房屋貸款違約率預測模型之研究與實證-以C銀行為例

消費者房屋貸款業務傳統上來為我國銀行放款經營獲利的重要來源，由於房貸為抵押擔保授信，信用風險相較於信用卡或小額信貸而言相對低很多，但前提是銀行的內部稽核及內部控制制度要能落實，才能降低或預防作業風險發生而引起的信用風險。傳統上銀行預測逾期損失時並不會考量與經濟指標的關係，主要會以過去的經驗來預測未來損失。本研究之方法是參考C銀行對於美國主管機關FRB所制定CCAR規範下，為2012年壓力測試情境所建立之模型為基礎，利用C銀行所建立最終模型所選定之自變數，以主計處、內政部及信義房屋之公開資訊，重新建立預測模型，來加以驗證C銀行模型在不同的因變數下，其自變數是否適用於其它銀行或全體銀行。實證結果顯示：整體銀行違約率與信義房屋指數有非常顯著的關聯。另外，從β統計值可得知，整體銀行違約率與信義房屋指數為負相關影響、整體銀行違約率與美元對台幣匯率及五大銀行新承作放款利率為正相關影響。在模型回溯測試結果中，本研究所建立之模型其累計錯誤率優於C銀行模型，這也代表本研究模型其預測能力優於C銀行模型。且預測出來結果更為精確及可靠。

關鍵字

房屋貸款違約率預測模型

並列摘要

The consumer housing loan business has traditionally been an important source of profit for Taiwan local banks. Since the mortgage is a secured product, the credit risk is relatively low compared to Credit Cards or Personal Loans, it’s only subject to the internal audit and internal controls and depends on whether they are well-implemented to reduce or prevent operational risks. Usually banks do not consider the relationship of mortgages with economic indicators when predicting overdue losses. They mainly use past experiences to predict future losses. This study is based on the model established for the stress test scenario of 2012 under the CCAR regulations formulated by the US competent authority FRB for C consumer bank, with independent variables selected by the final model established by the C bank. The data are from DBGAS, MOI and Xinyi House. We estimate the model to verify whether the C bank model is applicable to other banks or all banks with different dependent variables. The empirical results show that the overall bank default rate has a very significant correlation with the Xinyi Housing Index. In addition, it can be seen from the results that the overall bank default rate has a negative correlation coefficient with the Xinyi Housing Index, and the overall bank default rate has a positive correlation with USD to Taiwan dollar exchange rate, and the five major banks' new loan lending rates. In the model back-testing results. Further, the cumulative error rate of the model established by this research is better than that of the C bank model, which also indicates that the predictive ability our model is better, which is more accurate and reliable.