The active use of predictive information has the potential to enhance returns relative to those obtainable from fixed-weight “buy and hold” strategies. The authors use a conditional asset pricing framework to empirically examine portfolio management strategies that dynamically leverage predictive information while maintaining mean–variance efficiency from the perspective of an outside observer.
An active portfolio strategy that makes efficient use of conditional (predictive) information can appear inefficient to an outside observer—for example, an investor—who is attempting to evaluate performance unconditionally (without the benefit of that information) after the fact.
Hansen and Richard (1987) and Ferson and Siegel (2001) addressed this issue theoretically, demonstrating that it is possible to construct unconditionally efficient portfolios using conditional information. The authors enhance the relevance of this line of research for practitioners by empirically examining the efficiencies and economic benefits that can accrue from the use of these strategies.
How Is This Research Useful to Practitioners?
The authors empirically examine the in-sample (1960–2004) after-the-fact performance and economic benefits of three dynamically efficient portfolio strategies: maximum return (target volatility mean = 15%), minimum variance (target return mean = 15%), and maximum utility. They find that in the case of multiple assets using the Fama–French five-industry (FF5I) base portfolios, dynamic strategies effectively use predictive information to “decouple” from the market index and greatly outperform (reported betas approaching 0.30 and alphas higher than 9%).
In addition, they find that using term spread, credit spread, and inflation as predictive variables—both individually and collectively—earns excess returns that can support annual management premiums on the dynamic strategies (versus fixed-weight strategies) ranging from 2.5% to 6%. After transaction costs, strategies based on market index return, dividend yield, and inflation do not add to excess returns. The predictive power of factors varies by size and industry, with greater predictability in smaller-capitalization stocks. There is also a concern that return predictability has declined since 2004.
The authors also empirically examine out-of-sample performance and find that in the 1995–2004 subsample experiments, multiple-asset FF5I strategies based on term spread and credit spread outperform the fixed-weight strategies and strategies based on market index return by 6–12% for the 2000–04 (dot-com bubble collapse) period. A dynamically optimal maximum-return strategy that uses term spread and convexity as predictive variables avoids all losses and achieves a gain of about 40% during that period.
In the 2005–07 experiments, the authors report that in the single-asset case, the dynamically managed strategies achieve alphas between 5% (maximum-return strategy) and 7.4% (minimum-variance strategy). In the case of multiple assets using the FF5I base portfolios, all managed strategies underperform relative to the market index while the fixed-weight strategies achieve a respectable Sharpe ratio and a marginally positive alpha.
In the final 2005–June 2009 experiment, none of the strategies avoid losses and most of the dynamically managed strategies underperform relative to the market index in the final 18 months.
How Did the Authors Conduct This Research?
Using the theoretical conditional asset pricing framework established by previous researchers, the authors derive propositions for estimating a dynamically managed portfolio’s predicted efficiency—as measured by the Sharpe ratio—and for constructing dynamic portfolio strategies meant to be deemed efficient by an outside observer after the fact. They also use a utility-based framework to develop a formula to estimate the size of the management fee or premium that an investor is willing to pay for the predictive strategies.
The authors use monthly 1960–2004 data for the empirical tests and derive the formulas for calculating expected means and the variance–covariance matrix of base asset returns generated by using predictive variables. They also use bootstrap simulation to examine the robustness of calculated estimates.
For predictive instruments, they obtain the lagged return on the market index and the dividend yield on the index from CRSP; they construct all other variables using data from the economic database at the Federal Reserve Bank of St. Louis (FRED).
For base assets, the authors use the total return (including reinvested distributions) on the CRSP value-weighted market index for the single-risky-asset case. For multiple risky assets, they use the FF5I portfolios and the Fama–French 2 × 3 portfolios sorted by size and book-to-market ratio.
Prior to examining portfolio performance, the authors make in-sample, empirical estimates of predicted annualized Sharpe ratios and examine the impact of each of the predictive variables.
This research should be of interest to portfolio managers and those charged with evaluating manager performance.
The study of conditional asset pricing has a history of robust, proof-driven theoretical development. And the authors’ results—at least their in-sample results—indicate that dynamic portfolio management strategies based on this framework have the potential to enhance portfolio efficiency and returns.
In the course of my limited survey of the literature, however, I could not help thinking that it might be a challenge for a practitioner to correctly implement—or even validate the implementation of—the complex portfolio-weighting formulas used in this line of research.