The "Liquidity Grab" Fade, Measured: a Real Pattern That Loses a Third of a Pip Per Trade

If you have spent ten minutes on trading TikTok, you know this setup. London opens, price spikes through the overnight high, then slams back inside the range. The gurus call it a liquidity grab: the spike ran the stops sitting just above the level, the flow is spent, and price should now revert. So you fade it.

Unlike a lot of guru material, this one has a respectable pedigree. Carol Osler's New York Fed research documented that stop-loss orders cluster just beyond visible levels, and that running them produces exactly this spike-and- revert signature. The overnight high is the most universally watched level in FX, and the London open concentrates enough flow to reach the stops behind it. The mechanism is real. Nobody selling the pattern, however, quotes a number for what fading it actually pays.

So before writing a single line of trading-robot code, we measured it.

The setup, pinned down

Vague patterns produce unfalsifiable backtests, so we fixed every definition in advance:

• Range: the 03:00 to 07:00 UTC pre-London range on GBP/USD.
• Event: the first attempt of the day where price closes above the range high and then closes back inside within a few 15-minute bars. First attempt only; if it does not fail cleanly, there is no trade that day.
• The fade: short the failure, take-profit at the opposite end of the range, stop at the failed extreme, capped at two hours.
• Data: real Dukascopy tick data, 2019 to 2021, with the real spread and commission. That cost stack averages about 1.6 pips per round turn.

What 400 fades actually pay

Under the strict definition the hypothesis actually claims, we measured 400 trades, about 133 per year:

The uncomfortable part is the first row of results: the pattern IS real. Price genuinely drifts back after a failed break, by about 1.21 pips on average. The gurus are describing something that exists. It just costs 1.6 pips to collect, so you pay the market roughly a third of a pip for every trade, forever. And the yearly breakdown suggests the gross drift itself was fading as the years passed.

The definition-shopping trap

Our first pass at this measurement said something friendlier. Counting every outside close that later returned to the range, and reading the drift on a flat two-hour horizon, produced 502 events paying +2.07 pips gross, about 1.3 times the cost stack. That version of the strategy looked alive.

But that permissive definition counts extended breakouts that eventually drift back, which is not what the hypothesis claims. The claim is stop-run failures: the FIRST attempt of the day, failing within a few bars, traded with the pattern's own exit structure. Keeping the friendly definition because it survives would have been choosing the reading that flatters the data we can see, and paying for it on the data we cannot. Under every defensible reading, the harvestable edge sits at or below the cost stack.

Rejected at the measurement stage. No robot was built, and our untouched out-of-sample years stayed untouched, because the arithmetic was already final.

The coherent picture

This result pairs with one from the other side of the same event. Our Asian range breakout strategy, which FOLLOWS close-confirmed breaks on USD/JPY, held up out of sample and is forward testing on a demo account. Put the two findings together and the joint lesson is tidy: session-range breaks carry information on average, and their failures are noise, not fuel. The break is worth following; the failed break, after costs, is worth nothing.

That is not the lesson anyone sells, but it is the one the tick data pays for.

Backtests and measurements, not live results. Not financial advice.