Goal: Credit rating company wanted to improve the accuracy of their risk models and build model on more easily acquired data.
ODG Solution: We built credit risk models from the original data source and from new channels. We then compared their performances to determine whether the new channels produced equally accurate scores. Model scores were designed to clearly communicate risk levels to their clients.
Results: Using Champion-Challenger methodology, we refined the performance of both models until their lifts were nearly equivalent. In doing so, we successfully showed that the new channels, which had more easily acquired data, had the same predictive performance with reduced data collection costs.
The credit risk company had historically based their ratings off of surveys. This method, though accurate, was very expensive and time-intensive. With the rise of data analytics, they began to explore new data channels that were more cost and time effective.
What We Did
We set out to build a model from more accessible data in order to reduce data collection costs. To compare its performance, we developed a model based on the old data set and a similar model from the new one. From these two models, we generated a unified credit score that determined how likely a company was to default in 12 months.
The model from the new data channels proved as accurate as the model from the old one, saving the credit risk company money and time. In addition, the unified credit score provided the company’s clients with more reliable information.