Thesis
The news-sentiment strategy runs a simple, legible rule: buy names showing positive recent news sentiment, and exit when sentiment turns negative. It trades a 24-stock large-cap universe spanning tech, financials, healthcare, staples and energy — AAPL, MSFT, NVDA, JPM, JNJ, XOM and their peers. The appeal is interpretability: every position maps back to a readable signal, and the rule carries a built-in exit rather than leaving stops to chance.
Recent activity
Live trading has been quiet. The six scheduled runs between 14 and 21 October executed no trades; three of them logged a single rejected order, and cash sat unchanged at $2,036.71 throughout. The reported account total drifted between roughly $10,345 and $10,510 as existing holdings marked to market. The last real activity was a 23 September rotation — the book sold 35 shares of BAC at $55.96 and bought 4 shares of NVDA at $225.34 — following an accumulation phase from late May through July (AAPL, MSFT, BAC, UNH and GOOGL). The behaviour reads as a strategy that took on exposure mid-year and has since mostly held, declining to force trades when the signal is absent.
Backtest and validation
Here the picture is sobering, and worth stating plainly. Over 451 backtested days the strategy returned just 0.19% (CAGR ~0.1%), ending at $10,018.55 on total fees of $2. Sharpe was 0.74 and max drawdown a negligible 0.05%. But this came from only 2 trades, and the walk-forward validation did not pass. Of four folds, three produced zero trades and zero return; all performance was concentrated in a single fold (Dec 2025–May 2026), which alone posted the 0.19% return and a 1.5 Sharpe. Only 1 of 4 folds was positive.
Reading the numbers
Strengths first: the drawdown is tiny, fees are trivial, and the one active fold showed a respectable out-of-sample Sharpe of 1.5. The probabilistic Sharpe ratio (0.923) is encouraging — though the deflated Sharpe (0.551, adjusted for 6 trials) is far more modest and probably the more honest figure.
The risks dominate, though. A two-trade sample is not evidence; it is anecdote. With three of four folds completely inert, the signal fires too rarely to validate, and the headline return is statistically indistinguishable from flat. A reported 0% win rate alongside a positive total return suggests the bookkeeping is driven by open rather than closed positions. Validation failed for good reason.
Verdict
news-sentiment is a clean, well-motivated idea that has not yet proven itself. It is low-risk and low-activity to a fault — it simply doesn't trade enough to separate skill from a single lucky fold. Before we'd lean on it, it needs a sentiment threshold that engages more often and a sample large enough to make the Sharpe mean something. For now: watch, don't scale.