The thesis
news-sentiment runs a simple premise across a 24-name large-cap universe (AAPL, MSFT, NVDA, JPM, XOM and peers): buy when recent news sentiment turns positive, exit when it turns negative. It is a classic event-driven idea — let the flow of headlines, not price action, decide entries. The appeal is intuitive. The hard part, as the data makes clear, is getting the signal to fire often enough to matter.
Recent activity
The live paper account is quiet. Across six scheduled runs from 23–30 September, the strategy executed trades on only one day: 23 September, when it bought 4 NVDA at $225.34 and sold 35 BAC at $55.96. That BAC lot was opened on 12 June at $55.93 — effectively a round-trip to flat. Every other recent run logged "0 executed," and several carried one or two rejected orders, pointing to signals that failed risk or execution checks. Total portfolio value has drifted in a narrow band, sitting at $10,266.66 on 30 September against $2,036.71 in cash — stable, but hardly dynamic.
Backtest and validation
Over 451 days the backtest returned just 0.19% (19 basis points), a 0.1% CAGR, with a Sharpe of 0.74 and a negligible 0.05% max drawdown. Fees were a rounding error at $2. The telling figure is the trade count: two. A win rate of 0 alongside a positive return means the gains came from open mark-to-market, not from closed winners.
Validation did not pass. Of four walk-forward folds, three executed zero trades — the strategy was dormant for most of the test window — and only the final fold (Dec 2025–May 2026) was active, posting 0.19% and a 1.5 Sharpe on its two trades. One positive fold out of four is thin evidence. The statistics underline the tension: a probabilistic Sharpe ratio of 0.923 looks strong, but the deflated Sharpe of 0.551, which penalises the six trials run, is far more sober.
The bottom line
The good news is risk discipline. Drawdowns are negligible, turnover (36.5%) and costs are low, and the one active period performed respectably. The bad news is that there is almost no track record to assess. A strategy that trades twice in 451 days and sleeps through three of four validation folds has not really been tested — it has been left idle. The 1.5 Sharpe in fold four is encouraging but statistically meaningless at a sample size of two.
My read: treat news-sentiment as an unproven prototype, not a deployable edge. Before it earns capital, the signal needs to fire often enough to generate a real sample — and the gap between its optimistic PSR and its deflated DSR is a standing reminder not to mistake one good fold for destiny.