracy and price reflecting efficiency by analysing closed transactions in the residential real estate industry. The study addresses a central research question: “Does valuation represent a good proxy of real market price, which reflects all available information?”. To answer this question, a comparative analysis will be performed between traditional methods and AVMs to establish which od the two is more accurate hen determining value. The research begins by reviewing the traditional valuations methods; the income and sales comparison approaches, highlighting their reliance on historical data, human expertise and subjective assumptions. It then continues by observing the recent rise of AVMs, pushed by continues technological advancements related with AI and machine learning. Furthermore, introduces a hedonic price regression model to assess the price reflecting efficiency as a more data-driven approach traditional method. To carry out all the valuations, a dataset of 400 sold properties in the city of Indianapolis (USA) was extracted form Zillow. Data was organised, cleaned and interpreted to apply the income method and sales comparison approaches. Consequently, with the use of this data a hedonic model was built in python through Google Collab. Lastly, Zillow´s Zestimate, one of the most reputed AVMs in the industry, was used for the comparative analysis. Results showed how major discrepancies arise between methods. Traditional approaches obtained MAPEs of 16.41% for the income method and 18.14% for the sales comparison. Even though results aligned with industry benchmarks, they reflected limited predictive accuracy. The hedonic price regression model yielded an R-squared of 0.728, meaning that the model explained over 73% of the price variations according to the imputed variables. This reflected the importance of including as many price determinants as possible to perform valuations. On the other hand, Zillow´s Zestimate obtained a MAPE of 2.58% showing is robustness when reflecting prices and performing accurate valuations. The study concluded that whilst traditional methods remain important in many scenarios, they are quickly being surpassed by AVMs. However, due to the issues related with this data driven approaches, there is an urge to develop hybrid models which incorporate the strengths of both approaches and calls for further technological advancements so AVMs can incorporate property specific traits and ensure ethical standards and transparent procedures.