The principle of Occam´s razor is usually - and incorrectly - taken as meaning that the simplest hypothesis is most likely to be true, but really it's a pragmatic principle. It says that the most efficient way to explore the space of solutions is to start with the simplest ideas first. Only introduce more factors if you need them.
Unfortunately, in complex systems like financial markets there are too many variables, too many interacting agents and too much fluidity. If your model is simple and understandable it doesn't agree with reality, but if you try to make it more realistic you have too many possibilities to choose from.
The great advantage of deep machine learning is one doesn't have to construct explicit hypotheses and models in order to create predictions, and that means one can use more data and more variables. Such systems work with the information given, and speaking very loosely, optimize themselves. Deep machine learning allows you to cut off larger chunks with Occam´s Razor.
Our approach is to take this unique characteristic of ML and to apply it in a way that recognizes the other great truth of modern science: that not everything is always predictable! We have developed, and are developing further, a number of approaches that allow us to apply ML where and when it will be most effective.