Backtesting and optimization get you started in the right direction. However, given changing market dynamics over time, todays high performance EA may fail to perform at some point in the future. We all know this.

My preference is to backtest and optimize over a 1 year period, then test this on previous random years up to 8-9 years back to get a feel for overall performance. If I like the results, I will then zero-in to more recent trading history.

For example, I use MAs on the Daily USD/JPY. I take the slower MA and multiply this by 6 (7-day MA = a 6-week optimization period, for example). Then, at the end of each trading week I re-optimize by dropping the first week, and adding the week just completed. It is effectively a moving average applied to optimization.