Correlation and diversification

From Chapter 9 of The Analyst's Handbook, by Matt Lang and Bree Lang

Imagine that a fund holds three stocks, and that its committee is pleased to report that it has three separate bets. Whether that is true depends on how the three stocks move in relation to one another, and the number that measures this is called the correlation:

ρ = Cov(r_A, r_B) / (σ_A × σ_B)

Here r_A and r_B are the daily returns of two stocks, Cov is their covariance and the two σ are their standard deviations. The result always falls between −1 and +1. A correlation of +1 means that the two stocks move as one, and a correlation of 0 means that each ignores the other.

Suppose that we compute it for each pair in the fund over two weeks of daily returns. The first and second stocks have a correlation of 0.82, the first and third have −0.09, and the second and third have 0.02. This means that the third stock had essentially nothing to do with the other two, while the first two moved almost as a pair. The fund does own separate bets, but it owns two of them and not three.

In order to see what that costs, we need the formula for the variance of a portfolio of two stocks:

σ²_p = w_A²σ_A² + w_B²σ_B² + 2 w_A w_B ρ σ_A σ_B

Here w_A and w_B are the two weights. The first two terms are what each stock contributes by itself, and the third term is the only place where the correlation enters.

Say that the first stock has an annualized volatility of 13.0% and the second has 9.3%, and that we hold them in equal amounts. If the two were uncorrelated, the third term would be zero and the volatility of the pair would be 8.0%, which is below either stock taken alone and well below their simple average of 11.2%. That reduction is what diversification is. With the measured correlation of 0.82, the third term comes back, and the volatility of the pair is 10.7%.

Take a moment to compare the two answers. The pair we hoped for ran at 8.0%, the pair we have runs at 10.7%, and that is most of the way back to simply averaging the two. In other words, adding a second stock that moves with the first bought very little, and a fund can hold twenty stocks and still own only a handful of bets.

A correlation is also measured over a window that has already happened, so a high number is a reason to expect two stocks to stay related and not a promise that they will. Chapter 9 of the textbook builds one correlation by hand.

Read Chapter 9: Correlation, diversification, and factor tilts

Related topics: Performance attribution · Value at risk · All topics

The Analyst's Handbook is Sifter's free textbook for student investors, written by Matt Lang and Bree Lang. Sifter is free for students. This page as markdown.