The CAS Ratemaking Working Group is issuing a Request for Proposals (RFP) seeking research related to the development of scaling laws in pure premium models for use in the ratemaking process.

Casualty Actuarial Society (CAS)

The CAS was organized in 1914 as a professional society for the promotion of actuarial and statistical science as applied to insurance other than life insurance, such as automobile, liability other than automobile, workers compensation, fire, homeowners, commercial multiple peril, and others. Such promotion is accomplished by communication with those affected by insurance, presentation and discussion of papers, attendance at seminars and workshops, collection of a library, research, and other means. The membership of the CAS includes over 10,000 actuaries worldwide, employed by insurance companies, industry advisory organizations, national brokers, accounting firms, educational institutions, state insurance departments, the federal government, and independent consultants.

Ratemaking Working Group

The Ratemaking Working Group addresses actuarial issues of property and casualty insurance ratemaking including risk classification. The committee's charge includes furthering the development and dissemination of ratemaking theory and principles; identifying topics for research and discussion; monitoring professional developments and regulatory activities; and sponsoring panels, seminars, and other public forums on ratemaking issues.

Research Problem Description

Over the past 5 years, large language models have been the hottest area of machine learning research.  In the pursuit of improving these models, the primary debate has been how to best go about improving them.  Is it more data?  A larger neural network?  An entirely different type of model?  More training time?  A different network architecture?  A key piece of research that led to a push for ever increasing datasets and network sizes was Scaling Laws for Neural Language Models (2020, Kaplan et al).  In this work, the researchers were able to approximate power laws for the scaling of large language models which in turn were used to guide future model development in terms of volume of data and model size (generally, more data and bigger networks). 

Actuaries regularly face similar questions in their day-to-day work, often within their companies but also regularly between consultants and their clients or between regulators and carriers:  How much data is needed to build a 'good' model?  Have I saturated my model with variables?  How much data is needed to answer a business problem? 

While actuaries have had classical credibility to fall back on in the case of a rate indication (portfolio) to answer these questions, it becomes much more amorphous in a multivariate setting.  As such, it is proposed that actuaries carry out similar research for pure premium modeling in an insurance context.  The resulting scaling law would serve as a benchmark for all interested parties on model performance across a range of dataset sizes. 

Proposal and Work Product Requirements

We are seeking researchers to develop a scaling law for insurance ratemaking (pure premium modeling). Ideally, research would be conducted on a large dataset (>>1 million earned car years) but simulated data could be used in lieu and the final work products would include code, data, and a scaling law.  The CAS has a dataset that can be used for the purposes of this RFP on a confidential basis.  In selecting a dataset to use (CAS dataset, simulated, or other), respondents should consider trade-offs such as real-world vs. simulated data, the size of the dataset, the ability to know effect sizes and correlations in the dataset on an a priori basis.

Methods should have sound mathematical foundations, rooted in established principles of ratemaking. With that understood, the techniques within a method should be intelligible to an actuary working in insurance pricing. Concepts and techniques from the broader machine learning discipline are strongly recommended to be drawn on.

In addition to producing a research paper that will be published by the CAS, the selected researchers should also deliver an executive summary of the paper (two-three pages) that would be suitable as a blog post or magazine article. This summary should highlight the key themes of the paper and be understandable by a non-technical audience.     

The purposes of this are twofold - to better enable peer review of the research by the CAS and to facilitate the adoption of this research by CAS members.  To aid research adoption, the code will also be placed in the CAS’s GitHub repository, https://github.com/casact, under the MPL2.0 license.

Proposals should include a clear outline of the work that will be performed and the time frame in which it will be performed (including key dates). The proposal should be accompanied by the resumes of the researcher(s), indicating how their background, education, and experience bear on their qualifications to undertake the research.

The CAS contract will be awarded to the respondent who - in the judgment of the Ratemaking Working Group and entirely based on their written proposal - is best able to perform the work as specified herein. If the group determines that no proposal meets the requirements of the RFP, then no contract will be awarded.

Receipt of proposals will be acknowledged in a timely manner. Respondents who are not awarded the contract will be informed shortly thereafter.

Interested researchers should submit their proposals via email and any questions to:
Elizabeth Smith, Director of Publications and Research
Casualty Actuarial Society
esmith@casact.org

Timeline

Compensation

Compensation to researchers will be commensurate with the time required to carry out the work. Respondents should include an estimate of cost in their proposals. Costs of the proposal are encouraged to include considerations for the use of cloud computing resources.  Total cost should not exceed $45,000. 

Presentation, Ownership and Publication of Report

As a condition of selection, the CAS requires that all right, title, and interest, including copyright and patent, in and to the report and executive summary be owned by the CAS. The selected researcher/research team must sign a formal research agreement that assigns all such rights to the CAS. In any publication of the report and summary, the researcher(s) will receive appropriate credit with regard to authorship. The CAS may publish the report and summary in their entirety, or any sections thereof, in any format and medium as it finds fit, including but not limited to CAS publications, and electronic versions such as on its Web site or physical storage media.

The researcher(s) should make every effort to be available to present the report at a CAS meeting or seminar.