The authors ascertain the practical relevance of the predictability of long-run asset returns. They search for simple yet profitable dynamic asset allocation strategies based on authentic patterns of return in functions of valuation ratios that have superior prediction capability vis-à-vis strategies based on historical returns.
Although the global economy is gradually recovering, returns will continue to be constrained and volatile relative to the long-term bull market that started in the 1980s. As a result, asset allocation decision making is of paramount importance in the global financial industry. The authors demonstrate that historical returns are poor indicators of future returns. The current research provides some simple dynamic asset allocation methodologies based on a predictable pattern of returns.
How Is This Research Useful to Practitioners?
Historical returns are commonly used in portfolio decision making despite their irrelevance. Previous researchers have shown that forward-looking models that use market multiples for equities and initial yields to maturity for bonds are better predictors of long-term asset returns than the traditional approach used in the industry, in which trends are extrapolated from historical average returns. These forward-looking models better embed the concepts of mean reversion and structural changes in the return pattern than models based on historical returns.
Dynamic asset allocation strategies based on these models also deliver better risk-adjusted return compared with the conventional models. The expected return assumption used in portfolio decision making must be reliable for the optimal asset class mix to be accurate. Hence, asset allocators should abandon the static assumptions approach that uses historical average returns. Rather, they should routinely rebalance the portfolio toward an optimal asset class mix based on the relationship between long-term return and market fundamentals. The research is essential for institutional investors and portfolio managers.
How Did the Authors Conduct This Research?
The dataset is composed of US equity and bond returns from 1926 to 2010 culled from various sources. The authors form five models—namely, the Gordon dividend discount model, the inverted price-to-earnings ratio, the sum of parts, the return on equity, and the book-to-market ratio—for forecasting equity returns. To assess the difference, the authors benchmark these models against four historical models: the 5-year historical return, a variation of the dividend model, the 10-year future realized return, and the 1-year future return. The mean-squared error is also computed.
A bond return forecasting model is also formed based on yield to maturity. All of these models predict future realized returns better than the historical average return. Then, the authors apply these forecasts to the four US asset class portfolios. They assume a nine-portfolio matrix composed of three different levels of risk aversion coefficient and three different shortfall constraints. Then, for these portfolios’ Sharpe ratios, the difference between the historical return and the p-value of the t-test is estimated. With the aim of maximizing the satisfaction on the utility curve as well as considering shortfall constraints, the authors find that the portfolio returns based on forward-looking models generate better risk-adjusted returns compared with those of the portfolios that use historical returns.
In this era of global financial instability and uncertainty, asset allocation is of critical importance for investors—particularly for those who cannot take a long-term approach and those looking to enhance returns. Research findings reveal that institutional investors should adopt a dynamic asset allocation strategy. An important implication of the research is that steady-state return models deliver enhanced Sharpe ratios and reduce the need for a larger amount of rebalancing to arrive at the optimal portfolio mix compared with the results of those based on historical returns. Further research in this area on constraints (i.e., the impact of different volatilities and correlations) would be helpful.