Standard Deviation as a Risk Gauge
Standard deviation is a measure of how widely values spread around their average. Applied to price changes, it quantifies volatility, and gives a statistically grounded way to think about expected movement.
To compute daily volatility, take the daily percentage changes over a period, such as the last 20 days, and calculate their standard deviation. If it equals 0.5 percent, a typical daily move is about half a percent. Roughly two-thirds of days fall within one standard deviation of the average if returns were normally distributed, and about 95 percent within two. Real markets have fatter tails, so extreme moves occur more often than the normal curve suggests.
You can annualize the figure by multiplying by the square root of trading days, about 16 for a daily to annual conversion with 252 days. A daily deviation of 0.5 percent corresponds to about 8 percent annualized, which is typical for major currency pairs.
Uses include setting stops and targets in terms of expected movement, scaling position sizes inversely to volatility, and judging whether a move is unusual. A candle of three standard deviations is rare and often attracts attention.
Bollinger Bands and ATR are related tools, each with slightly different calculation.
Remember that volatility changes over time and clusters, with quiet and active periods.
Practical step: compute the 20-day standard deviation of daily returns for your pair, and convert it to pips at the current price.
Standard Deviation as a Risk Gauge
Standard Deviation as a Risk Gauge
It’s Fairman 
Re: Standard Deviation as a Risk Gauge
The fat tails point is important for risk. A three standard deviation day happens a lot more often in currencies than the bell curve says, so don't size as if it's a once a decade event. Under a normal distribution, a three sigma day should show up roughly once every year or so of trading days. In practice, major pairs tend to produce several of them a year, usually around central bank surprises, big data misses or sudden risk off moves. So when you check what a bad day could cost, use the biggest real moves from the last couple of years on your pair, not the textbook number. The real history is always a bit uglier.
It’s Fairman 
Re: Standard Deviation as a Risk Gauge
Converting the number to pips makes it much more practical. Then you can compare it directly with your stop and see if your stop is inside normal noise. Example: if the 20 day standard deviation on EURUSD is 0.45 percent and the price is 1.10, one standard deviation is about 50 pips a day. On a 15 minute chart you'd scale that down a lot, but the idea is the same. If your stop is smaller than the typical movement over the time you expect to hold the trade, you're likely to get stopped by noise. That's a useful sanity check before you blame your entries for every stop out.
It’s Fairman 
-
FXS Prop Desk
- Posts: 1689
- Joined: Wed Oct 07, 2026 9:52 am
Re: Standard Deviation as a Risk Gauge
One caution with the normal-distribution reading: currency returns have fatter tails than the bell curve suggests. Moves of three or four standard deviations happen noticeably more often than the theory predicts, especially around central bank decisions and in risk-off periods. Using the standard deviation for stops and sizing is sensible, but stress-testing the account against a move several times larger than the "typical" one gives a more realistic picture of what can go wrong.
FXS Official Team | The Idea Lab & Strategy
-
Shadow Trader
- Posts: 5594
- Joined: Wed Sep 09, 2026 5:19 pm
Re: Standard Deviation as a Risk Gauge
The square root rule also works in the other direction, which helps for intraday stops. If EURUSD has a daily standard deviation of 50 pips, an hour is roughly 50 divided by the square root of 24, about 10 pips, and a 15 minute window about 5.
That assumes every hour is alike, and they are not. Asian hours on EURUSD often run well below that figure and the London and New York overlap well above it, so measuring the deviation for the hours you actually trade gives a fairer number.
The 20 day estimate is noisier than it looks as well. A rough standard error for a standard deviation is the figure divided by the square root of twice the sample size, so with 20 days a reading of 0.45 percent could easily sit anywhere from about 0.38 to 0.52.
Sizing from it therefore deserves some rounding rather than false precision. A move from 0.42 to 0.47 between two weeks may be nothing more than sampling noise.
Does anyone here blend a short and a long window, such as 20 and 100 days? The short one reacts to clusters while the long one keeps a single wild week from swinging position size too far.
That assumes every hour is alike, and they are not. Asian hours on EURUSD often run well below that figure and the London and New York overlap well above it, so measuring the deviation for the hours you actually trade gives a fairer number.
The 20 day estimate is noisier than it looks as well. A rough standard error for a standard deviation is the figure divided by the square root of twice the sample size, so with 20 days a reading of 0.45 percent could easily sit anywhere from about 0.38 to 0.52.
Sizing from it therefore deserves some rounding rather than false precision. A move from 0.42 to 0.47 between two weeks may be nothing more than sampling noise.
Does anyone here blend a short and a long window, such as 20 and 100 days? The short one reacts to clusters while the long one keeps a single wild week from swinging position size too far.
-
Shadow Trader
- Posts: 5594
- Joined: Wed Sep 09, 2026 5:19 pm
Re: Standard Deviation as a Risk Gauge
Comparing the standard deviation with ATR says something about how a pair moves inside the day. Under a simple random walk, the average high to low range of a day is about 1.6 times the standard deviation of close to close moves.
So if EURUSD has a daily deviation of 50 pips, that model expects an average daily range of roughly 80 pips. An ATR close to 80 means the closes and the intraday swings are telling the same story.
An ATR of 110 against the same 50 says something else. Price is travelling a long way during the day and then giving much of it back by the close, the kind of behaviour that favours fading extremes over holding breakouts.
The reverse, an ATR of 65 with a deviation of 50, points to days that pick a direction and close near one end. Those are conditions where holding to the close tends to pay better than taking quick profits.
The ratio is also a neat way to compare pairs without converting everything into pips. A pair at 2.2 and another at 1.3 behave very differently even if their deviations match.
Has anyone looked at how that ratio changes through the year? It might be a cleaner regime filter than volatility on its own.
So if EURUSD has a daily deviation of 50 pips, that model expects an average daily range of roughly 80 pips. An ATR close to 80 means the closes and the intraday swings are telling the same story.
An ATR of 110 against the same 50 says something else. Price is travelling a long way during the day and then giving much of it back by the close, the kind of behaviour that favours fading extremes over holding breakouts.
The reverse, an ATR of 65 with a deviation of 50, points to days that pick a direction and close near one end. Those are conditions where holding to the close tends to pay better than taking quick profits.
The ratio is also a neat way to compare pairs without converting everything into pips. A pair at 2.2 and another at 1.3 behave very differently even if their deviations match.
Has anyone looked at how that ratio changes through the year? It might be a cleaner regime filter than volatility on its own.
-
Shadow Trader
- Posts: 5594
- Joined: Wed Sep 09, 2026 5:19 pm
Re: Standard Deviation as a Risk Gauge
Volatility estimation on short windows is noisier than on daily data, and the choice of estimator affects the result. The simplest approach uses the standard deviation of returns, but alternatives exploit more information within each bar.
Range-based estimators, such as Parkinson's or Garman-Klass, use the high and low in addition to the close, producing more efficient estimates from the same data. On tick data, summing squared returns over a window gives realized variance, which is the standard high-frequency measure.
Microstructure noise complicates matters. At very fine sampling, bid-ask bounce inflates measured volatility. Sampling at intervals of several seconds to a minute, or using noise-robust estimators, reduces the distortion. Scalpers can use realized volatility to size positions, adjust stops, and filter trades. Comparing current volatility to a rolling average identifies unusually quiet or active periods. Choosing an estimator that suits the data frequency and testing its stability avoids misleading readings that could skew risk decisions.
Range-based estimators, such as Parkinson's or Garman-Klass, use the high and low in addition to the close, producing more efficient estimates from the same data. On tick data, summing squared returns over a window gives realized variance, which is the standard high-frequency measure.
Microstructure noise complicates matters. At very fine sampling, bid-ask bounce inflates measured volatility. Sampling at intervals of several seconds to a minute, or using noise-robust estimators, reduces the distortion. Scalpers can use realized volatility to size positions, adjust stops, and filter trades. Comparing current volatility to a rolling average identifies unusually quiet or active periods. Choosing an estimator that suits the data frequency and testing its stability avoids misleading readings that could skew risk decisions.