Correlation and How to Compare Co-Movement
In this chapter
Intermediate
A correlation of +0.60 says two sets of returns have tended to move in the same direction relative to their averages, with some scatter around that pattern. The Pearson correlation used here measures linear co-movement. It scales covariance by the spread of both return series, so the result has no units and fits on a common scale from −1 to +1.
Put covariance on a common scale
Use paired returns over identical intervals. The formula for the sample is
\[ r_{xy}=\frac{s_{xy}}{s_xs_y}. \]
\(r_{xy}\) is sample Pearson correlation, \(s_{xy}\) is the sample covariance and \(s_x,s_y\) are the sample standard deviations of the same paired observations. All three quantities use the same sample conventions and dates. For a population, use its covariance and both population standard deviations. If returns are expressed as percentage points, pp² in the numerator cancels pp times pp in the denominator.
A value of +1 means the observations fall on an upward-sloping straight line. A value of −1 means they fall on a downward-sloping straight line. Zero means there is no linear association in the data. It leaves room for a curved relationship, and even a high correlation alone tells you little about which event caused the moves. For an investment example of shared drivers, see Diversification.
Worked case. Interpret a positive correlation
Here are four sample months of hypothetical nominal USD simple total returns including income, before fees and taxes. The two means are zero, so the numerical returns in percent also give the deviations in percentage points.
| Month | X return | Y return | Cross-product (pp²) |
|---|---|---|---|
| Jan | −3% | −1% | +3 |
| Feb | −1% | −3% | +3 |
| Mar | +1% | +3% | +3 |
| Apr | +3% | +1% | +3 |
Each series has a squared-deviation sum of 20 pp², so each sample standard deviation is \(\sqrt{20/3}\) pp. The signed cross-product sum is 12 pp², giving covariance \(12/3=4\) pp². Divide by the product of the standard deviations:
\[ r_{xy}=\frac{4}{20/3}=0.60. \]
All four pairs have the same sign here, yet the sizes vary. Correlation describes how closely the cloud of points follows a line. A reading of 0.60 does not assign a 60% chance that next month’s returns share a sign.
Change the scale or remove the variation
Worked case. Double one series, then try a constant series
First double every Y return as a mathematical transformation. The new series is −2%, −6%, +6%, +2%. Its covariance with X doubles from 4 to 8 pp² because each Y deviation doubles. Its standard deviation also doubles, so the denominator grows from \(20/3\) to \(40/3\) pp². The result is still \(8/(40/3)=0.60\). Converting returns from percentages to decimals also uses a positive multiplier and leaves correlation unchanged. Multiplying just one series by a negative factor reverses the direction of the line and the sign of correlation.
Now replace Y with a hypothetical position that earns exactly +1% in every month. Its mean is +1% and all four deviations are zero. Covariance with X is zero and Y’s standard deviation is zero. The calculation gives \(0/0\), which is undefined. A spreadsheet error or missing output in this case indicates there was no variation to standardise.
| Version of Y | Sample covariance with X (pp²) | Y sample standard deviation (pp) | Correlation |
|---|---|---|---|
| Original | 4.00 | 2.58 | +0.60 |
| Twice the original | 8.00 | 5.16 | +0.60 |
| Constant +1% | 0.00 | 0.00 | Undefined |
Calculate with the unrounded spreads.
Worked case. A clear curve gives zero linear correlation
Suppose X returns −2%, −1%, +1% and +2% in four months. A payoff that benefits from larger moves in either direction gives Y returns of 4%, 1%, 1% and 4% on those dates. X averages zero and Y averages 2.50%. The cross-products are −3, +1.5, −1.5 and +3 pp². They cancel, and both series have positive spread, so correlation is zero. Knowing the size of the X move still tells you Y is higher in the outer two observations. Pearson correlation is zero here because the pattern bends.
A four-month sample gives a lesson in the arithmetic. For a decision, check whether the relationship holds across other periods, including times of market stress, and whether any single paired move dominates it.
Practice questions
Choose one answer per question. Round only the final result where needed.
1. Standardise covariance
Sample covariance is 6 pp². The two sample standard deviations are 3 and 4 pp. What is correlation?
2. Interpret a negative result
What does a sample correlation of −0.80 show?
3. Change return units
Convert just X from decimal returns to percentages. What happens to Pearson correlation?
4. Handle a flat series
One asset’s return is +0.50% in every paired month. What is its correlation with a variable series?
5. Spot the bad input
Reported covariance is 15 pp² and the standard deviations are 3 and 4 pp, supposedly on the same basis. What should you do?
6. Read zero carefully
Both series have nonzero spreads and sample correlation is zero. What can you conclude?