Tradestation: how to run a Walk Forward process
This example code is found in the Ewfo folder.
C:\Program Files (x86)\EWFO\ iwm.eld
Install this eld into Tradestation
path is C:\Program Files (x86)\Trademaid Enhanced Walkforward optimizer
Create a custom time template.
set the slippage and commission.
In order to be able to do use EWFO the strategy has to be optimized. The output of this optimization in Tradestation is a set of files which are input to EWFO.
Here is how to optimize a strategy in Tradestation as a preparation for EWFO:
1. Prepare the workspace / chart / strategy.
2. Format the strategy and set the optimization values for your strategy and click ok:
Explanation of why slippage and commission should be used in T.S. / M.C.
T.S. & M.C when using genetic optimization will tend to optimize most around the areas that give max fitness. (Normally net profit). If the peak values of a system were around $100,000 profit, with $10 average trade, 10,000 trades - this would not be trade able due to the average trade being too low. This means most of the genetic calculations are around a parameter set that has no interest to us.
If we used $12.50 per side slippage & commission, this will shift the fitness to be looking for higher average trade profits, and less trades. This is much more desirable.
3. In optimization Details set Type=Walk-Forward, choose a method (Exhaustive or Genetic) and enter the Walk-Forward Test Name and click Optimize:
Optionally change method from brute force to genetic. I prefer 10,000 iterations as a minimum, or using genetic - 10% of the iterations of brute force. Whatever number is larger. Normally I keep generations at 100.
4. Once the optimization has completed the optimization files have been created within the Tradestation installation directory under WFO/Data:
Once optimization has completed continue with “EWFO: Preparation for usage“.
This concludes the necessary activities in Tradestation.
Run the icon on the desktop
Ewfo will install here.
C:\Program Files (x86)\EWFO\
Note the in sample and out of sample curves and metrics are not very similar. This is far from ideal.