Reproducibility package for "Explainable Multi-Horizon Stock Forecasting and Portfolio Allocation with an Attention-Enhanced Hybrid LSTM--SVR--ARIMA--GARCH Model and Multi-Source Sentiment"

This repository provides the public reproducibility package for the manuscript “Explainable Multi-Horizon Stock Forecasting and Portfolio Allocation with an Attention-Enhanced Hybrid LSTM--SVR--ARIMA--GARCH Model and Multi-Source Sentiment”. The package includes the exact walk-forward split definitions, model configuration files, schema-compatible sample datasets, results CSVs used to regenerate the reported manuscript tables, and the code framework for preprocessing, evaluation, backtesting, SHAP analysis, and table/figure regeneration. The empirical workflow covers nine large-cap US equities over 2020–2024, with forecast horizons of 1, 7, and 15 trading days under a five-fold expanding walk-forward protocol. The repository documents the hybrid model, evaluated baselines, direct horizon-specific forecasting setup, and the reported forecast, directional, statistical, portfolio, and explainability outputs. OHLCV market data are publicly obtainable via Yahoo Finance. Reuters, Bloomberg, and Twitter/X text data were used under licence or access restrictions and therefore cannot be redistributed publicly. For this reason, the repository does not redistribute raw licensed text corpora. Instead, it provides the public reproducibility structure, exact split definitions, model settings, sample schema-compatible data, and processed non-restricted outputs corresponding to the manuscript tables and figures. This Zenodo record is intended to support transparent methodological reproducibility and stable citation of the associated software and reproducibility materials.