The Emperor’s New Clothes—Balanced Portfolio Construction

The aggregate portfolios used by end-clients (e.g., pension, endowment, or retired couple) are most often broadly diversified across stocks, bonds, and alternatives. Such portfolios are constructed and maintained using a widely accepted portfolio construction approach. This approach is frequently shorthanded as "modern portfolio theory" or MPT. A simplified version of this approach assumes random variable distributions (to represent the future returns for each asset category) and associated defining summary statistics (means, standard deviations, and correlations). This is the approach upon which all retail investment software tools are based and underlies most institutional asset allocation and asset/liability modeling efforts. For this approach to be successful, the statistical behavior of the underlying asset categories must satisfy certain behavioral properties. Specifically, (a) means must be sufficiently stable, (b) return distributions must not suffer from regime change, (c) distributions must not deliver risk without return over excessively long time periods, (d) the implementor cannot ex-post cherry pick the best geography and most attractive time period and then base their results on that after-the-fact handpicked-selection, (e) return distributions must not be defined by extreme episodic eras, and (f) excessive levels of trending and momentum must not be present. The objective of this article is to provide meaningful statistical evidence that each of these six requirements is violated. If true, then our industry's generally accepted portfolio construction process is suspect. The article ends with three suggestions for portfolio construction techniques that are generally immune to the issues raised and suggests why forward-looking capital market assumptions are unable to overcome the challenge.