A Study Of Forecasting For Stock Index Based On ANN And GA

In recent years, it offers various new technologies and ways to forecasting stock index with the development of theory of Artificial Neural Network (ANN). At present, many scholars of the world has constructed the model of forecasting stock index based on ANN, and some other scholars have tried to combine artificial intelligence and ANN to build models used for stock index forecasting. But the forecasting capacities of all these models haven't reached the required precision. The thesis aims to deficiencies of current stock index forecasting ways, according to the character of non-linearity variation of stock index, constructs the model of forecasting for stock index based on ANN and Genetic Algorithm (GA) through the combination of them, which can improve the rationality of variables selection and forecasting precision, thus it provides a new way for stock index forecasting.In detail, the thesis includes seven chapters as following.Chapter 1 is introduction, which introduces the aims and significance of this thesis, and reviews national and international literatures of forecasting for stock index based on ANN. And it also introduces the content, research method and structure of the thesis in this chapter.Chapter 2 discusses the current situation of stock index forecasting and points out the problems of these ways. It mainly includes two kinds of ways of forecasting for stock index based on ANN. The first way is using single ANN to construct forecasting model, but the forecasting precision is poor because of the 'over-fitting'. The other kind way is combining various artificial intelligences for forecasting, which also has some default (unreasonable variables and algorithm selection and sample design), and the application of these models are often complicated, thus it can not be used practically well.Chapter 3 introduces relevant theories of ANN and GA. It mainly discusses the design of ANN basic structure, principle and application flow of GA. And it combines GA and back-propagation artificial neural network (BP-ANN), constructs the structure of forecasting for stock index based on ANN and GA according to characters of stock index variation in China.Chapter 4 analyzes stock index forecasting. On the basis of analyzing the difficulties in stock index forecasting, it pays attention to advantages of building forecasting model by combining BP-ANN and GA in this chapter. It not only avoids the over-fitting of single BP-ANN, but also improves the astringency of single GA, by using GA to optimize the weight of BP network. In other words, it can use the nonlinearity approximation of the model based on ANN and GA to forecast stock index.Chapter 5 constructs the model of forecasting for stock index based on ANN and GA. Firstly, it designs the structure of BP-ANN and initial weight; Secondly, it trains the BP network by using the training sample for first time, and using the testing sample to test the forecasting capacity of BP network; thirdly, it optimizes BP network's weight by using GA, and fixes the optimized weight and threshold, and trains the optimized BP network again, and, it constructs the model of forecasting for stock index based on ANN and GA after test its forecasting capacity again finally.Chapter 6 is forecasting for China Securities Index 300 (CSI 300). It chooses CSD00 with positive market representation, and uses the models constructed in fifth chapter to forecast CSI300. The result shows that the error of the model is small and the precision reaches the target, thus it can be used actual stock index forecasting and helps to investors in their investing decision.Chapter 7 is conclusion and expectation. The conclusion includes: firstly, it can improves the forecasting precision of BP network after optimized by GA; secondly, it needs to estimate the number of input units and hidden units rationally; thirdly, it also needs to choose the controls parameters of GA rationally. But, it may need to do more research on how to choose the fitness function of GA, and how to choose the input variables and sample size in the future.The innovation of this thesis is constructing the model of forecasting for stock index based on ANN and GA, and uses this model to forecast CSI300.