大中華貨幣單一化與總體經濟指標之關聯性-以模糊類神經和ARIMAX-GARCH模型分析

The Study of the Relationship between Chinese Currency Unification and Macroeocnomic Factors：The Analysis of Fuzzy Neural Network and ARIMAX-GARCH Model

鄭誌逸(Chih-Yi Cheng)

指導教授 ： 陳若暉

中原大學/商學院/企業管理學系/碩士(2008年)

https://doi.org/10.6840/cycu200800397

摘要

本研究參考歐元通貨籃之建構，以台灣、香港和中國大陸之每人生產毛額、出口總值和淨外匯存底加權，虛擬出1992年3月至2007年6月之大中華單一中心匯率（CCU），分別採取SDR、EURO和修正匯率模式，使用灰色關聯分析、模糊類神經和ARIMA-GARCH模型，欲從進口成長率、出口成長率、貿易餘額、工業生產指數、外匯存底、利率、通貨膨脹率、貨幣供給成長率、股價指數和國內生產毛額中，找出影響CCU關鍵因素和預測走勢，並找出何種模式績效較佳。
經灰色關聯分析結果發現，CCU分別受工業生產指數、國內生產毛額、股價指數、外匯存底等變數影響最深，再將篩選之前、後五變數以模糊類神經驗證預測績效優劣。研究結果發現，灰色關聯分析篩選之前五變數預測績效皆優於後五變數，且修正後匯率模式績效較SDR和EURO佳。在模糊類神經預測績效上，以修正台灣前五變數（國內生產毛額、股價指數、工業生產指數、外匯存底和貨幣供給成長率）預測績效最佳；若為ARIMAX-GARCH模型預測績效，以修正SDR建置CCU以中國大陸前五變數（工業生產指數、利率、出口成長率、股價指數和通貨膨脹率）預測績效最高；若為修正EURO建置，則以香港前五變數（工業生產指數、國內生產毛額、外匯存底、貿易餘額和股價指數）績效最佳。此外，綜合ARIMAX-GARCH分析結果，以工業生產指數、貨幣供給成長率和貿易變數影響最大且深遠。整體而言，ARIMAX-GARCH預測績效優於模糊類神經，而各經濟變數之效果則因CCU建置方式的不同而有所改變。實證結果期能在未來華元成立之際，提供兩岸三地政府和後續學者參考。

關鍵字

灰關聯分析 ； 類神經網路 ； ARIMAX-GARCH ； 模糊理論 ； 大中華單一貨幣

並列摘要

Refer to the structure of the EURO currency basket, the central rate of Chinese Currency（CCU） Unit was simulated from 1992/3 to the 2007/6 by the weights based on the GDP per capital, the exports, and the net foreign reserve of the Taiwan, Hong Kong, and China. By using each of the Special Drawing Rights（SDR） EURO, and modified-SDR method, the study utilized the analysis of grey relation, fuzzy neural network, and ARIMAX-GARCH model to find out the key factors from import growth rate, export growth rate, trade balance, industrial productive index, foreign reserve, interest rate,, inflation rate, monet supply growth rate, stock index and gross domestic product affecting CCU.
According the grey relational analysis by dividing the better five and the worse five factors, the CCU was affected by the industrial productive index, GDP, stock price index, and net foreign reserve. And the fuzzy neutral was used to test forecasting performance for each currency. The research found that the better five variables performed well comparing with the worse five. And the modified-SDR is better than SDR and EURO, except the Hong Kong for EURO method. As analyzing the CCU by fuzzy neutral network, the forecasting performance of Taiwan’s better five variables (gross domestic product, stock index, and industrial productive index, foreign reserve and money supply growth rate) is superior to other groups. If the CCU is simulated by modified-SDR method utilizing the ARIMAX-GARCH model, the forecasting performance of China’s better five variables (industrial productive index, interest rate, export growth rate, stock index, and inflation rate) is the best. If CCU is built by EURO, and the forecasting performance of Hong Kong’s better five variables (industrial productive index, gross domestic product, foreign reserve, trade balance and stock index) has best performance. According to the ARIMAX-GARCH model, the industry productive index, money supply growth rate, and trade factors significantly affect the CCU and its dynamic effect. Generally, the forecasting performance of ARIMAX-GARCH model is better than the neutral network, and the macroeconomic factors effect will be different if the CCU is built by the different way. Finally, this study will provide some valueable suggestions to policy makes and researchers for Taiwan, Hong Kong, and China if the CCU is created in the future.

並列關鍵字

ARIMAX-GARCH ； Grey Relationship Ananysis ； Fuzzy ； Neural Network ； Chinese Currency Unit

參考文獻

曹添旺、朱美麗（1993），股價與匯率的互動關係—臺灣的實證研究，行政院國科會專題研究計畫成果報告 (NSC81-0301-H001-029)。

葉柏村（2002），運用類神經網路預測匯率–以歐元為例，中原大學企業管理學系研究所碩士論文。

鄭婉秀、吳佩珊、陳君達、陳玉瓏（2005），「貨幣政策、匯率與股價關連性之探討：GARCH-IRF模型之應用」，朝陽商管評論，第4卷第2期，頁73-92。

王毓敏（1998），「台灣地區股票市場及外匯市場間報酬與波動性外溢效果」，台北銀行月刊，第28卷第12期，頁159-171。

王毓敏、徐守德（1998），「台灣地區股票市場與外匯市場間報酬與波動性外溢效果之研究」，中國財務學會年會暨學術研討會論文集，頁601-612。