| Abstract: | This dissertation will investigate the multi-variate statistical modeling of consumer behavior and its application to "true" cost of living indices (TCLI's). The archetypal approach, with problems of this genre, has been to impose constraints from neoclassical economic theories. However, some of the conventional assumptions are inadequate because, although successful in reducing complexity, they often drive likelihood ratios and their significance probabilities to zero. More particularly this study will analyze the sensitivities of the (TCLI's) to changes of prices and income, in particular the prices of monetary assets. This sensitivity analysis is applied to a TCLI that is based on a minimal set of theoretical assumptions. This intended minimal theory is achieved through the use of the nonparametric TCLI's based on the GFT form of direct utility function, which form does not rule out a priori any direct utility function that happens to conform to the data. Which is to say if the "true" functional form happens to be Leontief, Translog, the customary price dependent preference models, etc., this will show up in the likelihood ratio tests of their observational equivalent images which are nested within the GFT framework. A linear GFT-Translog, an observational equivalent form of the linear Translog whose image has been nested within the GFT framework, along with two other linear alternative functional specifications of the systematic changes of taste are considered. In addition different commodity groupings which contain monetary assets are used. |
|---|