整合獨立成分分析特徵萃取技術與支援向量迴歸建構匯率預測模式

匯率長久以來在國際貿易上扮演重要角色，匯率波動嚴重影響企業的績效表現，且由於匯率是財務時間序列資料的一種，具有高頻率、非定態與混沌等性質，因此有關匯率預測的研究一直受到產業與學術界的重視。本研究整合獨立成分分析（independent component analysis，ICA）特徵萃取技術與支援向量迴歸（support vector regression，SVR）建構匯率預測模式，先利用ICA具有將混合訊號分離出個別獨立來源訊號之能力，將資料轉換到特徵空間，利用特徵空間中的獨立成分突顯出隱含在原始資料中的資訊以及降低原始資料特性的影響；接著針對每一個獨立成分，以SVR建立預測模型，得到獨立成分的預測值；最後，再將所求得之獨立成分預測值還原到原始空間得到最終的預測值。為驗證所提方法之有效性，本研究以新台幣兌美元匯率收盤價進行實證研究，並與整合ICA與自我迴歸（autoregressive，AR）模式、單純SVR模式、單純AR模式及隨機漫步（random walk）模式之預測結果進行比較。實證結果顯示，所提之方法在趨勢預測準確度的表現上較其他四個方法為佳，並且整合ICA的預測模式，其預測績效均優於沒有整合ICA的預測模式，代表透過ICA的特徵萃取能提升預測績效。

關鍵字

獨立成分分析；支援向量迴歸；特徵萃取；匯率預測；財務時間序列

並列摘要

Forecasting exchange rates is important to the investors and the government. In this study, we present the application of using ICA as a preprocessing tool before building SVR time series prediction model. In the proposed method, we first use ICA to transform the input space composed of original time series data into the feature space consisting of independent components representing underlying information of the original data. The hidden information of the original data could be discovered in these ICs. Then, the prediction models will be built by using SVR for each IC. Finally, the predicted value of each IC will be transformed back into the original space for time series prediction. The proposed model can improve the forecasting performance compared to the SVR model without using ICA as a preprocessing tool since the hidden information of the original time series data can be explored through ICA preprocessing. Experimental results on the forecasting of NTD/USD exchange rate have showed that the proposed method outperforms the integrated ICA and autoregressive (AR) model, single SVR model, single AR model and random walk model.

並列關鍵字

independent component analysis ； support vector regression ； feature extraction ； exchange rate forecasting ； financial time series

國際替代計量

整合獨立成分分析特徵萃取技術與支援向量迴歸建構匯率預測模式

全文下載

主題瀏覽