The Application Of Artificial Neural Networks In The Shanghai Stock Market Trends Predict - Comparative Analysis And Time Series Forecasting

It has been over ten years since the stock market of our country was founded. Now the stock market has become one of the most important capital markets. Today, prediction of stock market has become a hot topic because that every part of the public pays more and more attention to the prospect of stock market. In the study field, the conventional statistics still is the main elements to forecast the trend of stock market. This dissertation elementarily studied the use of the artificial neural network on prediction of stock market, and compared it with time series models. After the study and comparison, we come to some valuable conclusions:First, we can come to some accurate predictions based on the artificial neural network. Artificial neural network not only can imitate the technical analysis and fundamental analysis, but also can consider technical factors and fundamental factors at the same time to analyze the prospect of stock market. The prediction of stock market based on the artificial neural network has almost the same precision as that based on time series models.Second, the process of modeling an artificial neural network has very high automaticity, for it can accomplish many inner processes automatically. Consequently, the artificial neural network can be integrated in software, and be offered to average investors.Third, however, the automaticity of artificial neural network can also be a weakness on the view from another angle. The automaticity of artificial neural network causes the inner process of network invisible, so that it is very difficult to analyze the inner influence and process, and to analyze the economic significance. This reduces the signification of artificial neural network on directing policy.Fourth, there are also many deficiencies in artificial neural network. It hasn't a perfect test system or an accepted criterion. For instance, artificial neural network uses sensitivity analysis to choose input variables, for it hasn't a much effectivemethod as the Granger causality Test in econometrics. Another example is that it hasn't an accepted standard on the distribution of samples.Fifth, the lack of uniform standard in identification and establishment of artificial neural network causes the instability of prediction. The results can be affected not only by the changes on the distribution of samples, but also by the changes on the time and method of training. This is different with econometric models.Therefore, artificial neural network should be consummated further. We can combine it with other theories. For instance, we can use the Granger causality test in econometric to help artificial neural network choose input variables, and use the principal factor analysis to help decrease the number of input variables, so that we can enhance the efficiency of artificial neural network.