Forecasting Fashion Sales With Social Media Signals: Insights From Amazon, Twitter, and Google Trends

This study validates that fashion features identified from social media are actionable for demand planning. Building on our prior work establishing the Twitter Trend Tool (3Ts) for discovering trending apparel attributes without prior knowledge, we demonstrate that incorporating Twitter‐derived features into forecasting models improves out‐of‐sample accuracy. Unlike Google Trends, which requires knowing what features to search for, our approach extracts new, rising, and existing fashion trends in real time. We integrate weekly Amazon Sales Rank with weekly counts of Twitter posts and Google Trends indices for 77 matched fashion‐attribute time series across 21 popularity buckets. Using a four‐pass experimental design that applies identical model families (VECM, VAR, and NNETAR) across univariate and multivariate settings, we isolate the contribution of social signals from modeling methodology. Across 4620 forecasts, the combined Twitter‐plus‐Google‐Trends model achieves rank‐1 accuracy for all 77 time series, with VECM attaining a mean RMSE of 3.47 versus 91,034 for the univariate baseline (97.7% improvement). In head‐to‐head comparison, neither Twitter nor Google Trends dominates alone (approximately 51%–57% win rate); their complementary information content drives the dramatic forecast improvement when combined. We discuss managerial implications for planning cycles, feature discovery, and trend‐responsive assortment.