Order Flow Imbalance Beyond The Touch

Queue imbalance at the touch — comparing size at just the best bid and best offer — is a strong signal, but it only looks at the very front of each queue. A book can have a thin bid queue at the touch backed by enormous size one and two ticks below, which top-of-book imbalance completely misses. Extending the same imbalance idea across several price levels captures more of the true resting supply-and-demand picture, at the cost of needing richer market data than just the best bid and offer.

The analogy: judging a crowd by the front row only

Imagine trying to estimate how many people are waiting to get into a concert by counting only the people visible in the front row of the line. If the line curves around a corner, the front row badly understates the true crowd. Someone who walks the whole visible queue, several rows deep, gets a much better estimate of true demand — at the cost of more effort to gather. Multi-level order flow imbalance is that fuller count: instead of just the best bid and offer, it sums resting size across the first several price levels on each side.

The formula, one symbol at a time

Let $Q_k^{bid}$ and $Q_k^{ask}$ be the resting size at the $k$-th price level from the top on the bid and ask side respectively, for levels $k = 1, \ldots, K$. A common construction weights nearer levels more heavily, since a resting order two ticks away is less immediately actionable than one at the touch:

$$I_K = \frac{\sum_{k=1}^{K} w_k \left(Q_k^{bid} - Q_k^{ask}\right)}{\sum_{k=1}^{K} w_k \left(Q_k^{bid} + Q_k^{ask}\right)} ,$$

with weights $w_k$ that typically decrease as $k$ increases — level 1 (the touch) weighted most, level 5 weighted least. In plain English: this is the same buy-minus-sell-over-buy-plus-sell ratio as top-of-book queue imbalance, just computed over the whole visible depth of the book instead of only the front row, and tilted toward the levels closest to trading.

Worked example: comparing touch-only versus five-level imbalance

A stock's book shows:

Level	Bid size	Ask size
1 (touch)	500	600
2	300	300
3	200	3,000
4	400	200
5	600	100

Touch-only imbalance ($K=1$): $I_1 = \frac{500-600}{500+600} = \frac{-100}{1100} \approx -0.09$ — a nearly balanced book, slightly ask-heavy.

Unweighted five-level imbalance: total bid size $= 500+300+200+400+600 = 2{,}000$, total ask size $= 600+300+3{,}000+200+100 = 4{,}200$. $I_5 = \frac{2{,}000 - 4{,}200}{2{,}000+4{,}200} = \frac{-2{,}200}{6{,}200} \approx -0.35$.

The touch alone suggests a nearly balanced book. Looking five levels deep reveals a much heavier ask side, driven almost entirely by a 3,000-share resting order at level 3 — a large seller sitting just behind the touch that top-of-book imbalance completely missed. Whether that deeper size is "real" (genuinely resting, cancelable at any moment) or a stale, spoofed, or soon-to-be-pulled order is a separate question multi-level imbalance can't answer on its own.

Think of the transform above as a stand-in for how depth reshapes the picture: the touch alone is a thin slice, like looking at a shape from one narrow angle, while multi-level imbalance stretches the view out across more of the underlying structure — sometimes revealing an asymmetry the narrow view hid entirely.

What this means in practice

Multi-level imbalance requires full order-book depth data, not just top-of-book quotes, so it's more expensive to compute and more sensitive to data quality — resting orders several levels deep are also more likely to be cancelled before ever being hit, making them a noisier signal than touch-level size. Desks that have access to full depth typically blend multi-level imbalance with touch-level imbalance rather than replacing one with the other, since the two capture complementary information about immediate versus slightly-delayed pressure.