Alternative approaches to modelling housing market segmentation : evidence from Istanbul.

There is a large literature on housing submarket definition and identification.
They did not address how to model submarkets once they have been identified.
Yet the modelling literature has produced several different approaches. These
approaches are being applied in different contexts at different times and using
different data sets. This thesis seeks to control some of this variation. It applies
four (market-wide hedonic model, hedonic models with submarket dummy,
separate hedonic models for each of submarkets, multi-level model) of the most
common methods to a data set comprising 2175 transactions in the Istanbul
housing market. The performance of these models is compared on the basis of
their accuracy in terms of proportion of estimated prices that fall within tolerable
range of the actual price. The results show that that the hedonic and multi-level
models with experts' submarket dummy variable can predict more accurately
than the models with a priori and cluster analysis stratified submarkets.
Similarly, the root mean square error test results indicate that the hedonic and
multi-level models with experts' submarket dummy variable show better
performance than other models. These test results show that both the hedonic and
multi-level models with experts' stratified submarkets dummy variable yields
better performance than market-wide hedonic models.