DataSplit training and trading

Hello and first, a big thanks for this wonderful work.


I was wondering what dates range are used when training with WFO.

The manual says "WFO uses a default training period of 85% and a default test period of 15%."

If data consists of 100 days, then the first 85 days will be used to train factors and parameters and the last 15 days for backtesting.

If I then decide to trade with this algorithm, then the optimize factors/parameters are missing the last 15 more recent days.

=> Is it true ? Are the days split randomly in the timeline ?

In the case you mentioned, after the first cycle the walk-forward window would shift forward by 15 bars and retraining would occur on the final 85 bars in the data set. The resulting parameters are then saved in a .par file and are not used in simulation, but are used for live trading. In this way, Zorro uses the most recent parameters for live trading.

From the documentation:

DataSplit
Splits the simulation in a training period (given in percent) and a following test period. F.i. when set at 60, the training period has a length of 60% and the test period has a length of 40%. This ensures that the test always uses out-of-sample data.

Also have a look at: http://zorro-project.com/manual/en/numwfocycles.htm

And for the general concept you might consider: Urban Jaekle, Emilio Tomasini: Trading Systems, which is recommended by jcl in his manual and is a short read, but afterwards many thinks gets clearer.

Thanks for your answer, it's now more clear.