即時台灣經濟成長率預估MIDAS模型與PCA整合應用

Abstract

金融數據係體現未來整體經濟環境之領先指標，故可藉其捕捉未來整體國家之經濟成長情形；此些金融變數雖具即時且快速之優點，然其公布頻率係以月為單位，且易受到市場氣氛波動而起伏。如欲使用此些金融變數以預測未來經濟成長變化，首先遭遇之問題即為如何將月資料型態之金融變數與季資料型態之經濟成長率相互結合。故本文擬利用另一種型態之回歸模型:即混合數據抽樣模型(Mixing Data Sampling, MIDAS)，進而解決資料頻率不對性之問題、輔以主成分分析方式(Principal components analysis, PCA)降低資料維度、從而提升分析品質，並將其應用於實證研究中，利用景氣對策信號及台灣加權股價指數報酬率對台灣實質經濟成長率進行預測。最終結果顯示運用MIDAS模型結合PCA預估經濟成長率，較使用簡單時間序列模型及平滑操作輔以簡單回歸分析方式(Ordinary Least Squares, OLS)更具效率，不僅大幅提升其解釋能力，亦可顯著改善長期預測誤差(Mean Square Error, MSE)。 綜言之，MIDAS模型可有效解決資料頻率不對稱之問題，並萃取出之解釋變數時間參數效果，PCA可降低資料維度，解決共線性問題，對於提升應變數之整體解釋與預估能力與有明確影響與貢獻。

Financial data is a leading indicator that reflects the overall economic environment in the future. It has the advantages of real-time and fast speed.However, the frequency of publication is in months.This article uses a mixed data sampling model (Mixing Data Sampling, MIDAS),Solve the problem of data frequency asymmetry,Supplemented by principal component analysis (PCA) to reduce data dimensions,To improve the quality of analysis,And apply it to empirical research,Forecast the real economic growth rate of Taiwan by using the Monitoring indicator and the return rate of Taiwan Capitalization Weighted Stock Index.The final result shows that the MIDAS model combined with PCA is used to estimate the economic growth rate,It is more efficient than using simple time series models and smoothing operations supplemented by simple regression analysis (Ordinary Least Squares, OLS),It not only greatly improves its interpretation ability, but also significantly improves the long-term prediction error (Mean Square Error, MSE). In summary,The MIDAS model can effectively solve the problem of data frequency asymmetry, and extract the effect of explanatory variable time parameters.PCA can reduce the data dimension and solve the problem of collinearity.It has a clear influence and contribution to improve the overall interpretation and estimation ability of dependent variable.