An Exploratory Approach to Composite Modeling for Real Estate Assessment and Accuracy

The role of automated valuation models within real estate assessment has gained in prominence over the last two decades. However despite these developments, investigations relating to ‘composite’ or ‘ensemble’ modelling remains an under researched area. Using 28,957 properties and 5,515 usable sales transactions between 2016 and 2021 obtained from the municipality of Helmond in the Netherlands made available by the Netherlands Council for Real Estate Assessment, this paper introduces three different approaches to composite modelling: the central tendency composition approach, the weighted composition approach, and the logical coupling composition approach to enhance mass appraisal assessment, accuracy and performance. The composition approaches developed are applied to three different automated valuation models, a traditional general linear model approach, an Artificial Intelligence based automated valuation model, and a geo-statistical based automated valuation model. To identify the performance of the composition modelling approaches, a framework comprising twelve scenarios is applied to the outcomes of the composition to compare their results against the input valuation models.