Forecasting Exchange Rate Using Support Vector Regression Base on Genetic Algorithm—A Case Study of EUR/USD

• Main Authors: Chen, Yu-Jung, 陳又榮
• Other Authors: Hung, Jui-Chung
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/hmh89v

碩士 === 臺北市立大學 === 資訊科學系碩士在職專班 === 107 === The foreign exchange market is composed of trading markets in different trading hours in various regions of the world. The overlapping area of trading hours between the London market and the New York market is the most frequent period of global foreign exchange transactions. The data is visualized and found in the overlapping areas of the two major trading hours. The exchange rate fluctuates a lot, so judging the overlapping time of the trading time will affect the exchange rate. Every weekend, the foreign exchange market may release unexpected news after the market closes, so the risk of holding foreign exchange before the market close is higher. After visualizing the data, it is observed that the exchange rate will have a significant rise and fall before and after the market close, so it is judged that the weekend will affect the exchange rate. In summary, this study explores the Support Vector Regression (SVR) model of forecasting exchange rate by selecting the euro-dollar as the research target, in addition to the general price-quantity attributes, and adding the factors of weekend and trading time overlap. The model parameters are adjusted by the Genetic Algorithm (GA) to obtain an optimized model. The experimental results show that the GA-SVR with the weekend and trading time overlap area attribute can reduce the Mean Absolute Percent Error (MAPE) by 9.4% compared with the GA-SVR using only the foreign exchange basic price attribute.