The Optimal Models of the Gold Volatility

• Main Authors: Keng-Yi Chen, 陳耕毅
• Other Authors: Chin-Chun Wu
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/46254587985901946397

碩士 === 義守大學 === 資訊管理學系碩士班 === 98 === This study use gray system, regression analysis and regression-type back-propagation neural network to predict the Bank of Taiwan in the gold price volatility, the regression-type back-propagation neural network use regression analysis of variables selected, then the independent variables results into the neural network to predict the gold price volatility and reduce energy consumption. Finally, MAE and MSE compare their performance. In the gray system model was found to establish the number of selected samples of the model prediction accuracy will have great influence in the regression analysis, excluding the seven outliers, stepwise regression analysis, the sample forecast performance of good, In the back-propagation neural network, the discovery of its predictive power for the sample regression analysis of the sample is larger than the accuracy of forecasting, back-propagation neural network play in the samples to search for powerful features. In three different models of forecasting performance, the regression analysis found that the best performance, followed by regression and back-propagation neural networks, the worst for the gray system. Select from the variable point of view, this study found that the gold book Ask the Bank of Taiwan to carry out two stepwise regression analysis, the New York NYMEX futures closing price of gold in recent months, and New York NYMEX futures closing price of silver in recent months as both opt-in model dependent variable, so do not rule out the selling price of the Bank of Taiwan and two gold book are strongly dependent variables. In this study, the sample of the forecast period be shortened to the first half of 2007, performance analysis results into a better regression back-propagation neural networks, back-propagation neural networks regression may not be suitable for long range forecasting.