With walkforward optimization, the Trader allows you to evaluate the out-of-sample performance of a Trading Strategy that is reoptimized regularly on newer data.
In this daily model for Deere we’ve loaded data back to 2006. We’re testing a system that goes long when the RSI indicator is less than 30 and sells short when RSI is greater than 70. The optimization is set to find the best parameters. On the dates tab, we’ve set the model to optimize on one year of data and test on 6 months of trading data. On the Advanced tab, we chose 6 walkforwards, and checked the box to carry our account balance from one walkforward to the next, as you would if you were actually trading this model.
The Trader divides the data up into 6 sets each shifted by 6 months, the amount of data we entered for the trading period. The Trader holds out the last 6 months of data for optimization period 0 so the model trains on the latest data for future trading.
Let’s see how this model worked. Walkforward #6 starts with the earliest data, with the optimization set from 10/17/2006 to 10/17/2007. The model built on that data is applied to 6 months of data from 10/18/2007 to 4/18/2008. Walkforward #5 shifts the data forward 6 months and starts the optimization on 4/18/2007 and runs until 4/18/2008. The model built from walkforward #5 is applied to the next 6 months of data, from 4/21/2008 until 10/20/2008. The pattern continues for the remainder of the walkforwards. The last 6 months of data are optimized and become walkforward 0, which is the model you will use for trading.
The results for this model look good for the majority of the Optimization and Trading sets, so you would expect that this model would continue into the future.