Microstructure noise: the jitter you see at the one-second level
Posted: Thu Oct 29, 2026 4:56 am
Zoom into a one-second chart and price looks wild. It jumps back and forth by fractions of a pip, with no obvious direction. Most of that movement is not information. It is microstructure noise.
Sources include the bounce between bid and ask, the arrival of quotes from different liquidity providers, and the delay between updates. None of these says much about where price is going over the next minute.
For scalpers working on short timeframes, the danger is mistaking noise for signal. An indicator crossing on a five-second chart might be reacting to a tiny quote fluctuation, not any real shift in supply and demand.
Filtering helps. Using mid-price instead of bid or ask removes part of the bounce. Averaging over a few seconds reduces jitter. Setting a minimum move threshold before acting prevents responding to every flicker.
The deeper point is that there is a floor below which price changes carry no information, and strategies working near that floor are mostly fitting randomness.
A quick test for noise is to compare the distribution of price changes on one-second bars with those on ten-second bars. If many one-second moves reverse immediately, the shorter bars carry little information beyond bounce.
How short a timeframe have you found still contains real signal on your pair, and how did you test it?
Sources include the bounce between bid and ask, the arrival of quotes from different liquidity providers, and the delay between updates. None of these says much about where price is going over the next minute.
For scalpers working on short timeframes, the danger is mistaking noise for signal. An indicator crossing on a five-second chart might be reacting to a tiny quote fluctuation, not any real shift in supply and demand.
Filtering helps. Using mid-price instead of bid or ask removes part of the bounce. Averaging over a few seconds reduces jitter. Setting a minimum move threshold before acting prevents responding to every flicker.
The deeper point is that there is a floor below which price changes carry no information, and strategies working near that floor are mostly fitting randomness.
A quick test for noise is to compare the distribution of price changes on one-second bars with those on ten-second bars. If many one-second moves reverse immediately, the shorter bars carry little information beyond bounce.
How short a timeframe have you found still contains real signal on your pair, and how did you test it?