Log In

Reset Password

AIR: More Bermuda reinsurers hire staff to question cat models

Reinsurers were taken by surprise last year by a slew of catastrophes in the Asia-Pacific region that were outside the scope of the computer models they use to evaluate the risks they take on.Now catastrophe modellers like AIR Worldwide, who create the complex models used by many of Bermuda’s property and catastrophe reinsurers, are working on filling in the gaps from the lessons learned.Of particular concern to reinsurers was the Tohoku earthquake and tsunami in March last year, a quake of greater strength and with greater resulting losses than any model had projected for that part of Japan.The impact of the Thai floods, which clocked up $12 billion of insured losses, also unsettled the industry.This week Jens Mehlhorn, head of Swiss Re’s flood unit, authored a report on the floods, pointing out: “The insured losses corresponded to 1,800 percent of the country’s total annual property premium.“This emphasises the difficulties the industry faces in creating an economically viable approach to flood insurance.”Ming Lee, CEO of AIR, which creates computerised models of disasters used by many Bermuda reinsurers, said last year’s Pacific losses had caused reinsurers to adopt a more global view of where the biggest losses can occur.“The industry had been focused on the peak areas, like US hurricane and European wind,” Mr Lee said in an interview at the Rendez-Vous reinsurance industry get-together in Monte Carlo.“I think last year gave the industry a realisation that you can have major catastrophes in non-peak zones that can have major impact on their businesses.“One of the main issues that keeps coming up in our meetings here are these non-modelled losses and what to about them. These non-modelled areas are certainly driving the way we view our priorities for the future.”For example AIR has been rebuilding its Japanese earthquake model.Even before last year’s events, some in the industry were suggesting that there is too much reliance on models. Mr Lee said it was crucial that models’ limitations are understood by underwriters.“In a row of identical houses in a hurricane, some will get damaged much worse than others,” Mr Lee said.“Why? Because there’s randomness in nature, so there will be uncertainty in model results. So these uncertainties should be understood by those using the models.”AIR senior vice-president Bill Churney said Bermuda reinsurers were better than most at understanding the limitations of the models.“What we see in Bermuda is more and more companies hiring staff to question the models,” Mr Churney said. “We’re getting more inquiries and we see that a very positive thing.“Understanding the models better helps them to really ‘own the risk’.”The effectiveness of any model depends on the quality of the data and assumptions put into it. As development occurs and landscapes change, so do probable losses.AIR has exposure databases for 90 countries, which are updated at least every year, and which try to incorporate the changing nature of risks in different places.“This element is combined with data from policyholders to create an accurate a picture as possible of what a given an event would look like in a given area.”AIR has also benefited from the steady growth of catastrophe bonds this year.“There have been 20 ILS [insurance-linked securities] transactions in the property catastrophe space this year and AIR has been the modeller for all of them,” Mr Lee pointed out.“As these structures evolve, we believe there will be more and more sponsors seeking protection in this space and there certainly is an appetite from investors.”Mr Lee said catastrophe models now had a 25-year track “The model is like a currency for risk transfer,” Mr Lee said.“ “Like any currency, it’s important to have confidence in the output.“Paper money would be worth little without confidence that you’ll be able to buy something with it.“We place a high priority on the science, engineering, data and data analysis that go into making the model and validating it, to make sure the model makes an unbiased, best estimate of what the risk truly looks like.”