Strategy Overview
news-sentiment operates on a clean, intuitive thesis: enter long positions when recent news flow turns positive for a stock; exit when sentiment reverses. The strategy scans a 24-name universe of large-cap U.S. equities spanning technology, financials, healthcare, consumer staples, and energy — household names such as AAPL, NVDA, JPM, and XOM. Execution is fully automated on a daily schedule.
Recent Activity
The past week has been quiet by design. Between July 1 and July 7, the strategy executed a single trade — a 5-share buy of GOOGL at $360.40 on July 1 — while rejecting one signal on July 7 and logging zero executions on every other day. This selectivity is intentional: the model declines to act when sentiment confidence falls below threshold rather than forcing trades.
As of July 7, the paper portfolio stands at $10,073.18 with $990 in cash. Active positions include AAPL (6 shares), MSFT (4), BAC (35), UNH (4), and GOOGL (5), accumulated between late May and early July 2026.
Backtest Performance
Over the 451-day backtest window the headline figures look measured rather than spectacular:
| Metric | Value |
|---|---|
| Total Return | 0.19% |
| CAGR | ~0.10% |
| Sharpe Ratio | 0.74 |
| Max Drawdown | 0.05% |
| Trades | 2 |
| Turnover | 36.5% |
| Total Fees | $2.00 |
Capital preservation is a genuine strength: a 0.05% maximum drawdown means the strategy has not taken a meaningful hit. Turnover at 36.5% is moderate — this is not a hyperactive signal chaser. The win rate of 0% deserves context: with only two trades still open and no closed positions, there are no realized winners or losers yet. The metric is undefined, not damning.
Validation: The Concerning Picture
The formal four-fold cross-validation is where the strategy's fragility becomes visible. Of the four folds spanning August 2024 through May 2026, three produced zero trades and zero return. All measurable performance — the 0.19% return and a fold-level Sharpe of 1.5 — originates entirely from fold 4 (December 2025 to May 2026).
The strategy failed validation on this basis. The Deflated Sharpe Ratio (DSR) of 0.551 adjusts for the six parameter trials run and reflects a genuine overfitting concern. The Probabilistic Sharpe Ratio (PSR) is more encouraging at 0.923 — suggesting the observed Sharpe is unlikely to be pure noise — but with a sample of two trades, that confidence interval is very wide.
Key Risk Factors
- Sparse signal: Three-quarters of the historical test window generated no actionable sentiment signal. Whether this reflects genuine market conditions or a calibration gap in the NLP pipeline is an open question.
- Recency concentration: The entirety of the strategy's track record comes from a single recent fold, which may reflect a specific news-driven environment that won't repeat.
- Sample size: Two trades cannot support statistically meaningful risk or return inference. Every metric shown carries wide error bars.
- Sentiment data quality: Headline sentiment diverges from article-level analysis; news latency and source bias are real failure modes that backtests may not fully capture.
Bottom Line
news-sentiment is a disciplined, live strategy with a coherent thesis and admirable capital preservation. But it is too early to draw confident conclusions from its performance record. The validation failure — rooted in a near-silent backtest history outside of one recent window — is the primary concern to watch. More trade history, and ideally a regime change that stresses the signal, will reveal whether this strategy has genuine edge or a narrow window of accidental fit.