Direct answer
Positive correlation between two currency returns is not a fixed property of the currencies themselves. It is a measured relationship that depends on the timeframe you use to observe returns and, indirectly, the timeframe implied by your holding period. When you change the observation window length (for example, minutes vs days) you often change what portion of market behavior you capture—noise, short-term dynamics, or slower-moving drivers—so the measured correlation can move noticeably.
Mechanism and definition
Positive correlation means that, over a defined period, the two return series tend to move in the same direction more often than chance. In practice, “correlation” usually refers to the correlation coefficient computed from returns over a sample of times.
Timeframe can affect the result in two main ways:
- What data you include (observation window): A short window typically includes more abrupt price changes and temporary effects. A longer window aggregates many events, making the relationship reflect more persistent co-movement.
- How you model the return horizon: If your holding period is longer, the relevant returns are sums over more intervals. Even if you compute correlation using the same sampling frequency, using longer-horizon returns changes the inputs.
This is a general statistical effect: correlation estimates are sensitive to the sample you use, and currency return relationships are not guaranteed to be stable through time.
Example: changing observation window
Consider a simplified setup with two time series of returns, where both series are influenced by (a) a slower common driver and (b) short-lived independent shocks.
- Short timeframe observation: The independent shocks represent a larger share of total variation, increasing randomness. This can weaken measured correlation or make it flip sign in some windows.
- Long timeframe observation: The short-lived shocks partially cancel out when returns are aggregated across many intervals, while the slower common driver continues to move both series together. Measured positive correlation can strengthen.
However, the opposite can also happen. If the common driver is temporary or regime-dependent, extending the timeframe can mix multiple regimes and reduce correlation.
Limitations and failure modes
Key limitations to keep in mind:
-
Non-stationarity (regime changes): The relationship can change over time. A timeframe that spans different regimes can produce a misleading “average” correlation.
-
Noise dominance at short horizons: At very short intervals, microstructure effects and irregular price updates can contribute substantial variability, making correlation estimates unstable.
-
Mismatched sampling and data handling: Missing observations, different trading session coverage, or inconsistent timestamps can distort returns and therefore correlation.
-
Costs and execution frictions (indirectly): Even though correlation is computed from price data, real outcomes from any strategy depend on spreads, fees, and execution quality. Those factors can change how a relationship matters in practice.
-
Historical correlation ≠ future correlation: Even if positive correlation appears over one timeframe, it does not guarantee the same relationship will hold later.
Verification and next question
A way to independently verify timeframe sensitivity is to recompute correlation across multiple observation windows and horizons using the same data-cleaning rules. Look for patterns such as: correlation strengthening with longer windows, correlation reversing across regimes, or correlation remaining consistently positive.
A useful next question is not only “Is the correlation positive?” but also “Is it robust across reasonable time windows and return horizons?” If the answer changes substantially as you vary the timeframe, then the observed positive correlation is likely driven by short-lived dynamics rather than a stable co-movement.