Quarterly foreign institutional shareholding forecasting and large-change detection: data, code and benchmark results (India, 2023–2025)

This reproducibility archive contains the pinned public input snapshot, corrected calendar-quarter shareholding panel, source and duplicate audits, experimental code and Google Colab notebook, per-observation held-out predictions, statistical model-comparison tables, exploratory resampling checks, ablations, conditional diagnostics, and generated figures for research on forecasting quarterly changes in foreign institutional shareholding and ranking subsequent large ownership movements among Indian listed issuers. The evaluation uses 2023–2025 chronological holdouts, a 45-day disclosure cutoff, and a no-change forecasting baseline. The no-change forecast has the lowest overall mean absolute error across the evaluated models, whereas selected tree-based classifiers have higher average precision for identifying movements of at least one percentage point compared with the previous-change ranking reference. Data were reconstructed from a publicly released third-party exchange-filing compilation. Original exchange XML values and source-reported submission timestamps have not been independently verified, and missing future filings can affect the observed-label analysis. Results do not establish causal effects, trading profitability or fully authenticated point-in-time operational performance. Source snapshot SHA-256 and upstream repository commit are documented in the package.