Volatility & bandsParkinson Volatility
A volatility estimate built from the high-low range of each bar rather than from closing prices.
Works in most conditionsEngine-computed on a fixed sample series
What it is
Parkinson volatility is a way of measuring how turbulent a market is by looking at how far price swings within each bar — from its high to its low — rather than only at where it closes. Introduced by physicist Michael Parkinson in 1980, it answers the question of how much a market is really moving, using the daily range as its raw material instead of close-to-close changes. The intuition is simple: a day that travels from a low of 98 to a high of 104 is clearly more volatile than one that quietly drifts from 100 to 101, yet if both close in the same place a close-only measure would miss the difference. By harvesting the extra information in the high and the low, Parkinson's estimator gets a sharper, more efficient read on volatility from the same number of days. For a beginner it is best understood as a smarter volatility ruler that watches the whole daily swing, not just the finish line.
How it is calculated
For each bar you take the natural logarithm of the high divided by the low, square it, and average those squared log-ranges over your lookback window. That average is multiplied by a scaling constant of one divided by four times the natural log of two (about 0.361), which calibrates the range-based figure to a true standard-deviation estimate under the assumption that price follows a random walk. Taking the square root gives a per-bar volatility, which is usually annualized by multiplying by the square root of the number of trading periods in a year, about 252 for daily data. The whole construction rests on the statistical fact that the expected range of a continuous random walk has a known relationship to its standard deviation, which is what lets the high-low spread stand in for volatility. Because it uses two data points per bar instead of one, it is roughly five times more statistically efficient than a simple close-to-close calculation.
Reading it, step by step
A higher Parkinson value means wider daily ranges and a more volatile, energetic market, while a lower value means price is traveling in tight, quiet bars. Because the output is typically annualized, it can be placed side by side with historical close-to-close volatility or with implied volatility from the options market to judge whether the market is calm or stressed. Rising Parkinson volatility often accompanies expanding, trending moves where each bar covers more ground, while a steady decline signals a market coiling into a quieter, contracting range. Traders read the direction and rate of change as much as the absolute level, since a sudden jump flags a regime shift toward turbulence. Compared with a close-to-close series the Parkinson line is smoother and less jumpy, because averaging ranges damps the noise that single closing prices introduce.
Best timeframes and settings
Parkinson volatility is most natural on daily bars, where the high-low range is meaningful and the annualization convention is well defined, and it is favored by swing and position traders and by options and risk desks. Common lookbacks are 10, 20, or 30 bars: a shorter window reacts quickly to a burst of turbulence but is noisier, while a longer window gives a stable, slow-moving read of the prevailing regime. The responsiveness-versus-smoothness trade-off is the central dial, since a 10-day estimate will spike fast around an event and a 30-day estimate will lag it but resist false alarms. It can be applied to intraday bars, but the shorter the bar the more the ignored overnight gap between bars distorts the picture. Most practitioners pick 20 or 30 days for a balanced regime read and shorten only when they need faster event detection.
When and where to use it
The estimator is best for gauging volatility regimes and sizing risk on instruments that trade in continuous, liquid sessions without large overnight jumps — broad equity indices, liquid futures, and major currencies. It pairs naturally with range- and band-based tools such as ATR, Bollinger Bands, and Keltner Channels that also key off how far price travels. Its major limitation dictates where to avoid it: because it ignores gaps and overnight moves entirely, it understates the true volatility of stocks that jump between sessions on earnings or news. It also assumes continuous trading with no drift, an assumption that breaks during shock events and around scheduled announcements. Use it when you want a more efficient volatility read from clean OHLC data, and reach for a gap-aware measure when overnight risk dominates.
Strategies that use it
The primary application is volatility-scaled position sizing — divide a fixed risk budget by the Parkinson estimate so you take smaller positions when the market is turbulent and larger ones when it is calm, keeping dollar risk steady. A second use is stop placement: set stops a multiple of the current range-based volatility away from entry so the stop breathes with the market rather than using a fixed distance. A third is regime detection — watch for the estimate to break out of a long contraction, which frequently precedes an expansion phase and a tradable trend, a logic that complements volatility-squeeze setups. In each case Parkinson volatility is a context and risk tool rather than a directional trigger, so it tells you how much to bet and how wide to stop, not which way to trade.
Combining it with other indicators
The Average True Range is a close cousin and a useful cross-check, since both measure travel, but ATR incorporates gaps while Parkinson does not, so a divergence between them highlights overnight risk. Bollinger Bands and Keltner Channels consume volatility directly, and Parkinson can inform how wide to set them or confirm when a squeeze is genuine. For a fuller volatility picture, the Garman-Klass estimator extends Parkinson by adding open and close information and gives a steadier reading. Directional and momentum tools then supply the trade direction that Parkinson deliberately omits. The recurring pattern is to let Parkinson quantify the volatility regime while separate tools handle entries, exits, and direction.
Where it fails
The signature failure is gap blindness — by using only the intraday high and low, Parkinson volatility completely misses the jump from one session's close to the next session's open, so it materially understates risk in gap-prone stocks. Its random-walk, no-drift assumptions also break during strong trends and news shocks, where the true volatility differs from what a symmetric range implies. On thinly traded instruments an unreliable or stale high or low corrupts the estimate, and discretely sampled extremes tend to fall short of the true continuous range, imparting a mild downward bias. The remedy is to treat it as one input among several, cross-checking against a gap-aware measure like close-to-close or ATR when overnight risk matters, and lengthening the window when single-bar noise is the concern. Traders also err by comparing a Parkinson figure directly to an implied-volatility number without recognizing the estimator's downward bias.
A worked example
Suppose over five days a stock prints high-low pairs of 101 and 99, 103 and 100, 102 and 100.5, 104 and 101, and 102.5 and 100, with all closes clustered near 101. A naive close-to-close volatility would look tame because the finishes barely moved, yet the daily ranges of 2.0, 3.0, 1.5, 3.0, and 2.5 points tell a more turbulent story. Parkinson takes the log of each high over low, squares it, averages the five values, scales by 0.361, and annualizes, producing a volatility reading meaningfully higher than the close-only figure and correctly flagging the churn. A trader sizing a position would use that higher number to trim exposure and widen stops. But note the blind spot: if that same stock had gapped from a 101 close to a 108 open on news, Parkinson would ignore the seven-point overnight leap entirely and understate the real danger — exactly when you would want to supplement it with a gap-aware measure.