Font Size: a A A

A Neural Network Learning Algorithm Based On The Extended Kalman Filter Applied To Stock Prediction

Posted on:2004-08-26Degree:MasterType:Thesis
Country:ChinaCandidate:F HeFull Text:PDF
GTID:2156360122467020Subject:Finance
Abstract/Summary:PDF Full Text Request
High return always goes with high risk in stock market. In order to pursue profit and avoid risk as much as possible, people work hard to study the inherent regulation of the market and explore efficient forecasting methods all the time. However, due to the enormous influential factors and complexity of the market structure, it is extremely difficult for us to discover the quantitative relation and conduct econometrical analysis. In addition, the data processing volume of stock market modeling and prediction is exceedingly huge. In the mean time, the requirement to precise and advanced algorithm is exigent. In this case, a modified learning algorithm for a multi-layered neural network based on extended Kalman filter (EKF) is proposed to predict the stock price. The main research work and achievement of the thesis include the following contents:Firstly, we introduce the basic theory of the neural network and network model, especially the explanation of the working process of the back-propagation learning algorithm. Then we illuminated the kalman filter and the extended kalman filter method. On the basis of the theories mentioned above, we put neural network and kalman filter all together. We have the neural network learning algorithm based on the extended kalman filter.In order to compare traditional BP and EKF learning algorithms, we use single-strap prediction and multi-strap prediction method and apply this algorithm in stock price forecast of PuFayinhang(Code: 600000), Handan-Gangtie(Code: 600001) and GuiguanDianli(Code: 6000236). The result shows that modified EKF learning algorithm has improved convergence and can provide much more accuracy learning results. Experiments in forecasting stock price are given to show the feasibility and efficiency of the proposed algorithm.
Keywords/Search Tags:Neural network, Extended kalman filter, UD factorization, Stock prediction
PDF Full Text Request
Related items