The CAS is pleased to announce prizes awarded for the best reserving research papers resulting from the CAS Reserves Working Group’s Call for Papers on Technology and the Reserving Actuary.
The winning papers are:
- “Navigating the Uncertainties: Robust Reserving Strategies for Catastrophic Events” by Olga Achkasova, ACAS, MAAA.
- “The Development and Use of a Claim Life Cycle Model” by Chris G. Gross, ACAS, MAAA.
- “Handling Sparse Data for Reserving Using Bayesian MCMC” by Michael R. Larsen, FCAS, MAAA.
These prize-winning papers, along with other outstanding papers stemming from the call for papers, are published in the Fall 2024 E-Forum.
Like other CAS research initiatives, CAS Call Paper Programs offer researchers the opportunity to support property-casualty actuaries by providing thought leadership and expanding actuarial knowledge. Papers are reviewed by CAS volunteers.
Three academics from Germany have won the ARIA Prize, a CAS-sponsored award established in 1997. The ARIA Board of Directors recently announced the paper “Detecting insurance fraud using supervised and unsupervised machine learning” by Jörn Debener, Volker Heinke and Johannes Kriebel in Volume 90 of the Journal of Risk and Insurance as winner of the 2024 ARIA Prize by the CAS. The ARIA Prize is awarded to the author(s) of a paper published by the American Risk and Insurance Association. ARIA publishes two academic journals quarterly (Journal of Risk and Insurance and Risk Management and Insurance Review). The winning paper provides a valuable contribution from the academic community to practicing casualty actuaries.
The paper acknowledges that insurance companies face a big problem with fraud, which has led to a lot of interest in using machine learning to solve it. While researchers have focused on using supervised learning to detect fraud, unsupervised learning hasn’t been studied much in this area. As a result, there’s not enough information to decide which type of machine learning is better for spotting insurance fraud. In this study, the researchers compare supervised and unsupervised learning using data from real insurance claims. They also partnered with an insurance company to test how well each method works in finding new fraudulent claims.
Some key findings include the following:
- Unsupervised learning, especially using isolation forests, can effectively detect insurance fraud.
- Supervised learning also works well, even though there are few labeled cases of fraud.
- Each method finds new fraudulent claims using different types of input information.
Based on these results, the researchers recommend using both supervised and unsupervised learning together rather than choosing one over the other.
The prize presentation took place at ARIA’s Annual Meeting this past August. One of the authors of the prize-winning paper will be invited to present the paper at a CAS meeting.
