Thesis and Approach
The news-sentiment strategy is straightforward by design: scan recent news headlines across a curated 24-name large-cap universe, enter positions when aggregate sentiment is positive, and exit when it turns negative. The watchlist covers the obvious blue-chip clusters — mega-cap tech (AAPL, MSFT, GOOGL, NVDA), financials (JPM, BAC, V, MA), healthcare (JNJ, UNH, PFE, ABBV), consumer staples and discretionary, plus a handful of industrials and energy names. Broad enough for diversification, concentrated enough to maintain signal quality.
Live Activity (May 31 – June 5)
The agent deployed with $10,000 on May 31 and immediately executed one trade: 6 shares of AAPL at $312.06, committing roughly $1,872 of capital. Four days later, on June 3, it added 4 shares of MSFT at $428.23 (~$1,713). That's it — five of six daily runs recorded zero executions.
Holding value has drifted modestly around the $9,940–$10,016 range over the observation window, with cash sitting at $6,414 since the MSFT buy. The strategy is not churning; it is waiting for conviction signals that largely haven't arrived.
Backtest and Cross-Validation
Over a 451-day backtest window, the strategy returned +0.19% on $10,000 starting equity, ending at $10,018.55. With only 2 trades, 36.5% portfolio turnover, and $2 in total fees, the picture is one of extreme selectivity — or inactivity, depending on your lens. The Sharpe of 0.74 is uninspiring on its own, and the win-rate statistic is undefined at this sample size.
The four-fold walk-forward validation tells a more nuanced story:
| Fold | Period | Return | Sharpe | Trades |
|---|---|---|---|---|
| 1 | Aug 2024 – Jan 2025 | 0% | 0 | 0 |
| 2 | Jan – Jul 2025 | 0% | 0 | 0 |
| 3 | Jul – Dec 2025 | 0% | 0 | 0 |
| 4 | Dec 2025 – May 2026 | +0.19% | 1.50 | 2 |
Only fold 4 generated any activity. The out-of-sample Sharpe of 1.50 for that fold is genuinely encouraging, but it rests on two trades across roughly five months — statistically thin ground.
Validation result: not passed. With only 1 of 4 folds positive and a Deflated Sharpe Ratio (DSR) of 0.551 against 6 backtest trials, the validation framework correctly flags this as insufficiently proven. The Probabilistic Sharpe Ratio of 0.923 is stronger, suggesting the observed Sharpe likely beats a zero benchmark — but DSR's trial-count penalty is the binding constraint here.
Strengths
- Low drawdown: max drawdown of 5% is well-contained for an equity strategy.
- Capital discipline: sitting in cash when there's no signal is a feature, not a bug. The strategy isn't forcing trades.
- Fold-4 Sharpe: when the model does act, the risk-adjusted return in the most recent period looks reasonable.
Risks and Open Questions
- Sparse signal: three of four backtest folds produced zero trades. Either the sentiment threshold is too conservative, the news feed lacks coverage depth for most tickers most of the time, or both.
- Sample size: two live trades cannot validate any hypothesis. Weeks more of observation are needed before live performance carries any inferential weight.
- Sentiment lag: headline-based sentiment is a well-crowded signal. Without knowing the news source latency and how the model processes it, there's a real risk of acting on information already priced in.
- DSR failure: the validation framework exists precisely to penalize over-optimized strategies with few trials. Passing validation requires more consistent fold performance — not just one good window.
What to Watch
The next meaningful checkpoint is whether fold-4-style conditions persist into June and July. If the agent continues executing selectively with contained drawdown, the validation picture improves organically. If it remains dormant, the thesis needs examination at the signal-generation layer — specifically whether the sentiment model is receiving and interpreting news data with sufficient frequency and breadth across the full 24-stock universe.