Business Objective
Our client is a re-insurer that underwrites insurance provided to commercial auto businesses.
One of their key needs is to ensure alignment of the reinsurance premia in line with the risk profile of the loan portfolios, which consist of commercial auto businesses with varying fleet ownership and behavioral profiles.
Challenges
- Fatalities per power unit (owned by the business), are rare events. Standard regression methods are not ideal to capture this phenomenon
- For the analyses to capture events with a usable degree of accuracy, data must be sourced from beyond traditional internal sources. Geospatial data trends over time had to be used.
Solution Methodology
- Performed exploratory analysis to understand relationship between fatalities per power unit (historical) and a range of firmographic variables (size of business, number of power units, individual DOT registrations etc.,)
- Acquired additional data from Commercial Auto Bureau to understand various characteristics such as speed violations, and other forms of traffic violations, indicating propensity of risky on-road behavior
- Modeled distribution of incidences using “Tweedy” distribution function as opposed to traditional statistical distributions for classification problems such as binomial logit models based on odds ratio
- Developed an accurate final model – a classification & regression hybrid, for incidence rates and mapped the modeled rates in existing risk buckets against premia realized in each bucket; finally, simulated the ideal premia for each risk bucket using modeled fatalities while also maintaining a consistent pricing trend across risk buckets.
Business Impact
- Established a potential gain in risk adjusted premia of close to USD 4MM on business underwritten in last 2 years.
- Validated the results with client’s Chief Data Scientist and presented to business stakeholders for pricing decisions.
