Why does EAs need to optimize?

<p id="isPasted">It depends...</p><p>First of all, try to make your EA less sensitive to changing market conditions by relating thresholds to ATR, express your SL/TP as a percentage of price rather than absolute number of pips/points, lot sizes as a percentage of current equity, etc. However, that brings new challenges: should I use standard ATR(14) which changes surprisingly a lot (you try it on a chart!), or use an ATR(252) as the average movement throughout the last year but slow to react to changing conditions.</p><p>To find a good lookback period for your EA, you could do Walk Forward Optimizations with various lookback periods and see what works best for you. I have been told some people use a moving window of the last 50 trades as a lookback period, others only 2 months. Personally, I typically use a moving window of the last 4 years as a lookback period.</p><p>The re-optimization interval should be irrelevant, as the settings should reflect the optimal settings over the current lookback period. In theory you could re-optimize after every tick, or every bar. That would require an unreasonable amount of computing power, and keeps finding the same optimal settings for many iterations before enough old data has shifted out and enough new data has shifted into the moving lookback window to make another setting more optimal.</p><p>Therefore, people usually re-optimize less frequent. I know people who do this every weekend, others do it once a month, but I also know someone who does this only when performance starts to deteriorate (which is typically after a year or so with his strategy). Preferably, if you can, make the re-optimization built-in your EA and have it re-optimize itself on the fly.</p><p>Alternatively, run a WFO with a ridiculously high re-optimization frequency (e.g. hourly). That will take a lot of time but gives you the benchmark of the best possible performance of your EA. Next, try WFOs with larger re-optimization intervals. Keep increasing the interval until you notice the performance starts to deteriorate. You have now found out what re-optimization interval you should use. It is important to note that if you get wildly different results depending on the re-optimization interval, then your strategy or optimization method is probably not robust and might be curve fitted.</p>