What “volatility in pair spreads” means
Pair spreads are the difference between the buy and sell price for a currency pair at a given moment. “Volatility in pair spreads” describes how strongly that spread fluctuates over time. It is a variability measure: it does not say whether spreads will widen or narrow next.
To measure it, start by defining what you treat as the spread value and at what times you record it. For example, you can sample the spread every minute, every hour, or at the end of each day. The measured volatility will change with that sampling interval.
Measurement choices: quantify variability, not direction
A straightforward way to measure spread volatility is to compute statistics from a time series of observed spread values.
1) Standard deviation (typical variation)
Assume you have spread observations (s_1, s_2, …, s_N) over a chosen time window, sampled at a fixed interval. Compute the mean (\bar{s}) and then the standard deviation:
[ \sigma = \sqrt{\frac{1}{N-1} \sum_{i=1}^{N}(s_i-\bar{s})^2} ]
Interpretation: higher (\sigma) means the spread varies more around its average during the window.
2) Rolling window volatility (how stability changes)
Instead of one (\sigma) for the whole window, compute (\sigma) over shorter rolling segments (for example, last 30 minutes or last 10 trading hours). This produces a volatility profile that can show regime changes.
Assumption: rolling windows use the same sampling interval and definition of spread throughout the series.
3) Range and percent changes (robustness to outliers)
Two simpler alternatives are:
- Range volatility: (\max(s_i)-\min(s_i))
- Percent range: ((\max(s_i)-\min(s_i))/\text{reference})
You must state the reference. Using the minimum, mean, or end-of-window value changes the number and the interpretation. Range measures can be sensitive to spikes, so you may prefer median-based measures in principle, but then you need a clear definition.
4) High/low spread moments (distribution-focused)
You can also measure spread variability using quantiles (e.g., interquartile range between the 25th and 75th percent spreads). This focuses on the “typical” spread variation and reduces sensitivity to extreme moves.
Mechanism: why the measurement depends on data and costs
Even if the definition of spread is simple, the observed time series can differ depending on the following variable factors:
- Data source/provider feed: different quotes can produce different spread observations.
- Sampling rule: a “minute sampled” spread series misses sub-minute widening.
- Execution and timing: spreads can differ between displayed quotes and what is actually filled.
- Time-of-day effects: volatility can cluster around specific market hours.
To keep the measurement meaningful, separate stable mechanics from variable conditions. The stable mechanic is “spread at time (t) minus spread at time (t)”. The variable parts are how quotes arrive, how frequently you sample, and whether your notion of “spread” includes additional components in your dataset.
Evidence or example (with explicit assumptions)
Suppose you record spread values in pips every 10 minutes for one week. You obtain (N) observations (s_i).
- Compute the standard deviation (\sigma) for the full week.
- Then compute (\sigma) for each rolling 1-day window.
If the full-week (\sigma) is high but the rolling 1-day (\sigma) is low most days, that indicates occasional short periods of unusually wide or unstable spreads. Conversely, if both are high most days, spreads are persistently variable.
This example measures variability, not direction. A high (\sigma) does not imply you will observe widening next; it only summarizes how much you saw spreads move during the measured window.
Limitations and risks (what can fail)
At least one material limitation is that spread volatility is not a stable property.
- Regime shifts: historical variability can change when liquidity conditions or market structure changes.
- Market microstructure effects: short, fast widening events may be missed or exaggerated depending on your sampling interval.
- Provider differences: two sources can show different spread series even for the same pair, so volatility numbers are not automatically comparable.
- Hidden cost effects: some datasets or reporting conventions may reflect more than the pure quoted bid-ask difference you intended to measure, depending on how the feed is constructed.
A failure mode: mixing definitions.