Correlation Has Personality

Correlations Have Personality, Too:

An Analysis of Correlations between Assets

Carlton J. Chin, CFA

Price Asset Management

Email: carat@alum.mit.edu

February 28, 2013

Abstract

The statistical measure “correlation” is one of the building blocks of creating

a well-diversified portfolio. But what do we know about correlation? Many

investment sources will compute the correlation of monthly returns between

two asset classes. Others may use daily or other time-frames, while some do

not even list the time-period used in their calculations. Does the time-period

matter? We study correlation in a variety of ways to obtain a more intuitive

feel – and a more meaningful understanding – of this elusive measure. In

addition to examining the time-period parameter, we review the long-term

cyclical nature of correlation between asset classes. The results show us the

fickle behavior of the correlation statistic and suggest that an active approach

to how we measure and apply correlation can improve portfolio optimization.

Results are based on more than fifty years’ worth of data, across several asset

classes, including stocks, bonds, and commodities.

PAST RESULTS ARE NOT NECESSARILY INDICATIVE OF FUTURE RESULTS.

1. Introduction

Investment practitioners constantly strive to improve expected return given

a certain level of risk. Some investment professionals focus on specific asset

classes or investment strategies. Others state that a diversified portfolio’s results

are driven mainly by the asset allocation policy (Brinson, Hood, and Beebower

1986).

Portfolio construction decisions and asset allocation policies – are driven by

risk, return, and correlation assumptions. In this research paper, we focus on the

statistical measure of “correlation.”

Correlation is one of the building blocks used to create a well-diversified

portfolio. The correlation between two asset classes is typically based on monthly

performance data. However, monthly...