This post is the second from a series on Decision tree for stock prediction. The first post is available here.
Before using decision tree on each stock separately to make its prediction, the stock universe has to be narrowed down. The “business” reason is that, in my case, we are only interested by big capitalization. The “technical” reason is computational power. Indeed, computing predictions for every stock among all possible US stocks would be very long. I will explain more about that later in the series of posts.
The filtering process starts from the US stock market (> 20’000 stocks). It is narrowed down to a number between 50 and 200. This number is typically a parameter of the system, that can be tuned (if we have enough calculation power of course). The filter is based on information such as the price of the stock, the volume of transactions and the current market capitalization of the company. This is done in order to select only the “big” companies.
As the system is based on a moving time window principle, the filtering is done at each movement of the time window. The next step is to make prediction on these, let say 100 filtered stocks and this will be the subject of the next post of the series.