Journal of Economic and Social Dynamics

Abstract

Traditionally, the process of credit risk adjudication has relied on periodic financial disclosures, supplemented with heuristic-based analyses. Although the process offers some level of oversight, the approach has significant drawbacks, such as information latency and cognitive bias. These issues are further compounded by the nature of the markets, which are known to be volatile, with opaque financial disclosures and significant governance risks. In the present research, the authors introduce CreditSentinel AI, a high-quality, machine learning-based approach to solvency analysis. The proposed CreditSentinel AI model utilizes non-linear eXtreme Gradient Boosting (XGBoost), incorporating high-frequency market data and fundamental financial statement-based financial information. The key differentiator of the CreditSentinel AI approach is the Graduated Penalty Mechanism (GPM), which utilizes severity-weighted risk adjustments, as opposed to binary covenant-based risk adjustments. In the proposed research, the CreditSentinel AI approach was tested using stress tests on listed firms on the National Stock Exchange (NSE), demonstrating the proposed approach's capability to detect structurally distressed capital profiles and leverage-based default risks. CreditSentinel AI has the potential to become a high-performance, automated first-line risk sentinel, capable of reducing decision latency.