Schmidt與Witte(1989)提出的區分母體持續時間模型，將母體分成「是否」和「持續時間」?部分；受限於積分沒有封閉式的限制，假設這?部分的決定是互相獨?的。本文沿用Leung與Yu(2002、2007)的計?技巧，?用蒙地卡?模擬估計法解決積分的限制，鬆綁「是否」部分和「持續時間」部分的決定是互相獨?的假設，提出互相依賴的區分母體持續時間模型。在此建構下，獨?的區分母體持續時間模型是互相依賴的區分母體持續時間模型的巢氏模型。我們將這個模型應用到探討農會信用部發生擠兌的問題上，以AIC值、概似比檢定、t-檢定比較各種?同的持續時間模型，結果顯示互相依賴的區分母體持續時間模型優於其他模型。實證結果發現，融通資?比、逾放比?愈高者，發生擠兌機?愈高且愈早發生擠兌；?動性比?愈高者，發生擠兌機?愈低且愈晚發生擠兌。而平均存款??較高者，擠兌危險?愈高，與Schumacher(2000) 的結果一致。

Schmidt and Witte (1989) present a split population duration model in which the decisions of whether and when to start an event must be independent due to the absence of a closed-form result for the integration. This paper following Leung and Yu (2002, 2007) develops an interdependently split population duration model which allows the interdependence of the decisions of whether and when to start an event. We apply the Monte Carlo simulation to resolve the integration problem. We show the independent split population duration model is hence nested in our interdependently split population duration model. Our model is applied to investigate the bank runs events of the credit departments of farmers' institutions. The results show the superiority of our model to other duration models according to the criteria of AIC, likelihood ratio test, and t-ratio test. The empirical results show that both the ratio of borrowing capital to total capital and the overdue ratio are positively correlated with the probability of runs and the hazard rate to run while the liquidity ratio is negatively correlated with them. More importantly, our result shows banks with higher deposit interest rates are easier to trigger runs, which is consistent with the finding of Schumacher (2000).