Generally a negative R-sq value means that the model predictions are worse than simply using the mean but I suspect that there is an issue in the code itself. If you could provide a code showing how you calculated it I can help more perhaps.