Data related to the credit loss experience (CLE) of financial institutions are widely used for a wide variety of research yet literature typically omits a discussion on the construction and selection of suitable CLE proxies.
Based on a comprehensive dataset of potential CLE proxies retrieved from original financial reports of 32 Australasian banks (1980 - 2005), this article analyses properties and interactions of potential proxies through time.
It is found that some commonly used proxies correlate rather poorly and bad debt provisioning is only partially matched by subsequent write-offs. Moreover, even derecognitions in the form of write-offs do not seem to be definite as up to 25% of loans written-off are ultimately recovered.
The results highlight the need for caution in the use of (and possible adjustments to) historical credit loss data proxies commonly used by researchers.