Whitepaper: Predicting Credit Defaults using Machine Learning
For Counterparty Credit Scoring & Risk Management
Imagine a Counterparty with a history of default, beginning to slip on its payments again; or a relatively newer Counterparty with steadily increasing exposures beyond company’s comfort levels. You’d want to see red flags on these and other such cases much before it actually hits the reality, right? But today’s systems and processes are just about geared towards generating either pure Bayesian framework based or pure score-card based models that fail to take into account several external factors that often play a large role in predicting these red flags.
Risk Edge has come up with a new whitepaper on using Machine Learning for predicting Credit Defaults, which can also be used for Credit Scoring. Although based on individual data (instead of corporate data), the model is quite similar to the one that can be used to predict Counterparty defaults.
What can you learn from this paper?
The paper shows multiple models that can be used for predicting defaults, along with their results on how they fared in terms of accuracy. The paper not only gives a brief about the models, so as to not give them a “black-box” feel, but also shows how they can be tweaked in multiple ways using advanced Machine learning techniques to improve their accuracy !
- Understanding the factors that affect play a role in predicting defaults and visualizing a couple of them to make sense of the data.
- Which factors are more important than others when it comes to building the model ?
- How to use different models like Decision Trees, Rules, Adaptive Boosting, etc. to build different equations that solve our problem?
- How can some of them be tweaked in a way that allows us to improve the way the model understands our specific problem?
You can download this paper for Free by submitting the form on the right. If you are an existing subscriber from Energy / Commodity company, you don’t need to fill this form again, you’ll receive this paper in your registered email id !
Please note that our Publications are shared selectively only with people from Energy / Commodity / Banking / Financial Services companies. We frequently get requests from generic domains (like gmail, yahoo, etc.) and sometimes, despite our best efforts to verify their identities, we are unable to do so. Such requests unfortunately have to be declined. So we would request you to use your official email id for downloading our publications – and be assured that you’re in good company !
Download Requests from vendors / consultants are also taken up separately, and approved only if there is a viable business case for doing so. Please write in separately to us on email@example.com – we read and respond to every personalized email !