Do Liquidity Sweeps Mark Market Tops? What 2,212 Day Highs Show

Measured against a proper control group, liquidity sweeps do not distinguish real market tops from ordinary swing highs. Real NQ day highs swept a level before reversing 54 percent of the time; ordinary swing highs that went nowhere swept 62 percent of the time. Sweeps fired more often on the failures.

That result comes from comparing 2,212 NQ daily highs against 8,848 control swing highs across nine years of one-minute regular-session data. The control group is the point: most structural claims are tested against nothing, and any feature looks predictive until you check how often it also fires where nothing happens.

The numbers

The three features most commonly taught as reversal markers, the sweep, the fair value gap, the break of structure, all fire at similar rates on swing highs that fail. One feature separates the populations cleanly: volume. Real day highs print on roughly 1.9 times the day's average volume; ordinary swings print on average volume.

Detail on the sweeps that did happen

Where real day highs did sweep a level, the median poke above it was 4.2 points, small enough to be consistent with a stop run rather than a genuine break. Which level was swept: a prior swing high 52 percent of the time, a round number 16 percent, the premarket high 13 percent, the opening range high 12 percent, the prior day high 6 percent.

Combinations carry modest information. Counting five fingerprints at once, sweep, stacked levels, FVG, break of structure and elevated volume, 64 percent of real day highs show three or more, against 52 percent of ordinary swings. Real, but far weaker than any single signal is usually credited with.

What this does and does not say

It does not say these concepts are worthless, and it does not indict discretionary traders who use them inside broader context that is not in this table. The claim is narrower: for anything mechanical, these features will not discriminate on their own, because they occur at similar rates on the highs that matter and the highs that fail. Volume will.

If you are using structure to find tops, the useful addition is not more structure. It is confirmation that anyone actually showed up. And whatever rule you settle on deserves the same treatment this one got: a control group, a real base rate, and a forward test on data the rule has never seen, which is the method we document publicly.