Out of sample


All the back tests were originated by the use of the WFA (Walk Forward Analysis) tool.

This technique splits analyzed data into a certain number of runs that will divide data in blocks. In each block (known as out-of-sample period) the parameters to generate signals will be provided by the optimization reached during the in-sample period, that time frame that ended the day before the run out-of-sample.

Only the results achieved during the various runs out-of-sample will contribute and take part to the development of the equity line while those pertaining to the in-sample period will be automatically discarded. This approach guarantees either a strong series of results and a serious reliability of the obtained back tests.


Let's take a look at this simple example.



Starting from the in-sample (in blue); our example covers four years and the system chose the most important parameters.

Since the fourth year, every year we train the system showing the in-sample period.

The parameters seen as 'dominant' will be applied for the following year (out-of-sample).

At the end of such year the process will be repeated, training the system by watching the last four years and the parameters identified as 'dominant' will be applied to the following year, and so on, up to today.

The resulting equity line will be obtained from the fourth year, that is after the end of the first in-sample period.

Thus, the risk of over-fitting is eliminated. Every forecast on the equity line will be based on values coming from the application of the system on non-optimized periods (out-of-sample) with a higher value compared to the opposite case.

Privacy Policy