The motivational use-case: signal prediction in trading
Back in 2014 I was leading a project, for a client, that had the “simple” purpose of automated trading for stock exchange. The project was quite focused in scope and was, in fact, targeting a discrete number of stocks, all these being clearly articulated/specified by our client. Therefore, following a short initial requirements analysis, we began to explore approaches and model architectures for this project — ranging from simple regression trees on top of MDPs, to ARIMAs and SVM ensembles up to recurrent DAGs.
Now comes the “human” part.