S-NODE-ANFRRC: JSE Equity Panel and Reproduction Package (2015–2026)
Domaine:
socioeconomicType de record:
Dataset and reproduction code for: "S-NODE-ANFRRC: A Continuous-Time Stochastic Neural ODE for Early-Warning Regime Detection in JSE Equities with Correction of a Systematic Distributional Artefact in GMM Baselines" (submitted to Expert Systems with Applications, 2025).
Contains: (1) jse_panel.csv — daily OHLCV for 17 JSE Top40 securities, 1 Jan 2015–30 Mar 2026, sourced from Yahoo Finance; (2) vix.csv — daily CBOE VIX, same period; (3) ANFRRC_corrected_analysis.py — full reproduction script, all models and statistical tests; (4) README.md with step-by-step instructions.
Key finding: log-transforming heavy-tailed financial features (realised volatility kurtosis = 51.6) is statistically required before GMM clustering. Failure to do so inflates ARI from 0.064 to 0.463 — a 7-fold artefact. On correctly transformed features, N-ODE-ANFRRC achieves ARI = 0.318, MCC = 0.438, BAC = 0.625. All experiments use seed = 42. Runtime ≈ 12 min on CPU.
Creative Commons Attribution 4.0 International (CC BY 4.0)