Winter Storm Risks Tough To Anticipate

In recent decades, winter storms have inflicted severe property damage across swaths of North America, resulting in considerable insured losses.

Although frequently overlooked as a major peril, winter storms constitute around 10 percent of the overall annual loss from natural catastrophes on average in the United States, and nearly one-third in Canada. For some locations–especially in the Northeast and Northwest–this figure can rise to as much as 60 percent.

Severe winter storms–such as the 1962 Pacific Northwest Windstorm, the 1983 Severe Arctic Temperature Freeze and the 1993 “Storm of the Century”–highlight the potential devastation that can occur from winter storms.

These complex, extra-tropical weather systems can produce various types and combinations of damage from snow, ice, freezing temperatures and extra-tropical winds.

Traditionally, insurers have used historical claims data and/or actuarial methods to establish their technical coverage rates and determine loss reserves from winter storms. However, there are often a limited number of years of high-quality data they can rely on. This raises problems, since winter storm activity is not static and insured losses vary wildly from year to year.

Since the active winter storm seasons of the mid-1990s, when insured losses reached nearly $20 billion in today's values, winter storm losses have generally been much lower–on average at around $1-to-2 billion annually. A majority of the high-quality claims data resides during this period of reduced activity and can lead risk managers to make decisions that are underestimating their exposure to winter storms.

Essentially, this decade of historical claims data is not a good representation of the next five-to-10 years, when activity may rise again.

Making decisions on insurance or reinsurance purchases for winter storms has also been a challenge. Given the relatively short historical record, estimating the possibility and potential costs of a severe or “tail event” have been difficult.

The most severe winter storm in the past 25 years was the 1993 “Storm of the Century.” From March 12-14, 1993, a powerful extra-tropical winter storm descended upon the eastern half of the United States, causing widespread damage from the Gulf Coast states to Maine and even into the eastern portions of Canada.

The winter storm caused record snowfalls across the Appalachian Mountains and mid-Atlantic states, hurricane-force winds, extremely low temperatures throughout the region, and even spawned tornadoes in Florida.

But while this was a large and costly event, which could cost as much as $5 billion if it were to reoccur today, how can we tell if it is representative of an extreme winter storm–a one-in-100-year event–given the relatively short historical record?

Another difficult question when using historical claims data is how to set technical rates in new geographical locations or for new lines of businesses. While locations may be geographically similar, there can be large variations in winter storm hazard and climatology, as well as building code designs and building stock that can affect the risk.

Likewise, expanding into new lines of business and setting technical rates based on a similar business can be quite misleading.

Historically, probabilistic risk models have been used for low- to mid-frequency and high-severity perils like hurricanes and earthquakes. The historical record is also incomplete for these perils and therefore cannot be relied on for risk differentiation and making decisions on portfolio management and reinsurance purchase.

Since these perils have the potential to cause significant insured losses, probabilistic models are used to assess the risks more effectively.

The challenge of quantifying winter storm risk is similar to hurricanes and earthquakes. Although winter storms are mid-high frequency and low-mid severity events, a probabilistic model can provide more insight to the causes of loss because all major sources of damage from snow, wind, ice and freezing temperatures can be explicitly captured, so it is possible to identify which individual winter storm peril or combination of perils are driving the risk.

This can be particularly useful for accurately assessing risk when writing insurance in new geographical areas.

One of the most significant challenges in risk management is identifying the likelihood and severity of extreme events to a portfolio. The way this is addressed through probabilistic modeling is by creating new and realistic winter storms that have not been historically observed in the short climate record.

This information allows insurers and self-insured risk managers to make more informed portfolio management decisions, particularly in assessing the need to mitigate risk through reinsurance.

There is a large discrepancy in the way buildings react during winter storms, due to their age and construction type, but this is rarely captured within historical claims data.

Through engineering studies and scientific research, vulnerability curves that specify the level of damage given a certain hazard value can be developed for each primary building characteristic in a probabilistic model for each peril.

Knowing how a building is likely to respond during an event will help insurers to set appropriate rates, as well as provide insight into how the risk could potentially be reduced through mitigation measures.

Advances in probabilistic modeling for winter storms will fill many of the gaps that currently exist from historical claims data.

While there has been a relative decrease in winter storm insured losses over the last 10 years and an elevation in hurricane activity, insurers and risk managers need to ensure they don't shift focus to the most severe peril at the expense of developing a comprehensive strategy to manage all risks impacting their portfolio.