The thesis
News-sentiment runs a simple, legible rule: buy names showing positive recent news sentiment, and exit when the tone turns negative. Its universe is a defensive, liquid slice of US large caps — 24 names spanning tech (AAPL, MSFT, GOOGL, NVDA), financials (JPM, BAC, V, MA), healthcare (JNJ, UNH, PFE, ABBV), staples (PG, KO, WMT, COST) and industrials/energy (CAT, HON, XOM, CVX). It is currently marked live.
The appeal is that the logic is easy to reason about and easy to audit. The risk, as the data shows, is that a sentiment gate this coarse rarely fires — and when it does, the sample is too thin to trust.
Recent activity
On the live book, the strategy was busy through early summer, executing five buys: AAPL (6 @ $312.06), MSFT (4 @ $428.23), BAC (35 @ $55.93), UNH (4 @ $402.85) and GOOGL (5 @ $360.40) between late May and July 1. Since then it has gone quiet. The last week of scheduled runs (Aug 24–31) executed zero trades, with a couple of orders rejected, leaving cash at $990.06 and total account value drifting in a narrow $10,310–$10,465 band. In short: fully invested, holding, and waiting for the next signal.
Backtest and validation
The backtest is where enthusiasm should cool. Over 451 days the strategy turned in a 0.19% total return (CAGR ~0.1%), a Sharpe of 0.74, and a tiny 0.05% max drawdown — but on just two trades and a 0% win rate. That is not a track record so much as a rounding error; the low drawdown reflects inactivity, not skill.
Walk-forward validation makes the point sharper. Across four folds, only one was positive — and it was the only fold that traded at all. Folds one through three (Aug 2024 through Dec 2025) produced zero trades and zero return; every dollar of the headline result comes from fold four, which posted 0.19% and a 1.5 Sharpe on two trades. The aggregate probabilistic Sharpe (PSR) looks healthy at 0.92, but the deflated Sharpe (DSR) — which penalises the six trials run — falls to 0.55. Unsurprisingly, validation did not pass.
Reading the tape
Strengths: the strategy is transparent, defensive in construction, and has kept drawdowns negligible. There is no evidence of reckless risk-taking.
Risks: the edge is unproven. A sample of two backtest trades cannot support a win-rate or Sharpe claim, and three of four out-of-sample windows generated no activity — the signal simply does not speak often enough to compound. The gap between five live buys and two modelled trades also suggests the live and backtested regimes are not yet aligned.
Verdict: treat news-sentiment as a live experiment, not a validated engine. Until the signal fires more often — and produces closed trades that survive walk-forward scrutiny — its clean drawdown chart is better read as silence than as strength.