Why Arbitrage Backtests Do Not Reproduce

Arbitrage extracts profit from price discrepancies in the same asset. On historical data such discrepancies are easy to find, which is exactly why a historical arbitrage backtest almost always lies.

A discrepancy on a chart and a discrepancy in reality

A historical quote records that a trade was possible at some moment. It does not say how long that possibility existed or what size it could absorb. An arbitrage discrepancy lives for fractions of a second and is exhausted by the first participants to reach it. On a chart it looks like a reliable, repeating pattern — because the chart shows traces of trades that happened, not opportunities available to you.

The competition is not with the market

An ordinary strategy competes with the market: it is right or wrong about future direction. An arbitrage strategy competes with other arbitrageurs for speed. The winner is decided by latency to the venue, connection quality and execution algorithm — quantities absent from historical data entirely. A backtest that does not model latency answers the question “was there an opportunity”, when what matters is “would you have got there first”.

Why a correction factor cannot fix it

In other cases data can be made pessimistic to produce a conservative estimate: add costs, widen slippage. That does not work here, because the question is not the size of the adjustment but whether the trade would have happened at all. There is no continuous path between “the trade earned less than expected” and “there was no trade”, and no coefficient converts one into the other.

What would be required

Order book data with millisecond timestamps, measured latency to specific venues, and a model of competition for the same opportunity. That is an infrastructure problem rather than an analytical one, and it is not solved by choosing a different strategy. We publish no arbitrage backtests because an honest result cannot be obtained from the available data, and a dishonest one is far too easy to produce.

We publish these explanations alongside the runs themselves. Rules that can be tested are tested and published with their numbers, including the negative ones — see the MACD backtest or the weekly range breakout. The absence of figures here is a result, not a gap.

This material is educational and is not individual investment advice. Backtested results do not guarantee similar results in the future. Trading forex carries the risk of losing capital.