Inter-exchange arbitrage is not about predicting prices, but about extracting profit from the difference in quotes of the same asset across different platforms. While one exchange prices a coin slightly lower and another slightly higher, an opportunity arises: buy on the first and sell on the second. In practice, however, it all comes down to speed: a typical spread is 0.01–0.5% and lasts only seconds. A human physically cannot react to such a window, so the task falls to algorithms.
How Arbitron works
Arbitron is a platform that automates the entire cycle of inter-exchange arbitrage. Instead of manually monitoring order books, the service scans about 10,000 trading pairs on 20 exchanges simultaneously. The key difference from simple screeners is accounting for trade feasibility. The raw quote difference is just the tip of the iceberg. Real profit is eaten away by four factors: taker fees on both legs, order book depth for the required volume, slippage during execution, and the decay of the spread itself. The platform calculates the price at which an order will actually execute for $25, $100, or $1000, rather than just showing a nice number on the screen.
Special attention is paid to infrastructure. Each account gets a dedicated AWS server with its own static IP, and the trader chooses the region themselves. This is critical: a request to Binance from Tokyo takes about 23 ms, while from Singapore it takes 206 ms. During that time, the order book of a liquid futures contract manages to fully update, and the order arrives at a book that no longer exists. In the team's measurements, such a delay cost nearly 0.8% in slippage—several times more than a typical spread.
Backtest instead of a showcase
The approach to the scanner deserves special attention. Instead of real-time quotes, the platform shows the result of a backtest with a 5-minute delay. The system runs the last 8 hours of recorded data through the same two-threshold strategy used by trading cards. Fees are deducted immediately, and execution delay is factored in at the worst price over the selected interval (0.5, 1, or 2 seconds). This filters out the main illusion of arbitrage screeners—pretty percentages that are impossible to reach.
Trading cards link a pair of exchanges, a specific coin, and two spread thresholds—for opening and closing. The algorithm works on divergence in either direction: if the spread moves up, one configuration opens; if it moves down, a mirrored one. Both orders are sent simultaneously, and in the event of partial execution, the system either fills the remaining volume or closes the position. Protection against delisting is also provided: the platform tracks official exchange announcements, cross-checks instrument lists, and reacts to a symbol disappearing from the feed.
Security and fees
Arbitron's model is non-custodial. API keys are stored using an "envelope" scheme with AWS KMS and encrypted with AES-256-GCM, while withdrawal rights are not requested at all. All trades are recorded in a full log: price, volume, fee, execution time.
The pricing structure is based not on trading volume, but on functionality. The Scanner plan at $39 per month provides data without execution. Trader at $99 adds a dedicated server and trading with your own keys, with a fee of 35% of realized profit. Prime at $299 (coming soon) reduces the fee to 30% and expands limits. The fee is charged only on profit, and losses are carried forward to subsequent periods—if you close the gap, no charges will apply.
My take: Arbitron is an attempt to solve the main problem of arbitrage—speed and feasibility—through engineering discipline. The honesty in assessing spreads is telling: a backtest that accounts for fees and delays is something most competitors lack. However, it is worth remembering that arbitrage is not "free cheese," but a competitive race where the winner is not the one who found the spread, but the one who managed to grab it before others. It is better to start with small amounts and careful threshold tuning, rather than expecting instant enrichment.