Finance education, free and open.An independent learning project
THEFINANCE ARCHIVES
Finance foundationsEquitiesFixed incomeDerivativesEconomicsFinancial statementsPortfolio management
Home / Quantitative methods / Covariance and How Paired Returns Move Together
Quantitative methods

Covariance and How Paired Returns Move Together

In this chapter

Intermediate

When two funds have unusually good months together and unusually poor months together, their covariance tends to be positive. When one tends to be above its own average while the other is below its own average, covariance tends to be negative. The calculation compares both returns on the same dates.

Multiply two deviations from their own means

Start with two columns of returns for the same intervals. Keep each pair on its date. Calculate each column’s own arithmetic mean and subtract it from each return in that column. Multiply the two deviations on each date. Two positive deviations or two negative deviations contribute a positive product. A positive and a negative contribute a negative one. If either deviation is zero, that pair contributes zero.

For a sample of \(n\) paired observations, use

\[ s_{xy}=\frac{\sum_{t=1}^{n}(x_t-\bar x)(y_t-\bar y)}{n-1}. \]

\(s_{xy}\) is sample covariance, \(x_t\) and \(y_t\) are the two returns for period \(t\), and \(\bar x\) and \(\bar y\) are their means on the matched dates. The sample calculation requires \(n\ge2\). If the matched rows are the entire equally weighted population you intend to describe, use the full-population covariance

\[ \sigma_{xy}=\frac{\sum_{t=1}^{N}(x_t-\mu_x)(y_t-\mu_y)}{N}. \]

Here \(N\) counts all the paired rows in that population and \(\mu_x,\mu_y\) are the respective population means. The sums add the signed products across periods. When both deviations are measured in percentage points, the product and covariance are in pp². With decimal returns, use decimal return units squared. A covariance’s numerical size depends on the spread and units of both inputs.

Worked case. Keep the calendar beside the returns

These are four consecutive months of hypothetical nominal USD simple total returns, including income and before fees or taxes. Treat them as a sample for illustration. Fund X averages 2.00% and Fund Y averages 1.00%.

MonthX returnY returnX deviation (pp)Y deviation (pp)Product (pp²)
Jan−1%−2%−3−3+9
Feb+1%0%−1−1+1
Mar+3%+2%+1+1+1
Apr+5%+4%+3+3+9
Sum00+20

Sample covariance is \(20/3=6.67\) pp². February contributes +1 pp² even though neither fund lost money. Both were below their own average. This is why checking signs of the raw returns can give the wrong answer.

Check what a date mismatch can do

Worked case. Reverse one column without changing its values

Say Fund Y was pasted from a sheet sorted newest first and its date column was left behind. The two lists contain the same values as before and keep the same means and individual spreads.

Row labelledX returnY return placed on that rowX deviation (pp)Y deviation (pp)Product (pp²)
Jan−1%+4%−3+3−9
Feb+1%+2%−1+1−1
Mar+3%0%+1−1−1
Apr+5%−2%+3−3−9
Sum00−20

The broken pairing gives \(-20/3=-6.67\) pp² and makes the funds look like they offset one another. Checking means and individual variances would miss the error. Rejoin by dates and verify both the start and end of each return interval. A join on month labels helps only if both entries represent the same span. If an asset has a February-to-April return because March was missing, that number cannot stand in for its April monthly return. If one observation is missing, recompute the means on the overlapping rows. Filling a missing return with zero inserts a price outcome you may never have observed.

A few large products can outweigh many small ones, so covariance is about the combined signed size of the moves. If you need a unit-free comparison of linear co-movement, scale covariance by both series’ standard deviations.

Practice questions

Choose one answer per question. Round only the final result where needed.

1. Find the sign from the deviations

On one date X returned +1% against a mean of +3%. Y returned −1% against a mean of −2%. What is that date’s cross-product?

Question 1 choices

C is correct

  • A. The two deviations point in opposite directions.
  • B. This multiplies the raw return numbers.
  • C. X is two points below its mean and Y is one point above.

2. Calculate sample covariance

Five paired dates give a signed cross-product sum of 24 pp². What is sample covariance?

Question 2 choices

A is correct

  • A. Divide the signed sum by \(5-1=4\).
  • B. Dividing by five describes the full five-row population.
  • C. Twenty-four is the sum before normalising by the sample denominator.

3. Check the date pairing

One return series was sorted from lowest to highest and its dates were lost. Its mean is unchanged. What’s needed before computing covariance?

Question 3 choices

B is correct

  • A. Means reveal nothing about whether individual observations are paired correctly.
  • B. Covariance uses paired deviations, so recover the shared return intervals first.
  • C. Sorting both by magnitude can manufacture co-movement.

4. Read the units

Sample covariance is −0.0012 in decimal return units squared. What is it in pp²?

Question 4 choices

C is correct

  • A. This rescales just one return unit by 100.
  • B. Changing units preserves the negative sign.
  • C. Both return units are scaled by 100, so multiply by 10,000.

5. Describe a complete set

A complete four-pair window has signed products −6, +2, +1 and −1 pp². What is the population covariance for that window?

Question 5 choices

A is correct

  • A. The sum is −4 pp². Divide by all four pairs.
  • B. Dividing by three gives the conventional sample estimate.
  • C. This adds the product magnitudes and loses the signs.

6. Handle a missing month

Fund X has January to April. Fund Y is missing February. You plan to calculate covariance using the three overlapping one-month intervals. Which means do you use?

Question 6 choices

A is correct

  • A. The deviations and their means must describe the same matched rows.
  • B. The unpaired February observation would still influence X’s baseline.
  • C. An absent observation cannot establish that its actual return was zero.

Find a finance concept

Press Enter to see all matching chapters.