This will probably be the last post I have regarding PAM. All the project files can be found on my github page. I don't think I'll be working with her anymore as I don't think theres anymore methodologies to drastically improve her error with JUST price action. It's not worth my time and I need to move on. I introduced some new features such as Gap, Previous ratios, etc. They did lower the errors again. I have come a long way from validation errors of approx 0.6938 to 0.4955. Out-of-sample pairs forecasting is still inconclusive as the model did marginally improve in errors but still quite high compared to validation set and training set errors which I found weird.
Regardless, I will be retiring the project and anyone who wants to improve Pam is welcome to do so. The last thing I believe I will do is wait a few weeks and run another trading simulation to see how it generalizes overall to new information.