Arbitrage Exists, but Not at Retail: Measuring the Retail Accessibility of Cryptocurrency Arbitrage

Abstract

We measure whether cryptocurrency arbitrage is accessible to a retail participant using commodity infrastructure, and report primarily negative results. Streaming Binance best-bid/offer data for 2.74 days (66,476,148 ticks) across five connected spot markets, an event-driven negative-cycle detector found zero executable triangular-arbitrage cycles once an execution-realistic simultaneity constraint (0.2-second maximum quote-age spread) was imposed. A detector without this constraint over-reports phantom opportunities—here up to +51.6 basis points—assembled from non-simultaneous quotes, reconciling prior work that finds arbitrage captured by co-located bots with prior work that finds it unexploitable at retail. On the funding-carry track, a two-year backtest over 151 perpetual-futures symbols reproduces a pitfall: a "no opportunity" conclusion under a curated major-coin universe reverses to positive validation-window profit under a volume-ranked broad universe, so the conclusion is driven by universe selection. A post-hoc placebo test (N = 1000, seed = 42) then finds the funding entry rule's profit not distinguishable from random entry timing (37.8th percentile; one-sided p = 0.622): in attribution terms, the return is beta to a carry factor, not alpha from timing skill. In our sample, arbitrage-like returns exist but are not accessible to retail as a skill-based edge.