AI 中文总结
研究中心化加密货币交易所中算法交易模式，提出新方法推断单向套利交易，在币安和Kraken交易数据中识别出大量相关序列，估算出利润，发现OWA速度变快但单个序列盈利能力下降，凸显定价差异并引发后续研究思考。
AI 中文摘要
中心化加密货币交易所（CEXes）实现了数百种加密货币之间快速的链下兑换。这些平台上出现何种算法交易模式仍是一个开放问题。衡量CEXes的一个主要挑战是其公开交易数据不包含可用于关联的地址或交易商标识符。我们提出了一种新颖的方法，用于从CEXes的匿名现货交易数据中推断单向套利（OWA）交易。我们在币安5年的交易中识别出402M个可能的OWA序列（在Kraken 9年的交易中识别出近2M个），分别占总交易量的0.94%和0.13%。我们估计币安的总利润为3120万美元，Kraken为97.5万美元，扣除交易费用后，单个OWA序列的平均利润不到1美元。我们还观察到OWA随着时间推移变得更快，而单个序列的盈利能力下降。我们的发现凸显了CEXes中经常出现定价差异，并为未来工作提出问题，以确定实现有利可图OWA的精确情况。
英文摘要
Centralized cryptocurrency exchanges (CEXes) enable fast off-chain conversions between hundreds of coins. It is an open question which algorithmic trading patterns occur on these platforms. A major challenge to measuring CEXes is that their public trade data does not contain addresses or trader identifiers allowing linkage. We propose a novel methodology to infer one-way arbitrage (OWA) trading in anonymized spot trade data from CEXes. We identify 402 M likely OWA sequences in 5 years of trading on Binance (and almost 2 M during 9 years on Kraken), accounting for 0.94 % and 0.13 % of the total traded volume, respectively. While we estimate total profits of $31.2 M on Binance and $975 k on Kraken, profits from individual OWA sequences are less than $1 on average after accounting for trading fees. We also observe that OWA has become faster over time, while the profitability of individual sequences has decreased. Our findings highlight that pricing discrepancies regularly occur in CEXes, and raise questions for future work to identify the precise circumstances that enable profitable OWA.
