Providing thought leadership with a continual focus on the essentials, CAS Research delivers quality research thanks to the dedication of nearly 200 volunteers who assure that every piece contributes to the overall body of knowledge for the P&C actuarial profession.
The top 15 most visited CAS research reports and papers feature fresh thinking on emerging topics such as artificial intelligence, while also showcasing modeling techniques for ratemaking, reserving and reinsurance. Other topics include finding bias in ratemaking, developing useful marketing models and taking a new look at autonomous vehicles and potential approaches to insuring them.
Not surprisingly, the most popular research is relevant — being released in 2023 or 2024. However, a couple of papers from 2021 are holding their own, demonstrating that older papers can maintain relevance.
1. Loss Modeling from First Principles
Pietro Parodi, Derek Thrumble, Peter Watson, et al., E-Forum, 2024
The authors establish a first principles approach that reshapes loss modeling and enhances clarity, precision and predictive power by balancing data fitting with dynamic risk. The methodology avoids complex, parameter-heavy methods, proposing those grounded in intuition and mathematics instead.
Why Read: Modeling from first principles can still be the best approach.
2. GLM for Dummies (and Actuaries)
David R. Clark, E-Forum, 2023
Offering insights that support robust, interpretable and adaptable ratemaking models, the paper addresses common modeling challenges, such as data sparsity, regulatory requirements and real-world variability. By calculating a fitted model so the weighted average of the fitted loss costs balances with the actual data, the paper strives to make the calculation more intuitive.
Why Read: Learn what anyone (including actuaries) would want to know about GLMs — but were afraid to ask.
3. Machine Learning and Ratemaking: Assessing Performance of Four Popular Algorithms for Modeling Auto Insurance Pure Premium
Sofia Colella, Harrison Jones, E-Forum, 2023
By integrating modern, evolving machine learning ratemaking techniques, such as XGBoost and neural networks, the authors discuss the competitive advantage that GLMs can provide. While offering actionable insights on tuning and performance, the paper can also help readers understand the trade-offs between accuracy and interpretability.
Why Read: Machine learning is powerful, but there’s always nuance.
4. A Simple Method for Modeling Changes over Time
Uri Korn, Variance, 2021
Using a regression-based state space model (RSSM) to enhance time series forecasting in reserving by blending penalized regression and time series components, the author demonstrates how to improve historical data interpretation and forecasting accuracy. By highlighting practical applications in reserving, profitability studies and insurance pricing — along with scalable solutions for big data — this paper stands the test of time.
Why Read: There is an innovative approach to time series forecasting.
5. Ultimate Loss Reserve Forecasting Using Bidirectional LSTMs
Lahiru H. Somarantne, E-Forum, 2022
By introducing models like recurrent neural networks (RNN) and long short-term memory (LSTM), the author reveals how machine learning can handle temporal components of loss data effectively, surpassing traditional methods such as chain ladder in accuracy, especially for volatile early loss development periods.
Why Read: Leveraging machine learning in predictive loss reserving can address age-old challenges.
6. Framework of BERT-Based NLP Models for Frequency and Severity in Insurance Claims
Shuzhe Xu, Vajira Manathunga, Don Hong, Variance, 2023
Introducing an innovative structure to leverage textual information in insurance datasets using bidirectional encoder representations from transformers (BERT)-based natural language processing (NLP), the paper demonstrates why integrating BERT with artificial neural networks enhances predictive accuracy and stability for claim frequency and severity — and also outperforms traditional approaches.
Why Read: Incorporating text data in complex models is made possible with a practical, powerful approach.
7. Capital Allocation Techniques: Review and Comparison
Qiheng Guo, Daniel Bauer, George H. Zanjani, Variance, 2021
Bridging theory and practice, the authors provide a critical review of capital allocation methods, exploring their underpinnings, practical implementations and stability through examples. This still sought-after paper also identifies key differences between methods, tail-focused measures and those considering entire distributions, covering the instability of methods such as value-at-risk (VaR) under certain conditions.
Why Read: The C-suite will always appreciate capital allocation supported by robust metrics for portfolio optimization and risk-adjusted return.
8. Recommender Systems for Insurance Marketing
Giorgio Alfredo, Giuseppe Savino, Variance, 2022
Just as e-commerce and entertainment industries use state-of-the-art recommender system algorithms to market their businesses, so can insurance companies — with help from actuaries, of course! The authors show how supervised learning models, such as gradient boosting and neural networks, can better predict insurance purchases compared to traditional techniques. As an added bonus, the paper shares insights to enhance cross-selling strategies, improve customer engagement and drive business growth.
Why Read: It never hurts to gain appreciation from the marketing department.
9. A Practical Approach to Quantitative Model Risk Assessment
Carole Bernard, Rodrigue Kazzi, Steven Vanduffel, Variance, 2023
Building a practical framework for quantitative model risk assessment, the authors highlight the risks from model assumptions, propose innovative tools to measure assumption contributions to model risk and introduce a formula for determining model risk capital. The research also helpfully addresses regulatory requirements and offers ways to enhance model reliability.
Why Read: Evaluating, mitigating and communicating model risks can bolster stronger financial decision-making.
10. Projection of On-Road Liability Losses for Autonomous Driving
Tetteh Otuteye, Corey Rousseau, Rafael Costa, et al., E-Forum, 2022
Integrating actuarial insights with autonomous vehicle (AV) technology and safety advancements, the authors cover the evolving challenges in assessing liability and offer ways to project risks, price coverage and anticipate reserves by considering liability exposure quantification, collision frequency, claim severity and loss distribution. The authors also discuss blending product and personal liability, regulatory framework variation and historical data scarcity.
Why Read: Staying up to speed on AV technology and the implications for insurers is future-critical.
11. An AI Vision for the Actuarial Profession
Ronald Richman, E-Forum, 2024
Highlighting the value of combining AI with traditional actuarial principles while addressing challenges including bias, ethics and regulation, the paper presents methods for improving efficiency, accuracy and innovation in actuarial disciplines such as pricing and reserving.
Why Read: AI will forever change the actuarial profession.
12. The Actuary Takes the Stand: Compensation for Personal Injury
Sule Sahin, Gary Venter Variance, 2024
Offering a transformative approach to litigation and compensation, the authors introduce a hybrid methodology combining systems in the U.K. and the U.S. to enhance fairness in calculating compensation through innovative age-earnings profiles.
Why Read: Enhanced methodologies are available to quantify loss of income and to potentially set reserves.
13. A Practical Guide to Navigating Fairness In Insurance Pricing (Part of Phase II of the CAS Research Paper Series on Race and Insurance Pricing)
Jessica Leong, Richard Moncher and Kate Jordan, 2024
Offering a framework for developing models more likely to comply with evolving regulations concerning unfair discrimination and bias, the authors discuss governance approaches — along with pros and cons of each — for bias mitigation in all stages in insurance modeling.
Why Read: Today’s regulatory climate demands that insurers consider bias that could be embedded in their pricing models.
14. Regulatory Perspectives on Algorithmic Bias and Unfair Discrimination (Part of Phase II of the CAS Research Paper Series on Race and Insurance Pricing)
Lauren Cavanaugh, Scott Merkord, Taylor Davis, et al., 2024
By providing the views of 10 state insurance departments, this critical report gauges regulatory commitment — and concerns — related to potential bias in insurance pricing. Is using race or ethnicity to conduct bias testing appropriate? When conducting testing, should there be one or multiple tests? What about using the Bayesian improved first name surname geocoding (BIFSG) as a proxy for racial data?
Why Read: Regulators may increasingly consider racial or ethnic disparities in insurance outcomes as well as how best to test for them.
15. An Actuarial Approach to Stochastic Modeling of Casualty Catastrophe Risk
Neil Bodoff, Eric Dynda, Brandon Stevens, et al., E-Forum, 2023
Natural catastrophic (CAT) models and scenarios have come a long way, but not so much for casualty CATs. That’s where stochastic modeling can come in, according to the paper, which also identifies flaws in current methods and emphasizes accurate tail risk quantification for ratemaking. Bonus features include actionable algorithms and critical insights for robust risk management.
Why Read: When a century-old household staple (talcum powder) leads to billions of dollars in losses, actuaries must be ready for surprises.
Finding CAS Research
Popular research tends to appeal to broad audiences, but with thousands of research reports and papers available on the CAS website, there is 100% likelihood that members will discover engaging content.
By topic, members can locate published CAS research by searching:
CAS research is also shared through presentations, so check out the Professional Education Library as well.
