I'm developing a trend following system. So far is good, but I need to make better filters for the trades. As any trend following system, the win rate is bellow 50% (way bellow) but Risk to Reward compensate for the losses.
I want to apply simple filters (no more than 3 input variables) in order to improve win% (this also will help get the drawdown "down").
So far I tested the system with the following filters for the trades:
a) Moving Averages compared to actual price
b) Moving Averages crosses
What other indicators or ways can I use to filter the trades? I have developed a system, but I'm asking for ways to filter better the trades that start with the trend.
To improve win%, you have to use more indicators that have to do with scalping in lower timeframes.
Thanks for responding. My worst fear of using to much indicators or to much inputs is the system is not robust.
What do you think about that? My fear is to have a really good system on backtesting but it doesn't work in the future. How you we make sure the system is going to perform similar to past? Do we only have forward testing as the only tool? Any information will be appreciated.
There is no way to ensure the performance in the future is similar to the past unless it is statistically invariant for every type of market structure.
Then there are people who build systems that only win for a specific market structure, rather than all of them.
In forex, you only need to specialise in one and one thing only, and you can make it great.
If you have a good backtest result, it could also mean the trading system is overfitting, and is unreliable to trade future prices.
Forward testing is the only way for you to truly check your system if it performs well.
Yes. It's easy to make a robot that looks good but is overfitting to historical -non repetitive- market behavior. Of course we can never know, as you said, how close will be history relative to the past or recent past.
I'm thinking on test many different market bahaviors (such as 2008 crisis up to more quiet times), and intend the system have at least 300 trades in the period. And be very careful with the optimization part.
In my humble opinion. The moment your strategy needs an indicator in order to *filter out* bad trades, the strategy should be scrapped and it's back to the drawing board.
Filters also filter out *good* trades, so it changes the whole dynamics and thus the strategy in the first place.
And i am not talking about a spread *filter*. For example not entering or exiting when spread is beyond a certain threshold limit or deviation. It is not really a filter in my book, just common sense.