05. Possible Solution 2

This idea is proposed in Marcos Lopez de Prado’s book, Advances in Financial Machine Learning, which is an interesting resource for further reading on this topic. A further question Lopez de Prado touches on is what bag size to use according to this method.

One recommendation is to reduce the size of each bag to be a fraction of the number of rows of the original dataset. As the fraction, use 1 divided by the number of labels that overlap at each time point, on average. So if you were using weekly returns, the fraction would be ⅕, or 0.2.