Establishment of the Inference Model to Predict Foreclosure Bidding Price.

• Main Authors: Ting- Fa Hsueh, 薛廷法
• Other Authors: Min-Yuan Cheng
• Format: Others
• Language: zh-TW
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/4yw6y4

碩士 === 國立臺灣科技大學 === 營建工程系 === 105 === Based on the latest statistical data from the Financial Supervisory Commission (FSC), the continuous increase in the mortgage ratio (from 0.15% at the end of 2014 to 0.18% at the end of 2015, and to 0.19% in April 2016) is a problem that already affects the stability of the national financial system. In recent years, economic recession and fierce competition in the banking industry have caused the mortgage ratio and mortgage amount to increase constantly. According to the FSC, the total amount of nonperforming loans (NPLs) at the end of February 2017 was NTD76.6 billion, an increase of NTD6 billion from the previous month. The rising bad debt (foreclosures) in the country has caused financial institutions to be conservative in granting loans. To quickly recover NPLs, the finance industry has been issuing foreclosures on defaulters, causing a rapid increase in the number of foreclosure cases. Foreclosures involve auctioning off the mortgaged asset under the supervision of the judicial court, bank, or the Taiwan Financial Asset Service Corporation. This study discusses the transaction history of foreclosure cases and their registered prices. Data from the Real Estate Information Platform of the Ministry of the Interior are analyzed to determine which antecedents could affect the foreclosure price. SPSS is employed for a correlation analysis on the antecedents and output variable (auction price), and critical factors affecting the bidding price in the auction are objectively selected as the input parameters for the research model. Various artificial intelligence (AI) theories are adopted to train the case database, and various inference models are applied to test and obtain the predicted outcome value of the auctioned house bidding price. To validate the predictive accuracy of the AI models, the predicted outcomes are compared using the mean absolute percent error (MAPE), root mean square error (RMSE), mean absolute error (MAE), and linear correlation coefficient (R) to evaluate the error measure of the predictive accuracy. Finally, a reference index (RI) is derived to serve as the overall evaluation criterion to validate conclusion. The optimal forecast can be obtained using a symbiotic organisms search-least squares support vector machine (SOS-LSSVM).