Thesis
The premise is simple and intuitive: buy names showing positive recent news sentiment, exit when sentiment turns negative. The strategy fishes in a 24-stock pond of familiar large caps — AAPL, MSFT, GOOGL, NVDA, the money-center banks, healthcare and staples like JNJ, UNH, PG and KO. It is currently running live in paper mode.
Sentiment-driven entries are a well-trodden idea, and the universe is liquid and defensible. The harder question is whether the signal actually fires often enough — and cleanly enough — to matter.
Recent activity
The live paper book has been quiet. Across scheduled runs from Aug 3 to Aug 10, the strategy executed zero trades on all but one day, which logged a single rejected order (Aug 5). Cash sat unchanged at $990.06 throughout, with total account value drifting between $10,466 and $10,573 and last marked at $10,506.38 on Aug 10 — a book that is almost fully invested and largely idle.
The positions themselves date back to a burst of buying between May 31 and Jul 1: AAPL (6 sh @ $312.06), MSFT (4 @ $428.23), BAC (35 @ $55.93), UNH (4 @ $402.85) and GOOGL (5 @ $360.40). Notably, no exits appear in the recent record — the "sell on negative sentiment" half of the thesis has not been exercised in the visible window.
Backtest and validation
Here the picture gets uncomfortable. The backtest posts a 0.19% total return over 451 days (CAGR ~0.1%), a Sharpe of 0.74, and a tiny 0.05% max drawdown — but on just 2 trades and a 0% win rate. Turnover was a modest 36.5%. A drawdown that small usually signals a strategy that is barely in the market, not one that is skillfully hedged.
Walk-forward validation failed. Of four folds, only the last (Dec 2025 – May 2026) traded at all; folds 1–3 produced zero trades and zero return. In other words, the entire result rests on a single window with two fills. The statistics reflect this fragility: a healthy-looking probabilistic Sharpe (PSR 0.923) collapses to a deflated Sharpe of 0.551 once you account for the 6 trials run — barely better than a coin flip. Only 1 of 4 folds was positive.
Verdict
Strengths: a clear, sensible thesis; a clean, liquid universe; and a discipline that keeps it out of the market when the signal is absent — hence the negligible drawdown.
Risks: the sample is far too thin to conclude anything. Two backtest trades and a validation that only came alive in one fold mean the 0.19% return is closer to noise than to evidence. The deflated Sharpe warns explicitly against over-crediting the headline number.
For now, news-sentiment is best read as a live probe rather than a validated edge. The next honest test is simple: does the signal fire often enough, over more windows, to build a trade count worth measuring?