Climate change Systemic risk Stress tests Valuation and portfolio optimisation

An asset-level analysis of financial tail risks under extreme weather events

How much can floods change a portfolio’s tail risk?

Ruben Kerkhofs, Mark Bernhofen, Marcin Borsuk, Moritz Baer, Nicola Ranger, Wim Schoutens and Gireesh Shrimali build a model of extreme weather losses in their paper “An asset-level analysis of financial tail risks under extreme weather events”.

They apply it to 424 coal, gas, oil, bioenergy and nuclear plants in India, using global flood hazard maps and river discharge data from 1979 to 2016 across 126 basins.

They fit copulas, then run Monte Carlo simulations of 25-year cash flows covering capital damage, business disruption and insurance premiums. Their main conclusions include:

  • An equally weighted portfolio loses about 0.7% of its value under every specification, from fully independent plants to every plant flooding at once.
  • The 99% value at risk is 1.01% under independence, 1.26% to 1.40% with calibrated copulas, and 3.86% when all plants flood together, three times the Gaussian copula estimate.
  • Assuming joint flooding inflates the 95% value at risk by 127% against a basin-level model in India, but by only 5% for a Thai portfolio clustered around Bangkok.
  • Direct damage to plants drives 83.2% of average losses and business disruption 16.8%, not taking into account failures of supply chains or critical infrastructure.
  • Insurance leaves averages unchanged, since premiums equal expected damage, but removing cover raises tail risk by 23% to 96%, and insurers leaving the market after three years adds a further 16% to 77%.
  • Under a Merton model, the 99th percentile rise in default probability grows from about 9 basis points under independence to 13 to 14 with copulas and 42 under joint flooding.

This paper introduces an open asset-level risk model offering an alternative to divergent scores and matching the EBA’s 2027 EU-wide stress test, which assumes a one-in-a-hundred-year river flood hitting every EEA country at once.

Most risks show up in tails rather than averages, especially insurance-related ones, so models without proper coverage in this dimension miss a material risk.

Supply chain and contagion effects are excluded, valuations only react once floods occur, and the vine copula proved unstable across 126 basins. Results are therefore a partial picture for one hazard and sector.