Unlike the number of claims, it is not easy to prove mathematically the optimal modeling technique for the claim severity. However, based on the typical distribution of the claim severity curve, the gamma, lognormal and Pareto³ distribution are the typical candidates to be considered.

In-depth industry knowledge and detailed analytical approach like curve fitting, etc. is required to find the optimal approach especially when the number of claims is small or some claims are of extremely large value (as is the case with BI or PD).

In our experience, Gamma or truncated Gamma functions give superior predictive results as compared to other regression techniques.

Merits of using these new statistical techniques are enormous. However model development is a complex and error-prone process. While many completed models work as planned, some models contain fundamental errors. Moreover, the internal logic of most models is usually very abstract and limiting, so it requires considerable judgment and expertise to apply model results in the appropriate business context.

(C) MODEL STRUCTURE
Traditionally, models are divided into a class plan and a tier plan, with no overlap of variables.

This leads to sub-optimal models as interactions between the two sets of variables are not captured. As a result, companies must adopt a holistic structure. Since every coverage has its own characteristic and behavior with regards to frequency and severity, each coverage should be modeled separately.

One could sometimes group coverages into low frequency high severity (e.g., bodily injury), and high frequency-low severity (e.g., collision). Figure 3 shows a typical distribution of severity and frequency for various coverages.

Once the national level models have been built at the individual coverage level, state specific versions have to be built as well to account for the difference in “regulatorily approved variable set”.

(D) USE OF EXTERNAL DATA
Third party data providers like Polk, Carfax, ISO etc. provide reliable information on driving history, new drivers in a household, car history, etc. to enable better decision making. The emergence of these sources has made collection of the same information from consumers redundant, thus simplifying the quote generation process as well. Figure 4 shows a list of such third party providers.

(E) ALLOCATION OF OTHER EXPENSES
An auto insurance company incurs various loss and non-loss related expenses. A robust pricing methodology should account for all the expenses incurred in addition to loss and LAE, i.e. acquisition
expenses and general indirect expenses. As pricing sophistication develops, all three cost elements should be handled scientifically and passed on to consumers appropriately based on the true cost drivers. While losses (including loss adjustment expenses) represent the biggest component of cost, companies no longer need to use a simple constant factor to allocate acquisition expenses. They can now use measures such as ease of acquiring a customer, conversion rates, his/her retention rate etc. to identify customers who should be allocated higher portion of the acquisition costs. Going forward, we can expect more companies to:

ADDRESSING KEY REGULATORY CHALLENGES
Despite all of these technological advances, companies continue to face challenges in the marketplace unique to the insurance industry.