Skip to content
 
 

Repository files navigation

Conformal Recalibration of Extreme Tail Quantiles under Temporal Dependence

The current R8 replication package is in R8/. It includes the intermediate numerical results needed to regenerate the paper's tables and figures, the original producers, a locked Python environment specification and a command that verifies the regenerated files against the deposited outputs.

conda create -y --prefix /tmp/r8-env --file R8/environment-osx-arm64.explicit.txt
/tmp/r8-env/bin/python -m pip install --no-deps -r R8/requirements-conda-overlay.txt
/tmp/r8-env/bin/python R8/reproduce.py replay --workdir /tmp/r8-replay --report /tmp/r8-replay.json

These exact binary builds target macOS on Apple Silicon, including FreeType 2.13.3. Package version pins alone do not fix the graphics renderer; details are in the R8 instructions. Choose a new work directory for each replay. The command starts without any pre-existing output tables or figures, runs all eight display producers, and compares all 67 generated files byte for byte. It also runs the statistical unit tests, the exact rational counterexample, and corruption controls.

The reproducible display release is tag R8-2026-09-13-repair1, which extends R8-2026-09-12-repro1 with the three studies of 13 September 2026 (see R8/README.md). The original R8-2026-09-12 tag is preserved as a historical deposit. The replay uses archived intermediate results; it does not retrain foundation models or rerun the full simulation and forecasting pipeline. Full native draws and fitted-path archives remain a separate research archive.

Older Quantlets and the data-v1 download belong to earlier revisions. They are retained for historical work and are not inputs to the R8 command above. The previous project description is archived in R8/history/.

Authors

Daniel Traian Pele, Vlad Bolovăneanu, Andrei Theodor Ginavar, Stefan Lessmann and Wolfgang Karl Härdle.

License

Source code is distributed under the repository's MIT license. Financial data originate from the providers identified in the manuscript and archived metadata.

About

Conformal VaR recalibration for time series foundation models and classical benchmarks — 10 models × 24 assets × 2000–2026

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages