Structural Models and Bank Credit Risk-- An Empirical Analysis

在既有的信用風險結構式模型實證研究中，基於資本結構之考量，研究樣本多數排除銀行業。因此，本研究首次以美國銀行業為對象，檢視四個著名結構型信用風險模型之信用風險預測能力，以瞭解其應用於銀行業信用風險評估之適用性。這四個模型包括Merton (M,1974)， Longstaff and Schwartz (LS，1995)， Leland and Toft (LT，1996)，and Collin-Dufresne and Goldstein (CDG，2001)。本文研究發現，對於資產預期報酬率低之銀行，CDG模型有較佳之預測能力；資產預期報酬率高之銀行，則適用LS模型；至於支付比率較高的銀行，則因為Leland and Toft (1996)模型有考慮支付比率於模型中，因此會有較佳之結果。Most existing empirical studies on structural form credit models exclude bank industry because of its high leveraged capital structure. In this study, among the first few studies, we examine the prediction performance of four famous structural form models in bank credit risk. The models are Merton (M, 1974), Longstaff and Schwartz (LS, 1995), Leland and Toft (LT, 1996) and Collin-Dufresne and Goldstein (CDG, 2001). We use the bank data from the US to do this investigation. We find that the CDG model has the best prediction ability for the banks with low mean asset returns and LS model predicts better for the banks with high mean asset returns. In addition, since LT model takes into consideration a firm’s payout policy, it performs well in banks with high payout ratios.CONTENTS I.Introduction- 1 - II.Data- 4 - III.The Models- 7 - 3.1 Merton Model (the M model)- 7 - 3.2 Leland and Toft Model (the LT model)- 7 - 3.3 Longstaff and Schwartz Model (the LS model)- 10 - 3.4 Colline-Dufresne and Goldstein Model (the CDG model)- 11 - IV.Parameters Estimation- 15 - 4.1 Merton Model (the M model)- 15 - 4.2 Leland and Toft Model (the LT model)- 16 - 4.3 Longstaff and Schwartz Model (the LS model)- 17 - 4.4 Colline-Dufresne and Goldstein Model (the CDG model)- 18 - V.Empirical Results- 19 - VI.Conclusion- 33 - References- 34