When Michaud Optimization Fails

The Markowitz and Usmen (MU) (2003) simulation study reported Michaud (1998) meanvariance (MV) portfolio optimization superior to Markowitz (1952, 1959) out-of-sample on average in each of thirty cases examined. However, a simplified replication of the MU test found thirty percent failures of Michaud relative to Markowitz. Instances of Michaud failures were associated with asset risk and return characteristics inconsistent with diversified portfolio risk management. Risk-return properties in professional asset allocations and large universe portfolio optimizations may often be similarly perverse. The simulation framework in Michaud (1998) can be a valuable diagnostic for risk-return estimate diversification perversity when appropriately applied. Our results underscore the necessity of investment sense oversight and validation for successful application of quantitative methods.