Arbitrage: a complete guide to the method
Arbitrage extracts profit from the divergence of prices of related instruments, not from the direction of the market.
| Parameter | Value |
|---|---|
| Simultaneous access | 3+ venues |
| Calculation and order-sending speed | < 5 ms |
| Capital across several pairs | Multi-currency account |
| Quote data | Tick-level |
How it actually works
Classic arbitrage buys the underpriced instrument and simultaneously sells the overpriced related one, profiting from the convergence of their prices. Market direction does not matter here — the profit comes from the relative move of the two positions, not from the absolute move of either.
Risk-free pure arbitrage has all but vanished on liquid markets: discrepancies are spotted and eliminated faster than a private trader can react, largely by those very high-frequency systems. The variant available to the private trader is statistical arbitrage: trading temporary divergences between historically related instruments in the expectation of a return to the usual ratio.
Statistical arbitrage shifts all the risk onto the stability of the relationship between the instruments. While the historical dependence holds, the strategy profits from its fluctuations; when the relationship breaks — and it breaks without warning — the divergence does not return, and the position turns into a loss on both legs at once.
Why this methodology cannot be honestly tested on our data
Arbitrage by definition requires at least two related instruments — profit is extracted from the ratio of their prices. The available data is a single pair, EUR/USD, so building an arbitrage position from it is impossible in principle, not for lack of history.
We deliberately show no backtest here: presenting attractive figures computed on unsuitable data would mislead the reader. An empty space is more honest than an invented result.
Pros and cons
- Profit does not depend on the direction of the market.
- Risk is lower than directional trading while the relationship of instruments holds.
- The position is market-neutral and reacts weakly to broad moves.
- Pure risk-free arbitrage has practically vanished on liquid markets.
- Statistical arbitrage collapses when the relationship of instruments breaks.
- Requires simultaneous trading of several instruments and precise execution.
Nuances and pitfalls
Statistical arbitrage is undone by the break of the relationship on whose stability the whole strategy is built. While the historical dependence holds, divergences return and bring profit, creating false confidence. But the relationship between instruments is not guaranteed by any law of the market — it is empirical and breaks without warning, for example on a structural shift or news on one of the instruments. At that moment both legs of the position go into loss simultaneously, and the “market-neutral” strategy turns out to be a directional bet.
Who this methodology suits
For traders who understand the statistics of relationships between instruments and have the infrastructure to execute several positions simultaneously. It requires constant monitoring of the stability of the relationship and readiness for its sudden break.
Frequently asked questions
Why can’t arbitrage be tested on a single currency pair?
Because arbitrage by construction works with the ratio of prices of at least two related instruments. Profit is extracted from their relative move, not from the move of one instrument. On a single pair an arbitrage position cannot be assembled in principle.
Does risk-free arbitrage still exist?
On liquid markets it has practically vanished: price discrepancies are spotted and eliminated in fractions of a second by automated systems. What is available to the private trader is statistical arbitrage — trading temporary divergences of historically related instruments — but it already carries the risk of that relationship breaking.
What is the main risk of statistical arbitrage?
The break of the relationship between instruments. The strategy assumes the divergence will return to its historical norm, but that norm itself is not guaranteed and can vanish on a structural change. The divergence then does not collapse back, and both legs of the position produce a loss at once.