Obtaining data on closeloop displaced with respect to the original

Well, I've noticed when it failed. In a matter of clarification I think I get very good results, with 2100 data is about R2 = 97% in both train, validation and test. When I do the loop represents the trends well, but I find that all the checks, the result gives me retarded, ie if the actual data I get a peak at day 5, the outputs predicted by me closeloop but gives the same peak on day 7. Should remove two delays to obtain the same results as the real ones?. The MSE in closeloop is 2.767 e-4, so I think in any case I overfitting occurs. Thank you very much.