RDP 2015-06: Credit Losses at Australian Banks: 1980–2013 2. Measuring Credit Losses

2.1 Accounting

Credit losses arise from borrower default. Banks value loans as the (discounted) value of the future repayments; as these fail to eventuate (or evidence emerges that they will not eventuate) accounting standards require banks to recognise the fall in the value of these loan assets.[2] Such losses are one component of a bank's overall profitability, so they affect capital and, in extreme cases, solvency.

This direct relationship with profitability makes the flow of credit losses the relevant quantity when attempting to understand the effect credit risk has on banks. Stocks of troubled assets, such as non-performing or impaired assets, are a frequently used alternative (see Gizycki (2001) and Salas and Saurina (2002)). But these assets only affect bank profitability and solvency through credit losses, and the relationship between these measures varies over time, and with loan type and bank behaviour. Most importantly, there is not a monotonic relationship between the measures. If one bank displays a higher level of non-performing assets than another bank during a year, this does not necessarily mean that the first bank experienced a higher level of credit losses during the year.[3]

In terms of accounting, there are three different ways in which banks can deal with credit losses:

  1. The most common way is to create an individual provision, a liability, equal in value to the expected credit loss.[4] This liability, and the loan (an asset) from which the credit loss stems, are intended to have a net value equal to the amount the bank expects to recover. The creation of the individual provision is funded through an expense item on the bank's statement of profit and loss. Provisions are generally raised immediately after a bank receives evidence that it is likely to incur a credit loss. The final stage of the credit loss process – the removal (or write-off) of the loan and accompanying provision from a bank's balance sheet – often occurs well after this, once the amount of the loss is known with more certainty. This final step does not affect profitability, as the credit loss has already been incurred through the creation of the provision. If the quantum of the loss increases from that expected when the provision was raised, the amount of the individual provision can be increased, or the additional loss can be written-off directly to the profit and loss (see below).
  2. Individual provisions are mainly used for credit losses on larger loans. For smaller loans, where it is not economic to assess the likely size of a credit loss at the loan level, banks raise collective provisions. These can be raised to cover, for example, expected credit losses on all small personal loans more than 90 days in arrears. The amount of the collective provision is usually based on past experience – for example, the average credit loss incurred on a particular category of loans in the past. Collective provisions are also used to cover likely future losses on the currently healthy portion of banks' loan books. Historically, this component of collective provisions has fluctuated in line with banks' expectations around future credit losses, creating a wedge between losses banks have accounted for through their profit and loss statement, and those that have actually occurred.[5]
  3. Credit losses can also be dealt with without raising provisions; they can be written-off directly to the profit and loss. This method can be used for loans where there is no prospect of recovering a significant portion of the loan amount, or if the quantum of the credit loss is immediately reasonably certain. It is often also used for lending where a high loss rate is expected and built into the interest margin (credit card lending is one example). Unlike where provisions have previously been raised, this type of write-off affects profitability.

This is a simplified overview of the accounting items that are needed to capture a bank's credit losses. Appendix A provides a complete list of the items needed to accurately measure credit losses. It also provides a detailed example of the accounting for a credit loss on a single hypothetical loan.

Most banks have, over time, used a combination of the above three methods to account for credit losses. I combine credit losses accounted for using the three methods above into three different aggregate measures of the overall credit losses incurred by a bank (the dashed lines in Figure 1). These three aggregate measures differ in the stage at which they capture credit losses accounted for under the three methods. Each has advantages and disadvantages:

Figure 1: Accounting for Credit Losses
  • Charge for bad and doubtful debts (CBDD) – This is the aggregate credit risk expense item that appears on banks' profit and loss statements. It is the net impact of credit risk on profitability, so is the most economically relevant measure. The weakness of this measure is that, as it captures the net charge to the profit and loss to fund collective provisions, it fluctuates in line with a bank's expectations around future credit losses on currently healthy loans.
  • Current losses (CL) – This measure modifies the CBDD in an attempt to capture only losses that have actually occurred. Instead of using the net charge to profit and loss to fund collective provisions, it includes only write-offs against these provisions. This change is intended to exclude provisions raised to cover likely future losses on currently healthy loans.
  • Net write-offs (NWO) – This captures write-offs against all provisions, as well as write-offs made directly to the profit and loss. It is less subjective than the CBDD and CL, because write-offs are usually made significantly after initial loss recognition, when the quantum of credit losses is more certain. But this long lag means that NWO lag the CBDD and thus the economic impact of losses on banks.

These dollar measures need to be scaled to be comparable across years. Following standard practice, I look at losses during each year as a share of loans outstanding at the start of the year (Foos, Norden and Weber 2010). This prevents mechanical exaggeration of loss rates by loan losses during a year lowering measured lending at the end of a year. I call the three resulting ratios the ‘bad debt ratio’ (CBDD/net lending), ‘current loss ratio’ and ‘net write-off ratio’, and denote them by (respectively) BDR, CLR and NWOR. The CLR is the focus of my analysis, as it provides a compromise between timeliness of economic impact and accuracy in measuring actual losses.[6]

2.2 Data

The main credit loss dataset used in this paper was largely compiled from banks' annual financial reports. This (public) source is the only one that provides credit losses right back to 1980 – collection of credit loss data by prudential regulators started later.[7] The dataset only covers whole-of-bank credit losses, rather than credit losses broken down by portfolio, as these are only available for a broad range of banks from 2008. The data is for parent banks, rather than consolidated groups. Parent bank data exclude lending by overseas subsidiaries, allowing me to concentrate on the credit risk from Australian loans.[8] Banks were chosen for the sample by looking at the ten largest banks at five-year intervals from 1980 to 2010; attempts were made to gather data over the full period for any bank that was in the top ten for any sub-period. The resulting dataset covers 26 banks, and is slanted towards larger banks (see Appendix B for a list of included banks). It is unbalanced, as banks enter, exit, and merge. On average, it covers around 80 per cent of bank lending in Australia over the sample period (Figure 2).

Figure 2: Sample Coverage

Where useful, I employ other credit risk data. For example, I use the portfolio-level (i.e. business, housing and personal) loss rates that the major banks have published in their (publicly available) Pillar 3 reports since 2008. I also make use of regulatory datasets, such as the long-run non-performing assets data (available from June 1990) and the quarterly credit loss data (available from 2003).

The major non-credit risk dataset used in this paper is the micro data underlying the measures of aggregate credit provided by financial institutions in Australia. This provides the share of each bank's lending that is devoted to business, housing, and personal lending at each point in time.


This discussion focuses on loans valued at amortised cost. This is the category of bank assets that has been most severely affected by credit risk over recent decades in Australia. Assets valued in different ways, for example at fair value, and assets that are not loans, for example derivative contracts, can also be affected by credit risk. [2]

This may be because the first bank's non-performing assets were residential mortgages, which are normally more highly collateralised than other types of lending. Alternatively, the second bank may simply have written off its non-performing assets more quickly than the first bank, in an attempt to display a healthier loan book to investors and ratings agencies. [3]

Under the Australian equivalents to the International Financial Reporting Standards (IFRS), ‘provisions’ are liabilities used to lower the value of loan assets to their recoverable value. In the credit losses literature, this term is commonly used for the flow of credit losses (an expense), reflecting its meaning under US Generally Accepted Accounting Principles. Prior to the adoption of IFRS in Australia, individual provisions were called specific provisions, and collective provisions were called general provisions. [4]

The adoption of IFRS in 2006 constrained the extent to which Australian banks could raise collective provisions to cover future loan losses. However, they still do this to some extent. This is dealt with in Appendix B. [5]

Current losses are the measure used for Australian banks by Esho and Liaw (2002), though these authors calculate and present it quite differently. [6]

I use regulatory data to measure the credit losses of three (unlisted) banks from 2002 onwards. [7]

This choice also excludes lending by banks' domestic finance company and merchant bank subsidiaries, many of which experienced substantial credit losses during the early 1990s. [8]