本文利用Stock與Watson所提出的動態因子分析法預測日本的通貨膨脹，透過動態因子分析法從數量龐大的變數中萃取出少數幾個因子 (Factor) 並且將其作為預測迴歸的解釋變數，用以建構預測。本文的主要研究目的是想探討是否CPI的細項資料在估計因子時扮演著相當重要的角色，因此透過這些因子作通膨的預測時，能有效的提升預測表現，更準確的預測通貨膨脹率。本文使用46個總體變數及103個CPI的細項資料，用以估計因子。資料期間的選擇為1986年至2009年。 由實證結果可以發現，有考慮CPI細項所估得之因子模型在向前一期、兩期與四期的預測表現中皆優於自我迴歸模型。而沒有考慮CPI細項所估得之因子模型則在向前一期、兩期與四期的預測表現中有惡化的現象產生。本文之實證結果說明了利用包含在CPI細項中的資訊確實有助於提升通貨膨脹的預測能力。 This study uses dynamic factor analysis proposed by Stock and Watson (2002a) to forecast inflation in Japan. This method summarizes large amounts of economic information by a few estimated factors, and uses them as predictors to construct the forecasts. The objective is to investigate whether CPI subcomponents play an important role in estimating factors to obtain better predictors to improve forecasting performance. The dataset composed of 46 macroeconomic series and 103 CPI subcomponents from 1986 to 2009. The results indicate that the subcomponent-included factor model outperforms the benchmark autoregressive model at the one, two, and four quarter ahead horizons and the factor model, which does not include the disaggregated CPI components, resulting in a deterioration of forecasting performance at the one, two, and four quarter ahead horizons. The results support the argument that using information from CPI subcomponents contributes to substantial improvements in accurate inflation forecasting.
