This is an interesting approach. You’ve clearly been quite thorough in laying out your process, but like some other commenters, I worry about the accuracy of this model beyond what you’ve shown here. Primarily using historical data to forecast in a volatile market has some serious drawbacks. My company Evo uses AI to anticipate demand in the fashion industry (another highly volatile market), and a core reason that we are able to get better results than traditional methods has been the inclusion of real-time data. I’d be curious to see how well this model performs when there is an unexpected market event.