About This Event
Predictive models rarely capture all of the possible signal in data. The best we can say is that the model validates well and predicts what can be predicted, given a certain modeling technique and set of inputs.
In this session, we'll discuss modeling on residuals to discover potential actions we can take. We will show five practical applications of residual modeling. These include using residual modeling to enhance predictive models, but also using them directly to create quick fixes and new options for insurers to improve performance. We will discuss applications to ratemaking, including modeling loss ratio - itself a residual metric. We will also show examples of how residual signal can be considered orthogonal to the base prediction, meaning that it adds to, rather than distorts, the accuracy of the underlying prediction.
