Distribution-Free Recalibration of Tail Quantile Forecasts under Temporal Dependence
Pele, D.T., Lessmann, S., Hardle, W.K. (2026)
| Quantlet | Description |
|---|---|
| CO_full_evaluation | Main pipeline: 9 models x 24 assets, Tables 1-8, Figures 1-5 |
| CO_baseline_comparison | Conformal vs 4 recalibration alternatives (Table 11) |
| CO_simulation_study | Monte Carlo: 5 DGPs x 500 reps (Section 5.7) |
| CO_coverage | Coverage recovery comparison (Figure 3) |
| CO_cross_sectional | Cross-sectional q_V vs asset characteristics |
| CO_cross_model | Cross-model threshold comparison (Figure 2) |
| CO_frontier | Coverage-efficiency frontier (Figure 5) |
| CO_garch_conformal | Parametric benchmark conformal correction |
| CO_heatmap | q_V heatmap across models and assets (Figure 6) |
| CO_multi_quantile_panel | Multi-quantile panel evaluation (Table 5) |
| CO_pipeline | Forecasting pipeline diagram |
| CO_quantile_scores | Quantile Score evaluation and DM tests (Table 7) |
| CO_raw_traffic_light | Basel traffic light matrix (Table 3) |
| CO_rolling_qV | Rolling q_V stability analysis (Figure 7) |
| CO_score_comparison | One-sided vs two-sided score comparison |
| CO_sharpness_penalty | Calibration-efficiency trade-off |
| CO_sign_diagnostic | q_V sign diagnostic heatmap |
All return series sourced from Yahoo Finance (24 assets, 2000-2026). Pinned TSFM checkpoints listed in Table 2 of the paper.
Python 3.10+, torch, chronos, timesfm, uni2ts, gluonts, arch, statsmodels, scipy, pandas, numpy
@article{pele2026conformal,
title={Distribution-Free Recalibration of Tail Quantile
Forecasts under Temporal Dependence},
author={Pele, Daniel Traian and Lessmann, Stefan
and H{\"a}rdle, Wolfgang Karl},
journal={Working Paper},
year={2026}
}MIT