Selection Effects of Lender and Borrower Choices on Risk Measurement, Management and Prudential Regulation

Empirical studies rely on features of observed data to gain new knowledge. However, data sets are often subject to selection bias. Recently, researchers have paid more attention to the impact of sample selection bias on outcome processes. In banking, selection is based on both lender and consumer choices and significantly affects outcomes of risk performance. This thesis presents three studies on the selection effects of lender and borrower choices on risk measurement, management and prudential regulation. The first study investigates the voluntary selection of banks to participate in a government guarantee scheme implemented during Global Financial Crisis 2008 – 2009 in Australia. First, we find strong empirical evidence that Australian banks that entered into the wholesale funding guarantee scheme offered by the Australian Government experienced a significant reduction in their funding costs and funding premiums. However, we also show that the subsequent removal of the guarantee scheme did not result in a full repricing of funding costs to normal levels. Further, the guarantee program did not cause excessive risk taking in terms of general bank risk, asset risk, or liquidity risk. Additionally, banks allocated the additional debt funding to residential mortgage loans coincided with a period of strong growth in house prices in Australia. The findings contribute to bank risk management on the liability side. The second study investigates the impact of prepayment selection on default likelihood. First, we document that prepayment and default are linked in a u-shaped pattern. Default risk is high for two distinct groups. The first group includes borrowers who have low prepayment risk as suggested by observed factors (unconditional effect). The second group includes borrowers who have high prepayment risk but did not refinance and remain in the sample post prepayment (selection effect). Second, the main cause for a high default rate in upturns is a selection effect, while that for high default in downturns is an unconditional effect. Third, industry practice models result in a significant error in default calibration. We propose a twostage model with a novel correction term to achieve a better default prediction than industry and literature models. The findings contribute to bank risk measurement and management on the lending side. The third study explores two approaches to predict prepayment risk and default risk in the multi-period setting: a life-cycle model and a forward model. Using data of US fixed-rate prime mortgages from 2000–2016, we find that both models perform equally well for prepayment and default predictions in the first three years, while the accuracy of both models decreases for longer periods. A life-cycle model provides a better calibration for later ages, while a forward model is more accurate in forecasts for periods beyond three years. We analyze the impact of prepayment selection on multi-period default predictions. We find that a default model, which controls for prepayment selection, provides more accurate default probabilities in long run than a model without selection. The mean absolute error can reduce by nearly 50% if controlling for prepayment selection. Our findings are useful for banks to assess more accurately mortgage risk over the loan lifetime and to implement loan loss provisioning changes under international accounting standards.