Order Flow - Where Orders Come From and Where They Go
TL;DR
Order flow is the stream of buy and sell orders entering a market. Not all flow is equal:
- Uninformed flow (retail) = Easy for MMs to profit from, highly valuable
- Informed flow (hedge funds, quants) = MMs often lose money against this
- Toxic flow (HFT arb, latency exploits) = Systematically picks off MMs
This explains:
- Why Citadel pays Robinhood for retail orders (PFOF)
- Why large trades go OTC instead of on-exchange
- Why MMs charge wider spreads when they can't identify the counterparty
Payment for Order Flow (PFOF)
What is PFOF?
Payment for Order Flow is when a broker sells its customers' orders to a market maker instead of routing them to a public exchange.
The most famous example: Robinhood → Citadel Securities
How Traditional Exchange Routing Works
┌─────────────────────────────────────────────────────────────────┐
│ TRADITIONAL EXCHANGE MODEL │
├─────────────────────────────────────────────────────────────────┤
│ │
│ You ──→ Broker ──→ NYSE/NASDAQ ──→ Executes on public book │
│ ↑ │
│ Many MMs competing │
│ for your order │
│ │
│ Your order is visible to everyone │
│ Best price wins (price-time priority) │
└─────────────────────────────────────────────────────────────────┘How PFOF Works
┌─────────────────────────────────────────────────────────────────┐
│ PFOF MODEL │
├─────────────────────────────────────────────────────────────────┤
│ │
│ You ──→ Robinhood ──→ Citadel Securities ──→ Internalized │
│ ↑ │ │
│ $0 commission Pays Robinhood │
│ ~$0.004/share │
│ │ │
│ Never touches │
│ public exchange │
│ │
│ Citadel sees your order BEFORE deciding to fill │
│ Trade happens privately between you and Citadel │
└─────────────────────────────────────────────────────────────────┘Key difference: In PFOF, the market maker sees your order first and then decides whether to fill it. On an exchange, MMs post quotes blind and anyone can trade against them.
Why Does Citadel Pay for Retail Orders?
This is the crucial insight: Not all order flow is equal.
| Flow Type | Source | Characteristics | MM Profitability |
|---|---|---|---|
| Uninformed | Retail traders | Random timing, no edge, trades for personal reasons | High - easy to profit |
| Informed | Hedge funds, quants | Has alpha, knows something MM doesn't | Low - gets picked off |
| Toxic | Latency arb, news algos | Systematically exploits stale quotes | Negative - loses money |
Retail flow is "uninformed":
When a retail trader buys Apple stock, they're usually not doing it because they know earnings will beat expectations. They're buying because:
- They like the company
- They got a bonus and want to invest
- They saw something on social media
- It's part of a regular savings plan
Their trades don't predict future price movements. This means:
- MM can quote tight spreads safely
- Low adverse selection risk
- Consistent, predictable profits
Institutional/HFT flow is often "informed" or "toxic":
When a hedge fund buys, they might know something. When an HFT arb bot trades, it's exploiting a price discrepancy the MM hasn't corrected yet. MMs lose money against these traders.
The Economics of PFOF
Citadel buys 1000 shares of AAPL from retail customer:
NBBO (National Best Bid/Offer): $150.00 / $150.02
(This is the best public price on NYSE/NASDAQ)
Citadel offers retail: $150.01 (1 cent "price improvement")
Citadel's cost to hedge: ~$150.00 (can buy at bid elsewhere)
Revenue:
Citadel profit on spread: $0.01 × 1000 = $10.00
Minus PFOF paid to Robinhood: - $4.00
─────────────────────────────────────────────────
Net profit to Citadel: $6.00
Scale: Millions of orders per day = massive profitsThe "Price Improvement" Debate
PFOF defenders say: Retail gets better prices than the public NBBO.
- NBBO spread: $150.00 / $150.02 (2 cents wide)
- Citadel fills at: $150.01 (1 cent "improvement")
Critics argue:
- The NBBO is artificially wide because HFTs profit from it
- If orders went to exchanges, competition might tighten the NBBO itself
- "Improvement" is measured against a flawed benchmark
- Retail never sees the "real" best price that exists in dark pools
Scale of PFOF
| Metric | Value |
|---|---|
| Robinhood PFOF revenue (2021) | ~$974M (half of total revenue) |
| Citadel Securities market share | ~41% of US retail equity orders |
| Top 3 wholesalers combined | 80%+ of all retail flow |
| Total US PFOF industry (2021) | ~$3.8B |
The Controversy
| Arguments For PFOF | Arguments Against PFOF |
|---|---|
| Enabled zero-commission trading | Conflict of interest (broker serves MM, not customer) |
| Price improvement vs NBBO | "Improvement" measured against artificially wide NBBO |
| Retail access to markets democratized | Best execution not actually guaranteed |
| MMs provide liquidity | Concentrated power in few wholesalers |
| Innovation and competition | Regulatory arbitrage (avoiding exchange rules) |
Regulatory Status
| Jurisdiction | Status |
|---|---|
| United States | Legal, but SEC has proposed restrictions |
| United Kingdom | Banned |
| Canada | Banned |
| Australia | Banned |
| European Union | Restricted under MiFID II |
PFOF vs Exchange Market Making
The Fundamental Difference
Where does the trade happen?
PFOF / Internalization:
Order → Broker → Single MM → Trade happens privately
- MM sees order BEFORE deciding to fill
- No competition for that specific order
- Price based on MM's discretion
- Never touches public order book
Exchange Market Making:
MM posts quotes → Order book → Anyone can trade
- MM commits to prices BEFORE seeing orders
- Competes with other MMs and all traders
- Price discovery is public
- All trades visible on exchangeDetailed Comparison
| Dimension | PFOF (Citadel/Robinhood) | Exchange MM (Wintermute on Binance) |
|---|---|---|
| Order visibility | MM sees order first, then decides | MM posts blind quotes |
| Competition | Usually 1 MM handles each order | Many MMs + all traders compete |
| Price discovery | Private | Public order book |
| Counterparty selection | Broker chooses which MM | Market chooses (best price wins) |
| Flow composition | Pure retail (by design) | Mixed (retail + institutional + bots) |
| Adverse selection risk | Very low (curated flow) | Higher (anyone can trade) |
| Spreads MM can quote | Tighter (low-risk flow) | Wider (must protect against informed) |
| Transparency | Limited (execution reports only) | Full (public order book) |
| Regulation | Securities law (FINRA, SEC) | Varies (crypto often less regulated) |
Why This Matters
In PFOF, the MM has an information advantage:
- Sees the order before committing to a price
- Knows it's retail (low adverse selection)
- Can choose whether to fill or route elsewhere
- Sets the price with full knowledge of the order
On an exchange, the MM is quoting blind:
- Posts prices without knowing who will trade
- Anyone can hit the quote (retail, HFT, hedge fund)
- Must protect against informed traders with wider spreads
- Competes purely on price and size
Is PFOF "Market Making"?
Technically yes - someone is providing liquidity. But it's a fundamentally different business:
| Traditional Exchange MM | PFOF Wholesaler |
|---|---|
| Earns spread by posting competitive quotes | Earns spread by purchasing curated flow |
| Risk: adverse selection from anyone | Risk: minimal (pre-selected retail) |
| Competition: other MMs on the same book | Competition: other wholesalers for broker contracts |
| Skill: pricing, speed, risk management | Skill: broker relationships, execution tech, scale |
Does Crypto Have PFOF?
Short Answer: Not exactly, but similar dynamics exist.
Why Crypto Doesn't Have Traditional PFOF
- No broker layer - Crypto exchanges ARE the brokers. Users trade directly on exchange order books.
- No regulatory framework - No equivalent of FINRA/SEC "best execution" requirements
- Fragmented markets - No consolidated tape like NBBO
But Similar Concepts Exist
1. OTC Desks (Cumberland, Wintermute OTC, etc.)
Large traders request quotes directly from market makers:
- MM provides price, trade happens off-book
- Similar to PFOF: private execution, MM sees order first
- Used for large blocks to avoid market impact
2. RFQ Systems (Request for Quote)
User requests quotes from multiple MMs:
- MMs compete to fill
- Hybrid: competitive bidding, but off-book execution
- Example: Backpack's Convert feature
3. DEX Aggregators + Private Market Makers
Some aggregators route to "private" MM liquidity pools:
- Similar flow segmentation dynamics
- Retail gets filled by designated MM rather than AMM
4. Exchange "Convert" / "Swap" Features
Simple swap interfaces on exchanges:
- Often filled by designated MMs behind the scenes
- User gets convenience, MM gets retail flow
- Spread embedded in the price
The Key Insight
The PFOF dynamic exists whenever:
- Flow can be segmented (retail vs institutional)
- Someone pays for access to "better" (less informed) flow
- Execution happens away from public price discovery
In crypto, this manifests differently (OTC, RFQ, Convert) but the economics are similar.
Toxic Flow
What Makes Flow "Toxic"?
Toxic flow = trades that systematically lose money for the market maker.
It's not about intent - it's about outcomes. If a counterparty's trades consistently move against the MM after execution, that flow is toxic.
Sources of Toxic Flow
| Source | How It Works | Why It's Toxic |
|---|---|---|
| Latency Arbitrage | HFT sees price change on Exchange A, hits stale quote on Exchange B faster than MM can cancel | MM's quote is outdated; they sell low or buy high |
| News Trading | Algo parses news/tweets faster than MM can react | MM provides liquidity into informed directional pressure |
| Statistical Arbitrage | Quant models predict short-term price moves | MM is consistently on wrong side of predicted moves |
| Momentum Ignition | Large aggressive orders trigger price cascade | MM provides liquidity into adverse price movement |
| Cross-venue Arb | Trader exploits price differences between exchanges | MM is arbitraged against other venues |
How MMs Measure Toxicity: Markouts
Markout = How much the price moves against you after a fill.
Example: MM sells BTC at $100,000
After 1 second, price is $99,950
→ Markout: +$50 (price went DOWN after you sold = good)
After 1 second, price is $100,050
→ Markout: -$50 (price went UP after you sold = bad)
If your markouts are consistently negative:
→ You're getting picked off
→ That flow is toxicMMs track markouts at multiple time horizons (100ms, 1s, 10s, 1min) to understand flow quality.
VPIN: Academic Toxicity Metric
VPIN (Volume-Synchronized Probability of Informed Trading):
- Tracks order imbalance relative to volume
- High VPIN = likely informed trading is occurring
- Used as early warning signal for toxic conditions
How MMs Defend Against Toxic Flow
| Defense | How It Works |
|---|---|
| Speed | Faster quote updates = less time to be picked off |
| Wider spreads | Charge more when uncertainty is high |
| Flow segmentation | Different prices for different counterparty types |
| Pulling quotes | Remove liquidity entirely during high-risk periods (news, volatility) |
| Inventory limits | Cap exposure to prevent catastrophic adverse moves |
| Counterparty analysis | Track which accounts are consistently toxic, quote wider to them |
The Arms Race
HFT/Arb invests in: MM invests in:
├─ Faster hardware ├─ Faster hardware
├─ Colocated servers ├─ Colocated servers
├─ Predictive models ├─ Defensive models
├─ News/data parsing ├─ News/data parsing
├─ Microwave towers ├─ Microwave towers
└─ Exchange connectivity └─ Exchange connectivity
→ Billions spent on infrastructure
→ Advantages measured in microseconds
→ Constant escalationWhy Retail Flow is Valuable (Redux)
This is why PFOF exists and why MMs pay for retail flow:
| Flow Type | Markout Profile | MM Wants It? |
|---|---|---|
| Retail | Random (no consistent direction) | Yes - will pay for it |
| Institutional | Often informed (negative markouts) | Depends on size/relationship |
| HFT/Arb | Systematically toxic | No - will quote wider or refuse |
OTC Market Making
What is OTC?
Over-The-Counter (OTC) = Trades negotiated directly between two parties, not on an exchange order book.
How OTC Works
Institutional Client: "I want to buy 500 BTC"
Cumberland (OTC MM): "I'll sell you 500 BTC at $100,050"
Client: "Deal"
Trade settles directly between parties.
Never touches Binance/Coinbase order book.
No one else knows this trade happened (until reported).Why Use OTC Instead of Exchange?
| Factor | Exchange | OTC |
|---|---|---|
| Market impact | Large orders move price (slippage) | Single price for entire block |
| Price certainty | Unknown until fully filled | Locked in quote |
| Visibility | Everyone sees your order | Private negotiation |
| Size | May not have enough liquidity | MM commits to full size |
| Timing | Immediate but uncertain execution | Negotiated, certain terms |
| Counterparty | Anonymous | Known relationship, credit terms possible |
OTC MM Business Model
Client asks for quote to BUY 500 BTC
OTC MM thinks:
Current market mid: $100,000
Can I source 500 BTC?
- My inventory: 100 BTC
- Exchange liquidity: ~200 BTC within 10bps
- Other OTC desks: can probably get 200 BTC
- Estimated average cost: $100,020
Risk premium for:
- Size: large block
- Timing: need to source quickly
- Market risk: price might move while I source
- Premium: $30
My quote: $100,050 (50 bps markup)
If client accepts:
- I'm committed to DELIVER 500 BTC at $100,050
- I go source 500 BTC (exchanges, inventory, other OTC)
- If I source at $100,020 avg → profit: $15,000
- If market spikes to $100,100 while sourcing → loss: $25,000OTC vs Exchange MM Risk Profile
| Risk Factor | Exchange MM | OTC MM |
|---|---|---|
| Quote lifetime | Milliseconds (can cancel instantly) | Minutes to hours (committed once client accepts) |
| Trade size | Small per individual quote | Large blocks (often $1M+) |
| Hedge timing | Immediate (quote and hedge simultaneously) | May take time to source full amount |
| Counterparty | Anonymous | Known (relationship and credit risk matter) |
| Information | Quotes are public | Private negotiation (info advantage) |
Who Uses OTC?
- Institutions - Hedge funds, family offices, corporate treasuries
- High-net-worth individuals - Large personal trades
- Token projects - Treasury management, token sales
- Miners - Selling block rewards without market impact
The Full Market Making Ecosystem
How It All Connects
┌─────────────────────────────────────────────────────────────────────┐
│ LIQUIDITY ECOSYSTEM │
├─────────────────────────────────────────────────────────────────────┤
│ │
│ RETAIL TRADERS │
│ │ │
│ ├──→ Direct to Exchange ──→ Trade against MM quotes on book │
│ │ ↑ │
│ │ Exchange MMs provide liquidity │
│ │ (Wintermute, Jump, GSR, etc.) │
│ │ │
│ ├──→ Via Broker (TradFi) ──→ PFOF to Citadel/Virtu │
│ │ (internalized off-exchange) │
│ │ │
│ └──→ "Convert" features ──→ RFQ to designated MMs │
│ │
│ INSTITUTIONAL TRADERS │
│ │ │
│ ├──→ Exchange (algo execution) ──→ TWAP/VWAP against book │
│ │ │
│ ├──→ OTC Desk ──→ Block trades with Cumberland/Wintermute │
│ │ │
│ └──→ Dark Pools (TradFi) ──→ Anonymous institutional matching │
│ │
│ MARKET MAKERS │
│ │ │
│ ├──→ Cross-exchange arb ──→ Keeps prices aligned across venues│
│ │ │
│ ├──→ Hedge on other venues ──→ Manage inventory risk │
│ │ │
│ └──→ Inter-dealer trading ──→ MMs trade with each other │
│ │
└─────────────────────────────────────────────────────────────────────┘The Economic Food Chain
WHO PAYS WHO GETS PAID
────────────────────────────────────────────────────────────────
Retail traders (spread + fees) → Market makers (spread capture)
Uninformed traders → Informed traders
Slow traders → Fast traders
Large orders (slippage) → Liquidity providers
Takers (cross the spread) → Makers (rebates)
Impatient traders → Patient tradersWhy Understanding This Matters
- As a trader: Know where your order goes and who profits from it
- As an exchange: Understand why MMs want (or don't want) your flow
- As a builder: Design systems that attract good liquidity
- As a regulator: Understand conflicts and market structure risks
Questions You Should Be Asking
About Any Market Maker
- Revenue model: Spread capture? Rebates? Prop trading? Token deals?
- Flow quality: Who trades against them? How do they segment flow?
- Risk management: How do they hedge? What's their max inventory?
- Infrastructure: Where are they colocated? What's their latency?
- Conflicts: Do they run prop strategies? Front-running risk?
About Exchange-MM Relationships
- What does the MM get? Fee rebates? Exclusive access? Data advantages?
- What does the exchange get? Spread commitments? Uptime guarantees?
- Who are the designated MMs? How many? What markets?
- What happens if an MM fails? Backstop arrangements?
About PFOF / Internalization
- Where is my order going? Public exchange or internalized?
- Who is filling my order? Which market maker?
- What is "price improvement" measured against? Real NBBO or stale quote?
- Can I opt out? Direct routing options available?
About Order Flow Generally
- How fragmented is liquidity? One dominant venue or many?
- What % is OTC vs on-exchange?
- Who are the dominant MMs? Concentration risk?
- What's the regulatory framework? Best execution requirements?
About Toxic Flow
- How does the venue identify toxic flow?
- Can MMs quote different prices to different counterparties?
- What protections exist against latency arbitrage?
- How transparent is the venue about flow composition?
Summary
Order flow is the lifeblood of markets. Understanding where orders come from, where they go, and who profits reveals:
- Why zero-commission trading exists (PFOF pays for it)
- Why large trades go OTC (to avoid market impact)
- Why MMs quote wider during volatile periods (toxic flow risk)
- Why retail flow is valuable (uninformed = profitable for MMs)
- Why speed matters (the arms race against toxic flow)
The key insight: Not all liquidity is equal, and not all flow is equal. Markets are a complex ecosystem of different participants with different information, different speeds, and different objectives - all interacting through the prices they're willing to trade at.