The same asset on different exchanges almost never costs the same. The difference in quotes is the fuel for arbitrage: buy cheaper on one platform, sell higher on another. But the devil, as always, is in the details. Spreads rarely exceed 0.5%, and they last only a few seconds. Manual trading loses to machines from the start: by the time a trader notices the discrepancy and places two orders, the market has already closed the window of opportunity.
The Arbitron platform automates this process, handling monitoring and execution. The service tracks order books and funding rates on 20 exchanges, independently placing both sides of the trade—the so-called "legs." Meanwhile, funds remain on the trader's own accounts, which is critical from an asset control perspective.
Why scanners lie, and backtesting doesn't
The main problem with classic arbitrage scanners is that they show the raw price difference, ignoring costs. Real profit is eaten by four factors: taker fees on both legs, order book depth, slippage, and the speed at which the spread itself decays. Fees are easy to calculate, but order book depth is a trap: at the best price, there is often only a couple of dollars of volume, and with a real order of $500, the spread simply disappears.
Arbitron solves this problem in an unconventional way. Instead of instant quotes, the scanner shows backtest results with a 5-minute delay. The platform runs the last 8 hours of recorded data through the same strategy used in live trading, with fees and execution delay already deducted. This filters out the main illusion of arbitrage scanners—the attractive percentages that are impossible to reach.
Speed as a decisive factor
The infrastructure here is thought out to the smallest detail. Each account gets a dedicated AWS server with its own static IP, and the trader chooses the region themselves. This is not a whim but a necessity: a request to Binance from Tokyo takes about 23 ms versus 206 ms from Singapore. A delay of 200–500 ms means a fully refreshed order book and slippage that is several times larger than the typical spread. The trading core is written in Rust, which eliminates unexpected pauses on the critical execution path.
Protection against delisting and honest monetization
The delisting protection system deserves special attention. When an exchange removes a contract from trading, one leg of the position disappears, leaving the other without a hedge. Arbitron tracks such events in three layers: official announcement APIs, periodic reconciliation of the full instrument list, and reaction to a symbol disappearing from the data feed. Market makers leave a doomed instrument in advance, widening the spread by 5–10 times, so the platform recommends closing positions 24–48 hours before the delisting.
Monetization is built on the principle of "you pay when you earn." A fee is charged only on realized profit, and losses are carried over to future periods. This is a fair model that aligns the interests of the platform and the trader: the service has no incentive to show false signals, since its income directly depends on the user's real results.
My verdict: Arbitron is a rare example of an infrastructure product that solves the real problem of speed and accuracy in inter-exchange arbitrage. However, it is worth remembering: even with perfect automation, arbitrage is not a "money printer" but a competitive battle against other algorithms. Success here is determined not only by technology but also by discipline in setting thresholds and managing risks.