الفهرس 
Title : Portfolio Management Using Reinforcement Learning
AbstractOnly 14 pages are availabe for public view
 |  | from | 116 |  |  |
| | | from | 116 | | |
Abstract
Portfolio management is the art of deciding which sectors to invest in to maximize
the total wealth. Both industry and academic researches are steadily working to develop
novel models to reach greater success and profit. One of the major options is to invest
in the financial markets, financial market trading is a complex process where traders aim
to maximize their expected return while minimizing associated risks. With the increasing
availability of digital historical records, using automated agents for stock market trading
becomes of a significant interest. The purpose of this thesis is to study a subsidiary problem of financial markets called optimizing execution costs using reinforcement learning.
Reinforcement learning is a machine learning branch which circumvents the problem of
defining explicit targets and tackles problems which require sequential decisions. Reinforcement learning has been applied in finance problems, yet execution costs optimization
problem among others still gets little attention. The optimization of execution order in
stock markets is a vital problem, where a trader wants to minimize the cost of buying a
predefined amount of shares over a fixed time horizon. In this study, we propose a novel
reinforcement learning Q-trade model to address the execution costs optimization problem. We tested the Q-trade model using historical data of the Egyptian stock market as
an example of a developing market and Nasdaq stock market as an example of a developed market, it showed in both markets a significant improvement (more than 60% for
some securities) over the compared strategies. Moreover, we develop a deep reinforcement learning model for trading (as a step towards optimal execution), the model managed
to outperform a major milestone in the literature over tested data. Finally, we adapt the
deep reinforcement learning model for optimal execution order. The model managed to
outperform compared strategies over tested data.