Deep Portfolio Model

Portfolio management is a well-known multi-factor optimization problem, which involves historical price data, macroeconomic data and market indicators. Many researches have been conducted to find optimal portfolio allocation using machine learning. However, some limitations still exist, such as incorporating multiple factors into a single model, the robustness for encountering economic events, and potential bias by using historical data. The objective of the project is to develop an adaptive portfolio selection method to achieve the desired objectives (e.g. risk minimization, return maximization, index tracking) based on deep learning. The procedures include: Investigate the current limitations and identify their causes; Propose solutions to the limitations through literature review and empirical experiments; Implement the solutions into the portfolio selection model; Achieve more desirable results than the current ones. The improved method would be able to assist in automating portfolio management and be adaptive to a diversity of objectives and different types of data.

Faculty Supervisor:

Deepa Kundur

Student:

Partner:

Osaka University

Discipline:

Engineering

Sector:

Education

University:

University of Toronto

Program: