Use this URL to cite or link to this record in EThOS:
Title: Applied machine learning for systematic equities trading : trend detection, portfolio construction and order execution
Author: Sethi, M.
Awarding Body: UCL (University College London)
Current Institution: University College London (University of London)
Date of Award: 2016
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
The systematic trading of equities forms the basis of the Global Asset Management Industry. Analysts are all trying to outperform a passive investment in an Equity Index. However, statistics have shown that most active analysts fail to consistently beat the index. This Thesis investigates the application of Machine Learning techniques, including Neural Networks and Graphical Models, to the systematic trading of equities. Through approaches that are based upon economic tractability it is shown how Machine Learning can be applied to achieve the outperformance of Equity Indices. In this Thesis three facets of a complete trading strategy are considered, these are Trend Detection, Portfolio Construction and Order Entry Timing. These three facets are considered in an integrated Machine Learning framework and a number of novel contributions are made to the state of the art. A number of practical issues that are often overlooked in the literature are also addressed. This Thesis presents a complete Machine Learning based trading strategy that is shown to generate profits under a range of trading conditions. The research that is presented comprises three experiments: 1- A New Neural Network Framework For Profitable Long-Short Equity Trading - The first experiment focusses on finding short term trading opportunities at the level of an individual single stock. A novel Neural Network method for detecting trading opportunities based on betting with or against a recent short term trend is presented. The approach taken is a departure from the bulk of the literature where the focus is on next day direction prediction. 2- A New Graphical Model Framework For Dynamic Equity Portfolio Construction - The second experiment considers the issue of Portfolio Construction. A Graphical Model framework for Portfolio Construction under conditions where trades are only held for short periods of time is presented. The work is important as standard Portfolio Construction techniques are not well suited to highly dynamic Portfolios. 3- A Study Of The Application of Online Learning For Order Entry Timing - The third experiment considers the issue of Order Execution and how to optimally time the entry of trading orders into the market. The experiment demonstrates how Online Learning techniques could be used to determine more optimal timing for Market Order entry. This work is important as order timing for Trade Execution has not been widely studied in the literature. The approaches that form the current state of the art in each of the three areas of Trading Opportunity (Trend) Detection, Portfolio Construction and Order Entry Timing often overlook real issues such as Liquidity and Transaction Costs. Each of the novel methods presented in this Thesis considers such relevant practical issues. This Thesis makes the following Contributions to Science: 1- A novel Neural Network based method for detecting short term trading opportunities for single stocks. The approach is based upon sound economic premises and is akin to the approach taken by an expert human trader where stock trends are identified and a decision is made to follow that trend or to trade against it. 2- A novel Graphical Model based method for Portfolio Construction. Standard Portfolio techniques are not well suited to a dynamic environment in which trades are only held for short time periods, a method for Portfolio Construction under such conditions is presented. 3- A study of the application of Online Learning for Order Entry Timing. Order Entry Timing for Trade Execution has not been widely studied in the literature. It is commonly assumed that trading orders would be executed at the day end closing price. In practice there is no real reason to trade on the close and it is shown that better execution may be obtained by trading at an earlier time which can be determined through the application of Online Learning.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available