What Is Forex Volatility? Measures, Examples and Currency Pairs

Forex volatility describes how much exchange rates vary over a stated period. A currency pair can make large movements in either direction and be volatile; a rising price alone does not establish high volatility. In statistical work, a common historical measure is the standard deviation of returns. Other tools report price ranges or typical absolute changes, so two figures labelled “volatility” may measure different things. CME explains the distinction between historical and implied volatility.

Start with the question you want to answer. “How far did this pair move each hour?” differs from “Did the euro strengthen?” and “Did two pairs tend to move together?” Keeping those questions separate makes a chart much easier to read.

Pearson correlation measures linear association, while the currency strength meter compares currencies with its defined basket. The correlation guide explains the separate question of co-movement. Neither should be read as a substitute for a volatility measure.

Economic news, political events and changing market expectations can alter currency prices. OANDA’s explanation identifies these influences. A release that changes expectations can matter more than an already anticipated announcement, but a price chart by itself cannot identify the cause of a particular move.

The observation schedule also changes what you see. A sample containing one unusually large hour can look different from a sample ending just before it. A week with quiet observations and a month containing a sharp adjustment are different datasets, even when the pair name is unchanged.

This is why “volatile today” needs a timestamp and a definition. It might mean a large latest hourly change, a wide daily high–low range, or a high standard deviation over several weeks. These descriptions are not interchangeable. A dated measurement is more useful than a permanent label attached to a currency.

1. Standard deviation of returns

For positive prices, one log return is gᵢ = ln(Pᵢ / Pᵢ₋₁). If yᵢ = 100 × gᵢ, the returns are expressed in log-percent units. Their sample standard deviation is s = √[Σ(yᵢ − ȳ)² / (N − 1)], where ȳ is the sample mean and N is the number of returns. NIST gives this sample standard-deviation formula.

This measures dispersion around the average return. The squaring makes large departures matter more. Always specify whether the inputs are hourly or daily and whether the result has been annualized. An hourly, non-annualized figure cannot be compared directly with an annualized options-volatility quote.

2. ATR and daily price ranges

True range for a bar is TR = max(H − L, |H − Cprev|, |L − Cprev|). ATR averages or smooths true ranges; its period and smoothing convention must be stated. MetaTrader’s documentation describes the required high, low and previous close.

Average daily range, when defined as the mean of daily high minus daily low, uses a different calculation: ADR = mean(Hday − Lday). Specify the trading-day boundary and source. Endpoint-only hourly observations cannot reconstruct either the true intrahour highs and lows or a full daily range.

3. Typical absolute hourly change

A different, readable summary asks how large a typical observed hourly endpoint change was, ignoring its sign. In the IFT explorer, xᵢ = 100 × |ln(Mend / Mstart)|, using positive mid-quotes (the midpoint of bid and ask) exactly one hour apart. The headline is median(xᵢ). It is called “typical absolute hourly change,” not standard deviation.

This is also not median absolute deviation. MAD centers the observations on their median: median(|gᵢ − median(g)|), with the chosen unit conversion applied consistently. Median(|gᵢ|) measures distance from zero. As a deliberately artificial example, if every hourly log return were +0.10%, the typical absolute change would be 0.10%, while sample standard deviation and MAD would both be zero. The price would grow at a constant log rate; the returns themselves would not vary around their mean. The distinction between scale measures is set out by NIST.

4. Historical versus implied volatility

Historical measures summarize observed prices. Implied volatility is inferred from option prices using a pricing framework. CME’s CVOL explanation describes that separate information source. It is not a certain future price range, and a quote-history chart does not calculate it without options data.

The choice of measure should follow the question. A high–low range is useful for describing extremes within a bar; standard deviation describes return dispersion; a median absolute change describes a typical endpoint move. A chart should name the chosen quantity rather than hide all three behind one score.

A pip is a price increment under the pair’s quotation convention. Its size and the starting exchange rate affect how a pip distance translates into a percentage. Comparing raw pip counts across pairs therefore answers a different question from comparing proportional changes. A return percentage is also not the percentage gain or loss on an account: position size, costs and the account currency are separate inputs.

Illustration, not a market quote: a change from 1.1000 to 1.1011 is a simple return of 100 × (1.1011 / 1.1000 − 1) = +0.1000%. Its log return is 100 × ln(1.1011 / 1.1000) ≈ +0.09995%. The numbers are close here, but they are not identical. The absolute log change is 0.09995%; removing the sign removes direction, not magnitude.

For a fair comparison, keep the return formula, source, bid/ask/mid basis, interval length and timestamps fixed. If one pair has missing hours, comparing its remaining quiet hours with another pair’s full sample can distort the ranking. A common set of valid intervals makes the comparison explicit.

Missing data should stay missing. A Friday observation followed by a Monday observation is not a one-hour return. A zero means two observed endpoints were equal; a blank means there was no valid measurement under the chosen rules. Neither tells you that the price was unchanged throughout the intervening time.

There is no useful answer without a measurement rule and a period. A ranking by daily pip range can differ from one by hourly percentage standard deviation or median absolute change. Changing the observed pairs can change the leader as well. A list of eight pairs cannot establish which pair is the most volatile across the whole market.

The following is a historical example from the dated IFT snapshot ending on 5 November 2026 at 12:00 UTC. It compares eight pairs on the same available, adjacent hourly intervals within each window. The headline metric is median absolute hourly log change. Each row uses its own stated window; the rows are not a single common-sample comparison.

The changing leader illustrates why a timeless “most volatile pair” label is misleading. These are historical results from one defined source, not a current ranking, a recommendation, or evidence about pairs that were not observed. The calculation can be reproduced from the dated IFT snapshot and methodology.

The 30-day sample runs from 6 October 2026, 12:00 UTC to 5 November 2026, 12:00 UTC. Its 522 common hourly intervals cover AUD/USD, EUR/GBP, EUR/USD, GBP/JPY, GBP/USD, USD/CAD, USD/CHF and USD/JPY. The fixed source is FxPro Standard on MT5: a live-server account without funds, using observed mid-quotes. These are quote observations, not executed customer trades. Snapshot evidence.

For reproducibility, the calculation sorts x from smallest to largest and uses linear interpolation. For percentile p, h = (N − 1)p, j = floor(h), f = h − j and Qp = x[j] + f × (x[min(j + 1, N − 1)] − x[j]), with zero-based indexing. The displayed figures are rounded to four decimal places.

The median is not an expected profit per hour. The 90th percentile is not a promise that the next change will be smaller, and an interpolated percentile need not equal an actual observation. The 522 accepted intervals also do not establish complete market-hours coverage: the expected number of tradable hourly intervals has not been established for this dataset.

Use the Forex volatility explorer to compare available observations and inspect individual hours. Select a pair and a period, then check the data timestamp and status. The historical results in this article remain tied to the snapshot ending on 5 November 2026 at 12:00 UTC; updates to the tool and comparison widget do not change this example.

A tall bar shows a large absolute change between two hourly observations. It is not necessarily the hour’s high–low range. In this snapshot, the largest EUR/USD absolute hourly log change occurred from 14 October 2026, 12:00 to 13:00 UTC: 0.5681% in magnitude, with a signed log change of −0.5681%. The corresponding simple endpoint return was approximately −0.5664%. Reproducible peak calculation.

Those numbers establish that the observed mid-quote fell between the two endpoints. They do not establish when the move happened inside the hour, whether the price rose earlier in the hour, what caused the move, or the price at which a customer could have traded.

• Read the pair, date, exact start and end, and UTC label before interpreting the bar. An interval ending at 00:00 belongs to two calendar dates.
• Check the signed change beside the absolute change. The tallest positive-height bar may represent a fall.
• Compare the bar with the distribution and its median or upper quantiles, using the same sample.
• Inspect blank periods as missing intervals, not zero activity. Do not connect a weekend gap into one hourly observation.
• Use the underlying observations to confirm the value. A hover tooltip, touch selection or keyboard readout should identify the same interval.

For a dense 30-day chart, the selected interval and its value matter more than the apparent width of a pixel. If several bars share a narrow screen area, use the keyboard or range control to inspect neighbouring real observations. A distribution chart answers “how often were changes this large?”; the time chart answers “when did they occur?”

Liquidity concerns the ability to transact; volatility describes price variation. They are related market characteristics, but one number cannot stand in for the other. OANDA states that volatility and liquidity changes can widen its spreads around news and market openings or closings. Its spread policy explains the conditions. This is not a universal formula for every broker or hour.

A mid-price movement is not an executable round trip. The midpoint is calculated from bid and ask; it does not show which side a transaction used, whether the quote was available for a particular size, or whether execution slipped. Our unfunded-account observations do not supply client fills or prove execution quality.

Once you understand the selected pair’s historical movement, a separate useful step is to inspect its observed spreads and commissions. Keep EUR/USD selected for this example, then check the provider, account type, observation time and commission convention. Broker availability and terms depend on the relevant entity and your country.

Do not divide a current spread by a historical movement statistic from another source or period and call the result a profitability score. Costs and movement need compatible units and observations before even a descriptive ratio would make sense. Broker reviews and country-specific comparisons can provide context for the separate account decision; the movement ranking cannot select a broker for you.

• What was measured: return dispersion, high–low range, true range or absolute endpoint change?
• What are the units: pips, price units, simple percent, log percent or an annualized percentage?
• Which source, price basis, interval length, dates and timezone were used?
• Were the same valid observations used for every pair, and how many remain?
• Are missing, flat and stale data shown explicitly?
• Am I reading a historical description, an options-implied estimate or an unsupported prediction?

A well-labelled number can answer a modest question precisely. It can describe the observed movement of a selected pair over a specified sample. It cannot, by itself, establish a safe trade, an optimal stop, a profitable strategy or a guaranteed future range.

Is volatility the same as a trend?

No. Direction describes whether a price rose or fell; volatility describes a specified kind of variation. A series can trend smoothly or fluctuate sharply while ending close to where it started. Read the signed return separately from an absolute-change or dispersion measure.

Which forex pair is the most volatile?

Specify the universe, metric, source and window first. In the historical IFT eight-pair example ending 5 November 2026 at 12:00 UTC, the leader by median absolute hourly log change differs between 24 hours, 7 days and 30 days. That does not establish a market-wide or current leader.

How do I calculate historical forex volatility?

One common method is the sample standard deviation of consistently sampled returns: s = √[Σ(yᵢ − ȳ)² / (N − 1)]. State whether y is a simple or log return, its units, the sampling interval and any annualization. ATR and median absolute change use different formulas.

Is median absolute hourly change the same as ATR or MAD?

No. Median absolute change measures median(|g|). MAD measures median(|g − median(g)|). ATR needs true ranges derived from highs, lows and previous closes. Endpoint-only hourly quotes do not provide the complete intrabar information required for ATR.

Does high volatility guarantee a better trading opportunity?

No. A larger historical movement does not establish a profitable entry or exit. It also does not specify the spread, commission, available liquidity, slippage or the financial effect of a position.

Can I use hourly changes to find the best hour to trade?

A timeline identifies the hours in which observed changes occurred. Establishing a recurring time-of-day pattern requires a separate, comparable grouping across days and an appropriate sample. One peak does not establish a reliable schedule or a recommendation.

Why can two volatility calculators give different answers?

They may use different sources, bid/ask/mid prices, bar boundaries, units, windows or formulas. Some summarize high–low ranges, others returns. Check those definitions and the accepted observations before treating the difference as an error.

Does a zero or missing bar mean the market did not move?

No. Zero means the two observed endpoints were equal under the calculation. A missing bar means a valid interval was unavailable. Neither reconstructs the price path inside that period.