Actuaries Argue For Credit Scoring

To The Editor:

Mr. Bock's letter challenging the validity of insurance scoring (NU, July 28, page 32) shows commendable actuarial instincts, but he draws a number of questionable conclusions, specifically that:

The predictive accuracy of scoring models is in question simply because all insurers do not reach the same score from the same information;

The scores emerging from models are tainted due to the relationship between creditworthiness and the average policy size purchased by the insured;

Insufficient information has been published to support or refute the first two conclusions drawn.

All three assertions are incorrect, and we can explain why.

As background, all scoring models are just distillations of the data on consumer reports. Some companies have chosen to use one, or sometimes several, consumer reports (such as credit, CLUE, and/or MVR) to derive a "custom" score tailored to predict policy performance for that insurer's book. Other companies have chosen to use "standard" scoring models to predict policy performance. These models are based solely on the credit report and developed for the industry as a whole by the resellers of credit information (e.g. Fair, Isaac and ChoicePoint).

Mr. Bock is correct that, depending on the existing and target markets of each insurer, a consumer may well receive a different score when evaluated in various insurance scoring models. Further, the range of scores, the average and the spread are all calibrated differently by each model. So a three-digit numerical score from one model is not comparable to the three-digit number from another.

What matters for underwriting and rating is where the consumer stands relative to the average risk in the insurer's existing or target market.

The key point is that there is no one right answer, precisely because insurers show a range of underwriting and marketing goals in our highly competitive marketplace. Just as some insurers consider low-limits liability risks ideal and others write only high-limits risks, credit scoring is influenced by insurer strategy.

Surely Mr. Bock would not be surprised to find that some insurers react differently to the presence of speeding tickets on a motor vehicle record than others, or give a larger claims-free discount than others because it is justified for their group of customers.

Mr. Bock correctly identifies the presence of relationships between traditional rating variables (such as policy limits and deductibles) and insurance score. Actuaries call this "correlation," and it has not escaped our attention in development of rating algorithms that use scoring as an input.

Several complex actuarial techniques are used to correct for this "overlap" between traditional variables and the new rating variable (insurance score) when introducing new rating plans.

Further, regulators typically require demonstration of our consideration of these effects when approving the new plans.

Mr. Bock also points out that loss ratios may, at first glance, appear better for high-scoring risks simply because they have a larger premium base. We work around this problem in several ways.

First, we adjust premiums used in calculating loss ratios "to base class," meaning we divide out the effects of increased limits factors and class factors to restate the premium "as if" it were written at the base class and limits. This technique is universally used in the analysis of traditional rating factors (such as driver class and territory) as well.

Second, we study many policy performance measures in addition to loss ratio. In particular, the measure described by Mr. Bock, number of losses to number of policies (or vehicles), is called "claim frequency" and is examined very carefully when setting score-based rates.

It turns out that most studies have concluded that insurance score predicts future claim frequency just as well as (and sometimes better than) it predicts future loss ratio.

Finally, it is possible that Mr. Bock may have experienced frustration in obtaining information on the predictive power of insurance scoring with regard to claim frequency and loss ratio, but we note that in many states (such as Florida), the rate filings which support the utility of scoring, containing detailed data and substantiating the resulting rating plans, are public record.

In addition, public domain studies such as the recent one by the University of Texas (www.utexas.edu/depts/bbr/) confirm the validity of insurance scoring using several measures of policy performance.

Other studies conducted by Casualty Actuarial Society members include:

Rade Musulin, ACAS, MAAA

John Rollins, FCAS, MAAA

Florida Farm Bureau Insurance Cos.

Gainesville, FL

(Editors Note: Mr. Musulin, an Associate of the Casualty Actuarial Society, authored an article titled "Lets Give Credit Where Credit is Due" for NUs Opinion page in the Aug. 26, 2002, edition.)

Reproduced from National Underwriter Property & Casualty/Risk & Benefits Management Edition, August 25, 2003. Copyright 2003 by The National Underwriter Company in the serial publication. All rights reserved.Copyright in this article as an independent work may be held by the author.

NOT FOR REPRINT

© Arc, All Rights Reserved. Request academic re-use from www.copyright.com. All other uses, submit a request to TMSalesOperations@arc-network.com. For more information visit Asset & Logo Licensing.