Criteria for determining the appropriate tax treatment of expenditures as repairs versus improvements : an empirical investigation of judicial decisions

This research effort centered on three objectives:

(1) to contribute to an operational refinement of the definitions of repairs and capital improvements, (2) to present an alternative methodology for quantitative analysis of tax cases, and (3) to compare the repair/capital improvement classification process of the United States District Court and the United States Tax Court. These objectives were accomplished through the use of quantitative models which were built from a random sample of sixty United States Tax Court Cases.

Objective one was accomplished through the relevant variables identified in a discriminant model and a logistic regression model. The variables identified by the models which indicate a capital improvement are: (1) Expenditure increases the value of the property; (2) Expenditure prolongs the life of the property; (3) Adequate accounting records are not present; and (4) Expenditure is in the form of a replacement, addition, or renovation. The variables which indicate a repair are: (1) There is a base or vestige of a usable asset present; (2) Expenditure is expected to be made repetitively; (3) Expenditure is an ordinary and necessary expense for the continued use of the asset; and (4) Expenditure keeps the property in an ordinarily efficient operating condition. Both models were capable of discriminating between repairs and capital improvements, as seen by their prediction rates in excess of 807o for a hold-out sample of sixty United States Tax Court Cases.

The alternative methodology was logistic regression. The higher prediction rate (90%) of the logistic regression model in comparison to the prediction rate (83.33%) of the discriminant analysis model makes logistic regression a viable alternative for the quantitative analysis of tax cases. This higher prediction rate was accomplished with fewer variables.

The final objective was accomplished by evaluating the logistic regression model against a sample of sixty United States District Court Cases. The model correctly predicted 88.337. of these cases. This high prediction rate helps refute some assumptions regarding the United States District Court which have caused the exclusion of these decisions from some prior studies.