Tiny tick-data reversals that are really just bid-ask bounce

[Shadow Trader]

Tick data shows a funny behavior. Even if the true price doesn't change, the quoted price jumps between bid and ask as trades hit one side or the other. The sequence of trade prices bounces up and down by roughly the spread.

This bounce creates negative serial correlation at the tick level. An up tick is more likely to be followed by a down tick, simply because of the bounce, not because of any information.

A strategy trading mean reversion on very short timeframes can mistake this for an edge. The reversals are real in the data, but impossible to capture, since capturing them requires paying the spread each time.

The effect is the main reason that tick-level backtests, done with trade prices rather than bid and ask quotes, can look better than reality. They show a reversal that the spread would consume.

Handling it means using separate bid and ask series, and simulating entries at the correct side. A buy happens at the ask, a sell at the bid, always.

It's one of the most common causes of "too good" results in high-frequency tests, and one of the easiest to fix.

Did a mean-reversion test look great right up until bid and ask were handled properly?