針對我國行政院主計總處在每季季中以「定基法」所衡量的實質國内生產毛額及其構成項目的當季成長率預測，此研究從衡量相對變異程度的預測能力指標、變動方向預測正確率、效率性、不偏性以及對應的多變量模型作了完整的預測績效評析。和文獻上的已知研究不同，我們除了分別探討構成項目的成長率預測外，也首度採用從2004年11月至2014年8月所公佈的40個不同時點的「時份資料」進行分析。分析結果顯示，實質國内生產毛額及其構成項目的當季成長率預測在當季實際變動方向上的誤判率平均高達20％以上，前期的景氣狀態與各項目的預測誤差實現值訊息也未被完全有效使用，因此仍有改善預測績效的空間。由於主計總處所採用的預測模型在目前採行的「連鎖法」與原「定基法」間應仍具有一定的相關與連貫性，此研究關於預測變數誤差間及與景氣狀態之間的關聯性多面向且完整的分析結果，應具有作爲進一步改善其預測模型的參考價值。

In this study, we systematically evaluate the forecasting performance of the Directorate General of Budget, Accounting and Statistics (DGBAS) of the Executive Yuan in Taiwan. We focus on the current-quarter growth rate forecasts of real GDP and its components (consumer spending, investment, government spending, imports, and exports), which are measured based on a fixed-based method. For the sample period from November 2004 to August 2014, the quarterly data including forecasts and realizations are collected from 40 different vintages. Some important findings are as follows. First, the examination of direction-of-change forecasts indicates that over 20% of DGBAS current-quarter forecasts predicted wrong directions for changes in real GDP and its components from the previous quarter to the target one. Besides, most forecast errors depended on their past values and on the previous economic state; that information was helpful to improve the forecasting performance. Although DGBAS currently constructs the real values via the chain-linked method instead of a fixed-based one, its reference forecasting models and methods are not supposed to be revised too much. Therefore, the results of this paper are expected to be helpful to improve DGBAS's models and methods for forecasting and real GDP and its components.