- 學位論文
以組合方法預測台灣通貨膨脹率
Forecasting Taiwan Inflation with Combination Methods
摘要
通貨膨脹率是總體經濟的重要指標之一，國內對通膨率預測的相關文獻多數以單純模型(naïve model)、貨幣模型、自我迴歸(autoregressive, AR)模型、菲利浦曲線模型(Phillips curve)為主。影響通膨率的變數很多，近年來外國學者指出組合方法在大量變數中擁有良好的預測性能，故本文以組合方法預測台灣通貨膨脹率。 本文參照Rapach and Strauss (2010)所建構的組合預測模型進行研究，以消費者物價指數(CPI)衡量通貨膨脹率，使用樣本期間涵蓋1982年1月至2020年1月(共457期月資料)的20個預測變數，在兩個不同樣本外評估期進行預測。實證結果顯示，依靠單一變數進行預測會因為時間推移而產生預測不穩定性，根據MSFE準則，多數時期表現甚至比AR基準模型差。組合方法在預測台灣通貨膨脹率能有效提升預測性能，其中又以均值法及折扣MSFE( )表現最為出色，主成分分析組合預測在多數時期表現不盡理想。
並列摘要
Inflation is one of the most important macroeconomic indices. We usually find lots of paper forecasting inflation rate with naïve model, AR model and Phillips curve model in domestic. Currently, foreign studies imply that combining forecast has better prediction performance among enormous variable. Therefore, this paper forecasts Taiwan inflation rate with combination methods. We employ combination forecast model proposed by Rapach and Strauss (2010), and use CPI to measure inflation rate. Our full sample is based on monthly data for 1982:01-2002:01. We consider several methods for combining forecasts with 20 individual ARDL model, and analyze two out-of-sample periods for the evaluation of inflation rate. Based on the MSFE metric, we find out single variable cannot perform well at all time. Our results show that combining methods based on mean and discount MSFE( ) produce the best forecasts of Taiwan inflation rate. Other combining methods, such as principal component often perform badly.