| Summary: | 碩士 === 東吳大學 === 經濟學系 === 91 === Abstract
The Taiwanese economic fundamental is based on the international finance Investment. Recent year, many researches use Artificial Neural Network to apply in types of stock market, currency exchange and company bankruptcy, etc. As the result, this method provides quiet accurate for prediction.
In this study, I first pick up 22 important variables from Taiwanese economic index, foreign index and technical index to be inputs. We use both the Traditional Multiple Regression (MR) and Artificial Intelligence methods to create four type prediction models that are including three processing steps.
First step is to use Genetic Algorithm Neural Network (GANN) and Stepwise Regression (SR) methods separately and to find the best input variables to enter. Second step is to use Genetic Algorithm (GA) to decide the best structure of Neural Network. Third step is based on the above 2 methods and to use Genetic Algorithm Back Propagation Networks (GABPN) and Stepwise Regression model in out-of-sample forecast. As the result, through Wilcoxon tests of performance, we found the best forecast model is combining the SR with GABPN method. These methods provide investors to make right decision on the foreign currency exchange prediction.
Through the Wilcoxon tests of performance we have discovered:
1. The best model is combining SR with GABPN method.
2. SR model performs well than the GANN method.
3. GABPN prediction ability is better than MR method.
This paper shows that the GABPN approach is effective not only for qualitative analysis but also for quantitative analysis of foreign exchange prediction in Taiwan.
Key words: Foreign Exchange, Multiple Regression, Genetic Algorithm, Artificial Intelligence, Stepwise Regression, Artificial Neural Network
|
|---|
