Stock markets are very important in modern societies and their behaviour have serious implications in a wide spectrum of the world’s population. Investors, governing bodies and the society as a whole could benefit from better understanding of the behaviour of stock markets. The traditional approach to analyze such systems is the use of analytical models. However, the complexity of financial markets represents a big challenge to the analytical approach. Most analytical models make simplifying assumptions, such as perfect rationality and homogeneous investors, which threaten the validity of analytical results. This motivates the use of alternative methods. For those reasons, the study of such markets is a fertile field to use the agent-based methodology. In this work, we developed an artificial financial market and used it to study the behaviour of stock markets. In this market, we model technical, fundamental and noise traders. The technical traders are non-simple genetic programming based agents that co-evolve (by means of their fitness function) by predicting investment opportunities in the market using technical analysis as the main tool. Such traders are equipped with

Abstract. The journal and in particular the resource reviews have been running for ten years. There are a number of activities being planned to celebrate. However it is a good time to revisit our original and updated goals again [1, 2], compare them with what the journal has achieved and make new plans. Section 2 onwards gives up to date statistics on the genetic programming and evolvable hardware literature and electronic resources. 1. Ten Years of Resource Reviews Excluding special issues and 10(3), every issue of GP/EM contained at least one review, making a total of 51. It was intended from the start that these would cover, not just books, but “resources ” in the wider sense, particularly, web pages, on-line resources, packages and products. We have reviewed 36 books, 10 edited collections, and two conference/workshop proceedings. (It has been agreed that a journal like GP/EM is not appropriate for immediate dissemination, and so we don’t expect to review any more conferences or workshops, since many of their results are quickly updated. Instead electronic newsletters, like SIGEvolution, have carried short reviews of a number of events.) We have also published an article on Internet-based resources, one on software (albeit covering three packages) and reviewed one product. Topics have included not only genetic programming (21) evolvable hardware (7) and genetic algorithms (5) but also particle swarm optimisation (3), artificial development and embryos (2) robotics (2), Ant colony Optimisation, evolutionary programming, data mining, evolutionary design and art. Reviews have also included books on DNA computing (2), quantum computing, cellular automata, intelligent bioinformatics and the history of artificial intelligence. Articles have been written by authors based in

Foreign exchange (forex) market trading using evolutionary algorithms is an active and controversial area of research. We investigate the use of a linear genetic programming (LGP) system for automated forex trading of four major currency pairs. Fitness functions with varying degrees of conservatism through the incorporation of maximum drawdown are considered. The use of the fitness types in the LGP system for different currency value trends are examined in terms of performance over time, underlying trading strategies, and overall profitability. An analysis of trade profitability shows that the LGP system is very accurate at both buying to achieve profit and selling to prevent loss, with moderate levels of trading activity.

Abstract. The journal and in particular the resource reviews have been running for ten years. There are a number of activities being planned to celebrate. However it is a good time to revisit our original and updated goals again [1, 2], compare them with what the journal has achieved and make new plans. Section 2 onwards gives up to date statistics on the genetic programming and evolvable hardware literature and electronic resources.

There has recently been a considerable amount of research into algorithmic traders that learn [7, 27, 21, 19]. A variety of machine learning techniques have been used, including reinforcement learning [20, 11, 19, 5, 21]. We propose a reinforcement learning agent that can adapt to underlying market regimes by observing the market through signals generated at short and long timescales, and by using the CHQ algorithm [23], a hierarchical method which allows the agent to change its strategies after observing certain signals. We hypothesise that reinforcement learning agents using hierarchical reinforcement learning are superior to standard reinforcement learning agents in markets with regime change. This was tested through a market simulation based on data from the Russell 2000 index [4]. A significant difference was only found in the trivial case, and we concluded that a difference does not exist for our agent design. It was also observed and empirically verified that our standard agent learns different strategies depending on how much information it is given and whether it is charged a commission cost for trading. We therefore provide a novel example of an adaptive algorithmic trader. i Acknowledgements First and foremost, I must thank Dr. Subramanian Ramamoorthy for supervising and motivating this research, and for providing sensible suggestions when I found myself low on ideas. I would also like to thank my family and friends for cheering me up and giving me moral support. ii Declaration I declare that this thesis was composed by myself, that the work contained herein is my own except where explicitly stated otherwise in the text, and that this work has not been submitted for any other degree or professional qualification except as specified.