Predictive modeling: Better decision making
Insurers and captives will benefit from more precise exposure data & loss data
By Michael J. Moody, MBA, ARM
"It's more complete electronic data that is the most significant change driving the advancements thus far in predictive modeling . . . this availability of data helps all the players along the pipeline."
—Michelle Bradley
Consulting Actuary
SIGMA Actuarial Consulting Group
Within actuarial science, one of the hottest topics today is predictive modeling. Much of the current commentary extols the virtues of this "new" approach to predicting future loss experience. According to one definition, predictive modeling "is a commonly used statistical technique to predict future behavior." That sounds like what actuaries have been doing for years. There is little new here; or is there?
Is it new?
In fact, predictive modeling is something new, and the major part of the new approach has to do with forms of datamining technology that can provide a much clearer picture of future outcomes. For the most part, in predictive modeling, data is collected, a statistical model is formulated and predictions are made. The model is then validated as additional data becomes available. The predictive analysis is most concerned with forecasting probabilities and emerging trends.
Michelle Bradley, ACAS, MAAA, ARM, CERA, notes, "Predictive modeling is about our ability to forecast loss costs, frequency and claims more accurately." Bradley, who is a consulting actuary at SIGMA Actuarial Consulting Group, acknowledges that actuaries have been doing this type of work for years. "It's just that the process has evolved so we can provide more advanced tools now." She believes that the models will evolve even further, thus allowing for more accurate prediction in losses.
Predictive modeling is made up of a number of predictors, which are variable factors that are likely to influence future behaviors or results. Improvements in information technology and datamining capabilities will continue to provide better, more current information for the statistical study.
Bradley points out, "It's more complete electronic data that is the most significant change driving the advancements thus far in predictive modeling." In the old days, she says, "frequently we were provided with several boxes of hard copy loss runs as our source for loss data." Today the data is provided in a format that is much better suited to more advanced statistical modeling. Now there is an abundance of recent and relevant data available in a wide variety of sources, she points out. "This availability of data helps all the players along the pipeline."
Further, she says, "It's not just loss projections that are benefiting from improved data." Quality data is being used in a number of different settings. "It is broader than just the actuarial profession." For example, she mentions, retailers are taking advantage of more complete and more timely data as well. "After you use your credit card to check out a few times, they know a lot about your buying habits." Data mining allows them to know just what coupon to provide you. She believes that this is just the beginning of other uses for all this data.
How will it affect insurance rates?
While more detailed information and data from predictive modeling will help develop more sophisticated models, the results should lead to more effective pricing mechanisms. As a matter of fact, not only should predictive modeling result in better pricing, the advanced results are just now scratching the surface of how they can be used to manage and improve insurance company operations.
Historically, as was noted previously, most predictive modeling has been limited by the lack of availability and difficulty in using various forms of data. The current abilities of the data mining methodology have greatly assisted the effort to secure more recent and relevant information. As a result, insurers will be able to quickly advance from models that were developed with less datadriven guidance. Not having enough data to make a prudent selection is not today's problem; it is finding a way to search through the volumes of data and find types of information that assist in the pricing task. Culling the list of variables will be one of the challenges going forward. But Bradley points out that it is not always advantageous to have more data. It needs to be relevant information.
She states that one of the benefits of the advances within modeling is that "overall, rates should more closely correlate with actual experience." However, she is quick to note that quality data is just half of the equation. She indicates that "in some instances we do not spend sufficient time obtaining comparable exposure data. In projecting losses, obtaining appropriate exposure base data will have just as much effect on the predictive outcome as proper loss data will."
While many of the above observations deal with insurance companies, similar reasoning also applies to alternative risk transfer mechanisms such as captives. When formation of a captive is being considered, it's critical for a corporate buyer and/or his broker to utilize the best available loss and exposure data. This will, notes Bradley, provide a more meaningful feasibility study. However, she cautions that since the exposure base may be relatively small for some captives, adjustments will need to be made to reflect this fact. But she believes that with current datamining capabilities, most organizations should be able to provide sufficient data if they are considering a captive.
Bradley also points out that the industry is finding new ways to use the results of predictive modeling. One is in claims management. "It can allow organizations to better manage challenging claims since methods could be used to predict or raise a 'red flag' on troubling losses." This approach would be used with a group of active claims, rather than closed claims and, she says, "It can provide management with information on how to most efficiently allocate their resources on open claims." This approach should help both insurance carriers and captive owners to make more informed decisions regarding open claims.
Conclusion
Long term, models that are based on more appropriate data and more advanced predictive models should yield better than average results. While the terminology may be changing, actuaries have been predicting losses based on past experience for years. Bradley points out that the changes that are occurring today are driven by "more complete and timely data and as such, well constructed models should provide projections (or predictions) that more closely correlate with actual experience." The profession has now advanced to the point that "the heart of predictive modeling is really focused on mining and incorporating relevant data into appropriate models." As a result, she says, "loss predictions for insurance companies or alternative risk transfer approaches such as captives should be more precise and offer more credible estimates."
