Monthly Ranges: What Two Years of Data Show and What They Don't

Everyone talks about the summer lull and thin December markets. We computed the median daily range by month across twelve pairs — and got a picture that illustrates the limits of the sample more than it shows seasonality.

Volatility seasonality by month: the result

For each pair we took the median daily range over the whole sample and the median range within each month, then expressed the second as a percentage of the first. A value of 100% means an ordinary day for that pair, lower is quieter, higher is wider. The table shows the median across twelve pairs.

• January: 106% of the pair's median range (43 bars)
• February: 106% of the pair's median range (40 bars)
• March: 112% of the pair's median range (43 bars)
• April: 109% of the pair's median range (42 bars)
• May: 98% of the pair's median range (43 bars)
• June: 91% of the pair's median range (43 bars)
• July: 86% of the pair's median range (47 bars)
• August: 102% of the pair's median range (43 bars)
• September: 97% of the pair's median range (43 bars)
• October: 92% of the pair's median range (46 bars)
• November: 104% of the pair's median range (41 bars)
• December: 94% of the pair's median range (44 bars)

Measured across 518 daily bars for each of 12 currency pairs. The sample is public, the methodology is in how we run backtests, and the calculation is reproduced by a script in the repository.

Monthly ranges: the order of magnitude matches expectations

July came out the quietest month at 86% of a pair's median range, March the widest at 112%. The summer months are indeed in the lower part of the list, consistent with the idea of a holiday-season lull. The spread between the extreme months is about a quarter, noticeably larger than the spread across weekdays we measured earlier.

Why you cannot build decisions on this

The key limitation is arithmetic: the sample covers two years, so each month has two observations. Two observations do not establish seasonality — what we see could equally be a property of these particular two years. A single volatile March in 2025 can move the March median, and on a sample this size that cannot be distinguished from a regularity.

What an honest conclusion would require

Ten to fifteen years of data, meaning ten to fifteen observations per month, plus a stability check on subperiods. That would separate seasonality from the quirks of individual years. Our sample does not allow it, and we publish the figures as an observation rather than a regularity.

What to do about it in practice

Change nothing in your rules. If seasonality exists at all, its magnitude is a few percent, whereas the difference between a quiet and an active day within one week can be twofold. The release calendar for a specific day says incomparably more about upcoming volatility than the number of the month: we reached the same conclusion checking days of the week.

This material is for educational purposes and is not individual investment advice.