Assets Write-off Prediction with Support Vector Machine Model

• Main Authors: Szu-Yin Wu, 吳思音
• Other Authors: Ching-Lung Chen
• Format: Others
• Language: zh-TW
• Published: 2009
• Online Access: http://ndltd.ncl.edu.tw/handle/08876347942544395220

碩士 === 朝陽科技大學 === 會計所 === 97 === Most of current researches about Statement of Financial Accounting Standards No. 35 (SFAS 35) focused on the motivations of assets write-off claims from companies, the effects on financial or managerial performance, the critical elements of losses, the incentives of accounting reports, the influences on revenue quality, etc. However, there hasn’t been any investigation on how the companies made their decisions on whether to declare write-offs and how to predict reasonable amount for declared ones. The present study enriches the reasonable predictions of decision and amount of assets write-off by the comparison of results from different predicting models. First, the classification performances on decision of write-off are studied with respect to Logit and SVM models. Second, reasonable magnitude of assets write-off prediction models for future reference is provided by the study of both linear regression model and Support Vector Regression model (SVR) for each write-off type respectively. In order to overcome the heterogeneity of write-off samples, the ensemble bagging approach is integrated into present prediction models. The data from year 2005 to 2007 are used for analysis. The empirical results show: the SVM model is mildly accurate to Logit model on the write-off decision prediction; the magnitude of assets write-off prediction performance by SVR model is better than traditional linear regression model under certain conditions; and the mean square errors are decreased year by year. The present study also displays that prediction performances of models relate with different write-off types, especially, the long-term investment write-off prediction is more feasible than others. Besides, the prediction performance could be more obvious by grouping samples with company growth options.