Do fair value gaps really fill?

A fair value gap is a three-bar move that leaves a price range untraded: the third bar's low is above the first bar's high, or the reverse. Popular teaching says price comes back to “fill” it. Does it, more than chance? And which way of trading a gap holds up?

• NQ, ES, gold, oil and Bitcoin futures, regular session, 8 Sept 2025 to 30 Oct 2026, about 147,000 gaps on 1, 5 and 15-minute charts.
• A gap is known only at the close of its third bar; every trade starts after that.
• The fill rate was compared with price travelling the same distance the other way.
• Three ways of trading: fade toward the gap; go with it at once; and retest, a limit order at the gap's near edge with the stop beyond its far edge and a target of twice the risk. Micro contracts, after costs, against random entries with the same exits, in two halves.

Fading toward the fill lost on every market and chart, no better than random. Going with the gap at once lost too.

The retest beat random entries before costs on almost every market. But on small gaps the stop is tiny and costs eat it: 1-minute gaps lost on every version. Retests only paid when the stop was at least about ten times the round-trip cost.

With that size, a retest in the direction of a day already trending that way made about +0.19 R a trade, positive on all five markets and in both halves. Without the trend filter, about +0.10 R.

A gap is a fast, one-sided move. A pullback into it is often traders who missed the move getting in, which is why a retest with the move works better than betting on a full reversal. On a day that is already trending, that pull is stronger.

A gap on its own is noise. If you use gaps, use them as a place to join a move that is already going your way, with a stop that is large next to your costs, and skip 1-minute gaps.

About 14 months. The gap-size cut-off was chosen after seeing results, so the retest finding is a lead to re-check on new months, not proof. A gap-retest idea is not yet a starting idea in our app.

Simulated results on past market data from 2025 and 2026. Research only, not trading signals or advice. Past results do not promise future ones; see the hypothetical performance disclosure at the bottom of the page.