Håkon Otneim1, Etienne Dunn-Sigouin2,3, Sondre Hølleland1, Mahsa Gorji1, and Geir Drage Berentsen1.
1: Norwegian School of Economics, Bergen, Norway.
2: NORCE Norwegian Research Center AS, Bergen, Norway.
3: Bjerknes Centre for Climate Research, Bergen, Norway.
Insurers are increasingly challenged by weather-related claims arising from property damage, yet they lack adequate tools for near-term planning because traditional actuarial models do not incorporate real-time weather forecasts. This study demonstrates that incorporating ensemble precipitation forecasts improves short-range (1-4 days ahead) predictions of property insurance claim counts, thereby enabling proactive risk management. We present a forecasting framework for two Norwegian cities, Bergen and Oslo, using precipitation forecasts to predict days exceeding operationally significant thresholds. The models are evaluated by their forecast skill and reliability in predicting claim surges, as well as by their economic value in a cost-loss decision context, illustrating the potential reduction in expected costs when early warning triggers are in place. Results show that weather-informed models substantially outperform baseline models based on climatology, improving discrimination of claim events and yielding up to 30-50% reduction in expected daily costs under various ideal warning scenarios. Two case studies of extreme events highlight how weather forecasts translated into early claims warnings could guide resource allocation and customer advisories. Overall, the presented framework highlights the practical benefit of integrating meteorological forecast information into insurance operations and offers a template for insurers to enhance climate resilience through improved risk communication and short-term decision support.
This repository contains necessary code for reproducing results presented in the paper by Otneim et. al (2026). Since we cannot share the insurance data, you will not be able to reproduce the specific results in the paper, but we have made a toy data set so that the code can be run with similar similar properties as one would get using the real data.
The script 0_run_scripts.R explains the order of how things should be run. The resulting predictions are stored the predictions folder and figures are stored in the figures folder. Figures that are very specific to the insurance data is not included here (i.e. for the two named storms Birk and Hans) or the descriptive figures for the insurance data. Some scripts are executed in Pyhon, and located in the folder called python.
The 51 ensemble forecasts and observed rain are stored in data/forecast_data.rds. The column obs is observed rain from the ERA5-reanalysis aggregated over the two cities (as described in the paper), while the columns '1'-'51' are the ordered ensemble forecasts from European Centre for Medium Range Weather Forecasts (Buizza et al., 2018) obtained from the MARS archive (European Centre for Medium-Range Weather Forecasts (ECMWF), 2024), also aggregated over the same area.
The insurance data is not publicly available, but a toy data set is generated by running the script R/0_making_toydata.R. This uses the estimated logistic regression model for Oslo and Bergen to generate a toy dataset.
- Buizza, R., Alonso Balmaseda, M., Brown, A., English, S. J., Forbes, R., Geer, A., Haiden, T., Leutbecher, M., Magnusson, L., Rodwell, M., Sleigh, M., Stockdale, T., Vitart, F., and Wedi, N.: The Development and Evaluation Process Followed at ECMWF to Upgrade the Integrated Forecasting System (IFS), ECMWF Technical Memorandum 829, European Centre for Medium-Range Weather Forecasts (ECMWF), https://doi.org/10.21957/xzopnhty9, 2018.
- European Centre for Medium-Range Weather Forecasts (ECMWF): Access to forecasts, https://www.ecmwf.int/en/forecasts/accessing-forecasts, accessed: 2024-09-05, 2024.
- Hersbach, H., Bell, B., Berrisford, P., Biavati, G., Horányi, A., Muñoz Sabater, J., Nicolas, J., Peubey, C., Radu, R., Rozum, I., Schepers, D., Simmons, A., Soci, C., Dee, D., and Thépaut, J.-N.: ERA5 hourly data on single levels from 1940 to present, https://doi.org/10.24381/cds.adbb2d47, [data set], 2023.
- Otneim, H., Dunn-Sigouin, E., Hølleland, S., Gorji, M., and Berentsen, G. D.: Short-Term Management of Water-Damage Claim Risk Using Ensemble Precipitation Forecasts, working paper, 2026.