Two ways a spread pays
Every spread trade is one of two kinds. In a spread deviation the spread sits near zero for weeks, then breaks away once: you open a single position and wait for it to converge. In HFT the spread swings inside a corridor, and the card opens and closes dozens of times. On a chart the two can look alike for the first hour, but they make money differently, lose it differently and need different settings.
Spread deviation
one trade- The spread breaks away once. You open one position and wait for it to converge.
- You hold for hours or days, through several funding settlements.
- You earn one spread, minus fees, slippage and funding.
- The main risk is a spread that never comes back.
HFT
dozens of cycles- The spread swings inside a corridor. The card opens at one edge and closes at the other, again and again.
- The card runs for hours or days, but each position lives minutes, so funding rarely reaches it.
- You earn the number of cycles times what one cycle clears after fees and slippage.
- The main risk is a small loss per cycle, repeated all day.
The rest of the knowledge base assumes you know which regime you are trading. Thresholds, order size, Depth, fees and funding mean something different in each, and a setting that is right for one can lose money in the other.
The whole article in one scheme
Each box opens the section that explains it.
why did it break?
Spread deviation: one trade, then wait
Take a pair whose spread has sat around zero for a month. Then something happens and the spread jumps. You open at +1.5% and close at +0.3% once it comes back. Between those two moments you hold a hedged position for hours or days, and every funding settlement on both exchanges passes through it. The chart below shows nine of them.
profit = spread − 4 fees − slippage − funding × N- spread
- What you capture between entry and exit: 1.5% − 0.3% = 1.2% in the example.
- 4 fees
- Taker fee on both legs at the open, and again at the close.
- slippage
- How far your fills land from the prices you saw, on both legs.
- funding × N
- Net funding on the two legs, paid at every settlement while you hold. N counts settlements, not days: three days on an exchange that settles hourly is 72 of them, each at the rate that exchange quotes for the symbol.
Funding can also be paid to you. Count it as a cost until the rates on both exchanges show otherwise.
Three things turn this trade into a loss.
Funding eats the income
The longer the spread takes to come back, the more settlements you pay. Hold long enough and funding takes the whole spread. The calculator below shows how fast that happens.
The spread never converges
If the cause is structural, there is no level for the spread to return to. The next section sorts the causes you can trade from the ones you can't.
One leg gets liquidated
Margin sits separately on each exchange. While the spread widens against you, the losing leg moves toward liquidation even though the pair as a whole is hedged. See Risk Management.
Spread deviation
About five weeks on one pair: a month near zero, then a single trade.
How long can you wait?
The 1.2% from the example, minus fees, slippage and net funding of 0.02% against you per settlement, settled every 8 hours.
left after everything
+0.79%
Funding settlements: 3
- you keep
- +0.79%
- funding
- −0.06%
- slippage
- −0.15%
- 4 fees
- −0.20%
After one day funding is a rounding error.
left after everything
+0.67%
Funding settlements: 9
- you keep
- +0.67%
- funding
- −0.18%
- slippage
- −0.15%
- 4 fees
- −0.20%
Three days: funding has taken a fifth of what fees and slippage left you.
left after everything
+0.43%
Funding settlements: 21
- you keep
- +0.43%
- funding
- −0.42%
- slippage
- −0.15%
- 4 fees
- −0.20%
A week: funding has taken half.
left after everything
+0.01%
Funding settlements: 42
- you keep
- +0.01%
- funding
- −0.84%
- slippage
- −0.15%
- 4 fees
- −0.20%
Two weeks: funding has taken everything. The spread came back and you earned nothing.
On an exchange that settles every hour, three days is already 72 settlements. Each costs the rate that exchange quotes for the symbol, often smaller per hour.
Enter or not: the cause decides
A spread that broke away is not a trade yet. Before you enter, find out why it broke. The causes fall into three groups, and each group has its own answer.
Event
trade it- News or a liquidation cascade swept the order book on one exchange.
- The exchange itself failed: matching engine, API, a trading halt. Trade the recovery once it is back: a frozen quote is not a price.
- The skewed exchange settles funding every hour, so the pressure that pulls the price back arrives eight times as often as on an 8-hour schedule.
- The two perps follow indexes built from different spot markets, so a move on one spot venue shows up on one perp only.
Capacity
cut the size- The book is thin: the spread is live for only $50.
- The exchange caps the total position you can hold.
- Arbitrageurs' margin is already tied up in other pairs.
- A market maker pulled their size, and holding the position became risky.
Structure
don't touch- Delisting: the contract usually goes close-only first, sometimes only half an hour before it is settled.
- Fresh listing: there is no common price yet.
- Deposits and withdrawals of the coin are frozen.
- Different contracts trade under the same ticker.
- A steady difference in funding is an equilibrium, not a deviation.
An event spread closes as the order book recovers, and that recovery is what you trade. A capacity spread is real but small: trade it at the size the book actually carries. A structural spread looks the same on the chart and never closes. Delist Detection catches the most common case for you.
Same jump, three stories
What the spread and the depth under it do after the break.
Event
trade itdepth under the spread
The order book refills in front of you.
Capacity
cut the sizedepth under the spread
The spread is there. The depth under it is not.
Structure
don't touchdepth under the spread
There is no level to return to.
HFT: dozens of cycles in a corridor
Here the spread swings inside a corridor. In the chart below the card enters at +0.8% and exits at +0.2%, and one cycle takes about three minutes: 13 cycles in 40 minutes. The card may run for hours or days, but each position lives only a few minutes, so funding rarely gets a chance to touch it.
profit = cycles × (spread − 4 fees − slippage)- cycles
- How many times the spread crossed the band and came back. Signal Strength measures exactly this.
- spread
- Entry minus exit: 0.8% − 0.2% = 0.6% in the example.
- 4 fees
- Taker fee on both legs at the open, and again at the close.
- slippage
- Measured at your order size, not at the best quote. This term decides whether the whole bracket is positive.
Funding is left out on purpose: a position that lives three minutes rarely sits through a settlement. On an hourly symbol a busy day still catches one or two, and a position open at that second pays the full rate.
Losses here come in small pieces, which is why they go unnoticed.
A losing cycle, 40 times over
If one cycle loses 0.05% after fees and slippage, the card does not lose 0.05%. It loses that on every cycle, all day.
Volume for a VIP tier is a cost
Cycles that lose a little to climb a fee tier are a payment for that tier. Count them as an expense, not as trading income.
The corridor ended, the card did not
When the spread moves to a new range or stops swinging, a card set for the old corridor keeps opening into it, and an HFT position turns into a spread deviation trade nobody planned. Auto Mode moves the band with the spread and recalculates its width every five minutes.
The spread was counted at the best quote
The spread you see is at the top of the book, and slippage at your size can eat all of it. Depth Multiplier prices the spread at your size before the card trades.
Thirteen cycles, three outcomes
Each cycle: 0.6% between entry and exit, minus 0.2% in fees, minus slippage at your size.
About 40 minutes: 13 cycles of roughly three minutes, too short for funding to matter.
running total, % of position
13 cycles × +0.30% = +3.90%
The chart and the account agree: every cycle clears its fees and slippage.
About 40 minutes: 13 cycles of roughly three minutes, too short for funding to matter.
running total, % of position
13 cycles × −0.05% = −0.65%. A 40-cycle day: −2.00%.
The spread chart is the same as in the healthy corridor. Only the fills differ: 0.45% of slippage at your size turns every cycle negative.
running total, % of position
7 cycles × +0.30% = +2.10%, then a position stuck at −0.90% with no exit in sight.
After cycle 7 the spread moved to a new level. The card opened at +0.8% as usual and is now holding a spread deviation trade nobody planned.
Size orders to the depth
This section and the next apply to both regimes. A card builds its position in sub-orders: each one is one Order Size on both exchanges, and Max Orders sets the ceiling. How you split the position decides the price you get.
Parts stay near the best price
Filled in parts, each order takes only the top of the book while it refills, so your average entry (VWAP) stays near the best price. One big order walks every level down to the worst one the moment the spread first touches your threshold.
Many small orders are safer
They give a better entry price, and each one puts less at stake: if an exchange rejects it, only one Order Size is left without a hedge.
Low-cap books refill
On a low-cap pair the book has time to fill back up between your orders. Order Cooldown sets how long the card waits between them.
No trades? Cut the size, not the depth
If the card stops trading on a thin book, lower Order Size and leave the Depth multiplier alone. A lower multiplier gets the card trading again, but at prices the depth check no longer covers: you swap no trades for bad fills.
Big orders fail on one leg
A big order is the one that gets rejected on one exchange and fills on the other, leaving the filled leg as an unhedged position.
Cut the size to the depth of the book, not to the type of pair.
On the card this comes down to four fields.
Order Size
Small enough that one sub-order fits into the top levels of the thinner book.
Max Orders
Raise it to match, so the full position still gets built, only from smaller pieces.
Order Cooldown
Long enough for a low-cap book to refill between orders.
Depth
Leave it where it is. It checks every sub-order against the real book, and lowering it to get trades back is the mistake described above.
One order or five
Buying 1,000 coins from the same ask side of a low-cap book.
One order of 1,000
- average price
- 1.0066 +0.66% above the best ask
- last fill
- 1.0200 +2.00%
- left open if the other leg is rejected
- 1,000 coins
Five orders of 200, with a pause
- order 1 1.000
- order 2 1.001
- order 3 1.000
- order 4 1.002
- order 5 1.001
- average price
- 1.0008 +0.08% above the best ask
- last fill
- 1.0010 +0.10%
- left open if the other leg is rejected
- 200 coins
The pause between orders is the card's Order Cooldown.
Stand as narrow as you can
Narrow means Upper and Lower sit close together. Each cycle earns less, but the spread crosses the band more often and the card trades more.
The flywheel
The narrower you stand, the more you trade. More volume earns a higher VIP tier and lower fees, and lower fees let you stand narrower still. It keeps turning only while each cycle still clears its fees: volume bought with losing cycles is the cost from the HFT section.
A rebate beats a tier
If you can get a fee rebate, use that first: its rate does not depend on how much you trade. See Exchange Fees.
Narrow exits fast
With a narrow band, Exit Mode On Loss finishes in hours rather than days: open layers reach the exit level sooner, as long as the spread still swings back to it.
Narrow is measured against the pair's own noise, not as an absolute number. A low-cap corridor is wider, so narrow starts at a larger spread there.
Small orders and a narrow band both add orders, and every order pays a fee. Both pay off only once your fee tier is sorted out.
Why narrow pays, and what narrow means
The flywheel
rebate: lower fees without the volume
Narrow, relative to the pair's noise
Both bands are narrow for their pair. On the low cap, narrow simply starts higher.