The thesis
news-sentiment is a straightforward idea: buy when recent news sentiment turns positive, exit when it turns negative. It operates over a 24-name universe of US large caps spanning tech (AAPL, MSFT, GOOGL, NVDA), financials (JPM, BAC, V, MA), healthcare, staples, and energy. The logic is intuitive and grounded in a real market phenomenon — sentiment drift around news flow. The question, as always, is whether the signal is strong and frequent enough to matter.
Recent activity
The strategy is live and running scheduled daily sessions. The last week of runs (Aug 27 through Sep 3) tells a consistent story: zero trades executed, with the occasional rejected order. Cash has been parked at $990.06 the whole time while total account value drifted between roughly $10,321 and $10,525 — movement driven by existing holdings, not new decisions.
Those holdings came from a cluster of buys earlier in the summer: AAPL (May 31), MSFT (Jun 3), BAC (Jun 12), UNH (Jun 18), and GOOGL (Jul 1). Since then, the signal has gone quiet. That is not necessarily a fault — a sentiment gate should stay flat when there is nothing decisive to act on — but it does mean the strategy is currently more a buy-and-hold basket than an active trader.
Backtest and validation
Here the picture demands caution. Over 451 days the backtest returned just 0.19% (final equity $10,018.55), a 0.10% CAGR — essentially flat. Sharpe was 0.74, max drawdown a mild 0.05%, and turnover 36.5%. But the headline caveat is trade count: only 2 trades, with a 0% win rate. You cannot draw meaningful conclusions from a two-trade sample.
The walk-forward validation failed, and the fold breakdown shows why. Of four folds, three saw zero trades and posted flat returns; the entire 0.19% came from fold 4 (Dec 2025–May 2026), which alone managed a 1.5 Sharpe. So only 1 of 4 folds was positive, and the strategy's whole track record rests on a single window. Statistically, the Probabilistic Sharpe Ratio looks encouraging at 0.923, but the Deflated Sharpe of 0.551 — which penalises for the 6 trials run — is the more honest number, and it sits barely above a coin flip.
Verdict
Strengths: a clean, defensible thesis; tight drawdowns; and no evidence of reckless risk-taking. Risks: the signal fires far too rarely to establish an edge, the sample is tiny, and performance is concentrated in one fold — a classic overfitting warning sign, which is exactly why validation gated it out.
news-sentiment is best read as a hypothesis under observation, not a validated engine. The sensible next steps are widening the sentiment trigger or lengthening the news lookback to generate more trades, then re-running validation. Until it trades enough to be measured, the jury stays out.