One aim of this study is to compare and evaluateprespecified and data-specific functional forms for theestimation of hedonic price functions. One prespecifiedfunction and three types of data-specific Box-Cox functions areused to estimate hedonic price functions. A second aim of thestudy is to test alternative monocentric and polycentrichypotheses according to a system of successive variablerejections constructed for this study, whereby requiredsignificance levels are influenced by the results of aselection of earlier hedonic price functions.
The possibility of constructing internally consistentcriteria for variable elimination is one of the merits ofemploying a prespecified functional form. Other advantagesinclude the simplicity and transparency of the derived results,and the robustness of key parameter estimates in the face ofmisspecified functions. Using a prespecified form will on theother hand as a rule not yield the likeliest functional form tohave generated the data and cannot be relied on to generatewell-behaved residuals.
Being assured of well-behaved residuals - as is the casewith data-specific Box-Cox functions - is advantageous, eventhough it makes the detection of potential heteroskedasticitymore difficult. Box-Cox models with more than onetransformation parameter are however especially prone to yielddescriptions of unique data sets at the expense of potentialreplicability. Because of the sensitivity of the t-values ofthe estimated independent variables to the estimated Box-Coxt:ransformation parameters - where the log-likelihood partlydepends on the set of independent variables - it becomesdifficult to avoid an arbitrary selection of independentvariables. This is true of all Box-Cox functions, althoughespecially of functions that estimate more than onetransformation parameter.
Although regression results provide detailed information onvariable effects, much of that information is not generalizableowing to interspatial and intertemporal instability. Because ofsupply and demand fluctuations and specification problems, itis argued that is preferable to draw qualitative rather thanquantitative conclusions.
An analysis of the Singapore condominium market illustratesthe conceptual discussion of various approaches to hypothesistesting, with an emphasis on testing hypotheses concerning thenumber and nature of urban centers. The data strongly supportthe realism of the monocentric model with one dominantemployment center. Alternative polycentric and service-centerhypotheses are unambiguously rejected. The signs, averageimplicit prices and t-values of the independent variables arerelatively robust over the range of estimated functional forms- whether pre-specified or Box-Cox - given the set of variablesto be included in the hedonic price equation.
Keywords:Hedonic prices, Box-Cox function, monocentric,polycentric, accessibility, hypothesis testing, rejection ofvariables, data-specific function, prespecified function.
Institutionen för infrastruktur och samhällsplanering , 1997. , 156 p.