The Data Revolution in Actuarial Science

There has been tremendous press coverage of "Big Data" and its implications for insurance companies, including captive insurers. Captive.com has participated in this coverage. And, while many of these stories have been interesting from a theoretical point of view, it's been difficult to quantify how beneficial this data revolution in actuarial science has been for insurers. However, a recent Milliman white paper, titled Individual Claim Reserving Models: Adding Value, piqued our interest because it began to crystalize how this information can start to have a tangible impact on individual insurance companies, including captive insurers.

Authors Alexandre Boumezoued, PhD, and Jeffrey A. Courchene, FCAS, MAAA, open with this sentence: "Individual claim reserving models are ready to complement aggregate triangle-based models, improving the reliability of reserve estimates and providing further insight into the drivers of claims and claim development." (For those readers who are not familiar with the basic actuarial loss reserve triangle concept, see a primer.)

For captive board members, the important thing to know is the basic loss reserve development triangle your actuaries use is typically comprised of industry data and your own data. The percentage of each type is primarily dependent on how much credible individual data your captive has concerning its losses. Younger captives will have fairly little credible data, while captives that have been in operation for 10 years or more, assuming they have a fairly constant flow of claims, will start to amass enough data to have it incorporated into the loss reserve model.

The Holy Grail for insurers is to have their loss development triangles be comprised solely of their own data since how claims are adjudicated and settled can vary significantly between insurers. In the Milliman paper, the authors refer to the traditional methodology as "aggregate triangle-based models (ATBMs)" and comment that they work well in relatively stable environments. They note, "However, for the vast majority of general insurance business analyzed, such stability is rarely observed, which means an actuary's confidence in the reserve estimates based on ATBMs can be low...." This can be especially true for captive insurers, where frequency and severity of individual claims can vary widely.

In discussing the use of ATBMs, the authors also list a series of limitations that can be important. While these are probably well known within the actuarial community and even by certain members of senior management or the captive management team, this editor (John Foehl) has never seen them in aggregate before. For captive board members, we have reproduced the list below because understanding these limitations can help when questioning your actuary about the loss reserve estimates being produced for the captive.

1. Loss of information when aggregating original claims data details for use in ATBMs.
2. ATBMs use a rigid structure of cumulative amounts with consistent cohorts of claims that are evaluated at consistent intervals of time.
3. ATBMs are overly dependent on the average age within a cohort of claims, while alternative explanatory variables may possess significant signal.
4. ATBMs are poorly equipped to identify and account for changing levels of estimation bias.
5. ATBMs exhibit large estimation errors for the least mature cohort of claims.

(Source: Individual Claim Reserving Models, by Alexandre Boumezoued and Jeffrey A. Courchene, published by Milliman, February 27, 2019.)

Note: for those who may not be familiar with all of the terminology above, IRMI has an extensive, free Glossary of Insurance and Risk Management Terms.

They go on to say actuaries have developed techniques to deal with some of these indications. As a captive board member, the next time you are listening to a presentation by your actuary, think about this issue and ask your actuary to describe how it deals with these limitations in the context of arriving at your own loss reserve estimates.

So, what exactly is an individual claim model (ICM)? As the name implies, an ICM uses individual claims level data, instead of aggregating the data, to estimate loss reserves. The Milliman paper says, as follows.

Evolution in technology with respect to efficient data collection, storage and analysis has made ICMs more accessible. To date, there is convergence neither with respect to an ICM analysis framework nor to the universe of model parameter assessment and validation techniques. Further, the amount of expert judgment required in an ICM analysis can be substantial. Nevertheless, the use of valuable information embedded in individual claim data is a promising feature of the approach that should lead to more reliable loss reserve estimates. Highly reliable loss reserve estimates are accurate, reproducible and consistent over time.

For captive insurers, this suggests the need to ensure the amount of data they are capturing concerning individual claims is as complete as possible. Many captives outsource their claims handling and management. Therefore, captive board members would be advised to have a discussion with either their captive manager or directly with the claims third-party administrator concerning the ability of the claims data as collected to be utilized in an ICM. It is never too early to begin planning for the future.

We would also encourage readers to download the entire Milliman white paper and read it in its entirety. It contains a list of references where those with further questions on this evolving technology can look for additional information.