Direct answer
Volatility in USD/CAD vs AUD/USD can be measured by quantifying how much exchange rates fluctuate over time. The most common approach is to convert price changes into returns, then calculate a statistical dispersion measure such as rolling standard deviation (realized volatility). A second approach is to use options-implied volatility, which reflects what an options market implies about future fluctuation, but only if options data is available. In both cases, the key is to define the inputs (which price, which time step, which window length) and state the assumptions, because the “volatility” number changes when these choices change.
If your goal is to compare USD/CAD and AUD/USD, use the same measurement method for both series, otherwise differences may reflect methodology rather than market behavior.
Mechanism or definition
“Volatility” is usually defined as variability of returns, not variability of raw prices. A return is a change in exchange rate over a chosen time step. For example, if you sample daily rates, a common choice is the log return between today’s rate and yesterday’s rate. Once you have a sequence of returns, you can compute realized volatility.
Option A: realized volatility (historical)
A typical realized-volatility workflow is:
- Select the reference exchange rate for each pair (for instance, a mid-market rate or another consistent provider quote).
- Choose a sampling frequency (daily, hourly, etc.).
- Choose a rolling window length (for example, 20 or 60 observations).
- Compute volatility as the standard deviation of returns inside the rolling window.
- (Often) annualize if you want a comparable long-run scale; annualization requires an assumption about how many periods fit in a year.
Assumption example: If daily observations are used, annualization usually assumes 252 trading days. If you instead use calendar days, a different count applies. The volatility estimate is not “wrong,” but it is based on the chosen scaling assumption.
Option B: implied volatility (options-based)
If options are traded on relevant contracts, implied volatility is derived from option prices using a pricing model. It is a market-implied measure rather than a direct measurement of past dispersion. The practical limitation is that implied volatility availability depends on contract existence, liquidity, and the specific model and inputs used to back out implied volatility.
Evidence or example
Below is a concrete example of how two estimators can be compared without predicting movement.
Example setup (assumptions stated):
- You use the same time step for both USD/CAD and AUD/USD (e.g., daily sampled rates).
- You compute log returns.
- You use a rolling window of 30 daily returns.
What you compare:
- The resulting rolling realized volatility time series for USD/CAD and for AUD/USD.
- Summary statistics such as average volatility over a chosen subperiod.
Interpretation (without forecasting): If USD/CAD’s rolling realized volatility is higher than AUD/USD’s over the same dates, that means USD/CAD returns showed greater historical dispersion under your definitions. It does not guarantee similar behavior will continue.
A second evidence angle is to compare realized volatility changes against major “regime” changes in the data (for instance, periods with abrupt jumps). Even then, you should treat the relationship as descriptive: volatility spikes can come from sudden repricing, but the cause is not determined solely by the volatility metric.
You can also compare realized volatility computed at different sampling frequencies (daily vs hourly). If hourly volatility is much larger, that indicates higher short-term variability or the presence of microstructure effects (bid-ask effects, timing differences, or more frequent price jumps). This helps you understand what part of “volatility” you are actually measuring.
Limitations and risks
1) Historical volatility is not future volatility
Realized volatility summarizes what already happened in the selected window. It does not establish future results, because the next period’s dispersion can differ due to new macro conditions, policy actions, or market microstructure changes.
2) Measurement choices can change the number
Volatility estimates are sensitive to:
- Window length (short windows respond faster to changes; long windows smooth them).
- Sampling frequency (more frequent sampling captures more short-term variation).
- Return definition (log returns vs simple returns).
- Scaling/annualization assumptions.
- Data quality (missing timestamps or irregular observation spacing can distort results).