The thesis
The idea behind news-sentiment is refreshingly plain: buy names when recent news sentiment turns positive, and exit when it turns negative. It runs over a 24-name large-cap universe spanning tech (AAPL, MSFT, GOOGL, NVDA), financials (JPM, BAC, V, MA), healthcare (JNJ, UNH, PFE, ABBV), and staples/industrials (PG, KO, WMT, COST, CAT, HON, XOM, CVX). That's a liquid, well-covered set — exactly where sentiment signals have the most raw text to work with. The strategy is currently flagged live.
Recent activity
The headline from the last week or so is inactivity. Across six scheduled runs between 2026-09-03 and 2026-09-10, the strategy executed zero trades and rejected one or two candidate orders on each run. Cash has sat unchanged at $990.06, while total paper value has drifted between roughly $10,277 and $10,525 — closing the 2026-09-10 run at $10,304.52, a little above the $10,000 starting line.
The last actually executed orders are older: a GOOGL buy on 2026-07-01, and before that UNH, BAC, MSFT, and AAPL positions opened across May and June. Every recorded trade is a buy — there are no exits in the log yet, which matters for reading the performance figures below.
Backtest and validation
Over a 451-day backtest the strategy returned just 0.19% (final equity $10,018.55, ~0.1% CAGR) on a Sharpe of 0.74, with a shallow 0.05% max drawdown, 36.5% turnover, and only 2 trades. The reported win rate is 0% — but with just two open buys and no logged exits, that reads as an artifact of unclosed positions rather than two clear losers.
Validation is where caution is warranted. The 4-fold walk-forward did not pass: only 1 of 4 folds was positive, and folds 1 through 3 recorded zero trades at all. The entire result comes from fold 4 (2025-12-16 to 2026-05-29), which posted the 0.19% return and a 1.5 out-of-sample Sharpe on two trades. The probabilistic Sharpe (PSR) of 0.923 looks healthy, but the deflated Sharpe (DSR) of 0.551 — adjusted for 6 trials — is far more sober.
Strengths and risks
Strengths: the logic is transparent and easy to reason about, drawdown has been negligible, and the strategy is not overtrading or bleeding fees ($2 total in the backtest). When it did fire out-of-sample, the risk-adjusted number was respectable.
Risks: the sample is simply too thin to trust. Two trades over 451 days, three empty folds, and a failed validation mean we have almost no evidence the signal generalizes — the single good fold could be luck. The recent live behavior compounds the worry: the strategy keeps finding candidates but rejecting them, so it is neither validating the thesis nor accumulating the track record it needs.
Verdict: a sound idea starved of data. Until the signal fires often enough to fill more than one fold, news-sentiment should be treated as an experiment, not a conviction position.