Testing and Improving Commercial Real Estate Market Segmentations With Cluster Analysis and Neural Network Techniques

Market segmentation is a standard concept in both strategic marketing and investment analysis. The standard approach in the real estate market context is to segment markets by regions and property types. While this approach provides investment analysts with a powerful basic classification grid, its capability of predicting the performance and risk characteristics of direct investments is rather limited. The corollary of this is that a classification based on the two criteria yields segments that are too heterogeneous in their investment performance. As this is an issue of potential importance to investment analysts, this paper tests the predictive power of existing segmentations. In a further step, we apply a two-step cluster algorithm to generate new clusters based on additional investment characteristics and information relating to the tenant base and lease structure of a property. To this aim, we analyze the very large IPD commercial real estate database for the UK over the period 1980-2006. Finally, we apply both discriminant analysis and a non-parametric neural network estimation to test the ability of various segmentations to predict total returns. Both methods confirm that the segments determined by cluster analysis yield superior explanatory power. We conclude that the new segments are a potentially useful tool for commercial real estate investment analysis.