Using market and news data to predict price movement of stocks
Seo, Won Seob (2019)
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-201903143018
https://urn.fi/URN:NBN:fi:amk-201903143018
Tiivistelmä
Until recently, investing use to be the domain of human experts. However, many domains where it was thought that artificial intelligence could not compete with human experts turned out that the assumption was wrong. Especially when think about the investing, decision making is backed by huge amount of data, processing that much data is where computers excel at, not humans. One could easily imagine algorithms running on computer someday out perform in profits for investments than the likes of Warren Buffett or Peter Lynch, just like what AlphaGo did to Lee Sedol, the south Korean Go champion. Indeed, there are many current ongoing attempts on computerized investing. Two Sigma is one of those companies who is using machine learning to invest.
Recently, a Kaggle competition that was hosted by Two Sigma challenged machine learning practitioners to predict the movements of stock prices in ten days using market and news data. By participating in this competition, the author tried both deep learning and gradient boosting decision tree models and positioned rather in the upper bracket judging by public leader board of the competition. The implication of this result is machine learning can be effective even if it’s done by a regular bachelor student in engineering field. It does not take PhD or teams of machine learning veterans to come up with nice results.
Thus, in this thesis the author aims to lay out the process and learnings of his Kaggle com-petition.
Recently, a Kaggle competition that was hosted by Two Sigma challenged machine learning practitioners to predict the movements of stock prices in ten days using market and news data. By participating in this competition, the author tried both deep learning and gradient boosting decision tree models and positioned rather in the upper bracket judging by public leader board of the competition. The implication of this result is machine learning can be effective even if it’s done by a regular bachelor student in engineering field. It does not take PhD or teams of machine learning veterans to come up with nice results.
Thus, in this thesis the author aims to lay out the process and learnings of his Kaggle com-petition.