ONLINE PORTFOLIO SELECTION
HAN EN CHOU
HAN EN CHOU
Citations
Altmetric:
Alternative Title
Abstract
This paper proposes an online portfolio selection strategy named "Box Theory Portfolio Selection
(BTPS)". There is an increased interest from the Artificial Intelligence and Machine-Learning community to apply learning techniques to perform the portfolio selection in a sequential manner over a period of time. Empirical evidences shows that stock prices follow the
Mean-reversion phenomenon in the short run and are trend-following in the long-run. We borrow the trading idea of Nicolas Darvas Box System to exploit these two behaviours. We make
use of an online passive aggressive learning framework to put this trading idea to practice. To
more comprehensively evaluate BTPS, we benchmark its performance against existing strategies.
We used six datasets from existing literature and introduced ve new datasets from Asian
stock markets. Our empirical results on these real markets showed that performance of existing
strategies lack consistency across various datasets. While not a top-performing strategy, BTPS
has shown consistent performance and a strong ability to stop loss in adverse market conditions
across all datasets. In addition, BTPS has a superior computational run-time that is desirable
for online portfolio selection.
Keywords
Machine Learning, Data Mining, Computational Finance, Portfolio Optimisation
Source Title
Publisher
Series/Report No.
Collections
Rights
Date
2016
DOI
Type
Thesis