Description: In any model, assumptions are necessary to generate output. Typically, the model is run with varying assumptions to determine how sensitive the output is to the input. But what if the philosophy of generating assumptions is flawed?

Description: Complex models can provide a false sense of security, hiding the evidence that the entire range of indications may hinge on one or two key assumptions. Use data-driven assumptions, making sure the time series includes stressed environments when possible. If a model of underwriting risk indicates that the probability of accident year combined ratios experienced from 1998 to 2000 is remote, it is not a realistic model