Trading and Arbitrage Across Crypto Exchanges: Large Gaps That Persist
Can a crypto exchange spread be captured after transfer, fees, and execution delay?
Compute the apparent spread at synchronized timestamps, then subtract both exchanges? fees, realistic depth-based execution, transfer cost and settlement delay. A quoted spread is not captured profit when inventory cannot be moved or simultaneously hedged.
Observed cross-exchange spreads depend on transfer, execution and settlement frictions.
Evidence map
| Aspect | Finding |
|---|---|
| What it is | The systematic measurement of price differences for the same cryptocurrency across exchanges and across countries, and of why they are not immediately competed away. |
| Key result / formula | Makarov and Schoar collect order book and trade data from many exchanges in several countries and document price deviations that are large and persist for extended periods, far beyond anything observed in mature markets. |
| Why it matters for backtesting | For a user of Stochastly the practical consequence comes before the trading idea: the price series depends on which venue produced it, and two providers' bars for the same pair are not the same series. |
What it is
It is the paper that turns the folklore about the Korean premium into a measured fact with an explanation.
Key result / formula
The deviations are much larger across countries than within a country, and they move together across currencies within a country, which points at the capital flow between fiat systems rather than at anything specific to a coin. The size of the deviation is related to the difficulty of moving money in and out of the local banking system, and episodes such as the Korean premium coincide with periods of tighter capital controls. Within-country arbitrage is easier and the corresponding deviations are smaller but still present, limited by the time to move coin between venues and by counterparty and withdrawal risk. The authors also document that order flow is largely idiosyncratic to each exchange, which is consistent with segmented markets.
Why it matters for backtesting
A strategy calibrated on one venue's bars and traded on another inherits a basis that can exceed the strategy's entire edge. The check is available to anyone with two sources: align the two bar series in the same time zone and plot the difference, which measures the basis directly and shows whether it is stable. The arbitrage itself is not testable in Stochastly, because it requires simultaneous positions on two venues, transfer times and withdrawal limits, none of which a bar-based backtest represents; an agent asked about it should state this limitation and should not produce a backtest of an untradable spread. The transferable lesson is about data provenance, which belongs with each crypto result Stochastly produces.
Source
Makarov & Schoar, "Trading and arbitrage in cryptocurrency markets", Journal of Financial Economics 135(2), 2020, 293-319. Primary source