RDP 2022-01: MARTIN Gets a Bank Account: Adding a Banking Sector to the RBA's Macroeconometric Model 1. Introduction

Despite the real economy and banking system being inextricably intertwined, within the Reserve Bank of Australia (RBA) they are currently modelled separately. In the RBA's macroeconometric model, known as MARTIN (Ballantyne et al 2019), the difference between banks' mortgage rates and the RBA's cash rate is treated as exogenous (i.e. determined outside the model). This means that economic downturns in the model do not feed back into the banking system and change the interest rates banks charge borrowers or the amount they are willing to lend (i.e. there is no feedback to ‘credit supply’). Analogously, the RBA's bank stress testing model treats the macroeconomic scenario as exogenous, and therefore does not consider the effect a stressed banking system may have on the macroeconomy (RBA 2017).

In MARTIN, treating the banking system as exogenous does not typically lead to large model inaccuracies. This is because the interest rates banks charge borrowers ensure they are sufficiently profitable to weather most downturns without a deterioration in their capital levels. However, when faced with large downturns – such as what some countries experienced during the global financial crisis or what was feared could result from COVID-19 – loan losses may eat into banks' capital, and they may respond by increasing their loan interest rates and/or reducing the amount they are willing to lend. This response from banks amplifies the downturn – higher interest rates and reduced lending lead to lower housing prices, lower business investment and lower consumption, thereby leading to higher unemployment and a further increase in loan losses – leading to an amplified response from banks and even further amplification of the downturn. This amplification cycle is known as a ‘financial accelerator’ mechanism.

While the existence of financial accelerator mechanisms was known before the global financial crisis (Kiyotaki and Moore 1997; Bernanke, Gertler and Gilchrist 1999), it was this crisis that highlighted the failure of modern macroeconomics to fully appreciate their size and likelihood of occurring (Lindé, Smets and Wouters 2016; Gertler and Gilchrist 2018). So it is important for the RBA to have a modelling framework that incorporates at least some of these accelerator channels. Unfortunately, the best way to incorporate these channels is not obvious, as the literature has advanced in a myriad of different directions (see Appendix A for a literature review).

To fill this gap in the RBA's modelling repertoire, we build an extension to MARTIN that incorporates one of the key financial accelerator mechanisms – a banking sector that endogenously and nonlinearly changes credit supply in response to loan losses and/or changes to their funding costs. Consistent with the way MARTIN was originally designed, our approach captures how RBA staff currently model the banking system; we achieve this by basing our extension on the existing modelling frameworks used within the RBA.

The determinants of banks' funding costs are based on the RBA's funding cost model (Davies, Naughtin and Wong 2009; Brassil, Cheshire and Muscatello 2018). The determinants of household loan losses are based on the micro-simulation model developed by Bilston, Johnson and Read (2015) and updated by Kearns, Major and Norman (2020). How banks respond to these losses is based on the RBA's bank stress testing framework (RBA 2017).

Incorporating these disparate frameworks into MARTIN, while making minimal changes to the original MARTIN design, obviously requires compromises. For example, the lack of recent financial crises in Australia prevents us from using time series data to estimate the necessary relationships (the method used in the original MARTIN design). Instead, the additional MARTIN equations are calibrated from the aforementioned RBA frameworks, microdata from the Survey of Income and Housing, and the Australian Prudential Regulation Authority's (APRA) stress testing results. We also have less modelling freedom than the aforementioned frameworks, as MARTIN already includes equations for some important relationships – the relationship between credit growth and interest rates, for example.

By combining a large and complex macroeconometric model (MARTIN) with a micro-simulation model, and nonlinear stress testing and funding costs frameworks, our approach moves beyond the existing macroeconometric frontier (see Appendix A for a literature review).

After providing a summary of the new banking sector mechanisms (Section 2) and detailing the new components (Section 3), we illustrate how this new modelling framework (henceforth, banking-augmented MARTIN or BA-MARTIN) improves our ability to model the economy and allows us to answer important policy questions that could not previously be adequately answered. Specifically, we show:

1. how the inclusion of a ‘financial accelerator’ mechanism changes how large shocks are transmitted through the economy (Section 4); and
2. how the pass-through of monetary policy to banks' lending rates depends on the level of interest rates and the state of the economy (Section 5).

1.1 BA-MARTIN's financial accelerator mechanism

We illustrate the financial accelerator mechanism by showing how one of the more pessimistic economic scenarios that could have resulted from COVID-19 might have affected the banking sector, and subsequently the supply of credit to the Australian economy. We do this by feeding the downside scenario from the May 2020 Statement on Monetary Policy (RBA 2020c) into BA-MARTIN and then explore how much worse predicted economic outcomes would be with the additional financial accelerator mechanism. Had this scenario eventuated – with GDP 12 per cent below pre-COVID-19 forecasts and the unemployment rate surpassing 10 per cent – we estimate that loan losses would have been sufficient to reduce banks' capital. Without additional public policy support, housing prices two years after the onset of the crisis would have fallen an additional 3 per cent and 16,000 fewer people would have been employed (a 0.1 percentage point increase in the unemployment rate).

We have received feedback that the effect on the economy from our financial accelerator mechanism seems small considering how much damage the global financial crisis did to the global economy (Guerrieri and Uhlig 2016; Bernanke 2018). But there are two important distinctions between the Australian banking sector and those overseas. First, the Australian banking sector has an ‘unquestionably strong’ capital framework (APRA 2017), tends to be highly profitable (FSRC 2018), and is lower risk than other countries' banking sectors (RBA 2012). So it can weather larger storms.

Second, the majority of assets held by Australian banks are loans that can be repriced at short notice. Most loans are variable rate, and even the fixed-rate loans tend to have rates fixed for less than three years (far less than the 30-year mortgages common in the United States). Moreover, Australian variable-rate loans are not explicitly indexed to any market rate, unlike the ‘tracker’ loans that are common in some European countries (Lea 2010). This allows Australian banks to spread the cost of unexpectedly high losses across both new and outstanding loans, which leads to a much smaller effect on economic activity than if the cost were borne by new loans only (e.g. by the banks tightening lending standards, which was a common response by US and European banks during the global financial crisis (Maddalonia and Peydró 2013; Bassett et al 2014)).

That said, BA-MARTIN has the flexibility to explore what would happen if banks restricted new loans only. In the same downside scenario, and without additional policy support, housing prices two years after the onset of the crisis would have fallen an additional 12 per cent and 59,000 fewer people would have been employed (a 0.4 percentage point increase in the unemployment rate).

To be absolutely clear, these amplified numbers are scenarios not forecasts, and are designed solely to show the quantitative importance of the financial accelerator mechanism. Forecasted amplifications would be smaller, as some combination of fiscal, monetary and regulatory policies would respond to the amplified downturn.

1.2 BA-MARTIN's state-dependent monetary policy pass-through

Pass-through is affected by the level of interest rates because deposit rates tend to have a lower bound around zero – due to the possibility of holding physical currency instead (Garner and Suthakar 2021; Hack and Nicholls 2021). With deposit rates unable to move below zero, cash rate reductions cause smaller reductions in banks' funding costs than when this lower bound does not bind.

Pass-through also changes when the banking system is stressed (i.e. when losses are sufficient to reduce banks' capital). When the banking system is stressed, further reductions in net interest income or increased loan losses will lead to further contractions in credit supply. Cash rate reductions lead to reduced net interest income but they also reduce losses, with the latter effect typically dominating during periods of stress. As a result, pass-through can be greater than 100 per cent because the cash rate cut moderates the credit supply contraction in addition to the usual reduction in banks' funding costs.