By Henry Gale

July 10 - (The Insurer) - Traditional catastrophe models for severe convective storm $(SCS)$ are struggling to predict future losses because they fail to incorporate data on smaller events, which sometimes produce more than half of carriers' losses, according to Demex.

Vendor models "have to rely on input data to train them, just like all models do", Demex CEO Bill Clark told Parametric Insurer. He said this typically comes from the National Oceanic and Atmospheric Administration's billion-dollar event database and Verisk's Property Claim Services (PCS) loss data.

"They only get data on events of a certain loss size, and they don't collect data for the smaller ones," Clark said, explaining that both datasets only include events that exceed a certain threshold for insured or economic losses.

"And they also don't get data from every carrier, nobody does. So when they're trying to come up with a model to use for underwriting reinsurance for these much more frequent events, they're a bit constrained by the fact that they don't have sufficient data to do a really high-quality job of being able to predict."

One of the top three largest U.S. carriers told Demex that 40% of all their SCS losses are from events that do not meet the thresholds to be reported by PCS or in NOAA's billion-dollar database, Clark said.

He also showed The Insurer the results of an analysis Demex had done of four U.S. carriers' SCS losses in Texas. The proportion of their losses that came from events not reported by PCS ranged between 14% and 62%.

"That starts to get to why reinsurance providers have been reluctant to underwrite" frequency SCS risks, Clark said. "They lost money on it because the models were not excellent at predicting future losses."

Matt Coleman, chief risk officer at the SCS-focused analytics firm, added: "Traditional catastrophe models have some skill predicting rare hurricanes, rare earthquakes, but they have less skill on the frequency side.

"In the very rare instances when reinsurers actually sell frequency loss covers, they often completely disregard the traditional catastrophe models because they have such low skill in predicting those aggregating losses. And instead they price on an experiential basis based upon the losses that they think that insurer has actually sustained."

Demex creates indexes from weather parameters calibrated to a cedant's historic loss experience, which it uses to offer a modelled-loss reinsurance product for severe convective storm risks.

Seven U.S. carriers purchased its reinsurance product at this year's January 1 renewals, first reported by Parametric Insurer last month, representing around $76 million in limit.