Drought to Deluge

For the past five years, the property and casualty insurance industry has been talking more and more animatedly about the flood peril in Canada, with a marked sharpening of focus since the Calgary and Toronto events of summer 2013.

Statistics relating to industry losses from these two unprecedented events, along with trends in the annual average water losses paid through insurance claims, are thrown around with heightened frequency and greater urgency.

While economic and industry estimates vary, the Calgary event generated more than $6 billion of economic losses while the insured losses were approximately $2.4 billion, largely due to overland flood exclusions and very low sub-limits on sewer back-up coverage for residential property policies.

Yet, in spite of all of this attention and the latest unfortunate examples of how material the losses associated with flooding can be, the industry has seen little tangible progress towards an actual solution.

THE COST EQUATION

In order to offer insurance coverage for any peril, a company or industry needs a sense of how often these losses take place, how large the associated loss for a given policy can be and what risk factors influence that likelihood or magnitude of loss. In addition, for natural catastrophe perils such as flood or earthquake, the insurance industry as a whole needs an understanding of the correlated nature of the risk, whether through physical proximity of exposures to one another or as a result of common susceptibility of exposures to the same set of external conditions.

These elements need to be taken into account, both at the individual insured location level, as well as at the aggregated portfolio level, to ensure that both profitability targets and solvency considerations will be satisfied over the long run.

On the surface, this should not pose a challenge to the industry for the flood peril. After all, the industry has addressed these needs for many other perils already through the use of sophisticated catastrophe models. These probabilistic models allow the Canadian p&c industry to evaluate, rate and offer peace of mind to clients and shareholders alike.

For Canada, tornado, hailstorm, thunderstorm, winter storm, hurricane and earthquake and fire-following earthquake are all perils that have already been addressed through the development and use of such models. Why then has no such model been developed in the past for Canadian flood?

The answer lies in the information and technology required to develop these models and allow them to produce results within a reasonable timeframe.

UNDERSTANDING THE HAZARD

The first requirement in developing a catastrophe model is an understanding of the hazard. This means a proper sense of the likelihood of an event of a given intensity.

For earthquake, the Canadian Geological Survey, founded in 1842, started taking seismographic readings in Toronto and in Victoria in 1897 and 1898, respectively. Since then, the Canadian National Seismograph Network has expanded to approximately 125 seismographs across the country and there are almost 200 such devices in Canada when academic and research institutions are included.

There are extensive scientific publications on ground motion attenuation through various soils, as well as soil maps and studies of the country’s fault lines. Combining these elements creates a strong understanding of the likelihood of various ground-shaking intensities in Canada.

For meteorological perils, such as severe thunderstorm, tornado, hailstorm and winter storm, a long history of rainfall measurements and wind speed and direction is available.

For hurricane, the National Oceanic and Atmospheric Administration tracks sea surface temperatures, climate cycles and the formation of storms. There is sufficient data and scientific knowledge to permit forecasts of whether or not the upcoming season is likely to be active, as well as forecasting whether any given storm forming over water is likely to intensify or dissipate and narrow down the potential paths that a given storm may take.

For flood, the hazard is not well-understood, particularly in a country as large as Canada.

Flood hazard is managed at a “local” level, often by municipalities. This local approach is taken because flood is highly subject to local topographic features and man-made defences, which change and shape the likelihood of any given location to be inundated.

However, management at the local level means that maps of the hazard may be of varying vintage from one community to the next, as well as of varying technology and resolution. In this way, progress both helps and hurts the industry’s overall understanding of the peril.

Unlike other perils, the risk of flooding may diminish with investment in defences, but increase with expansion of built-up areas into more flood-exposed places or beyond the capacity of the associated infrastructure, such as sewer systems. Ignoring the development of defences radically changes the likelihood of flood events arising, since, as a society, steps are taken to reduce the risk of events that can be foreseen.

Incorporating this factor represents one of the biggest obstacles to modelling the flood peril in Canada.

The next major challenge relates to the size of Canada, almost 10 million square kilometres. To compute hazard over such a large area can be a daunting, if not prohibitive, undertaking. And yet this has not been a major obstacle for other catastrophic perils.

The reason for this difference is a question of resolution. For earthquake, the attenuation of ground motion arising from an event is continuous. Typically the intensity of ground shaking decays smoothly away from the epicentre. Likewise, with wind, rain, hail or snow, the wind speeds and intensity of precipitation taper off at the edges of an event.

With flood however, there is an all-or-nothing aspect to the peril: if the water comes right up to the driveway and stops three feet from the front porch, there may be no damage to the house. If it comes a few feet closer and crosses the threshold, it could cause a flooded basement and a severely damaged ground floor.

The likelihood and degree of damage is, therefore, very sensitive to estimates of the location of the flood boundary. As a result, the objective is to use the finest resolution possible given the data and computational limitations. The desired resolution is in metres (or finer) not kilometres.

Even if the focus is solely on the populated area of Canada, which is just over 3% of its total land area, and aimed for 10-metre resolution, this would produce a grid of more than 3 billion cells over which to overlay calculations of likelihood and magnitude of loss.

Fortunately, graphics processing unit (GPU) computing technology has enabled companies specializing in modelling the flood peril to combine the considerable processing power of multiple GPUs together with multiple CPUs to churn through these endless calculations to determine the hazard on an order of square metres, rather than square kilometres.

The advent of this computing breakthrough has enabled undefended flood hazard maps to be generated at a fine resolution for the whole of Canada, using the historical rainfall record, river gauge data, hydrologic models and hydraulic models. This, in turn, has made it worthwhile for companies to put in the extraordinary effort required to assess the impact of flood mitigation infrastructure on the undefended hazard.

Some companies have taken the last required step to generate a stochastic event set to allow insurers to assess the impact of all potential flood events to understand not only the likelihood of flood, but the degree of damage following a given event.

This positions insurers to perform pricing by line of business, exposure type and risk zone, to undertake location-level underwriting, and to practice sound and proactive portfolio management.

NEW
REVENUE STREAM

In today’s market, a pre-condition for the industry to offer insurance protection for any catastrophic peril is the ability to conduct statistical analysis and to perform stochastic modelling. The lack of current coverage available to homeowners and the risk of extra-contractual payments being imposed has been unsatisfactory and frustrating for insureds and insurers alike.

With the recent breakthroughs in modelling, tools are now available that allow insurers to increase their knowledge of flood, which could lead to opportunities to generate a new revenue stream for residential property writers.

The Canadian p&c industry has discussed the risk of flood in Canada for years and is now increasingly in a position to formulate a solution.

With several insurers’ appetite to write flood insurance already growing, the coverage gap for homeowners is starting to shrink.