Thesis
The news-sentiment strategy operates on a straightforward premise: scan recent news across a 24-stock universe of large-cap US equities, buy when aggregate sentiment turns positive, and exit when it sours. The universe skews toward mega-caps — Apple, Microsoft, Nvidia, JPMorgan — where news flow is deep and frequent, in theory providing a reliable sentiment signal without thin-market noise.
The logic is defensible. Academic literature consistently links short-term price drift to earnings surprise and news tone, and large-caps have enough analyst coverage to make the sentiment signal meaningful rather than random. The question is whether this implementation is sensitive enough to act — and that is where the data gets interesting.
Recent Activity
Over the past week, the agent ran daily and executed exactly one trade: a 4-share buy of MSFT at $428.23 on June 3rd. The prior week saw a 6-share AAPL buy at $312.06 (May 31st). Both remain open positions. Every other daily run — June 1st through June 8th — passed without a single execution, with cash sitting at $6,414.72 out of a $9,888–$10,016 total portfolio.
This is the defining characteristic of the strategy so far: extreme selectivity. The agent is not trigger-happy, which is good for risk control, but it also means the portfolio has deployed less than 35% of its capital across 451 backtest days and only two live signals.
Backtest & Validation
The headline numbers look innocuous rather than impressive:
| Metric | Value |
|---|---|
| Total return | +0.19% |
| CAGR | +0.10% |
| Sharpe (backtest) | 0.74 |
| Max drawdown | 0.05% |
| Trades (451 days) | 2 |
A max drawdown of 0.05% is pristine — but mostly because the strategy rarely commits capital. Two trades over 15 months is not a risk-managed strategy; it is an idle one.
The walk-forward validation tells the harder story. Across four folds covering August 2024 through May 2026, only Fold 4 (Dec 2025–May 2026) generated any activity. Folds 1–3 returned zero trades, zero return, zero drawdown — effectively dead periods. The strategy's validation status is failed: only 1 of 4 folds was positive, and you cannot validate an edge from a single fold that contains just two trades.
The out-of-sample Sharpe of 1.50 looks attractive, but a two-trade sample makes it statistically meaningless. The Deflated Sharpe Ratio of 0.551 — which adjusts for multiple testing across 6 trials — is the more honest number, and it sits well below the conventional acceptance threshold of 1.0.
Strengths and Risks
Strengths:
- Capital preservation is excellent; no material drawdown observed.
- The PSR of 0.923 suggests the backtest Sharpe is unlikely to be pure noise, modestly.
- When the signal does fire, it picks recognizable large-caps with genuine news catalysts.
Risks:
- Sample starvation: two trades cannot support any conclusion about edge, positive or negative. The strategy needs at least one full market cycle of consistent signals before meaningful evaluation.
- Sentiment threshold may be too tight: three consecutive silent folds suggest the model's positive-sentiment bar is set so high it rarely triggers, or the signal extractor is not processing enough news sources.
- Opportunity cost: $6,400 sitting in cash while the market moved through a significant stretch of 2025 is a real cost, even in paper trading terms.
Outlook
news-sentiment is not broken, but it is unproven. The priority should be diagnosing whether the signal drought in Folds 1–3 reflects a deliberate design choice or a data pipeline gap. If the sentiment extractor was simply not ingesting enough news volume in those periods, a fix could unlock a far richer signal history. Until then, treat current metrics as preliminary — neither a green light nor a red flag.