NEW YORK-- April 11, 2022 -- Akur8 released today a new research paper for the actuarial community entitled “Credibility and Penalized Regression.” Developed by a senior team of actuaries and data scientists at Akur8, the purpose of this new paper is to provide practitioners with key concepts and intuitions that demonstrate how Penalized regression blends Generalized Linear Models (GLMs) with Credibility-like assumptions.

In recent years a number of adaptations to GLMs have been developed to address some limitations, such as their inability to incorporate Credibility-like assumptions. These adaptations are widely adopted within the Machine Learning community, however they have not been very popular within the actuarial world. Credibility methods and GLMs are part of the standard actuarial toolkit of predictive modeling, but the actuarial literature describing how Penalized regression blends Credibility with GLMs is not equally developed.

“By exploring how Penalized regression (and Lasso in particular) can be interpreted from the perspective of both Credibility and GLM frameworks, this paper’s objective is to familiarize practitioners with Penalized regression as an extension of established actuarial techniques, instead of considering it one among several new modeling techniques from the Machine Learning and Data Science literature” noted Guillaume Beraud-Sudreau, Co-founder & Chief Actuary at Akur8.

“Our team of actuaries and data scientists at Akur8 worked closely together to produce this comprehensive research paper on Credibility and Penalized Regression. We are excited to publish this information in an effort to help expand the literature available to the actuary community on this important topic” stated Samuel Falmagne, Co-founder & CEO at Akur8.

Specifically developed for actuaries and predictive modelers, Akur8’s solution enhances insurers’ pricing processes by automating technical and commercial premium modeling with proprietary transparent machine learning technology. The core benefits for insurers include a reduction in data preparation and modeling time which effectively accelerates time to market and the production of more predictive models, while ensuring full transparency and control of the models created.