本篇論文在panel data 的模型基礎下，將Hansen (1999)所提出的 panel threshold model與Arellano and Bond (1991)的panel dynamic model進行結合，成功推導出panel dynamic threshold model的模 型設定，本模型的特色在加入內生解釋變數落後項的動態解釋下， 進行門檻效果的檢定，以觀察實證資料是否存在顯著的門檻效果。 我們並進一步運用此panel dynamic threshold model 對於金融發 展與經濟成長間的互動關係進行實證分析，發現若不採用門檻變數 將實證資料進行區分，所獲得的實證結果將會受到金融發展程度較 高國家的資料所影響，而金融發展程度較低的國家資料卻因此喪失 其原本應具有的影響力。此外，若比較採用股市發展變數做為門檻 變數的實證結果和以銀行發展變數做為門檻變數的實證結果，則於 金融發展變數對於經濟成長影響的差異性上，以銀行發展變數做為 探討金融發展對於經濟成長影響性的門檻變數應是較為適合的選擇。 / The aim of this article is to investigate whether finance development has positive or negative relationship with the economic growth. We argue that the past studies do not reach consensus between finance development and the economic growth because the relationship may be indeed nonlinear. We plan to examine whether the effect of finance development on the growth depends on the threshold variable of bank development and stock market development. To examine this hypothesis, we develop a panel dynamic threshold model (PDTM) to test this hypothesis. The PDTM is a direct extension of the recent two research strands when using panel data. One is the dynamic panel model of Arellano and Bond (1991), who consider the lagged dependent variable into the panel data. The other is the non-dynamic panel threshold model of Hansen (1999), who takes threshold effect into the panel data model. Past studies, however, did not take dynamic and threshold into account simultaneously. Our PDTM thus incorporates the threshold into the dynamic panel data. Results demonstrate that the relationship between finance development and economic growth is nonlinear.

本篇論文在結構性時間序列模型基礎下，將中國大陸29省市自治區1978-2005年實質人均GDP，拆解出其長期趨勢變動軌跡中的水準值與斜率值，對照傳統上直接利用實質所得數據，以動態縱橫資料方法進行經濟成長條件收斂假說的檢定。本文特色在於加入潛在GDP長期趨勢項的水準值和斜率值，並利用內生解釋變數落後項動態分析。除可驗證隨著時間經過，中國相對貧窮省區是否終將逐漸趕上相對富有省份所得水準外，其次，根據GDP趨勢項一階與二階條件的收斂與否，可進而確認實質GDP收斂的本質。 我們發現，實質人均GDP收斂的本質關鍵在於潛在趨勢水準收斂，潛在GDP趨勢斜率的成長率將左右區域間實質所得收斂速度。大部分樣本中，擴大的Solow模型或考慮不同經濟開放程度因素下的內生成長模型，支持條件收斂假說，而後者設算出的收斂係數明顯較為低。此外，考慮採用Arellano and Bond(1991)的the first difference GMM估計式可能存在弱工具性問題(a weak instruments problem)，以Blundell and Bond(1998)發展出的the system GMM估計式，作為探討初始所得與經濟成長收斂的關係應是較為適合的方法。 / This research examines the economic growth conditional convergence hypothesis. Using the data of 29 provinces in Mainland China between 1978 and 2005, this study applied the structural time series model to deconstruct the provinces’ real GDP per capita into two parts - the level and the slope of trend movement. The characteristics of this paper are to include the level and the slope of trend of potential GDP and to consider the lagged dependent variables into the panel data. This study intends to validate whether the income level of relatively poor provinces will gradually catch up that of the relatively affluent provinces in Mainland China eventually. In addition, this study, based on the convergence or divergence in the first-order and second-order conditions of GDP tendency, will confirm the essence of the convergence in real GDP. The findings are that the essential key of the convergence in real GDP per capita is the convergence of the potential level of GDP. The growth of potential GDP tendency slope would affect the converging speed of real income in regions. The testing results of either the augmented Solow model or the endogenous growth model which considered different economic opening degrees both support the conditional convergence hypothesis in most sample sets, while the estimated convergence coefficients of the later are significantly lower than those of the former. In addition, considering the possible weak instruments problem in the first difference GMM estimator developed by Arellano and Bond (1991), the system GMM developed by Blundell and Bond (1998) should be a more suitable way to observe the relation between initial income level and economic growth convergence.