Skip to content

KF-R01 / FLOW QUALITY / 8 SEPTEMBER 2026

Four numbers for a book

Spread, depth, entropy and imbalance describe most of what a trader needs to know about an order book. Here is how we compute them, and how each one gets gamed.

HELM / 6 MIN READ

One row of the Flow Terminal, column by column.

Spread is the entry fee

Cross the book and you pay the spread. Everything else about a venue matters less to a taker than this one number, which is why it sits in the first column of our board. We quote it in basis points of the mid so that a 0.4 dollar gap on ETH and a 12 dollar gap on BTC can sit in the same column.

s=paskpbidm×104,m=pbid+pask2s = \frac{p_{ask} - p_{bid}}{m} \times 10^{4}, \qquad m = \frac{p_{bid} + p_{ask}}{2}
Spread in basis points of the mid.

A spread read on its own flatters a venue. One lot on each side produces the same number as fifty, and a maker who wants to look good on a screenshot can post the minimum size at the tightest price and cancel the moment anyone reaches for it. That is why the next column exists.

ETH spread, live
0 venues
  • reading the books...
Live spreads on ETH, sorted from tight to wide. The board updates every few seconds.

Depth answers how much

Depth sums the dollars resting inside a band around the mid. We report three bands, 10, 25 and 50 basis points, because the shape of the answer matters: a book with 400,000 dollars inside 10 bps and 450,000 inside 50 bps is a different animal from one that goes 40,000 then 900,000.

Db=pm(1b/104)pqp  +  pm(1+b/104)pqpD_{b} = \sum_{p \,\ge\, m(1 - b/10^{4})} p\,q_{p} \;+\; \sum_{p \,\le\, m(1 + b/10^{4})} p\,q_{p}
Notional resting within b basis points of the mid, both sides counted.

Kyle wrote the version of this that economists cite, where the depth of a market is the order flow required to move the price by one unit (Kyle, 1985). Our column is the practical cousin: the money you can move before the price moves 0.1 percent.

ETH depth within 10 bps, live
0 venues
  • reading the books...
Consolidated depth inside 10, 25 and 50 bps, and the venues contributing to it.

Entropy describes the arrangement

Two books can hold the same dollars inside 25 bps and behave nothing alike. One spreads size across forty levels. The other parks 90 percent of it on a single level that vanishes when the tape speeds up. Shannon entropy separates them.

H=ipilogpi,pi=vijvjH = -\sum_{i} p_i \log p_i, \qquad p_i = \frac{v_i}{\sum_j v_j}
Entropy of resting notional across 1 bps buckets, normalised by log N so the reading lands between 0 and 1.

Corsetti, Lafarguette and Mehl measured this on foreign exchange data and found that a more evenly distributed quote population absorbed macro news with less damage to effective spreads (ECB Working Paper 2300, 2019). The mechanism carries over to crypto venues. A book with high entropy has many hands in it, and no single cancel empties the level a taker was aiming at.

ETH book entropy, live
0 venues
  • reading the books...
Normalised entropy per venue. Higher means size is spread across levels rather than stacked on one.

Imbalance is the tilt

Imbalance compares the two sides inside a band. It runs from minus one, everything offered, to plus one, everything bid.

I=D25bidD25askD25bid+D25askI = \frac{D^{bid}_{25} - D^{ask}_{25}}{D^{bid}_{25} + D^{ask}_{25}}
Bid minus ask depth inside 25 bps, over their sum.

A single reading says little. A reading that holds one way for minutes while the price does not follow says the book is absorbing size from someone who keeps coming back. Makers watch that pattern closely, because it usually ends with the price going where the pressure was.

ETH imbalance within 25 bps, live
0 venues
  • reading the books...
Live imbalance inside 25 bps, venue by venue.

Where the four numbers lie

Three traps catch people who read these columns for the first time.

  • Quote currency. A USDT book and a USD book are not the same instrument. We convert USDT venues at the live USDT/USD mid before anything goes into the consolidated view; skip that step and a stablecoin wobble looks like an arbitrage.
  • Instrument class. A perp and a spot book trade different things. Funding and basis push a perp mid away from spot, so the board keeps a venue out of the consolidated book while its mid sits more than 25 bps from the median, and names it.
  • Phantom size. Depth counts orders, and an order is a promise, not a trade. Size that cancels the instant a taker appears counts in the column and not in the fill. Comparing the depth column with the impact column is the cheapest way to catch it.

Read together, the four numbers answer the question a trader actually asks: what does it cost me to do size here, right now, and what happens if I am not the only one asking. The rest of our research builds on those four readings.

References

  1. 01Kyle, A. S. (1985). Continuous Auctions and Insider Trading. Econometrica, 53(6), 1315-1335.
  2. 02Corsetti, G., Lafarguette, R., & Mehl, A. (2019). Fast Trading and the Virtue of Entropy: Evidence from the Foreign Exchange Market. ECB Working Paper Series No. 2300.
  3. 03Shannon, C. E. (1948). A Mathematical Theory of Communication. Bell System Technical Journal, 27(3), 379-423.