ENCHANCING A MACHINE LEARNING-BASED METHOD TO EVALUATE THE PROFITABILITY OF CONTRACT EXECUTION GUARANTEES

Protasov P. S., Potekhina E. V.

The present study is focused on investigating the intricacies associated with the concurrent utilization of two distinct machine learning models, each developed with different specifications and dedicated to forecasting the likelihood of contract non-performance during the issuance of a bank guarantee. Within the context of this research, a comprehensive approach for evaluating the profitability of a bank guarantee portfolio is proposed, taking into account the magnitude of the commission remuneration linked to the issuance of the guarantee, the expenditure incurred for acquiring the requisite data for model development and replication, and a quantitative assessment of the risk of contract default, under which the bank guarantee is provided. Conclusions are drawn concerning the connection between the scale of data acquisition costs pertinent to model development and replication and the overall yield of the guarantee portfolio. These elucidated methodologies offer a framework for devising strategies to determine the optimal threshold of the model assessment based on the profitability metric, while factoring in the pertinent business metrics that characterize the specific operational context.

Keywords:binary classification, gradient boosting, confusion matrix, profitability.