Designing Index Insurance Contracts to Manage Hydrologically-Driven Financial Risks Using Machine Learning

Abstract:

Index insurance has been identified as an effective tool for managing the financial risks of hydrologic variability for water utilities. Furthermore, pooling these risks using the same index has shown to reduce index insurance prices relative to a scenario in which individualized indices are designed. However, one tradeoff associated with pooling risk across many utilities is increased basis risk, a measure of the degree to which the index (and therefore payouts) and financial losses are uncorrelated, as each water utility has a unique set of characteristics. The usefulness and demand for index-based contracts will increase as basis risk decreases. This research examines methods to identify improved multivariate indices that reduce basis risk through the use of various machine learning models. Data for the predictor variables are publicly available, provide insight into hydrologic conditions at various spatial and temporal scales, and provide a common approach for identifying an effective index across water utilities. Candidate variables for inclusion in the index are selected via parameter optimization and cross-validation, and all machine learning models are applied to a set of 325 water utilities that are geographically dispersed across the United States. Results will describe the reduction in basis risk from using generalized contracts to machine learning-based contracts for the set of water utilities. This work will also describe the tradeoffs in price, effectiveness, and interpretability of these different index insurance contracts designed to manage financial losses associated with hydrologic variability.