### Long-run Test of a Tactical, Tractable MPT

**January 15, 2015** - Momentum Investing, Strategic Allocation

Does a cross-asset class, momentum-driven, simplified version of Modern Portfolio Theory (MPT) offer reliably strong performance over the long run? In their December 2014 paper entitled “A Century of Generalized Momentum; From Flexible Asset Allocations (FAA) to Elastic Asset Allocation (EAA)”, Wouter Keller and Adam Butler present an asset allocation strategy based on five concepts:

- MPT is a sound framework for portfolio construction.
- Momentum, a form of trend measurement, is a generally effective way to estimate key inputs to MPT: asset returns (R), return volatilities (V) and return correlations (C).
- Crash protection based on excluding assets with negative past returns is a reasonable corollary of reliance on trends.
- Tractability requires compromise to strict MPT, such as calculating return correlations relative to a single index (the equally weighted average returns of all assets).
- Recognition of differences in import among inputs means weighting R, V and C inputs differently according to their elasticities (how much small changes in R, V and C affect the optimal portfolio weight for the asset).

The fifth concept is the innovation relative to the Flexible Asset Allocation (FAA) predecessor (see “Asset Allocation Combining Momentum, Volatility, Correlation and Crash Protection”), which weights expected R, V and C inputs based on a simple scoring system. The new Elastic Asset Allocation (EAA) strategy each month scores all assets in a universe by: (1) calculating expected R, V and C for each asset as geometrically weighted averages of past values; and, (2) weighting the expected values of R, V and C by their respective elasticities. For R, they use average total monthly excess (relative to the 13-week U.S. Treasury bill yield) returns over the last 1, 3, 6 and 12 months. For V and C, they use the last 12 monthly returns. To test the EAA strategy, they each month reform a long-only portfolio of the top-ranked assets weighted by their respective scores. They replace a fraction of the portfolio with 10-year U.S. Treasury notes (selected empirically as the best “cash” asset) according to the fraction of assets in the universe with non-positive excess returns. They apply a nominal one-way index switching friction of 0.1%. They consider three universes of 7, 15 and 38 asset classes. They emphasize Calmar ratio (focusing on drawdown) as a key optimization metric, but also consider Sharpe ratio. To mitigate data snooping, they optimize elasticity parameters during April 1914 through March 1964 and test it out-of-sample during April 1964 through August 2014. Using monthly returns for the three sets of financial asset indexes as available during April 1914 through August 2014, *they find that:*