Modelling credit risk of small and medium sized enterprises using transactional, accounting and market variables
MetadataShow full item record
This thesis comprehensively explores the credit risk of Small and Medium Sized Enterprises (SMEs) using transactional characteristics, financial variables and market information. It contributes SMEs credit risk modelling by exploring a range of soft features, such as management capability, industrial sectors, entity type, etc. It is the first study of investigating the concept of management capability through quantitative transactional information. Firstly, models are proposed to assess the credit risk of SMEs by identifying the significant factors. To fulfill this, two studies are carried out. In the first study, logistic regression, survival analysis and ordinal regression are used to model the relationship between transformed financial variables and probability of default. Both the traditional AUROC measure and Hand Statistic are used to evaluate the performances of the models, and they both indicate that logistic regression on weights of evidence transformed data yields the best prediction. Survival model takes an extra element of the time dimension into consideration. Ordinal regression performs poorly possibly due to impact of sample sizes. The factors appeared with highest frequencies are ratios associated with liquidity and growth. The other study predicts the credit risk (‘good’ ‘bad’ and ‘indeterminate’) of the SMEs using transactional characteristics. 35000 SMEs are clustered by different clustering algorithms. It is notably found that most ‘indeterminate’ observations are clustered with ‘bad’ observation, which is different from industry habit of combining ‘indeterminate’ and ‘good’. Logistic regression performs better than ordinal regression according to AUROC measure. In addition, some key points raised in focus group interview with bank managers are seen in the modelling process as significant variables, such as sector belonging to, entity type, region/location, time associated with bank, and account conduct. Secondly, the informational bases of two major models, which are accounting based credit scoring models and Merton type models, are explored to figure out aspects which affect SMEs’ credit risk. 33 financial variables covering nine financial categories are considered. It employs other modelling frameworks rather than the often-used linear regression, which are linear regression with interactions and the Cox proportional hazard model. It is found that weak relationship exists between these two models. The two major models capture different aspects of corporate information, it is suggested that a hybrid model, which incorporate both sources of information, might be considered to predict SMEs financial health. Thirdly, management capability of SMEs is elicited by applying principal component analysis to their transactional characteristics. Management capability is a qualitative idea, and its manifestation in quantitative variables was not explored in previous research. This study indicates some success in determining management capability. It is found that financial measure (credit turnover and debit turnover) and the performance measure (number of days in excess of the account) could be considered as reflecting management capability. Good management can identify trends at a very early stage and take action to mitigate the issue.