Separating Tradable Crypto Price Gaps from Stale Quotes
A crypto arbitrage system must decide in under 100 milliseconds whether to buy an asset on one exchange and sell it on another, with positive expected profit after fees and worse-than-quoted execution prices, even for small trades. Quotes arrive with different, unpredictable delays, so the apparent price gap may never have existed at the same moment. Available orders can disappear before execution, and the buy and sell execute independently: if only one fills, the trader is left exposed to price changes. Checking more data reduces false signals but takes time during which other traders can capture the gap or prices can converge.
Borrowing passive sonar from submarine hunting, we read public exchange data before sending orders to judge whether a price gap will survive long enough to trade profitably.
How it works
Buying crypto on one exchange and selling it for more on another looks simple until the quoted prices disappear. Even a small test order can prompt market makers to withdraw their offers. We borrow the submarine hunter’s practice of listening before sending a revealing sonar ping. The system first checks public trading data, accounting for differences in feed delay, to separate gaps caused by buying and selling from apparent gaps caused by news reaching exchanges at different times. It then estimates whether available liquidity is stable, thinning, withdrawing or absent from how quickly quotes change and depleted orders are replaced. That estimate gives a time window before quoted depth disappears. The system trades only when the estimated chance of completing both trades produces a positive expected profit after fees, slippage and the cost of reversing an unmatched trade. It reserves small test orders for uncertain cases with enough time left to act. Compared with a simple price-gap trigger, this approach checks whether the opportunity is likely to remain executable before risking money.
Borrowing delay tracking and linked traffic limits from multipath networking, we estimate how much of a cross-exchange price gap comes from stale quotes and size trades around the chance of completing both sides profitably.
How it works
Buying on one exchange and selling on another only pays if both prices are still available when the orders arrive. With feeds arriving at different speeds, an apparent bargain can come from comparing an old price with a new one. We borrow from multipath networking, where engineers manage connections across routes with different delays. The engine tracks each feed’s delay, blocks trades on unreliable feeds, and uses delay uncertainty and recent price volatility to estimate the chance that a price gap survives after costs. It also predicts how much displayed buying and selling capacity will remain when orders arrive. Before trading, it weighs all four outcomes: both orders complete, only the buy completes, only the sell completes, or neither completes, including the cost of closing an unwanted position. Linked trade-size limits grow after successful paired trades and shrink across affected pairs after misses. Compared with a fixed minimum price gap, this approach can admit small, credible opportunities while rejecting larger gaps caused by stale data.