Research Discussion Paper

David Rodgers

RDP 2015-06

The contents of this publication shall not be reproduced, sold or distributed without the prior

consent of the Reserve Bank of Australia and, where applicable, the prior consent of the external

source concerned. Requests for consent should be sent to the Head of Information Department

at the email address shown above.

ISSN 1448-5109 (Online)

The Discussion Paper series is intended to make the results of the current economic research

within the Reserve Bank available to other economists. Its aim is to present preliminary results of

research so as to encourage discussion and comment. Views expressed in this paper are those

of the authors and not necessarily those of the Reserve Bank. Use of any results from this paper

should clearly attribute the work to the authors and not to the Reserve Bank of Australia.

Enquiries:

Phone: +61 2 9551 9830

Facsimile: +61 2 9551 8033

Email: rbainfo@rba.gov.au

Website: http://www.rba.gov.au

David Rodgers

Research Discussion Paper

2015-06

May 2015

Economic Research Department

Reserve Bank of Australia

I would like to thank Malcolm Edey, Luci Ellis, Christopher Kent, Gianni La Cava,

Vanessa Rayner, Matthew Read, John Simon, Robert Street, Grant Turner, and

James Vickery for helpful comments. The views expressed in this paper are those

of the author and do not necessarily reflect the views of the Reserve Bank of

Australia. The author is solely responsible for any errors.

Author: rodgersd at domain rba.gov.au

Media Office: rbainfo@rba.gov.au

i

Credit risk – the risk that borrowers will not repay their loans – is one of the main

risks that financial intermediaries face, and has been the underlying driver of most

systemic banking crises in advanced economies over recent decades. This paper

Australian banking system. It does so using a newly compiled dataset covering

bank-level credit losses over 1980 to 2013.

The Australian credit loss experience is dominated by two episodes: the very large

losses around the early 1990s recession and the losses during and after the global

financial crisis. The available data indicate the above-average losses during both

periods were on lending to businesses. Credit losses on housing loans during and

after the global financial crisis were minimal in Australia. Consistent with this, an

econometric panel-data model that properly accounts for portfolio composition

indicates that conditions in the business sector, rather than those in the household

sector, drove credit losses in Australia during the period studied. The data also

indicate that the very worst credit loss outcomes – including those that led to the

failure of several state government-owned banks in the early 1990s – were driven

by poor lending standards.

JEL Classification Numbers: G01, G21, G33

Keywords: banking, credit losses, lending standards

ii

1.

Introduction 1

2.

Measuring Credit Losses 4

2.1

Accounting 4

2.2

Data 8

3.

Descriptive Analysis 9

3.1

Credit Losses over Recent Decades 9

3.1.1

The early 1990s 11

3.1.2

The global financial crisis 19

3.2

Other Aspects of Credit Losses 22

3.2.1

Timing 22

3.2.2

Relationships between credit risk measures 23

4.

Econometric Analysis 24

4.1

Modelling Approach 25

4.2

Initial Models 26

4.3

Using Portfolio Composition 29

4.4

Lending Standards 36

5.

Summary and Policy Implications 41

Appendix A: More Accounting 44

Appendix B: Data 49

Appendix C: More Regressions 58

References 61

Credit risk – the risk that borrowers will not repay their loans – is one of the main

risks that financial intermediaries (such as banks) face. Credit risk has been the

underlying driver of most systemic banking crises in advanced economies over

materialises and borrowers fail to make repayments, banks are forced to recognise

the reduction in current and future cash inflows this represents. These ‘credit

losses’ reduce a bank’s profitability and can affect capital. In extreme cases, credit

losses can be large enough to reduce a bank’s capital ratio below regulatory

requirements or minimum levels at which other private sector entities are willing to

deal with a bank, so can cause banks to fail.

This paper explores the historical credit loss experience of the Australian banking

losses of larger Australian banks over 1980 to 2013. Portfolio-level credit loss data

– data that break losses down by type of lending (e.g. business, housing and

personal lending) – are available for a broad range of banks only from 2008

onwards, so this paper mainly uses total loan portfolio data.

This paper provides the first narrative account of banking system credit losses in

Australia that includes both the early 1990s and global financial crisis episodes.

Credit losses rise sharply during economic downturns, and are the main influence

on banking system profitability during such periods. The Australian credit loss

experience over the past three decades is dominated by two episodes: the very

large losses around the early 1990s recession and the losses during and after the

global financial crisis. During both episodes, banks’ credit losses appear to have

had a close relationship with changes in business sector conditions (such as

commercial property prices and the business sector’s interest burden). Losses

during the earlier period totalled around 8½ per cent of lending; losses during and

after the global financial crisis were around 2½ per cent of lending. The earlier

episode was a more severe downturn – business sector conditions declined to a

greater extent – but anecdotal evidence indicates that differences in lending

2

standards also played a role in the different levels of credit losses across these two

episodes.

As well as macroeconomic conditions and lending standards, portfolio composition

turns out to be important for credit losses. The very limited portfolio-level data

available for the early 1990s indicate losses during this episode were incurred

mainly on business lending. The better data available for the global financial crisis

episode make it clear that the elevated losses during this episode were almost

entirely incurred on business lending. Credit losses on housing loans during the

global financial crisis episode were minimal.

experience of Australian banks. Gizycki (2001) modelled bank-level measures

related to credit losses – impaired asset and return-on-asset ratios – over periods

that end in 1999. She found the interest burdens of the household and business

sectors, real credit growth, the real interest rate, the share of construction in GDP,

as well as commercial and residential property prices, to be the macro-level

conditions that influenced credit risk measures. This is informative, but the

dependent variables that Gizycki used do not have straightforward relationships

with credit losses, so these conclusions are not directly transferable to credit

consider some of the macro-level variables that Gizycki found to play key roles,

particularly financial variables. Esho and Liaw (2002) is the only paper on credit

losses in Australia that considers banks’ portfolio composition. These authors use

measures of portfolio composition from capital data as stand-alone explanatory

variables in a model for credit losses over 1991–2001. They found residential

mortgage lending to be indistinguishably risky from bank lending to governments,

and much less risky than lending to businesses and (non-housing) personal lending.

The econometric models of banks’ credit losses in this paper add to past Australian

work in several ways. As the new dataset covers 1980–2013, they include both the

early 1990s episode and the global financial crisis. They also consider a wide range

of macro-level variables as potential explanators of credit losses. Most importantly,

the main econometric model presented in this paper allows the effect of macro-

1 As an example, impaired assets are not a sufficient statistic for credit losses. See Section 2.1

below.

3

level variables on bank-level credit losses to vary depending upon each bank’s

portfolio composition. An example of the underlying intuition is that a fall in the

profitability of the business sector should lead to more credit losses (as a share of

each bank’s lending) for banks with a higher share of their portfolio devoted to

business lending. This variability is achieved using interactions between bank-level

portfolio composition variables and macro-level variables. This modelling strategy

exploits the panel nature of the newly compiled credit loss dataset, as well as that

of a regulatory dataset – the bank-level data underlying the aggregate measures of

business, housing and personal credit. Interaction variables are clearly suggested

by the available data on portfolio-level loss rates – which indicate losses on

different portfolios respond differently to macro-level conditions – but a systematic

approach of this type is novel in the literature. Pain (2003), Gerlach, Peng and

Shu (2005) and Glogowski (2008) allow interactions between the share of one

portfolio and a limited number of macro-level variables; I interact all macro-level

variables with portfolio shares.

This model with portfolio interactions explains bank-level credit losses over recent

decades reasonably well. The macro-level conditions that are statistically and

economically significant are business sector conditions. As these variables are

interacted with the shares of each bank’s portfolio made up by business lending,

this indicates business lending has been the main source of credit losses over recent

decades. Analogous interactions between household sector conditions and the

shares of banks’ portfolios made up by housing or personal lending are not

significant in the model. This result is consistent with the narrative account of

credit losses in Australia over this period.

The econometric models in this paper do not explain all of the variation in credit

losses. For example, they cannot explain why credit losses were so large at several

state government-owned banks during the early 1990s. This accords with the

omission of most of the variation in lending standards – roughly, the average

riskiness of a bank’s borrowers – from the models (quantitative measures that

comprehensively summarise bank lending standards are not available). It also

accords with anecdotal evidence that state government-owned banks had belowaverage

lending standards. An alternative measurement strategy, based on quantile

regressions, indicates that credit losses at banks with similar portfolios can respond

very differently to macro-level downturns, providing further support for the

importance of lending standards. While this evidence is not definitive, it suggests

4

that poor lending standards may have been the cause of the very worst credit loss

outcomes seen in Australia over recent decades.

As well as underlining the importance of lending standards, these findings have

practical implications for the conduct of financial stability monitoring and stress

testing. However, past performance does not necessarily predict future

performance. A point of caution in projecting forward past patterns of credit losses

is that the residential mortgage market has developed considerably since the early

1990s and now represents a much larger proportion of banks’ lending activity.

The next part of this paper, Section 2, sets out the way I measure credit losses.

Section 3 provides the narrative account of credit losses in Australia since 1980.

Section 4 contains the econometric analysis of credit losses. Section 5 summarises

my conclusions and discusses the implications for stress-testing practice and

broader financial stability policy.

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

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

2 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.

5

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

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

credit losses:

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

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

3 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.

4 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.

6

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

3. Credit losses can also be dealt with without raising provisions; they can be

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:

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.

5 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.

7

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.

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

Net write-offs

Charge for bad

and doubtful debts

Current losses

A credit loss on a loan

1

Raise individual

provision

[impact on profit]

Write-off loan and

individual

provision

[no impact on profit]

2

Raise collective

provision

[impact on profit]

Write-off loan and

collective

provision

[no impact on profit]

3

Write-off directly

to profit and loss

[impact on profit]

8

provides a compromise between timeliness of economic impact and accuracy in

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

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

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).

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.

6 Current losses are the measure used for Australian banks by Esho and Liaw (2002), though

these authors calculate and present it quite differently.

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

8 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.

9

As at September

Sources: Annual reports; APRA; RBA

The Australian bank credit loss experience since 1980 is dominated by the very

high rate of losses before, during, and after the early 1990s recession, as well as the

smaller losses during and after the global financial crisis (Figure 3). Losses around

the early 1980s recession were much lower. Relative to lending, credit losses

during the early 1990s far exceeded those incurred by banks during and after the

global financial crisis. Current losses between September 1989 and

September 1994 totalled around 8½ per cent of the average value of banks’ lending

during this period. In comparison, current losses during September 2007 to

September 2012 were equivalent to around 2½ per cent of average lending over

this period.

1983 1989 1995 2001 2007 2013

0

5

10

15

0

25

50

75

Share of bank lending

(RHS)

no %

10

Sample aggregate, as at September

Sources: ABS; Annual reports; APRA

The average sample aggregate CLR during 1980–2013 was 56 basis points. The

median, less influenced by the high levels in the early 1990s, was 34 basis points,

which was also the 2013 level.

Credit losses have strongly influenced the profitability of the Australian banking

system during the sample period. This can be seen by decomposing changes in

Credit losses were the largest contributor to the cycles in profitability during the

9 The data used in this exercise differ somewhat from the credit losses dataset: it is consolidated

data for Australian-owned banks only.

10 Decomposing changes in a ratio requires choices as to the ordering of the decomposition. I

have used the ordering that minimises the contribution of credit losses to the change.

-2

0

2

4

-1

0

1

2

%

2013

%

GDP growth

(LHS, year-average)

Bad debt

ratio

(RHS)

Current loss ratio

(RHS)

Net write-off ratio

(RHS)

1983 1989 1995 2001 2007

11

Australian-owned banks, consolidated data, as at September

Source: Annual reports

indicate that the bulk of credit losses were incurred on lending to businesses rather

than households. Two major banks published usable portfolio breakdowns of their

net write-offs in their annual reports for some or all of the early 1990s, but the

on non-construction housing loans were minimal (these fall within the ‘Real estate

– mortgage’ category). Loans to individuals for construction of housing probably

fell within the ‘Real estate – construction’ category, but this category also contains

lending for commercial property. Losses on this category were significant, but only

make up around 13 per cent of reported losses for these two banks. The key point

is that most of the losses reported by these two banks fall in the ‘Other business’

category. Losses on personal lending, such as credit cards and non-housing term

loans, were non-negligible, but appear to be less cyclical than losses on business

lending.

11 These two banks, CBA and NAB, accounted for 33 per cent of bank lending at

September 1991. CBA’s write-offs include those made within the State Bank of Victoria’s

loan book after its acquisition in November 1990.

0

10

20

0

10

20

1987 1992 1997 2002 2007 2012

-20

-10

0

10

-20

-10

0

10

Due to other components

Change in ROE

Due to credit losses

Total change

%

%

%

12

Note: (a) Mainly owner-occupied housing lending

Source: Annual reports

Portfolio-level data are available on all banks’ non-performing assets from

mid 1990 to mid 1994, and these support the conclusion that losses were incurred

to businesses that was non-performing far exceeded the share of their lending to

households (including non-mortgage personal lending) that was non-performing.

12 No similar data were collected before June 1990, and the regulatory collection from

September 1994 onwards did not have a portfolio breakdown. These rates are slightly

downward biased. The numerator uses non-performing assets data from the Australian

operations of all banks’ consolidated groups. In contrast, the denominator includes all lending

done by financial intermediaries in Australia, including lending done by non-bank financial

companies not owned by banks.

150

300

450

600

150

300

450

600

1989 1990 1991 1992 1993 1994 1995

0

300

600

900

0

300

600

900

NAB – net

CBA – gross

$m

$m

$m

$m

1 200 1 200

na na

13

All banks, share of lending by type

Source: RBA

Contemporary accounts of the period also indicate that credit losses were primarily

on lending to businesses. Trevor Sykes’ (1994) classic account of corporate and

Edna Carew’s (1997) account of Westpac’s experience during the period indicates

its losses were concentrated in business lending, and more specifically, in property

development lending. The dominant role of business lending is also suggested by

contemporary accounts from industry participants (Phelps 1989; Lee 1991).

Various authors have set out potential reasons why credit losses were so large in

the early 1990s (Battellino and McMillan 1989; Fraser 1994; Sykes 1994;

Carew 1997; Conroy 1997; Ullmer 1997; Gizycki and Lowe 2000). There was a

recession during 1990–91, and downturns in financial and property markets, but

losses were many times greater than those seen in earlier (and later) downturns,

suggesting other factors at play. A short version is that deregulation of the banking

sector in the 1980s was accompanied by very fast business lending growth and

declining lending standards, all during a period of strong economic and financial

0

4

8

12

0

4

8

12

Lending to businesses

1994

% %

Lending to households

1990 1991 1992 1993

14

was the result. In more detail:

1. Deregulation allowed banks to extend credit to meet demand from borrowers

(Battellino and McMillan 1989). The rates and terms at which banks could offer

deposits were liberalised over the 1970s and first half of the 1980s. Prior to this,

banks passively accepted deposit flows and restricted lending during periods of

deposit outflow. The change allowed banks to actively manage their funding to

match the demand for credit, and was accompanied by the removal of interest

rate caps on lending products and requirements for banks to lend to certain

borrowers. In addition, in 1985 foreign banks were allowed to enter the

Australian banking market as retail deposit-takers for the first time in over

40 years (Fraser 1994). The net result of these changes was a market where

banks competed intensely to grow their loan books and maintain market share.

Annual growth in nominal business credit rose above 20 per cent in

September 1984, and didn’t fall below this level again until June 1989

(Figure 7).

13 I use a broad definition of lending standards in this paper: non-price differences in borrower

definition below, but it is important to note that I do not include changes in portfolio

composition between business, housing, and personal loans within my definition.

15

Year-ended

Note: (a) Deflated using the domestic final demand deflator

Sources: ABS; APRA; RBA

2. In part due to the competitive pressures unleashed by deregulation, bank lending

standards loosened considerably over the 1980s (Macfarlane 1991; Sykes 1994;

Conroy 1997; Ullmer 1997). From the late 1970s, banks departed from the

practices of earlier decades and began lending to large companies on an

unsecured basis, and accepting riskier forms of collateral (such as equity in

subsidiaries and mortgages over unfinished developments). Banks also relaxed

covenants around the use of borrowed funds, loan-to-valuation and interestcoverage

ratios. Another major driver of the losses over this period was a lack of

transparency on borrowers’ total use of debt finance. When borrowers entered

financial difficulty, banks would sometimes discover total debt was higher than

thought, and even their own group exposures were higher than thought, due to

lending by subsidiary finance companies and merchant banks. Remuneration

was one potential driver of the fall in lending standards: corporate lending

officers in banks were frequently remunerated on the basis of volume, with little

consideration of long-run asset performance. Arguably, this loosening of lending

standards occurred because banks, emerging from an era of tight regulation,

lacked the proper corporate governance and sophisticated credit risk

-10

0

10

20

30

-10

0

10

20

30

2013

% %

1983 1989 1995 2001 2007

Nominal

16

management frameworks that have come to be seen as necessary for prudent

banking in a deregulated financial system.

3. Macroeconomic and financial conditions facilitated these developments. Real

GDP grew at an average rate of about 4¼ per cent over the five years to

September 1989. Equity prices rose by almost 50 per cent per annum from late

1984 until the crash in October 1987. Commercial property price growth rose

above 10 per cent per annum at the start of 1986, and accelerated in subsequent

years. This price growth was accompanied by an exceptional amount of nonresidential

construction, particularly of offices (Figure 8; Kent and Scott 1991).

Commercial property was a key form of collateral for the business loans that

were secured.

Note: (a) Capital city CBD prices: based on Adelaide, Melbourne, Perth and Sydney prior to June 1984,

includes Brisbane and Canberra after

Sources: ABS; JLL Research; RBA

4. Immediate triggers for the rise in credit losses are easier to discern than the

underlying reasons why they were so large. Business interest rates rose from

around 13 per cent at the start of 1988 to over 20 per cent by the end of 1989,

due to rises in official rates. Together with slowing business profits growth and

the significant growth in business debt, this meant that the aggregate business

0.0

0.3

0.6

0.9

-50

-25

0

25

Office construction

(LHS, per cent of GDP)

2013

% %

(RHS, year-ended)

1983 1989 1995 2001 2007

17

sector interest burden was very high (Figure 9). By early 1990, large highly

geared companies across a range of industries were unable to meet their

increased loan repayments and defaulted on their debts (Sykes 1994). This,

together with a weakening in the commercial property market, exposed banks to

a first round of credit losses (Gizycki and Lowe 2000). These losses broadened

as business profits began to fall and Australia entered a recession around the end

of the year. By September 1991, large additions to the supply of office property

had combined with flat or falling demand to sharply raise vacancy rates and

drive prices down by over 20 per cent on an Australia-wide basis; some banks

were forced to recognize significant credit losses on commercial property

lending (Carew 1997).

Note: (a) Business sector interest payments on intermediated debt divided by profits

Sources: ABS; Annual reports; APRA; RBA

Of the banks in the long-run dataset, the one that incurred the highest rate of credit

losses during the early 1990s was a small foreign-owned bank (Figure 10). These

losses were equivalent to a significant proportion of this bank’s capital, but it was

recapitalised by its parent entity. Of the groups that make up larger portions of the

sample, state government-owned banks experienced the highest credit loss rates

over this period. Two, the State Bank of South Australia (SBSA) and the State

Bank of Victoria (SBV), effectively failed, in that they had to rely on extraordinary

0

15

30

0

15

30

1983 1989 1995 2001 2007 2013

0

1

2

3

0

1

2

3

Profits growth

%

Current loss ratio

Sample aggregate, as at September

%

%

Average interest rate

18

financial support from their state government owners (Fitz-Gibbon and

Gizycki 2001). The major banks also experienced large credit losses over this

period. Two major banks – Westpac and ANZ – reported large overall losses in

their annual reports for 1991. The other banks in the sample, primarily smaller

Australian-owned banks, incurred significantly lower losses over the period. These

banks’ portfolios were generally more concentrated in lending to households.

Individual bank current loss ratios, as at September

Source: Annual reports

Even during a period in which system-wide lending standards loosened, there are

indications lending standards at state government-owned banks, particularly SBSA

and SBV, were below average:

SBV grew their lending at rates of 43 and 27 per cent per year between 1985 and

1990, versus growth in total credit of around 18 per cent per year over this

period. This fast growth was driven by business lending – the share of these

banks’ portfolios made up by business lending increased by over 20 percentage

points over the same period. State government owners encouraged fast lending

growth, both to support state economies and to provide a new source of revenue

for state coffers, and installed aggressive managers (Sykes 1994).

0

2

4

6

8

10

0

2

4

6

8

10

Foreign

banks

2000

% %

1980 1984 1988 1992 1996

State government-owned

banks

Other banks

19

Auditor-General of South Australia’s report into SBSA stated:

… the Bank’s corporate lending … was poorly organised, badly managed and badly

executed. Credit risk evaluation was shoddy. Corporate lending policies and procedures

were not even compended into a credit policy manual until 1988, and even then contained

serious omissions. The ultimate loan approval authority - the Board of Directors - lacked

the necessary skills and experience to perform its function adequately. Senior

management’s emphasis was on doing the deal, and doing it quickly.

(MacPherson 1993, p 1-24)

supervision by the Reserve Bank, though they had given undertakings to comply

with the Reserve Bank’s prudential regulations. Despite this, there were

1990s (Gizycki and Lowe 2000). The liabilities of state government-owned banks

were always explicitly guaranteed by their owners. The banking system as a whole

remained well-capitalised; partly due to some banks raising equity, the aggregate

capital ratio actually rose over this period (Fraser 1994). Both ANZ and Westpac

maintained capital ratios above regulatory minima, despite their losses in 1991.

There were short-lived deposit outflows at some small banks, but these were

quickly ended by Reserve Bank assurances about their solvency.

The elevated credit losses experienced during and after the global financial crisis

were due to business lending; the better data available for this period make this

clear (Figure 11). Losses on household lending barely rose over the period. Losses

on the business loan portfolio were much lower than those incurred during the

14 Sykes (1994) provides the examples of a large exposure and a related-party transaction that

were undertaken by SBSA contrary to Reserve Bank advice. The SBV failed to meet the

Reserve Bank’s capital adequacy standards during the late 1980s (Victoria 1991).

20

rates during the early 1990s (and, presumably, even higher business loan write-off

rates at that time).

Annual net write-off ratios

Notes: (a) Consolidated data for three major banks

(b) Includes all banks with housing loans > $1 billion; eighteen banks as at December 2013

Sources: APRA; Pillar 3 reports

The low loss rate on lending to households over this period was driven by very low

losses on housing loans, which made up around 90 per cent of bank lending to

households over this period. The net write-off ratio on housing lending averaged

households during this period arose from personal lending (credit card and other

personal lending) (Figure 12). Though personal lending has a relatively high loss

rate, it appears to be significantly less cyclical than business lending, and anyway

only makes up around 5 per cent of bank lending in Australia.

15 This loss rate is after the effect of lenders mortgage insurance (LMI), which Australian banks

hold on a significant portion of their housing loans (estimates suggest LMI covers roughly

one-quarter of housing loans). Reserve Bank estimates suggest the annual loss rate faced by

lenders mortgage insurers averaged 3 basis points over 1984 to 2012.

0.0

0.2

0.4

0.6

0.8

1.0

1.2

2013

0.0

0.2

0.4

0.6

0.8

1.0

1.2

Business and personal

% %

2009 2011 2013 2009 2011

21

Around one-fifth of Australian-owned banks’ consolidated assets are offshore, so

the consolidated Pillar 3 data used in Figure 11 (left panel only) and Figure 12

reflect overseas credit risk to some extent. Australian banks’ credit losses on

offshore lending were significant during the GFC (see, for example, RBA (2010)),

but domestic credit risk is the focus of this paper.

Consolidated data for three major banks, annual net write-off ratios

Source: Pillar 3 reports

One part of the explanation for the lower credit losses experienced during the

global financial crisis is the less severe nature of this episode: GDP fell for only a

single quarter and office property prices fell by around a quarter, compared with a

peak-to-trough decline of around one-half in the early 1990s (see Figure 8). Bank

lending to businesses grew at around 15 per cent per annum over the five years up

to mid 2008; this was around 8 percentage points below its growth rate over the

five years up to mid 1989 (higher inflation in the earlier period only accounts for

around half of this gap). This smaller rise in debt, together with structurally lower

interest rates that fell quickly in response to large cuts to the cash rate, meant the

business sector’s aggregate interest burden peaked at around 17 per cent of profits

during the global financial crisis, well below its level in the early 1990s (see

Figure 9).

0

1

2

3

4

0

1

2

3

4

Residential mortgage lending

2013

% %

2008 2009 2010 2011 2012

Credit cards

Other personal lending

22

There is also evidence that more conservative business lending standards were a

key contributor to the better credit loss experience during this episode. Partly in

response to the problems in the early 1990s, and partly in response to the

imposition of risk-based capital requirements and other regulatory pressures, banks

had improved their management of credit risk by the start of the global financial

crisis according to many observers (Eales 1997; Ullmer 1997; Gray 1998;

APRA 1999; Laker 2007). Better IT systems were put in place to assess and

monitor credit risk, and the governance of credit risk decisions within banks had

improved.

This section explores the timing of credit losses with respect to the economic

cycle, and relationships between credit risk measures. If credit losses peak quickly

after troughs in output, this means the financial strength of the banking sector may

start to improve soon afterwards – a key consideration for economic policymakers

after the global financial crisis. Likewise, if credit losses peak before nonperforming

assets, they might provide an early signal of future improvement in the

financial strength of the banking sector.

The temporal relationship between credit losses and output was reasonably similar

during the early 1990s and global financial crisis episodes. The peak in current

losses in the early 1990s, as measured by the long-run dataset (which provides

annual losses as at September of each year), was in 1991. The trough in annual

GDP during this episode was in the December quarter of 1991. APRA’s quarterly

credit loss data for all banks (available from 2003), allow more precise

measurement of timing. Quarterly credit losses, a volatile series, peaked in the

same quarter as the trough in quarterly GDP during the global financial crisis

episode (Figure 13). Losses rose noticeably three years before their peak in the

early 1990s, while they were only slightly elevated a year before their global

financial crisis peak.

23

Peak = 100

Sources: ABS; APRA

The relationships between different measures of credit losses differed somewhat

across the two main episodes (see Figure 3). The BDR exceeded the CLR in the

years immediately prior to both the downturns, indicating that banks were

increasing collective provisions in anticipation of a deterioration in loan

performance. During the global financial crisis, banks continued to increase

collective provisions during the downturn itself, perhaps owing to an overly

pessimistic view of future developments. The profile of credit losses was a

relatively symmetric hump in the early 1990s, but credit losses generally declined

more slowly in the years following the global financial crisis. This may reflect

economic conditions over this period, or banks adjusting their behaviour in

recognising and disposing of troubled loans. This difference makes comparing the

delay between initial losses and final write-offs between the two episodes difficult;

but, in aggregate, the net write-off ratio peaked two years after the other two ratios

in the early 1990s, and a year after in the global financial crisis episode.

-8 -6 -4 -2 0 2 4 6 8

0

20

40

60

80

100

0

20

40

60

80

100

Quarters from trough in GDP

index index

Current loss

ratio – annual

Current loss

ratio – quarterly

24

Credit losses in Australian banking have generally peaked before non-performing

assets (NPA) and impaired assets (IA), though these measures have risen in tandem

at the start of downturns. APRA’s quarterly credit loss data for all banks show a

during the global financial crisis episode (Figure 14); the lead is five quarters

are arguably zero and two quarters (respectively), given the June quarter 2009

value for this variable is very close to its peak in the March quarter of 2010.

All banks

Source: APRA

The narrative account in Section 3 reveals a range of features of credit losses in

Australian banking. Aggregate credit losses clearly have a relationship with the

economic cycle, but appear to be affected by macro-level factors other than just

output growth. Business sector conditions, such as commercial property prices and

business indebtedness, look to have played the key role. The composition of banks’

portfolios also appears important: credit losses look to have been incurred mainly

on business lending. But the direct evidence for this is based on data from only a

0.0

0.2

0.4

0.6

0.8

0.0

0.5

1.0

1.5

2.0

Current loss ratio

– annual

(LHS)

2013

% %

2005 2007 2009 2011

Current loss ratio

– quarterly

(LHS)

NPA ratio

(RHS)

IA ratio

(RHS)

Peak in NPA and IA ratios

25

few banks for the largest episode of credit losses. Cross-sectional differences in

bank-level credit losses have been large, and there are suggestions that these are

driven by variation in lending standards (as well as portfolio composition).

Section 4 uses a panel data modelling framework to explore these issues further.

Consistent with the international literature, I model the relationships between banklevel

credit losses and both macro-level and bank-level factors (see

literature, I use the fixed-effects (within) estimator, which removes time-invariant

heterogeneity is unobservable and may be correlated with the explanatory variables

of interest. Relevant unobservables include the average risk appetite of a bank’s

managers and, relatedly, its average lending standards (both of these probably also

vary within banks over time, and this is explored in Section 4.4).

business sector profits and growth in the household sector’s disposable income, all

measures of changes in borrowers’ incomes. The level of the cash rate, as well as

the interest burdens of the whole economy, household sector and business sector,

System-wide nominal credit growth, and growth in nominal housing and business

credit, are intended to capture system-wide changes in lending standards (see, for

example, Keeton (1999)). Residential and commercial property are used as

collateral for housing and business loans in Australia, so changes in the prices of

these assets are included to capture changes in the value of this collateral. Details

16 A survey is available in Glogowski (2008). Salas and Saurina (2002) are a commonly cited

precedent when using macroeconomic and bank-level explanatory variables to model banklevel

credit risk outcomes.

17 The economy-wide interest burden is equal to the estimated interest payments on all

intermediated debt in the economy divided by GDP. The business and household sector

interest burdens are defined similarly: see Table B2.

26

of variable construction, descriptive statistics and correlations for all explanatory

variables are in Appendix B (Tables B2, B3 and B4).

devoted to business and personal lending, and bank-level loan growth.

I include all variables contemporaneously, except for interest burden (which I lag

one year) and the credit and loan growth variables (I include lagged terms covering

the past four years for these variables). These variables are excluded

contemporaneously because of the mechanical impact of credit losses on the level

of credit and loans (these measures are calculated net of identified losses). I

exclude bank-year observations on banks making up less than 1 per cent of total

bank loans to prevent idiosyncratic risk in very small loan portfolios from clouding

my results. This leaves 328 observations on 26 banks covering 1982 to 2013.

Table 1 reports regression results from two alternative forms of Equation (1).

Model A uses mainly economy-wide macro-level explanatory variables. Model B

uses variables specific to the household and business sectors.

These models indicate that the drivers of credit losses over 1982–2013 are largely

those highlighted by previous Australian work using shorter time periods. At the

macro level, interest burdens, sectoral credit growth measures, and growth in

residential and commercial property prices appear to influence losses, and

measures specific to both the business and household sectors appear important.

credit risk at Australian banks over periods ending in 1999. Banks with mainly

business lending appear to have incurred higher credit loss rates than banks with

mainly housing lending, in line with Esho and Liaw (2002). The model that uses

only sectoral macro-level variables (Model B) explains credit losses slightly better

than the one that uses primarily economy-wide variables (Model A). This is useful

for the development of the main model of this paper – presented in Section 4.3 –

which includes interactions between portfolio shares and macro-level variables.

27

Variable Model A Model B

Constant –1.576*** –2.749***

Observations 328 328

Within R-squared 0.48 0.54

Adjusted within R-squared 0.45 0.51

AIC 725 697

BIC 785 781

Notes: All models are estimated with bank fixed effects and standard errors are clustered by bank; ***, ** and *

denote significance at the 1, 5 and 10 per cent level respectively

28

These simple models offer a number of other insights:

signed, but only significant at the 10 per cent level. Output growth has been

found insignificant in international studies using models that include a range of

cyclical macro-level variables (Davis and Zhu 2009). At the sectoral level,

business profits growth has a negative relationship with credit losses that is

significant at the 1 per cent level, while growth in household disposable income

does not appear to be a significant explanator of credit losses. This difference is

consistent with the relative importance of developments in these sectors in the

narrative account of credit losses in Australia.

with credit losses in Model A. A one standard deviation increase in this variable

(roughly 2 percentage points) is associated with a 26 basis point rise in credit

loss ratios. The economy-wide aggregate interest burden is the weighted average

of borrower-level interest burdens across the economy; a rise in the former must

represent some increase in risk at the borrower level. Interest burdens within

both the business and household sectors appear to underlie this aggregate

relationship – both are significant in Model B. This is consistent with

in Australia. Default and financial distress among household borrowers does not

feature prominently in this.

credit losses. Again, while consistent with Gizycki (2001), this is not entirely

consistent with the narrative account of credit losses in Australia. Residential

property has primarily served as collateral for housing loans in Australia, and

the available evidence indicates banks have not incurred significant credit losses

on such loans over recent decades. Residential property now also collateralises a

significant amount of small business lending in Australia, but this makes up only

a small proportion of total business lending and it is unclear how prevalent this

arrangement was in earlier decades.

in this model, though they have opposite effects over short horizons. Positive

relationships between longer lags of credit and loan growth and losses are the

29

most common finding in the international literature, and are generally thought to

operate through increases in credit supply that involve lending to less financially

sound borrowers (see Section 4.4). Though there is an explanation for the

estimated negative relationship between business credit growth and losses –

more easily available credit (signalled by strong credit growth one year ago)

may make refinancing easier for weaker borrowers who would otherwise default

– this estimated relationship should be interpreted cautiously. Demand for credit

probably weakens during downturns (which in turn cause credit losses), so the

estimated relationship may not be causal.

relative magnitudes that accord with the portfolio-level loss rates shown in

Section 3.1.2. A bank with more business lending and less housing lending

incurs higher credit losses. More (non-housing) personal lending has a similar,

but stronger, effect. Higher loan growth raises losses at the individual bank level

with a multi-year lag. The estimated coefficients on this variable are of a similar

This section contains the primary econometric model for credit losses presented in

this paper. It uses the same modelling framework as the simple models presented

in Section 4.2, but relies on explanatory variables that are interactions between

macro-level variables and bank-level portfolio shares. Equation (2) shows a model

simply those thought to cause credit losses on business lending: business profits

growth, the business sector interest burden, business credit growth, and

commercial property price growth. The other macro-level variables – those that

1 1

1

α

ε

− −

−

= + ʹ′ + ʹ′

+ ʹ′ + ʹ′ +

β

Γ

(2)

30

The key idea is that requiring macro-level variables to affect credit losses through

the portfolios they are related to should provide better identification of the drivers

of credit losses. For example, the mechanism through which falls in residential

property prices are thought to cause credit losses is by lowering the value of the

collateral backing housing loans. A model that requires changes in residential

property prices to act on credit losses through banks’ housing lending should

distinguish this causal channel from mere correlation between house prices and

other macro-level conditions that cause credit losses (cross-correlations between

my explanatory variables are shown in Appendix B). The limited international

literature that models credit losses at the portfolio level generally finds each

A necessary condition for the unbiased estimation of Equation (2) is that the

portfolio shares are independent of the error term. One reason why this assumption

may not hold is correlation between (within-portfolio) lending standards and

portfolio shares. But arguments can be made for both positive and negative

relationships; banks that do more business lending should be better at selecting

businesses to lend to, but banks with a lot of business lending may have arrived at

that position by accepting borrowers other banks did not. The dataset contains a

wide range of variation in portfolio composition, in part due to the regulatory

distinctions between savings banks (which mainly concentrated on housing

lending) and trading banks (which mainly concentrated on business lending) over

the late 1980s and early 1990s. The state government-owned banks that received

extraordinary government support in the early 1990s had shares of business

lending in the middle of the sample range.

Table 2 shows the estimation results from the model (Model C). The key insight

from Model C is that business sector conditions appear to have been the main

driver of credit losses in Australia over recent decades. The household sector

interest burden, residential property prices, and housing credit growth no longer

18 For example, using data on Greek banks, Louzis, Vouldis and Metaxas (2012) found nonperforming

personal loans to be very sensitive to interest rates, business lending sensitive to

GDP growth, and mortgages not very sensitive to macroeconomic developments. Hoggarth,

Logan and Zicchino (2005) estimate models for sectoral write-offs from UK banks’ business,

personal, and housing portfolios that are each driven by a different set of macro-level

variables.

31

Constant –0.183

Observations 328

Within R-squared 0.62

Adjusted within R-squared 0.58

AIC 642

BIC 733

Notes: Model C is estimated with bank fixed effects and standard errors are clustered by bank; ***, ** and *

denote significance at the 1, 5 and 10 per cent level respectively

(a) BSL = business share of lending, HSL = housing share of lending and PSL = personal share of

lending; portfolio measures used in interactions are lagged one period

32

have significant relationships with credit losses when required to interact with

credit losses through household lending. In other words, studies showing macrolevel

correlations between measures of household sector financial conditions (such

as housing prices) and future financial crises might actually be picking up a

correlation between housing prices and the actual drivers of financial distress, not a

causal link from housing prices to financial instability.

Business profits growth, the business sector interest burden, business credit

growth, and commercial property price growth are all significant at the 5 per cent

level in Model C. This model also has a better statistical fit than Models A and B,

indicating that incorporating portfolio interactions is a valid choice.

Most of the statistically significant relationships in Model C are economically

significant. The median, mean and standard deviation of current losses in the

dataset used for this regression are 24, 57 and 107 basis points respectively, and

one standard deviation changes in key macro-level variables generate losses that

in business interest rates and business profit growth appear to be the most

important for credit losses. Both affect losses through changes in the business

sector interest burden, as well as directly in the case of business profits. The model

implies that the level of business debt relative to interest rates and profitability is

an important state variable for losses. Assuming an initial business sector interest

burden equal to the average over 1981–90 (21.4 per cent), rather than the sample

average (16.4 per cent), leads to the larger effects on losses shown by the lighter

bars in Figure 15.

19 Sample means used in Figure 15 are: business interest burden (16.4 per cent), business

interest rate (10.5 per cent), commercial property price growth (5.2 per cent), business profits

growth (7.4 per cent), and business credit growth (10.8 per cent). One standard deviation

shocks are: business interest rate (+4.5 percentage points), commercial property price growth

(–12.3 percentage points), business profit growth (–6.1 percentage points), and business credit

growth (+9.5 percentage points).

33

Effects of one-year, one standard deviation changes

Notes: All macro variables evaluated at sample averages; business share of lending = 0.4 and housing share of

lending = 0.5; I sum the simultaneous direct effect of business profit growth on the credit losses and its

effect one year later through the interest burden, I do the same for the four lags of business credit growth

Another way to look at the influence of macro-level variables is to examine the

Model C. Figure 16 plots the contribution of each macro-level variable and its

contributions of all household macro-level variables are shown as an aggregate, as

are the contributions of the variables in the model that are not interacted with

fits actual losses quite closely (the RMSE is 0.15), so this model provides a macrolevel

explanation that, while suffering from the same limitations as all models,

quite closely fits the actual experience.

,1990 1 ,1990 1991

ˆ

the appropriate weight for each bank (each bank’s share of sample loans, based on loans

outstanding in 1990).

21 This is shown as the ‘Other variables’ contribution in Figure 16.

Business credit

growth (rise)

Commercial property

price growth (falls)

Business profits

growth (falls)

Business interest

rate (rises)

0.0 0.1 0.2 0.3 0.4 0.5

Current loss ratio

Higher

business

debt

34

From Model C

Notes: Use of interactions means part of the effect of changing portfolio composition is captured

estimated fixed effects

Sources: APRA; Annual reports

been the macro-level driver of aggregate credit losses (as they have been for banklevel

credit losses). Household macro-level variables, even in aggregate, have

made only small contributions to changes in the aggregate CLR. Rising business

indebtedness placed upward pressure on credit losses during the first decade of the

sample. During the 1980s, this was offset by fast growth in business profits and

commercial property prices. Slowing growth in (and eventually falls in) profits and

commercial property prices, in combination with the high business sector interest

burden, triggered the large rises in credit losses in the early 1990s. A similar, but

smaller, dynamic underlies the rise in the fitted CLR between 2007 and 2009.

During both episodes, sharp slowdowns in business credit growth also contributed

to higher aggregate losses.

1983 1988 1993 1998 2003 2008 2013

-3

-2

-1

0

1

2

3

-3

-2

-1

0

1

2

3

Fitted

% %

Actual

35

By construction, Figure 16 attributes most of the variation in aggregate credit

losses to macro-level variables, as it aggregates the contribution of changes in each

macro-level variable and changes in the portfolio share with which it is interacted.

Figure 17 takes the alternative approach of applying changes in macro-level

variables while holding the banking system constant. This is done for two

reference years, 1991 and 2008. For example, the line for 1991 shows the

aggregate CLR predicted by Model C for the 16 banks in the sample in 1991,

applying the actual macro-level variables experienced in each year, but freezing

each bank’s portfolio composition and past loan growth at 1991 values. The

distance between each of counterfactual lines and the actual fitted values from

Model C shows how changes in the banks in the sample, and their characteristics

(from the reference year), contribute to aggregate losses.

Figure 17 indicates that different macro-level experiences do not explain all of the

difference in credit losses between the early 1990s and global financial crisis

episodes. For example, the banking system in 2008, if subjected to the macro-level

conditions present in the early 1990s, is predicted to incur credit losses of around

5½ per cent over the period (the area under the 2008 line between 1990 and 1994).

This is 3 percentage points below the actual credit loss ratio incurred over this

environment and changes in the structure of the banking system explain the large

difference in credit losses between the early 1990s and global financial crisis

episodes.

Caution should be used in giving causal interpretations to the relationships

estimated by these econometric models. While most of the estimated relationships

are intuitive – the business sector interest burden, for example, has a very natural

relationship with credit losses – reverse causality may be present. A good example

of this is the United States during the global financial crisis, where credit losses on

residential mortgages destabilised large financial intermediaries with consequent

impacts upon broader economic and financial conditions (Hall 2010;

Mishkin 2011). This causal chain is likely to have been less important in Australia

during my sample period, mainly because of the robust position of the Australian

2012. This is smaller (around ½ percentage point). The large difference between these

estimates is an inherent drawback of the structure of Model C.

36

banking system over the whole period (including while it was subject to large

credit losses in the early 1990s). This argument is not that there is no casual

channel from credit losses to macroeconomic conditions in Australia, but rather

that it was not triggered during my sample period. Public actions during crisis

periods have also dampened this channel in Australia.

From Model C

Note: (a) Fitted losses holding banks in the sample and their characteristics at the values in the indicated year

The results of Model C are robust to a number of alternative specifications,

including alternative portfolio interactions, alternative lag structures, and different

sample periods (see Section C.1 of Appendix C). Omitted variable bias is probably

the greatest statistical concern: lending standards have not been discussed in the

context of the models, but are likely very important for credit losses.

The econometric models above treat all business lending as having equal

propensity to cause credit losses (conditional on the macroeconomic environment),

regardless of whether it is business lending in the early 1990s or in the mid 2000s,

and regardless of the bank doing the lending. But there is evidence that this is not

an accurate assumption; that, for example, lending standards were worse in the late

1983 1988 1993 1998 2003 2008 2013

-1

0

1

2

3

-1

0

1

2

3

% %

loss ratio

37

1980s than in the early 2000s. And, internationally, empirical work has shown that

lending standards vary over time and played a role in the global financial crisis

(see, for example, Lown, Morgan and Rohatgi (2000); Maddaloni and

Peydró (2011); and Dell’Ariccia, Igan and Laeven (2012)). This section of the

paper attempts to quantitatively explore the effect lending standards have had on

credit losses in Australia.

Lending standards, particularly for lending to businesses, are not well-defined in

business, housing and personal lending as a component of lending standards. Some

examples of changing lending standards were given in Section 3.1.1. But my

definition encompasses other differences, such as the industry composition of a

bank’s business lending. A business lending portfolio with a higher share of

commercial property lending, which has historically been riskier than other

business lending (Ellis and Naughtin 2010), could be described as of a lower

standard under my definition.

Changes in average lending standards over time are captured reasonably well by

Model C, given its close fit at the aggregate level. Several of the macro-level

variables in this model likely act as proxies for lending standards. As shown in

Figure 15, the long-run relationship between business credit growth and credit

losses is positive, and this probably captures increases in credit supply that involve

lending to less financially sound borrowers (see, for example, Keeton (1999)).

Jiménez and Saurina (2006) use loan-level data to show that, controlling for

macroeconomic conditions, loans originated while a bank is growing faster are

more likely to default and less likely to be collateralised.

The business sector interest burden is also likely acting as a proxy for average

lending standards. Banks extending business loans often place a contractual limit

upon businesses’ interest burdens (at origination and/or over time). The aggregate

business sector interest burden, the weighted average of the interest burdens of all

businesses in the economy, captures some portion of the time series variation in

this lending standard. Firm-level data illustrate this clearly. Looking at the largest

300 listed companies at each point in time, firm-level interest burdens were higher

in 1990 than in either 1982 or 2008 (Figure 18). For example, around half of the

38

largest 300 listed firms had an interest burden above 50 per cent in 1990, while less

than one-fifth of the largest 300 listed firms in 1982 had an interest burden above

this level. This variation is partly captured by the aggregate business sector interest

ranking also correlates with the relative magnitudes of the credit losses

experienced during the downturns that began in each of these years.

CDF for largest 300 listed non-financial firms in each year

Notes: CDF denotes cumulative distribution function; firms with no debt have been excluded; loss-makers with

debt are assigned an interest burden of 120 per cent

Sources: Morningstar; RBA; Statex

In contrast, Model C does a poor job of explaining cross-sectional variation in

lending standards. Model residuals during the early 1990s are very large for some

state government-owned banks, and the narrative evidence presented in Section 3

indicates that these banks had below-average lending standards (Figure 19). The

bank-level variables included in Model C, portfolio composition and bank-level

23 The aggregate and firm-level measures differ slightly. Aggregate interest burden captures

interest on intermediated debt only, while the firm-level measure captures interest on all debt.

The aggregate measures of business profits, gross operating surplus (for private non-financial

corporates) and gross mixed income (for unincorporated enterprises) differ somewhat from

the firm-level measure, earnings before interest and tax.

0.0

0.2

0.4

0.6

0.8

1.0

0 20 40 60 80 100

Interest burden (interest expense / EBIT) – %

1990

2008

1982

39

loan growth, do not explain why the credit loss ratios experienced by these banks

were so much larger than those at other banks. This omission of lending standards

is partly responsible for the higher RMSE of Model C at the bank level: this

statistic drops from 63 to 44 basis points if state government-owned banks are

excluded.

Lines show residuals from Model C for individual banks, as at September

There has almost certainly been more variation in lending standards than is

indicated by this state government-owned/non-state government-owned distinction.

But little other hard information on lending standards is available. Quantile

regression is one strategy that has been used to assess the effect of unobserved

of credit losses, conditional on the explanatory variables. It provides a more

complete description of relationships than least squares regression, which is based

upon estimating only the conditional mean of a dependent variable. If unobserved

differences in lending standards are the primary determinant of the conditional

distribution of credit losses, estimated relationships with macro-level variables at

higher (lower) quantiles can be interpreted as being for banks with worse (better)

lending standards. If other factors (e.g. idiosyncratic risk) are the primary

24 Bitler, Gelbach and Hoynes (2006), for example, examine quantile treatment effects in a

labour economics context. ‘Quantile’ is a synonym for ‘percentile’.

1986 1988 1990 1992 1994

-4

-2

0

2

4

-4

-2

0

2

4

State governmentowned

banks

Other banks

ppt ppt

40

determinants of the conditional distribution of credit losses, this interpretation does

not hold and quantile regression estimates merely show the range of possible

responses to changes in macro-level variables.

Quantile regression is also more robust to outlier observations than least squares

methods (Cameron and Trivedi 2009). For example, the significance of

commercial property prices in the above models could be driven entirely by a very

strong relationship for a subset of banks that experienced credit losses well aboveaverage.

But in a quantile regression, this should be apparent in the lack of a

relationship between the variables and some parts of the credit losses distribution.

Quantile regression generates estimated relationships with macro-level variables

that vary widely across the distribution, and do so in a way that is statistically

significant in some cases (Figure 20; full quantile regression outputs are in

significant effects, signed in line with estimated coefficients in Model C, on losses

across almost all of the distribution. For example, the effect of the business sector

interest burden upon losses at the 90th percentile is more than three times as large

as at the 10th percentile. More concretely, a one standard deviation rise in business

sector interest burden (5.5 percentage points) raises credit losses by 12 basis points

at the 10th decile of the distribution, while it raises credit losses by 43 basis points

at the 9th decile (for a bank with 40 per cent business lending). The comparable

impact from Model C is 30 basis points.

25 I use a parsimonious version of Model C for the quantile regression, and I drop the fixed

effects.

41

Note: Shaded regions are 95 per cent confidence intervals

Credit losses in Australian banking in the post-deregulation period have been

concentrated in two episodes: the very large losses around the early 1990s

recession and the smaller losses during and after the global financial crisis. They

have a close temporal relationship with the economic cycle, peaking close to

troughs in GDP during downturns. A narrative account attributes the key roles in

driving credit losses to business sector conditions such as business indebtedness

and commercial property prices. The available data on portfolio-level losses

indicate that elevated losses during these downturns stemmed from banks’ lending

to businesses, rather than their lending to households. Data available from 2008

onwards indicate losses on housing loans barely rose (from very low levels) during

the global financial crisis, even though housing prices and employment fell

noticeably in some geographical areas.

One of the main contributions of this paper is an econometric panel-data model

that properly controls for bank-level portfolio composition. This model indicates

business sector conditions, rather than household sector conditions, have been the

0.1 0.3 0.5 0.7 0.9

-0.1

0.0

0.1

0.2

0.3

0.3 0.5 0.7 0.9

Quantile

0.3 0.5 0.7 0.9

-0.24

-0.18

-0.12

-0.06

0.00

growth

RHS

Commercial property

price growth

RHS

Business sector

interest burden

LHS

coeff

42

driver of domestic credit losses over the period studied. The relevant business

sector conditions – interest burden, profitability and commercial property prices –

are indicators of the ability of this sector to service its debts and of the value of the

collateral behind these debts. As a corollary, the model indicates that most losses

over the past three decades were incurred on banks’ business lending, and

household losses were largely unresponsive to economic conditions in that period.

Unlike past work, these results are consistent with the narrative account of credit

losses in Australian banking.

Descriptive accounts attribute the scale of losses during the early 1990s to poor

lending standards, and the data support this. One piece of evidence, based on

quantile regressions, indicates that changes in macro-level conditions have had

very different impacts upon banks with similar portfolios (in terms of the shares of

business, housing and personal lending). Most compellingly, standard models

cannot explain the extremely high credit losses experienced at some state

government-owned banks in the early 1990s. Given the anecdotal evidence that

these banks had below-average lending standards, this is consistent with the

conclusion that poor lending standards have caused the very worst credit loss

outcomes over recent decades.

These conclusions have practical implications for stress testing. The credit loss

models in this paper that use least squares estimation, and include bank-level

variables, are unable to explain, and so unlikely to predict, the very worst credit

loss outcomes. Many stress-testing exercises use similar (and in some cases

simpler) econometric models (see, for example, IMF (2012)). As the worst credit

loss outcomes are the most relevant when stress testing, this suggests that

alternative models are needed. Covas, Rump and Zakrajsek (2013) show that a

type of quantile regression (quite different to that in this paper) can provide out-ofsample

forecasts that encompass the credit losses experienced by the US banking

system during the global financial crisis. In an Australian context, Durrani, Peat

and Arnold (2014) show that allowing variation in credit risk outcomes across

banks, rather than applying the same average risk parameters to all banks, can lead

to significantly larger loss estimates.

Stress-testing models could also be improved by incorporating better data on

lending standards. The Federal Reserve collects and makes use of loan-level data

on borrower characteristics in its annual stress tests of the largest US banks (Board

43

of Governors 2014). This captures some aspects of the risk profile of borrowers;

more work is probably needed to make it possible to systemise and accurately

record banks’ lending standards.

The historical experience of credit losses at Australian banks this paper describes

should help to guide overall understanding of the credit risk they currently face. It

supports a continued focus on the analysis of the financial health of the business

sector (one output of this work is a chapter of the Reserve Bank’s semiannual

peak before asset performance measures, potentially providing an early signal of

future improvement in financial system stability.

The lack of a historical relationship between household sector conditions and credit

losses should be used cautiously in contemporary debates on the riskiness of

housing lending. It indicates that the macroeconomic shocks experienced by the

household sector during the past three decades have been small relative to the

lending standards in place for housing lending over this period. Future

macroeconomic shocks may, however, have a larger impact on households. There

have been, for example, no large nationwide falls in house prices during recent

decades. In addition, a rise in unemployment on par with that in the early 1990s

could be expected to have a more severe influence on household credit losses,

given the large rise in household indebtedness over the intervening period. A

corollary of this rise in household indebtedness is the greater share of banks’

lending now made up by housing and personal lending. These considerations

suggest that any weakening in lending standards in these areas could have a larger

systemic impact than in the past.

44

Banks’ financial reports provide a number of different items which capture credit

losses:

when a bank identifies that it has incurred a loss on the loan, it must raise an

individual provision (a liability) in order to reduce the carrying value of the loan

to the amount it expects to recover. This reduction in net assets is a loss and

incorporates the release of individual provisions held against loans that no

longer require them (because, for example, a borrower has recommenced

making repayments).

collective provisions are held against incurred losses on loan portfolios with

similar characteristics (usually retail loans that are too small to deal with

individually), and the losses historical experience suggests are likely on the

portfolio of currently healthy loans (to an extent, see Appendix B).

all provisions through collective provisions at first. When losses on individual

loans are identified, appropriate amounts are transferred from collective

provisions to individual provisions to cover these losses, and an amount

necessary to replenish the collective provision to appropriate levels is charged to

the profit and loss though item 2.

of removing troubled assets from the balance sheet, and, for larger loans, are

made after losses have been recognised through individual provisions and

recoveries of any collateral made. The troubled loan, individual provision and

amount recovered should cancel out so that there is no impact on the profit and

loss account at this stage.

45

collectively provisioned for are written off against collective provisions.

prospect of recovery can be written off immediately upon evidence of loss

emerging. As there are no provisions held against such assets, their write-off has

an impact upon the profit and loss account. This category sometimes also

captures losses on loans that are not adequately covered by existing provisions.

credit losses: the sum of 1, 2 and 6. The charge for bad and doubtful debts is the

net reduction in the value of a bank’s assets due to credit losses in a given

period. It appears in a bank’s profit and loss account and so is a component of

the net change in a bank’s capital position over each period.

Unlike the charge for bad and doubtful debts, the other two aggregate measures of

in banks’ financial reports. Net write-offs are the sum of 4, 5 and 6; current losses

are the sum of 1, 3, 5 and 6.

‘Performing’ and ‘non-performing’ are a classifications applied to the assets banks

hold at amortised cost (mainly loans and some other credit-type assets). These

terms are, in most countries, defined by accounting standards and rules set by

banking regulators. In Australia, NPAs include two categories:

covered by sufficient collateral such that no loss is expected (well-secured).

doubtful and which are not well-secured.

46

The accounting surrounding credit losses, and the relationship between NPAs and

credit losses (and profitability and capital), can be illustrated via the souring of a

hypothetical $200 million loan to a commercial property development company

(see Figure A1):

1. At December 2007 the loan was being repaid on time and was otherwise within

its conditions.

2. During early 2008, pre-sales of residential units within the development began

to dry up. As a result, the borrower (a property developer) began to run out of

cash. It was unable to meet its required loan repayment in January 2008, and

also failed to make required repayments in February and March.

3. In accordance with prudential standards, immediately upon aggregate loan

arrears reaching 90 days of repayments (in April), the bank assessed its expected

future cash flows from the loan. It ascertained that the developer company was

in poor financial shape and would be unable to complete the development. The

bank thus decided that the sole loan recoveries likely to be made were those

from exercising its rights to repossess and sell the unfinished development (the

security for the loan). The bank assessed the market value of the development as

$100 million, and thus classified the loan as impaired and raised an individual

provision of $100 million against it. This was funded through a charge to the

bank’s profit and loss account (item 1 in Section A.1).

4. The unfinished development did not attract any offers to buy it over the

remainder of 2008. At the end of the year, the bank decided that property market

conditions had deteriorated, and had the property re-valued. The valuation this

time came to $50 million, so the bank increased its individual provision against

the loan by $50 million, again funded through a charge to the profit and loss

account.

5. In March 2009, the property sold for $50 million. The bank received this amount

and wrote-off the $200 million loan and $150 million individual provision (with

no impact upon the profit and loss account). The write-off of individual

provisions is item 4 above.

47

6. In June 2009, the liquidation of the development company was completed. In the

final distribution of assets to its creditors, the bank unexpectedly received

$10 million. The bank acknowledged this amount via a recovery direct to profit

and loss, which reduced both the cumulative charge for bad and doubtful debts

and cumulative net write-offs from the loan to $140 million. This is the net

impact on the bank’s profitability and capital of the credit losses caused by the

non-repayment of this loan.

This example closely follows how banks have actually dealt with troubled loans

over recent years, but there are ways in which the bank could have managed the

loan that would have led to different relationships between NPAs and credit losses.

For example, the bank could have immediately sold the development for the best

available price and written off the necessary amount against its profit and loss

account, without ever raising any provisions. Alternatively, the bank could have

kept the impaired loan on its books and delayed sale of the development (perhaps

for several years) until the market recovered sufficiently to allow a higher sale

price. A different asset type could also have changed the relationship: if the asset in

question was $200 million worth of residential mortgages, which are normally

much better-collateralised, the time profile for NPAs may have been exactly the

100

200

100

200

M J S D M J S D

0

100

200

0

100

200

Cumulative

net write-offs

2009

Non-performing assets

$m

$m

$m

$m

Credit losses

Cumulative charge for bad and

doubtful debts/current losses

Individual provisions

Impaired assets

2008

48

same (except made up by past-due, rather than impaired, assets) but credit losses to

the bank may have been much lower.

49

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.

50

Name of

bank

Bank

In

sample

Precursor

entities

Becomes Banks acquired

during sample

period

Alternative

names

ANZ (trading) Trading 1980–91 ANZ

ANZ Combined 1992–2013 ANZ (trading)

& ANZ

(savings)

CBA (trading) Trading 1980–92 CBA

Commonwealth

Savings Bank

Savings 1980–91 CBA

CBA Combined 1993–2013 CBA (trading)

&

Commonwealth

Savings Bank

Bankwest (from

2013)

State Bank of

NSW (from

2000)

State Bank of

Victoria (from

1990)

NAB (trading) Trading 1980–92 NAB

NAB Combined 1993–2013 NAB (trading)

& NAB

(savings)

Westpac

(trading)

Trading 1982–93 Westpac

Westpac 1994–2013 Westpac

(trading) &

Westpac

(savings)

St. George Bank

(from 2010)

Bank of

Melbourne

(from 1998)

St. George

Bank

Combined 1989–2009 St. George

Building

Society (until

1992)

Advance Bank

(from 1997)

State Bank of

South Australia

Combined 1985–94

State Bank of

NSW

Trading 1981–99 Colonial State

(from 1990)

State Bank of

Victoria

Savings 1982–90

Bankwest

Combined

1983–2012

Rural and

Industries

Bank (prior to

1990)

51

Name

Bank

In

sample

Precursor

entities

Becomes

Banks acquired

during sample

period

Alternative

names

Advance Bank Savings 1986–96

Metway Bank Savings 1980–96 Suncorp

Metway

Bank

Suncorp

Metway Bank

1997–2013 Suncorp

Metway Bank

& QIDC

Adelaide Bank 1994–2007 Adelaide

and

Bendigo

Bank

Bendigo Bank 1994–2007 Adelaide

and

Bendigo

Bank

Adelaide and

Bendigo Bank

2008–13 Adelaide Bank

& Bendigo

Bank

Bank of

Melbourne

Trading 1989–97

Bank of

Queensland

Trading 1980–2013

Deutsche Bank

Australia

Trading 1987–92 Changed to

branch

status in

1994

Macquarie Bank Trading 1986–2013

ING Bank

1995–98,

2002–13

ING

Mercantile

Mutual Bank

(until 1998)

HSBC Bank

1987–99,

2002–13

Notes: ANZ = Australia and New Zealand Bank; CBA = Commonwealth Bank of Australia; NAB = National

Australia Bank; QIDC = Queensland Industry Development Commission; Westpac = Westpac Banking

Corporation

(a) This column is blank for banks that entered the sample after 1993, given the distinction is not

meaningful after this period

(b) Data are not available for the missing years

52

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.

53

Aggregate current loss ratio, three major banks, as at September

Source: Annual reports

0.0

0.1

0.2

0.3

0.0

0.1

0.2

0.3

Current loss ratio

2013

% %

2001 2004 2007 2010

Current loss ratio –

adjusted to remove

discounting

54

All variables observed at September of each year

Variable Detail Sources

Real GDP growth Year-on-year growth ABS

Cash rate Average over year RBA

Economy-wide interest burden (Average intermediated credit outstanding

× average cash rate) / GDP, average over year

ABS; APRA;

RBA

Inflation Trimmed mean from 1993, average over year ABS; RBA

Total credit growth Annual growth APRA; RBA

Business sector interest burden (Average intermediated business credit

outstanding × average large business interest

rate) / business profits, average over year

ABS; APRA;

RBA

Business profits growth Private non-financial corporations’ gross

operating surplus + unincorporated enterprises’

gross mixed income, year-on-year growth

ABS

Average interest rate for large

businesses

Average over year APRA; RBA

Commercial property price

growth

Capital city CBD office property, weighted

using ABS shares, annual growth

ABS; JLL

Research; RBA

Change in the unemployment rate Annual ABS

Household sector interest burden (Average intermediated household credit

outstanding × standard variable mortgage rate) /

household disposable income, average over

year, year-on-year growth

ABS; APRA;

RBA

Household disposable income

growth

Before interest payments, year-on-year growth ABS

Standard variable mortgage rate Average over year APRA; RBA

Residential property price growth Annual growth REIA

Business credit growth Annual growth APRA; RBA

Personal credit growth Annual growth APRA; RBA

Housing credit growth Annual growth APRA; RBA

Share of system lending

Business share of lending Share of portfolio that is business lending APRA; RBA

Personal share of lending Share of portfolio that is personal lending APRA; RBA

Housing share of lending Share of portfolio that is housing lending APRA; RBA

Loan growth Winsorized at the 5th and 95th percentiles APRA; RBA

Dummy variable for state

government ownership

Author

55

Variable Mean Standard

deviation

Minimum Maximum

GDP growth 3.22 1.61 –1.91 5.97

Cash rate 8.24 4.41 2.91 17.19

Economy-wide interest burden 9.22 1.87 6.57 13.38

Inflation 4.27 2.99 –0.40 12.40

Total credit growth 11.25 6.23 –0.40 23.99

Business sector interest burden 16.22 5.45 10.01 29.24

Business profits growth 7.43 6.07 –4.62 27.69

Average interest rate for large

businesses

10.34 4.55 4.75 20.50

Commercial property price growth 5.95 12.67 –22.81 30.00

Change in the unemployment rate –0.02 1.02 1.61 2.90

Household sector interest burden 7.92 1.94 5.67 12.71

Household disposable income

growth

7.77 4.08 1.95 17.50

Standard variable mortgage rate 9.77 3.29 5.80 17.00

Residential property price growth 7.93 8.31 –2.30 41.66

Business credit growth 10.18 9.76 –5.90 30.55

Personal credit growth 8.66 7.38 –5.51 22.32

Housing credit growth 12.91 4.47 4.67 21.58

Share of system lending

Business share of lending 0.42 0.23 0.00 0.99

Personal share of lending 0.10 0.09 0.00 0.56

Housing share of lending 0.48 0.28 0.00 1.00

Loan growth 14.47 13.89 –40.53 93.28

Loan growth (winsorized) 14.69 11.34 –0.98 46.27

Dummy variable for state

government ownership

0.13 0.34 0.00 1.00

GDP growth

Business profits growth

Household disposable

income growth

Commercial property price

growth

Residential property price

growth

Cash rate

Average interest rate for

large businesses

Standard variable mortgage

rate

Economy-wide interest

burden

Business sector interest

burden

Household sector interest

burden

Total credit growth

Business credit growth

Housing credit growth

Change in the

unemployment rate

Business profits

growth 0.61

Household

disposable

income growth

–0.09 0.17

Commercial

property price

growth

0.30 0.37 0.57

Residential

property price

growth

0.24 0.23 –0.07 0.29

Cash rate –0.09 0.14 0.65 0.29 0.02

Average interest

rate for large

businesses

0.01 0.17 0.58 0.25 0.05 0.97

Standard

variable

mortgage rate

–0.03 0.16 0.57 0.24 0.03 0.95 0.96

Economy-wide

interest burden –0.27 –0.04 0.05 –0.27 –0.05 0.35 0.29 0.44

56

Business sector

interest burden –0.13 0.06 0.08 0.24 0.11 0.68 0.69 0.77 0.68

Household

sector interest

burden

–0.25 –0.15 –0.08 –0.18 –0.20 –0.37 –0.46 –0.33 0.58 –0.18

Total credit

growth 0.53 0.58 0.55 0.71 0.43 0.54 0.55 0.48 –0.19 0.08 –0.43

Business credit

growth 0.42 0.47 0.60 0.74 0.33 0.63 0.60 0.56 0.00 0.16 –0.26 0.94

Housing credit

growth 0.49 0.41 0.01 0.08 0.32 0.09 0.20 0.17 –0.32 0.14 –0.55 0.49 0.21

Change in the

unemployment

rate

–0.86 –0.57 0.15 –0.36 –0.25 0.25 0.15 0.16 0.39 0.29 0.23 –0.43 –0.28 –0.49

Inflation –0.16 0.17 0.75 0.52 0.05 0.82 0.79 0.76 0.09 0.30 –0.32 0.55 0.61 0.01 0.20

57

58

The results of Model C are robust to a number of alternative specifications. In

particular:

portfolios is supported statistically. If interactions between the household share

of the portfolio and these variables are added to a parsimonious version of

not significant. Some lending to small businesses in Australia is collateralised by

residential property, but an interaction between the business share of lending and

residential property price growth is not statistically significant if added to

Model C.

one year are not a common finding in the literature (see, for example, Hess

with both contemporaneous and lagged values of business profits growth,

household disposable income growth, and property prices, leads to very similar

estimated coefficients and significance. The only exception to this is that the

first lag of household income growth becomes significant at the 5 per cent level.

But this variable is only marginally economic significant – a one standard

deviation fall in household income raises credit losses by 13 basis points for a

representative bank, versus 24 basis points for a one standard deviation fall in

business profits.

and standard errors very similar to the fixed effects version of Model C in

Table 2. The only exception is the importance of the interaction between the

personal share of lending and the household sector interest burden. These

models estimate a strong and significant (at the 5 per cent level) relationship

between the household sector interest burden and losses on personal lending.

This result is intuitive, but is not particularly important at the bank level, given

the low portfolio share of this type of lending. A one standard deviation increase

59

in this macro-level variable increases credit losses by roughly 5 basis points for

a representative bank (with 10 per cent personal lending).

rate and inflation are insignificant if added to Model C, even if interacted with

portfolio shares.

restricted to a period that excludes the early 1990s downturn. For example, in a

model estimated using data for 1997–2013, the business sector interest burden

and commercial property prices remain statistically significant and have

coefficients similar to Model C, but business profits growth and bank-level loan

growth become insignificant.

the Arellano and Bond (1991) estimator. The lagged dependent variable has a

positive estimated coefficient and was significant at the 10 per cent level, though

unlike for stocks of non-performing assets, there is no strong reason to expect

true state dependence over short horizons in the flow of current losses.

Estimated coefficients on key explanatory variables in this model were

quantitatively similar to Model C. Banks learning from their mistakes may

create a negative relationship between credit losses in an earlier downturn and

those in a later downturn – something akin to the ‘institutional memory

hypothesis’ of Berger and Udell (2004). But only a small number of the banks in

the sample during the global financial crisis episode were not present during the

early 1990s downturn, so testing this hypothesis is difficult.

substantive change in results. This approach is likely not entirely robust, given

the number of clusters in both dimensions is below 50 (Cameron and

Miller 2013).

60

Constant –0.050

Constant –0.021

Constant 0.181

Notes: Robust bootstrapped standard errors clustered by bank; ***, ** and * denote significance at the 1, 5 and

10 per cent level respectively

(a) BSL = business share of lending; lagged one period

61

Basel Committee on Banking Supervision, Credit Risk Modelling: Current

<http://apra.gov.au/adi/Documents/Credit-Risk-Modelling-Current-Practices-and-

Applications-October-1999.pdf>.

Their Implications for Financial Aggregates’, RBA Research Discussion

Paper No 8904.

13(4), pp 458–495.

Statement before the Financial Crisis Inquiry Commission ‘Too Big to Fail:

Expectations and Impact of Extraordinary Government Intervention and the Role

of Systemic Risk in the Financial Crisis’, Washington DC, 2 September.

Framework and Results’, Report, March.

Inference’, Unpublished manuscript, University of California – Davis, October.

62

Press, College Station, Texas.

Reserve Bank of Australia, Sydney, pp 21–23.

Companies: A Dynamic Panel Quantile Regression Approach’, Board of

Governors of the Federal Reserve System Finance and Economics Discussion

Series No 2013-55.

pp 1341–1359.

Mortgage Portfolios by Decomposing the Mean’, Paper presented at the 12th

INFINITI Conference on International Finance ‘Global Finance – Integration or

Mere Convergence’, Prato, 9–10 June.

Conference, Reserve Bank of Australia, Sydney, pp 50–68.

Lending be Reduced? Evidence from Australian Banks’, Australian Prudential

Regulation Authority Working Paper 2002-02.

63

Other Support for Troubled Financial Institutions in Australia’, RBA Research

Discussion Paper No 2001-07.

pp 16–22.

for International Settlements, Basel, pp 481–497.

Profitability’, RBA Research Discussion Paper No 2001-06.

Proceedings of a Conference, Reserve Bank of Australia, Sydney, pp 180–215.

Losses – Results of a Panel Data Study’, National Bank of Poland Working

Paper No 53.

64

pp 392–408.

Stability Assessment’, IMF Country Report No 12/308.

pp 57–75.

1985/86 to 1988/89’, RBA Research Discussion Paper No 9106.

Australia, Sydney, pp 300–317.

Specific Determinants of Non-Performing Loans in Greece: A Comparative Study

Impact of Commercial Credit Standards on Lending and Output’, Federal Reserve

65

Reserve Bank of Australia, Sydney, pp 175–199.

Supervision, and Low Interest Rates: Evidence from the Euro-Area and the U.S.

Panel Investigation’, Bank of England Working Paper No 177.

Conference ‘Damage Control: Credit Risk Management for Banks and Financial

Institutions’, Organised by Australian Investment Conferences, Sydney,

4–5 September.

22(3), pp 203–224.

& Unwin, Sydney.

Bank of Australia, Sydney, pp 5–20.

66

Government Printer, Melbourne.

in Mature Economies’, Basel Committee on Banking Supervision Working

Paper No 13.

2nd edn, MIT Press, Cambridge.