Financial Metabolomics Series (IST-04): Gaussian-Weighted Swin Networks with Metabolic Saliency for systemic stress detection in JSE equity markets. PyTorch implementation + reproducible figures.

IST-04 Financial Metabolomics Series

Zenodo DOI: 10.5281/zenodo.19339552

Series: IST-04 Financial Metabolomics
Institution: North-West University, Faculty of Economic and Management Sciences
Author: Prof. Ntebogang Dinah Moroke · ORCID: 0000-0001-8545-1860

Papers in this Series

Paper	Framework	Journal	Status
Paper 1 — Gaussian-Weighted Swin Networks	GWS-STNet	Mathematics (MDPI)	Under review
Paper 2 — Financial Metabolomics & KL-Divergence	MetabolicNet	Entropy (MDPI)	Under review

Data

All data in this repository is from publicly available sources. Raw JSE individual security returns are proprietary and available under licence from the Johannesburg Stock Exchange (www.jse.co.za).

File	Description	Source	Period
shredi_combined_clean.csv	JSE ALSI, R186 bond, Gold, Platinum, CRB, VIX, Brent, Eskom stage, regime label	Yahoo Finance + SARB + Eskom	Jan 2015 – Apr 2026
yahoo_returns.csv	JSE ALSI index + global commodity returns	Yahoo Finance	Sep 2017 – Apr 2026
r186_proxy.csv	South African 10-year government bond yield	SARB (public)	Jan 2015 – Apr 2026

Variable Descriptions

Variable	Description	Unit
JSE_ALSI	JSE All Share Index log return	Daily %
R186_Bond	R186 bond yield change	%
Gold	Gold spot price log return	Daily %
Platinum	Platinum spot price log return	Daily %
CRB	CRB Commodity Index level	Index
VIX	CBOE Volatility Index	Index
Brent	Brent crude oil log return	Daily %
eskom_stage	Eskom load-shedding stage (0–6)	Integer
regime	Market regime label (resilient/stressed)	Categorical

Code

File	Description
IST04_descriptive_stats.py	Descriptive statistics + summary figure
reproduce_figures.py	Reproduces all paper figures
requirements.txt	Python dependencies

Pre-computed Outputs

File	Description
ist04_descriptive_stats.csv	Panel summary statistics
ist04_eskom_summary.csv	Eskom stage distribution
IST04_Figure_DataSummary.png	Summary figure

How to Run

pip install -r requirements.txt
python IST04_descriptive_stats.py    # descriptive stats
python reproduce_figures.py          # all figures

Data Availability Note

The GWS-STNet and Financial Metabolomics models were trained on a panel of 15 JSE blue-chip securities (2015–2026). This data is proprietary to the JSE. Researchers requiring access should contact the JSE directly (jse.co.za). All model code and public proxy data are provided in this repository.

Licence

CC-BY 4.0

Contact

ntebo.moroke@nwu.ac.za · ORCID: 0000-0001-8545-1860

Version 2 adds public market proxy data (JSE ALSI, R186 bond yield, commodities, Eskom load-shedding stages) and replication scripts. Raw JSE individual security data is proprietary — cite DOI 10.5281/zenodo.20008530 for the JSE panel dataset.