• Abstract: No, and often it helps !

• Paper: The paper

• Forecasting annual U.S. equity premium: Sixteen predictors described by Welch and Goyal (2008).

  

  The predictive R square as the number of predictors p increases. The original dataset contains p0=16 predictors. So all the remaining p-16 predictors are artificially added standard normal noises. The predictive R square stabilizes at around 10 percent after 300 ~ 6,800 noises have been artificially added. The predictor is simply pseudo-OLS:

  x'_new(X'X)⁺X'Y

  where the pseudo inverse is used to replace the ordinary inverse in OLS.