Same signal, three hold times: +$2,134, −$2,512, −$408

AlgoProven Research · June 2026 · #7 in a series on why backtests lie · read #6

We exported every fill from our own broker — 82 round-trips across a practice and an evaluation account over two weeks — and sorted them by the one variable most traders never chart: how long the position stayed open. The gap between the hold-time buckets was wider than the gap between our best and worst instruments. Hold time, not symbol picking, was the edge.

The three buckets

Same strategy, same account, same fortnight. Split only by how long each trade was held:

Hold timeNet P&LWin rateTrades
Held > 30 min+$2,13457%30
Held 2–30 min−$2,51235
Scalped < 2 min−$4087

The evaluation account — a smaller, independent sample — told the identical story: everything held longer than 30 minutes made +$1,625 (four trades, four winners); everything shorter lost −$244. One variable, two accounts, same sign.

Why the middle bucket bleeds

The signal underneath these trades is a mean-reversion / overnight edge. Its natural horizon is hours — sometimes a full session — not minutes. Held to that horizon, it pays. Chopped into intraday churn, every extra entry and exit pays the spread and the commission again, and the per-trade edge is simply too thin to clear that toll a dozen times over.

Fees make it concrete. Over one stretch, $463 in commissions turned a −$321 gross result into a −$785 net one — the fees were 144% of the loss itself. At roughly $5–$7.50 per round-trip on micro futures, hold time is the difference between paying the toll once and paying it on every nervous re-entry.

It wasn't the symbol — until it was the fill

By instrument, the liquid index micros did their job: MES +$1,257, MNQ +$1,234 — tight spreads, clean fills. The worst line was micro crude, MCL −$2,012 on a 15% win rate. That wasn't a broken signal; it was a fill problem — a wider spread on a thinner book turning marginal entries into losers before the trade had a chance. Hold time and liquidity are the same lesson seen twice: an edge only survives where execution is cheap enough to let it through.

A backtest measures the signal. Your broker measures the signal minus execution — and execution is mostly spread and hold time. One of those you can't change. The other you choose on every single trade.

What we did about it

Our live bots are tuned to hold to the mechanism's horizon instead of looking busy, and short-churn entries are the first thing we cut when an account starts bleeding. It's also why every number we publish is the broker's number, not the notebook's. The signal was never the problem here. The clock was.