Analytics refers to the process of utilizing a series of mathematical techniques to gain insights through data. As an actuary who lives and breathes analytics in daily work, it is inspiring to see it appear as a common theme across various sessions during the 2024 CAS Annual Meeting.
Using Predictive Analytics to Price Med Mal Risk
Presented by William E. Burns, ACAS, MAAA, and Matt Koerlin, MBA
Although many industries have benefited from the use of data and analytic techniques, Burns revealed a different journey in pricing physicians’ liability risk. Advancements in research and analytics seem to have limited usage in the malpractice field, as medical professional liability (MPL) pricing has been based on the same set of criteria (specialty and territory of practice) over the past 50 years. Tradition and politics are two key factors that keep this field operating as it has for decades.
Burns then shared a brief history and current state of pricing medical malpractice (med mal) risk, and Koerlin showcased the recent success in leveraging analytics for med mal pricing and risk management. This innovation could make underwriting easier in an actual business environment.
Connecting the model to medical and pharmaceutical billing transactions, Koerlin relayed how the model utilizes this big dataset to predict the likelihood of each doctor having a claim during the next year. Koerline also explored implications related to cross validation against big data and to benchmarking for risk management and behavioral change.
Math works perfectly on the actuarial side in this scenario, Koerlin commented, but operational considerations are the real battlefield. A couple of common challenges include recency bias, correlation versus causation and misconception around “high score is bad risk.”
The session closed on the outlook for the future. Even if it is not always easy to embrace change, analytics does open the door to further standardize and automate the underwriting and renewal process for med mal risk.
Model Weighting, Ensembling and Stacking
Presented by John M. Shoun, CPCU
With increasing analytics tools available nowadays, it is natural to ask how we could choose and consolidate different models into a “super model” to optimize performance.
Shoun explained how including the average of historical results can make models more accurate. In fact, if there is any standalone model or expert judgment that has low correlation with the current modeling technique, it is often better to include it through model weighting or ensembling. That is, more diversified opinions can improve the model outcome.
Shoun also demonstrated how to use feature-weighted stacking to estimate unpaid claims by selecting different techniques between expected loss ratio, Bornhuetter-Ferguson or chain ladder based on ages of claims. He pointed out that the stacking concept is nothing new to actuaries. It allows us to understand how model stacking weights vary among different circumstances. However, model stacking is not always a better solution in terms of cost and benefit assessment given the complexity in its implementation against a marginal model improvement.
Conclusion
In one way or another, all the speakers emphasized that analytics can enlighten ideas for future work. However, because data does not reveal secrets easily, analytics without context is meaningless. We should keep business and operational considerations in mind while conducting any analytic work. In addition, no matter how complicated the model becomes, understanding the statistical basis and fundamentals is just as critical as implementing the model.
Yuhan Zhao, FCAS, is a senior actuarial manager with Aviva Insurance Company of Canada. She is a member of the AR Working Group and the Monograph Editorial Board.
