Estimating expected lifetime of revolving credit facilities in an IFRS 9 framework

Abstract (Swedish)

This paper sets out to estimate expected lifetime of revolving credit facilities
(e.g. credit card products) and is motivated by the introduction of the International
Financial Reporting Standard 9 (IFRS 9) and its requirements for loan
impairments. The reporting entity is required to estimate lifetime expected
credit losses for certain nancial instruments. In practice, maximum contractual
period for revolving credit facilities cannot be used in dening lifetime for
the facility and credit risk mitigation actions need to be considered. A data
set for a retail credit card portfolio was provided by a Nordic bank and for
the lifetime denition derived, a model based on a conditional Markov chain
was selected. Expected lifetime was estimated and an analytical expression for
expected lifetime of revolving credit facilities was derived and validated.

Popular Abstract

What period should credit losses be estimated over in IFRS 9?
When the standard on how to account for credit losses moves to an expected loss approach, there is a need to find out how far into the future to look for losses. This is known as the expected lifetime, and I have dived into interpreting how the IFRS 9 standard can be implemented in a model for expected lifetime for credit cards and similar instruments. I propose and validate a model, along with a methodology for estimation.
The Great Financial Crisis had a grand effect on most economies. Banks were assessed and it was recognized that in order to reduce effects of future downturns, provisions that were more forward-looking and higher were needed for credit losses. IFRS 9 is the answer to this (for reporting purposes) and the provisioning based on expected credit losses (rather than incurred) is the ingredient enabling this. A relevant consideration in this context is the expectation of lifetime for the instrument (e.g. a credit card) and since the phase of IFRS 9 concerning provisioning for loan losses is recently published, how to measure expected lifetime is not yet established.
For this purpose, I introduce a concept called “End of lifetime event”. This concept brings together a common form of credit risk model for credit card portfolios and similar (where risk rating is based on how late the borrower is on a payment) with events related to expectations of how this type of instrument is managed. The result is a list of End of lifetime events that will function as absorbing states in the Markov chain implemented to estimate lifetimes.
The Markov chain is a popular form of model, suitable for this task, both in its connection to common credit risk models and broad applicability. The selection methodology is based on testing if different candidate models possess the Markov property (that the distribution of future state depends on the past only in the present state), and finding a good trade-off between complexity (as low as possible) and accurate modelling of the data. For this, two statistical tests are used and a first-order Markov chain is selected, shown to be dependent on maximum historical risk of a borrower (which is reasonable, since this would explain a lot about a borrower beyond how late he or she currently is on a payment).
The parameters for all allowed transitions in this extended model are estimated based on data from a portfolio of credit cards provided by a bank, where transitions between internal risk ratings (or states) were used for estimating the transition probabilities in the model. It is convenient at this point that a Markov chain is used, since expected lifetime now corresponds to what is known as expected absorption time, a straight-forward tool in analyzing absorbing Markov chains.
There is a need to modify the way expected lifetime for this model is calculated, since the End of lifetime event for removal of what is known as the undrawn commitment component (the difference between the credit limit on e.g. a credit card and how much has been drawn) occurs with a delay from when the process reaches the corresponding absorbing state in the Markov chain. So, (expected) credit losses should be estimated over the period given by the following expression:

• author: Berglund, Jonas
• supervisor: Magnus Wiktorsson
• organization: Mathematical Statistics
• course: FMS820 20161
• year: 2016
• type: H2 - Master's Degree (Two Years)
• subject: Mathematics and Statistics
• language: English
• id: 8568367
• date added to LUP: 2016-01-26 11:46:26
• date last changed: 2016-01-26 11:47:09