Thesis
news-sentiment runs a simple, legible rule: buy names showing positive recent news sentiment across a 24-stock large-cap universe — megacap tech, financials, staples, healthcare and energy — and exit when sentiment turns negative. It is currently live in paper trading. The appeal is interpretability: every decision maps back to a headline signal a human can inspect.
Recent activity
The live book is holding, not churning. The last six scheduled runs (Aug 6–13) each executed zero trades, leaving cash pinned at $990.06 and the account nearly fully invested. Its positions were opened earlier — buys in AAPL, MSFT, BAC, UNH and GOOGL between late May and early July. Total equity has drifted lower recently, from roughly $10,506 on Aug 10 to $10,366 on Aug 13, a mid-month pullback of about 1.3%. A week of near-idle runs is consistent with the thesis: no fresh negative sentiment means no exits, and no new signals means no rotation.
Backtest & validation
Here the picture gets thin. Over 451 days the backtest returned just 0.19% (final equity ~$10,019), a 0.1% CAGR, on only 2 trades and 36.5% turnover, with a shallow 0.05% max drawdown and a Sharpe of 0.74. A near-flat curve built on two trades is not a track record — it is a sample too small to trust.
Walk-forward validation makes that explicit, and it did not pass. Of four out-of-sample folds, three produced zero trades and zero return; the entire result rests on a single fold (Dec 2025–May 2026) that booked 0.19% at a fold Sharpe of 1.5. The probabilistic Sharpe ratio is a flattering 0.923, but the deflated Sharpe — which penalizes the six trials behind this design — falls to 0.551. One positive fold out of four is not evidence of a repeatable edge.
Risks and verdict
The strengths are real but modest: a tiny drawdown, a clean interpretable rule, and a live book behaving as designed. The risks dominate. The strategy trades so rarely that its statistics are essentially anecdotal, and a reported 0% win rate against a positive return suggests the sample is still open or mismeasured — either way, not something to lean on. Failed validation and a sub-1.0 deflated Sharpe argue against sizing this up.
For now, news-sentiment is best read as a promising hypothesis on probation: worth keeping live to accumulate trades, not worth real capital until it proves it can act often enough to actually be measured.