Hi Smruti, thank you for dropping me a line. I am basically filtering out the tickers with poor prediction power, and training my model to predict the movement of filtered tickers only — this is about 150 tickers if I am not mistaken. This is essentially the same as manually picking tickers, but here I am using the average correlation as a metric to help me choose.
One way look forward bias can be introduced is if I had not partitioned the training, validation, and test sets incorrectly. However, I made sure I have done this part correctly.
In that regard, I can’t see look forward bias being introduced in my approach. Please correct me if I am missing something. Happy to learn and correct my approach.