Bid-ask bounce: the noise that makes tiny timeframes lie

[Amanhasnoname]

On very short timeframes, price often looks like it's oscillating when it isn't. Trades alternate between hitting the bid and lifting the ask, so recorded prices flip back and forth by the width of the spread.

That oscillation is called bid-ask bounce, and it creates negative autocorrelation in short-term returns. In plain terms, a tick up is slightly more likely to be followed by a tick down, purely because of the quoting structure.

For scalpers, this has real consequences. Indicators on one-second or tick data pick up fake mean reversion. A strategy that fades each tiny move seems to work in backtests using last-trade or mid prices, and then fails live because every round trip crosses the spread.

Using mid prices smooths it out but hides the cost. The honest approach is to test with bid and ask separately, and to charge the spread on each entry.

If a backtest shows a smooth edge on tick data, the first question should be whether the bounce is producing it.

A simple experiment makes the effect visible: take a day of tick data, compute returns on mid prices and again on trade-side prices, then compare the first-order autocorrelation. The second number is usually strongly negative, and that gap is the bounce at work.

Has a short-term edge ever vanished once spread was charged properly?