“Good judgment comes from experience, and a lot of that comes from bad judgment.” – Will Rogers

Based on public filings, we know many U.S. companies experienced an increase in Personal Umbrella claims severity leading up to 2020. We have seen these trends in our book as well. The pandemic’s favorable effects on claims frequency due to the reduction in miles traveled have overshadowed the consistent presence of claims severity. We expect severity to regain its prominence and become even more problematic as we emerge from COVID‑19.

To better understand these adverse trends, we looked to the typical known and measurable Personal Umbrella loss drivers, such as claims involving youthful operators, excessive speeding and drunk driving. We found that while these exposures continued to drive claims, less was known about how these and other variables interact and contribute to claims severity. For this analysis, we used a machine learning model. With this approach, a computer algorithm is created that automatically learns and improves from experience without explicit programming.

Our data analysts worked with our Underwriting and Claims teams to identify the right dataset to use for this exercise. In addition, we text-mined claim file narratives and accessed external data sources to enhance our analysis.

The model identified the following conditions as key drivers of claim severity:

• An increase in the annual poverty rate (see graph)
• An increase in opioid prescription rates
• An increase in fatal accidents
• Brain injuries
• Attorney representation
• Injuries involving a fatality
• Multiple claimants

Other notable, albeit less impactful, predictors were correlated with higher claims severity, including:

• Laws permitting recreational marijuana
• The lack of laws requiring helmets for motorcyclists

Some of these findings are more obvious – such as claims involving brain injury and death – and their effects are duly quantified by the model. Others are more unexpected, such as the effects of poverty rates, opioids usage and marijuana laws. All will facilitate deeper client interaction on this line of business.

These findings will inform our underwriting and pricing models and help us in evaluating and reserving claims. As for next steps, we will continue to improve and standardize the claims data we collect. One example of this is the zip codes associated with accident locales. This information is often inconsistently captured in claim files, but knowing it will enable the model to pinpoint trends at a more granular level.

With this new view of Personal Umbrella claims severity drivers, perhaps we can rescript Will Rogers’ adage to read: Good judgment comes from experience, and a lot of that comes from better judgment based upon an improved understanding of the data.