Estimated realisation price (ERP) by neural networks: forecasting commercial property values

Abstract

This work shows that it is possible to link various economic and property attributes to the value of a commercial property over time in a particular market, and arrive at a valuation pattern which can be used to give a short‐term forecast of valuation fluctuations using longitudinal rather than cross‐sectional analysis. Shows that it is possible to do this by using a novel process we have termed “backtrack valuations” or “backtracking”. The method proposed creates a simulated historic record of valuations, from which a neural network can be trained and then used as a model to estimate a forward trend. This is allied to the requirement in the RICS Appraisal and Valuation Manual (Red Book) whereby the valuer may be instructed to provide Estimated Realisation Price which depends on completion taking place on a future date as compared with Open Market Value where achievement of completion is assumed at the date of valuation. There is also the new definition of “Forecast of Value” in the RICS Red Book and we suggest that the valuer would find the technique of forecasting from backtracked time series of interest and use in both these particular circumstances. The source of data for the investigation was Richard Ellis, International Property Consultants, who provided monthly valuations of 16 major commercial properties in Central London. Our forecasts are presented alongside the subsequent Richard Ellis valuations. The results confirm that in the conditions obtaining in this market, it is feasible to predict capital valuations in the short term. The method is being extended and tested in the wider commercial markets.