The Complement of Credibility in Predictive Modeling (Intentional or Not)
MuSigma Webinar Series, July 28th, 2021
Speakers: Chris Gross, Gross Consulting
Sometimes the concept of partial credibility is explicitly considered when predictive models are being built, for example through shrinkage methods or by insertion of credibility adjustments into a Baily Minimum Bias algorithm. More often, the concept of credibility is treated in a binary manner, with individual model parameters found to be either statistically significant (credibility = 1) or not (credibility = 0).
Whether credibility is explicit or implicit, understanding the complement of credibility is important and has real-world implications for model construction. This presentation will focus on this key actuarial concept and how it is often overlooked, with negative real-world consequences.