Pullback to VWAP or value

It is 10:50 ET (22:50 ICT, winter). ES opened at 6,000, drove to 6,028 by 10:30 and has stayed above its RTH VWAP all morning. Now it is falling back toward VWAP near 6,015.75. One trader sees the trend ending and wants to sell. Another bought 6,027.75 at the high and is now 12 points down. A third has a written plan for exactly this moment. This lesson is that plan: what it checks, what it costs to be wrong, and how much of it anyone has actually measured.

A pullback (nhịp hồi) is a move against the session’s direction so far that stops before that direction is undone Definition. Here the pullback is to VWAP (giá bình quân theo khối lượng), the average price traded since the 09:30 ET open (M3.07), or, for longs, to the developing value-area low (VAL).

Chart first: one pullback on a one-sided morning

The setup, written as rules

Every part below is something you can check on the chart as it happens. That is the point: a rule you can check is a rule you can later score.

• Context: a one-sided session so far. Price has held above VWAP since the open or since a reclaim (M4.04), and VWAP is rising Context. “No one-sided context, no setup” is a house rule House rule. Note what this does not say: it does not say “today is a trend day”. A trend day is a label you can only give at the close, and in the public NQ data, forecasting the day type by 11:00 ET from the open, the IB and the location showed no clear edge (inconclusive, n = 625) Lore. In a balanced session price crosses VWAP again and again, and buying each cross mostly pays commissions.
• Trigger: price reaches VWAP (or the developing VAL) and the selling stops. Written so you can check it: a 5-minute bar trades at or through VWAP and closes back above it; the pullback’s delta shrinks bar by bar. These are observable Definition. That they predict the bounce is not shown: “responsive buyers at the edge reverse price” and “absorption predicts reversal” were both tested on NQ and not supported Lore, and footprint signals in general widen the move both ways rather than pick its direction PlausibleOne trader's study (M5.14). Treat the trigger as a filter that keeps you out of the pullbacks that never pause, not as a forecast.
• Invalidation: acceptance back below VWAP. Closes holding under the line (time, closes and volume, M4.01) mean the morning’s description has changed Definition. The stop goes where that would be true, not at a round number of points.
• Target: the prior leg extreme. It is a price the session has already traded, so it is a sane first objective House rule. It is not a level price “wants” to reach: in our lab’s tests, pre-market levels as a group did no better than random prices. Anything past the extreme is a trailing decision (M5.11), not part of the plan.

Reward to risk is arithmetic, not an edge

The appeal of the pullback is the location. Buying near VWAP puts the stop close (just under the line) and the target far (the prior extreme). Buying the breakout at the extreme puts the same invalidation far away and leaves no target the session has already traded.

The arithmetic Definition: break-even win rate before costs = risk ÷ (risk + reward). A trade risking 4 points to make 11 breaks even if it wins 4 ÷ 15 = 26.7% of the time; costs push that up a little (M6.08). A trade risking 15 points to make 8.25 needs 15 ÷ 23.25 = 64.5%.

That is a real difference in what each trade needs. It says nothing about what each trade gets. A close stop under VWAP gets hit more often than a wide one, and the pullback entry only fills on the days price comes back, which includes the days the one-sided morning is ending. The ratio tells you the hurdle; only a logged hit rate tells you whether you clear it.

First touch, later touches: a house rule

Myth / hypothesis: “The first pullback to VWAP in a trend is the highest-probability entry. Each later touch is weaker because the defenders’ orders get used up, so by the third or fourth touch VWAP is about to break” Lore. A sensible-sounding inventory story with no test behind it for VWAP: not yet measured. The nearest test cuts the other way: in the public NQ study, levels in general were not more likely to break the more often they were tested (evidence box below).

What survives is a risk rule, and it is worth keeping as one House rule: full planned size on the first pullback of a one-sided session, half on the second, and skip the third unless your own log says otherwise. The reason is not that later touches are known to fail. It is that each extra touch means the session has spent longer near its average, which is what balance looks like, and the one-sided context the setup needs is getting harder to claim. Write the rule down, log every touch, and let your numbers decide (drill below).

What the data says

Worked example: long 6,016.75 vs the chase at 6,027.75

Synthetic session, not a real date (the figure above). ES, 5-minute chart, tick 0.25 points, $50 per point per contract.

Context: price has held above VWAP since the open and VWAP is rising; it drove from 6,000 to 6,028 by 10:30 ET (22:30 ICT, winter). Check ✓. At 10:50 ET price pulls back to VWAP, about 6,015.75. The 10:50 5-minute bar trades down to 6,014.75, one tick under the line, and closes at 6,016.75, back above it; the pullback’s delta has shrunk over the last three bars. Trigger ✓.

• Entry: long 1 ES at 6,016.75, on the trigger bar’s close.
• Stop: 6,012.75. Two closes under VWAP would mean acceptance below it; a stop 3 points under the line and 2 under the bar’s low is where the plan is plainly wrong. Risk 4 points, $200 before costs.
• Target: 6,027.75, one tick under the 6,028 leg extreme. Reward 11 points, $550. Reward to risk 2.75; break-even win rate 26.7% before costs.

In the figure, price reaches 6,028 by 11:40 and the target fills.

Now the chase. A trader bought 6,027.75 at 10:30, at the leg extreme. With the same invalidation (acceptance under VWAP, 6,012.75) the risk is 15 points, $750. There is no traded price above to aim for, so the target is a guess, say 6,036.00 (8.25 points, $412.50). Break-even win rate 64.5%. Same session, same direction: the chaser needs to be right more than twice as often as the pullback trader to break even. That is what location buys. It does not tell you which of the two traders, over a hundred sessions, makes money.

The mirror and the failure

The short mirror. Flip the figure to the short side. Price has held below a falling VWAP all morning; it rallies back to the line, a 5-minute bar trades through it and closes back below; short on that close, stop above VWAP where acceptance above would be true, target the morning’s low. Same rules, same grades.

The failure. Same morning, another path. At 10:50 the trigger bar closes back above VWAP and you are long 6,016.75. The next two 5-minute bars close at 6,014.00 and 6,012.50: acceptance below VWAP, stop hit at 6,012.75, $200 plus costs. By 11:30 price is rotating around VWAP and the “one-sided morning” was only a morning. Nothing in the trigger could have told you which path you were on. Until the setup is scored against a baseline, assume this ending is common and size for it.

Prop-firm note. A limit order resting at VWAP fills in a simulated account whenever price touches it; live, you stand behind the queue and a touch-and-go often leaves you unfilled (SIM fills vs live fills, MKT2.03). A pullback plan that only works with touch fills will look better in SIM than it is. Prop firms’ prohibited-conduct rules can also cover strategies that exploit simulated fills (Prop Firm course, PROP1.04). Your stop is the number the firm’s loss limits see; check your firm’s current rules.

Common mistake: Treating “price is above VWAP” as a forecast that today is a trend day. The public NQ result describes trend days after the fact; it does not identify them in the morning. The context check tells you which plan is allowed, not that it will work.

Common mistake: Reading the 2.75 reward to risk as an edge. It is the hurdle (26.7% before costs), not the hit rate. A setup with a great ratio and a 20% hit rate loses money.

Common mistake: Adding size because “VWAP has held three times”. That treats an untested story as fact in the risky direction. The house rule runs the other way: later touches get less size, not more.

Drill: log your VWAP touches

For 20 sessions, on replay or live sim, log every pullback to RTH VWAP in a session that was one-sided at the time of the touch (price above a rising VWAP since the open or a reclaim, or the mirror).

1. Write the rule first: context, trigger bar, stop (acceptance back through VWAP), target (the prior leg extreme). Do not change it during the 20 sessions.
2. For each touch log: date, time ET, touch number in the session (1st, 2nd, 3rd), whether the trigger fired, and the outcome: target first, stop first, or neither by 12:00 ET.
3. Compute the hit rate for first touches and for later touches, and compare each with the break-even rate of its own ratio.

What we don’t know yet

• Whether pullbacks to VWAP in a one-sided session continue more often than pullbacks to a random price at the same distance: not yet measured.
• Whether first touches beat later touches: not yet measured. The general level-testing result points slightly the other way.
• Whether the developing VAL works better or worse than VWAP as the pullback line: not yet measured.
• The open leads in the module are different trades and are graded Lore until a forward test: a VWAP ±2σ stretch with delta divergence at a level reversing (a fade, the opposite of this setup), a counter-direction new 1-hour extreme, delta divergence into a level’s first touch, and the multi-day balance breakout (M5.07).

Recap:

• The setup needs a session that has been one-sided so far, checked in real time; “trend day” is a label you only get at the close.
• Trigger: a bar through VWAP that closes back on the session’s side; invalidation: acceptance back through VWAP. The trigger is a filter, and its predictive value is not shown.
• Location changes the break-even win rate (26.7% vs 64.5% in the example), not the hit rate. A better ratio is not an edge.
• “Size down on later touches” is a house risk rule; as a market claim it is graded Lore, and on NQ, levels in general did not break more often with each test.

Further reading:

• Volume-weighted average price: VWAP mechanics and its use as an execution benchmark.
• Berkowitz, Logue and Noser, “The Total Cost of Transactions on the NYSE”, Journal of Finance (1988): the paper that made VWAP a standard benchmark for judging execution.
• Order Flow, Measured (public practitioner study): the NQ tests of VWAP band reversion, trend days on one side of VWAP and repeated level tests.
• Our lab: the review of that study’s VWAP claims (band reversion Lore, the one-side-of-VWAP result descriptive only) and the level placebo study (721 sessions: pre-market levels no better than random prices).

This lesson connects to: M4.03 (one-sided days) · M4.04 (VWAP as a reference and the reclaim) · M5.11 (trend-day continuation) · M5.14 (entry patterns: limit vs confirmation at the pullback). Next, M5.09 asks the same question of a footprint signal: does a continuation trigger add anything the location does not.

Educational content only — not financial advice. Futures trading involves substantial risk of loss.