Model Flood Risk Without Historical Data – An Innovative Solution!

(Carl Lambert, Vice-President of Business Intelligence at The Co-operators)

In June 2013, Canada suffered one of its most severe floods in recorded history. 32 towns in southern Alberta were flooded for total damages exceeding C$5 billion. At the time, the insurance industry did not offer flood insurance.  Sewer backup losses were covered and the total cost for the industry was C$1.7 billion. Yet, in early 2015, Canada remained the only G7 country where residential flood insurance coverage was not available.

At Co-operators, we were already working on launching a flood product.  Those events reinforced the demand for residential flood coverage and put more pressure on the industry to develop a solution.

The Co-operators was the first in Canada to launch such a new coverage. Significant effort was required across the organization to ensure we implemented the proper solution that would answer an unmet need, while focusing on making Canadian communities more resilient to flooding. This blog will focus on only one piece of the work, the development of the risk assessment and the pricing. The BI-Research team and Actuarial pricing team collaborated on a non-traditional pricing solution. 

The Approach - Research

Learn & Partner with Canadian Universities

We started by reaching out to our network of partners in Canadian universities. This helped us better understand important concepts around flood hazards, flood plains, and damage functions. Our Statisticians and Actuaries have learned to work with Hydrologists, Geologists, Hydraulic Engineers and Civil Engineers.

Seek out Third Party Vendors

We then started a long process of seeking out and assessing existing flood models and data sources. We learned to speak with modeling firms, and gradually built enough expertise internally to be able to assess the credibility and value of third party vendors.

Leverage Open data & Big Data

We then sought out external available data.  There is a lot of information available and the challenge was to identify the ones that are usable for that purpose.  By usability of information, we mean reliability, predictability and frequency of updates.  

We have tested dozens of external sources and a significant number of them have been used.  For example, we used elevation data at every 5 meters Canada Wide.  (30 meters in rural areas).  We also used the Soil type across Canada to better model water dispersion and evaluate how long the flood will last.  We also used Historical River flows, with numerous lecture points of all rivers in Canada, available every minute, for at least 50 years.  We even used a database showing historical Tectonic Plates movements.

Assessment of the Risk

There were three sources of flood risk to model: Fluvial flood, pluvial flood and coastal flood. Each of them has their own specificities and therefore have different models.

It was important for us to provide adequate and flexible coverage for all Canadians, whether they are in a high risk zone or not, at a price that accurately reflects the true risk. For that reason, we needed a model that was accurate, precise, and consistent.

Our model is customized to use different sources of insight that complement each other. Vendor models will sometimes fail our quality standards and, most of them also ignore a significant amount of local flood defense structures such as dikes and reservoirs. On the other side, our internal models were not always based on enough data to be fully credible. 

With extensive R&D efforts, we were able to leverage the large amount of data available in Canada to bridge that gap and create a national flood risk model that meets our standards.

Assessing the risk means developing models for the following 3 phenomena:

Model Flood Water amounts

Hydrological models are used to determine the probability that a water body will flood. 

Model Water Dispersion

Hydraulic models are used to determine how those water volumes flood the landscape. 

Model the « submersion depth »

Submersion depth models use the results of water dispersions and combine them with other sources of information to determine models for submersion. Furthermore, the required use of rooftop geocoding of the exact location of the insured building complicated the availability of information since many possible sources did not have the geocode.

Convert the « submersion depth » into building & content damage

Many factors impact the amount of damages:  the submersion depth, the type of building, the expected duration of the flood, the temperature of the water and many more. In order to build both content and building predictive models, we have used text mining on notes coming from past sewer backups claims and integrated that with external probabilistic models. 

Pricing Policies

Flood models estimate the flood risk but they don’t calculate an insurance premium. For example, third party damage curves work well at estimating flood damage but cannot be directly applied to insurance claims, because the latter includes elements of client behavior as well as the effect of limits and deductibles. Furthermore, our comprehensive water insurance product offers our clients unprecedented flexibility regarding their water coverage, which also provided pricing challenges. In times when the “buzz word” in technology and science is Minimal Viable Product, in the case of flood insurance the bar for a viable product is very high.

In the end, a key to our success is to take a scientific approach to modeling the flood risk, for each and every house, farm, and building. It is what allows us to provide insurance at the right price, for everyone. It is that analytical mindset, combined with a lot of determination and innovation, that is and will continue to be the Co-operators’ advantage. 

This blog post has been written by Carl Lambert, who is vice-president of Business Intelligence at The Co-operators. Carl completed a Master's degree in Actuarial in 1994. He joined The Co-operators in 2009, where he launched a Research team that now consists of over 65 professionals in Mathematics, Statistics, IT and Actuarial. The team is responsible for the development of Analytics throughout the organization.

Carl Lambert is a panelist at CatIQ’s Canadian Catastrophe Conference (C4 2017) on the How to Create an Inventory of Canadian Hazard Data session during the conference.