How to optimize a trading system without curve-fitting it
There’s a fine line between “optimizing” and “curve-fitting”.
You have to make sure that your don’t over-optimize.
In the following example I’ll show you how to determined the “best” parameter without curve-fitting a
system.
The underlying system is a simple breakout system. The system contains a parameter called <strong>TF_Param</strong>.
In the following I will explain how we optimized the parameter and why we selected 0.7 as the current value for the parameter.
First I run the optimization and look at the <strong>net profit</strong>, since that’s one of the most important figures.
As you can see, any parameter between 0.35 and 0.8 produces robust results.
Next I am looking at the
max drawdown.
Any parameter above 0.4 produces a rather low drawdown, so using the combined information I would pick a
TF_Param
between 0.4 and 0.8.
Now I am looking at the
average profit per trade:
The parameter range between 0.45 and1.2 looks promising, so I am limiting the
TF_param to 0.45-0.8.
Now I am looking at the
winning percentage. The higher, the better:
Anything above 0.5 produces a quite robust winning percentage above 50%.
As it seems, anything above 0.6 is quite good.
Combining all the previous results we should look for a TF_Param between 0.6 and 0.8
As a last test we look at the
number of trades. Again: The higher, the better:
No surprises here: The lower the parameter, the more trades we get.
By combining all the above findings we see that a
TF_Param between 0.6 and 0.8 produces best results.
That’s why 0.7 is definitely a good choice. Even if the market changes slightly, the system would still produce excellent
results.
