When industry experts gathered to consider the state of flood modelling, they discovered a disconnect between theory and reality that can leave insurers under some expensive water. So what’s next for modelling approaches?
Insurers are worried about their flood exposure. The huge floods that accompanied Hurricane Harvey last year have focused minds even further, but the simple fact is flood modelling is still an inexact science. Model outputs often fail to reflect the reality of indemnities as they treat weather events as wholly separate: there’s a windstorm, followed by a flood. Circumstances show this is not what occurs and that models and weather do not behave the same way.
This basic disconnect leaves insurers without an accurate guide to what might happen in the future. To help them better understand the evolving tools surrounding this complex, costly peril, Willis Re hosted an industry technical event – the Willis Re Flood Club – attended by catastrophe model users, developers, and risk advisors.
Three critical topics were discussed:
- The definition of a flood “event” and the poor alignment of scientific definitions and re/insurance policy terms.
- The inability of current flood models to manage multi-peril correlations.
- The uncertainty that comes when there are weak links in flood modelling.
Defining a flood event
Model users and developers have different approaches to defining flood events. The former seek insurance coverage controls, the latter an event identification algorithm based on physical occurrences. And in practice, hours clauses in re/insurance contracts dictate the loss, rather than any physically defined parameter.
Fellows of the Willis Research Network at Newcastle University outlined new research on a data-driven approach to event identification that’s less reliant on model assumptions and more reflective of the physical characteristics of each event.
Definitions are important, but so is exposure, which can determine loss frequency as well as severity. New flood modelling technology enables efficient, continuous hydrological event simulation that eliminates the need for users to define a flood event before running the model. It also allows users to apply their own definitions, but it’s still important for the insurance industry as a whole to agree on what constitutes a flood event.
Many weather events combine wind and flood damage, but expressing the relationship between the two isn’t easy. Better understanding is imperative and has become a high priority for many organizations including CatInsight, which shared its model simulations that look at links between flood losses and wind-induced losses in the U.K. Willis Re is coordinating a research project that extends this multi-peril correlation analysis and presented early findings at the WRN Spring Series seminar this month.
Catastrophe models are imperfect tools for decision-making. Users are trying to understand what their exposure is and developers are supposed to understand what non-modelled flood risks look like. There are attempts to unite the two, however.
OASIS showed which methodological choices can make catastrophe models so uncertain, and Lloyds Banking Group has suggested that uncertainty be categorized into four distinct areas from missing data and information all the way to climate, enabling us to understand the nature of the problem so it can be fixed. Again, Newcastle University has been active in this area, creating the CityCat model to improve the ability to define degrees of risk through more accurate modelling and quantification of uncertainty.
Keeping the good and discarding the bad
Flood modelling remains an imperfect science, but can still make an important contribution. Models do what they were designed to do: adeptly quantify risk differentials year-on-year, portfolio versus portfolio and region versus region. Even though Willis Re event shows that new ways of working hold out exciting possibilities, we need to remember that modelling works very well in certain contexts, avoiding a reliance on over-engineering.
By combining multiple models and cutting-edge science, along with powerful data and analytics capabilities, Willis Re helps insurers to create the best possible views of risk, whether for specific portfolios or their entire universe of risk. Through initiatives like Flood Club, we support clients by forging direct connections between experts and customers, and putting the latest advanced modelling knowledge into their hands.
Nalan Cabi is a flood specialist at Willis Re.
Bastian Manz is a catastrophe model developer at Willis Re.