In this study, a Prediction Accuracy Based Hill Climbing Feature Selection Algorithm (AHCFS) is created and compared with an Error Rate Based Sequential Feature Selection Algorithm (ERFS) which is an existing Matlab algorithm. The goal of the study is to create a new piece of an algorithm that has potential to outperform the existing Matlab sequential feature selection algorithm in predicting the movement of S&P 500 (^GSPC) prices under certain circumstances. The two algorithms are tested based on historical data of ^GSPC, and Support Vector Machine (SVM) is employed by both as the classifier. A prediction without feature selection algorithm implemented is carried out and used as a baseline for comparison between the two algorithms. The prediction horizon set in this study for both algorithms varies from one to 60 days. The study results show that AHCFS reaches higher prediction accuracy than ERFS in the majority of the cases.

Over the last few years there has been a surge of activity within the physics community in the emerging field of Econophysics—the study of economic systems from a physicist's perspective. Physicists tend to take a different view than economists and other social scientists, being interested in such topics as phase transitions and fluctuations. In this dissertation two simple models of stock exchange are developed and simulated numerically. The first is characterized by centralized trading with a market maker. Fluctuations are driven by a stochastic component in the agents' forecasts. As the scale of the fluctuations is varied a critical phase transition is discovered. Unfortunately, this model is unable to generate realistic market dynamics. The second model discards the requirement of centralized trading. In this case the stochastic driving force is Gaussian-distributed "news events" which are public knowledge. Under variation of the control parameter the model exhibits two phase transitions: both a first- and a second-order (critical). The decentralized model is able to capture many of the interesting properties observed in empirical markets such as fat tails in the distribution of returns, a brief memory in the return series, and long-range correlations in volatility. Significantly, these properties only emerge when the parameters are tuned such that the model spans the critical point. This suggests that real markets may operate at or near a critical point, but is unable to explain why this should be. This remains an interesting open question worth further investigation. One of the main points of the thesis is that these empirical phenomena are not present in the stochastic driving force, but emerge endogenously from interactions between agents. Further, they emerge despite the simplicity of the modeled agents; suggesting complex market dynamics do not arise from the complexity of individual investors but simply from interactions between (even simple) investors. Although the emphasis of this thesis is on the extent to which multi-agent models can produce complex dynamics, some attempt is also made to relate this work with empirical data. Firstly, the trading strategy applied by the agents in the second model is demonstrated to be adequate, if not optimal, and to have some surprising consequences. Secondly, the claim put forth by Sornette et al. that large financial crashes may be heralded by accelerating precursory oscillations is also tested. It is shown that there is weak evidence for the existence of log-periodic precursors but the signal is probably too indistinct to allow for reliable predictions.

Over the last few years there has been a surge of activity within the physics community in the emerging field of Econophysics—the study of economic systems from a physicist's perspective. Physicists tend to take a different view than economists and other social scientists, being interested in such topics as phase transitions and fluctuations. In this dissertation two simple models of stock exchange are developed and simulated numerically. The first is characterized by centralized trading with a market maker. Fluctuations are driven by a stochastic component in the agents' forecasts. As the scale of the fluctuations is varied a critical phase transition is discovered. Unfortunately, this model is unable to generate realistic market dynamics. The second model discards the requirement of centralized trading. In this case the stochastic driving force is Gaussian-distributed "news events" which are public knowledge. Under variation of the control parameter the model exhibits two phase transitions: both a first- and a second-order (critical). The decentralized model is able to capture many of the interesting properties observed in empirical markets such as fat tails in the distribution of returns, a brief memory in the return series, and long-range correlations in volatility. Significantly, these properties only emerge when the parameters are tuned such that the model spans the critical point. This suggests that real markets may operate at or near a critical point, but is unable to explain why this should be. This remains an interesting open question worth further investigation. One of the main points of the thesis is that these empirical phenomena are not present in the stochastic driving force, but emerge endogenously from interactions between agents. Further, they emerge despite the simplicity of the modeled agents; suggesting complex market dynamics do not arise from the complexity of individual investors but simply from interactions between (even simple) investors. Although the emphasis of this thesis is on the extent to which multi-agent models can produce complex dynamics, some attempt is also made to relate this work with empirical data. Firstly, the trading strategy applied by the agents in the second model is demonstrated to be adequate, if not optimal, and to have some surprising consequences. Secondly, the claim put forth by Sornette et al. that large financial crashes may be heralded by accelerating precursory oscillations is also tested. It is shown that there is weak evidence for the existence of log-periodic precursors but the signal is probably too indistinct to allow for reliable predictions. / Science, Faculty of / Physics and Astronomy, Department of / Graduate

In this work, we propose and investigate a series of methods to predict stock market movements. These methods use stock market technical and macroeconomic indicators as inputs into different machine learning classifiers. The objective is to survey existing domain knowledge, and combine multiple techniques into one method to predict daily market movements for stocks. Approaches using nearest neighbor classification, support vector machine classification, K-means classification, principal component analysis and genetic algorithms for feature reduction and redefining the classification rule were explored. Ten stocks, 9 companies and 1 index, were used to evaluate each iteration of the trading method. The classification rate, modified Sharpe ratio and profit gained over the test period is used to evaluate each strategy. The findings showed nearest neighbor classification using genetic algorithm input feature reduction produced the best results, achieving higher profits than buy-and-hold for a majority of the companies.