RDP 2022-07: The Term Funding Facility: Has It Encouraged Business Lending? Appendix A: Data Description

Our analysis primarily draws on confidential data reported to APRA by banks and non-banks. We draw mostly on data collected under the EFS collection, which have been available since mid-2019.

For our analysis examining whether the incentive in the TFF's additional allowance was effective, we largely draw on data from EFS form ARF 742_0 A/B, which contains data on business credit stocks, flows and interest rates. This form also contains breakdowns of business credit by firm size and industry (e.g. manufacturing). Banks and non-banks that have $2 billion of business credit or more are required to report under this form; this threshold is estimated to capture over 95 per cent of total business credit.

For our first analysis comparing SME and large business credit growth for banks, we draw on data on outstanding business credit by size from form ARF 742_0 A/B.^[23] We supplement this with data reported directly to the RBA for the purposes of calculating the TFF additional allowance by banks who did not report on form ARF 742_0 A/B; 15 banks chose to report SME and large business credit outstanding directly to the RBA from November 2019 until April 2021.^[24],[25] These data are adjusted to account for breaks in the series, mostly to account for reclassifications between business sizes. In total, 50 banks reported business lending data by size throughout the drawdown period, although our baseline model in the first analysis excludes 2 banks (outliers) that reported extremely large growth rates for some periods.^[26]

Form ARF 742_0 A/B also contains a number of subcomponent breakdowns for business credit outstanding, including lending to businesses in each of 25 different industries. We group these industries into four broad groups (household services, business services, goods production and goods distribution) due to data volatility at the more granular industry level.^[27],[28] The industry dataset allows us to run a robustness check on our baseline estimates using data at the bank–industry level. The key rationale for running models at the bank–industry level is to control for uneven effects of the pandemic across industries, based on the notion that businesses within these broad industry groups were similarly affected by COVID-19 restrictions. Unlike our first analysis, the industry-level data are not break adjusted to account for reclassifications between business sizes, owing to a lack of readily available data.

The ARF 742_0 A/B form also includes subcomponent breakdowns for business credit outstanding by loan type. From this, we construct a dataset that is limited to fixed-term loans (defined as loans extended for a fixed period and with a maturity date by which the loan must be repaid). Our rationale for running models with just fixed-term lending is to look through larger businesses drawing down on lines of credit for precautionary reasons; banks would have likely expected these loans to be short term and to not influence the final additional allowance. Indeed, these firms generally repaid these funds within a couple of months.

Data on bank characteristics are sourced from various APRA forms to examine whether the effect of the TFF on business credit growth differed by bank characteristics such as size, liquidity and capital ratios.^[29] We use data as at December 2019 (prior to the TFF) to avoid endogeneity issues. Descriptive statistics for a subset of banks are in Table A1; these statistics cover the 50 banks that report the breakdowns of business credit by firm size, although some of these statistics cover a smaller sample of banks due to a different set of reporting thresholds than EFS forms ARF 720_1 and ARF 742_0.

For our analysis examining whether the availability or take-up of the TFF affected lending, we use data from EFS form ARF 720_1 A/B, which contains data on loans and leases for a larger set of banks than ARF 742_0 A/B. Form ARF 720_1 A/B includes data on total business credit (outstanding loans and finance leases) to community service organisations, non-financial businesses and financial institutions. This form captures monthly data from most banks, as well as non-banks with more than $400 million in total assets. While this form does not include a breakdown of business credit by firm size, its lower thresholds relative to ARF 742_0 A/B means it has the advantage of a larger number of reporting institutions. We use a balanced panel of 114 banks and 48 non-banks in our analysis.

Aside from data reported to APRA through the EFS forms, we use RBA data on the availability (or adequacy) of banks' self-securitised assets as the instrumental variable in our analysis of how banks' participation in the TFF affected lending. To avoid endogeneity issues we use data as at May 2020, prior to some banks establishing self-securitisations to facilitate accessing the TFF. Of the 114 banks in our sample, 46 had self-securitised assets established prior to the TFF.

Footnote

Under the EFS collection, businesses with turnover greater than or equal to $50 million are classified as large businesses. For businesses with turnover less than $50 million, when the lender has an exposure of more than $1 million, the business is classified as medium. When the exposure is less than $1 million, the business is classified as small. [23]

Data reported by these 15 banks are used in our baseline SME regression (Equation (1)) and our triple-difference regression (Equation (5)). As banks with less than $2 billion in business credit outstanding could choose whether to report SME and large business credit directly to the RBA, this poses a potential self-selection bias (as these banks may have been more likely to increase lending anyway). As a check, we ran these regressions without these 15 banks and also found no statistically significant effects of the TFF on business credit. [24]

This results in a break in two of our regression results (Equations (1) and (5)). In our coefficient figures, we have chosen to present the full set of reporters for a smaller period of time. Coefficient results up to October 2021 are presented in our regression tables in Appendix C, but include a series break at May 2021. [25]

These banks did not access the TFF and had very volatile growth rates. We also test the robustness of our results to outliers by running weighted regressions. [26]

Banks may reclassify loans between industries following data reviews. Grouping into broader industry groups may reduce the risk that reclassification leads to big movements in the data (if industries are more likely to be reclassified to other industries within the broad group). [27]

Lending to the following industries are excluded from our analysis: financial and insurance (lending to the RBA, authorised deposit-taking institutions, registered financial corporations and central borrowing authorities); and public administration and safety. [28]

This includes forms ARF 720_0 A/B, ARF 720_1 A/B, ARF 720_2 A/B, ARF 330_0_l, ARF 330_1_L, ARF 110_0_1 and ARF 323_0. [29]