# Table 3 In-sample predictability of US GVD for stock returns

Panel A: Predictability of US stock market excess returns
h=1h=3h=6h=12
1980$$\hat {\beta }$$R2(%)$$\hat {\beta }$$R2(%)$$\hat {\beta }$$R2(%)$$\hat {\beta }$$R2(%)
0.42**0.940.52***4.150.43***5.130.35**6.47
(p value)(0.02) (0.00) (0.01) (0.02)
19900.41**1.000.46**3.490.40***4.630.35**6.41
(p value)(0.04) (0.01) (0.02) (0.04)
19990.310.520.37*2.020.36*3.200.375.84
(p value)(0.16) (0.08) (0.08) (0.12)
Panel B: Predictability of Swiss stock market excess returns
$$\hat {\beta }$$R2(%)$$\hat {\beta }$$R2(%)$$\hat {\beta }$$R2(%)$$\hat {\beta }$$R2(%)
1980– 0.080.030.010.000.040.040.070.20
(p value)(0.62) (0.47) (0.25) (0.35)
1990– 0.080.03– 0.020.000.060.070.130.66
(p value)(0.57) (0.50) (0.37) (0.28)
19990.010.000.040.020.110.330.201.81
(p value)(0.49) (0.43) (0.33) (0.23)
1. Notes: This table presents OLS estimates from univariate regressions of h-month ahead US (panel A) or Swiss (panel B) stock market returns on the US version of GVD over different sample periods that all end in December 2017. The starting points of the different sample periods are January 1980 (row “1980”), January 1990, and January 1999. The US GVD is z-standardized. We compute heteroskedasticity and autocorrelation robust p values (in parentheses below the estimates) from a wild bootstrap procedure that tests the null hypothesis of $$\hat {\beta }^{h}=0$$ against the alternative that $$\hat {\beta }^{h}>0$$ because high values of the US GVD predict high excess returns. *, **, and *** indicate significance at the 10%, 5%, and 1% level, respectively