SV · Volatility estimators
Realized volatility
Also called: historical volatility, HV, Parkinson volatility, Garman–Klass, Yang–Zhang
Realized volatility measures how much an asset actually moved; close-to-close, Parkinson, Garman–Klass and Yang–Zhang estimators use different price data.
Definition
Realized (or historical) volatility is the standard deviation of an asset's returns over a past window, usually annualized. Unlike implied volatility, which is a price, it is a measurement of movement that has already happened.
Several estimators exist. They differ in which prices they use from each bar (close only, or open, high, low and close) and therefore in how efficiently they extract information from the same number of bars.
Intuition
Close-to-close volatility ignores everything that happened during the bar. A day that swings wildly but closes unchanged contributes nothing. Range-based estimators use the high and low as well, so they capture intrabar movement and reach the same precision with fewer bars.
The right estimator depends on the data: overnight gaps, drift and microstructure noise each bias some estimators more than others.
How it's calculated
With n bars, log returns ri, and O, H, L, C the open, high, low and close of each bar (variances per bar, annualize by multiplying by bars per year, for example 252 for daily):
- Close-to-close:
σ² = 1/(n−1) · Σ (ri − r̄)² - Parkinson (1980):
σ² = 1/(4n ln 2) · Σ [ln(Hi/Li)]². Uses the range; assumes no drift and no opening gaps. - Garman–Klass (1980):
σ² = 1/n · Σ { ½[ln(Hi/Li)]² − (2 ln 2 − 1)[ln(Ci/Oi)]² }. Adds open and close; still ignores gaps between bars. - Yang–Zhang (2000):
σ² = σo² + k·σc² + (1 − k)·σrs², combining overnight (close-to-open) variance σo², open-to-close variance σc² and the Rogers–Satchell term σrs², withk = 0.34 / (1.34 + (n+1)/(n−1)). It handles both drift and opening gaps.
How to read it
Use the same estimator and window when comparing assets or periods. Short windows react quickly but are noisy; long windows are stable but lag regime changes.
For assets that trade around the clock, overnight gaps matter less and simple range estimators work well; for exchange-traded stocks with large opening gaps, Yang–Zhang or close-to-close is safer than Parkinson.
Common mistakes
- Mixing annualization factors, for example 252 for stocks and 365 for crypto, without saying so.
- Using Parkinson on gapping assets, which understates volatility.
- Comparing realized volatility over one window with implied volatility for a different horizon.
- Forgetting that thinly traded bars inflate range-based estimates through bid-ask bounce.
In Saqrid
In SVPulse, clicking any asset opens a volatility profile with 10, 20 and 60-day realized volatility, a vol cone and a regime score.
Open SVPulse.
Go deeper
Saqrid Institute lesson: Measuring Historical Volatility — and Putting It in Context. Lesson text is open to read without an account.