FORECASTING INFLATION IN TAIWAN: MODEL COMPARISON AND ANALYSIS

• Main Authors: Feng-Chien Chung, 鍾豐鍵
• Other Authors: Ching-Yi Chiang
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/25972182985272275279

碩士 === 銘傳大學 === 經濟學系碩士班 === 98 === This paper investigates forecasts of Taiwan inflation at the 1- and 4-quarter horizon. We compare four prototype models, including autoregressive model, random walk model (Atkeson and Ohanian, AO, 2001), generalized Phillips curve model and static factor model (Stock and Watson, 2002) with 110 economic indicators, applied to forecasting four different inflation measures (CPI, core CPI, GDP deflator, PCE deflator), using quarterly data from 1981:1 – 2009:4. The simulated out-of-sample forecasting methodology (recursive and rolling simulation) is adopted to evaluate the model performance. We found that the AO forecast outperforms the AR forecast and the Phillips curve forecast at the 4-quarter horizon in most cases, this is consistent with the findings in AO (2001). However, forecasts using generalized Phillips curve based on macroeconomic variables generally have been more accurate than univariate models at the 1-quarter horizon. Another empirical result is, forecasting inflation using static factor model based on a large number of predictors does not improve upon univariate AR model.