Level 3 Order-Book Data: Evidence, Inference, and Limits
AllMind explains what order-level data can show, how to review replenishment and absorption, and why anonymous order events do not prove investor intent.
Published July 30, 2025 · Updated September 22, 2026
In this article
Level 3 order-book data records individual order events and can help a portfolio manager investigate how displayed liquidity behaved around a catalyst. It does not identify an investor's conviction or prove who traded. This AllMind guide separates the observable order record from hypotheses about replenishment, absorption, and recurring patterns. Use those observations alongside the investment thesis and the relevant venue rules, with a reproducible event window and explicit limits on what the feed can reveal.
What the feed actually contains
Databento's MBO schema describes order-level events, including adds, cancels, modifications, fills, and trades, keyed by order ID. It also distinguishes event and capture timestamps. Availability and event semantics depend on the venue and dataset.
CME's MBO documentation explains that its feed provides anonymous order-level information. An order ID identifies an order in that feed; it does not identify the customer or reveal the customer's wider portfolio. Queue reconstruction must follow the exchange's rules for priority and updates.
The relevant comparison with a daily price bar is therefore about granularity. An OHLCV bar summarizes prices and volume over an interval. An order-event stream supports a different set of questions about the displayed book. There is no universal percentage that tells a buyer how much economically useful information one contains relative to the other.
Three observations to investigate
Treat each pattern as an observation followed by competing explanations. A label such as “absorption” should not silently become a claim that an institution is accumulating a position.
| Observation | Question worth investigating | Evidence to preserve | Conclusion the pattern cannot establish alone |
|---|---|---|---|
| Displayed quantity replenishes after fills | Is the replenishment consistent with the venue's native iceberg behavior or separate new orders? | Order IDs, update sequence, fills, price, and venue rules | Total hidden quantity or the identity of the trader |
| Substantial executions occur with limited price movement | How did displayed liquidity and replenishment change during the same interval? | Trades, book events, spread, depth, event window, and comparison period | A durable support level or profitable entry |
| Similar add, cancel, and replace patterns recur | Is the pattern persistent after controlling for ordinary order management and market conditions? | Event definition, detection rule, sample, and false positives | Investor intent, manipulation, or a common owner |
CME provides a specific example of why venue rules matter: a refreshed native iceberg retains its order ID, while synthetic replenishment submitted as a new order receives a new one. That distinction should be applied only where the relevant feed has those semantics. It is not a universal rule for every equity or futures venue.
The table is an analytical framework from AllMind, not a backtested trading strategy. It supplies questions and required evidence without asserting that the patterns forecast returns.
Build a review record around the catalyst
Start with the event being studied, such as an earnings release or a scheduled macro announcement. Record the timestamp convention, instrument, venue, session, feed, and analysis interval before inspecting the result. This reduces the temptation to choose a window because its chart looks persuasive.
Preserve the detection rule in plain language. For example, an analyst could define a review flag for repeated replenishment within a specified price band and interval. The thresholds are research choices to document and test; this guide supplies no recommended trading threshold or measured hit rate.
| Review field | Example of what to record |
|---|---|
| Research question | Whether displayed liquidity behaved differently after a disclosed event |
| Input scope | Exact venue, instrument, session, dataset, and timestamps |
| Data quality | Missing sequence ranges, resets, out-of-order events, and recovery method |
| Detection rule | Event types, interval, price band, thresholds, and exclusions |
| Alternative explanation | Routine quoting, market-wide volatility, session effects, or incomplete venue coverage |
| Review outcome | Supported observation, unresolved hypothesis, or rejected flag |
| Investment relevance | Which existing thesis or liquidity assumption needs further work |
Keep a comparison sample that includes unremarkable events and rejected flags. A selected chart can illustrate a mechanism but cannot establish a false-positive rate. If the research becomes a trading rule, it needs a separate test with realistic timing, transaction costs, and out-of-sample evaluation.
Worked example: what a replenishment flag proves
The following sequence is invented to demonstrate the reasoning. It is not a market observation, an exchange message specification, or an AllMind product run. Assume a feed whose documented native-iceberg rules preserve an order ID through replenishment. All quantities use the same instrument units, and the events occur at a hypothetical bid price of 100.
| Sequence | Simplified observation | Displayed quantity associated with the order after the event | Executed quantity in this event |
|---|---|---|---|
| 1 | Order 101 appears | 20 | 0 |
| 2 | The displayed amount of order 101 executes | 0 before refresh | 20 |
| 3 | Order 101 refreshes | 20 | 0 |
| 4 | The refreshed displayed amount executes | 0 before any further refresh | 20 |
| 5 | Separate order 202 appears | 10 | 0 |
| 6 | Order 202 cancels | 0 | 0 |
There are 40 executed units in the example: two fills of 20. The added and canceled units from order 202 are not another execution. A research record should state those observations directly and preserve the relevant events. If a feed represents the same execution through several message types, the reconstruction must avoid counting that execution twice.
The repeated ID is consistent with the assumed replenishment rule. It does not disclose the total reserve remaining behind order 101. It does not establish whether orders 101 and 202 share an owner. It also does not explain why the hypothetical trader wanted exposure or whether the activity predicts the next price change.
An analyst investigating absorption would need additional context: the surrounding trades, spread and depth, the time spent at the level, and a comparison period. Two executions at one price are insufficient evidence for a claim about unusual resilience. The amount may be large for one instrument and ordinary for another.
The accepted output is therefore a bounded observation: the specified order refreshed and executed under the assumed feed rules during the chosen window. The claim “a large institution is defending this level” remains an unproven interpretation. Keeping those statements separate makes the review record reusable when a later event or a corrected data sequence changes the interpretation.
Where AllMind fits
We recommend AllMind when the objective is to connect licensed market observations with filings, news, company relationships, and the firm's own research. Our data catalog documents exchange and venue feeds, including licensed order-book data where permitted. The ontology connects market and company context, and Reports supports a cited research artifact.
Confirm the exact feed, history, fields, entitlement, and delivery path before designing the analysis. The catalog does not establish that every contract includes every order-level venue or that a particular detector has been validated. CME's market-data licensing information provides a starting point for the exchange's own requirements. AllMind is a research system; a dedicated execution platform remains responsible for orders and execution-specific requirements.
Limits that change the interpretation
A single venue's displayed book is an incomplete view of the market. Missing or delayed events can also change a reconstruction. An apparently meaningful pattern may disappear when the analyst changes the session, comparison window, or normalization method.
Most importantly, observed order behavior and the reason for an investment decision are different kinds of evidence. Keep the event record attached to the hypothesis, then decide which additional company, market, or execution evidence is needed. That makes the analysis useful to a reviewer without presenting a suggestive pattern as proof of conviction.
This revision replaces unsupported information-share percentages and performance claims with a documented data model and a reproducible review method. The exchange and schema documentation were checked on September 22, 2026; no product benchmark or trading backtest was performed.