Home / Core Volatility / Realized Volatility
Core Volatility Intermediate What the market actually did Backward-looking

Realized Volatility RV

The volatility you were actually short — and the reason your estimator and mine disagree on the same 60 days.

Quick answer: Realized Volatility is the volatility an asset actually delivered over a past period, measured from its realised price path — the quantity an option seller is short and a delta-hedged book is paid on — and it is estimator-dependent, so close-to-close and high-low methods can report different numbers from the very same days.

In simple words

Realized volatility is the volatility that really happened — the movement the market actually put on the board, measured after the fact from its own price path. It sounds like it should be a single, settled number, and it is not, because 'measured how?' is a real question with several answers. The simplest estimator uses only closing prices and is identical to close-to-close historical volatility. But the market did not politely stand still between the closes: it traded a high and a low every day, and a method that looks at that daily range can extract more information from the same 60 days and land on a different figure. Realized volatility is the honest name for 'what the market did', and the honesty includes admitting that your measuring stick shapes the answer.

Here is why it matters beyond bookkeeping. Realized volatility is not an abstraction — it is the thing an option seller is actually short. When you sell an option and hedge it, your profit or loss over its life comes down to how much the underlying really moved, its realised volatility, versus the implied volatility you sold it at. If you sold at 15% implied and NIFTY realises 12%, you were paid for movement that never came; if it realises 20%, the movement you underwrote arrived and cost you. Realized volatility is the scoreboard that settles that bet, which is why measuring it accurately is not academic.

Not to be confused with: Historical volatility, which in its usual sense is the specific close-to-close estimator, and implied volatility, which is a forward-looking price rather than a backward measurement. Realized volatility is the general idea of measuring what actually happened, of which close-to-close is only the crudest version; the phrase is most often used precisely when someone wants a better estimator than close-to-close, such as a high-low or intraday one.

The path the market actually traced

Realized volatility is measured from the path, and the path has more in it than the closes

A NIFTY intraday path with the daily closes marked, showing how much travel happens between them.

22,50023,00023,50024,00024,50025,00025,5000d15d30d45d60dclose-to-close volatility: 14.3% · Parkinson high–low volatility: 20.6%Trading dayNIFTY levelclosing priceeach bar is that day's high–low range
The dots are the closes a close-to-close estimator sees; the line is everything it misses. On the choppy days the price travelled far intraday and returned near its open, so the closes barely moved while the path swept a wide range. Any estimator that reads only the dots will call these days calm; any estimator that reads the range will not. That gap between the dots and the line is the entire estimator problem.

Professional explanation

Close-to-close is blind to everything between the closes

The default realized-volatility estimator uses one price per day and measures the return from close to close, which makes it identical to standard historical volatility and gives it the same blind spot: the entire trading day. Consider a session that opens at 24,000, spikes to 24,300 on a rumour, sells off to 23,950 when it is denied, and closes back at 24,010. To a close-to-close estimator that is a flat day — the return from 24,000 to 24,010 is 0.04% — even though the index travelled a 350-point range and a delta-hedged book would have been trading frantically the whole session. Close-to-close does not underestimate that day's volatility slightly; it misses it almost entirely, because the only two numbers it looks at happened to end up next to each other. This is not a defect to be tuned away — it is the definition of an estimator that samples once a day at a moment the market chose arbitrarily.

The Parkinson high-low estimator sees the range, and reads differently on the same days

The Parkinson estimator throws away the closes and uses each day's high and low instead, on the logic that the range a market traces contains far more information about its volatility than the single close-to-close jump does. On the choppy day above — high 24,300, low 23,950 — the log range ln(24,300 ÷ 23,950) = 0.0145 implies a daily volatility of about 0.87% and an annualised 13.8%, against the close-to-close estimator's near-zero. Run both methods over the same 60 sessions of a range-bound but internally busy market and they will disagree, often by several percentage points, because they are reading different features of the identical price history. The Parkinson estimator is also statistically more efficient — for a given number of days it has lower estimation error than close-to-close — which is why it is valued even when the two happen to agree. But it has its own blind spot, and it is an important one.

Every estimator is blind to something — Parkinson cannot see an overnight gap

The mirror image of close-to-close's intraday blindness is Parkinson's overnight blindness. Because Parkinson uses only the high and low within a trading day, it cannot see a move that happens between one day's close and the next day's open — a gap. Suppose BANKNIFTY closes at 52,000, overnight news breaks, and it opens at 50,700, then trades quietly in a 50,600 to 50,900 band all day and closes at 50,800. Parkinson looks only at that tight intraday range, ln(50,900 ÷ 50,600) = 0.0059, and reports a placid 5.6% annualised — while the real story is a 2.5% overnight gap that a close-to-close or open-to-close estimator would capture and Parkinson simply cannot. This is the honest conclusion the page has to reach: there is no single correct realized volatility, because every estimator makes the same trade — it sees some part of the price path perfectly and is structurally blind to another. Combined estimators like Garman–Klass and Rogers–Satchell exist precisely to use opens, highs, lows and closes together, and even they assume no jumps.

Higher-frequency sampling cuts estimator variance, then microstructure noise ruins it

If sampling once a day throws away information, the obvious fix is to sample more often — every hour, every five minutes — and sum the squared intraday returns to build a much more precise estimate. Up to a point this works beautifully: more observations mean lower estimation error, and high-frequency realized volatility is a genuinely sharper measurement of a day's movement than any daily estimator. But push the frequency too high and it breaks in a specific, well-documented way. At very fine intervals the returns you are measuring are dominated not by the market moving but by the bid and offer bouncing back and forth — microstructure noise — so the estimate stops rising toward the true volatility and instead inflates without bound. Plot realized volatility against sampling frequency and you get the famous 'volatility signature': accurate in the middle, ruined at the fast end. Practitioners sample at an interval fine enough to gain precision but coarse enough to escape the noise, typically several minutes, and treat anything faster as measuring the plumbing rather than the price.

Realized volatility is what you were short, and what a delta-hedged position is paid on

The reason a whole discipline cares about measuring realized volatility precisely is that it is not merely descriptive — it is the settlement quantity of the entire volatility trade. When you sell an option and delta-hedge it continuously, your profit and loss over its life is, to a close approximation, proportional to the gap between the volatility you sold and the volatility the underlying subsequently realised: you collected implied and you paid out realised. The Greek that governs this is gamma, and the daily rent it charges or pays is set by how far the underlying actually moved that day — realised volatility, day by day. So realized volatility is the thing an option seller is genuinely short and an option buyer is genuinely long, independent of direction. This is why the difference between implied and realized volatility has its own name, the volatility risk premium, and why a delta-hedged trader stops caring which way NIFTY goes and starts caring only how much it moves.

Formula

Parkinson high-low realized volatility

σ_Parkinson = √( (1 / (4 ln 2)) × mean( ln(H_t / L_t)² ) ) × √252

The Parkinson (1980) estimator uses each day's high and low instead of its close. The constant 1 ÷ (4 ln 2) ≈ 0.3607 rescales the mean squared log-range so that, under the assumption of continuous geometric Brownian motion with no drift and no gaps, it is an unbiased estimate of the daily variance. It is more statistically efficient than close-to-close, but it sees only intraday range and is therefore blind to overnight gaps, and it tends to understate volatility on days where the true high and low occurred while the market was thin.

  • σ_ParkinsonThe Parkinson estimate of realized volatility, annualised, as a percentage.
  • H_tThe intraday high price on day t.
  • L_tThe intraday low price on day t.
  • ln(H_t / L_t)The log-range: the natural logarithm of the day's high divided by its low — the estimator's raw signal.
  • 4 ln 2The scaling constant 4 × ln 2 ≈ 2.7726; its reciprocal ≈ 0.3607 makes the mean squared log-range an unbiased variance estimate under the model's assumptions.
  • mean(·)The average over all days in the window.
  • √Square root — of the scaled variance to get volatility, and of 252 to annualise.
  • 252Approximate number of trading days in an Indian market year — the annualisation factor.

The close-to-close estimator, for comparison

σ_close = √( (1 / (n − 1)) × Σ (r_t − r̄)² ) × √252, with r_t = ln(C_t / C_{t−1})

The standard close-to-close estimator, identical to historical volatility: the annualised sample standard deviation of daily closing-price log returns. It uses only the close C_t, so it captures overnight gaps that Parkinson misses but is entirely blind to the intraday range that Parkinson is built to see. Running both on the same window and comparing them is one of the most direct ways to detect whether a market's movement is happening intraday or across the close.

How to measure realized volatility, and how to choose an estimator

  1. Decide what you are trying to see. If overnight gaps and event risk matter to your position, you need an estimator that reads the close; if intraday range is the story, you need one that reads the high and low.
  2. For a close-to-close estimate, take the daily closing prices, compute log returns between consecutive closes, take their standard deviation, and annualise by √252. This is the baseline and equals historical volatility.
  3. For a Parkinson estimate, take each day's high and low, compute ln(H ÷ L) squared for every day, average them, multiply by 1 ÷ (4 ln 2) ≈ 0.3607, take the square root, and annualise by √252.
  4. Run both over the same window and compare. If Parkinson sits well above close-to-close, the market is travelling intraday and returning near its open; if close-to-close sits above Parkinson, the movement is happening across the close, in gaps Parkinson cannot see.
  5. If you have intraday data and want precision, sum squared returns sampled at a fixed intraday interval — but do not go too fine. Below roughly a few minutes, bid-ask bounce inflates the estimate, so back off to where the estimate stabilises.
  6. Annualise consistently and label the estimator. '60-day Parkinson RV = 12%' and '60-day close-to-close RV = 9%' are both complete statements; 'RV = 12%' hides the one thing the reader needs to know.
  7. Interpret against implied volatility, not in isolation. Realized volatility earns its keep when compared with the implied volatility that was charged for the same period — that comparison is the whole point.

Practical example

NIFTY worked example

Take one deliberately choppy NIFTY day. It opens at 24,000, spikes to a high of 24,300, sells off to a low of 23,950, and closes at 24,010. A close-to-close estimator sees only the two closes: ln(24,010 ÷ 24,000) = 0.00042, a 0.04% move that annualises to under 1% — a dead-flat day, as far as it can tell. The Parkinson estimator sees the range: ln(24,300 ÷ 23,950) = 0.01451; square it to get 0.0002105; multiply by 1 ÷ (4 ln 2) = 0.3607 to get 0.00007593; the square root is 0.008714, a daily figure, and annualised by 15.87 that is 13.8%. Same day, same data, and the two estimators disagree by roughly fourteen percentage points — one calls it calm, the other calls it a normal-to-busy session. Now extend that character over 60 such days, where the index keeps closing near its open while sweeping a wide range: close-to-close might land near 9% while Parkinson lands near 14%. Interpret the gap, do not just note it. It tells you the market's movement is happening inside the day and getting erased by the close — information a close-to-close number structurally cannot carry, and exactly the movement a delta-hedged option book is trading against all day.

BANKNIFTY worked example

BANKNIFTY makes the opposite blind spot concrete. Suppose it closes one evening at 52,000, overnight news breaks, and the next day it opens far lower at 50,700, then trades in a narrow 50,600 to 50,900 band and closes at 50,800. The Parkinson estimator reads only that tight intraday range: ln(50,900 ÷ 50,600) = 0.005911; squared, 0.00003494; times 0.3607, 0.0000126; square root 0.00355; annualised, a serene 5.6%. But the day was not serene — the underlying gapped 2.5% overnight, a move a close-to-close or open-to-open estimator captures in full and Parkinson never sees, because the gap happened while the market was shut and left no trace in the day's high-low. On the same session, close-to-close would report a large single-day move and Parkinson a quiet one, and the honest reading is that both are right about the part of the path they can see. The lesson generalises: on a gap-prone, event-driven index like BANKNIFTY, a range-based estimator can systematically understate the risk you actually carry through the close, and choosing an estimator that ignores gaps is choosing not to measure the risk that hurts most.

Lot sizes used above (NIFTY 75, BANKNIFTY 30) are those in force at the time of writing; NSE revises them periodically. Figures exclude brokerage, STT, exchange charges, stamp duty and GST. Examples are teaching scenarios built on round numbers — they are not historical quotes, not backtests and not trade calls.

Risk note. Because a range-based estimator like Parkinson is more efficient and usually reads lower and smoother than close-to-close on gap-prone markets, it is tempting to adopt it as 'the' realized volatility and size risk from it. On an event-driven index that can be a quiet way to under-measure exactly the risk that matters: Parkinson is blind to overnight gaps, which is where an index's most violent, position-destroying moves often occur. An estimator chosen for its low reading, rather than for what it can and cannot see, will flatter the recent past and leave a book under-provisioned against the gap that no intraday range ever recorded.

Advantages & limitations

What it is good for

  • It measures what actually happened, which is the settlement quantity of the volatility trade. Realized volatility is the exact thing an option seller is short and a delta-hedged book is paid on, so it is the number that decides those outcomes rather than merely describing them.
  • Range and intraday estimators are more statistically efficient than close-to-close. For a given number of days, Parkinson and its relatives produce a lower-error estimate, so they extract a cleaner reading from a shorter history.
  • Comparing estimators is itself diagnostic. Running Parkinson against close-to-close on the same window reveals whether a market's movement is happening intraday or across the close — information neither estimator gives alone.
  • It is the natural benchmark for implied volatility. The whole judgement of whether options were rich or cheap is implied volatility versus subsequently realized volatility, so realized volatility is what makes the volatility risk premium measurable.
  • High-frequency versions give genuinely precise daily measurements. When sampled at a sensible interval, intraday realized volatility pins down a single day's movement far more tightly than any once-a-day estimator can.

Where it breaks down

  • It is estimator-dependent, so there is no single 'the' realized volatility. Close-to-close, Parkinson, Garman–Klass and intraday methods can all report different numbers from the same days, and quoting RV without the method is quoting nothing.
  • Close-to-close is blind to the entire trading day. A session that swept a wide range and closed near its open registers as calm, so the estimator can miss most of the movement a hedged book actually traded against.
  • Range estimators are blind to overnight gaps. Parkinson cannot see a move that happens while the market is shut, so on gap-prone, event-driven indices it can systematically understate the risk that does the most damage.
  • High-frequency estimates break down at fine sampling intervals. Below a few minutes, bid-ask bounce and other microstructure noise inflate the estimate without bound, so more data past that point makes the measurement worse, not better.
  • It is backward-looking. Realized volatility describes a period that has finished and, like all past-based measures, tells you nothing directly about the volatility of the period you are about to be exposed to.
  • Estimators that combine open, high, low and close still assume no jumps. Garman–Klass and Rogers–Satchell improve efficiency but rest on a continuous-diffusion model, so a genuine jump violates their assumptions and biases the result.

Common mistakes

  • Treating realized volatility as a single settled number. It is estimator-dependent, so 'RV = 12%' is incomplete — close-to-close and Parkinson on the same 60 days can differ by several points, and the reader cannot interpret the figure without knowing the method.
  • Using close-to-close on a whippy, range-bound market and concluding it was calm. A day that travelled a 350-point range and closed near its open reads as flat close-to-close, so a book that took real intraday risk sees a number that denies it.
  • Adopting Parkinson on a gap-prone index because it reads lower and smoother. It reads lower partly because it is blind to overnight gaps, which is exactly where an event-driven index's worst moves live, so the low reading is a measurement failure, not calm.
  • Sampling intraday returns as fast as the data allows. Below a few minutes the estimate is dominated by bid-ask bounce, not by the market, so a tick-level realized volatility measures the plumbing and inflates without limit.
  • Comparing a Parkinson realized volatility to an implied volatility without matching conventions. Implied volatility is close-to-close in spirit; benchmarking it against a range-based realized volatility introduces an estimator gap that can be mistaken for a volatility risk premium that is not there.
  • Forgetting that realized volatility is the thing you are short when you sell an option. Sizing a short-option position off implied volatility alone, without a view on what the underlying is likely to realise, is trading the price you collected while ignoring the quantity that will settle the trade.
  • Ignoring the no-jump assumption in combined estimators. Garman–Klass looks authoritative, but a real overnight jump violates the continuous-diffusion model it assumes, so on an event day its precision is spurious.

Professional usage

Volatility desks live and die by realized volatility because it is the settlement side of every volatility position they hold. A volatility arbitrage trader forecasts realized volatility, compares it against the implied volatility the market is charging, and buys or sells options accordingly while delta-hedging continuously, so that the residual profit and loss is driven by the gap between implied and realised rather than by the direction of NIFTY. To measure the realised side they rarely trust one estimator: they run close-to-close, a range-based estimator like Parkinson or Garman–Klass, and a high-frequency estimate sampled at a few minutes, and they read the disagreements between them as information about where the movement is concentrated — intraday, across the close, or in gaps. Getting that measurement right is not pedantry; it is the difference between correctly and incorrectly marking whether a position made money.

On the sell side and in risk, realized volatility feeds the forecasting and margining machinery directly. Realized-volatility time series are the training data for the whole apparatus of volatility forecasting — HAR models built on daily, weekly and monthly realized volatility, and GARCH-family models — because a good forecast of future realised volatility is what a market maker needs to quote a fair implied volatility and what a risk manager needs to set margin that anticipates movement rather than trailing it. The candid caveat the best desks state plainly is that every one of these estimators and models assumes the price path is a diffusion without jumps, and the days that matter most — the gap, the crash, the limit move — are exactly the days that assumption fails, so realized volatility is measured most confidently in the regimes where it matters least.

Key takeaways

  • Realized volatility is the volatility a market actually delivered, measured after the fact from its price path — and it is estimator-dependent, so there is no single correct number.
  • Close-to-close uses only the closes and is blind to the whole trading day; the Parkinson high-low estimator sees the intraday range and reads differently, often by several points, on the very same days.
  • Every estimator trades one blindness for another: close-to-close misses intraday travel, Parkinson misses overnight gaps, and even combined open-high-low-close estimators assume no jumps.
  • Higher-frequency sampling lowers estimation error up to a point, then microstructure noise takes over and inflates the estimate, so precision has a sweet spot at a few minutes, not at the tick.
  • Realized volatility is the quantity an option seller is short and a delta-hedged book is paid on, which is why the gap between implied and realized volatility — the volatility risk premium — is the whole game.

Realized volatility is where the abstraction of volatility becomes a settlement number. It is not a chart annotation; it is the exact thing you are short when you sell an option and hedge it, the quantity gamma rents out day by day, and the benchmark that decides whether the implied volatility you traded was rich or cheap. The hard, unmarketable truth the page has to end on is that 'what the market actually did' is not one number but a family of them, each estimator seeing part of the price path clearly and part of it not at all — and the parts they cannot see, the choppy intraday reversals and the overnight gaps, are disproportionately the parts that decide who wins. Choose the estimator for what it can see, state which one you used, and never confuse a low reading for a low risk.

Frequently asked questions

What is realized volatility in simple terms?
Realized volatility is the volatility that actually happened over a past period, measured from the market's own price path. It is the movement the market really put on the board, as opposed to the movement an option's price implied. The catch is that measuring it requires choosing an estimator, and different estimators give different answers from the same days.
What is the difference between realized and implied volatility?
Realized volatility is measured backward from prices that already happened; implied volatility is extracted forward from option prices and describes a period that has not begun. Realized is what the market did; implied is what the market charged you to bear the uncertainty. The average gap between them, with implied usually higher, is the volatility risk premium.
What is the difference between realized and historical volatility?
They overlap: close-to-close realized volatility is exactly historical volatility. But 'realized volatility' is a broader term that includes better estimators — Parkinson's high-low, Garman–Klass, and intraday high-frequency methods — that see movement close-to-close misses. People usually say 'realised' when they specifically want an estimator sharper than the plain close-to-close one.
Why do close-to-close and Parkinson give different numbers?
Because they read different features of the same price history. Close-to-close sees only the move between closing prices; Parkinson sees the intraday high-low range and ignores the close entirely. On a day that swept a wide range and closed near its open, close-to-close reports near zero while Parkinson reports a substantial figure — the same data, two answers.
What is the Parkinson volatility estimator?
The Parkinson estimator measures realized volatility from each day's high and low instead of its close, using the formula σ = √((1 ÷ (4 ln 2)) × mean(ln(H ÷ L)²)) × √252. It is more statistically efficient than close-to-close because the range carries more information than a single jump, but it is blind to any move that happens overnight while the market is shut.
Why is the constant 1 over 4 ln 2 in the Parkinson formula?
It rescales the average squared log-range so that, under the model of continuous geometric Brownian motion with no drift, the result is an unbiased estimate of the true daily variance. The daily range of a diffusion is statistically related to its standard deviation by that factor, 4 ln 2 ≈ 2.77, whose reciprocal is about 0.3607.
Which realized volatility estimator is best?
None is best in general, because each trades one blind spot for another. Close-to-close misses intraday movement but catches gaps; Parkinson catches intraday range but misses gaps; combined estimators use both but assume no jumps. The right choice depends on whether your risk lives intraday, across the close, or in events — so match the estimator to the movement you need to see.
Is realized volatility the same as the volatility I am short when I sell an option?
Yes, essentially. When you sell an option and delta-hedge it, your profit and loss over its life depends on the gap between the implied volatility you sold and the realized volatility the underlying subsequently delivers. You are short realized volatility, day by day, through gamma — which is why measuring it accurately decides whether the trade actually made money.
Why does higher-frequency sampling improve realized volatility, then hurt it?
Sampling more often gives more observations and lower estimation error, up to a point. But at very fine intervals the returns you measure are dominated by the bid-ask bounce rather than by the market moving, so microstructure noise inflates the estimate without bound. The result is a sweet spot at a few minutes, with tick-level sampling measuring the plumbing instead of the price.
What is microstructure noise?
Microstructure noise is the distortion in high-frequency prices caused by the mechanics of trading — bid-ask bounce, discrete tick sizes, order-flow effects — rather than by genuine movement in the asset's value. It is negligible over a day but dominates over seconds, which is why realized volatility sampled too finely balloons upward and stops measuring the real volatility.
Can Parkinson volatility be lower than close-to-close?
Yes, specifically when a market's movement happens in overnight gaps rather than intraday. Parkinson only sees the intraday range, so on a day that gapped hard at the open and then traded quietly, it reports a low figure while close-to-close, which captures the gap, reports a high one. A Parkinson reading below close-to-close is a signal that gaps are driving the movement.
What is the volatility signature plot?
It is a plot of realized volatility against sampling frequency, and it has a characteristic shape: roughly flat and accurate at moderate intervals, then rising sharply as the interval shrinks toward the tick, because microstructure noise takes over. Traders use it to choose a sampling frequency on the flat part of the curve, fine enough for precision but coarse enough to avoid the noise.
Is realized volatility backward-looking?
Yes, entirely. It measures a period that has already finished, from prices that have already printed. It is often used as an input to forecast future volatility, because volatility clusters, but the measurement itself contains no forecast — it is a scoreboard of what happened, not a prediction of what will.
What is the Garman–Klass estimator?
Garman–Klass is a realized-volatility estimator that combines the open, high, low and close of each day to extract more information than either close-to-close or Parkinson alone, making it more statistically efficient than both. Its cost is a stronger model assumption: like Parkinson it assumes the price is a continuous diffusion with no jumps, so a real overnight gap violates it.
How does realized volatility relate to gamma?
Gamma is the Greek that converts realized volatility into profit and loss for a delta-hedged option. Each day, a long-gamma position earns roughly in proportion to how far the underlying actually moved squared, and a short-gamma position pays it — so the daily realized volatility is exactly what gamma is renting out. Realized volatility is the quantity gamma exposure is a bet on.
Why compare realized volatility with implied volatility?
Because that comparison is how you judge whether options were rich or cheap. Implied volatility is the price you paid or received; realized volatility is what the movement turned out to be worth. If implied exceeded subsequently realized, sellers were overpaid on average — the volatility risk premium — but the exceptions, when realised blows past implied, are large enough to make the average dangerous to lean on.
Does realized volatility tell me market direction?
No. Like every volatility measure it is sign-blind — it is built from squared or ranged moves, so an up day and a down day of the same size contribute equally. It tells you how much the market moved, never which way, which is exactly the property that makes it the right quantity for pricing and settling bets on movement.
How many days should I use for realized volatility?
It depends on the horizon you are measuring and the estimator. Range and intraday estimators are more efficient, so they need fewer days than close-to-close for the same precision. As with historical volatility, the window is a real choice that shapes the answer, and it must be stated alongside the number and the estimator for the figure to mean anything.
Why does close-to-close call a whippy day calm?
Because it only looks at the two closing prices, and a day can travel a huge intraday range yet close right back near where it opened. If NIFTY spiked 300 points and gave it all back by the bell, the close-to-close return is near zero, so the estimator records a calm day even though a hedged book traded against real movement all session.
Is a higher realized volatility always worse?
It depends on your position, not on the number. Higher realized volatility is good for a long-option, long-gamma position, which is paid when the underlying moves, and bad for a short one, which pays. Realized volatility is directionless and position-neutral in itself; whether a high reading helps or hurts is entirely a function of which side of the volatility you are on.
Can realized volatility be computed without intraday data?
Yes — the close-to-close estimator needs only daily closing prices, and the Parkinson estimator needs only the daily high and low, both of which are widely available. True high-frequency realized volatility requires intraday tick or bar data, but the daily estimators are perfectly usable and are what most retail platforms display.
Why do combined estimators still assume no jumps?
Because Garman–Klass, Rogers–Satchell and their relatives are derived under a continuous-diffusion model of prices, in which the path never leaps. A real jump — an overnight gap, a crash, a limit move — breaks that assumption, so on precisely the days that matter most the estimators' efficiency becomes spurious and their readings can mislead. Their precision is real only in calm, gap-free regimes.

Voice search & related questions

Natural-language questions people ask about realized volatility.

What is realized volatility?
Realized volatility is the volatility that actually happened, measured after the fact from the market's price path. It is the real movement the market delivered, as opposed to the movement an option's price implied ahead of time. How much it comes out to depends on which estimator you use, which is the whole subtlety of the concept.
Why do two people measure different realized volatility for the same week?
Because they used different estimators. One may have used close-to-close, which sees only the closing prices, and the other Parkinson, which uses the daily high and low. On a week of choppy, range-bound trading those two methods can differ by several percentage points, and both are correctly measuring the part of the path their method looks at.
Is realized volatility the thing I actually lose money to when I sell options?
Yes. When you sell an option and hedge it, what settles the trade is how much the underlying really moved — its realized volatility — against the implied volatility you sold. If the market realises more than you sold, the movement you underwrote arrives and costs you; realized volatility is the scoreboard, not an abstraction.
Does using more frequent data always give a better realized volatility?
Only up to a point. Sampling more often lowers your estimation error at first, but if you go too fine — down to seconds or ticks — the bid-ask bounce takes over and the number balloons, measuring the trading mechanics instead of the market. Most practitioners sample every few minutes, on the part of the curve where precision is high and noise is still low.
Can an estimator miss a big move entirely?
Yes, and this is the key thing to internalise. Close-to-close misses a day that swept a wide range and closed flat; Parkinson misses an overnight gap that happened while the market was shut. Each estimator is perfectly blind to one kind of movement, and unfortunately those blind spots — intraday reversals and gaps — are often where the real risk lives.
Should I trust a low Parkinson reading on BANKNIFTY?
Be careful. Parkinson only sees the intraday range, so on a gap-prone, event-driven index like BANKNIFTY it can read low precisely because it cannot see the overnight moves that do the most damage. A low range-based number is not the same as low risk; check a close-to-close estimate alongside it before you conclude the market was quiet.
Why does everyone compare realized volatility to implied volatility?
Because that comparison is the entire point of measuring realised in the first place. Implied volatility is what you were charged; realized volatility is what the movement turned out to be worth. The gap between them tells you whether options were rich or cheap, and it is the quantity — the volatility risk premium — that option sellers are trying to harvest and option buyers are trying to catch.

Sources & references

Last reviewed 10 July 2026. Educational content only — not investment advice.

← Previous
Historical Volatility
Educational content only — not investment advice. Every diagram on this page is generated from the site's own model, using illustrative inputs rather than live quotes. Options and futures carry substantial risk, including loss exceeding your deposit on short-volatility positions. See our Risk Disclosure and SEBI Disclaimer.