亞洲國家近年來積極推動亞洲區域整合,若區域經濟發展程度相近的國家,能建立共同通貨區或單一貨幣,可降低交易成本,促進經濟成長,降低區域性的貨幣危機發生可能與國家匯率的異常波動。 貨幣危機通常為過快的金融開放、經濟的泡沫化、外債過高、過度寬鬆的貨幣政策和金融監管能力,使得短期資本有炒作的機會。以有效的資訊與總體經濟指標架構,建立早期預警模式(Early Warning System),評估國家經濟脆弱程度,有助於緩減貨幣危機發生的可能性。 本研究以1992年至2009年期間,採用Logit、Probit、支援向量機(SVM)與貝氏網路,研究亞洲單一貨幣中心匯率(ACU)與貨幣危機的關係。探討何種為較佳外匯壓力指標(EMP)與貨幣預警模型(EWS)及重要影響的變數。 實證結果指出:在所有的外匯壓力門檻下,實行亞洲單一貨幣中心匯率(ACU)制度較各國對美元匯率制度,可以降低貨幣危機發生的可能。在貨幣危機發生前,政府當局可以藉由外匯存底、外債占外匯存底比例等總體經濟指標,進行適度的調整,降低貨幣危機發生的可能性。在模型選取,以Logit篩選變數後使用貝氏網路模型精準值最高,外匯壓力指標STV型一與型二的錯誤率最低,對於貨幣危機預警模型具有較佳的預警能力。
The advantages of establishing a common currency area or common currency basket peg within the regional countries which have similar levels of economic development are lower trade cost, lower probability of suffering currency crisis, lower unusual vibration of exchange rate and better economic growth. Therefore, in recent years Asian countries are persistently implementing the Asia regional integration. The reasons for the currency crisis are financial over-liberalization, economic bubble, the over-leverage of government debt, the ineffective monetary policy and financial supervision capacity, making short-term capital become hot money. With effective information collection and analysis architecture and suitable macroeconomic economic indicators, we can assess the vulnerability of national economies and establish the early warning models to reduce the likelihood of suffering currency crisis. In this paper we adopt the Logit model, Probit model, Support Vector Machines (SVM) model and Bayesian networks model to analyze the relationship between the Asian Currency Unit (ACU) and the currency crisis from 1992 to 2009. By so doing, we try to find out the better exchange market pressure (EMP) indicators, early warning system (EWS) models and the key factors. The empirical results show: firstly, the governments which implement ACU central rate can reduce the probability of currency crisis under all exchange market pressure by an appropriate adjustment of important macroeconomic variables such as foreign reserves and the ratio of government debt to reserves. Secondly, in the model selection, the combination of Logit model and Bayesian networks model gives the best prediction power than any other models. The type I and type II errors for Exchange market pressure index STV(Sachs, J., Tornell, A. and Velasco, A. (1996)) method have the minimum rate and therefore have better early warning capability.