Geography modelling plays a vital role in property and casualty (P&C) insurance pricing, especially in motor and property insurance. By incorporating spatial patterns into risk and demand models, insurers can ensure premiums align with varying risk levels and behaviours across different geographic areas. This process leads to more accurate and fair pricing. Dynamic pricing software specialists Quantee delves into the most common approaches to geography modelling together with their pros and cons.
Considered as a fundamental practice in the field, the principle of incorporating geography information and applying geographic smoothing in the pricing process is considered essential to actuaries.
But there are still cautionary tails for actuaries looking to implement the software.
For instance, they should avoid overloading models with redundant information. For instance, using both REGION and POSTCODE data in motor coverage risk modelling can lead to overfitting, as these data types often overlap.
Instead, the focus should be on granular, fair data sources that provide unique insights, such as third-party data about social information or natural catastrophe (NatCat) risks. However, care must be taken to prevent unintended discrimination, particularly when using geo-social factors.
Available methods for geography modelling
Two leading approaches within the insurance pricing community are 2D thin-plate splines and credibility-weighted residuals smoothing.
Both methods enhance the precision of Generalised Linear Models (GLM) and Generalised Additive Models (GAM) by incorporating geographic data into modelling.
2D thin-plate splines
This method assigns latitude and longitude coordinates to regional variables and applies an interpolation process over a two-dimensional surface.
The algorithm balances goodness of fit and smoothness, ensuring the model captures non-linear relationships effectively without overfitting.
Advantages of thin-plate splines include their reliance solely on geo-coordinates, ease of implementation during GLM/GAM fitting, and capacity to handle complex relationships between geographic and target variables.
However, challenges arise in isolated regions like islands and when working with large datasets, as only advanced software can implement this method effectively.
Credibility-weighted residuals smoothing
This technique blends spatial smoothing with actuarial credibility theory. The process involves building a GLM/GAM model without geographical variables, calculating residuals, smoothing these residuals based on distances or adjacency, and clustering the smoothed results to create a regional mapping. This mapping is then incorporated into the model.
Advantages of credibility-weighted smoothing include its effectiveness in borders and isolated regions and its relative simplicity compared to thin-plate splines.
However, it requires extensive manual work, careful parameter tuning, and additional steps for extrapolating postcodes not included in the training dataset.
Making the decision
Both 2D thin-plate splines and credibility-weighted residuals smoothing offer unique strengths and challenges.
The choice between these methods depends on specific use cases, with considerations such as implementation complexity, data characteristics, and modelling goals influencing the decision.
Read the full blog from Quantee here.
Copyright © 2025 InsurTech Analyst
