Order Latency: What Happens Between Clicking and Getting Filled
The chart showed $80,000 when you clicked, but the fill was $80,050. This is often called slippage, yet a simpler cause comes first: the click and the fill happen at different times. The interval is latency.
The path from click to fill has several stages
An order passes through several stages. Each takes time, and their sum is the actual delay. Without knowing which stage is slow, you may try to fix the wrong thing.
① Input: from the displayed price to your click
② Transmission: your device → exchange server
③ Queue: waiting for the matching engine
④ Matching: execution against the order book
⑤ Response: the result returns to your screen
─────────────
You observe the start of ① and the end of ⑤.
The execution price is determined at ④.
You therefore click using a price from the recent past.
Stage ④ happens after stage ①. The screen displays a price the exchange sent earlier, and the book may have changed by the time your order reaches the matching engine. A different fill price is a consequence of this sequence, rather than necessarily an error.
What 0.4 seconds can cost
Milliseconds are difficult to picture. Converting delay into price movement makes it concrete.
BTC $80,000; quantity 0.1; notional $8,000
Assumptions: display refresh 0.15 seconds + transmission 0.10
+ queue/matching 0.05 + allowance 0.10
= 0.40 seconds
Assume price moves adversely by 0.03% during that interval:
80,000 × 0.0003 = $24
Cost to the position = 0.1 × 24 = $2.40
─────────────
10 trades per day → $24
20 trading days per month → $480
For comparison, a 0.10% round-trip fee is
8,000 × 0.001 = $8 per trade.
The assumed latency cost is about 30% of that fee.
Fees appear on statements, so traders notice them. Latency costs do not appear as a separate line and are easily treated as nonexistent. They can nevertheless accumulate with frequent trading. Read round-trip trading costs when calculating breakeven including this effect.
These figures illustrate the structure. Actual latency can vary severalfold with the connection, device, exchange and time of day.
Market and limit orders suffer differently
The same delay has different consequences for different order types. Distinguishing them helps identify an appropriate response.
Market order: the price moves away.
The ask has already risen when the order arrives.
→ A more expensive fill.
→ The cost is realized immediately.
Limit order: queue position moves back.
Other orders reach the same price first.
→ Your order joins the back of the queue.
→ It may never fill.
─────────────
In the guide's comparison, market orders prioritize entry at a worse price.
Limit orders preserve the price limit but may miss entry.
The limit-order cost is less visible: no fill means no recorded loss, yet missing an intended opportunity still has a cost. See limit-order queue priority for why another order fills first at the same price, slippage and ways to reduce it for adverse execution prices, and limit versus market orders for the choice between them.
Latency grows when speed matters most
The difficult feature is its distribution, rather than its average. Delays are often short in quiet conditions and long during sudden moves.
Normal
Latency 0.3 seconds; price movement during it 0.02%
→ Price difference $16 at BTC $80,000
Volatile
Order congestion increases latency to 1.5 seconds.
Price moves 0.40% during that interval.
→ Price difference $320
─────────────
The original comparison states five times the latency,
twenty times the interval's price movement,
and a twentyfold price effect.
Both delay and movement increase at the same time.
Stop exits often cluster in precisely these conditions. During a crash, many participants sell together while order processing slows and liquidity thins. A stop planned only around normal conditions may execute far below the intended level. See why conditional orders may not trigger for issues between activation and execution.
Measuring your own latency
Someone else's benchmark is not enough. Measure your own connection, device and exchange.
1) Submit 20 small market orders under the same conditions.
2) Record two timestamps for each:
· Submission time on your device
· Fill time reported by the exchange
3) Calculate differences and inspect their distribution.
─────────────
Look beyond the average:
· Median: the typical delay
· Maximum: the worst observed delay
Caution: the two timestamps use different clocks.
A misaligned computer clock distorts the measurement.
A computer clock off by a few seconds can produce negative or implausibly large latency. With an exchange API, query server time and account for the offset first; enable clock synchronization. Slow responses during a request surge may also involve request limits, so check API rate limits and 429 errors.
What can and cannot be reduced
Latency cannot reach zero. Breaking it into controllable and largely uncontrollable stages makes the practical steps clearer.
Often reducible
· Use a wired connection for greater stability than wireless.
· Close unnecessary browser tabs and programs.
· Reduce chart indicator load.
· Prepare quantity and price fields in advance.
Difficult to reduce
· Transmission time imposed by physical distance
· Internal exchange processing
· Queues during order surges
─────────────
The largest opportunity is often ① input.
Human decision and click time can be
much longer than technical latency.
Defining entry conditions and placing conditional orders beforehand can save more time than trying to trim 0.05 seconds from transmission. Watching, deciding and clicking commonly takes more than a second in the guide's comparison.
Design around latency
A practical response is to make the plan robust to delay rather than expect delay to disappear.
· Include a latency allowance in loss planning.
Planned stop 0.5% → assume actual execution at 0.6%.
· Place conditional entries and stops in advance.
This removes the manual click-reaction interval.
· Avoid strategies whose result hinges on a 0.1% difference.
Delay alone can reverse the outcome.
· Reduce quantity during sudden volatility,
when execution-price differences can grow.
─────────────
The guide's heuristic:
If latency costs exceed 10% of expected profit,
the strategy is too sensitive to delay.
If expected profit is $20 per trade and delay costs $5, execution conditions can undermine the strategy regardless of its idea. Longer holding periods or wider targets may reduce delay's relative contribution. This is a common source of divergence between backtests and live trading, so include execution costs during validation; see backtesting.
For bots, polling intervals can dominate latency
Automation removes human click time but can introduce another delay.
Price polling interval: 3 seconds
+ decision and order creation: 0.05 seconds
+ transmission and matching: 0.20 seconds
─────────────
Worst case: 3.25 seconds
Three seconds come from a setting.
· A three-second poll can learn of a change three seconds late.
· A live stream delivers changes as they occur in the guide's comparison.
This bot may be slower than a human click taking one second or more.
Developers can optimize code while overlooking the polling interval. Periodic queries inherently discover changes only at the next query. For a speed-sensitive strategy, inspect that interval first. See automated-trading expectations and reality for more.
Key points
② Latency stages: input, transmission, queue, matching and response.
③ Even 0.4 seconds can accumulate into fee-sized costs.
④ Market orders lose on price; limit orders lose queue priority.
⑤ Latency and volatility can rise together, making stop exits difficult.
⑥ Compare submission and fill timestamps; inspect median and maximum.
⑦ Different clock references can invalidate measurements.
⑧ The largest delay is often human decision and click time.
⑨ Plan for delay through loss allowances and advance orders.
⑩ A bot's polling interval can dominate its delay.
Decisions are always based on a price from slightly earlier. Account for the unavoidable interval in the plan. Choosing a strategy that does not compete over tenths of a second is another way to do so.
Caution
The delay durations, price movements, fee rates and PnL figures are hypothetical examples, not measurements from a particular exchange or asset. Actual latency varies substantially by connection, device, location, exchange and time, so measure your own conditions. Leveraged trading can lose all principal. Decisions and their consequences remain your responsibility.
NOONOO TRADING invites you to follow live trading in our free chat.
Start in the bot📈 OKX trading fee discount for new registrations
Register for the OKX Fee Discount →