We tested news sentiment. It doesn't predict the market.

One of the most-sold ideas in retail trading is trading news sentiment: buy the positive, sell the negative. Sounds logical. We tested it for real.

We ran a classifier (FinBERT) on real-time catalyst news and let the paper trading run. The honest verdict: mean of −14.1 bps per trade, p-value 0.96. Indistinguishable from zero (and, with cost, negative).

Widening to the full sample: +5.7 bps/trade but p=0.07 — also not significant, and fragile. There's no exploitable signal at this horizon. The wall, as almost always at short horizons, is cost.

We publish this because honesty is the product. What works in our research is cross-sectional ranking at ≥1 month (lightly shorted), not news-based market timing. If someone sells you news signals with hype, ask for the p-value.