以動態因子模型預測台灣總體經濟變數

A Dynamic Factor Model for Forecasting Macroeconomic Variables in Taiwan

張雅涵(Ya-Han Chang)

指導教授 ： 陳俊志

國立臺北大學/社會科學學院/經濟學系/碩士(2012年)

摘要

本研究藉由Stock and Watson(2002) 的動態因子模型 (dynamic factor model) 以預測台灣的實質經濟活動及通貨膨脹，並比較動態因子模型、AR模型以及VAR模型對台灣總體經濟變數之預測能力是否有所差異。以台灣及中國總體經濟變數為研究對象，分析比較加入中國總體經濟變數後是否有較佳之預測效果。將資料型態分為指數資料和年增率資料，並檢定是否因不同之資料型態而產生不同之預測表現。蒐集790筆台灣及中國總體經濟變數，研究樣本為2001年1月至2010年12月的月資料，以2001年至2007年來建立預測模型，再以2008年至2010年來檢視模型的預測準確性。採用Harvey et al. (1997) 所提出之修正Diebold-Mariano檢定來評估預測準確性。結果顯示，利用動態因子模型預測台灣實質經濟活動和通貨膨脹僅有少數情況會顯著優於AR模型以及VAR模型。若模型中加入中國總體經濟變數，反而獲得較差之預測效果。表示加入中國總體經濟變數，無法得到較佳之預測表現，至於資料的運用上，本研究發現在短期的預測下，利用年增率資料之預測效果會優於指數資料。

關鍵字

擴散指標；動態因子模型；預測；主成份分析

並列摘要

This research will estimate Taiwan’s real economy activity and inflation rate by using Stock and Watson’s (2002) dynamic factor model. The further research will compare the difference between the estimating ability of dynamic factor model, of AR model and of VAR model when predicting the Taiwan’s macroeconomic variables. In the end, by studying Taiwan and China’s macroeconomic variables, this research will try to identify if including China’s macroeconomic variables will lead to better predicting result. The research will classify data into two categories, index data and annual growth rate data, and to test if different types of data will lead to different predicting result. The data will include 790 observations from Taiwan and China’s macroeconomic variables since 2001 to 2010. The research will establish the model based on data from 2001 to 2007 and use data from 2008 to 2010 to examine the accuracy of model. Using modified Diebold-Mariano test from Harvey et al (1997) to evaluate the model’s precision. According to the result, using dynamic factor model to estimate Taiwan’s real economy activity and inflation rate has powerful explanation than using AR model and VAR model under certain circumstance. If we consider China’s macroeconomic variables inside the model, we will get poorer estimating outcome. In other words, China’s macroeconomic variables will not provide better forecasting capability when we include them inside the model. For the data usage, this research find out that using annual growth rate has better estimating ability than index data in the short-term studying period.

並列關鍵字

Diffusion index ； Dynamic factor model ； Forecasting ； Principal components

參考文獻

徐士勛、管中閔、羅雅惠 (2005)，「以擴散指標為基礎之總體經濟預測」，台灣經濟預測與政策，頁1-28。

Artis, M., A. Banerjee and M. Marcellino (2005), “Factor Forecast for the UK,” Journal of Forecasting, 24, 279-298.

of Monetary Transmission,” The American Economic Review, 82, 901–921.

Brisson, M., B. Campbell and J. W. Galbraith (2003), “Forecasting Some Low-Predictability Time Series Using Diffusion Indices,” Journal of Forecasting, 22, 515-531.

Bruneau C., O. D. Bandt, A. Flageollet (2008), “Measuring Co-movements in the Euro Area Using a Nonstationary Factor Model,” Applied Economics Letters, Taylor and Francis Journals, vol. 15(10), 781-785.