| In this article, we research whether basic economy environment, industry factor and company statements have influence to prediction of company stock price in short period with artificial neural network technology. At first, we choose Thomson listed in New York Stock Exchange (NYSE) as our research target. Then, we want to find a fit artificial neural network that can be used to forecast correctly stock price after one week. We apply three layers back propagation network. Finally, wecompare forecasting error rate with two sets of input variables. These input variables are: 1. the whole economy factor, industry factor, company statements and real time stock market variable; 2. only real time stock market variable.Result from the research of this article shows: 1. Predict accuracy of Levenberg-Marquardt training algorithm is good. But predict accuracy of the algorithm between Levenberg-Marquardt and conjugate gradient is almost same; 2> There's no relationship between hidden units number and predict accuracy; 3> In this case, predict accuracy of logarithm sigmod transfer function is better than tan-sigmod transfer function; 4> The whole economy factor, industry factor, company statements are all statistic data during certain period. They only become obvious different after some times. Furthermore, single stock is not more sensitive than the whole stock market. So we input the variable of economy environment when we predict single stock price in long run. If we need to forecast single stock price in a week, that's enough that we only input real time stock market variable. Then we input real time stock market variable, and run 13 circles containing 96 group of data training and stock price forecasting of next week. We found that the average of prediction accuracy is 3.4% and absolute error value is between USD0.27~1.94.The research of this article should be creative. To predict single stock price in one week, it is no use to input the variables of economy environment.
|
PDF Full Text Request