Date of Award:

5-2020

Document Type:

Thesis

Degree Name:

Master of Science (MS)

Department:

Mathematics and Statistics

Advisor/Chair:

Kevin Moon

Co-Advisor/Chair:

D. Richard Cutler

Third Advisor:

Tyler Brough

Abstract

The correct assignment of trades as buyer-initiated or seller-initiated is paramount in many quantitative finance studies. Simple decision rule methods have been used for signing trades since many data sets available to researchers do not include the sign of each trade executed. By utilizing these decision rule methods, as well as engineering new variables from available data, we have demonstrated that machine learning models outperform prior methods for accurately signing trades as buys and sells, achieving state-of-the-art results. The best model developed was 4.5 percentage points more accurate than older methods when predicting onto unseen data. Since finance and economics departments pay thousands of dollars in annual data service subscriptions they are often reluctant to fund purchase of additional data containing trade signs when methods for predicting these signs exist. The use of our best trade signing model as an alternative to the purchase of additional data has the potential to collectively save universities millions of dollars in additional subscription fees, facilitate more reliable research, and lighten the burden of data processing for researchers.

Checksum

6ada7fac7d123490b679935eab9805ca

Share

COinS