| 摘要: | 為驗證現行景氣領先指標構成項目是否能有效預測景氣循環，本文針對行政院經濟建設委員會發佈的七項景氣領先指標構成項目進行實證分析。 為避免景氣領先指標構成項目受到短期偶發事件，或無法預測景氣循環之因素的影響，本文使用由黃鍔所提出的Hilbert-Huang Transform (HHT)方法中的經驗模態分析法 (Empirical Mode Decomposition)，將最短期波動從變數中抽離出來，以構成新的景氣領先指標內容。 實證結果發現，在落後期數較短VAR模型下，七個領先指標構成項目當中，只有製造業存貨量指數與核發建照面積 (住宅、商、辦及工業倉儲)兩項領先指標構成項目在Granger因果關係檢定上，屬於有效的領先指標。在落後期數較長VAR模型下，七項領先指標構成項目在Granger因果關係檢定的表現很差，均無法做為景氣循環領先指標。 本文修正後的七個領先指標構成項目，在落後期數較短VAR模型下，外銷訂單指數、製造業存貨量指數及股價指數等領先指標構成項目在Granger因果關係檢定上，屬於有效的領先指標，在落後期數較長VAR模型下，本文修正後的七個領先指標構成項目，在Granger因果關係檢定表現良好，均能當作景氣循環的領先指標。 本文最後使用樣本外預測力來進行修正後領先指標構成項目的預測評估，研究結果支持本文建議的新景氣領先指標構成項目，其對景氣循環的預測準確性較佳。 In this paper, we examine whether the current business leading indicator components can forecast business cycle effectively. This study proceeds with empirical analysis about seven business leading indicator components and business cycle published by CEPD (Council for Economic Planning and Development). In order to prevent business leading indicator components from the influence of short-term incidents and factors which can not forecast business cycle, this paper uses the Empirical Mode Decomposition, the primary method of Hilbert-Huang Transform developed by Huang, extracts the shortest-term fluctuations from the variables, and then compose new business leading indicator components. As a result, we find that there are only two leading indicator components, Index of Producer’s Inventory and Building Permit (including housing, mercantile, business and service, industrial warehousing), in empirical analysis under shorter lag periods VAR model, are effective leading indicators in Granger causality test. While under longer lag periods VAR model, all of the seven leading indicator components behave poor in Granger causality test. However, under shorter lag periods VAR model, there are three leading indicator components (including Index of Export Orders, Index of Producer’s Inventory, and Stock Price Index) are effective leading indicators in Granger causality test within seven modified leading indicator components recommended by this paper. Under longer lag periods VAR model, all of the seven leading indicator components behave well in Granger causality test, which can be business leading indicators. Finally, this paper uses out-of-sample test to evaluate the modified leading indicator components. This result proves that new business leading indicator components suggested by this paper forecast business cycle more accurately. |
