Comparison of Scalable Analytics Correlation and Classical Correlation in High-Frequency Finance

Measures of covariance and correlation between returns of different financial assets are of great interest in the finance industry. In the high-frequency domain, raw price data is filled with numerous bad data points to which traditional definitions of correlation by Pearson are very sensitive to these outliers, and thus should not be directly applied to raw high-frequency data. Robust measures of correlation less sensitive to outliers can be used to improve the performance of popular financial methods. The objective of this project, in partnership with Scalable Analytics, a cutting-edge startup company whose mandate is to provide real-time analytic tools for companies using high-frequency financial data in electronic trading, is to study properties and assess the usefulness of robust covariance/correlation calculations with intra-day data for financial applications. The project contains two phases. The first phase of the project is an in-depth study of high frequency correlation time series data to ascertain fundamental statistical properties of such data sets, called “stylized facts” in the finance literature. The second phase of the project is to draw on results from the first phase to design and test canonical pair trading strategies that use robust correlation.

Faculty Supervisor:

Dr. Rachel Kuske

Student:

Jieren Wang

Partner:

Scalable Analytics Inc.

Discipline:

Mathematics

Sector:

Finance, insurance and business

University:

University of British Columbia