Backtest precision asks, for one specific ticker: of all the past moments the model flagged a BUY, what share were followed by a real gain that hit the target threshold (default +5% within about 63 trading days)? A precision of 70% means seven of every ten past BUY flags reached the +5% bar.
Precision is measured per ticker, and any single name usually has only a handful of past BUY signals to score. With a sample that small, a high reading is suggestive, not proof — a run of three-for-three is nowhere near enough to conclude the model has a durable edge on that stock. Treat a strong number as a hint worth more scrutiny, not a "strong track record" you can lean on. (Example, illustrative: 3 wins out of 4 signals reads as 75% precision, but 4 signals tells you almost nothing statistically.)
Precision only means something once you know the bar it is measured against. Change the threshold (say +3% instead of +5%) or the horizon (six months instead of three) and the same history produces a different precision. A high number under a soft threshold is not the same achievement as a high number under a demanding one. And a backtest is a rewind of history, not a guarantee of live results.
Precision is one of the inputs to the Kelly position-sizing calculation, alongside win rate and average return. Noise in a small-sample precision figure propagates straight into a suggested position size. It is precisely because these inputs are noisy that the sizing is confidence-scaled and hard-capped rather than taken at face value — the system is built to distrust its own edge estimates, and so should you.
By default, a return of +5% within about 63 trading days (roughly three months). A past BUY signal counts as a hit only if the price reached that bar within the window, so precision is always tied to a specific threshold and horizon.
No. Precision is measured per ticker on a small sample — often just a handful of past signals — so a high number is suggestive, not proof. It is a hint that deserves more scrutiny, not a track record to lean on, and a backtest is not a guarantee of live results.
It is one input to the Kelly position-sizing formula, so noise in a small-sample precision figure flows into the suggested size. That is exactly why the sizing is confidence-scaled and capped rather than taken at face value.
Browse all S&P 500 tickers to see this metric applied to individual companies.
Educational research only — not investment advice.