Identifying Profitable Day-Trading Opportunities Based on Machine Learning

ROLE: DATA ANALYST

01/21 - 05/21

Academic project completed at Cornell University with Yiheng Dong, Erhao Zhao, Zili Zhou

Instructor: Professor Yudong Chen

ABSTRACT

Day trading, as a speculative trading style that involves the opening and closing of a position on a daily basis, can be affected by all sorts of variations in the market. It is desirable to build a model to predict whether a transaction can benefit at all, given the entry time, the stock information, and real-time market situations. We implemented different supervised learning models for that purpose, with a substantial amount of minute-level trading opportunity data for US stocks. This paper describes data preprocessing, modeling methodologies, comparison and evaluation of several classifiers, and further improvement and insights into the modeling result. The overall performances of all models do not differ significantly, but certain models may be recommended based on different risk-return preferences according to our performance analysis.

REFERENCE

“Cleaned Day-Trading Training Data.” Kaggle. https://www.kaggle.com/dawerty/cleaned-daytrading-training-data.

Milovanović, M., Minović, M., Štavljanin, V., Savković, M., & Starčević, D. (2012). Wiki as a corporate learning tool: Case Study for Software Development Company. Behaviour & Information Technology, 31(8), 767–777. https://doi.org/10.1080/0144929x.2011.642894

Yahoo Finance Python Module. https://github.com/lukaszbanasiak/yahoo-finance