CAS Releases the 2022 Winter E-Forum Featuring a Machine Learning Working Party Report and Ratemaking Essays

The CAS is pleased to release the Winter 2022 E-Forum, which includes a report of the CAS Machine Learning Working Party titled “Machine Learning in Insurance.” This edition also features two submissions in response to the 2021 Call for Ratemaking Essays, conducted by the CAS Ratemaking Working Group.

Published research on machine learning in an insurance context is sparse — a fact that the CAS Machine Learning Working Party identified as a barrier to for actuaries drawn to machine learning wishing to use it in their work. “Machine Learning in Insurance” gives references and descriptions of current research and is a guide for actuaries wanting to learn more about this advanced field.

In the ratemaking essay, “Decision Trees and Categorical Independent Variables with Many Levels,” Chao Guo discusses categorical independent variables with several levels — a common concern about decision tree models. In examining the decision tree algorithm and running numerical simulations in R, Guo determines that categorical variables with many levels do not always cause trouble and offers other factors to consider.

“Alternative to Tweedie in Pure Premium GLM,” by David R. Clark, FCAS, MAAA, is a ratemaking essay that proposes using the quasi-Negative binominal (QNB) as an alternative to the Tweedie distribution. Clark maintains that both can be interpreted as collective risk models, but the QNB has a variance structure that is used more commonly in other actuarial applications.

The Winter 2022 E-Forum embodies the CAS’s on-going commitment to advancing the field of actuarial science and preparing its members for opportunities and challenges in the future.