I'm running a ridge regression in a rolling window and I tried a subset of window sizes to see which would better fit out of sample. I thought that the greater the training set size, the lower the RMSE in the test set. But what happened was the opposite. Is there any possible explanation?
The subset I used for the window is w = [12,16,20,24,32]. What happens is that the greater the w, the greater the RMSE out of sample.
It may be important to say that I have a very large ammount of features, with size of 300 being much greater than the training set sizes above. Is that a problem for the ridge?
Would someone help me, please?
