Novel information in estimating loss given default in Brazil

The Basel Accord regulates risk and capital requirements to ensure that a bank holds capital proportional to the exposed risk of its lending practices. Basel II allows banks to develop their own empirical models based on historical data for probability of default (PD), loss given default (LGD) and exposure at default (EAD). Brazil was among the first emerging market countries to release a timetable for the implementation of the Basel II Accord and aimed to apply it uniformly to all Brazilian financial institutions from 2005 to 2011. Within this context, the necessity arises of conducting research that could assist the financial institutions in improving the accuracy of their models. This thesis has three objectives. The first is to develop a macro-economic model to predict the behaviour of the aggregate delinquency in Brazilian consumer loans. The model consists in testing co-integrating relationships and then estimating a short run error correction model. The results based on monthly data from 2000 to 2012 show that the delinquency rate is particularly sensitive to shocks on GDP and to the variation of workers’ income. The analysis then shifts to micro or account level to model the behaviour of borrowers and certain novel types of information that can be used for prediction. Second, customers fail to make loan repayments for a number of reasons, ranging from simple forgetfulness to deliberate attempts. For this reason, the second objective is to investigate the reasons for default and to explore ways of incorporating these variables into Recovery Rate (RR = 1 - LGD) models, since the standard approach overlooks real reasons for default and uses proxies for them such as marital status and length of employment. Customers who failed to repay their loans were interviewed in order to discover the causes for this failure. In addition, the interviews included questions aimed to measure the customer’s personality traits and their financial knowledge in relation to the reasons for default. The empirical results show that the variables proposed in this study, namely, reason for missing payment, financial knowledge and risk taken, improve the prediction of the recovery rate. Thirdly, it is known that recovery depends on the debt collection process and on the different options or actions that collection departments can take. Yet there is practically no literature exploring the impact of the lender’s collection actions on RR/LGD. This work fills this gap by investigating the role of different collection actions at the loan-level for a retail credit product, and by estimating LGD models using Panel Data regressions