With all due respect, I saw this technique in the book , Advances in financial machine learning, but I found that it acts like a filter for the trades only. And it seems doing the job of overfitting past data by filtering out those bad trades ... I just don't get it why meta labeling could give a realistic help... could someone help me on the topic please?
I am late here, but I will give my answer to help any one with same question. Ml tries to find patterns in the data, and if you give labels based on trend this will not be enough. There has to be a strategy so the algorithm can connect the dots. The reason is why meta labeling is done is that you think that for example moving average cross over works but you don’t know when? You filter the bad trades to teach the algo where moving average worked. By doing that you ask the machine to identify the environment of when to invest and when not..