Advertisement IC Markets

Rolling 20-Day vs 60-Day Correlation: Which Window to Trust

Optimize MetaTrader 4, MetaTrader 5, cTrader, and TradingView for speed. Discuss Level II Market Depth (DOM), custom hotkeys, and volume indicators.
Post Reply
FXS Prop Desk
Posts: 1689
Joined: Wed Oct 07, 2026 9:52 am

Rolling 20-Day vs 60-Day Correlation: Which Window to Trust

Post by FXS Prop Desk »

Correlation between currency pairs changes over time. The window used for calculation affects how quickly the number reacts.

Comparison:
  • 20-day correlation: reacts quickly to new relationships, but is noisier.
  • 60-day correlation: more stable, but slower to show changes.
Uses:
  • Risk management: 20-day shows current exposure better.
  • Strategy development: 60-day or longer shows structural relationships.
  • Comparing both: if they differ a lot, the relationship is changing.
Example: EURUSD and GBPUSD often show high positive correlation, but it drops during UK-specific events.

In practice, many traders use the 20-day value to size exposure and the 60-day value to decide which pairs to trade together. When the shorter window turns sharply lower, it often signals that a currency-specific story has started.

Which correlation window do you use, and why?
FXS Official Team | The Idea Lab & Strategy
Recommended broker for automated trading & scalping IC Markets
Shadow Trader
Posts: 5594
Joined: Wed Sep 09, 2026 5:19 pm

Re: Rolling 20-Day vs 60-Day Correlation: Which Window to Trust

Post by Shadow Trader »

I run both windows and act on the gap between them rather than picking one. When the 20 day sits more than 0.25 under the 60 day, I treat EURUSD and GBPUSD as separate trades again.

Using daily returns rather than price levels matters as much as the window. Price based correlation can read 0.9 in a month when daily moves barely agree.
img_13733_r1.png
img_13733_r1.png (83.91 KiB) Viewed 114 times
The illustration uses synthetic daily returns with a UK specific episode: the 20 day falls from about 0.87 to 0.14 while the 60 day only bottoms at 0.61, and the gap flag stays on for 11 sessions. The 60 day keeps sliding after the story ends, purely because old days are still inside its window.

That lag cuts both ways. By day 157 the short window was back above 0.8, yet the long one stayed under 0.75 until day 173, so sizing off 60 days meant treating the pairs as diversified for over two weeks too long.

For risk I use whichever window reads higher, not the 20 day alone as the post suggests. Correlations snap back toward 1 in a broad dollar move, which is exactly when a low reading turns two positions into one.
Shadow Trader
Posts: 5594
Joined: Wed Sep 09, 2026 5:19 pm

Re: Rolling 20-Day vs 60-Day Correlation: Which Window to Trust

Post by Shadow Trader »

Sample size is what I would add on the 20 day side. With only 20 daily returns, a reading around 0.8 has a standard error of roughly 0.08, so a drop from 0.85 to 0.75 is mostly noise and not a new relationship.

For sizing scalps I moved to intraday returns instead. Fifteen minute returns across the last ten London sessions give about 320 observations, and that number reacts within days while staying far steadier than the daily 20 day reading.

The daily close you pick also changes the result. Closes at 17:00 New York and at midnight London can give noticeably different figures for EURUSD and GBPUSD, because one version splits the London morning across two days.

Holidays are the other trap. A UK bank holiday gives a thin, quiet sterling day, and two of those inside a 20 day window drag the figure down for reasons unrelated to how the pairs actually move together.

Has anyone compared session based correlation with the daily windows over a full year? I would like to know whether the intraday version gives earlier warning or just more noise.
Shadow Trader
Posts: 5594
Joined: Wed Sep 09, 2026 5:19 pm

Re: Rolling 20-Day vs 60-Day Correlation: Which Window to Trust

Post by Shadow Trader »

Correlation coefficients between currency pairs are often quoted as if they were fixed properties, yet they drift considerably. EUR/USD and GBP/USD are usually strongly positively correlated, but the strength changes with the news cycle, the session, and whether the move is driven by dollar flows or by a currency-specific story.

Estimates also depend on the measurement window and sampling frequency. Correlation computed on one-minute returns can differ noticeably from correlation computed on daily returns, partly because of asynchronous quoting and differing liquidity. Short windows produce noisy numbers that swing widely.

During stress, many dependencies tend to tighten, and a set of positions that looked diversified can behave like one large position. Two trades on different pairs that share the same dollar leg may effectively double a single exposure.

A more reliable approach is to think in terms of underlying currencies rather than pairs. Long EUR/USD and short USD/CHF are largely the same dollar view. Tracking rolling correlation with an eye on its variability, and sizing as though correlated trades are one trade, avoids the false comfort of nominal diversification. Correlation also says nothing about timing, since one pair frequently leads another by a short interval that is hard to exploit after costs.
Post Reply