1. Inicio
  2. Atrás
Título: Data mining and knowledge discovery in financial research : empirical investigations into currency
Autores: Wu, Qionglin, 1964-
Fecha: 2001
Publicador: McGill University - MCGILL
Fuente:
Tipo: Electronic Thesis or Dissertation
Tema: Economics, General.
Information Science.
Artificial Intelligence.
Computer Science.
Descripción: Since there exist drawbacks for linear models such as those based on regression techniques, which have been the basis of traditional statistical forecasting models, neural networks are used in this thesis to train and test the input data. This thesis presents a feedforward backpropagation neural network approach to univariate time series analysis. Real world observations of foreign exchange rates in twenty currencies against U.S. dollars have been applied as a study in this experiment. Feedforward connectionist networks have been designed to model daily exchange rates over the period from January 4, 1999 to October 20, 2000. The values of the root mean square error (RMSE) is used as the criterion of selecting the parameters of the training set, testing set, the numbers of the hidden nodes and epochs, the momentum terms and the learning rates. The models obtained in this study by using this method can be used to forecast the movement of these exchange rates.
Before analyzing neural network techniques, data preprocessing and correlation analysis are presented. It is found there exist three correlation situations: the currencies between member countries in European Economic Community (EEC) have very strong correlation relationship; the correlations between Chinese Renminbi and the other currencies are very weak; and the correlations between the other currencies are variable with the change of the time period. They are related to the different finance policy, economic situation and the other factors of each country with the different time period.
Idioma: en