pcmci

Experimental causal-inference research on financial regimes: PCMCI+, ICP and causal forests over market data (work in progress)

General-purpose Python toolkit for horizon-wise forecastability analysis using interchangeable dependence scorers, with AMI/pAMI support, rolling-origin benchmarking, and reproducible reporting.

Causal discovery (PCMCI) applied to ERA5 winter fields to test whether they predict the September Arctic sea ice minimum beyond trend and persistence. Causally selected predictors beat using all features, but no model beats a naive trend+persistence baseline consistent with the spring predictability barrier.