The event history analysis of debt portfolios concerns the repayment behaviours of
accounts under the management of a debt collection team. Typically, an account
can experience a series of events throughout the course of the debt recovery process,
such as payment commencement, missing a payment, settlement, etc. In this
thesis, we aim to provide a new perspective of modelling the evolution of the process
and evaluating the collection performance using various statistical techniques
in survival analysis and event history analysis.
In the first three chapters, we describe the consumer debt purchase and collection
industry, explore the data sets and review the related statistical methods to be used
in the thesis. In Chapter 4, we investigate the directly settled accounts, which form a
special group of accounts settled at the beginning of the recovery process. The time
until commencement of repayment, which is considered as an important indicator
for accounts’ repayment performance, is studied extensively in Chapter 5 using a
number of survival analysis techniques.
For accounts that have started to make monthly repayments, missing a payment
is an interesting event to investigate. As such events may occur more than once, a
specially structured multi-state model is explored in Chapter 6. Performance covariates
are also introduced to the modelling procedure to reflect the series of historical
events experienced. With an increased number of covariates, a tailored model
selection procedure is proposed to achieve improved interpretability of regression
results.