Speed Bumps - Intentional Latency in Market Microstructure
TL;DR
- Speed bump = An intentional delay added to order processing, originally pioneered by IEX with a 350 microsecond delay
- Problem it solves: Latency arbitrage — fast traders race to pick off stale quotes from market makers before they can update prices
- How it helps: Gives market makers time to update quotes → tighter spreads → better prices for everyone
- Symmetric (IEX) delays all orders equally; asymmetric delays only aggressive orders while letting cancels through instantly
- 100ms is massive — 285x longer than IEX's delay. At that timescale it's not anti-HFT, it fundamentally reshapes who can provide liquidity
- Crypto context: Latency arb across CEXs is rampant. DEXs use batch auctions and MEV protection as blockchain-native alternatives
- Key insight: A speed bump is a policy decision about who the exchange's microstructure favors — makers or takers, fast or slow
New to these concepts? See the Glossary for definitions of every term used in this doc.
Table of Contents
- What Is a Speed Bump?
- The Problem: Latency Arbitrage
- How a Speed Bump Helps the Order Book
- Symmetric vs Asymmetric Speed Bumps
- What a 100ms Speed Bump Means
- Speed Bumps in Crypto
- Relevance to Backpack Exchange
1. What Is a Speed Bump?
A speed bump is an intentional delay inserted into the order processing pipeline of an exchange. Instead of matching orders as fast as the hardware allows, the exchange holds incoming messages for a fixed duration before acting on them.
The IEX Origin Story
IEX (Investors Exchange) introduced the concept in 2016 with a 350 microsecond (0.35ms) delay. They implemented it physically: 38 miles of coiled fiber optic cable in a small box that every order must traverse before reaching the matching engine.
Traditional Exchange:
Order → [Matching Engine] → Fill
Latency: ~microseconds
IEX:
Order → [38 miles of coiled fiber] → [Matching Engine] → Fill
Added delay: 350 microseconds350 microseconds is imperceptible to any human trader. You can't feel it, you can't see it. But it's an eternity in the world of high-frequency trading, where firms spend hundreds of millions to shave off single microseconds.
Why Would an Exchange Deliberately Slow Itself Down?
Because raw speed creates a predator-prey dynamic in the order book. The predators are the fastest firms. The prey are the market makers whose quotes become stale when prices move. The speed bump neutralizes the predators' edge.
2. The Problem: Latency Arbitrage
How It Works
Imagine BTC is trading at $100,000 on both Exchange A and Exchange B. A large sell order hits Exchange A and pushes the price to $99,950. What happens next:
Timeline (microseconds):
t=0 Large sell on Exchange A. Price drops to $99,950.
t=5 HFT firm sees the price move on Exchange A.
Their algorithms know Exchange B still shows $100,000.
t=10 HFT sends a market buy to Exchange B at $100,000.
They're buying from the market maker's resting bid.
t=50 Market maker on Exchange B sees the move on Exchange A.
Wants to cancel their $100,000 bid. Too late.
t=51 HFT's buy order fills against MM's stale $100,000 bid.
HFT bought at $100,000, true price is $99,950.
HFT profit: $50/BTC. MM loss: $50/BTC.The HFT firm didn't provide any liquidity or discover any price. They simply had a faster connection and raced the market maker to the punch. This is latency arbitrage — also called stale-quote sniping.
Adverse Selection: The Tax on Market Makers
Every time a market maker gets sniped, they lose money. This is called adverse selection — the market maker's fills are disproportionately toxic (they buy right before prices drop, and sell right before prices rise).
Market makers aren't stupid. They respond rationally:
- Widen spreads — Charge more per trade to cover sniping losses
- Reduce size — Post smaller orders so each snipe hurts less
- Reduce time in market — Quote for shorter periods (flicker quotes on and off)
- Invest in speed — Spend millions on co-location and faster hardware
All four responses make the market worse for regular traders:
| MM Response | Effect on Regular Traders |
|---|---|
| Wider spreads | Pay more on every trade |
| Smaller size | Can't fill large orders without moving price |
| Quote flickering | Phantom liquidity, worse execution |
| Speed arms race | Exchange costs increase, passed to all participants |
This is the core insight: latency arbitrage is a hidden tax on everyone. The HFT profits come directly from market makers, who pass the cost along to all traders through worse quotes.
3. How a Speed Bump Helps the Order Book
The Mechanism
A speed bump gives market makers a window to update their quotes after a price move, before aggressive sniping orders can reach the book.
Without Speed Bump:
t=0 Price moves on Exchange A
t=5 Sniper sends buy order to Exchange B
t=10 Sniper's order arrives and fills MM's stale quote ← MM loses
t=50 MM's cancel arrives. Too late.
With 350μs Speed Bump:
t=0 Price moves on Exchange A
t=5 Sniper sends buy order to Exchange B
t=10 Sniper's order enters the speed bump (held for 350μs)
t=50 MM's cancel arrives and executes immediately (or also delayed, depending on design)
t=360 Sniper's order exits speed bump. MM's quote is already updated.
Sniper gets the new price or nothing. ← MM protectedThe speed bump doesn't prevent trading. It prevents racing. Legitimate buyers and sellers still get filled — they just can't exploit the microseconds of information asymmetry.
What Improves
IEX published research showing measurable improvements across key metrics:
| Metric | What It Measures | Effect |
|---|---|---|
| Quoted spread | Best bid-ask gap on the book | Tighter (MMs quote more aggressively) |
| Effective spread | Actual cost paid by traders vs midpoint | Lower (better execution) |
| Realized spread | MM's actual profit after price moves | More stable (less adverse selection) |
| Depth | Size available at top of book | Deeper (MMs post more size) |
| Fill rates | Percentage of limit orders that execute | Higher for non-HFT participants |
The logic is straightforward: if market makers get sniped less, they can afford to quote tighter and in larger size. Tighter quotes and deeper books mean better prices for everyone.
Who Loses?
The firms that lose are the latency arbitrage specialists — the ones whose primary strategy is racing to stale quotes. They can no longer profit from being 5 microseconds faster. This is a feature, not a bug. These firms extract value from the market without contributing price discovery or liquidity.
4. Symmetric vs Asymmetric Speed Bumps
Symmetric (IEX Model)
IEX delays all inbound messages equally — new orders, cancels, and modifications all go through the same 350μs delay.
Symmetric Speed Bump:
New Order → [350μs delay] → Matching Engine
Cancel → [350μs delay] → Matching Engine
Modify → [350μs delay] → Matching Engine
Everyone waits. No exceptions.Pros:
- Simple and transparent — same rules for everyone
- Harder to game (can't exploit cancel speed)
- Regulators like the fairness argument
- Already proven in production (IEX has been live since 2016)
Cons:
- Market makers can't cancel stale quotes any faster than snipers can hit them
- Only helps if the MM sees the price move before the sniper's order enters the bump
- Doesn't fully solve the problem if the sniper and the MM see the move at the same time
Asymmetric (TSX Alpha, Aequitas NEO)
Some exchanges delay only aggressive (taking) orders while processing passive orders (cancels, modifications, new resting orders) instantly.
Asymmetric Speed Bump:
Aggressive Order → [delay] → Matching Engine
Cancel → [no delay] → Matching Engine
New Passive → [no delay] → Matching Engine
Modify → [no delay] → Matching Engine
Makers get a head start.Pros:
- Directly protects market makers — they can always cancel before a sniper's order arrives
- More effective at reducing adverse selection
- MMs can quote tighter with more confidence
- Eliminates the "who sees it first" problem
Cons:
- Favors one side of the market (makers over takers)
- Can be gamed: submit passive orders that you intend to be aggressive (race to the front of the queue, then get crossed)
- Regulators and market participants debate fairness
- More complex to implement and explain
The Key Trade-off
| Property | Symmetric | Asymmetric |
|---|---|---|
| Fairness perception | Higher (same rules for all) | Lower (makers favored) |
| MM protection | Partial | Strong |
| Complexity | Low | Higher |
| Gaming risk | Low | Moderate |
| Spread improvement | Moderate | Larger |
| Regulatory acceptance | High | Debated |
The choice comes down to philosophy: do you want to slow everyone down equally, or do you want to explicitly protect the side of the market that provides liquidity?
5. What a 100ms Speed Bump Means
Scale Comparison
IEX speed bump: 0.35 ms (350 microseconds)
Human blink: ~150 ms
100ms speed bump: 100 ms
100ms is ~285x longer than IEX's delay.At 350μs, you're shaving the edge off HFT latency races. The delay is invisible to everyone except the fastest firms running co-located servers with FPGA-based trading systems.
At 100ms, you're in a completely different regime. Every trader on the planet can react in 100ms. A person sitting at home with a Python script and a decent internet connection can observe a price move and send a cancel within 100ms.
What Changes at 100ms
1. Market making becomes democratized
With IEX's 350μs bump, you still need institutional-grade infrastructure to market make effectively. At 100ms, a much wider range of participants can viably provide liquidity:
- Small quant shops running cloud-hosted bots
- Individual algorithmic traders
- API traders with basic monitoring scripts
- Even manually assisted semi-automated trading
The speed advantage of co-location and FPGA hardware becomes nearly irrelevant.
2. The book becomes much "stickier"
100ms is long enough that quotes don't flicker. Orders placed on the book tend to stay there because:
- MMs have ample time to manage positions
- No one needs to race to cancel (they have 100ms of protection)
- Less incentive for "ghost liquidity" that disappears on touch
This means the displayed book is more real — what you see is closer to what you can actually get.
3. Aggressive strategies get throttled
Any strategy that depends on reacting to public information faster than others gets much harder:
- Latency arb across venues becomes difficult (100ms is an eternity)
- Momentum-ignition strategies lose their edge
- Quote-stuffing becomes pointless
4. Second-order effects
| Effect | Implication |
|---|---|
| Retail gets better fills | MM quotes are more stable, spreads tighter, less phantom liquidity |
| Sophisticated takers get worse fills | Can't snipe stale quotes, can't arb as easily |
| Inventory risk changes | MMs hold positions for longer per quote cycle, need more capital |
| Cross-venue arb slows | Prices may diverge across venues for longer (100ms is a long time in markets) |
| Information incorporation slows | Book reflects new information with ~100ms lag vs microseconds |
5. The honest trade-off
A 100ms bump makes the exchange friendlier for makers and retail, but it also means prices update slower. In a fast-moving market, the book could be 100ms behind reality. For most retail traders, this doesn't matter. For institutional traders moving large size, it could mean the displayed price isn't the true price during volatile moments.
The question isn't whether 100ms is "better" or "worse" — it's who benefits and who doesn't:
Winners: Losers:
+ Market makers (less sniping) - Latency arb firms
+ Retail traders (better fills) - HFT stat-arb strategies
+ Small quant shops (can compete) - Cross-venue arb desks
+ Exchange (attracts MMs/retail) - Traders who need instant price discovery6. Speed Bumps in Crypto
The Latency Arbitrage Problem in Crypto Is Worse
Crypto markets have properties that make latency arbitrage even more profitable than in traditional markets:
- 24/7 trading — No opening/closing bells. Arb opportunities happen around the clock.
- Fragmented liquidity — Dozens of exchanges (Binance, Bybit, OKX, Coinbase, Backpack, etc.) all running separate order books for the same assets.
- Volatile assets — Crypto moves 5-10% days are common. More movement = more stale quotes to snipe.
- Unregulated co-location — No SEC rules about fair access. Exchanges can (and do) sell preferential latency.
- Global venues — Exchanges in different continents mean physical latency (speed of light) creates persistent arb windows.
BTC trades on 50+ venues simultaneously.
A price move on Binance creates stale quotes on every other exchange.
The fastest firm to reach each exchange captures the arb.
This happens thousands of times per day.CEX Approaches
Most centralized crypto exchanges don't have formal speed bumps, but some are experimenting:
| Approach | Description | Used By |
|---|---|---|
| Rate limiting | Cap messages per second per user | Most CEXs |
| Batch processing | Process orders in small time batches rather than continuous matching | Some newer exchanges |
| Random delay | Add random jitter (e.g., 1-5ms) to order processing | Experimental |
| Asymmetric fees | Charge takers more, pay makers rebates (economic speed bump) | Most CEXs |
| Maker protection windows | Short delay on aggressive orders only | Under discussion at several venues |
The fee structure (maker rebates / taker fees) acts as an economic speed bump — it doesn't prevent sniping but makes it less profitable. If the spread is 1bp and the taker fee is 2.5bp, the arb needs to be at least 2.5bp to be worth taking.
DEX / On-Chain Approaches
Decentralized exchanges face the same problem but with different constraints. Blockchain-native solutions include:
Batch Auctions (Frequent Batch Matching)
Instead of continuous matching, collect all orders over a time window and execute them all at a single clearing price.
Continuous Matching:
Orders arrive and match instantly. Faster = better.
Batch Auction (e.g., every 250ms):
t=0 Batch opens. Collect orders.
t=250ms Batch closes.
t=251ms All orders matched at single clearing price.
Arrival order within the batch doesn't matter.Examples: CoW Protocol (Ethereum), Penumbra, Frequent Batch Auctions research
Pros: Completely eliminates latency advantage within a batch. Fair price discovery. Cons: Adds latency for everyone. No continuous price. Harder to hedge against.
MEV Protection
MEV (Maximal Extractable Value) is the blockchain equivalent of latency arbitrage. Validators/sequencers can reorder transactions to front-run traders. Solutions include:
- Encrypted mempools — Orders are encrypted until included in a block (can't be front-run if you can't read them)
- Fair ordering — Protocols that enforce first-come-first-served at the consensus layer
- Private transaction submission — Send orders directly to block builders who commit to not extracting MEV (e.g., Flashbots Protect)
- Application-specific sequencing — DEXs run their own sequencer with anti-MEV rules
Commit-Reveal Schemes
Traders submit a hash of their order (commit), then reveal the actual order after a delay. Prevents anyone from seeing and front-running the order before it's included.
Hybrid Models
Some crypto platforms blend CEX speed with DEX fairness:
- Off-chain matching, on-chain settlement — Match orders on a centralized server (with speed bump if desired), settle to blockchain for finality
- App-chain DEXs — Run a dedicated blockchain for the exchange, control the sequencer, implement batch auctions or delays at the consensus level
- Intent-based systems — Traders submit intents ("buy 1 BTC at market"), solvers compete to fill them. No order book, no latency race.
7. Relevance to Backpack Exchange
Why This Matters for Backpack
Backpack operates a centralized order book exchange in a market where latency arbitrage is a significant source of toxicity for market makers. A well-designed speed bump mechanism could:
1. Attract and retain market makers
Market makers choose where to deploy capital based on the toxicity of flow they expect to face. An exchange with lower adverse selection lets MMs quote tighter and in larger size, which creates a virtuous cycle:
Less adverse selection
→ MMs quote tighter spreads and deeper size
→ Better execution for traders
→ More volume
→ More attractive for MMs
→ Even tighter spreads2. Differentiate from competitors
Most crypto CEXs have no formal speed bump. Implementing one (especially an asymmetric model that explicitly protects makers) would be a concrete differentiator that appeals to the firms providing the most liquidity.
3. Improve displayed book quality
With a speed bump, the quotes sitting on Backpack's book would be more real. Traders could trust the depth they see, reducing the "phantom liquidity" problem where displayed size evaporates the moment you try to trade against it.
4. Reduce the infrastructure arms race
Without a speed bump, MMs must invest in co-location, dedicated servers, and ultra-low-latency connections to Backpack's matching engine. With a speed bump, a wider range of participants can provide competitive liquidity, which is good for decentralization and market resilience.
Design Considerations
| Decision | Options | Trade-off |
|---|---|---|
| Bump duration | 350μs (IEX-like), 1-10ms (moderate), 50-100ms (aggressive) | Shorter = less market impact, longer = more MM protection |
| Symmetric vs asymmetric | Delay all orders vs delay only aggressive | Fairness perception vs MM protection strength |
| Per-market tuning | Same bump for all pairs vs different bumps for different volatility profiles | Simplicity vs optimization |
| Transparency | Published delay vs undisclosed | Trust and regulatory compliance vs gaming prevention |
| Scope | All participants vs exempt certain programs (e.g., designated MMs) | Simplicity vs targeted protection |
Practical Impact
For a crypto exchange like Backpack, even a modest speed bump (say 1-5ms) would meaningfully change the dynamics:
- Cross-exchange arb becomes harder — Firms that profit by racing Binance price moves to Backpack's book would face a meaningful delay
- Market maker P&L improves — Less adverse selection means MMs can offer better prices
- Spread compression — With more confident MMs, spreads tighten, reducing trading costs for all users
- Book depth increases — MMs can post larger size knowing they have time to manage risk
The speed bump is not a silver bullet. It needs to be paired with good fee structures (maker rebates), reliable infrastructure (low jitter, deterministic matching), and clear market maker programs. But as a microstructure tool, it's one of the most impactful levers an exchange can pull to improve order book quality.