Further Empirical Examination of an Improved Sales Comparison Approach

Abstract

Despite the significant advances in applying regression analysis into property valuation, the main features of the sales comparison approach lack thorough research. A series of works have endeavoured to retain the essence of the sales comparison approach, while at the same time take advantage of regressions to derive not only the implicit values of property attributes, but also the degree of similarity between properties. Despite these improvements, the determination of the best regression forms and the piecemeal-type of price adjustment remain vexing problems. The nearest neighbours method assumes that the effects of all attribute differences between the subject and comparable properties are captured by the Mahalanobis distance. The indicated market value of the subject property is simply a weighted average of the actual selling prices of the comparable properties. This method sidesteps the above vexing difficulties and seems worth employing. The present study extends the application of the nearest neighbours method to high-density residential properties, which have not previously been examined. In terms of both the average and coefficient of variations for prediction errors, neither the conventional regression nor the nearest neighbours method outperforms the other. Nevertheless, the distribution of the accumulated prediction errors suggests that the nearest neighbours method is superior over the regression analysis approach. Our empirical findings are, therefore, in favour of further pursuit along the small sample (comparables) methods.