Machine learning (ML) is a hot topic nowadays. There are numerous benefits to using this technology, and many companies are jumping on board as soon as they can. Machine learning is also leading a new era of credit risk analytics.
What Is Credit Risk Analytics?
Risk professionals have been using analytics solutions for years. Credit scoring, one type of analytics solution, is a discipline developed in the 1960s and widely adopted by financial institutions by 1990. Originally a manual underwriting process, it later shifted to statistically driven scorecards for mortgage applications, credit cards, and auto lending.
The major objective for applying analytics methods in risk management was to separate the “good” accounts, the ones that will pay their obligations, from the “bad” accounts, the ones who will default and cause losses for lenders. The quality and value of an analytical solution in risk management measured how well it can distinguish these two categories.
Connecting ML and Credit Risk Analytics
In the last few years, new statistical algorithms have become very popular. Traditional scorecards were based on one decision tree, or “logistic regression.” The newer algorithms represent a combination of hundreds of decision trees instead of one single tree. These algorithms also provide much more accurate predictions compared to traditional methods. The current hype around machine learning methods typically revolves around these algorithms in particular: random forests, XgBoost, and deep learning.
The risk community has become very interested in these methods, as each additional point of separation between good and bad accounts translates into significant ROI. But this being a highly regulated industry, lenders struggle to embrace these methods because of an increased level of complexity and the challenge in explaining results with high level of transparency.
What We Are Doing
Dun & Bradstreet is currently investing in advanced machine learning for credit scoring, with the goal of delivering more precision. If executed correctly, the technique should make ML-based scores and decisions more intuitive and auditor-friendly.
We are iteratively experimenting and innovating with ML-based risk solutions. In applications such as customer delinquency models, failure score, and fraud risk index, machine learning methods have already provided immense value and complemented existing methodologies. Using proprietary methods, the analytics team combined the accuracy of ML methods with full explanation of scores (similar to the scorecard approach).
Machine learning and artificial intelligence have the potential to change the benchmark for what could be achieved with analytics, helping customers assess risk with much higher precision. There is a lot of potential in these methods, and we are constantly encouraged as we uncover new potential across different use cases.