Social media sentiment and stock return forecasting in customer-supplier links

Traditional models capture little of the variation attributable to investor sentiment, limiting the accuracy of short-term stock return forecasting. Using Apple–TSMC as the starting relationship, we examine 13 verified customer–supplier links and ask whether each firm’s own StockTwits discussion, read from the Bullish or Bearish stance tags that message authors attach, contributes differently to downstream and upstream firms’ one-day-ahead return-direction forecasts. Our primary paired estimate averages the within-link downstream-minus-upstream improvements equally across the 13 links. Against a financial-plus-history benchmark, this link-equal log-loss gap is 0.0085 (Newey–West p = 0.0065 ; Benjamini–Hochberg q = 0.0129 ). The corresponding classification-accuracy gap is about 1.4 percentage points. A repeated-firm sensitivity check that counts each distinct firm once gives a positive gap of 0.0076 ( p = 0.0145 ). The paired gap is largest in 2024, remains smaller but statistically significant in 2025, and is positive but imprecisely estimated in January–April 2026. Thus, the forecast contribution of investor discussion differs across the two roles in the sampled links, and it follows a position read from the disclosure rather than a firm type assigned by the researcher. This observed difference does not by itself establish supply-chain role as the causal mechanism.