The Honest Backtest: How DataQuant Measures Itself

Every signal service shows you its winners. The only numbers worth your attention are the ones computed by a rule that couldn't cherry-pick — and published in the months they look bad.

The measurement rule

DataQuant's backtest is deliberately simple enough to audit in one paragraph. Take the first signal per stock in the window (no doubling up when a name re-fires). Enter at the closing price on the signal date — not an idealized intraday fill. Exit at the close 20 trading days later, or at -8%, whichever comes first. Weight every trade equally. Disclose every signal — not a curated subset. And never annualize: a 30-day sample is a 30-day sample.

The good window — and the bad one

The window of May 22 – June 20, 2026: 22 signals, +6.9% average return, 64% win rate. Live production data, all signals disclosed.

Then July: roughly a 37% win rate with a slightly negative average return. Same rules, same discipline, same scanner. What changed was the market — a choppy tape where indexes held up while institutions distributed underneath, precisely the regime where breakout entries fail most. Breakout systems are regime-dependent. That isn't a flaw to hide; it's the central fact of the strategy, and it's why the Market Health gauge exists: to tell you which regime you're in before the month's results do.

The parts most services don't mention

• Survivorship honesty. In one recent window, 41 candidate signals produced only 19 evaluable trades — the rest lacked complete price history (delistings, data gaps) or hadn't finished their 20-day forward window. That attrition is stated, not hidden, because silently dropping unmeasurable trades is how backtests flatter themselves.
• The stop is part of the arithmetic. Capping every loss at -8% is what lets a strategy survive a 37% win-rate month. The edge of breakout trading was never accuracy — it's asymmetry: losses capped small, winners given 20 days of room.
• Closes, not wishes. Entries and exits at real closing prices mean the numbers include gaps and slippage-adjacent reality, not perfect fills at the exact pivot.

The asymmetry that makes it work

It's worth spelling out why a strategy can survive — even prosper through — win rates that would sink a coin-flip bettor. The -8% stop puts a hard floor under every loss, while winners get twenty trading days of unbounded room. When the average winner is several times the size of the average capped loss, a win rate well under 50% still compounds; when the tape trends, a 64% month is a windfall. The discipline isn't decoration on the strategy — the discipline is the strategy. Accuracy fluctuates with the regime; the asymmetry is the part you control.

How to read any backtest (including this one)

Ask three questions. Could the rule cherry-pick? (First-signal-per-stock, all-disclosed answers that.) Is the sample one regime? (30 days always is — which is why months are published as they accumulate, wins and losses both.) Does the method survive its own bad month? (The -8% cap is the answer here — a losing month costs little; a trending month pays.) A service that shows its worst window next to its best is offering you evidence. One that shows only the best is offering you marketing.