Contribution to Financial Modeling and Financial Forecasting

This thesis consists of three chapters. Each chapter is independent research that is conducted during my study. This research is concentrated on financial time series modeling and
forecasting.
On first chapter, the research aims to prove that any abnormal behavior in debt level is a signal of future unexpected return for firms that is listed in indexes in this study, hence it is a
signal to buy. In order to prove this theory multiple indexes from around the world were taken into consideration. This behavior is consistent in most of indexes around the word.
The second chapter investigate the effect of United State president speech on value of United State Currency in Foreign Exchange Rate market. In this analysis it is shown that during the time the president is delivering a speech there is distinctive changes in USD value and volatility in global markets. This chapter implies that this effect cannot be captured by linear models, and the impact of the presidential speech is short term.
Finally, the third chapter which is the major research of this thesis, suggest two new methods that potentially enhance the financial time series forecasting. Firstly, the new ARMA-RNN
model is presented. The suggested model is inheriting the process of Autoregressive Moving Average model which is extensively studied, and train a recurrent neural network based on it to benefit from unique ability of ARMA model as well as strength and nonlinearity of artificial neural network. Secondly the research investigates the use of different frequency of
data for input layer to predict the same data on output layer. In other words, artificial neural networks are trained on higher frequency data to predict lower frequency. Finally, both stated
method is combined to achieve more superior predictive model.