No, you don't. Just try it. No matter which time period you use and how deep you nest MA functions, they are always either less smooth or far more laggy than equivalent lowpass filters. That's why the 9 lines script that I posted would not work with an MA instead of a lowpass filter. Just replace LowPass(...) with SMA(...) or EMA(...) and you'll see what I mean.
That's a website by our Gamestudio group. We're attempting to develop trade systems that are more robust and based on better mathematics than the usual systems. As we see this as research and not as a commercial operation, all systems and tools are free to anyone who's interested.
This is the equity curve of the 9-lines lowpass filter system, using EUR/USD with the current spread, rollover, and slippage of FXCM: The is the equity curve of the same system with an EMA instead: At a first glance the EMA system seems to be also profitable, but that is deceiving. It has less profitable trades - the green lines - and they are longer, meaning more risky. When you change its time period, the performance is all over the place. The lowpass filter system on the other hand is far less sensible to time period changes. The drawdown of the EMA system is far higher. It has only half the profit and a Sharpe ratio of 0.54 - that means it is absolutely not tradable. The filter system is also not tradable in this simple form of course, but you can see that it's more promising than the EMA system.
If the green/red points represent entries/exits, then in either case it looks like fewer than 100 trades???? Perhaps fewer than 50???? If so, potentially dangerous based on these two samples to draw too many conclusions about which is "better" IMHO ....
It's 332 trades. It's a little hard to see in the plot because the scale is very small. But we have also run tests with several 1000 trades and filters turned out to be better than MAs. Which is the whole point of this thread.
I think you are the one who misundestands it. You are always trading based on past market conditions. You cannot know future market conditions but only after they happen. If this is not clear to you you should know it. If you think that walk-forward testing eliminates parameters and curve-fitting you are very much mistaken. It just breaks down a global curve-fit to a series of local curve-fits. Your statement is quite deceiving intentionally or not.
I admit that I also don't really get the meaning of what you want to say. Do you think that automated strategies can not work? Or that they can not be tested? Or that they can not be walk forward tested? If so, how would you test them instead?