Development and Evaluation of an Explainable Hybrid Decision-Support System for Indian IT and AI Stock Forecasting

Replication package for: “Development and Evaluation of an Explainable Hybrid Decision-Support System for Indian IT and AI Stock Forecasting” (manuscript MA18938m3, submitted to Investment Management and Financial Innovations). This package contains the code, data and output files needed to reproduce the results in the manuscript and its supplementary material. Contents code/: Python scripts for data collection, forecasting models, evaluation, model confidence set, stacking validation, explainability (TreeSHAP and permutation importance), extra analyses, rolling evaluation, and the panel, backtest and COVID-19 analyses. data/: the daily market and panel data used in the analysis (sample period 2019-09-11 to 2026-03-20), conditional volatility series, and model settings. forecasts_and_losses/: the out-of-sample forecast and loss series for all models. results/: the output tables and figures’ underlying numbers. requirements.txt: exact package versions (Python 3.13). README.md: the order in which to run the scripts. Methods covered: eight forecasting models (ARIMA, Holt–Winters, random forest, gradient boosting, MLP, LSTM, Transformer, and Ridge stacking), evaluated out of sample against zero-return and historical-mean benchmarks using Diebold–Mariano, Clark–West and Giacomini–White tests and the model confidence set. Licence: code under the MIT License; data and results under CC BY 4.0. JEL: G11; G12; G17; C53; C58