What Historical Data Does Not Tell Us About Emerging Liabilities

Actuaries and underwriters rely heavily on the law of large numbers: as observations increase, actual results tend to converge toward expected outcomes. For high-frequency, low-severity claims losses, this principle is one of the foundations of insurance pricing and risk selection. The challenge is that long-tail risks, which have the potential to take a long time to materialize or slowly escalate to catastrophic loss levels over time, do not behave the same way.

“Severe losses are rare, highly variable, and often shaped by factors that historical data cannot fully capture,” said Joanne Fung, Large Casualty, Senior Regional Underwriting Manager at Safety National. “The events that matter most frequently have the smallest and least credible datasets. As a result, historical averages can create an illusion of certainty. The limitation is not that the data is wrong, but that it does not capture the full range of possible outcomes.”

Risk Versus Uncertainty

A useful distinction exists between risk and uncertainty. Risk represents what can be reasonably estimated using available information. Uncertainty represents what remains unknown, evolving, or insufficiently understood. Most organizations spend enormous effort refining expected loss estimates. Far less attention is often devoted to evaluating how much confidence should be placed in those estimates.

For exposures such as artificial intelligence, autonomous vehicles, and emerging product liability, historical experience may be limited, rapidly evolving, or not representative of future outcomes. In these situations, the challenge is not only estimating expected loss but understanding how much confidence to place in that estimate.

Two insureds may have nearly identical historical loss experience yet present dramatically different levels of uncertainty regarding future outcomes. Strong underwriting, effective brokerage, and sound risk management all recognize and account for that distinction.

For brokers, uncertainty creates a unique challenge. Clients often expect decisions grounded in historical results, but emerging risks require a broader view. The broker’s role increasingly involves helping clients understand exposures that may not yet appear in their loss runs, identifying potential coverage gaps, and facilitating discussions around risks that traditional metrics may not fully capture.

Why Long-Tail Risk Requires a Different Mindset

Traditional underwriting often focuses on predicting the most likely outcome. Tail-risk thinking focuses on understanding what happens when key assumptions fail. For brokers, underwriters, and risk managers alike, that means looking beyond expected loss picks and historical trends and asking different questions like:

• Could a single event generate multiple claims?
• Could a common-cause failure create losses across numerous insureds simultaneously?
• Are there hidden aggregation exposures within the portfolio?
• Have legal, social, or regulatory changes outpaced historical data?

Many of today’s largest casualty losses arise not because operations suddenly become more hazardous, but because the liability environment evolves. Litigation financing, social inflation, changing jury attitudes, and evolving liabilities have altered the severity landscape beyond what past losses may suggest. In other words, a clean loss history does not automatically equal low risk. Sometimes, it simply means the underlying assumptions have not yet been tested.

What Autonomous Vehicles Reveal About Data Limitations

Autonomous vehicles illustrate this challenge particularly well. If the technology performs as intended, accident frequency may ultimately be lower than that of human-driven vehicles. From a purely frequency perspective, the long-term outlook could be positive. However, lower frequency does not eliminate tail risk.

Traditional automobile losses typically result from isolated driver errors. Autonomous vehicle systems introduce the possibility of software defects, systemic failures, and common-cause events capable of affecting numerous vehicles simultaneously. These represent fundamentally different risk characteristics than those reflected in historical driving data.

The question is not whether autonomous vehicles will be safer. The question is whether historical experience is enough to understand the potential tail risk. Emerging risks rarely announce themselves through deteriorating loss experience. Many begin as highly uncertain exposures that appear manageable until legal, regulatory, technological, or social developments fundamentally alter the loss landscape.

Confidence Is a Risk Indicator

One lesson repeatedly appears across emerging risks and severe casualty events: the risks that deserve the most attention are not necessarily those with the highest expected losses. They are often the risks where confidence exceeds knowledge. For underwriters, this means understanding the limitations of historical loss data. For brokers, it means helping clients identify emerging exposures before they become claims. For risk managers, it means ensuring that strategic decisions account for uncertainty, not just expected outcomes.

That requires humility, curiosity, and a willingness to challenge even the most credible-looking data. The organizations that navigate uncertainty most effectively are rarely those that perfectly predict the future. They are the ones that recognize uncertainty early and prepare for the possible outcomes.