Trading strategies for autonomous agents in financial markets

Electronic trading is relatively new to the long history of financial markets. The typical traditional way of trading in a financial market was that traders gathered in the market floor and exchanged their financial assets in a manner of open outcry. In the past twenty years or so, financial markets have been reshaped dramatically due to the use of electronic trading. Innovations in computing and Internet eliminate the need for direct person-to-person contact via trading floors and make possible global electronic order routing, broad dissemination of trade information, and e effective market operations. Electronic trading systems for financial trading allow to match buy and sell orders automatically without human intervention. Such a system lowers trading costs, bypasses human intermediaries, such as dealers, and also offers faster trade execution. The automation of market operations makes possible for traders to trade automatically. The advances of agent technology have resulted in the emergence of agent-based systems that automate business activities. Ambitious attempts have been made by academia and a number of companies to create automated programs, known as trading agents, which are capable of autonomously trading on behalf of human traders in financial markets. It has been widely recognised that the automation of traders' decision making creates more challenges than machinizing market making. The most crucial decision-making for either a human trader or an autonomous trading agent is when, how much and what to buy or sell an asset in a financial market, known as trading strategies. This thesis aimed to shade a light on the research of autonomous trading agents by introducing an approach of trading strategy de- sign and analysis. To mimic the way of decision-making by human traders, we investigated a number of financial technical indicators, including Simple Moving Average, Exponential Moving Average, Moving Average Convergence and Divergence, Relative Strength Index and Stochastic Oscillator, that have been widely used by human traders for market analysis and used them as the major signals for design of bidding and investment strategies. By utilising our Jackaroo Trading Agent Platform (JTAP), we first analysed the statistic properties of each technical indicator using the real data of stock exchange taking from the Australian Stock Exchange and then implemented a range of trading strategies based on the technical indicators we have investigated. The trading strategies we designed can be divided into two categories: bidding strategies and investment strategies. A bid- ding strategy deals with when, how many and how much to buy or sell a financial product, whereas, an investment strategy decides how much fund is allocated to each financial product. A simple bidding strategy can be easily designed based on either trend based indicators or momentum based indicators, and an investment strategy can be implemented based on the enforcement learning with returns as re- wards. However, our experiments showed that a trading strategy can perform much better if we combine a number of technical indicators and market returns. This is because such a combined signal can be more stable under volatile market conditions and more responsive to market returns. Each trading strategy we implemented has been fully tested against the real market data and simulated market data