總體變數之領先、同時與落後性質之認定與指標構成項目之選取－－LARS 方法的運用

AN EVALUATION OF COMPONENT SERIES OF BUSINESS INDICATORS: AN APPLICATION OF THE LARSMETHOD

本文利用Efron et al. （2004）所介紹的least angle regression （以下簡稱LARS）方法，試圖認定總體數列的領先、同時與落後性質，並依此結果從中挑選出適當的變數編製景氣指標。在實證分析中， 我們先從眾多總體資訊中選出103 筆與景氣循環相關的資料， 再透過LARS 方式進行兩階段的篩選。在第一階段的篩選過程中， 我們主要是將103 筆資料區分成領先、同時與落後群組（groups）， 而第二階段的目的則是從各個群組中挑選出重要且對基準循環數列有較佳解釋力的變數。我們發現， LARS所篩選出的結果與經建會所認定的結果大致相同， 特別是同時指標的構成項目， 兩造所認定的結果均一致。若就第二階段的篩選結果來分析， 我們發現不同樣本期間變數的排序順位可能會改變； 但也有某些變數不受樣本期間改變的影響， 顯示這些變數與基準循環數列之間的關聯性具有穩健的特性。最後， 我們也嘗試依據第二階段的篩選結果來編製新的領先指標， 我們發現新的指標在近期的領先效果比現行公佈的領先指標還要好。

As an alternative to themethod currently used by the Council for Economic Planning and Development (CEPD), we apply the least angle regression (LARS) method proposed by Efron et al. (2004) to select themacroeconomic series used to construct composite business indicators. We use a two-step approach. First, we obtain 103 macroeconomic series and assign each of them to one of three subgroups – leading, coincident and lagging – based on their ability to predict the reference cycle series. Second, within each subgroup, we rank the assigned series based on their ability to predict the reference cycle series and construct a composite business indicator based on these rankings. Our results suggest that themacroeconomic series selected via the LARSmethod tend to agreewith those currently used by the CEPD. However, the power of the selected macroeconomic series to predict reference cycle series varies with the size of the sample selected. Moreover, the composite leading indicator constructed from the macroeconomic series selected via the LARS method predicts turning points of the business cyclemore effectively than the existing CEPD leading indicator.