All-Moscow seminar "Mathematical methods of decision analysis in economics, finance and politics"

On Wednesday, September 17, the National Research University Higher School of Economics hosted a regular meeting of the All-Moscow seminar "Mathematical methods for analyzing optimal solutions in economics, business and politics".

Seminar leaders:
Doctor of Technical Sciences, prof. Aleskerov Fuad Tagievich
Doctor of Technical Sciences, prof. Podinovsky Vladislav Vladimirovich
Doctor of Technical Sciences, prof. Mirkin Boris Grigorievich


Speaker: E.V. Lashkevich (HSE), Yu.A. Zelenkov (HSE)
Topic: Generating algorithmic recommendations to prevent financial insolvency of a company


Annotation:

This paper examines an approach that combines explainable artificial intelligence (XAI) and synthetic data generation methods to generate algorithmic recommendations (algorithmic recourse) for changing a firm's risk profile to reduce the risk of financial insolvency. Unlike traditional methods, the proposed approach uses synthetic data generated based on the joint distribution of features, which improves the quality of counterfactual explanations (CE) and ensures the relevance of recommendations even with limited initial data.
The study includes a comparative analysis of various methods for generating synthetic samples and assessing their effectiveness for binary classification problems and constructing counterfactual scenarios. The developed methodology includes constructing a bankruptcy probability assessment model, generating a set of potential counterfactuals using Bayesian networks, and selecting recommendations based on both CE metrics and user requirements. Particular attention is paid to assessing the implementation costs of the proposed measures and integrating expert judgment into the decision-making process.
The solution, tested using data from Finnish small and medium-sized businesses, provides not only highly accurate but also realistic recommendations, significantly increasing their practical value for financial risk management.