Specification error analysis in econometrics.

Essay 1. An umbrella test for general misspecification. A simple test designed to detect a wide range of misspecifications is developed expanding upon results derived by Ochshorn (1986) in an unpublished dissertation. These results show that a variety of specification error tests can be classified as RESET type tests. The unifying element behind this analysis is that these procedures can be implemented by testing for the significance of additional test variables in a linear regression. These conclusions are used to produce a general test by simultaneously augmenting the regression with selected test variables from various RESET type tests. The power of this general procedure against a variety of misspecifications is exhibited through an extension of a Monte Carlo study undertaken by Thursby (1989). The results show that this umbrella procedure is superior to any of the specification error tests considered by Thursby. Essay 2. Diagnostic checking through specification analysis: An evaluation of Hendry's consumption function. This essay expands upon the work of Kramer and Sonnberger (1986) who describe the theory and application of a number of specification error tests. A variety of these tests, including the umbrella procedure developed in the previous essay are used to analyze a consumption function for nondurable goods and services proposed by Davidson, Hendry, Srba and Yeo (1978) for Great Britain. The results indicate a number of misspecifications and provide information that forms the basis for possible improvements in the model. Essay 3. Specification testing as a means for model comparison: An empirical analysis. The methodology described above is used to compare the specification of a consumption function for nondurable goods proposed by the Research Seminar in Quantitative Economics at the University of Michigan with the specification for nondurables and services developed by Davidson, et. al. (1978). In order to provide data that were not used to estimate or test either model, the tests are employed using Canadian data. The equation proposed by Davidson et. al. is rejected more frequently which indicates that the RSQE model is the preferred specification.PhDEconomicsSocial SciencesUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/128714/2/9124103.pd