Digital investor sentiment and gold ETF demand: introducing an audience reaction index with explainable machine learning

Purpose This study examines whether digitally expressed investor sentiment, reflected in YouTube comments, explains daily Gold Exchange Traded Fund (ETF) demand in India beyond conventional macro-financial sentiment indices. A novel Audience Reaction Index (ARI) is developed and evaluated using machine learning and explainable artificial intelligence (XAI). Design/methodology/approach Using 25,666 YouTube comments from 14 leading Indian financial influencer channels (from April 2019– to March 2025), the study constructs the ARI by integrating sentiment polarity and audience engagement. The ARI is analysed alongside macro-financial sentiment indicators within an explainable machine-learning framework. Robustness is assessed through separate sentiment and engagement models and Granger causality tests. Findings Gold ETF demand exhibits strong persistence, with lagged sales emerging as the strongest predictor. Economic policy uncertainty increases Gold ETF demand, reinforcing gold's safe-haven role. The ARI provides incremental predictive information beyond conventional sentiment measures, while robustness analyses support its predictive relevance. SHAP and permutation importance analyses confirm that digital investor sentiment captures unique behavioural drivers of ETF demand. Originality/value The study introduces a multidimensional, engagement-weighted sentiment index derived from YouTube discourse and demonstrates the usefulness of explainable machine learning in identifying behavioural drivers of commodity-specific ETF demand in an emerging market.