Ideally, the set of optimal parameters and decisions for the first part of the backtest period is also the optimal set for the second period, but things are rarely this perfect. The performance on the second part of the data should at least be reasonable. Otherwise, the model has data-snooping...