Use this URL to cite or link to this record in EThOS:
Title: An examination of the stability of forecasting in failure prediction models
Author: Lin, Lee-Hsuan
ISNI:       0000 0001 3610 3341
Awarding Body: University of Warwick
Current Institution: University of Warwick
Date of Award: 1992
Availability of Full Text:
Access from EThOS:
Access from Institution:
The main focus oi this study is an empirical examination of the stability of company failure prediction models based on accounting information. Stability of forecasting in failure prediction models is tested using industry relative ratios and unadjusted ratios. Three homogeneous economic periods are defined : expansion, recession, and recovery. The stability and quality of forecasting models developed in these three different economic environments is tested using the industry relative ratios previously derived. The study also compares the stability of forecasting of both the industry- specific models and the aggregate model for each of the five years before failure. Specific industries include Contracting. General-Engineering. Textile. Other Manufacturing, and Miscellaneous. Finally, the ability of economy-wide indicators and year-dummies proxying calendar events to predict failure is examined. Industry adjusted and unadjusted ratio models, business cycles models (adjusted and unadjusted ratios) and specific industry models are reported. Each model is developed using multivariate discriminant analysis. An examination of the stability of forecasting in failure prediction models in terms of the classification accuracy, proportional chance criterion, expected cost, relative cost ratios, and Conover (1971) T test is performed. Finally, comparison graphs for each model are plotted. Industry relative (mean) ratios were preferred to unadjusted ratios because they reduce the heterogeneity of companies' data. This results in improved stability of forecasting both in the within-sample (ex post) and out-of-sample (ex ante) context. Subsequent, industry relative ratios are used to control for industry differences and different economic environments are used to control for time-inconsistency. The empirical findings of the study are that use of industry relative (mean and median) ratios and business cycles provides more stability and gives better predictive ability than use of unadjusted ratios and uncontrolled economic environments. In general, segmentation of the sample according to industry produced models that performed better than ones based on aggregate data across industries. Because each industry has different financial characteristics we conclude that industry-specific models should be developed if data is available. We find that industry specific and different economic conditions models are robust with respect to variation in prior probability and misclassification costs. In the context of failure prediction, accounting information appears to be more useful than macro-economic variables. The 4 macro-economic and 11 year-dummy variables are shown not to play an important role, adding only marginal discriminating power to the models.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: HB Economic Theory ; HG Finance