REP Research Seminar - Presenting in person - Dr Rainer Schulz, University of Aberdeen. Title: "Making and Assessing Real Estate Valuations".

This is an internal seminar and attendees should use their University email account to join the meeting if they are joining via MS Teams. You are welcome to share this invite internally but please do not share it beyond the University. Please direct any enquiries from those outside of the University to: repschooloffice@reading.ac.uk.

Abstract:

Fitting an automated valuation model requires a loss function. The very same loss and its related identification function should then be used to assess out-of-sample predictions. This consistency requirement also works in the opposite direction. If predictions represent a particular functional of the price distribution, such as the mean or the median, prediction errors should be assessed under the loss that elicits this functional. Real estate practice and research do not follow this consistent approach. We clarify the problem and consider in an empirical application different loss functions, including those motivated by economic losses in real estate applications. Deep neural networks adapt well to the functions, and conditional predictions fulfil the consistency requirement and perform always best. The variation between the predictors - each representing a different functional - is sizeable. Further research should examine how prediction errors translate into economic losses.

Bio:

Rainer Schulz is a senior lecturer at the University of Aberdeen Business School. He received his doctoral degree in Economics and Econometrics from the Humboldt-Universität zu Berlin. His research has been published, amongst others, in the Journal of Economic Behaviour and Organization, Journal of Real Estate Finance and Economics, the Journal of Urban Economics, and Regional Science and Urban Economics.