Good point regarding the macroeconomic factors. It doesn't look like such datasets are readily available (not for free at least) so you'd probably a script to aggregate this information from different sources, but it's doable in theory. In fact that was also my thinking when I built this bot as a prototype.

People generally refer to market movements as chaotic meaning that sometimes they just move for no reason at all. I'm quite certain that what we actually call chaos, is simply data that we haven't observed or measured yet.

I'm that if you somehow measured all the variables that might affect the price of an asset, and ratio of how much each variable contributes to the movement and feed it all to a machine learning algorithm, it would perform quite well. This is pretty much what you suggested above in your day traders example.

The performance and correct prediction ratio per asset is interesting, thanks. How do you see this data work? Just as a tangential script or would the machine learning somehow ingest this in order to improve its own predictions?