透過您的圖書館登入
IP:54.160.243.44
  • 期刊

消費金融無擔保客戶違約協商後毀諾-資料探勘技術之應用

A Study on Default after Debt Negotiation of Unsecured Loans of Consumer Finance-Data Mining Approach

摘要


2005年底爆發卡債風暴後,銀行業受卡債風暴衝擊,信用卡、現金卡及個人信用貸款總體業績呈直線下滑;金管會與銀行公會便於2006年推行「消費金融無擔保債務協商機制」。「協商」原在協助銀行確認風險來源,並協助客戶還款;然「協商」程序也極耗成本及時間,且客戶申請協商後能夠正常履約者,所佔比率亦低。故本研究希望以資料探勘(Data Mining)建構模型,區別違約協商後毀諾的客戶與正常繳款客戶,以提升銀行成本與人力的使用效能。 研究中採用「決策樹」、「類神經」二種模式進行探討。結果發現在訓練集上,模式準確率上,以「類神經」的96.07%為最優,次為「決策樹」92.94%;而在測試值上以「決策樹」92.53%較佳,次為「類神經」92.47%。研究中亦以傳統的「區別分析」進行比較,結果發現其準確度分別為75.15%、75.14%,準確率與資料探勘技術差異甚大。以模糊矩陣得分值的結果進行比較,亦顯示「決策樹」與「類神經」的結果,遠較「區別分析」的得分值為佳。鑑於三種模式都選取了債務協商後的相關變數,如利率、失業率、正常繳款6個月及協商註記等;因此,銀行對於消費金融無擔保戶的信用狀態,應結合動態的後續管理與信用調整措施,才能有效管理授信風險。

並列摘要


At the end of year 2005, the banking sector were severely affected by the outbreak of credit card debt crisis, and the performance of most banks in credit cards, cash cards and personal credit loans were declined significantly. The Financial Supervisory Commission and the Bank Association then jointly promote and implement the ”Consumer Finance Unsecured Debt Negotiation Mechanism” in year 2006 to help banks identifying its source of risk and assisting default clients to repay their debt. However, Negotiation process is also extremely cost and time consuming, and in reality, the percentage of normally repay clients to the total negotiating clients is also relatively low. In order to help banks enhance the performance of human resource and decrease the operation cost, this study adopt Data Mining Techniques (DMT) to construct the prediction model to discriminate the client repay its debt normally from the client fail to fulfill its obligation after negotiation. In this study, ”Decision Tree Approach” (DTA) and ”Neural Network Approach” (NNA) were used to establish the prediction models and test the relative performance of different models. The results demonstrated that, on the relative accuracy rate of training set, 96.07% of accuracy rate of NNA is far better than the 92.94% of DTA; and for the testing set, 92.53% of accuracy rate of DTA is relatively higher than the 92.47% of NNA, both accuracy rates were over 90%. We also adopted traditional ”Multivariate Difference Analysis” (MDA) to compare the performance with two DMT approaches, the accuracy rate were only 75.15% and 75.14%, respectively, in both training set and testing set, significantly lower than the performance of DMT approaches. Scores of fuzzy matrix from the different models were also calculated; and the results suggested that the performance of DTA and NNA are superior to the performance of MDA model. The variables such as interest rate, unemployment rate, debt repays over 6 months and negotiation remark, etc., which selected in three models, also suggest that bank should apply more dynamic credit risk management and follow-up adjustment measures to monitor the credit status of unsecured client of consumer financing.

參考文獻


中央銀行(2009),金融統計月報,2009 年7月。
中華民國銀行公會,「無擔保債務協商相關統計資料」,http://www.ba.org.tw/ExpenditureFinanceDebtChk.aspx。
謝邦昌(2003),「簡介Data Mining」,http://cdms.stat.fju.edu.tw/。
Abinzano, I.,Muga, L.,Santamaria, R.(2010).Is the Momentum Effect Exclusive of High Default Risk Firms?.Journal of Finance and Accounting.39(147),445-470.
Agarwal, S.,Chomsisengphet, S.,Liu, C.(2010).The Importance of Adverse Selection in the Credit Card Market: Evidence from Randomized Trials of Credit Card Solicitations.Journal of money credit and banking.42(4),743-754.

被引用紀錄


蔡宜芳(2012)。資料探勘應用於天然資源類型基金連動性之研究〔碩士論文,淡江大學〕。華藝線上圖書館。https://doi.org/10.6846%2fTKU.2012.00538
陳威頤(2014)。應用Dagging集成式學習演算法改善分類準確度之研究〔碩士論文,國立虎尾科技大學〕。華藝線上圖書館。https://doi.org/10.6827%2fNFU.2014.00169
彭健瑜(2013)。應用集成式技術的推進法於資料分類準確度之研究〔碩士論文,國立虎尾科技大學〕。華藝線上圖書館。https://www.airitilibrary.com/Article/Detail?DocID=U0028-1408201314530700
鍾政旭(2017)。應用集成式學習於不孕症治療成功率預測之研究〔碩士論文,國立虎尾科技大學〕。華藝線上圖書館。https://www.airitilibrary.com/Article/Detail?DocID=U0028-2806201711132300

延伸閱讀