Weird results from my LSTM prediction

Hey everyone,
I am using deeplearning4j for stock predictions. I train my results and then predict both on the train data and the test data. When i print out the results of the output() method, it seems to work normally at first, but then my problem occurs.

    Real Train Prices: [[[  750.0000,  750.3100,  748.4000,  ... 1352.6200, 1356.0400, 1349.5900]]]
Predicted Train Prices: [[[ 1043.1039,  979.5354,  944.9826,  ... 1167.1135, 1167.1205, 1167.1156]]]

Here the price fluctuation starts off normally, but as it gets to the later time steps, the fluctuation becomes almost non existant. Likewise for the test data, it does this the whole way through. Here is the result for the test data:

    Real Test Prices: [[[ 1348.8400, 1343.5601, 1360.4000, 1351.8900, 1336.1400, 1337.0200]]]
Predicted Test Prices [[[ 1167.0906, 1167.1364, 1167.1282, 1167.1149, 1167.1169, 1167.1235,]]]

The predicted prices do barely move at all. Now i know i am predicting stocks and it will be fairly inaccurate, but that is not my problem. Does this issue point to anything in particular? After training my model and evaluating on my test data, i get pretty good results, so i am confused. But regardless of what my eval stats look like, i get the same problem each time. Here is what this attempt’s evaluation results give:

INFO: Column    MSE            MAE            RMSE           RSE            PC             R^2            
col_0     3.58377e+00    1.68689e+00    1.89308e+00    4.15568e+00    -6.46589e-03   -3.15568e+00   

Thank you all for the help.