Putting prior on the constant separately in a regression

Yes, I noticed that. I switched and put prior on std instead. I settled on Half Cauchy. The model does well now, though it still has a good deal of autocorrelation, even with QR. I really wonder where it might come from.