> If you actually try to use the prediction of MA(3) you will find it is often very hard to use, because it's always just too late. I wonder if the machine lear

> If you actually try to use the prediction of MA(3) you will find it is often very hard to use, because it's always just too late. I wonder if the machine learning version had the same problem, or not. Judging from the graph in the article, it looks like it doesn't. The model didn't reinvent the moving average, it invented it without the inherent delay. Though it does seem to diverge from real price over time.