I left of having fully trained and tested models, graphed their feature distributions, and tested on a fully different set of stocks to generalize to real-world results.
I don't expect to beat the MSE of our baseline because that would mean that a couple of college kids built a tool that can beat the market in a semester, a task that mega-corps have been trying to do for decades.
However, we can still do more to get our model's accuracy closer to the baseline. If you've noticed the outliers in the feature distributions, I'm wondering if we could improve our accuracy by just getting rid of those outliers. They can be very far away and mess up our scaling.
Please test out anything else you can think doing to improve our model and give us more to talk about in the final report.