Towers Watson again hails predictive modeling. National Underwriter summarizes:
Nearly 90 percent of U.S. insurance companies said the use of predictive modeling enhanced rate accuracy in 2010, according to the results of a survey conducted by Towers Watson.
The global professional services company reported that 70 percent of companies said the same in 2009. The use of predictive modeling in the U.S. is up about 10 percent across all lines of business except commercial property/business owners packages.
The survey indicates the use of predictive modeling continues to gain momentum. For example, 76 percent of the companies surveyed said they realized an improvement in loss ratio compared to 57 percent the previous year, and 68 percent said it improved profitability compared to 55 percent in 2009.
Neither NU nor Towers Watson mention it directly, but modeling is not only penetrating deeper into its original market, personal auto, but insurers are also broadening its use across lines, including into commercial business. Now, 32% of workers comp companies use predictive modeling, vs. 18% a year ago.
The emergence as a commercial lines tool is an important change. Commercial lines has used a (very) crude form of predictive modeling for decades, schedule credits. Underwriters of course, credit and (occasionally) debit insureds for indiosyncrasies in their business or the quality of their risk controls. Predictive modeling just puts quantitative controls on the process, precisely granting the credit and monitoring its effectiveness.
Of course, some could see the model as a threat to underwriters’ discretion, and that might help explain why BOP was the only line of business where use didn’t increase more than 10 percentage points. But really the models supplement the underwriter’s work more than replace it.
Finally, the Towers survey shows that regulators are good about approving models, but not so good about keeping details secret:
When asked how often they have encountered difficulty securing regulatory approval for new iterations of pricing predictive models, nearly half (49%) said less than 10% of the time, while an additional 31% said it occurs only 10% to 25% of the time. However, when filing for regulatory approval, 48% said they face challenges in keeping the details of their predictive models proprietary and confidential at least 25% of the time.
(Towers summarizes its survey here.)