Sparse machine learning methods for financial signal processing
Ever since stock trading came into force, financial economists are keen on identifying optimal methods that track stock movements and make a prediction on future prices with a high degree of accuracy. One such research problem is portfolio optimization. Ever since then an extensive research has bee...
Saved in:
Main Author: | |
---|---|
Other Authors: | |
Format: | Theses and Dissertations |
Language: | English |
Published: |
2017
|
Subjects: | |
Online Access: | http://hdl.handle.net/10356/72617 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Nanyang Technological University |
Language: | English |
Summary: | Ever since stock trading came into force, financial economists are keen on identifying optimal methods that track stock movements and make a prediction on future prices with a high degree of accuracy. One such research problem is portfolio optimization.
Ever since then an extensive research has been carried on in the space of portfolio optimization in financial engineering. A several of signal processing techniques like Time Series analysis, Regression analysis and Bayesian inferences have found their deep inroads into this particular area of research.
Investors are keen on holding portfolios that maximize the returns and minimize the risk. And research of portfolio theory was primarily driven as a mean – variance optimization problem where in mean mimicked the returns and variance the risk.
Current active research in portfolio optimization is for finding efficient mathematical techniques that solve mean – variance optimization problem efficiently and accurately with less computation burden. The new edition to current research in portfolio theory is sparsity.
Sparsity explores the new space of achieving mean – variance optimization with minimum possible asset combinations. The prime motivation of this thesis is to find applications of decision sciences and machine learning techniques to solve the mean – variance optimization problem to construct portfolios that are optimal and simple with the aid of less computation intensive methods.
Current thesis work aims to address the above problem from decision theory perspective. It employs a widely acknowledged approximation techniques of Bayesian inference like variational methods and exploits the advantages of such methods in constructing portfolios that are optimal and simple.
This research shall find wide applications in financial industry from risk – management perspective. Sparsity which implies minimal combination of assets to form a portfolio significantly lessens the operational costs and transaction costs involved in holding a bulky portfolio in open markets. This shall be an advantage to end investors and enables to better manage the risk of a portfolio with minimum assets in place. |
---|