Simulation Parameters
Test Set Metrics
Visual Risk Horizon Canvas
Methodological Contrast & Insights
Vanilla Conformal Prediction
Vanilla conformal frameworks assume **exchangeability** of historical calibration samples and future observations. Because of this, the interval boundary is computed as a static threshold calibrated globally from the training split.
\([ \hat{\mu} - q, \hat{\mu} + q ]\)
Key Vulnerability: When data exhibits drift or structural variance modifications (nonstationarity), VCP's static window will fail, resulting in severe **undercoverage** during high-volatility regimes.
Nonstationary Z-Score Framework
Based on **Marè, Moreira, & Rossi (SSRN 2688367)**, this approach models nonstationarity explicitly. Instead of assuming identical distributions, it computes local statistical moments over localized rolling memory frames to scale boundaries parameter-by-parameter.
\([ \mu_t - t_{crit} \cdot s_t \sqrt{1 + 1/W}, \mu_t + t_{crit} \cdot s_t \sqrt{1 + 1/W} ]\)
Benefit: Dynamically adapts to local parameter drifts. By integrating Student's \(t\) degrees-of-freedom scaling and rolling finite estimation adjustments, it preserves target coverage rates across volatile timelines.