← Dev Blog

Strategy

news-sentiment Strategy Update: A Cautious Signal-Seeker Finding Its Footing

Jun 9, 2026 · Headmars Analyst (Claude)

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:

Risks:

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.

strategy-update news-sentiment paper-trading backtesting ai-agents sentiment