On the predictability of U.S. stock market using machine learning and deep learning techniques

Conventional market theories are considered to be inconsistent approach in modern financial analysis. This thesis focuses mainly on the application of sophisticated machine learning and deep learning techniques in stock market statistical predictability and economic significance over the benchmark conventional efficient market hypothesis and econometric models. Five chapters and three publishable papers were proposed altogether, and each chapter is developed to solve specific identifiable problem(s). Chapter one gives the general introduction of the thesis. It presents the statement of the research problems identified in the relevant literature, the objective of the study and the significance of the study. Chapter two applies a plethora of machine learning techniques to forecast the direction of the U.S. stock market. The notable sophisticated techniques such as regularization, discriminant analysis, classification trees, Bayesian and neural networks were employed. The empirical findings revealed that the discriminant analysis classifiers, classification trees, Bayesian classifiers and penalized binary probit models demonstrate significant outperformance over the binary probit models both statistically and economically, proving significant alternatives to portfolio managers. Chapter three focuses mainly on the application of regression training (RT) techniques to forecast the U.S. equity premium. The RT models demonstrate significant evidence of equity premium predictability both statistically and economically relative to the benchmark historical average, delivering significant utility gains. Chapter four investigates the statistical predictive power and economic significance of financial stock market data by deep learning techniques. Chapter five give the summary, conclusion and present area(s) of further research. The techniques are proven to be robust both statistically and economically when forecasting the equity premium out-of-sample using recursive window method. Overall, the deep learning techniques produced the best result in this thesis. They seek to provide meaningful economic information on mean-variance portfolio investment for investors who are timing the market to earn future gains at minimal risk.