Thesis
The news-sentiment strategy is built on a straightforward premise: buy when recent news flow around a stock turns positive, exit when it turns negative. The universe spans 24 large-cap U.S. names across tech, financials, healthcare, consumer, and energy — liquid, well-covered names where news signal tends to be abundant and fast-moving.
The appeal is intuitive. Institutional sentiment often leads price, and with LLM-grade NLP now cheap enough to run on a schedule, harvesting that signal at paper-trading scale is a reasonable first experiment.
Recent Activity
The agent was deployed on 31 May 2026 with a $10,000 paper balance. It opened an initial position immediately — 6 shares of AAPL at $312.06 — and added 4 shares of MSFT at $428.23 on 3 June. The two days in between (1–2 June) saw no executions, meaning the sentiment filter found nothing actionable in the universe on those sessions.
As of the 3 June run, the portfolio sits at $9,987.04 total with $6,414.72 in cash — roughly 36% deployed across two positions. Turnover in the backtest was 36.5%, which is consistent with a strategy that trades selectively rather than churning the book.
Backtest & Validation
Over a 451-day backtest window, the strategy returned +0.19% — essentially flat — with a Sharpe of 0.74 and a maximum drawdown of just 0.05%. Total fees were $2, FX costs zero. On paper those risk figures look clean, but the return itself offers little to celebrate.
The cross-validation tells a more pointed story. Across four folds covering August 2024 to May 2026, only the final fold (December 2025 – May 2026) generated any activity at all — two trades, a 0.19% return, and a Sharpe of 1.5. The three earlier folds posted zero trades, zero return, and zero drawdown. The validation framework has correctly flagged this: validation did not pass.
The probabilistic Sharpe ratio (PSR) sits at 0.923, which is encouraging — it suggests the observed Sharpe is unlikely to be pure noise. But the deflated Sharpe ratio (DSR) of 0.551, adjusted for the six trials run, pulls that optimism back down. With most of the signal concentrated in a single fold and a two-trade live history, there simply isn't enough data to draw confident conclusions.
Strengths
- Minimal drawdown: 0.05% max drawdown across the backtest period is genuinely low, even if it partly reflects inactivity rather than active risk management.
- Selective execution: the strategy isn't firing randomly. Going three days without a trade in live conditions suggests the sentiment filter has some discrimination.
- PSR > 0.90: for a strategy this early in its life, a PSR of 0.923 is a reasonable starting point.
Risks & Open Questions
- Near-zero activity in early folds: if the strategy couldn't find trades in three of four historical windows, it may be overfitted to conditions that only emerged recently — or the news data coverage for those earlier periods was simply thin.
- Two live trades is not a sample: the live track record is days old. Any performance assessment at this stage is speculative.
- News latency and source quality: the strategy's edge lives or dies on how fresh and accurate the sentiment input is. That pipeline isn't visible in these metrics.
- Flat overall return: +0.19% over 15 months is below any reasonable hurdle rate. Even as a paper account, the opportunity cost benchmark matters.
Outlook
News-sentiment is best understood right now as a live hypothesis under observation, not a validated edge. The next 60–90 days of live execution — particularly how it navigates earnings season and macro events — will be far more diagnostic than the backtest. Watch for whether trade frequency increases as the strategy finds its footing, and whether the positions it takes show any pattern of outperforming the benchmark on a per-trade basis.