Paper: Distribution-Free Recalibration of Tail Quantile Forecasts under Temporal Dependence
git clone https://github.com/QuantLet/Conformal_Oracle.git
cd Conformal_Oracle
pip install numpy pandas matplotlib scipy openpyxl
python run_conformal_analysis.py # LLM conformal analysis
python run_conformal_garch_ewma.py # GARCH/EWMA conformal analysis
python run_simulation_study.py # Monte Carlo simulation (5 DGPs)Conformal_Oracle/
├── README.md
├── run_conformal_analysis.py # Main LLM conformal analysis
├── run_conformal_garch_ewma.py # GARCH/EWMA conformal analysis
├── run_simulation_study.py # Monte Carlo simulation (5 DGPs)
├── run_all_three_tasks.py # Extended sims + quantile scores
├── CO_conformal_agnostic/ # Cross-method q_V bar chart
├── CO_universality/ # Universality figure (q_V vs pi)
├── CO_simulation_study/ # Monte Carlo simulation
├── CO_calibration_plot/ # Calibration plot
├── CO_garch_conformal/ # GARCH/EWMA conformal correction
├── CO_quantile_scores/ # Quantile score evaluation
├── CO_sharpness_penalty/ # Calibration-efficiency table
├── CO_conformal_garch_ewma/ # Notebook: GARCH/EWMA
├── CO_conformal_llm/ # Notebook: LLM analysis
├── CO_dual_correction/ # VaR correction time series
├── CO_coverage/ # Coverage comparison
├── CO_cross_model/ # Cross-model thresholds
├── CO_GARCH_comparison/ # GARCH benchmark
├── CO_frontier/ # Coverage-efficiency frontier
├── CO_freq_magnitude/ # q_V diagnostic
├── CO_heatmap/ # q_V heatmap
├── CO_rolling_qV/ # Rolling q_V stability
├── CO_traffic_light/ # Traffic light matrices
└── CO_*/ # Other QuantLet figures
| Method | Mean q_V | SD(q_V) | Green/9 |
|---|---|---|---|
| GARCH-N(250) | +0.004 | 0.009 | 7/9 |
| GAS-t(250) | -0.007 | 0.008 | 6/9 |
| EWMA-DCS(120) | +0.002 | 0.004 | 9/9 |
| GPT-3.5+CP | +0.002 | 0.007 | 8/9 |
| GPT-4+CP | +0.024 | 0.015 | 9/9 |
| GPT-4o+CP | +0.020 | 0.014 | 9/9 |
| DGP | T=1000 q_V | T=5000 q_V | T=5000 Green% |
|---|---|---|---|
| Normal | 0.0001 | 0.0000 | 96.2% |
| Student-t(5) | 0.0014 | 0.0013 | 98.2% |
| Student-t(3) | 0.0016 | 0.0013 | 97.2% |
| Stoch. Vol. | 0.0009 | 0.0006 | 93.2% |
| Regime Switch | 0.0122 | 0.0114 | 95.8% |
@article{pele2026conformal,
author = {Pele, Daniel Traian},
title = {Distribution-Free Recalibration of Tail Quantile
Forecasts under Temporal Dependence},
year = {2026},
url = {https://github.com/QuantLet/Conformal_Oracle}
}MIT