1. What it is
A fair value gap is a three-bar pattern in which the first and third bar do not overlap. The space between them is a price range that the market moved through quickly, leaving little trade behind.
2. How it is computed
For a bullish gap, the low of the third bar is above the high of the first. For a bearish gap, the high of the third is below the low of the first. The gap is the distance between those two levels.
Most implementations add a minimum size filter, typically a fraction of ATR, because in fast markets the raw pattern occurs constantly. A gap is usually considered closed once price trades back through it, though "back through" is defined differently from tool to tool — some require a full fill, some only a touch.
3. What it rests on
The assumption that price tends to return to areas it passed through too quickly — that an imbalance left behind acts as a magnet.
This is a widely held idea, and it is not a law. Nothing in the mechanics of an order book requires price to revisit a range. The argument is behavioural: participants who missed the move want to transact at the earlier price, and that creates orders there. Plausible, but not demonstrated by the pattern itself.
4. Where it is wrong
In a fast market gaps form constantly, and most of them fill within minutes without meaning anything. The pattern is common enough that finding examples where it worked is trivial — which is exactly why examples prove nothing.
Three specific failures:
The minimum size filter decides almost everything. Set it low and you get noise; set it high and you get a handful of instances a month. Where to set it is not determined by the pattern.
Most tools track only the most recent gap in each direction. Older unfilled gaps stay drawn but stop being used, which means the chart shows more than the tool is actually watching.
And "the gap filled" is not the same as "the gap predicted the fill". Price covers a lot of ground in a session. A range it passed through will often be passed through again simply because it is nearby.
5. Is there a measurement
Registered, not yet measured. The claim is that price returns into a gap more often than it continues without returning. The measurement is defined; the data has not been collected.
Price returns into a fair value gap more often than it continues without returning.
In the measurement log