國內有關於通膨之文獻，多著重於使用特定變數進行預測。然而特定變數的選擇，多半是根據傳統經濟理論認定與通膨相關的變數，侷限了研究者發現其他重要變數的可能性。目前國內鮮有文獻探討高維度資料集於通膨預測之應用，因此本文參照 Forni et al. (2005); Giannone et al. (2004); Stock and Watson (2002a, 2002b, 2012b)，嘗試由上而下 (top-down)，利用過擴散指數預測法 (diffusion index forecasting) 預測台灣通膨。 本文蒐集 2000 年至 2021 年間，近 100 個對於台灣通膨具有潛在影響力變數，探討不同降維方法所萃取之潛在因子 (latent factor) 對模型預測力的影響，發現使用監督式的降維方法有助於提升模型整體預測能力。本文採納 Stock and Watson (2002b) 之建議，事先將變數分為11 大類後再進行預測。發現在分類前預測力最好的偏分量迴歸 (PQR) 於分類後模型之預測力有了更進一步提升。本文接著探討預測過程中的關鍵變數、不同的時空背景下 11 大類別相對重要性之消長，最後建構通膨 (縮) 預警模型，做為台灣央行制定貨幣政策時的參考依據。 ;Past literature on Taiwan’s inflation forecasting mostly confines to only several theory-specific variables, which limits the possibility of roles played by other potential important variables. In view of the superior forecasts from the diffusion index method via incorporating large dimension information via PCA as in Forni et al. (2005) ; Giannone et al. (2004) ; Stock and Watson (2002a, 2002b, 2012b), this paper extends the framework to allow for linear/nonlinear, supervised/unsupervised dimensionality reduction methods. We collected nearly 100 potential variables, from the period of 2000 to 2021, in order to extract the hidden common factors and for inflation forecasting. Among the examined 4 approaches, our results indicate that the supervised partial quantile regression (PQR) dominate the other 3 approaches in anticipating inflation. Once we further divide variables into 11 categories and extract category-specific factors for the subsequent forecasting as in Stock and Watson (2002b), we found that the predictability of PQR became even better. Based on these results, we not only investigate the importance of each category toward inflation across time, but also establish an early warning model for monitoring the arrival of radical inflation/deflation and adjusting for policy interventions.
