Abstract
Successful trading in modern markets depends on the ability to monitor and anticipate changes in exchange rates, share, bond and derivative prices, and to make effective decisions to buy, sell, withdraw or hold based on these changes. Rapid changes in market conditions necessitates consistent but flexible strategies to inform, evaluate and take these decisions. The potential of Artificial Intelligence (AI) based systems to augment, and where necessary, replace human decision-making is clear in this field.While AI-based methodologies for trading and successful decision-making are discussed in academic publications, the tools (whether “black box” and producing output by “hidden” algorithmic or statistical means, or “white box” and employing explicit causal mechanisms) are typically proprietary, i.e. exact algorithms, heuristics and modelling archetypes employed are not in the public domain. Hence, for the trader, bespoke development is required to apply these methodologies to provide the information advantage sought.
The work described in this thesis derives from an industrial collaboration between the University of South Wales and OSTC Wales Ltd., a trading company specialising in the trading of interest rates, commodity futures and other derivatives. The thesis addresses the exploitation of market behaviour by combining artificial intelligence with financial trading heuristics, to design and produce a working prototype system which quantifies and categorizes trading strategies.
This thesis therefore details the design of, and examines the further implications and applications for, an early warning detection system for identifying ‘bad’ traders using machine learning. It is assumed traders can be evaluated using the same performance metrics as the trading strategies they employ. That is, a good trader is one that uses a successful strategy. The system is “white box” and hence amenable to analysis and practical experimentation.
The system identifies a set of criteria by which the success of a trading strategy may be judged. It examines these criteria in order to “score” the traders employing these. The trading company, proprietary or otherwise, employing the traders can therefore make better decisions about the amount of funds to allocate to each trader.
The methodology is intended for use by those who manage traders, but can also be used by traders themselves. The detection system found may be used by other concerns in the financial sector whereas the principles examined can be extended to other areas in which decision-making is critical. Examination of the finished system and the process of its composition will provide useful material for academic analysis.
Date of Award | Dec 2017 |
---|---|
Original language | English |
Supervisor | Paul Roach (Supervisor), John Wyburn (Supervisor) & Hugh Coombs (Supervisor) |