Skip to content

FRM Exam Part I · Measuring Return, Volatility, and Correlation

Correlation Pitfalls and Dependence Measures in FRM Part 1

Updated 11 October 2026 · Fact-checked

Linear (Pearson) correlation measures only straight-line dependence. Outliers distort it, it can be zero when variables are strongly dependent, and it can be high for unrelated trending series (spurious correlation). Spearman's rho and Kendall's tau use ordering instead of values, so they capture monotonic dependence and resist outliers.

Understand Correlation Pitfalls and Dependence Measures

Linear correlation (Pearson) is covariance divided by the product of the two standard deviations. It runs from -1 to +1 and tells you how well a straight line describes the relationship between two variables. That is all it tells you.

This causes several pitfalls. First, nonlinearity: if Y = X² and X is symmetric around zero, Y depends completely on X but the correlation is zero. Zero correlation does not mean independence. Independence implies zero correlation, but not the other way round. The exception is the bivariate normal distribution, where zero correlation does imply independence.

Second, outliers. Pearson uses actual values, squared deviations in particular, so one extreme observation can raise, cut or even flip the estimate. Third, spurious correlation: two series that trend or are nonstationary, such as price levels, can show high correlation with no real link. A common driver or pure chance can also produce it. Correlation also changes over time. In a crisis, asset correlations often rise, so a single full-sample number can understate risk when you need it most.

The alternatives are rank correlations. They replace values with ranks or with pair orderings. Spearman's rho is the Pearson correlation of the ranks. Kendall's tau compares every pair of observations and counts how many pairs move in the same direction (concordant) and how many in opposite directions (discordant).

Rank measures are unchanged by any strictly increasing transformation of the data, for example taking logs. They are far less sensitive to outliers. They still measure only monotonic dependence, so a U-shaped relationship can still give a value near zero. Correlation, of any kind, does not prove causation.

Key formulas to remember

Pearson correlation
ρ(X,Y) = Cov(X,Y) ÷ (σX × σY)
Measures linear dependence only. Range -1 to +1. Zero does not imply independence.
Spearman rank correlation (no ties)
ρs = 1 − 6 × Σd² ÷ [n × (n² − 1)]
d is the difference between the two ranks of each observation, n is the number of pairs. With ties, compute Pearson on the ranks.
Kendall's tau (no ties)
τ = (nc − nd) ÷ [n(n − 1) ÷ 2]
nc is the number of concordant pairs, nd the number of discordant pairs. The denominator is the total number of pairs.
Kendall's tau shortcut
τ = 1 − 2 × nd ÷ [n(n − 1) ÷ 2]
Valid when there are no ties, because nc + nd equals the total number of pairs.
Concordant and discordant pair
Concordant if (xi − xj)(yi − yj) > 0; discordant if < 0
Compare every pair of observations i and j.
Independence and correlation
Independent ⇒ ρ = 0, but ρ = 0 does not ⇒ independent
Only for the bivariate normal does zero correlation imply independence.

How to solve Correlation Pitfalls and Dependence Measures questions

Use this method for any question on correlation limitations or dependence measures.

  1. 1Identify what the question asks: a pitfall to name, a statistic to compute, or a measure to choose.
  2. 2If it describes a data feature, match it to a pitfall: a curved relationship (nonlinearity), an extreme point (outlier), trending levels (spurious), or a crisis period (unstable correlation).
  3. 3If it asks you to choose a measure, ask whether the relationship is only monotonic or has outliers. If yes, prefer Spearman or Kendall. If the data are bivariate normal and clean, Pearson is fine.
  4. 4For Spearman, rank each variable (average ranks for ties), find d for each pair, square and sum, then apply the formula.
  5. 5For Kendall, list the pairs. Count the discordant pairs, which is usually the smaller number, then use the shortcut or the full formula.
  6. 6Sanity check: the sign and size should match the data's pattern. A fully reversed ordering gives -1 and a fully matching ordering gives +1.
  7. 7Pick the answer that states the condition correctly. Avoid options that say zero correlation proves independence or that high correlation proves causation.

Quickest way: Rank and count method

When to use it: Computation questions with 5 to 8 pairs where ranks or orderings are given or easy to build.

  1. Sort the pairs by X so that X ranks run 1, 2, 3, ... Write the Y ranks in that order.
  2. Spearman: take the difference in each position, square it, add them up, then plug into 1 − 6Σd² ÷ [n(n² − 1)].
  3. Kendall: for each Y rank, count how many later Y ranks are smaller. The total is nd.
  4. Compute τ = 1 − 2nd ÷ [n(n − 1) ÷ 2].
  5. Check that |τ| is usually smaller than |ρs| for the same data. If it is not, recheck your counts.

Common mistakes in Correlation Pitfalls and Dependence Measures

  • Saying zero correlation means the variables are independent.

    Correlation is often taught as a general measure of dependence.

    Fix: Remember it captures linear dependence only. Use the Y = X² example. Zero correlation implies independence only for jointly normal variables.

  • Treating a high correlation between two trending price series as evidence of a real relationship.

    Shared trends make levels move together even when changes are unrelated.

    Fix: Think spurious correlation. Check whether the series are nonstationary and use returns or changes instead of levels.

  • Forgetting to square d in the Spearman formula, or using the wrong n.

    Rushing under time pressure.

    Fix: Write d, d² and the sum in a row. Use n as the number of pairs. Check the answer lies between -1 and +1.

  • Using the wrong denominator for Kendall's tau.

    Students divide by n or by n² instead of the number of pairs.

    Fix: The denominator is n(n − 1) ÷ 2. For n = 5 it is 10.

  • Assuming rank correlations detect any dependence.

    They are described as more robust, so students overstate them.

    Fix: They capture monotonic dependence. A U-shaped relationship can still give a value near zero.

  • Assuming correlation is stable and that one estimate applies in all market conditions.

    Correlation is quoted as a single number in portfolio formulas.

    Fix: State that correlation can change over time and often rises in stress, so diversification benefits may shrink when they are needed.

Worked examples

Example 1

Five assets are ranked by two analysts. Sorted by analyst X (ranks 1 to 5), the Y ranks are 2, 1, 4, 3, 5. Compute Spearman's rho and Kendall's tau.

Show the solution
  1. Spearman. Differences d = X rank − Y rank: 1−2 = −1, 2−1 = 1, 3−4 = −1, 4−3 = 1, 5−5 = 0.
  2. d² = 1, 1, 1, 1, 0. Σd² = 4.
  3. ρs = 1 − 6 × 4 ÷ [5 × (25 − 1)] = 1 − 24 ÷ 120 = 1 − 0.2 = 0.8.
  4. Kendall. There are 5 × 4 ÷ 2 = 10 pairs. For Y order 2, 1, 4, 3, 5, count later values smaller than each: 2 has one (the 1), 1 has none, 4 has one (the 3), 3 has none, 5 has none. So nd = 2.
  5. nc = 10 − 2 = 8.
  6. τ = (8 − 2) ÷ 10 = 0.6. Check with the shortcut: 1 − 2 × 2 ÷ 10 = 0.6.

Answer: Spearman's rho = 0.8 and Kendall's tau = 0.6.

Example 2

X takes the values −2, −1, 0, 1, 2 with equal probability, and Y = X². Compute the correlation of X and Y and explain what it shows.

Show the solution
  1. Mean of X = (−2 − 1 + 0 + 1 + 2) ÷ 5 = 0.
  2. Y values are 4, 1, 0, 1, 4. Mean of Y = 10 ÷ 5 = 2.
  3. Cov(X,Y) = average of (x − 0)(y − 2) = [(−2)(2) + (−1)(−1) + 0 × (−2) + (1)(−1) + (2)(2)] ÷ 5 = (−4 + 1 + 0 − 1 + 4) ÷ 5 = 0.
  4. Since the covariance is 0 and the standard deviations are positive, ρ = 0.
  5. Yet Y is completely determined by X.

Answer: The correlation is 0, even though Y depends perfectly on X. Zero linear correlation does not imply independence. Pearson correlation misses nonlinear dependence.

Exam tips

  • Expect conceptual MCQs that ask which statement about correlation is true. Eliminate any option claiming zero correlation proves independence or correlation proves causation.
  • Know the Spearman and Kendall formulas cold, and practise with 5 or 6 pairs so you finish in under two minutes.
  • When a question mentions trending price levels, think spurious correlation and the fix of using returns or differences.
  • If a question mentions outliers or a nonlinear but monotonic link, rank correlation is usually the intended answer.
  • For crisis or stress wording, the point is that correlations are unstable and tend to rise, which weakens diversification.

Practice questions from Measuring Return, Volatility, and Correlation

Correlation Pitfalls and Dependence Measures in other exams

The same ground in other exams, if you are preparing for more than one or want another angle on it.

Correlation Pitfalls and Dependence Measures: frequently asked questions

What are the main limitations of linear correlation in finance?

It captures only linear dependence, so it can miss nonlinear links. It is sensitive to outliers and can be spurious for trending series. It is also unstable over time and often rises in market stress.

What is the difference between Spearman and Pearson correlation?

Pearson uses the actual values and measures linear dependence. Spearman is the Pearson correlation of the ranks, so it measures monotonic dependence and is less affected by outliers. It also does not change under increasing transformations such as logs.

How is Kendall's tau calculated?

Count concordant pairs (both variables move the same way) and discordant pairs (they move opposite ways). Then compute (nc − nd) ÷ [n(n − 1) ÷ 2]. With no ties, this equals 1 − 2nd ÷ the total number of pairs.

What is a spurious correlation example?

Two unrelated series that both trend upward over time, such as the price levels of two unrelated assets, can show a high correlation. The link comes from the shared trend or chance, not from a real relationship. Using returns or changes usually removes it.