Causal Bandits Podcast

Strait of Hormuz: Causal Models for Rare Events | Alexander Denev S2E11 | CausalBanditsPodcast.com

Jun 1, 2026
Guest
Alexander Denev
Alexander Denev, CEO of Turnleaf Analytics and author on Bayesian networks, explores using causal models to reason about one-off shocks. He explains why correlation-based risk breaks down and how Bayesian networks can hold through time. The conversation covers using LLMs as scalable expert surrogates, historical analogies like the Bosphorus, and a three-way robustness test for scenario models.
Ask episode
AI Snips
Chapters
Books
Transcript
Episode notes
INSIGHT

Causal Structure Remains Stable For Unique Shocks

  • Unique shocks lack historical repetition so statistical models fail; causal Bayesian networks use stable transmission mechanisms to reason about one-off events.
  • Alexander Denev reused a 2012 scenario structure because propagation channels stayed stable, letting causal structure 'hold in time'.
ADVICE

Use LLMs To Rapidly Construct Bayesian Scenarios

  • Use LLMs to accelerate scenario building by providing methodology context and public-source evidence, reducing expert whiteboarding from weeks to minutes.
  • Denev fed his books and prompted Claude iteratively, then performed sensitivity checks and reprompting to refine outputs.
ADVICE

Triple Check LLM Outputs To Prevent Hallucinations

  • Never accept an LLM-built network at face value; perform multiple reprompts and cross-checks to avoid hallucinations.
  • Denev recommends at least ten reprompts, sensitivity analysis and triple-checking assumptions before publishing or using institutional outputs.
Get the Snipd Podcast app to discover more snips from this episode
Get the app