RDP 2015-06: Credit Losses at Australian Banks: 1980–2013 Appendix B: Data

B.1 Credit Losses Dataset

The sample of banks for this dataset was selected by obtaining lists of the ten largest banks at each of 1980, 1985, 1990, 1995, 2000, 2005 and 2010, and then attempting to gather data for these banks for the longest possible period. This was not always possible. For example, data were unavailable for many large savings banks during the 1980s, as these were part of banking groups that only published data on their trading bank and consolidated results. The 26 included banks are listed in Table B1. Data for Bankwest, HSBC, and ING from 2002 onwards come for regulatory reports. Major bank time series are split around the incorporation of savings banks in the early 1990s. Time series for other banks are separated around major mergers. For example, Adelaide Bank and Bendigo Bank are separate from the merged Adelaide and Bendigo Bank. The sample includes some observations on banks that are very small in relation to the whole system (less than 1 per cent of total lending). These are excluded from most of the regressions in the paper.

I test for attrition bias in my unbalanced panel in the manner suggested by Wooldridge (2010). I add a dummy variable indicating exit in the next period to Model C. This is insignificant, in the cases in which I define the indicator to capture (i) all attrition from my sample (1 instance through becoming a non-bank asset management company (SBSA), 14 instances through merger, and 5 instances via missing data or falling below the 1 per cent of lending threshold); (ii) just failures and mergers (15 instances); and (iii) the attrition of only SBSA, SBV and Bankwest. These latter three cases are (arguably) the only cases of exit under stress in my sample.

Current loss rates appear to be stationary. The test statistic is less than the 1 per cent critical value in an Im-Pesaran-Shin test of the null hypothesis that credit loss rates are non-stationary. This is also the case in the version of this test that accounts for serial correlation. This result is in line with that of Pain (2003) for UK banks.

The credit losses of SBSA and SBV have been adjusted to remove the economic impact of support these banks received from their state government owners during the early 1990s. SBV received an indemnity from the Victorian government for losses on a proportion of its loan book. The value of this indemnity has been added to SBV's credit losses, as it removed the requirement for SBV to raise an equivalent amount of provisions. Direct payments and indemnities given to SBSA have been dealt with in the same way. A large portion of SBSA's troubled loans were transferred to the state government in 1994 and run-off over the subsequent decade. I have treated this transfer as a write-off by SBSA of these loans and the associated specific provisions.

Changes in accounting standards affect the comparability of credit loss data over time in two main ways. The change from the previous standards to the Australian equivalents to IFRS in the 2006 financial year had an effect on banks' collective provisions (which were referred to as general provisions before the shift). Compared to the previous standards, IFRS allows less scope for banks to hold provisions against expected future losses. It requires that provisions only be held against losses that have been ‘incurred’, in the sense that they are supported by objective evidence. The five largest banks in Australia reduced their general/collective provisions by around 20 per cent as a result of the change. This outflow from provisions, which was generally absorbed through an increase in shareholders' equity, has been removed from the charge for bad and doubtful debts in the long-run sample (it does not affect the other two measures of credit losses). Accounting advice and banks' public statements about their accounting policies indicate that, using various mechanisms, banks continue to raise collective provisions to cover likely future losses under IFRS.

Unlike the previous standards, IFRS requires banks to discount expected future recoveries under impaired loans, at the original interest rate applying to the loan. Thus, upon initial loss recognition, banks must raise higher dollar amounts of provisions. The extra provisions are run down over the period until recovery is made, and this flow is recognised in interest income. The overall effect of the change is that both credit losses and interest income are higher under IFRS than under the previous standards. Some banks provide the amount of the flow to interest income from provisions in their annual reports. Figure B1 presents the aggregate current losses for three banks that publish this data. The average difference between the adjusted and unadjusted ratios is less than 2 basis points.

B.2 Other Data