Risk-Based Asset Allocation: A New Answer to anOld Question?

WAI LEE is the director of research and CIO of the Quantitative Investment Group at Neuberger Berman in New York, NY. [email protected] The global financial crisis in 2008 caused investors to question what went wrong with many of their portfolios, which were believed to be diversified. Mean-variance optimization (MVO), 60/40, modern portfolio theory (MPT), and others seem to have been put on trial by practitioners and critics alike for their apparent underdiversif ication and accused failure to provide risk control. A list of “new paradigms” or “next generation solutions” has been declared to displace MPT. A growing amount of literature on portfolio construction approaches focused on risks and diversification rather than on estimating expected returns, collectively called risk-based asset allocation in this study, has been documented. On the topic of strategic asset allocation, we have been seeing more writings on the various versions of risk-based approaches applied to a global universe of assets, especially in cases of pension and endowment management. Allen [2010] and Foresti and Rush [2010] provide good examples. A common finding among these studies is the superior risk-adjusted return of a portfolio that is constructed in such a way that assets are expected to contribute equal risk to the whole portfolio—an approach commonly labeled risk parity. In a risk parity approach, only risk forecasts are used as inputs, and no forecasts of returns of any assets are required. In recent years, we have also witnessed a growing literature documenting, in particular, that some equity portfolios—naively diversified portfolios (e.g., equally weighted portfolios), portfolios that are constructed to achieve the minimum volatility possible given the universe of risky assets (e.g., stocks), in order to achieve “maximum diversification” subject to the definition of diversif ication, or portfolios that are dubbed as risk parity— outperformed on a risk-adjusted basis both the market capitalization–weighted portfolio and portfolios that ex ante are constructed to be mean-variance optimal as derived by application of the Markowitz optimization. In some studies, these portfolios even outperformed the market portfolio on an absolute return basis. Examples include Clarke, de Silva, and Thorley [2006]; DeMiguel, Garlappi, and Uppal [2009]; Behr, Güttler, and Miebs [2008]; Martellini [2008]; and Choueifaty and Coignard [2008]. We have also seen the parallel development in the industry of the growth of product offerings and client interest in investment vehicles built upon these findings, especially in equities ( Johnson [2008]). One common characteristic across all of these portfolios is that the only input required to determine the portfolio compositions is a model of risk, which is typically measured by the covariance matrix, while explicit modeling of expected returns is not required. Some of these studies argue