Assessing the ability of dividend yields to predict stock returns using out-of-sample forecast errors

In Introduction to Econometrics, 3rd Edition, by Stock and Watson, there is a short example about evaluating the ability of using dividend yields (current dividends over price) to predict future stock returns. The regression featured uses monthly data from 1960:1 to 2002:12 on the logarithm of the dividend-price ratio for the CRSP value-weighted index of stock returns. (See pp. 564-565.)

They present several regressions. The main one of interest is a ADL(1) regression (auto-regressive distributed lag regression) $$\text{excess return}_{t} = \rho_0 + \rho_1 \text{excess return}_{t-1} + \gamma_1 \ln(\text{dividend yield}_{t-1}) + \epsilon_t.$$

The mention

One way to evaluate the apparent predictability found in column (3) of Table 14.7 is to conduct a pseudo-out-of-sample forecasting analysis. Doing so over the out-of-sample period 1993:1-2002:12 provides a sample root mean square forecast error (RMSFE) of 4.08%. In contrast, the sample RMSFE of a "constant forecast" (in which the recursively estimated forecasting model includes only an intercept) is 3.98%. The pseudo out-of-sample forecast based on the ADL(1,1) model with the log dividend yield does worse than forecasts in which there are no predictors!

The regression results are given in the following table:

Question: Does anyone have a reference (a paper or something else) that calculates this same average out-of-sample forecast error on more recent data? Ideally, this would include data up until 5 years ago or so.