Predictability of bond markets
Johan Duyvesteyn, PhD, CFA, Martin Martens, PhD
Some of the most influential scientific papers on the predictability of bond markets connect theory with the tested predictive power of the variables of Robeco’s duration model. A small sample of academic evidence on the predictability of fixed income markets is discussed in this paper and its link to the model is illustrated. The sample, which we selected from many articles, consists of three key academic publications that have strong links to three of the model’s variables. The findings of the publications are discussed in brief and compared with Robeco’s duration model.
The duration model
Robeco’s duration model has been in use to predict fixed income markets for more than 12.5 years. The model has proven its predictive power over time and has an important role in the fixed income investment process of Robeco. The model consists of six indicators which have an intuitive relationship1 with fixed income markets. When Robeco developed the duration model in 1994, existing evidence of predictability in interest rate markets was taken into account. Since then, Robeco’s quant researchers have closely followed new academic publications in the area of predictability to improve the model.
The academic evidence
We briefly discuss the findings of academic publications on interest rate predictability. We selected three key articles by -1- Eugene Fama and Robert Bliss (1987), -2- Antti Ilmanen (1995) and -3- Eugene Fama (2006), covering two decades of academic research.
Fama and Bliss (1987) find a mean reversion in 1- to 5-year interest rates for the period from 1953 to 1986 in the US. The forecast power even increases when the forecast horizon is extended. Fama and Bliss attribute the forecast power to interest rates reverting back to a long-term mean, also known as mean reversion. The intuition behind the factor is that after a sharp rise of the interest rate, it will be followed by a decrease back to the longer term average. Mean reversion is one of the six variables in Robeco’s duration model.
Ilmanen (1995) finds that a set of four global instruments is able to predict the returns of long maturity bonds for six different countries. The four factors are the stock market return (called “inverse relative wealth” by Ilmanen), time varying risk as measured by the bond beta, and two expected bond risk premium factors as measured by the yield spread and the real yield respectively. The factor with the highest predictive power, the stock market return, is a second variable in Robeco’s duration model. Our intuition is that a good performance of the stock market means that investors are expecting higher corporate earnings. This suggests a positive economic climate and indicates inflationary pressure, and therefore the potential for higher 1 In the research paper ‘Robeco's duration model, 2006’ the model’s variables, their background an the predictive power are extensively discussed interest rates. Thus, when the stock market increases, a negative return for interest rates investments is expected. The bond risk premium is a third variable incorporated in Robeco’s duration model. The risk premium is measured by the slope of the yield curve. A steeper slope implies a larger risk premium and a larger expected return. The bond beta does not show significant forecasting power in Ilmanen’s research and is not incorporated in the duration model.
Fama (in press) updates the prediction model that was developed in Fama and Bliss (1987). Fama (in press) evaluates the original model using additional data from 1986 to 2004 and observes that the predictive power of the model is higher in the new period. Fama also shows more insights in the mean reversion property of the strategy. The mean reversion is not caused by the general increase of US interest rates in the period from 1953 to 1985 followed by the large decrease in the period from 1986 to 2004. On the contrary, the explanation for the predictability is mean reversion in yield differentials between countries which is caused by business cycle variation. For example, the US business cycle is known to lead the business cycle of the EMU. When the US economy contracts and therefore the yield decreases, it will be followed by a decrease of yields in the EMU. As mentioned before, mean reversion is one of the six factors in Robeco’s duration model . This factor is used in the duration model only for the country allocation decision, not for the overall duration timing, consistent with the local nature of mean reversion Fama describes.
Robeco’s duration model was developed in 1994 and has shown a strong track record since its inception 12.5 years ago. The model is based on six variables that have shown predictive power for the fixed income markets, are based on an intuitive relationship to fixed income markets and are backed by academic literature. The academic literature presented here is a sample of influential academic evidence on the predictability of fixed income markets that has clear links to several of Robeco’s duration model’s variables.