The thesis
News-sentiment is about as legible as a strategy gets: buy names whose recent news sentiment turns positive, exit when it turns negative. It runs over a 24-name mega-cap universe spanning tech (AAPL, MSFT, GOOGL, NVDA), financials (JPM, BAC, V, MA), healthcare, staples and energy — liquid, heavily-covered stocks where a sentiment feed actually has material to chew on. The logic is intuitive and the universe is sensible. The question, as always, is whether the signal fires often enough and cleanly enough to matter.
Recent activity
On the live paper book, the strategy currently sits around $10,323 total with $990.06 in cash — most capital is deployed. The positions came from a cluster of buys between late May and early July 2026: AAPL, MSFT, BAC, UNH and GOOGL. Since then the scheduled runs have gone quiet. The last week of runs (Aug 11–18) executed zero trades, with a single rejection on Aug 18. In other words, the model took its shots early, then went dormant — consistent with a signal that only speaks when the news flow is decisively one-sided.
Backtest and validation
This is where enthusiasm should cool. Over 451 days the backtest returned just +0.19% (CAGR ~0.1%), on 2 trades, with a Sharpe of 0.74 and a shallow 0.05% max drawdown. A win rate of 0 alongside a positive total return is a red flag that the sample is simply too small to read.
The walk-forward tells the real story. Of four folds, three produced zero trades and zero return — the signal did not fire at all from Aug 2024 through Dec 2025. Only fold 4 (Dec 2025–May 2026) traded, and it alone delivered the entire +0.19% at a flattering Sharpe of 1.5. Every headline out-of-sample number is that one fold wearing a disguise. The probabilistic Sharpe ratio looks healthy at 0.923, but the deflated Sharpe — which penalizes the 6 trials run — drops to 0.551, and the validation gate correctly returned passed: false.
Strengths and risks
Strengths: the concept is sound, drawdowns are tiny, turnover is modest (~36%), and costs are negligible ($2 in fees). Nothing here blew up.
Risks: the evidence base is dangerously thin. A two-trade, single-fold track record is indistinguishable from luck, and the strategy's own validation says as much. That it is running live despite failing that gate is the headline concern — it is deployed on hope, not proof. Before trusting this signal, we'd want to see it fire across multiple regimes and accumulate a trade count large enough to make a win rate meaningful. For now, treat news-sentiment as an interesting hypothesis on probation, not a validated edge.