Understand latency arbitrage — Stock-Market notes

WHAT is latency arbitrage?

Latency = the time delay between an event happening and you learning/acting on it.
Stale quote = a resting order priced on old information that hasn't updated yet.
It is a statistical/mechanical arbitrage: the "risk-free" window is tiny (microseconds) but the edge is near-certain if you win the race.

WHY does the opportunity exist?

Three ingredients must all be present:

Fragmentation — the same asset (or correlated assets, e.g. SPY vs. E-mini futures) trades on multiple venues.
Finite propagation — updates reach venues at different times.
Passive resting liquidity — slow participants leave orders that are momentarily mispriced.

HOW is the profit built up — derivation from scratch

We derive the expected profit per event, not just quote it.

Step 1 — The price move. Suppose the fair price jumps up by $m$ (a "tick move") on the fast feed. Why? Arbitrage needs a change; a stable price gives nothing to exploit.

Step 2 — The stale quote. The slow venue still shows an ask (someone willing to sell) at the old price $p$ . The new fair value is $p + m$ . Why this step? The mispricing is exactly the resting order that hasn't been cancelled yet.

Step 3 — The trade. The fast trader buys at $p$ (lifting the stale ask), then sells at $p+m$ once everyone re-prices. Gross gain $= m$ per share. Why? You lock in the difference between the stale price and the new true price.

Step 4 — Subtract frictions. Real edge: $\pi = m - c$ where $c$ = fees + market-impact + adverse-selection cost. Why subtract $c$ ? Exchanges charge; sometimes the "stale" quote was cancelled a nanosecond before you hit it, giving you nothing or a loss.

Step 5 — Probabilistic expected value. You only win the race with probability $P_{win}$ ; if you lose, you may pay a small cost $\ell$ : $\boxed{\; \mathbb{E}[\pi] = P_{win}\,(m-c) - (1-P_{win})\,\ell \;}$ Why multiply by $P_{win}$ ? Many fast firms race for the same stale quote — only the first wins. Speed increases $P_{win}$ , which is why firms spend fortunes on microwave towers.

Worked example Numerical walk-through — SPY vs. E-mini

Chicago futures (E-mini) tick up by $m = \$ 0.01$ per SPY-equivalent share.
NY shows a stale SPY ask of 10,000 shares at the old price.
Fees + slippage $c = \$ 0.002 $/share; you win the race with$ P_{win}=0.7 $; loss if you miss$ \ell = $0.001$.

Per share: $\mathbb{E}[\pi] = 0.7(0.01 - 0.002) - 0.3(0.001) = 0.0056 - 0.0003 = \$ 0.0053 $. **Why this step?** We plug into the boxed formula; each term is (probability)×(payoff). **Total on 10,000 shares:**$ 0.0053 \times 10{,}000 = $53$ in a few microseconds. Repeat thousands of times a day → the business.

Common mistakes (steel-manned)

The 80/20 — what actually matters

Forecast-then-Verify

Recall Forecast before revealing

Q: Two microwave firms cut Chicago–NY latency from 4.0 ms to 3.9 ms. Does per-trade profit $m-c$ change? What changes?

Predict, then reveal: Per-trade payoff $m-c$ is unchanged (depends on the tick move). What changes is $P_{win}$ — the 0.1 ms firm wins more races, so expected profit $\mathbb{E}[\pi]=P_{win}(m-c)-(1-P_{win})\ell$ rises via a higher $P_{win}$ . This is exactly why speed spending never stops.

Feynman: explain it to a 12-year-old

Recall Explain like I'm 12

Imagine two shops in two towns sell the same trading cards, and they always match prices by phone. One day the "real" price goes up. The phone call to Town B takes a few seconds. A kid with a walkie-talkie hears the new price first, sprints to Town B, and buys the card that's still marked at the old cheap price, then sells it for the new higher price. He didn't cause the price change — he just ran faster than the phone call. That head-start is latency arbitrage.

Mnemonic

Flashcards

What is latency arbitrage?

Profiting from speed differences in receiving/acting on market data by trading against stale quotes on slower venues before they re-price.

Why does the opportunity physically exist?

Price info travels at finite speed, so fragmented venues hold different (stale) views of the same price for a brief window.

Formula for the arbitrage window?

\Delta t = L_{slow} - L_{fast}

; profit only exists when this is positive.

Expected profit per event?

\mathbb{E}[\pi] = P_{win}(m-c) - (1-P_{win})\ell

, where

m

=price move,

c

=cost,

\ell

=loss on miss.

Does a bigger latency edge increase per-trade payoff?

No — per-trade payoff is

m-c

(depends on the price move). A bigger edge raises

P_{win}

, i.e. how often you win the race.

Why isn't latency arbitrage risk-free?

Execution risk (quote cancelled), queue risk (someone faster wins), and signal risk (move reverses);

P_{win}<1

Why microwave links over fiber?

Air-microwave is ~1.5× faster than light-in-fiber and travels straighter, cutting propagation distance/time.

What is a stale quote?

A resting order priced on old information that hasn't updated to the new fair value yet.

Who bears the cost of latency arbitrage?

Slow market makers, via adverse selection; they respond by widening spreads.

What is co-location?

Placing your servers physically inside the exchange datacenter to minimize propagation latency.

Connections

High-Frequency Trading (HFT) — latency arb is a flagship HFT strategy.
Market Microstructure — stale quotes and adverse selection live here.
Co-location and Network Speed — the infrastructure that wins the race.
Bid-Ask Spread — widens as makers defend against being picked off.
Statistical Arbitrage — the broader family of probabilistic edges.
Market Fragmentation — the structural precondition.
SPY vs E-mini Futures Basis — classic Chicago–NY latency arb pair.

Concept Map

Hinglish (regional understanding)

Intuition Hinglish mein samjho

Dekho, latency arbitrage ka funda simple hai: ek hi stock kai jagah (venues) pe trade hota hai — jaise SPY New York mein aur E-mini future Chicago mein. Jab price change hoti hai, wo news har venue tak ek saath nahi pahunchti, kyunki information ki speed limited hai (light bhi fibre mein delay karti hai). Toh ek chhota sa time gap banta hai jismein ek venue ka quote purana (stale) reh jaata hai. Jo trader sabse fast hai, wo naya price pehle dekh leta hai aur us stale cheap quote ko utha leta hai — buy sasta, sell mehenga. Yehi hai latency arbitrage.

Profit ka maths yaad rakho: per trade tumhe milta hai roughly $m - c$ , jahan $m$ = price ka move aur $c$ = fees/slippage. Lekin race har koi lad raha hai, toh sirf jo pehle pahunche usko fill milta hai. Isliye expected profit hota hai $\mathbb{E}[\pi] = P_{win}(m-c) - (1-P_{win})\ell$ . Important baat: agar tum apni latency aur kam kar lo, toh per-trade payoff nahi badhta — badhta hai $P_{win}$ , yaani tum zyada races jeetne lagte ho. Isi wajah se firms crore lagati hain microwave towers aur co-location (server ko exchange ke andar rakhna) pe.

Ek galatfehmi door kar lo: ye "risk-free" arbitrage nahi hai. Ho sakta hai stale quote cancel ho jaaye, ya koi tumse fast nikal jaaye, ya move ulta ho jaaye — isiliye $P_{win}$ hamesha 1 se kam hota hai. Ye ek statistical edge hai, guaranteed profit nahi. Aur jo slow market maker apna order chhod ke gaya tha, uska nuksaan hota hai (adverse selection), isliye wo apna spread bada karke apna bachaav karta hai. Bas yaad rakho: fragmentation + finite speed = stale quote = fast banda kha jaata hai.

Test yourself — HFT & Advanced Concepts

Connections