What it means
The measure most people mean is the Pearson correlation coefficient, usually written as r. It compares how far each observation sits from the average of its own series, and asks whether the two series deviate in the same direction at the same time.
Because the result is standardised, you can compare an r calculated on dollars with one calculated on units or percentages. Interpretation depends heavily on context, so fixed thresholds are misleading.
In finance, an r of 0.6 between two asset returns is a strong relationship, while in a controlled engineering measurement the same value would be considered weak. What matters is comparing correlations within a consistent context rather than against a universal rule.
The best-known application is portfolio diversification. Combining assets with low or negative correlation reduces the volatility of the overall portfolio, because losses in one holding are partly offset by gains in another, which is the mathematical basis for not concentrating in one sector.
A common trap is that correlations between assets tend to rise sharply in a crisis, exactly when the diversification benefit is needed most. Outside investing, correlation is a standard tool for commercial analysis.
Marketers test the relationship between advertising spend and enquiries, retailers test temperature against product sales, and finance teams test volume against overhead cost to see how variable a cost really is. It is usually the first step before building a regression model, which goes further and estimates the size of the effect.
Two limitations deserve emphasis. Correlation only detects linear relationships, so a strong curved relationship can produce an r close to zero, and it is highly sensitive to outliers, where a single unusual month can dominate a small sample.
Plotting the data on a scatter chart before trusting the number takes a minute and prevents most misinterpretations. The most repeated warning is also the most ignored: correlation does not establish causation.
Two series can move together because one drives the other, because both are driven by a third factor such as the economic cycle, or through pure coincidence in a small sample. Establishing cause requires a controlled test or a credible mechanism, not a higher r.
In practice
Real-world examples.
Example
A pension fund reviews two equity funds and finds their monthly returns have a correlation of 0.93. Holding both adds very little diversification, so the trustees replace one with an infrastructure fund whose correlation to equities is 0.25.
Example
A garden centre chain correlates weekly average temperature with sales of outdoor furniture and finds an r of 0.78. It uses the relationship to build a weather-adjusted forecast, which improves stock planning in the spring peak.
Example
A subscription business tests the correlation between support ticket response time and customer churn across 200 accounts and finds an r of 0.41. The link is moderate rather than decisive, so the team runs a controlled trial on faster response times rather than immediately reallocating budget.
Think of it
“Correlation measures how two things move together-do they go the same direction or opposite?
Formula
Calculation
Pearson's correlation coefficient is: r = sum of (x - mean of x)(y - mean of y) divided by the square root of [ sum of (x - mean of x) squared x sum of (y - mean of y) squared ].
A company examines five months of marketing spend (x, in $000) against revenue (y, in $000). Spend was 10, 20, 30, 40 and 50; revenue was 100, 140, 150, 190 and 220.
Mean spend = 150 / 5 = 30. Mean revenue = 800 / 5 = 160.
Deviations in spend: -20, -10, 0, 10, 20. Deviations in revenue: -60, -20, -10, 30, 60.
Products of the deviations: 1,200 + 200 + 0 + 300 + 1,200 = 2,900.
Sum of squared spend deviations: 400 + 100 + 0 + 100 + 400 = 1,000.
Sum of squared revenue deviations: 3,600 + 400 + 100 + 900 + 3,600 = 8,600.
r = 2,900 / square root of (1,000 x 8,600) = 2,900 / square root of 8,600,000 = 2,900 / 2,932.6 = 0.99.
An r of 0.99 shows an almost perfectly linear relationship across these five months. The analyst should still be cautious: five data points is a very small sample, and the correlation alone does not prove that the spend caused the revenue.Case study
Seen in the real world.
Fenwick Outdoor Group is an invented company used purely for illustration, selling camping equipment through its own website. Its marketing team reported a correlation of 0.88 between weekly paid search spend and weekly revenue over two years, and used it to argue for a $600,000 budget increase.
A finance analyst plotted the data and noticed that both series peaked every summer and collapsed every winter. When she measured the correlation within each season separately, the relationship fell to 0.31, because most of the apparent link came from seasonality driving both variables rather than spend driving revenue.
In this fictional example, Fenwick ran a regional holdout test instead, cutting paid search in two comparable areas for six weeks. Revenue in those areas fell by only 4%, so the company redirected most of the proposed increase into range expansion, illustrating why a high correlation is a prompt for a proper test rather than a conclusion.
Watch out
Common mistakes.
- Reading correlation as proof of cause. A strong relationship may reflect a third variable such as seasonality or the economic cycle affecting both series at once.
- Calculating it on a handful of data points. Small samples produce unstable coefficients, and a single unusual month can swing the result dramatically.
- Assuming a low coefficient means no relationship. Pearson's measure only detects straight-line patterns, so a strong curved or threshold relationship can still show an r near zero.
Questions
People also ask.
What counts as a strong correlation?
It depends entirely on the field; in financial markets values above about 0.7 are usually treated as strong, but the sensible comparison is against other correlations in the same context.
What is the difference between correlation and covariance?
Covariance measures whether two series move together but its size depends on the units used, while correlation divides out the units to give a comparable figure between -1 and +1.
What is R-squared and how does it relate?
R-squared is the correlation coefficient squared, and it expresses the proportion of variation in one variable that moves in line with the other, so an r of 0.6 gives an R-squared of 0.36, or 36%.
From the founder's library
Take it further with the book.
Build your financial confidence beyond this definition. Shihan's full-length guide, Accounting Fundamentals, takes the same plain-English approach and turns it into a complete, practical playbook for non-finance managers, business owners and students - with chapter-end quiz answers and presentation slides included.
25% off with code MMHQ25, applied at checkout. Priced in USD - checkout may show the equivalent in your local currency.
View the book and save 25%Related