yield gap of -3.16 percent. Therefore, the strategy would have resulted in a 15.8 percent short stock position and 15.8 percent long bond position at the beginning of January 2000. As shown in Figure 25.1, the success of this simple strategy is striking. The information ratio on the earnings-yield-gap timing model is 0.43, which yields a highly significant probability value of 0.01 percent due to its extremely long history. In addition, the strategy is correctly positioned coming out of the bear market of 1973-1974, and is substantially underweight during the crash of 1987. The earnings yield gap is negative, though, for the four-year period starting at the beginning of 1996, thus missing out on one of the greatest continuous market appreciations on record. We believe this is one reason why many TAA managers struggled in the late 1990s. One important facet of any forecasting model is robustness. By robust, we mean that the same factor or investment theme forecasts asset returns both across and within countries, across various asset classes, and across different time periods. Robust forecasting variables are less subject to data-mining biases and therefore are more likely to perform well out of sample. The more intuitive and consistent the factor, the more likely future performance will be strong. As shown earlier, the valuation investment theme is a very robust strategy. Predictability Using Momentum Measures To further support the case of predictability, let us turn to the evidence on predictability using momentum. Momentum means using measures of short-term performance to predict future performance. Recent past returns should forecast 4.0 t-------------------------------------------------------------------------------------------------------------------------------------- , 1925 1929 1933 1937 1941 1945 1949 1953 1957 1961 1965 1969 1973 1977 1981 1985 1989 1993 1997 2001 FIGURE 25.1 Cumulative Excess Return on a U.S. Stock/Bond Timing Strategy