KEML Week One Blog — Curiosity

As a senior student who’s accepted a job offer already for after this academic year, I’ve started to think about life after college. Apartments to rent, to buy a car or not, what benefits package to accept. Imposing ideas to the 22 year old mind.

One of the things that I’ve started considering is investment in the stock market and finance in general. Having to balance loans, rent, food, other payments, and the general day to day spending of being a functioning member of society will be enough to worry about, sure, but even more so will long-term wealth management and planning worry me. I became curious about the stock market and smart investing because I have this futuristic outlook on where I want to be not only in one year, but five or ten or twenty years.

I wasn’t surprised when I logged onto and began researching to find that Bloomberg was using machine learning to predict potential stock market futures based upon historical data, world event inputs, and current financial statuses of many, many entities. I decided to investigate a bit further and found this ( paper from Stanford University. As a bonus feature of the paper, the researcher had used Bloomberg data for their example stock futures model, which tied further into my initial interest in the stock market.

I was impressed with the Stanford paper for a number of reasons. First off, it truly used a long set of historical data (approximately 5.5 years of daily stock prices, or 1471 data points), each containing statistics on 16 different features used to train the machine to predict the futures. It was intriguing to see how the accuracies improved from the initial next day futures prediction (highest accuracy in the paper was 58.2% for an optimal next-day stock prediction, which is basically equivalent to flipping a coin) to 79.3% for a window of a 44-day stock future prediction.

The researchers went on further to develop a stock-trading strategy based upon their ideal 44-day stock future prediction. As both a student of computing and as an interested party in the investment and financial markets, this was the perfect section for me: the application of computing concepts and data to make informed, real-time, high-risk decisions around financial futures and (hopefully) benefit both me and those I am working for in the long run.

The end of the paper was the most interesting portion. At the conclusion, using the model developed and comparing it to the 3M stock value, the machine learning model actually outperformed on annual return, which leaves me curious as to how this strategy can be implemented in other ways centered around financial markets. I want to do more research and see what other applications in the finance area I can use machine learning in.