Price pattern detection via functional biclustering

The detection and the forecast of recurrent patterns is of utmost importance in financial time series. For example, it is useful to predict the explosion of price bubbles in advance in order to mitigate the consequences for the society. This project constitutes a further development of an existing research project on bubble detection via unsupervised learning methods (already financed by the SSHRC). Specifically, the proposed project aims to detect repetitive typical local patterns in asset price time series, such as oscillations or trends, that are crucial for portfolio managers, traders and retail investors in financial markets. We achieve this goal by adapting and applying to time series an innovative biclustering-inspired functional motif discovery algorithm which is able to capture repeated fragments in a noisy curve. In so doing, we expect to discover recurrent price anomalies and predict future price directions and unattended changes.

Faculty Supervisor:

Federico Severino

Student:

Partner:

Pennsylvania State University

Discipline:

Sociology

Sector:

Education

University:

Université Laval

Program: