QM015 Hands-on Lab
Multivariate Probabilistic Forecast Evaluation
Objective
Compare two synthetic two-asset predictive clouds with Energy Score, Variogram Score, marginal CRPS, PIT rank, and interval coverage.
Run
python python/qm015_probabilistic_evaluation.pyCheck
The script prints one line per model. The point is not that one model must dominate every metric; inspect which diagnostics disagree and explain what dimension each one is measuring.
Experiment
Change the off-diagonal covariance in Model B while keeping its marginal variances fixed. Observe how the Variogram Score reacts relative to the componentwise CRPS.
The data are synthetic and carry no empirical performance claim.