本文欲利用方向距離函數 (DDF) 來驗證中東歐國家銀行業之效率。不同的國家之銀行業者會由於不同的文化、資源稟賦和環境而採取不同的經營模式。因此，本文藉由共同邊界方向距離函數，其允許我們在跨國間不同的技術下，得以估計出銀行的技術效率並加以比較。使用方向距離函數，得以讓我們的模型中納入非意欲產出，此外，其亦允許銀行廠商同時增加產出與縮減投入和非意欲產出，相較於傳統模型，方向距離函數屬於較有彈性之模型。重要的是，不良貸款被視為是貸款過程中之副產品，其可能會降低銀行的利潤和績效。因此，為了減少不良貸款之產生，銀行管理者必須花費額外的成本，以確保借貸者是否有良好的信用，此舉亦可能影響銀行的績效。 本文試圖發展新的共同邊界隨機方向距離函數，其不同於Battese et al. (2004) 所提出的方法，該方法是屬於線性規畫法。此數理規畫法是屬於確定邊界，其無法針對有興趣之參數估計出該標準誤，因此，無法做有效地統計推論。因為本文提出的新共同邊界方向距離函數是隨機的，所以參數之標準誤可以被估計，其亦允許我們建造信賴區間和假設檢定。此外，共同邊界方向距離函數之無效率項可以被進一步設定成環境變數之函數，即Battese and Coelli (1995)所提出之模型。 / This study plans to employ directional technology distance function (DDF) to examine bank efficiency of Central-East European countries. Banks from different countries choose to operate under distinct technologies due to their differences in culture, endowments, and environments. A metafrontier directional distance function will be established, which allows for calculating comparable technical efficiencies for banks under different technologies relative to the potential technology available to the industry across nations. The salient feature of the DDF is its ability to include undesirable outputs into the model. In addition, it allows for a bank to simultaneously expand outputs and contract inputs, as well as undesirable outputs. It is important to note that the non-performing loans (NPL) can be regarded as a by-product of various loans granted, which lowers a bank’s profitability and performance. To reduce the occurrence of NPL bank managers have to spend extra costs to confirm whether the potential applicants for loans have good credit before granting loans to them. This may also affect the bank’s performance. This study attempts to develop a new metafrontier DDF in the context of the stochastic frontier approach, which differs from the one proposed by Battese et al. (2004) who suggest the use of a linear and/or a quadratic programming technique. The mathematical programming technique is known as deterministic, which is unable to estimate the standard errors for the parameters of interest. Hence, no statistical inference can be made. As our new metafrontier DDF is stochastic, the standard errors of the parameters are estimable, which permits establishing confidence intervals and hypotheses testing for the parameters. Moreover, the inefficiency term of the metafrontier DDF can be further specified as a function of several environmental variables of the form proposed by Battese and Coelli (1995).