Short answer: A crypto arbitrage bot watches the same or linked assets on several markets and trades when a price gap exceeds the full cost of capturing it. That cost includes taker fees on every leg, spread and slippage, transfer or rebalancing costs and, for futures, funding. Most visible gaps shrink or vanish once those costs and order book depth are counted, and what remains tends to go to the fastest, lowest-fee participants. Arbitrage carries execution, inventory and counterparty risk and can lose money.
Calculate it with your own figures: Crypto trading fee calculator and Crypto arbitrage calculator.
Price differences for the same coin across exchanges are easy to spot, and they invite an obvious idea: buy where it is cheap, sell where it is dear, and let a bot do it around the clock. Research complicates the conclusion. Makarov and Schoar, writing in the Journal of Financial Economics in 2020, found periods of large, recurrent arbitrage opportunities across crypto exchanges, with deviations much larger across countries than within them, and linked this to capital controls that limit how freely arbitrage capital can move: the largest gaps sat where capital could least easily cross.
This guide explains how arbitrage bots work, the main types, the full cost stack with fees taken from official pages as of October 2026, a tested worked example, and the operational and regulatory limits. It does not recommend any exchange or strategy.
How does a crypto arbitrage bot work?
A crypto arbitrage bot runs a loop: stream order books from several markets, compute the executable edge for a given size after all costs, send both legs at once when the edge clears a threshold, then reconcile and rebalance.
- Data. Order books arrive over WebSocket from each venue and must be kept in sync (see maintaining a local order book). Symbols, tick sizes and minimum order sizes differ by exchange and need normalizing, as described in crypto exchange API integration.
- Edge calculation. The bot uses the ask it would buy at and the bid it would sell at, walked through the book for the intended size, never the last traded prices.
- Execution. Both legs go out simultaneously from balances already held on each venue. Order conditions keep legs from resting on the book: in Binance's API, an IOC order fills as much as it can before expiring, and an FOK order expires if it cannot be filled in full.
- Leg management. If one leg fills and the other does not, a predefined rule closes or hedges the open position.
- Rebalancing. Inventory drifts toward one venue and has to be moved back, at a cost.
Speed decides who captures the gap. A Richmond Fed summary of research by Budish, Cramton and Shim reports that the median duration of arbitrage opportunities between E-mini S&P 500 futures and the SPY ETF fell from 97 milliseconds in 2005 to 7 milliseconds in 2011, while their profitability stayed constant.
Types of arbitrage trading bots
Four types dominate crypto: cross-exchange, triangular, spot-futures basis and funding-rate arbitrage; statistical arbitrage is sometimes added but is not arbitrage in the strict sense.
| Type | Mechanics | Main risk |
|---|---|---|
| Cross-exchange | Buy where the asset is cheaper, sell where it is dearer, from balances held on both | Leg risk, inventory imbalance, transfer cost and time |
| Triangular | Three trades around a loop on one exchange, such as USDT to BTC to ETH to USDT | Three sets of fees; prices move between legs |
| Spot-futures basis | Buy spot and sell a dated future priced above fair value, hold to expiry | Margin and liquidation on the short leg, capital cost |
| Funding rate | Hold spot long and a perpetual short to receive funding while the rate is positive | The rate turns negative; the short leg is liquidated |
| Statistical | Trade deviations between historically related assets | The relationship breaks and both legs lose |
Funding arbitrage depends on exchange rules and on how funding rates work. As of October 2026, Binance states that funding on its futures occurs by default every 8 hours, at 00:00, 08:00 and 16:00 UTC, that the interval can change during extreme volatility, that longs pay shorts when the rate is positive, and that the payment equals the nominal value of the position times the funding rate. Binance does not charge fees on funding payments, but entering and exiting both legs incurs trading fees, and the rate can change sign.
The full cost stack of cross-exchange arbitrage
The visible gap must pay for at least five costs, and the first two alone can exceed it.
| Cost | What it is | Notes |
|---|---|---|
| Taker fees | Charged on each leg | Binance lists 0.100% for regular spot users; Kraken lists 0.80% spot taker at its entry tier, falling to 0.18% from 1 million US dollars of 30-day spot volume |
| Spread and slippage | Your size walks the order book | The edge at the best price is not the edge at your size |
| Withdrawal and network fees | Fixed amounts per asset and network | Weigh heavily on small trades; check the live fee pages |
| Transfer time | Deposits wait for confirmations | Binance Academy notes confirmations can take 30 minutes or more depending on network traffic |
| Inventory and capital | Pre-funded balances on every venue | Capital sits idle and is exposed to each platform |
Because transfers are slow, arbitrageurs usually keep funded accounts on several exchanges and trade both legs at once. That removes transfer risk during the trade but replaces it with inventory imbalance: one venue accumulates the asset, the other accumulates cash, and rebalancing costs fees and time. Futures-based strategies add margin and funding that can turn against the position.
Is crypto arbitrage profitable after costs? A worked example
Only if the gap still exceeds every cost at your size and fee tier, and nothing guarantees that: a visible gap of half a percent can become a loss at the wrong fee tier or the wrong size.
The Python example below walks two synthetic order books. The best ask on exchange A is 100.00 USDT and the best bid on exchange B is 100.50 USDT, a visible gap of 0.50%. The taker fee on A is 0.10%; B is tested at 0.10% and 0.80%, the published Binance regular rate and Kraken entry-tier taker rate. Prices and depth are hypothetical.
def fill_price(levels, qty):
"""Average price for qty, walking the book level by level."""
left, total = qty, 0.0
for price, size in levels:
take = min(left, size)
total += take * price
left -= take
if left <= 1e-12:
return total / qty
raise ValueError("not enough depth")
# Synthetic books: buy on A (asks), sell on B (bids)
asks_a = [(100.00, 50), (100.06, 100), (100.15, 200)]
bids_b = [(100.50, 40), (100.40, 100), (100.25, 200)]
def net_edge(qty, fee_a, fee_b):
cost = fill_price(asks_a, qty) * qty * (1 + fee_a)
proceeds = fill_price(bids_b, qty) * qty * (1 - fee_b)
return proceeds - cost, (proceeds - cost) / cost * 100
for fee_b in (0.001, 0.008): # taker fee on B
for qty in (40, 140, 300):
pnl, pct = net_edge(qty, 0.001, fee_b)
print(f"fee_b={fee_b:.3f} qty={qty:>3}: {pnl:+8.2f} USDT ({pct:+.2f}%)")Output:
fee_b=0.001 qty= 40: +11.98 USDT (+0.30%)
fee_b=0.001 qty=140: +26.53 USDT (+0.19%)
fee_b=0.001 qty=300: +11.37 USDT (+0.04%)
fee_b=0.008 qty= 40: -16.16 USDT (-0.40%)
fee_b=0.008 qty=140: -71.89 USDT (-0.51%)
fee_b=0.008 qty=300: -199.33 USDT (-0.66%)At low fees the net edge falls from 0.30% to 0.04% as size grows, because larger orders consume worse price levels; in these synthetic books the largest result is at a medium size, which is not a sizing suggestion. At the higher fee every size loses. Rebalancing transfers, latency and failed legs, all ignored here, make real results worse.
Hypothetical triangular check: 10,000 USDT buys BTC at 60,000, the BTC buys ETH at 0.05000 BTC, and the ETH is sold at 3,010 USDT, returning 10,033.33 USDT: a gross gap of 0.33%. With a 0.1% fee on each of the three trades the result is 10,003.26 USDT, or 0.03%. Had ETH/USDT been 3,005, the loop would end at 9,986.65 USDT, a 0.13% loss. At 0.1% per leg the break-even gap is about 0.30% (1 / 0.9993 is roughly 1.0030), before any slippage between legs.
Latency, inventory and operational risk
Arbitrage looks market-neutral on paper, but in operation it carries leg, latency, inventory and platform risk.
- Leg risk. One order fills and the other does not, leaving a directional position at the moment prices are moving.
- Latency. Faster participants close the gap first, and reconnects create blind spots: Binance documents that a market data connection is valid for only 24 hours, so the bot must expect disconnects and stop trading on stale books.
- Downtime and suspensions. Maintenance windows and paused withdrawals can strand inventory on the wrong venue exactly when gaps are widest.
- Counterparty risk. A balance on an exchange is a claim on that exchange. In December 2022 the SEC charged FTX's founder, alleging the undisclosed diversion of FTX customers' funds to his crypto hedge fund, Alameda Research.
Theory anticipates these limits. Shleifer and Vishny's work on the limits of arbitrage describes arbitrage as carried out by relatively few specialized investors using other people's money, and shows that it can become ineffective in extreme circumstances. The recurrent cross-country gaps documented by Makarov and Schoar fit the same picture: gaps last where capital cannot move freely.
Monitoring vs execution, and the regulatory side
Start with monitoring: a system that measures executable edge after costs over several weeks tells you whether execution is worth building at all.
A monitoring system records order books from each venue, computes net edge at several order sizes using your actual fee tier, and logs how often opportunities appear and how long they last. Historical spot order book data is not part of Binance's public data collection, whose spot files cover trades, aggregated trades and candles (as of October 2026), so a recorder usually has to run before an honest backtest is possible. A scenario of this kind is described in the multi-exchange arbitrage monitoring project. Execution then adds inventory limits per venue, a cap on unhedged leg exposure, a kill switch and reconciliation, covered in trading bot risk management. This guide explains how arbitrage works; for the software itself, see what building an arbitrage bot involves.
Which platforms you may use depends on your jurisdiction and each exchange's terms. In Turkey, Article 99/A of Capital Markets Law No. 6362 treats activity by foreign crypto-asset platforms directed at Turkish residents as unauthorized. Never trade against your own orders to manufacture volume: wash trading is market abuse, and exchange APIs offer self-trade prevention settings.
Is crypto arbitrage legit? Warning signs in arbitrage software
Arbitrage itself is a legitimate trading technique, but an offer that promises a return from crypto arbitrage software is a warning sign, whatever it is called.
A January 2024 CFTC release lists crypto-asset arbitrage algorithms among the AI-assisted tools scammers promote and says that promised high returns, with or without a guarantee, are red flags of fraud. Labels change nothing: free crypto arbitrage software still needs your capital and often your API key, and a paid VIP tier does not make a gap any wider after the costs above.
- A stated return. Daily or monthly percentages or a 100% win rate. Net edge depends on fees, depth and speed, so no seller can know it in advance.
- Deposits to the seller. Software that needs your funds sent to its own wallet or platform is holding your money, not running your rules.
- Withdrawal rights. A vendor has no need for a key that can move funds out; see API key security.
- Strangers promoting it. The CFTC says such claims from strangers online should be ignored.
- A history you cannot check. Screenshots and dashboards prove little. The CFTC suggests researching the company or trader, checking how old the website's domain is and getting a second opinion.
In the United States the CFTC takes complaints at cftc.gov/complaint and the FBI at ic3.gov; elsewhere, use your national regulator. If you want to test whether an edge exists at all, the monitoring approach above costs far less than any software promising one.
Frequently asked questions
Is crypto arbitrage legit?
The technique is legitimate, but software sold with promised returns is a fraud warning sign. A January 2024 CFTC release lists crypto-asset arbitrage algorithms among the tools scammers promote and calls promised high returns, with or without a guarantee, a red flag.
Is crypto arbitrage risk-free?
No. Fees, slippage, failed legs, transfer delays, exchange downtime and counterparty failure can each turn a visible gap into a loss.
Why do arbitrage opportunities disappear so quickly?
Many automated participants watch the same prices and trade the gap away. In traditional markets, research by Budish, Cramton and Shim found the median duration of ES-SPY arbitrage opportunities fell to 7 milliseconds by 2011.
Can I run arbitrage without moving coins between exchanges?
Yes, by holding balances on both venues and trading both legs at once. The cost is capital tied up on each platform and periodic rebalancing transfers.
Is triangular arbitrage easier for a bot?
It avoids transfers because all three trades happen on one exchange, but the gap must cover three fees: about 0.30% at 0.1% per trade, before slippage between legs.
Is crypto arbitrage legal?
Buying on one market and selling on another is ordinary trading, but which platforms you may use depends on your jurisdiction and each exchange's terms. Wash trading and other manipulation are prohibited.
Sources
- Makarov, Schoar: Trading and arbitrage in cryptocurrency markets (Journal of Financial Economics, 2020) (opens in a new tab)
- Shleifer, Vishny: The Limits of Arbitrage (NBER Working Paper 5167) (opens in a new tab)
- Federal Reserve Bank of Richmond: Research Spotlight on Budish, Cramton, Shim (Econ Focus, 2015 Q3) (opens in a new tab)
- Binance: Trading Fee Rate (spot, regular user) (opens in a new tab)
- Kraken: Fee schedule (spot crypto tiers) (opens in a new tab)
- Binance: Introduction to Binance Futures Funding Rates (opens in a new tab)
- Binance Academy: Arbitrage (glossary) (opens in a new tab)
- Binance Spot API: Enum definitions (timeInForce, order status) (opens in a new tab)
- Binance Spot API: WebSocket market streams (opens in a new tab)
- Binance Public Data: README (data.binance.vision) (opens in a new tab)
- SEC press release 2022-219: SEC Charges Samuel Bankman-Fried with Defrauding Investors in Crypto Asset Trading Platform FTX (opens in a new tab)
- CFTC Release 8854-24 (January 25, 2024): customer advisory on AI investment scams (opens in a new tab)
- Capital Markets Law No. 6362, Article 99/A (Turkish) (opens in a new tab)
This guide is general information about software and markets, not investment, legal or tax advice. Automated trading can lose money, past or backtested performance does not guarantee future results, and the rules that apply depend on your jurisdiction and your exchange's terms. Sources, checks and corrections: see the editorial policy.