Correlation measures how strongly two variables move together. In market research, the most common numerical measure is the Pearson correlation coefficient, r, which standardizes covariance by the two variables’ standard deviations. The result is dimensionless and ranges from −1 to +1: +1 indicates a perfect positive linear relationship, −1 a perfect negative linear relationship, and 0 zero linear correlation. That last case is easy to overread. A Pearson coefficient near zero can coexist with a strong nonlinear relationship.
One number can hide several relationships
A correlation coefficient is a property of the chosen data and sample, not a market mechanism. It does not establish causation, reveal which variable leads the other, or show that the relationship will persist. Two markets can move together because both are responding to a third factor. A full-sample estimate also averages across potentially different regimes. Rolling Correlation addresses that time variation by recalculating the statistic over a moving window. Cointegration asks a different question: whether a linear combination of non-stationary series is stationary, which concerns a long-run statistical relationship rather than the strength of contemporaneous co-movement.
The construction is part of the statistic
Daily-return correlation can differ from weekly-return correlation, and price levels, arithmetic returns and logarithmic returns need not tell the same story. With non-stationary time series, high association in levels can be spurious unless the time-series properties, including possible cointegration, are handled explicitly. Outliers matter too: ordinary Pearson correlation can be materially influenced by extreme observations. Sample dates, missing observations and continuous-futures construction can therefore change the estimate enough for two competent researchers to report different correlations for what appears to be the same market pair.
Correlation also removes scale, which creates a practical trap. Two assets can be almost perfectly positively correlated while one moves twice as much as the other. A one-for-one hedge would then leave material residual risk. In minimum-variance hedge sizing, relative volatility enters alongside correlation; high correlation alone does not determine the hedge ratio. The useful question is therefore not simply “What is the correlation?” but “Correlation of which variables, transformed how, over what sample, and for what decision?”