Why one-second volatility looks bigger than it really is
Posted: Sat Oct 31, 2026 6:41 pm
Realized volatility measured at very short intervals gets inflated by microstructure noise: bid-ask bounce, rounding, and stale quotes. A volatility estimate from one-second returns can exceed the estimate from one-minute returns scaled to the same horizon.
The volatility signature plot shows this. Plot annualized or scaled realized volatility against the sampling interval. At long intervals it settles near a stable value. At short intervals it climbs.
For scalping, the message is to choose the sampling interval carefully. Too short, and the number mostly measures noise. Too long, and it misses relevant dynamics. A range of 5 to 30 seconds often balances the two for majors, but the best value depends on the feed.
If volatility drives position sizing, noise inflation causes systematic undersizing. Risk is overestimated and positions are smaller than the true risk warrants, though occasionally the reverse happens.
Compute the signature plot for each pair and select the shortest interval at which the curve stabilizes.
Where does the volatility signature flatten for the pair being traded?
The volatility signature plot shows this. Plot annualized or scaled realized volatility against the sampling interval. At long intervals it settles near a stable value. At short intervals it climbs.
For scalping, the message is to choose the sampling interval carefully. Too short, and the number mostly measures noise. Too long, and it misses relevant dynamics. A range of 5 to 30 seconds often balances the two for majors, but the best value depends on the feed.
If volatility drives position sizing, noise inflation causes systematic undersizing. Risk is overestimated and positions are smaller than the true risk warrants, though occasionally the reverse happens.
Compute the signature plot for each pair and select the shortest interval at which the curve stabilizes.
Where does the volatility signature flatten for the pair being traded?