CHAPTER 14
An Empirical Study of
Time-Varying Return
Correlations and the
Efficient Set of Portfolios
Thadavillil Jithendranathan
14.1 INTRODUCTION
Modern portfolio theory was first introduced in 1952 (Markowitz, 1952), and
since then it has been the mainstay of asset allocation models. In the mean-
variance paradigm of Markowitz, an efficient set of portfolios is estimated by
maximizing the expected return of the portfolio and minimizing its risk, as
measured by the standard deviation. For practical purposes, efficient port-
folio construction requires estimation of expected returns and variances of
expected returns of individual assets in the portfolio, as well as the covari-
ance matrix of the asset returns. The most widely used method of estimating
these inputs into a portfolio model is to use the past return data for a period
of five years and use the historic average values of returns, variances and
co-variances as proxies for expected values. One of the implicit assumptions
in this method of efficient portfolio construction is that the variances and
co-variances are time-invariant during the holding period of the portfolio
(Jobson and Korkie, 1981).
Despite its theoretical appeal, practitioners are generally cautious in
applying mean-variance optimization models in practice. As pointed out
by Michaud (1989), the optimization tends to give higher weights for secu-
rities with large expected returns, low variances and negative correlations
265