LSTM-Based Detection of Structural Breaks in Property Insurance Loss Reserving: A Climate-Informed Approach

This white paper introduces a research program that evaluates the use of Long Short Term Memory (LSTM) neural networks for detecting structural breaks in property insurance loss reserving, particularly in the context of climate change. The study utilizes over 15 years of regulatory triangle data from Florida and Louisiana, enhanced with NOAA hurricane intensity indices and sea surface temperatures, aiming for a targeted improvement of 15-20% in reserve accuracy compared to traditional methods like Chain Ladder and Bornhuetter Ferguson. The findings provide a theoretical framework for LSTM's performance in this domain, offering formal guarantees that address the challenges posed by the limited number of catastrophic events during the testing period, which is crucial for practitioners seeking to enhance predictive accuracy in insurance reserving.