Workspace / Research tools
Guide

Why crypto arbitrage spreads disappear

An arbitrage scanner answers a narrower question than many traders assume: which observed prices differ? To evaluate an actual trade, you need to know how much can be bought and sold, whether both quotes still exist, and what the complete operation costs. A visible spread is the beginning of that calculation.

Explicit assumptionsFormula & methodology includedNo account required

Start with executable sides of the market

Buying consumes asks; selling consumes bids. Comparing the latest trade on one exchange with the latest trade on another mixes two historical events. Those trades may have occurred at different times and sizes. Even a comparison of current best ask and best bid applies only to the quantity resting at those prices.

Binance documents its depth response as price-and-quantity levels. That is the information needed to estimate an order's average fill. Read enough levels to cover the proposed base-asset quantity, then calculate the buy cost and sale proceeds separately. If the returned book cannot cover either leg, label the estimate incomplete instead of extending the last price indefinitely.

A positive headline spread can produce a negative result

Consider an illustrative purchase of 0.1 BTC at an average ask of 60,000 USDT and a simultaneous sale of 0.1 BTC at an average bid of 60,120 USDT. The gross difference is 12 USDT, or 0.20% of the 6,000 USDT purchase value. These are invented example prices, not current quotes.

At an assumed taker fee of 0.10% on each leg, the purchase costs 6 USDT in fees and the sale costs 6.012 USDT. The result is already negative by 0.012 USDT before latency, withdrawal charges or inventory rebalancing. Another 3 USDT of estimated execution cost takes the result to minus 3.012 USDT. Actual fee tiers and the asset used to pay fees must replace these assumptions.

Transferring after buying creates a different exposure

A pre-funded route starts with quote currency on the buy venue and BTC on the sell venue. It can attempt both legs without waiting for an on-chain transfer. After execution, its balances have moved in opposite directions, so repeated trades eventually require rebalancing or a profitable reverse route.

Buying first and transferring the purchased BTC before selling leaves the sale price unknown during the transfer. A quoted spread does not reserve that future price. Withdrawal processing, required confirmations and deposit availability therefore belong in the trade model, alongside network compatibility and the cost of restoring usable balances.

Measure what your estimate leaves unresolved

Save each source timestamp, snapshot receipt time, intended size, estimated average fills and assumed costs. Reject materially mismatched snapshots. A rapidly changing book can invalidate even a recently retrieved estimate, and a partial fill on one exchange can leave an unmatched position on the other.

A useful comparison is net proceeds for several sizes: 100, 1,000 and 10,000 USDT, for example. The smallest route may clear its costs while a larger route consumes worse prices. Use a break-even spread calculator to identify the required threshold, then a depth calculation to test whether enough quantity is actually visible above it.

Questions about this tool

Is a larger percentage spread always a better opportunity?

No. It can reflect thin liquidity, stale information, different quote currencies or restrictions on moving balances. Compare executable quantity and total costs before comparing percentages.

Does pre-funding eliminate arbitrage risk?

It removes the need to transfer between the two immediate trade legs. It does not eliminate partial fills, exchange exposure, balance requirements or future rebalancing costs.

Sources and further reading

Binance Spot API: order-book prices and quantities ↗Coinbase Exchange: product-book levels and snapshots ↗

Find a tool or explanation