過去的文獻經常採用只允許位置參數變化的MSAR模型,直接以估計結果解釋景氣循環,但普遍發現此類模型對台灣曾經發生過的多次景氣循環不具長期的解釋能力。本文說明,這個實證現象有可能來自於「模型設定」與「未區分長短期狀態變換」這兩個計量問題。我們藉由區別檢驗決定出適當的MSAR模型,用以解釋長期狀態變換效果,並以此狀態轉折取代饒秀華等(2001)先驗判斷的樣本分期。之後,再更進一步定義出只具短期效果的「分段標準化」成長率序列,用以重新估計傳統的MSAR模型。本文的實證顯示,對台灣實質GNP年成長率的長期原始序列而言,區別檢定非常顯著地拒絕只允許位置參數變化的MSAR模型;相反地,接受了同時具有位置變化與波動變遷之狀態意義的MSAR模型。台灣過去的經濟背景說明,計量檢定接受的MSAR模型所解釋的是成長率在隨著不同經濟發展階段而異的長期位置變換與波動變遷,故非傳統文獻所欲直接解釋的景氣循環現象。而以分段標準化成長率序列估計傳統MSAR模型所得出的結果,支持適當的處理前述的兩個計量問題,是使傳統MSAR模型恢復對景氣循環之長期解釋能力的重要關鍵。文中亦與部分相關文獻進行過詳細的比較分析。
The location-MSAR models, which impose the Markov chain on the location parameters but not other parameters, are widely used in the empirical studies of business cycles. These simple models are found unable to interpret the business cycles of Taiwan in the long-term, however. By the use of Taiwan real GNP growth rate series, we illustrate that this problem is likely due to two econometric deficiencies of such models: the misspecification of models and the inability to discriminate between the long-term and short-term regime-switching behavior involved in the raw growth rate series. (The long-term effect is driven by economic development and the short-term effect is determined by business cycles.) To circumvent these two deficiencies, we first conduct model specification tests to determine suitable long-term MSAR models for the raw series. Then, we use these long-term models to define a ”segment-standardized” growth rate series that includes only the short-term effect. This segment-standardization may serve as an alternate to the sample-splitting method of Rau et al. (2001, Academia Economic Papers). Finally, we utilize this standardized series to re-estimate a simple location-MSAR model. The empirical study shows that our method is useful for recovering the performance of this simple model on explaining the business cycles of Taiwan in the long-term.