Date of Award

1985

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Abstract

In the past the problem of financial distress has been investigated mainly through discriminant analysis and conditional response (logit, probit) techniques. With the use of such models, inference is made about the future status of a company as failure or non-failure conditional upon its observed financial attributes. Although response models (and discriminant models under certain assumptions) can be used to estimate the probability of failure of a firm as a function of its observed characteristics, neither group of techniques can provide estimates of the failure rate (hazard) of a population as a function of time. In certain situations, when the time to failure is an important determinant of the payoffs, knowing the failure rate over time becomes critical. Expected payoffs, under different investment or lending decision policies, can be estimated when a model of the evolution of failures over time is available. This study provides a functional method of modeling the empirical survivor function of a corporation over a period of at least five years, conditional upon the corporation's observed financial characteristics. The survivor function S(t,z) (which provides the probability that a firm of z financial attributes will survive for at least t years) was estimated through the proportional hazards model. The covariates employed in the formation of the hazard function were chosen from accounting variables and financial ratios constructed from the information contained in the annual statement of publicly traded manufacturing and retail companies. The survivor function leads to the estimation of the probabilities of failure by time intervals of interest, inside the study period of five years. The significance of this feature is that one does not need to be confined to the probability of binary response (i.e., failure on non-failure) within the whole study period; the probabilities of failure over finer time segments are provided. This is in contrast to the information provided by discriminant analysis and other binary response models which by themselves provide little insight into the way explanatory variables affect survival.

Pages

107

Share

COinS