AI in Credit Risk Management Part 1 By Panos Skliamis (@SPINANALYTICS1) This post discusses my point of view regarding the debate initiated by the US Treasury Department on a radical overhaul of financial technology regulations, so as for innovative fintech solutions to be facilitated. It should be noted though, that my opinion is focused exclusively on the Credit Risk Management sector and the relative regulations. In recent years, great progress has been made on Credit Risk Management methodologies and techniques. However, Credit Risk modelling and management is both a science and an art, thus, most modelling and management work is essentially still performed manually. Pure Machine Learning techniques have been proven inappropriate for the specific problem, primarily because the data samples that relate to Credit Risk are always significantly biased and problematic, for several reasons:

• credit rejections
• dependent variables that highly relate to the specific treatment of the credits by each bank or department
• long horizons of predictions
• inconsistency of risk management practices between institutions and over time
• different and continuously evolving regulatory regimes,
• economic cycle

Moreover, the contexts of the sectors where Machine Learning techniques succeed are usually stable over time. A fingerprint, an iris and a character retain their shape and, thus, if an ML model is trained to recognize them, then it can repeat the recognition in the future. The context of Credit Risk Management, on the other hand, is highly fluid and extremely unstable over time. For instance, the current revenue, income, leverage, industry status and macro environment in no case can be considered stable for the duration of a credit. If you liked what you just read, why not sign up to our weekly newsletter? Get all the content straight to your inbox here.