Credit risk prediction using support vector machines

Abstract

Support vector machines, Credit risk prediction, Default classification, Estimation of probabilities of default, Training sample size, Accounting data, C14, G33,

Predicting SME loan delinquencies during recession using accounting data and SME characteristics: The case of Greece

“…SVMs are derived from statistical learning theory and follow a structural risk minimization principle (Boser et al ., ; Cortes and Vapnik, ). To obtain classifiers, these powerful learning systems merge efficient algorithms from the optimization theory and elements of statistical learning theory (Chen & Li, ; Trustorff et al, ).…”

“…They found that the probabilistic neural network model showed the highest degree of association with the data and the lowest total misclassification cost. Finally, Trustorff et al (), using a data set of more than 31,000 German companies from 2000 to 2006, found that SVMs significantly outperform LR models, particularly under the condition of small training samples and high variance of the input data. To the best of our knowledge, there is no study that compares simultaneously the predictive capacity of various credit‐scoring models (i.e.…”

“…Financial institutions mostly used the LR model because of its robustness and transparency (Louzada et al, 2012). 2 Concerning the artificial credit-scoring models, we identify four main methods: artificial neural networks (West, 2000;Zopounidis & Doumpos, 2002), DTs (Jiang, Li, & Cai, 2009), case-based reasoning (Shin & Han, 2001), and SVMs (Trustorff, Konrad, & Leker, 2011). In recent years, neural networks and SVMs are widely used in credit scoring because of the possible non-linear relationship between independent variables.…”

“…optimization theory and elements of statistical learning theory (Chen & Li, 2010;Trustorff et al, 2011).…”

Model Selection for Financial Distress Prediction by Aggregating TOPSIS and PROMETHEE Rankings

“…However, some assumptions of the statistical procedures, such as the multivariate normality for explanatory variables, are frequently violated in practice, thus making them theoretically invalid for finite samples [10]. During the last decade, efforts have focused on the deployment of data mining techniques such as artificial neural networks [3,5,14], support vector machines [4,20,28] and classifier ensembles [19,29,31,32], to design and implement solutions for financial risk prediction. In contrast with statistical models, data mining methods do not assume any specific prior knowledge, but automatically extract information from the examples available.…”