A US asset manager asked if we could create a machine learning model that predicted US equity prices. They wanted the model to utilize a specific set of inputs to make the predictions, but they lacked the mathematical and computational expertise to create it.
We drafted a proposal that outlined our plans, including:
Since we made use of code that formed part of our internal project, the client agreed to only own the custom code that we developed for them. We agreed on a fixed cost, and estimated that the project would take three months.
Processing the input data took longer than we thought, due to idiosyncrasies in file formats. We also had to battle memory issues as the data sets were large. Though some of us worked overtime to try to meet the deadline, we were late by a month and a half.
Fortunately, the clients were happy with the results of the machine learning model. They obtained historical predictions of each stock from the model, and ran several backtests which came out favourably. The client, who wishes to remain anonymous, have since asked us to complete a follow on project.