Drought parametric insurances by a Two-Step machine learning approach under climate change scenarios

Speaker: Dr. Hirbod Assa, Founding team | Quantitative researcher, Edge Technologies; Director and founder, Model Library Ltd.

Abstract: In this paper, we utilize data from the IPCC to develop a predictive model for the Palmer Drought Severity Index (PDSI). Our approach involves a two-step modeling process: initially applying a random forest regression, followed by a linear regression correction, achieving a forecasting accuracy exceeding 94%. This method aims to predict the drought index using a minimal set of climate indices, with projections extending to the year 2100 based on CORDEX CMIP5 models. The analysis focuses on selected U.S. states, assessing the impacts of different climate scenarios and climate change under various RCP pathways. On that basis we design a parametric insurance on drought index. If time permits, we will also explore the implications of these drought projections on supply chain risks in the beef commodity market.