Executive summary

Machine learning (ML) — the most prominent arm of artificial intelligence (AI) — cuts both ways for the financial services industry, where its applications are getting wider by the day. 

The benefits are obvious. ML models are trained to learn from results just as the human brain does and can execute complex tasks at a scale and speed humans simply cannot. 

But perils abound. Complexity of the models is a risk. Many can be opaque and obscure, notorious for being black boxes. And when non-transparent models malfunction, things could go out of hand. 

In extreme cases, it could even lead to financial institutions failing, with systemic consequences for the entire economy. 

Regulators are, therefore, highly concerned about ML models. Globally, regulatory scrutiny of firms using ML models for purposes such as capital plans and stress testing has not only become tighter, but full-fledged guidelines are also on the anvil. 

For financial institutions, the challenge is to make ML models adhere to the existing principles and best practices of model risk management. While all statistical models need validation, ML models call for a specialised approach. 

The effective use of ML, thus, entails familiarity with its proven applications, anticipating roadblocks, and understanding the broader analytics environment. 

This white paper situates the discussion on ML models in finance in this context. It examines in detail the regulatory landscape for such models, the choice of model validation techniques available, and the bottlenecks faced by firms in validation. It also presents how CRISIL can help financial institutions meet these ever-growing challenges and prepare for the soon-to-come regulatory tightening around ML.