Towards an intelligent residential appraisal model

In the UK, and indeed in many countries, Direct Capital Comparison (DCC) remains central to the practice of residential property valuers. Theoretically well founded, statistically or heuristically-based alternatives, usually embodying regression techniques, have failed to penetrate professional practice despite long pedigrees. For several years, neural networks, in which values are not so much derived or assigned but discovered, have also been propounded as potential alternatives. Yet DCC clings to its pre-eminent position because it is readily understood and it is thought to produce more accurate results. In order to improve upon DCC, complementary or alternative methods will need to enhance accuracy, and be equally intelligible and transparent. This paper reports empirical findings, discusses some of the obstacles that will need to be overcome and some of the constituents that may comprise an improved model.