Aggregation of Enterprise Bankruptcy Risk Assessments Based on Logit Complex—Mamdani Models and Fuzzy Logic Inference

Abstract

The problem of increasing the predictive accuracy of bankruptcy forecasting in the analysis of data arrays of various nature is considered. The analysis of existing approaches showed their limitations. The goal of the study is to develop a methodology that allows to aggregate estimates of the bankruptcy risk of an enterprise, obtained on the basis of a combination of logit models. The proposed approach is based on the results of the Mamdani fuzzy-logical conclusion and, when constructing complex estimates, possesses such properties as variability, the ability to quickly adapt to a specific task by varying weight coefficients, ease of implementation, and wide possibilities for taking into account expert opinions in the model. The adjustment of the model for industry and time specifics is carried out not by recalculating the coefficients in the original logit—models, but by clarifying the weighting coefficients of the models, based on expert estimates and recommendations. The aggregation of enterprise bankruptcy risk assessments obtained on the basis of five different models (Altman-Sabato, Lina-Pesse, Gruzczynski, JuHa-Tehong, and Zhdanov) was carried out; the initial estimates are characterized by different quantities, opposite dynamics, and do not allow an unambiguous interpretation by conventional methods. The aggregated estimates are constructed over the five considered years, taking into account the significance of the models used. The proposed model is new in the view of developing fuzzy-multiple methods of complex estimates for the analysis of financial statements.