The thesis
The news-sentiment strategy runs a simple, intuitive rule across a 24-name large-cap universe (AAPL, MSFT, GOOGL, NVDA, JPM, and peers): buy when recent news sentiment turns positive, exit when it turns negative. It is currently live as a paper-traded book. The appeal is obvious — sentiment is a fast-moving, event-driven signal that traditional price-only strategies ignore. The catch, as the data shows, is that a rule this selective only speaks up rarely.
Recent activity
The live book sits at roughly $10,321 total as of 27 Aug 2026, with about $990 in cash. The last week of scheduled runs is telling: six consecutive sessions executed zero trades, and three of them logged a rejected order. The strategy is holding, not trading. The most recent executions on record are a cluster of buys from late May through early July — GOOGL, UNH, BAC, MSFT, and AAPL — after which activity went quiet. That pattern is consistent with a sentiment signal that fired during a news-rich stretch and has stayed dormant since.
Backtest and validation
Here is where balance matters. On the headline backtest the strategy returned just +0.19% over 451 days (final equity ~$10,019), with a Sharpe of 0.74, a max drawdown of only 0.05%, and — critically — just 2 trades. A near-flat curve with a barely-there drawdown looks calm, but it is calm because almost nothing happened.
The walk-forward validation confirms this and, importantly, did not pass. Across four folds, only one was positive: folds 1–3 executed zero trades and returned exactly 0%, while all of the strategy's activity and return landed in fold 4 (Dec 2025–May 2026), which posted +0.19% on an out-of-sample Sharpe of 1.5. So the entire track record rests on a single active window.
Reading the risk metrics
The probabilistic Sharpe ratio (PSR 0.923) looks encouraging, and the deflated Sharpe (DSR 0.551) survives the six-trial adjustment — but both are computed over a sample of essentially two trades. That is far too thin to trust. The reported win rate of 0% likely reflects open, unrealised positions rather than a string of losers, but it underlines the same problem: there is not enough closed activity to judge the edge. One positive fold out of four is not evidence of skill; it is evidence of a signal that is silent most of the time and occasionally lucky.
Verdict
news-sentiment is a reasonable idea implemented conservatively, with genuinely small drawdowns and a clean out-of-sample fold. But it failed validation for the right reasons: three of four folds produced no trades at all, and every metric hangs on a two-trade sample. Before this earns real conviction, it needs either a looser trigger that fires across more regimes or a longer live run that actually populates the trade log. Until then, treat the numbers as a promising hint — not a proven edge.