I distilled liquid gold from 9 years of MBO data.

Main ideas, concepts, and lessons

• Candle prices are summaries; the “tape” matters.

  • Retail traders often interpret each candle as if it fully represents who won (buyers vs. sellers).
  • The key claim: the real predictive information lies in individual trades—their size, price, and especially who initiated them—which is available via Level 3 (market-by-order / complete order book) data.

• Research question: do the biggest trades in each minute predict the next move?

  • Using ~9 years of Level 3 MBO data (May 2017 to July 2026), the research tests whether the largest trade executed within each 1-minute interval (“whale”) forecasts the next price movement.

• Backtesting approach is standardized to measure “information,” not strategy performance.

  • Trades are entered/exited in a uniform way so the average profit per trade becomes a clean measure of informational edge.

---

Methodology / experiment setup

Data

• 9 years of Level 3 market-by-order (MBO) data from May 2017 to July 2026
• Equivalent scale: about 3 million 1-minute bars

Trading signal definition

For each 1-minute bar:

• Identify the biggest trade (largest “print” in that minute)
• Use the direction of that biggest trade:
  • If the biggest print is buy-initiated → go long
  • If sell-initiated → go short
• Trigger a trade at the open of the next candle

Trade lifecycle

• Holding time: exactly n minutes (tested across multiple values)
• No stops, no targets, no discretion
• Every trade uses the same “bet structure,” with variation only in the signal rules (baseline vs whale vs filters)

Profit measurement philosophy

• Uses gross results (before costs) to measure predictability first.
• “Signal performance” is evaluated via:
  • Average profit per trade
  • Win rate
  • Profit factor (for the final configuration)

Baseline test (“candle-only” behavior)

• Rule:
  • Go long after every green candle, short after every red candle
  • Hold for 1 minute
• Result:
  • ~3 million trades
  • 41 cents per trade
  • Statistically real but economically negligible (spread is much larger than the average edge)

Whale test (biggest trade each minute)

• Rule:
  • Keep the same 1-minute hold
  • Only trade when the minute’s biggest print is ≥ 100 contracts
  • Direction follows the print (long on buy-initiated, short on sell-initiated)
• Result:
  • Average trade: $1.88
  • About 4.6× the baseline informational value (with the same mechanical bet)
  • Win rate: 49%
• Key insight: the edge isn’t mainly about being right more often; it’s because winners run further than losers.

---

Surprises / findings (what changed and what it meant)

Surprise 1: Market digest time is ~15–20 minutes (not seconds)

• Procedure: optimize holding periods n = 1 to 30 minutes (MultiCharts)
• Observed pattern:
  • Edge grows to a plateau roughly from minute 13 to 23
  • Peak around minute 19
  • Declines by minute 30
• Interpretation:
  • It appears the market takes 15–20 minutes to “digest” information from a large block-like trade.
  • This challenges scalping order-flow beliefs that everything matters instantly at sub-second/sub-minute horizons.

Surprise 2: There’s a “ceiling” to informative size; beyond that, size becomes exhaustion

• Procedure:
  • Optimize largest-print thresholds:
    • size from 10 to 300 contracts (step 10)
  • Also test holding horizons 5 to 30 minutes (step 5)
• Observed pattern:
  • For below ~50 lots: biggest print provides no useful predictive edge
  • For ~70 to 190 contracts: edge increases with size
  • For above ~200 contracts: edge disappears
  • Relationship may even become inverse for “monster” prints
• Interpretation:
  • Very large prints may reflect blocks, stop cascades, re-pricing events, and other dynamics that can partially reverse the effect.
  • Conclusion: size is informative only up to a point; beyond that, it becomes exhaustion.

Surprise 3: Context matters—time of day, where within the candle, and order book balance

• Regular vs overnight session:

  • 79% of relevant large prints occur during regular trading hours
  • Overnight prints “predict nothing” (attributed to plumbing, rolls, hedges, spread legs)

• Holding time sensitivity:

  • With a 15-minute hold:
    • average win grows dramatically to $24.31

• Location within the candle:

  • “Where” matters:
    • Whale during the candle’s week (top extreme):
      • ~$34 per trade at 15 minutes
    • Whale in the body/mid:
      • ~$8 per trade

• Order book balance:

  • If the resting book is balanced: whale print is predictive
  • If the book is already leaning: the whale loses predictive value

• Interpretation:

  • Most informative large trades are those the market didn’t already price in (i.e., occur when context suggests the move was not fully anticipated).

---

Final “best configuration” test (integrating all lessons)

The speaker integrates the findings into a final mechanical configuration:

• Consider only prints larger than 100 contracts
• Trade only during regular trading hours
• Require:
  • Balanced order book
  • Large trades located on the candle’s week (extreme position)
  • Hold for 15 minutes
  • Direction follows the print

Results:

• 1,587 trades
• $53 per trade (average)
• Profit factor: 1.24
• Long and short contribute equally

Important caveat:

• Not claimed as a fully deployable strategy:
  • Average trade win may be too small after commissions and slippage
  • Edge is stronger at the longer horizon

---

Practical takeaways (explicit “what you should do” rules)

• Focus on large trades:

  • 100+ contracts
  • Below 50 lots: biggest minute print tells you nothing

• Size zone of usefulness:

  • Informational edge begins around 60–70
  • Strongest range roughly 70–190
  • > ~200 contracts: edge disappears (ignore/avoid)

• Avoid monster prints

  • Interpreted as blocks, stop cascades, climactic prints, etc.

• Trade timing:

  • Prefer regular trading hours
  • Overnight prints predict nothing

• Hold duration:

  • At 1 minute, edge is below spread (not enough)
  • After 15–20 minutes, you may overcome trading costs

• Location and context:

  • Whale at the candle extreme (“week”) carries edge; whale in the body carries much less/no information
  • Whale is predictive when the order book is balanced; it loses meaning when the book is already leaning

• Direction rule:

  • Trade with the large print direction:
    • Large seller → sell
    • Large buyer → buy
    • Even if the candle disagrees

---

Sources / speakers featured

• Speaker: Matt Conte

  • Market maker (claimed 7 years)
  • Former employer: a large investment bank (claimed 30M+ euro profit)
  • Runs a hedge fund automating trading strategies
  • Builds/uses a tool to connect Level 3 data with MultiCharts backtesting

• Data / systems referenced (not individuals):

  • Nasdaq futures (study subject)
  • CME order book / Level 3 MBO data
  • MultiCharts (backtesting software)
  • “Institutional protocol” / “institutional protocol tool” (as referenced by the speaker)