Using open-source and freely available data science tools, we took a historical book of catastrophe-exposed commercial property insurance and assessed the segments of the market in terms of risk and profitability. In close collaboration with both the underwriting and actuarial teams, we help them set future strategic goals for that line of business.
Situation
Rate pressure in the global commercial property market was a concern for a Lloyd’s Managing General Agent (MGA) with a large exposure to these risks. With an aim to improve their risk selection and mitigation procedures, they wanted to assess their historical book with a view to altering the mix of the business over the next few renewal cycles.
Solution
An extensive amount of data engineering, cleansing and exploration work was required in the initial stages of the project, combining additional data sources to enrich the data. Locations covered by each policy were available, but claims were linked to policies. Policy terms such as attachment points and limits were accounted for within the model.
Business Outcomes
Visual exploration of the policy data and concentrations of risk
A number of simple dashboard-style slice-and-dice tools were built to assist future underwriting
Identification of loss making segments
Strategic underwriting objectives for renewal identified
Project Outputs
Reproducible research methods allows for full auditability of the data engineering and exploration work critical to the project
Efficient data processing software avoided need for ‘big data’ infrastructure to perform analysis
Location less important than expected for attritional losses
Implications
Rate pressure had significantly eaten into expected profit of the book
Number of market segments below attritional loss-cost
Total premium written could be reduced without affecting profitability
Key Benefits for Client
Reproducible research methods understood within the business
Improved audibility of work eased its take-up when revisited a year later
Enabled strategic decision-making based on quantitative reasoning rather than qualitative judgements
