Displayed · traded · replenished · realized
ES Liquidity Pockets: Measuring the Order Book Without Folklore
A large ES bid can disappear before a sell order reaches it. The same price may later absorb repeated trades, or the market may move through it without hesitation. A screenshot cannot tell those states apart, and none proves a permanent “institutional liquidity pocket.”
- Book
- Sequenced messages
- Trades
- Executed quantity
- Response
- Price + replenishment
- Actor identity
- Not inferred
Do not ask bar data to answer order-book questions
Match the Claim to the Data Layer
CME’s historical catalog separates trades, top-of-book, depth and market-by-order products. Each supports different statements. Choose the minimum sufficient layer and document feed depth, sequencing, timestamps and recovery behavior.
| Data layer | What can be observed | What remains unknown | Eligible question |
|---|---|---|---|
| Bars | Open, high, low, close, aggregate volume | Spread, message order, queue and depth | Where did price and volume travel? |
| Trades | Transaction price, size, time, corrections | Unexecuted displayed interest | Where and how much traded? |
| Top of book | Best bid/ask, quantities and changes | Capacity beyond the inside | How did inside width and size evolve? |
| Market depth | Displayed levels through the feed’s depth | Individual order identity where aggregated | How much displayed capacity sat near price? |
| Market by order | Order-level add, modify, cancel and execute sequence | Trader identity, private intent and hidden decisions | How did the visible queue evolve? |
The CME futures and options data catalog describes data resolutions from top-of-book to full depth and market-by-order. CME Globex is a central electronic marketplace where participants can view the book and enter orders; a vendor’s reconstructed view may still differ by entitlement, latency and processing.
One dropped message can manufacture a pocket
Reconstruct a Valid Book Before Measuring It
Order-book analysis is a state-reconstruction problem. A snapshot and subsequent incremental messages must align. Sequence gaps, duplicate messages, late corrections or timestamp drift can leave phantom size in the book.
Identify
Resolve venue, product, dated contract, channel, schema and session.
Seed
Load an authoritative snapshot or start at a documented recovery point.
Sequence
Apply messages in exchange sequence; quarantine gaps and duplicates.
Join
Synchronize trades and book events using source timestamps and sequence rules.
Validate
Reject crossed, negative, impossible or stale states before analysis.
If a feed supplies both, retain both and label them. Local receipt timestamps measure the observer’s path and cannot substitute for the authoritative event timestamp required to sequence the source book.
Describe a state, then its response
Measure Pockets as Timestamped Conditional Distributions
A candidate pocket is a repeatable record of displayed capacity, trading and response inside a defined state—not a horizontal line copied forward forever. Normalize distance in ticks or volatility units and report the sample and expiry.
Width
Time-weighted spread
Measure inside spread in ticks and time spent locked, crossed or unavailable. Quote counts alone overweight message bursts.
Capacity
Depth and sweep cost
Sum displayed quantity through fixed tick bands and estimate the cost to execute declared sizes on each side.
Durability
Age and replenishment
Track dwell time, executed fraction, cancellations and quantity returning after depletion. State the matching assumptions.
Response
Impact and recovery
Measure signed price response after an aggressive quantity, the time to refill and whether spread/depth normalize.
Candidate pocket record
Store the exact interval, contract, price-relative location, book depth K, median and tail displayed quantity, time-weighted spread, executed quantity, cancel-to-add measures, replenishment after depletion, size-conditioned sweep cost and subsequent price response. Attach event, volatility, session and roll labels.
Define the candidate on a training sample and test on later observations. If the price-relative location moves when volatility or index level changes, a fixed historical level should expire. If current depth is outside the trained distribution, the pocket is unscorable rather than “still valid.”
The same trace supports multiple stories
Test Rival Explanations Before Naming Behavior
Displayed size that cancels is not proof of manipulation. CME Rule 575 prohibits specified disruptive practices and turns on conduct and intent; public book data alone does not reveal the owner, purpose or full order context. Use neutral event descriptions.
| Observed trace | Possible explanation A | Possible explanation B | Safe label |
|---|---|---|---|
| Large quantity cancels near touch | Risk or quote update | Strategic cancellation | Displayed quantity canceled before execution |
| Repeated replenishment | Multiple independent orders | Reserve/algorithmic execution | Visible quantity replenished after trades |
| High volume, falling depth | Fast two-sided risk transfer | Liquidity withdrawal during repricing | Trade activity rose while displayed depth fell |
| Price stalls at a level | Balanced flow | Large passive interest | Conditional price response was small |
CME Rule 575 and its advisory provide the official disruptive-practices framework. An educational chart label is not a regulatory determination.
Translate evidence into conservative permissions
Use the Book as an Execution Gate, Not a Direction Oracle
Declare size, urgency, order type, maximum spread, minimum near-touch depth and tail sweep cost before looking at the signal. The current state must match the validated sample. Directional forecasts require separate evidence.
Go
State represented
Sequence is valid, current spread/depth fit the sample and modeled capacity covers the order with stress.
Reduce
Capacity supports less
A preregistered lower quantity passes while the requested size fails the sweep or impact limit.
Wait
Transient instability
Replenishment, spread or event state is changing too quickly for the model’s observation horizon.
Reject
Evidence incomplete
Sequence gaps, stale quotes, out-of-distribution depth or unknown contract state make the order unscorable.
- Quoted, traded and experienced liquidity remain separate.
- All metrics are conditioned on order size and time horizon.
- Event, holiday and roll states are attached before measurement.
- Price response uses forward-only windows and reports both sides.
- Actor identity and intent are not inferred from public order messages.
- Realized fills calibrate, and can invalidate, the pretrade capacity model.
No original ES liquidity-pocket location, imbalance threshold, queue signal or after-cost execution advantage is reported. This page defines observable states, candidate metrics and rejection rules.
Sources, methods and editorial disclosure — reviewed August 28, 2026
- CME Globex for central electronic-book and platform context.
- CME futures and options data catalog for trade, top-of-book, depth and market-by-order data distinctions.
- CME DataMine Market Depth FAQ for historical book-reconstruction scope.
- CME DataMine Equity Execution Statistics FAQ for width, depth, fill and slippage measurement categories.
- CME Rule 575 advisory for official disruptive-practices language and the boundary against casual intent attribution.
Sources were reviewed August 28, 2026. The page separates official market/data structure from proposed metrics and reports no proprietary empirical result.