Climate-Conditioned Catastrophe Modeling for Dynamic Risk Assessment

This repository contains code and data necessary to reproduce the results presented in the article "Climate-Conditioned Catastrophe Modeling for Dynamic Risk Assessment" (preprint available here). The following files are needed for this: Data Publicly Available The DATA/ folder contains the publicly available input data: DATA/ILW-prices.csv: Prices quoted for different Industry Loss Warranties (ILWs). DATA/indexation-factors.csv: Loss indexation factors, used to adjust historical losses to present-day values based on inflation and estimated changes in exposure. This follows the method used in Comola et al., 2024 DATA/landfall.csv: Landfall region of historical storms. This file can be used to benchmark modeled landfall frequencies against historical data. DATA/Reask_climate-indices.parquet: Season-by-ensemble values of August-October means for various indices calculated from the gridded weather model output used to force the Reask Unified Tropical Cyclone (UTC) model. These are used to generate Figure 2b-d. DATA/Reask_gates.parquet: Landfall gates used in CBRA. DATA/Reask_smoothing-mapper.parquet: By default, nearby gates are smoothed to improve convergence in CBRA. This crosswalk file defines which gates should be smoothed to which other gates. DATA/CBRA_YLT/STD_1951-2020_[APRIL,JUNE]: These folders contain the seasonally adjusted loss tables for each seasonal forecast from 1985-2024, for both the April and June forecast initialization dates. They are produced by the Climate-Based Risk Adjuster (CBRA), using data from Reask_landfall-data.parquet and Verisk_STD_YLT.csv (see below). The adjusted loss tables provided in this data repository contain anonimized event IDs that do not correspond to those in Verisk_STD_YLT.csv. Restricted Data This data in the DATA/RESTRICTED/ folder is subject to commercial confidentiality agreements and cannot be publicly disclosed. Contact the corresponding author ([e-mail]) for additional information. DATA/RESTRICTED/PCS_regional-split.csv: The regional distribution of historical losses from the industry-standard source for claims data: Verisk's Property Claim Services (PCS). DATA/RESTRICTED/Reask_landfall-data.parquet: Long-term and seasonal landfall simulations: Reask's generative Unified Tropical Cyclone (UTC) model generates synthetic event sets forced by weather model output. For this analysis, we use an event set forced by the ERA5 reanalysis over 1951-2020 (used to define our "baseline" climate) and event sets forced by the European Center for Medium-range Weather Forecasting (ECMWF)'s 5-member SEAS5 ensemble forecast initialized in April and June of each year from 1985-2024. This represents a backend database for the Climate-Based Risk Adjuster (CBRA) tool used to adjust loss tables reflect forecasted risk in each season (see adjust-ylt.py). While this database is not publicly available, the anonymized outputs of the CBRA process, e.g. loss tables for each season, are included in the public data associated with this repository. DATA/RESTRICTED/Verisk_STD_regional-split.csv: The regional distribution of modeled losses in the static loss model used in this analysis. DATA/RESTRICTED/Verisk_STD_YLT.csv: The static Year-Event Loss Table (YELT) used in this analysis (AIR North Atlantic Hurricane Model, as implemented in TouchstoneRe v13). DATA/RESTRICTED/event-mapping.csv: Mapping file to match the event IDs in the adjusted YLTs with those in the original YLT. This mapping makes it possible to split the industry losses in the adjusted YLTs geographically in the same proportions as in the original YLT. Code CODE/environment.yml: A conda environment that will load the packages you need to execute adjust-ylt.py and investment-simulation.R. CODE/adjust-ylt.py: A python script that adjusts the long-term View of Risk (VoR), represented by the industry standard catastrophe model output in the form of a Year-Event Loss Table (YELT) (Verisk_STD_YLT.csv). This output is adjusted to season-specific YELTs for two different forecast initialization months. June (used for the main results of the paper) and April (shown in the supplementary information). CODE/investment-simulation.R: the R script that simulates the investment strategies based on the seasonally-adjusted risk model output and the long-term risk model output. The script reads in the output of adjust-ylt.py and other input data contained in the DATA/ folder, performes the investment calculations, and generates the plots shown in the manuscript (with the exception of Fig. 2a, c, e, which are generated from the CBRA package). Replication To replicate the findings in the paper, you will need access both to CBRA (if you wish to run the optional step 3 below) and to the restricted data (see above). CBRA is made available to Reask clients and for additional noncommercial research purposes on a case-by-case basis. To request access to CBRA, email [e-mail] with a description of your use case. The restricted data is subject to third-party commercial confidentiality agreements and may not be available. Obtain all data listed above. Install and activate the conda environment with the necessary packages a. First, put your Reask account credentials in a file at ~/.netrc. This will allow pip to install CBRA from the private pyPI server on which the package is hosted. If you do not wish to run the optional step 3, CBRA is not needed and you may remove the corresponding line from environment.yml and skip to step 1b. Once otained, your ~/.netrc file should look like this: machine pypi.reask.earth login your_username password your_password b. Next, create and activate the conda environment: conda env create -f environment.yml conda activate ilwpaper (optional) Run python adjust-ylt.py. To do this, you will need Reask_landfall-data.parquet. This step is not necessary, as the outputs for this step are already provided in the publicly available data under the DATA/CBRA_YLT/ folder. Run Rscript investment-simulation.R. This will simulate the investment strategies and generate all of the figures and tables presented in the manuscript. License See the LICENSE.txt file for the license associated with all source code provided within this repository. It does not apply to CBRA, which has its own license.