Thesis
The news-sentiment strategy follows a deliberately simple rule: buy on positive recent news sentiment, exit on negative. It trades a 24-name large-cap universe spanning tech, financials, healthcare, staples and energy — AAPL, MSFT, NVDA, JPM, V, JNJ, XOM and peers. The appeal is intuitive: lean into names the news cycle is treating kindly, step aside when the tone sours. The risk is equally well known — sentiment is noisy, often lagging, and can whipsaw a strategy into buying tops and selling bottoms.
Recent activity
The strategy is live. Its scheduled run on 2026-10-29 executed three buys with zero rejections: 66 shares of PFE at $29.92, 9 of NVDA at $213.49, and 4 of V at $410.29. After that deployment it held $4,405.24 in cash against a total book of $9,942.53 — fractionally below a $10,000 starting line, consistent with fresh Season 2 positions still finding their footing.
Approval came from an automated review on 2026-05-31: "Approved (risk 0). Built-in hand-written strategy." This is a vetted, rules-based entry rather than an experimental LLM-authored one.
Backtest and validation
Here the picture turns cautionary. Over 451 days the backtest returned +0.19% (final equity $10,018.55), a CAGR near 0.1%, with a Sharpe of 0.74 and a shallow 0.05% max drawdown. Turnover was 36.5%, fees a trivial $2, FX cost zero. But those figures come from only two trades, and the reported win rate is 0% — the net gain was not driven by a majority of winning positions.
Cross-validation makes the fragility explicit. Across four folds, only one was positive, and that single fold (Dec 2025–May 2026) is the sole source of all activity: the other three folds placed zero trades and returned nothing. The strategy simply did not fire for most of the tested history. Unsurprisingly, validation did not pass (passed: false).
The statistics cut both ways. The Probabilistic Sharpe Ratio is a healthy 0.923, and out-of-sample Sharpe reads 1.5. Yet the Deflated Sharpe Ratio — which penalises for the 6 trials run — falls to 0.551, and with a two-trade sample those ratios rest on almost no data.
Verdict
news-sentiment is clean, cheap to run, and low-drawdown by construction. But it is effectively untested at scale: a coherent thesis whose evidence base is two trades in one fold. The near-silent folds suggest the sentiment trigger is too rarely met to judge. Watch the fresh PFE/NVDA/V positions closely — this live run may finally generate the sample size the backtest never had.