Imbalance & stacking

It is 10:12 ET (22:12 ICT, winter). NQ pushes through the overnight high at 18,028.00 and the 5-minute footprint lights up three ask imbalances in a row near the top of the bar. Many footprint courses call this a bullish stack and say the zone it leaves behind will act as support on the way back.

Most traders who discover imbalances make the same trade for a month: they buy the highlighted cells and lose about as often as they win. This lesson is the mechanics of what the footprint is actually highlighting, what it tells you for certain, and what the public tests say about the support-and-continuation story.

Key terms

Imbalance (mất cân bằng): on a footprint, the ask volume at price P compared with the bid volume one tick lower, at P − 1 Definition. When the ask side is at least a set multiple of that diagonal bid, the platform flags a buying imbalance. The mirror: the bid at P against the ask one tick higher, at P + 1, flags a selling imbalance.

Threshold (ngưỡng): the multiple that counts as “imbalanced”, plus any minimum-volume filter, so that 4 contracts against 1 does not light up Definition. It is a setting. ATAS X ships its built-in imbalance rate at 150%; the OrderflowSuite stack study defaults to 300%; 300% and 400% are common community choices. None of these numbers was derived from a test.

Stacked imbalance (xếp chồng): several consecutive prices in one bar that all carry an imbalance on the same side Definition. Three is the usual minimum; some traders use two, some four. That count is a convention too.

Chart first: the stack

Price      Bid  x  Ask      Diagonal read (ask at P vs bid at P−1)
18,036     90   x  350  ┐  ask 350 vs bid 80 (18,035)  = 4.4×  imbalance
18,035     80   x  310   │  ask 310 vs bid 70 (18,034)  = 4.4×  imbalance
18,034     70   x  280   │  ask 280 vs bid 60 (18,033)  = 4.7×  imbalance
18,033     60   x  240   │  ask 240 vs bid 200 (18,032) = 1.2×  no imbalance
18,032    200   x  180  ┘  (two-sided trade below: the push started here)

Rows are 1 point apart for readability; on a real 1-tick NQ footprint they are 0.25 points apart. Read the diagonals: 18,034, 18,035 and 18,036 each clear a 300% threshold, three in a row, so this bar holds a three-level buying stack. 18,033 does not: 240 against 200 is two-sided trade.

How it works

Why the diagonal

The comparison follows the spread. When the best offer is 18,029.00, the best bid is 18,028.75. An aggressive buyer at that moment trades at 18,029.00 and lands in the ask column; an aggressive seller at the same moment trades at 18,028.75 and lands in the bid column one row lower. So the two aggressors competing across the same spread sit one row apart: ask at P against bid at P − 1.

A same-price comparison pits buyers who lifted 18,029.00 against sellers who hit 18,029.00 at a different moment, when the whole spread sat one tick higher. It mixes two market states. That is why most footprint platforms, and the OrderflowSuite study, compare diagonally Definition. Some platforms offer a horizontal comparison as an option, and the two definitions flag different cells, so check which one yours uses before you compare notes with anyone.

What a stack describes

One 5× print can be a single large order. Three or four consecutive imbalanced prices say that aggressive buyers kept paying up through several prices within one bar. Whether that is one participant working an order or many algorithms chasing the same move, the footprint cannot tell you. What it records for certain is narrow: in that bar, at those prices, aggressive buying heavily outweighed aggressive selling across the spread Definition. That is a description of what already happened, in the same bar. It is the footprint cousin of delta (M3.04), and like delta it explains the bar’s move rather than forecasting the next one.

Myth / hypothesis: “A stack becomes support (or resistance) when price comes back to it, and a single strong imbalance predicts continuation.” Lore Both are widely taught, and both were tested on NQ in a public practitioner study and not supported (box below). Retests of three-plus stacks held no more often than comparable prices without a stack, and single imbalances were not followed by continuation more often than bars without them. Treat both as hypotheses you could test, not rules.

Myth / hypothesis: “A bullish stack that stalls into a prior high traps its buyers, and their forced selling drives the reversal.” Lore The trapped-trader story is coherent, but the closest public test (heavy buying at a high that fails to extend, then reverses) was inconclusive on NQ: +1.9 percentage points, 95% CI −0.4 to +4.3, n = 11,846. The “mapped high” part does not rescue it either: in our lab, pre-market level families were touched, respected and stalled at the placebo rate (QQQ/SPY as the proxy for NQ/ES, 721 sessions).

Why it is still worth knowing

You still need the vocabulary. A stack is the clearest footprint picture of initiative: aggressive traders paying up through several prices. It tells you where a move started, which is where a plan built on that move is plainly wrong if price returns and stays there. It helps you describe the auction (M1.03) without guessing intent. And it is exactly the kind of claim you will learn to test in EV1.04: write the rule, pick a baseline, then decide.

Worked example: the stack that broke out, then the stack at the top

Synthetic session, built for this lesson, not a real date. NQ, 5-minute footprint, tick 0.25 points, $5 per tick per contract. Rows quoted in whole points for readability.

At 10:12 ET a bar breaks above the overnight high at 18,028.00. Its footprint shows a three-level buying stack at 18,034–18,036, each ask roughly 4× the diagonal bid. Price reaches 18,040.00.

The next bar’s top prints three consecutive selling imbalances at 18,040–18,042: the bid at each price is 4–6× the ask one tick above. Now you have two descriptions, not two signals. Aggressive buyers paid up through 18,034–18,036 during the break. Aggressive sellers hit bids hard at 18,040–18,042 at the high. Neither stack says which side wins next.

You do not buy the first stack or sell the second. You write the plan:

• If a 5-minute bar closes above 18,042.00 with positive delta, the selling at the high has been traded through and the break is being accepted. Long 1 contract at 18,042.25.
• Stop 18,035.25, under the middle of the buying stack: 28 ticks, $140 per contract before costs.
• Invalid if a 5-minute bar closes back below 18,028.00: the break failed and you stay flat. That describes failed acceptance; it is not a forecast that the buyers get forced out.

In this synthetic session price chops between 18,036 and 18,041 for 25 minutes and never closes above 18,042.00. No trade. That is one ending, not the expected one.

The mirror and the failure

Short side. A selling stack forms as price breaks below a prior low: the bid at each of three prices is 4× or more the ask one tick above. The plan mirrors: short only on a close below the next selling imbalance with negative delta, stop above the middle of the stack, invalid on a close back above the broken low.

The failure. The figure shows the common one. The stack’s zone is retested and bounces once, which is what the support story predicts. On the second retest, six minutes later, price goes straight through it without pausing. A trader who bought the first retest with a stop under the zone took a small win if they exited early and a full stop if they held. The public test found retests of stacks held no more often than comparable prices without a stack, so neither ending is the stack’s doing.

Do it on the platform

Common mistakes

Common mistake: Treating a stack as an edge because it “looks strong”. A stack records that aggression happened. In the public NQ test, retests of stacks held at the baseline rate. Strength of the print and edge of the signal are different questions.

Common mistake: Tuning the threshold until the signals look good. Raising the ratio from 300% to 400% changes which cells light up, not whether the idea works; the threshold check in the box was inconclusive. A threshold chosen by looking at past charts is fitted to those charts (M3.13 covers calibration).

Common mistake: Reading the same stack three ways and calling it context. “Inside balance it is rotation, at the edge it is a breakout, at a high it is a trap” sounds like analysis, but none of the three branches has been tested, and the levels themselves matched placebo in our lab. Write the if-then plan and its invalidation instead.

Drill: score your own stacks against a baseline

Over five sessions of NQ replay on a 5-minute footprint, mark every buying or selling stack of three or more levels at 300%. Before you look at what happens next, write your outcome rule: for example, “held” means price returns to the stack’s zone within 30 bars and then moves 8 ticks away from it on the stack’s side before 8 ticks through it.

Then mark the same number of returns to ordinary prices from the same bars (no stack) and score them with the same rule.

1. Log each stack: time (ET), side, prices, number of levels, the largest ratio.
2. Score every return as held, broke or never returned.
3. Compare the held rate of stack zones with the held rate of the ordinary prices.

What we don’t know yet

• Our lab’s own replication of the public stack and imbalance results on NQ is planned but needs an NQ tick export: not yet measured.
• Whether stacks help as a stop-placement tool (smaller losses at the same win rate) rather than a direction signal: not yet measured.
• Whether the horizontal and diagonal definitions give different results on NQ: not yet measured.

Recap

• An imbalance compares the ask at P with the bid at P − 1 (and the bid at P with the ask at P + 1), because those are the two aggressors trading across the same spread.
• The ratio threshold and the minimum stack count are settings; 150%, 300% and 400% are all conventions.
• A stack describes sustained aggression in one bar. Stacks as support/resistance on a retest, and single imbalances as continuation signals, are both graded Lore on NQ: tested in a public study, no edge shown.
• Use a stack to describe the auction and to place your invalidation, and write an if-then plan instead of trading the cells.

Further reading

• Order Flow, Measured: All claims. A public, pre-registered practitioner study on NQ, 2018–2026. Read the imbalance rows (FP-06, FP-07, FP-08) and the trapped-buyer row (DA-06).
• ATAS Knowledge Base: Stacked Imbalances. The platform’s own description of a diagonal stack indicator.
• Cont, Kukanov and Stoikov (2014), “The Price Impact of Order Book Events”, Journal of Financial Econometrics 12(1). Order-flow imbalance explains price changes over the same interval; that is explanation, not forecast (EV3.03).
• Our lab: no NQ footprint test of its own yet; the pre-market level placebo study found mapped levels touched and respected at the placebo rate.

Next: M3.04 turns the bid/ask split into one number per bar, delta, and shows the same pattern: tied to the bar it describes, not to the next one.

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