Imperial College London

 "Advanced methods of portfolio optimization employing dynamic factors models along with the Kalman filter estimation procedure"

​

​

​

Abstract 

In the topic of portfolio management, a substantial amount of study has been carried out over the course of the previous few decades. There has been research done on static factor models for predicting portfolio returns, however static factor models presume that factors remain stationary throughout a certain timeframe. As a result, dynamic factor models have been created to investigate the trade-off between returns and risk and make use of both observable and unobservable components in the process of forecasting. This thesis project aims to research and construct dynamic factor models that properly anticipate the returns of portfolio assets so that it may be used for investment purposes. In this project, two dynamic techniques are constructed based on the principles and applications of Convex Optimization Theory and Semidefinite Programming. Both theories and their applications are discussed. The first method stacks state-space matrices and converts dynamic observable variables into static observable variables. It is founded on the principles of the Model Predictive Control framework, which is used to carry out these two tasks. The second strategy involves putting into action the Expectation-Maximization algorithm on real-time data to investigate the unobservable elements that are affecting the performance of the assets. The aforementioned models are incorporated to the Kalman Filtering estimating method, and they are evaluated with regard to a variety of macroeconomic parameters. When evaluated with synthetic data and data taken from the real-time financial data, the created algorithms are reliable and produce correct results.