Derivative Lasso: Credibility-based signal fitting for GLMs
Building transparent and accurate models for insurance risks is challenging due to the complex correlations and non linearities of the modeled effects. Generalized Linear Models (GLMs) are an essential tool to handle correlations and build transparent models. However they require a lot of iterative work to incorporate nonlinearities and develop robust and credible models.
The white paper suggests that these drawbacks can be effectively solved within the Penalized Regression framework, in a manner that does not change GLM’s input hypotheses’ and table based output. Furthermore, the approach is sound as it relates to intuitive statistical assumptions that integrate credibility and nonlinear effects in the modeling. We will also provide the insurance practitioner with guidance on how to build and analyze these models with examples from publicly available data.
Register
Our team will contact you with the event details.
Derivative Lasso: Credibility-based signal fitting for GLMs
Building transparent and accurate models for insurance risks is challenging due to the complex correlations and non linearities of the modeled effects. Generalized Linear Models (GLMs) are an essential tool to handle correlations and build transparent models. However they require a lot of iterative work to incorporate nonlinearities and develop robust and credible models.
The white paper suggests that these drawbacks can be effectively solved within the Penalized Regression framework, in a manner that does not change GLM’s input hypotheses’ and table based output. Furthermore, the approach is sound as it relates to intuitive statistical assumptions that integrate credibility and nonlinear effects in the modeling. We will also provide the insurance practitioner with guidance on how to build and analyze these models with examples from publicly available data.
Learn more about the speakers
Max Martinelli
Max Martinelli is a Managing Director in the Insurance and Actuarial Advisory Services practice of EY, where he focuses on P&C pricing, actuarial data science, applied AI and external rating. He has over a decade of experience in actuarial and data science roles, primarily in predictive modeling for personal and commercial lines, and a background in machine learning and computational mathematics. A dedicated collaborator with the Casualty Actuarial Society, he co-designed and led the CAS AI Fast Track bootcamp and co-hosts Almost Nowhere, the CAS Institute podcast exploring AI and data science in insurance.
