Financial Stability Review – October 2020Box B: Business Failure Risk in the COVID-19 Pandemic – Technical Appendix

This Technical Appendix outlines the assumptions underpinning the scenario analysis in Box B. It also discusses some of the ABS BLADE data definitions and limitations and how the data are cleaned prior to analysis.

The Business Longitudinal Analysis Data Environment (BLADE)

The BLADE database has information on sales, employment and value-added for sole traders but not balance sheet information, such as assets and liabilities. Because of the focus on balance sheets in Box B, sole traders are excluded from the analysis.

Following ABS definitions, the analysis focuses on ‘active’ businesses that pay the goods and services tax (GST) which means the sample is restricted to businesses that make at least $75,000 a year in revenue (this is the cut-off for GST payment). Extreme observations are deleted based on the following criteria: firms reporting negative sales or assets or expenses; the cash-to-assets ratio is greater than 100 per cent; the ratio of current assets to current liabilities exceeds 1,000 per cent; the ratio of cash to total expenses exceeds 10,000 per cent; the ratio of debt to assets exceeds 1,000 per cent; and the return on assets is greater than 1,000 per cent or less than −1,000 per cent.

A direct measure of firm failure or insolvency is not available in BLADE. Instead, a firm is assumed to fail when it exits, holds debt and is in negative equity. A direct measure of cash holdings is also not available in BLADE as this information is not collected in tax returns. Instead, ‘cash’ is meant to measure both cash on hand and bank deposits, and is defined as total current assets less accounts receivable and the closing stock of inventories.

Scenario assumptions

The analysis considers three scenarios:

  1. a COVID-19 shock scenario (with a sharp decline in business revenue and no policy intervention);
  2. a COVID-19 policy scenario (with the same decline in revenue and policy intervention); and
  3. a counterfactual ‘normal times’ scenario based on firms' 2017/18 balance sheets.

For the ‘shock’ scenario, the economic downturn is modelled as a decline in business cash flow due to both lower revenue and a limited ability for firms to adjust expenses. The scenario is based on a wide range of sources, including timely indicators of business revenue, Bank forecasts and liaison. Judgement has also been applied as revenue forecasts that exclude the effect of the policy measures are not readily available.

In each year, there is a ‘shock’ to the revenue of each firm that has both an industry-specific component and a random firm-specific component. So some industries get hit harder than others by the downturn and even within the hardest hit industries there are some firms that suffer more.

Those firms operating in ‘more affected’ industries face larger negative revenue shocks in the June quarter 2020. These hard-hit industries include accommodation and food services and arts and recreation services. All industries are assumed to have no recovery in revenue during 2020/21.

Firms can partly offset changes in revenue by changing their variable costs, including labour, intermediate input and even some fixed costs. The sensitivity of these input costs to changes in revenue is estimated from historical relationships (with elasticities of around 0.75, 1.10 and 0.40 for labour costs, intermediate input and other costs, respectively).[1]

For the ‘policy’ scenario, four separate income support policy interventions are considered:

  • JobKeeper (JK): This program reduces labour costs for eligible businesses. The share of businesses in each industry that received JK payments in April or May 2020 are included in the BLADE data environment (Graph B.7). For the firms that report receiving JK, labour costs are assumed to decline by the amount of the subsidy ($1,500 per worker per fortnight in the June quarter 2020, $1,500 in the September quarter 2020, $1,000 in the December quarter 2020 and $800 in the March quarter 2021).
Graph B.7
Businesses Receiving JobKeeper by Industry
  • Cash Flow Boost for Employers: This program gives between $20,000 and $100,000 to businesses with annual turnover of less than $50 million and that employ staff, so long as they have made payments to employees subject to tax withholding. Eligible businesses receive two one-off payments (the first in the June quarter and the second in the September quarter). These payments are equivalent to business PAYG tax withholdings for the year capped at $100,000, and all eligible firms receive at least $20,000 in total.
  • Loan repayment deferrals: the interest expenses of businesses with debt are assumed to be equal to zero in the June and September 2020 quarters and then revert to their 2017/18 level in the December 2020 quarter (and beyond). This assumes that every indebted business is given a loan repayment deferral, and the lender does not capitalise the interest. This could be interpreted as an upper bound on the income support given that about one-fifth of firms have received loan repayment deferrals. However, this assumption also does not allow for relief in terms of principal payments, which works in the opposite direction and may underestimate the effect of the deferrals on business cash flow.
  • Rent reductions: The rent expenses of each business is reduced in proportion to their estimated fall in revenue consistent with the principles set out in the Government's Mandatory Code of Conduct. The Code of Conduct entitles eligible tenants facing financial hardship to request rent reductions during the pandemic.

Firm-level model

The probability of failure at the firm-level is estimated as a function of various balance sheet indicators, which are outlined below.

  • Cash flow (expressed as a ratio to total assets): over the long term, business survival is not possible without cash flow. Hence, a firm with low cash flow is more likely to cease operations due to poor performance.
  • Gearing (ratio of debt to assets): firms that have relatively high gearing are more vulnerable to asset devaluations as they have a relatively small equity buffer. This makes them more likely to go bankrupt. Moreover, high gearing is likely to be associated with more debt serviceability issues. Debt is defined as total non-current liabilities, as debt is not reported separately in the data.
  • Liquidity (ratio of current assets to current liabilities): measures a company's ability to finance its short-term debts (obligations payable within the coming year). All else being equal, a lower ratio implies that a firm will have more trouble meeting its financial obligations and is therefore likely to be associated with having a higher failure risk.
  • Size (log of number of employees): small firms may be more likely to fail than large firms as small firms are more likely to be credit constrained and presumably have less bargaining power with suppliers and creditors. All else being equal, size should be negatively associated with the probability of failure.

The distribution of business failure rates across the various balance sheet indicators used in the model are shown below (Graph B.8).[2] Overall, the results indicate that, prior to the pandemic, firms were more likely to fail when they were smaller, younger and less profitable, had fewer liquid assets and were more leveraged.

Graph B.8
Business Failures

The firm-level logit model is specified as:

l o g ( F A I L U R E i t 1 - F A I L U R E i t ) = β C A S H F L O W i j t + γ C O N T R O L S i j t + μ j + ε i j t


  • FAILURE is a dummy variable that is equal to one if firm i in industry j failed at time t and is zero otherwise.
  • CASHFLOW is the cash flow of the business and is measured as the ratio of cash flows to total assets in the current financial year. This is a key variable of interest and so is separated from the other balance sheet indicators. Cash flow enters the regression as a series of dummy variables to allow for non-linear effects between firms' profitability and likelihood of failure.
  • CONTROLS consists of other time-varying balance sheet indicators such as liquidity, leverage and firm size. It also includes some business demographics such as age (and its square) as well as a dummy variable that equals one if the firm is part of an enterprise group and is zero otherwise.
  • µ are industry fixed effects which control for unobserved time-invariant industry characteristics at the 4-digit ANZSIC level that explain why some industries are inherently riskier than others.

The model is estimated on a sample period covering the financial years from 2002/03 to 2015/16 and comprises more than 6 million observations. The results are shown in Table B.1 and discussed in Box B.

Table B.1: Business Failure Model Estimates
  Probability of failure
Logit OLS
Coefficients dy/dx Coefficients
Cash flow (per cent)
< −50 Excluded category
−50 to −25 0.3976***
0.007 −0.0018
−25 to −10 0.367***
0.006 −0.0032*
−10 to −5 0.193***
0.003 −0.0093***
−5 to 0 −0.042
−0.0006 −0.0143***
0 to 5 −0.649***
−0.007 −0.0213***
5 to 10 −0.878***
−0.008 −0.0228***
10 to 25 −1.068***
−0.009 −0.0237***
25 to 50 −1.290***
−0.010 −0.0244***
50+ −1.468***
−0.011 −0.0247***
Gearing 0.543***
0.0029 0.0205***
Liquidity −0.180***
−0.001 −0.0014***
Age −0.098***
−0.0005 −0.00094***
Age*Age 0.0014***
0.0000 0.00002***
Size −0.103***
−0.0006 −0.00077***
Conglomerate 0.112
0.0006 0.0006
Company 0.159***
0.0009 0.003***
Partner 0.254***
0.0015 0.002***
Constant −4.203***
Year fixed effects No No
Industry fixed effects Yes Yes
Pseudo R-squared/Adjusted R-squared 0.1232 0.0259
N 6,717,933 6,717,933

Note: ***,**,* denote statistically significant at the one, five and ten per cent levels, respectively. Standard errors are shown in parentheses and are clustered at the industry level. Margins are calculated at the means.

Sources: ABS; RBA


In practice, these elasticities are allowed to vary by industry. [1]

The graph shows unconditional correlations between failure rates and each of cash flow, liquidity and leverage. It does not show the conditional relationships that will be captured in the regression model. The model allows for the fact that cash flow, liquidity and gearing can be correlated. [2]