Further Evidence on the Accuracy of Residential Vacancy Chain Models

Abstract

Residential vacancy chain models simulate the transfer of vacant housing opportunities among sectors of an urban housing market. The Markov model simulates forward-reaching chains in response to vacancy initiations. The Leontief model simulates backward-reaching chains in response to vacancy absorptions. Each simulates residential mobility among housing sectors as a by-product. The accuracy of these models was assessed in earlier work by the authors, using 1975-80 data from Gävle, Jonkoping and Vasteras, Sweden, to project intra-urban residential mobility in each town during the 1980-85 period. Using log-linear analysis to compare projected moves with observed moves, they found projection errors ranging from 3-12 per cent. In this paper, data from the 1985-90 period are used first to repeat these assessments over the subsequent time-period and then to extend the projection period from 5 to 10 years. Projection errors range from 5-11 per cent for the 5-year period and from 8-18 per cent for the 10-year period. Both the Markov and the Leontief models perform equally well. Models with more homogeneous housing sector definitions produce more consistent results.