Stock Forecasting Model under the Effect of Presidential Election

• Main Authors: Wen-Kai Li, 李文凱
• Other Authors: Hsu-Che Wu
• Format: Others
• Language: zh-TW
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/5rkmr9

碩士 === 國立中正大學 === 會計與資訊科技研究所 === 106 === This study wants to find a suitable time point for investment and investment targets, in view of a short period of time after the presidential election stock prices rose higher, so this study determines the period and the event, expect to find good screening of investment targets screening model. In this study, we found all the variables that have been proved to be effective from the past studies as the input variables of this study. After the selection of variables, we make neural network prediction. In addition, this study also tried to use event research to confirm that the presidential election has a significant impact on the stock market. After the event research method used to confirm that the presidential election has a significant impact on the stock market. The study conducted during the election period, hoping to obtain better forecasting ability. Use the 15 days prior to the election as a restriction to do the collection of variables as much as possible that used to predict the election 15 days after the stock price. The results of this study are mainly due to the good or bad predictors of the variables. The predictions result of stepwise regression in the three screening methods is bad. The remaining principal component analysis and random forest two methods, the principal component analysis to 80% of the main components to make better than 90% of the main components of the results are slightly worse than the original model, did not enhance the effect of predicting ability. The random forest is better than the principal component analysis and even beyond the unscreened variable model, which is the best prediction model in this study.