ISO Review Forum: Analytical Opportunities across the Property/Casualty Life Cycle

Insurers Can Augment Traditional Ratemaking with Advanced Predictive Modeling

By Mark Richards

Pricing is a classic area for application of analytics. "Insurance rate" is simply an analytic product. Increasingly, advanced predictive modeling techniques are being used to augment traditional ratemaking. Among the key benefits of predictive modeling methodologies is the ability to analyze significant volumes of diverse data and discover patterns at fine levels of granularity. For instance, traditional territorial ratemaking and credibility theory require territorial geographies not only to be homogeneous enough but also large enough for the derived statistics to be significant. These classical approaches are often of a hierarchical decomposition nature — decompose the country into states and states into territories to ensure the territories satisfy credibility requirements. Attention is not paid to the fact that risks or sub-geographies in a certain territory may be extremely similar to those in another. The credibility requirements are expected to be satisfied within each territory, not across.

Predictive modeling expands the horizon, inherently discovering statistically significant risk patterns across geographies or other risk groups used in traditional ratemaking. This allows carriers to develop precise ways of segmenting and pricing their risks.

However, building a predictive model–based pricing structure is a daunting exercise. Risk scoring at a fine granularity requires consideration of numerous external data sources, such as weather, geological maps/features, census characteristics, locality characteristics, traffic densities and patterns, vehicle characteristics, home characteristics, and so forth. Identifying, exploring, and including such data sources can be significantly costly, both financially as well as in time and resources. Specialized tools to geocode locations are required, as are significant computing and storage capabilities. When processed, the data sources often yield thousands of predictor attributes requiring algorithms for sampling and variable selection. In all, such initiatives are easily multiyear efforts even for large analytics teams.

SmartCo looks to make quick headway in innovative risk scoring and pricing. Consequently, the company chooses a buy-and-customize option instead of a build option. It licenses ISO Risk Analyzer® — a sophisticated rating paradigm that leverages predictive modeling methodologies and a number of unique sources of data. The analytics and actuarial teams at SmartCo use the ISO Risk Analyzer solutions in a number of ways.

For their personal auto rating, SmartCo actuaries choose the power of ISO Risk Analyzer Personal Auto (RAPA) and leverage its accurate loss cost estimates at a granular level. The powerful differentiation and refinement offered by the product are clearly visible in their territorial maps, as illustrated in Figure 3. The map on the left shows their current loss costs by territory in New Jersey, while the one on the right shows the RAPA loss costs at a block-group level. The spotty map provides clear evidence that very few of the current territories are "pure." Most territories have sub-geographies that are less or more risky than the territory as a whole.

Figure 3		Figure 4
Click to enlarge	Click to enlarge	Click to enlarge
Current loss costs by territory in New Jersey	RAPA loss costs at a block-group level	New territories from multidimensional clustering using RAPA loss cost estimates

SmartCo's analytics and actuarial teams analyze the model results on their book of business and develop two broad strategies for adoption. In states with significant books of business and updated territories, the analysts assume the territorial loss costs are accurate. They perform revenue-neutral analysis at the territorial level, developing, and consequently filing, rating relativities for block groups within the territories. When applied to new business, each garaging address is geocoded and its block group identified. The corresponding block group–level relativity is then applied to the corresponding territorial loss cost estimate to produce the effects of micro-segmentation.

In states with thin books of business or outdated territories, the analysts perform multidimensional clustering using RAPA loss cost estimates (and other attributes) to develop homogeneous and credible territories along with new territorial loss cost estimates. For instance, new territories (with yellow boundaries) for New Jersey are illustrated in Figure 4.

To support SmartCo's expansion into new states such as Florida and Georgia, the analytics and actuarial teams follow a similar approach. They leverage the granular location loss cost estimates from ISO Risk Analyzer to craft homogeneous territories and estimate territorial loss costs. They then file the territories and loss costs for use in pricing new business onto their books.

The examples provided illustrate the use of advanced analytics in pricing. Since pricing is a function of estimated future losses based on past loss experience, it is critical to consider attributes that increase or decrease loss severities or payouts. For instance, fraudulent claims can drive up loss payouts, thereby impacting not only current losses but also estimates of future losses. By controlling loss expenses and inappropriate loss payments, a carrier can significantly improve its competitive pricing and possibly market share. The next article describes these opportunities in claims analytics.

Mark Richards is director of analytics at ISO Innovative Analytics (IIA), a unit of ISO focused on delivering advanced predictive analytic tools to the property/casualty insurance industry.