Time & Sales

M3.09Orderflow ToolsPractitionerDefinition≈ 15 min

Learning objectives

Read the tape: time, price, size and aggressor side of each print, and the pace at which prints arrive.
Explain why one aggressive order can show as several small prints or one large print, depending on the feed's aggregation setting.
Define "large" by percentiles of your own session and time window, not by fixed contract tiers.
Read the evidence grade of "large prints predict direction" and use the tape to describe an event at a level, not to forecast it.

It is 10:12 ET (22:12 ICT, winter). NQ has drifted down into 18,030.00, a level on your map. The tape has been quiet for minutes: one to four contracts at a time, two-sided. Then four sell prints arrive in three seconds, 34, 51, 28 and 44 contracts, each one a tick lower. Price stops at 18,029.25 and the tape goes quiet again.

Was that an institution showing its hand? Is the level broken, or defended? This lesson is about what the tape can tell you for certain in that moment, and what it cannot.

Key terms

Time & Sales, “the tape” (băng giao dịch): the list of every trade, with its time, price, size and aggressor side (sold at the bid or bought at the ask) Definition. Footprint, delta, CVD and big-trade bubbles are all computed from it. The DOM and the heatmap are not: they show resting orders, the advertised side, while the tape shows what actually traded.

Print (lệnh khớp): one line of the tape, one fill or one group of fills, depending on how your feed aggregates them Definition.

Tape speed (tốc độ khớp lệnh): how many prints or contracts trade per second Definition. It describes pace. Whether a change of pace forecasts anything is a separate question (below).

Sweep (quét lệnh): one aggressive order that fills at three or more prices in the same match event Definition. M3.11 covers sweeps in full.

Term you need: percentile. The 99th percentile (p99) of print size is the size that 99% of prints are smaller than. In a window of 4,000 prints, a p99 filter shows about 40 of them.

Chart first: how big is a big print?

ANIMATED FIGURE

Print sizes in one half hour: a fixed tier vs percentiles

Definition

WHAT TO SEEAlmost every print is one to four contracts; the tail is long and thin. Move the floor: a fixed 20-lot tier keeps about 2.7% of this sample, near its 97th percentile. The same 20 lots would sit at a different percentile in a quieter or busier window, which is why this lesson sizes prints by percentile. Synthetic sample of 9,992 NQ-like prints; the real NQ distribution by time of day is not yet measured in our lab.

Synthetic data

How it works

Reading a print

Each line answers four questions: when, at what price, how many, and who was in a hurry. A print at the ask means the buyer crossed the spread, so the buyer was the aggressor; a print at the bid means the seller was. That aggressor side is what delta adds up (M3.04).

Two cautions before you trust the side column. Your platform may infer the side from prices rather than read the exchange’s own aggressor flag, and on a fast tape the two can disagree. Whether ATAS delta matches the exchange flag on NQ is not yet measured in our lab. And the tape only shows trades: an order that rests and is cancelled never appears, which is why the DOM (M3.08) and the tape answer different questions.

One order, many prints: aggregation

An aggressive buy of 65 contracts that lifts 12, 15, 18 and 20 contracts at four prices is one decision. On an unaggregated feed it prints as four lines with the same timestamp; on an aggregated feed it prints as one line of 65 Definition. The exchange reports a match event as a summary plus its fills, and platforms choose which to show.

This matters for every size filter. With a 30-lot filter, the unaggregated feed shows nothing and the aggregated feed shows one big print. An aggregated sweep through four prices can also look like one large print absorbed at a single price. Before you read anything into size, know which setting your tape and your big-trade study use. MKT1.07 covers market data and aggregation in depth.

Size: percentiles, not tiers

Trade sizes are heavy-tailed: most prints are tiny and a few are very large. This is measured for US stocks (Gabaix and co-authors, 2003) MeasuredOther market, and the NQ tape shows the same shape on any screen. Because the tail is long, an average or a fixed cut-off says little.

Traders often quote fixed tiers such as “large is 20–49 NQ contracts, very large is 50 or more, roughly half that overnight”. These are conventions from one practitioner’s study, not measured thresholds Lore. A 20-lot is routine at 09:31 ET and rare at 12:30 ET. The fix is simple: define “large” as a percentile of the prints in the same half hour of recent sessions (p90, p99), and let the number move with the market.

Speed: a description, not a verdict

A burst of prints says someone needed size now; a dead tape says nobody did. That is what speed describes Definition. What it forecasts is a different matter. In the public practitioner study, volume surges at the end of a move were tested as exhaustion markers and leaned slightly against the claim (M3.02) Lore. Whether a burst of tape speed at a level predicts the next move on NQ is not yet measured.

What a big print does not tell you

Myth / hypothesis: “Big prints are institutions, and their side shows where price goes next.” Lore Two problems. First, a print’s size says nothing reliable about who sent it: large traders usually slice orders into small pieces, and one aggressive order can show as many small prints. Second, as a direction signal it was tested on NQ and not supported (box below). Treat it as a hypothesis you could test with your own percentile rule.

Myth / hypothesis: “A burst followed by steady same-side prints is a campaign and starts a leg; a burst followed by silence is one order and fails.” Lore Follow-through describes what happened after the burst. Whether it forecasts what happens next has not been measured on NQ, and in hindsight every leg seems to have started with a campaign.

Sweeps have a measured result, and it runs against the intuition: in our lab’s NQ event study, sweeps through three or more prices continued slightly less often over the next minute than single-price bursts of the same size and side PlausibleNQ. Three limits apply: the forward test is pending, the outcome uses trade prices (the bid-ask bounce was checked on three days only), and the effect is smaller than a round trip. M3.11 has the numbers.

Why it is still worth knowing

The tape is the source every other order-flow tool summarises. Reading it tells you, at your level, whether business is being done there at all, at what pace, and on which side. It shows the exact price where aggression stopped, which is where a stop is cheap. And it lets you check that your footprint and bubbles show what traded rather than what a filter let through.

What the data saysLoreOne trader's study

Large single prints on the tape predict which way NQ goes next.

Tested and not supported: no edge shown, and the estimate leans slightly against the claim.

Result: −1.3 on the study's primary measure (positive would support the claim)
Sample size: 137,875
Interval: [−1.9, −0.8]
Sample: NQ, every trade 2018–2026 (claim FP-09)

Three caveats: one practitioner's pre-registered study; intervals not clustered by session, so too narrow; a study like this caps at Plausible even when it supports a claim.

The same study's print-size tiers (NQ 20–49 large, 50+ very large) are a rule of thumb, not a measured threshold. Our lab's sweep work sizes prints by the 90th and 99th percentile of each 30-minute window instead.

Our lab has not yet run its own test of large prints on NQ: not yet measured.

Worked example: the sell burst at 18,030, long side

Synthetic session, built to the shape of the hook and the figure, not a real date. NQ, tick 0.25 points, $5 per tick per contract. In your pre-market you measured the prior sessions’ 10:00–10:30 ET prints: p90 is 8 lots and p99 is 46 lots on your aggregated feed.

10:12:04  18,030.00   34  BID
10:12:05  18,029.75   51  BID   <- above p99
10:12:06  18,029.50   28  BID
10:12:07  18,029.25   44  BID
   ... 60 seconds of 1-4 lot prints, both sides ...
10:13:10  18,029.75    3  ASK

Read it in plain words. Someone sold 157 contracts in three seconds, pushing three ticks through the level; one print was above your p99 and three were between p90 and p99. Then the selling stopped and price settled one tick under the level. That is a description: urgent selling that did not extend. It is not a forecast. Large prints did not predict direction in the public study, and nothing in this burst says which way the next move goes.

You do not buy the burst, and you do not sell it. You write the plan in the confirmation format of M4.09, a house rule rather than a measured edge House rule:

If a 1-minute bar closes back above 18,030.25 with positive delta, the push through the level has failed. Long 1 contract at 18,030.50.
Stop 18,028.75, two ticks under the burst low of 18,029.25: 7 ticks of risk, $35 per contract before costs.
Target 2R, 14 ticks, at 18,034.00. Manage it per M6.
Invalid if another sell burst of p90-plus prints trades below 18,029.25: the level is giving way and you stay flat.

In this synthetic session the 10:14 bar closes at 18,030.75 with delta +210 and the order fills at 18,030.50. That is one ending, not the expected one.

The mirror and the failure

Short side. A buy burst lifts the offer through a level above you and stops. The plan mirrors: short only if a bar closes back below the level with negative delta, stop two ticks above the burst high, invalid if a second buy burst trades above it.

The failure. Same burst, another synthetic morning. At 10:12:40 a second wave prints 22, 47 and 31 at the bid down to 18,028.50, and price never closes back above 18,030.25. The trader who “faded the big seller” at 18,029.50 with a stop at 18,028.75 loses 3 ticks plus costs; the trader who sold the first burst at 18,029.25 made money this time and would have lost it on the first morning. The tape showed the event both times. It did not choose the ending.

Do it on the platform

OrderflowSuiteOrderflowSuite

Size big prints against the recent tape, not a fixed number

Add the suite's Big Trades indicator (group Order Flow) to your NQ chart.
Data › Trade Feed › Print Grouping: Aggregated Trades, so one match event counts as one trade.
Data › Trade Feed › Aggregation (ms): 1 (one match event).
Data › Baseline › Baseline Window (min): 30. Each trade is judged against the trades of the last 30 minutes, so "big" moves with the session.
Signals › Filter › Filter Mode: Adaptive and Adaptive Intensity: Medium. Keep them fixed while you collect a sample, so your log is comparable from day to day.

Setting	Value	Why
Print Grouping	Aggregated Trades	a sweep counts once, at its full size
Baseline Window (min)	30	the window your "large" is relative to
Filter Mode	Adaptive	a rolling threshold instead of a fixed contract count

The adaptive threshold is a display choice, not a calibrated signal. The bubbles mark where size traded; they do not say which way price goes next.

ATAS XStock ATAS X

Without the suite: a Smart Tape with a size filter you calibrated

Open a Smart Tape window for the NQ front month.
Show the columns Time, Price, Volume and BuySell (aggressor side).
Measure your p90 for the coming half hour from recent sessions (drill below), then set the tape's minimum-volume filter to that number.
Note whether your tape shows cumulative (aggregated) trades or separate prints, and use the same choice in your big-trade study.

Setting	Value	Why
Columns	Time, Price, Volume, BuySell	the four facts of a print
Minimum volume filter	your p90 for the window	a percentile you measured, not a fixed tier

Common mistakes

Common mistake: Treating a big print as an edge. “A 50-lot hit the bid, so sellers are in control” reads intent and direction into one line. In the public study large single prints did not predict direction; without a baseline, a vivid print is an anecdote.

Common mistake: Using one fixed size tier all day. A 20-lot filter is busy at the open and nearly silent at lunch, so your “big print” count measures the clock, not the participants. Use the percentile of the same half hour.

Common mistake: Reading the tape all day for discovery. It is exhausting, and most prints are routine two-sided business. Watch it at your pre-mapped levels (M4.07), where it answers a narrow question: is business being done here, at what pace, and where did aggression stop?

Drill: measure your own percentiles

Use replay (PLAT2.06) on five recent NQ sessions, with the same aggregation setting you trade with.

Export or record every print between 09:30 and 10:00 ET and between 12:00 and 12:30 ET.
For each window, compute p90 and p99 of print size (a spreadsheet’s PERCENTILE function is enough).
Compare them with fixed tiers of 20 and 50 contracts: what share of prints does each tier keep in each window?
Switch the tape to the other aggregation setting for one session and repeat step 2.

Solution

Expect the open window’s percentiles to sit well above the lunch window’s, so a fixed 20-lot tier keeps a different share of prints in each. Expect the aggregated feed’s p99 to be larger than the unaggregated feed’s, because sweeps count once at their full size. Write your p90 and p99 per window into your pre-market sheet; those are the numbers your size filter should use. The exact values depend on the period you choose: our lab has not yet published NQ percentiles by time of day.

What we don’t know yet

NQ print-size percentiles by half hour and by aggregation setting: not yet measured in our lab.
Whether a burst of tape speed at a mapped level predicts the next move: not yet measured.
Whether ATAS’s aggressor side matches the exchange’s own flag on NQ: not yet measured.
Whether reading the tape at a zone improves outcomes over not reading it: not yet measured (M4.09 writes the rule so it can be scored).

Recap

The tape lists every trade: time, price, size and aggressor side. Footprint, delta and bubbles are built from it; the DOM is not.
One order can print as many small lines or one large line; know your aggregation setting before you read size.
“Large” is a percentile of the same window, not a fixed tier; fixed tiers are conventions.
Size and speed describe urgency. As direction signals, large prints are graded Lore on NQ; use the tape to describe the event at your level and let a written rule decide.