NEURONET LOGISTIC DYNAMIC MODEL TO DIAGNOSE AND FORECAST THE STAGES OF DEVELOPING BANKRUPTCY OF CORPORATIONS

Abstract

Purpose of Study: The object of the study is the problem of financial management, in particular, the problem of forecasting the stage of developing bankruptcy of corporations-loaners and decision-making on the restructuring of credit debt. The subject of the research is the development of a dynamic model of bankruptcies with continuous time in conditions of high uncertainty and noise data, which allows diagnosing the stages of bankruptcy of the simulated object at any time (between the “time slices” in the data), as well as to predict the probability of bankruptcy in time ahead for a given horizon. Uncertainty is understood as a specific characteristic of the simulated class of dynamic bankruptcy problems –incompleteness and uncertainty in the data: in the training sample in the “time slices” only the boundary values of the probability of bankruptcy are specified (P=0 or P=1), i.e. there is no information about the intermediate values of P in the interval [0;1]. The uncertainty is determined by legal reasons: until the corporation is declared bankrupt by the arbitration court or the tax authorities, for it P=0, although objective accounting data may show proximity to bankruptcy. Methodology: The purpose of the study is to create an effective mathematical tool for predicting corporate bankruptcy to support decision-making on the financial management of corporations, which is focused on complex real-world modeling conditions. Results: On the basis of the system-wide law of inertia of the simulated dynamic system (or object), the original neuronet iterative logistic dynamic method (NLDM) is proposed, which allows eliminating the above incompleteness and uncertainty in the training sample and operate with continuous time in the procedures of diagnosis and prediction of bankruptcy stages of corporations-loaners. The adequacy of the dynamic model of bankruptcies is comprehensively investigated. The probability of correct identification of bankruptcies on the test set is not worse than 90%. The convergence of iterative procedures in the NLDM algorithm is investigated.