Introduction
The old paradigm
Till this day, there has always been an assumption that machine learning algorithms can rely on past data to create a model which would predict the future. The flaw with this view of machine learning is that by relying on such data we're assuming that the world doesn't change and patterns don't evolve. While in some cases this could actually be true, in complex environments like the financial world it simply cannot be. The world of finance evolves all the time and the effect of each change ripples through the entire world of finance.
The new paradigm
So what is the solution? Two key additional piece of information should be added to the historical data in order to make better predictive/prescriptive models. The first piece is considering the relationship between entites (e.g. companies/assets). This will ensure that whatever algorithm we build on top of this data will be aware of the mechanism with which the data evolves. The second piece of information we need to consider is the present data. In other words, we should separate between what is historical patterns of change and what is the present state of the world. This is simply because, the present state of the world has much more influence in shaping the future direction of the finance than any other historical data.
