This is a summary of “Predicting Bond Returns: 70 Years of International Evidence” by Guido Baltussen, Martin Martens, and Olaf Penninga, published in the Third Quarter 2021 issue of the <i>Financial Analysts Journal</i>.
Listen to an audio version of this summary.
This study uses a testing framework based on financial trading strategies on data spanning 70 years in six countries and finds consistent and robust evidence of bond return predictability.
What Is the Investment Issue?
Government bonds represent approximately 30% of overall market capitalizations across asset classes. Can the excess returns of government bonds be predicted?
Much of the existing research into bond return predictability uses limited data samples, usually ones that cover only a single country or the post-1980 period. In this study, the authors set out to examine a more extensive historical sample. They also develop a testing framework based on financial trading strategies rather than commonly used predictive regressions.
How Do the Authors Tackle the Issue?
The authors consider 10-year government bonds across six major markets: the United States, the United Kingdom, Germany, Japan, Canada, and Australia. Their prediction sample covers 70 years of data, from 1949 to 2019, totaling 7,497 monthly return observations.
To examine bond return predictability, they test four predictor variables: yield spread, bond trend, equity returns, and commodities returns. Each variable is transformed into a real-time trading signal.
The authors also assess the four variables’ joint forecasting power and use an equal-weighted average across the countries to calculate a “Global” strategy performance.
The trading signals are updated monthly. Bond return predictability is determined by evaluating the signals’ performance. The analysis is divided into two periods. One is an in-sample period from October 1981 to May 2019—three decades characterized by mostly falling yields. The other is an out-of-sample period from January 1950 to September 1981—a time of mostly rising yields. The authors note that this earlier out-of-sample period has not been studied in previous research.
They test their findings for robustness by assessing subperiods within both the in-sample and out-of-sample periods, variations in testing choices, and government bond markets in nine additional countries.
Finally, the authors ask why predictability in government bond markets might arise. They explore three classes of explanations: risk-based explanations, market frictions, and market inefficiency.
What Are the Findings?
The authors find “consistent and ubiquitous evidence” of bond return predictability.
For example, the Global strategy during the 1981–2019 in-sample period produces a Sharpe ratio of 0.73 and an annual excess return of 1.5%. Each of the four predictor variables demonstrates this predictability.
The 1950–81 period shows similar results, with a combined Sharpe ratio of 1.09 and an annual excess return of 1.8% for the Global strategy. This finding—that bond returns are also predictable in the earlier decades—provides no evidence of out-of-sample decay. It suggests that the in-sample findings do not result from statistical errors such as false positives and p-hacking.
A breakdown of the results by government bond market shows positive Sharpe ratios in all six countries for the four variables, both when the variables are considered separately and when they are combined. An equally weighted average across countries produces a higher Sharpe ratio than in any individual country, indicating that bond market predictability strategies benefit from diversification.
The robustness tests show that bond predictability is consistent over time periods and market episodes, including during extended periods when rates are rising or falling. Moreover, results from nine additional countries are consistent with those of the six countries in the main sample. These findings lead the authors to conclude that “bond return predictability is a very persistent empirical phenomenon in markets, not driven by statistical type I errors, data mining or p-hacking effects.”
The authors conclude that predictability in bond markets is unlikely to be a consequence of market or macroeconomic risks. Nor is it meaningfully explained by investment frictions such as transaction costs and short-selling constraints. The authors posit that instead, the most likely reason for predictability is market inefficiency.
What Are the Implications for Investors and Investment Managers?
The authors conduct asset allocation tests and find that a bond market timing strategy can substantially raise the Sharpe ratio of a traditional equity–bond portfolio. Moreover, they find that the predictability of bonds still holds after accounting for realistic transaction costs. They therefore conclude that the timing of international bond market returns presents an opportunity for investors. Active management of government bonds could add value by predicting the direction of yield changes.