Four ways to measure the same volatility, four answers
Close-to-close volatility uses two prices per session and discards the high, the low and the open. Four better-known estimators use more of each bar. Run all of them on the same 30 sessions of Nike and they return 19.9, 20.4, 20.8, 24.1 and 25.6 percent, a spread of 5.7 volatility points on identical data.
This is the page that changes how you read every realized-volatility figure on every platform, and it is one table long.
The same 30 sessions, five estimators
| Estimator | Uses | Annualized | Per day |
|---|---|---|---|
| Garman-Klass | open, high, low, close | 19.9% | 1.25% |
| Rogers-Satchell | open, high, low, close | 20.4% | 1.29% |
| Parkinson | high, low | 20.8% | 1.31% |
| Close-to-close | close only | 24.1% | 1.52% |
| Yang-Zhang | open, high, low, close, and the gap | 25.6% | 1.61% |
Now put that beside the thing you were going to use it for. The variance risk premium on this name was about 1.9 volatility points. Your choice of estimator moves the answer by 5.7. The measurement error is three times the signal.
Pick Garman-Klass and Nike's 26 percent implied looks like a fat six-point edge. Pick Yang-Zhang and the option is trading below realized and there is no trade at all. Same stock, same window, same option.
What each one is doing
Close-to-close takes log returns between consecutive closes. It uses two numbers out of every session and ignores everything in between. A day that opened at $74, ran to $77, broke to $72 and closed at $74.05 registers as a 0.07 percent day. That is obviously wrong, and it is the default everywhere.
Parkinson (1980) uses the high and the low of each session instead. The intuition is direct: a wide range means a volatile day whatever the close did. It is roughly five times more efficient than close-to-close on the same amount of data, meaning you need far fewer sessions for the same precision. Its blind spot is the overnight gap, which it cannot see at all.
Garman-Klass (1980), published five pages later in the same journal issue, adds the open and the close to Parkinson's high and low. More efficient again. Same blind spot on gaps, and it also assumes the stock has no drift.
Rogers-Satchell fixes the drift assumption, so it does not overstate volatility for a stock that is trending steadily. Still blind overnight.
Yang-Zhang (2000) is the one that handles the gap. Yang and Zhang built it as a combination of the overnight move, the open-to-close move and a Rogers-Satchell term, and claim in the paper that it is unbiased in the continuous limit, independent of the drift, and "consistent in dealing with opening price jumps," with "the smallest variance among all estimators with similar properties."
Why Yang-Zhang comes out highest here
Because Nike gaps at the open, and it is the only estimator on the list that is looking.
That gap is not noise to a premium seller. It is the single most important part of the distribution. Overnight is when earnings land, when guidance gets cut, when a competitor announces something, and it is the window in which you cannot close a position, cannot hedge and cannot use a stop. An estimator that scores 19.9 percent by ignoring the overnight session is measuring the risk you can manage and skipping the risk that assigns you.
Which makes the ordering in that table meaningful rather than academic. The range-based estimators are lower precisely because they are blind to the thing that hurts.
Which one to use
If you sell options and hold them overnight, Yang-Zhang. Your risk is the full 24 hours and it is the only estimator that measures the full 24 hours. That it is usually the highest reading is a feature.
If you need precision from a short window, Parkinson or Garman-Klass. Estimating volatility from ten sessions with close-to-close is nearly hopeless. The range-based estimators get you a usable number from far less data.
If you are comparing against implied, be consistent and be explicit. This is the real rule. Any of these is defensible. Switching between them, or comparing your Yang-Zhang figure against a platform's close-to-close figure, is not. Write down which one you use and never change it mid-comparison.
If your platform does not say, assume close-to-close. It is the default nearly everywhere, and now you know it is running about four points light on a gappy name.
The point of the whole page
Realized volatility looks like a fact and behaves like an opinion. There is no single true number for what a stock did last month, only a number produced by a method, and the methods differ by more than the edges most people are trading.
So when a screen tells you an option is trading three points above realized, the only correct first question is: realized measured how? If nobody can answer, the three points are not information yet.
OptionsKing scores every candidate strike on a deterministic 0 to 100 scale, blends that score with the return on the capital the trade ties up, and shows you the highest-ranked handful. There is no minimum score. How it works covers what the ranking does and does not tell you.
Questions people actually ask
What is the Yang-Zhang volatility estimator?
A realized volatility estimator combining the overnight move, the open-to-close move and a Rogers-Satchell term. Yang and Zhang published it in 2000 and claim it is drift-independent, handles opening jumps, and has the smallest variance among estimators with those properties.
Why do volatility estimators give different answers?
Because they use different parts of each price bar. Close-to-close sees two prices a day, range estimators see the high and low but not the overnight gap, and Yang-Zhang sees all of it. On one Nike window the five span 19.9 to 25.6 percent.
Which volatility estimator should an option seller use?
Yang-Zhang, if you hold positions overnight, because the overnight gap is the risk you cannot hedge and it is the only common estimator that measures it. The important part is picking one and never switching mid-comparison.
Is Parkinson volatility better than close-to-close?
More efficient, so it needs far less data for the same precision, but blind to the overnight gap. That makes it good for estimating from short windows and misleading for measuring the risk in a position held through the close.
Sources
Rules and thresholds above were checked against these documents on August 4, 2026. Exchange and broker rules change. Confirm anything you are about to act on with your own broker.
Keep reading
Do the math on your own trade
Every price in this article is an illustrative worked example, not a quote. Read Volatility for the rest of the series, and the disclaimer before you act on any of it. Selling options carries real risk of loss, and the loss can be far larger than the premium you collected.