Choosing Pearson or Spearman Correlation for Factor Screening

Summary

This note compares Pearson and Spearman correlation for assessing whether proposed economic factors relate to portfolio assets. Pearson measures linear association, while Spearman measures monotonic association based on ranks. These measures can differ because they capture different forms of dependence; Spearman is not a general detector of every nonlinear relationship. The discussion frames correlation as a preliminary way to assess candidate factors before estimating regression betas.

A response recommends Pearson correlation matrices as a common first step when the intended model is linear regression, particularly to screen for endogenous or multicollinear variables among the candidate inputs. The material does not give a universal rule for selecting one statistic, nor does it provide diagnostic results for the portfolio in question. Correlation alone does not establish that a factor is relevant or determine regression specification; the appropriate measure depends on the relationship being modeled and the purpose of the analysis.

Key ideas

• Pearson correlation measures linear association between variables.
• Spearman correlation measures monotonic association using ranks.
• The two measures can produce different results because they describe different relationships.
• Pearson correlation matrices are commonly used as an initial screen for multicollinearity in linear regression.
• Correlation screening alone does not establish factor relevance or determine a full regression model.

Tags

# Choice of factor model based on correlation


# Choice of factor model based on correlation












I have a portfolio of assets. Each assets have been discretionally (based on investment manager experience) related to economic factors like (like exchange rate inflation spread etc). Now for each asset I want to see if effectively those fator are relevant in order to make a regression and find the betas (coefficient of regression). My question is: what correlation should I use? Pearson which is only a linear based approach or Spearman which can also deal (I am right) with non linear relations and outliers? I did both but clearly I get different results so I need to choose among them. Thanks.Luigi

## Answer by John (score 1)

https://quant.stackexchange.com/a/55570

Indeed Pearson correlation coefficients measure only linear relationships. Spearman correlation coefficients measure only monotonic relationships

There is a comprehensive article there.

## Answer by Chris (score 0)

https://quant.stackexchange.com/a/55644

Assuming you're going to be fitting a linear regression, creating a correlation (Pearson) matrix of all assets is a common first step to filter endogenous (or multicollinear) variables from your test set.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.