RDP 2022-08: The Consequences of Low Interest Rates for the Australian Banking Sector 3. The Literature through the Lens of Banks' Balance Sheets

Following Brassil et al (2018), a bank's balance sheet can be split into the components over which it has little control over pricing (henceforth, ‘non-discretionary’) and those over which it has some pricing power (henceforth, ‘discretionary’). Components could be non-discretionary because the bank is a price-taker in the relevant market (e.g. wholesale debt funding markets) or because its determinants are mostly outside the control of the bank (e.g. losses on outstanding loans). Table 1 shows a stylised version of a bank's balance sheet split into non-discretionary and discretionary components. The remainder of this section will look at how low interest rates might affect each part of banks' balance sheets and will explore the domestic and international evidence.

In Australia, more than 60 per cent of the major banks' funding comes from domestic deposits. And of this, just under half are owned by households (Fitzpatrick, Shaw and Suthakar 2022). The non-major banks have a slightly lower share of domestic deposit funding (around 55 per cent on average).

3.1 Non-discretionary

3.1.1 Low-interest retail deposits

If the interest rates banks paid on the deposit accounts of households and small businesses went sufficiently negative, these savers may choose to instead hold their liquid assets in physical currency. This risk tends to prevent banks from reducing the interest rates on these accounts too far below zero; the interest rates on these accounts effectively have a lower bound.

Deposit accounts pay a range of interest rates. Some accounts always pay rates close to zero, while others typically pay some spread above the cash rate (or expected future cash rates). Therefore, as the level of interest rates falls, an increasing share of banks' deposit accounts will hit the effective lower bound (ELB). Once these rates hit the ELB, any further reductions in the cash rate will not be passed through to these accounts. As a result, when the level of interest rates is low, the spread between banks' cost of deposit funding and the cash rate can increase with further reductions in the cash rate.^[5]

For Australian banks, estimates from Brassil et al (2022), based on data from Garner and Suthakar (2021), suggest that it is once the cash rate falls below 1.5 per cent that the share of deposits at the ELB begins to increase (Figure 1). Brassil et al (2022) further show that the effect this ELB has on the pass-through of monetary policy is significant; at a cash rate of zero, the ELB would cause only 20 basis points of any subsequent 25 basis point reduction to be passed through to banks' debt (including deposits) funding costs.

The international evidence on the existence of a retail deposit ELB is consistent with the Australian evidence. In jurisdictions with a negative policy rate, retail deposit interest rates remained close to zero (albeit negative). In Denmark, where the policy rate was –50 basis points or lower between 2015 and 2022, banks only introduced negative rates on large retail deposit balances (Kuchler, Spange and Krogstrup 2020). While in the euro area, only 5 per cent of retail deposits faced negative rates by the end of 2020 (Altavilla et al 2022).

In some jurisdictions, banks have increased fees to offset the cost of the ELB (CGFS 2019). To date, there is no evidence of Australian banks increasing fees to compensate for the deposit rate ELB (Sparks and Garner 2021).

3.1.2 Wholesale debt and deposits

Wholesale debt and deposits are less likely to have an ELB near zero. This is because the values of these bank debts/deposits are typically much larger than retail deposits, making the cost for the creditor/depositor of storing an equivalent volume of physical currency, and the associated increased transaction costs, sufficiently prohibitive that physical currency is not a cost-effective substitute at small negative interest rates.

The international experience with negative interest rates is consistent with this assumption. In the euro area, 35 per cent of non-financial corporate deposits paid negative rates of interest by 2020 (Altavilla et al 2022), with the equivalent share in Denmark being three-quarters (Kuchler et al 2020). International evidence suggests complete pass-through of negative policy rates to money market rates (CGFS 2019).

The Australian experience with low (but positive) rates has been similar to the international experience. Australian banks' wholesale debt and deposit funding is typically either directly referenced to bank bill swap rates (BBSW) or hedged to these rates (Fitzpatrick et al 2022). And there has been full pass-through of policy rate changes to BBSW (Aziz et al 2022).

3.1.2.1 Unconventional policies typically used at low interest rates

While the cash rate target has remained above zero in Australia, the RBA implemented several unconventional policies designed to stimulate the economy. These unconventional policies are typically used by central banks when interest rates are low, and so should be considered a part of the central bank's toolkit at low rates.

The Term Funding Facility (TFF) provided banks with a source of funding that was significantly cheaper than issuing bonds of the same maturity (Graph 10 in Fitzpatrick et al (2022)). Banks therefore issued fewer bonds in favour of borrowing directly from the RBA via the TFF. In addition to the direct effect on banks' costs of funding, the lower bond issuance reduced the supply of banks' bonds, thereby lowering yields on the bonds that were issued during 2020–21 (Fitzpatrick et al 2022).

The quantitative easing measures implemented by the RBA partially operated through a ‘portfolio rebalancing’ channel that would further reduce the cost of bond issuance (CGFS 2019; Finlay, Titkov and Xiang 2022). Moreover, to the extent that bank debt is perceived as riskier than government debt, the purported ‘search for yield’ behaviour of investors at low interest rates would have further reduced banks' costs of funding (Simon 2015).

The RBA's unconventional policies were able to reduce the major banks' funding costs by a sufficient amount that the deleterious effect of the retail deposit ELB was (potentially more than) offset (Fitzpatrick et al 2022). Therefore, over the decade to 2021, the major banks' overall costs of debt funding fell by a similar amount to the cash rate, even though the pass-through of cash rate changes to these funding costs was more muted.

3.1.3 Central bank deposits

Central bank deposits in Australia are remunerated at a rate that moves with the cash rate target. If these central bank deposits are funded by liabilities with interest rates that also move with the cash rate, then banks' net income from holding central bank deposits does not necessarily change at low interest rates. This was the case during the pre-COVID-19 period in Australia, when deposits held at the RBA were mostly funded by short-term repurchase agreements and foreign exchange swaps (Dowling and Printant 2021) (Figure 2).

Unconventional policies implemented during COVID-19 changed how central bank deposits are funded. Central bank deposits are now mostly funded by selling government bonds to the RBA, and the TFF (Dowling and Printant 2021). In turn, these policies increased the value of banks' deposit funding (Fitzpatrick et al 2022). As a result, the lower pass-through of monetary policy to retail deposit rates lowers the net income from holding central bank deposits as the cash rate falls.

Several jurisdictions that have implemented negative interest rates have offset this reduced net income by implementing a ‘tiered’ reserves system, whereby only a subset of central bank deposits are remunerated at the negative rate (Fuhrer et al 2021; Hack and Nicholls 2021).

3.1.4 Securities

As with wholesale debt and deposits, the pass-through of policy rate changes to newly purchased securities is expected to be no different at low interest rates. The market value of fixed-coupon securities already held by banks will also change as interest rates change, leading to two channels through which lower interest rates could affect the banking sector.

The first is the capital gains banks would receive from these securities as interest rates fall. According to the theoretical model of Brunnermeier and Koby (2018), these capital gains provide a valuable, albeit one-off, boost to banks' profits that staves off the drastic changes in their lending behaviour that are due to net income reductions from their lending operations. Such an effect was found in an empirical analysis of euro area banks (Altavilla, Boucinha and Peydró 2018).

The second channel is interest rate risk. As the level of interest rates falls, the duration – broadly defined as the sensitivity of a bond's price to interest rate changes – of fixed-interest securities increases. The neutral rate (discussed in Section 2) is typically modelled as a random walk with constant shock variance. Assuming this is a reasonable characterisation of the behaviour of the neutral rate, then for a given asset portfolio, the combination of higher duration and constant neutral interest rate volatility means the long-run interest rate risk of the securities held by banks is higher when the level of interest rates is lower.^[6] Unlike the one-off capital gains during the transition, this higher interest rate risk persists as long as the average level of rates remains low, and could require banks to hold more capital.

These channels are likely to be small in Australia, as Australian banks' holdings of fixed-income securities in their trading books amount to only 3 per cent of their assets (RBA 2022).

3.1.5 Losses

When assessing the effect of low interest rates on banks' loan losses, it is helpful to distinguish between economic downturns (that may become more frequent and/or severe in a low-rate world), and the new low-rate steady state.

3.1.5.1 Economic downturns

Holding interest rates constant, an economic downturn increases losses. Business revenues will fall, unemployment will increase and property prices will fall, all of which reduce both the ability of borrowers to repay their loans and the values of the assets these loans are secured against. Recent Australian studies exploring the quantitative relationship between macroeconomic outcomes and loan losses include Hess, Grimes and Holmes (2009), Bilston, Johnson and Read (2015), Rodgers (2015), Cummings and Durrani (2016), Brassil et al (2018), Bergmann (2020), Kearns, Major and Norman (2020) and Garvin et al (2022).

The easing of monetary policy that results from the central bank responding to the macroeconomic deterioration reduces losses (relative to a counterfactual in which interest rates were not lowered). The policy easing increases aggregate demand, thereby increasing business revenues and property prices, and reducing unemployment. In Australia, with variable-rate loans more common than fixed-rate loans, the policy easing also directly reduces borrowers' interest payments, thereby improving their ability to repay their loans (this direct channel is not a feature of US and some European jurisdictions with high shares of long-term fixed-rate loans).

For the period 2002–17, Brassil et al (2018) found that a 100 basis point reduction in the cash rate reduced the major banks' annual loss provisioning rates by 7 basis points. While this may seem small, when compared with an average annual provisioning rate of just 23 basis points during this period, monetary policy is quite effective at reducing loan losses. In order to construct a more forward-looking and nonlinear response of losses to interest rate changes, Brassil et al (2022) use the micro-simulation model of Kearns et al (2020) to determine the extent to which monetary policy may be able to reduce losses in future downturns. They find that a 100 basis point reduction in the cash rate is expected to moderate any downturn-induced increase in losses by between 3 and 17 per cent (with the larger effectiveness occurring when the economic deterioration is larger).

The international literature has also found that interest rate reductions reduce banks' loan losses. Consistent with the preponderance of longer-term fixed-rate loans in the United States, Bikker and Vervliet (2017) find a 100 basis point short-term interest rate reduction is associated with a 3 basis point reduction in loss provisioning (compared with 7 basis points in Australia). Using a panel of international banks from 14 advanced economies, Brei, Borio and Gambacorta (2019) find an asymmetric effect that is largest at low interest rates (13 basis point reduction in provisioning with a 100 basis point short-rate reduction to zero per cent). Older studies by Albertazzi and Gambacorta (2009) and Bolt et al (2012) find similar results.

With a focus on the euro area, Altavilla et al (2018) stress the importance of properly controlling for the endogeneity of interest rates to both the current and expected future state of the economy.^[7] In a stylised macro model designed to determine the effectiveness of monetary policy at low interest rates (including unconventional policies), an expansionary unconventional policy that reduces 10-year yields by 100 basis points reduces annual loss provisioning by a peak of 10 basis points. This is similar to the estimated effects of conventional policies discussed in previous paragraphs, suggesting that the ability of central banks to moderate loan losses does not diminish at low interest rates.

3.1.5.2 The new low-rate steady state

All of the previous studies reflect the effect monetary policy would have on losses during economic downturns (i.e. when loss provisioning would otherwise be high), which may be more frequent in a low-rate world to the extent that there would be more economic volatility in an environment where monetary policy is more frequently constrained. But banks' general level of loss provisioning could also be different in a low-rate world.

If credit risk is higher in the low-rate world, banks' interest spreads on new loans should be higher to compensate, and their general level of provisioning should also be higher. There could also be a higher risk of banks' provisioning increasing sharply given that the duration of the assets used to secure their loans will be higher (as explained in Section 3.1.4).

Given that the transition to the new low-rate world is either still continuing or had only just finished prior to the COVID-19 shock, recent studies may not yet fully capture the ongoing effect of low interest rates on banks' loan losses. That said, in Australia just prior to the COVID-19 crisis, banks' provisioning for loan losses were no higher than prior to the global financial crisis (despite the lower level of interest rates).

If higher wealth inequality is either a cause of (Mian, Straub and Sufi 2021b), or consequence of (Dollman et al 2015), the lower level of interest rates, this could eventually lead to more of the debt in the economy being held by those who are more at risk of being unable to service this debt, thereby increasing credit risk. That said, such an outcome is not a forgone conclusion; empirical studies produce mixed results (Coibion et al 2014; Kumhof, Rancière and Winant 2015; Papadopoulos 2019), and face substantial endogeneity challenges (Bazillier and Hericourt 2017). Recent estimates continue to suggest that the majority of debt in Australia continues to be held by those who are most able to service this debt (RBA 2021).

It is important to remember that the low-rate world results from structural changes in the economy to which the central bank responds.^[8] The economy would be in much worse shape if central banks tried (in vain) to return interest rates to their previous levels. So the potential for increased credit risk in a low-rate environment to lead to higher losses for banks should be considered second-order relative to the benefit of setting interest rates appropriately for the changed macroeconomic environment.

3.2 Discretionary

Although limited by competitive pressures and regulations, banks have some discretion over the interest rates they set for their loans, the credit quality of the loans they offer, and the interest rates they pay on high-interest deposit accounts. When combined with the interest rates on their non-discretionary assets and costs of funding, these decisions combine to form banks' net interest margins (NIMs, a key component of profitability). Because banks' discretionary loan and deposit decisions map directly to changes in profitability, I will not evaluate the various discretionary components separately. Instead, I will discuss how low levels of interest rates might affect the discretionary decisions of profit-maximising banks (including how their constraints may change).

Low interest rates reduce the pricing discretion banks have on their deposits, due to the retail deposit ELB. This occurs both because the rates paid on existing at-call accounts progressively hit the ELB as the policy rate falls, and because depositors switch from the less liquid term deposits to at-call accounts as the interest rate differential compresses. Both of these mechanisms featured in Australia as interest rates fell (Figure 3).

For over a decade to 2021 (including during COVID-19), Australia's major banks set interest rates on their loans to maintain a relatively stable spread between the interest received on these loans and their costs of debt (including deposit) funding (Figure 4) – henceforth, the ‘lending spread’. This is despite the level of interest rates falling several percentage points during this period.

Interestingly, this broadly stable lending spread does not seem to be a feature of many other jurisdictions. With the literature highlighting three main reasons:^[9]

• Competition from institutions with lower reliance on deposit funding means the marginal lending rate is more closely tied to wholesale funding costs (rather than banks' overall cost of funding) in other jurisdictions. As a result, deposit-reliant banks' lending spreads compress as more deposit accounts hit their ELB.
• Hedging of interest rate risk by Australian banks – either by maturity matching or via derivatives – means changes in the slope of the yield curve do not affect their lending spreads (Brassil et al 2018; RBA 2022) (Figure 5).^[10] In jurisdictions where hedging is less prevalent (e.g. the United States (Begenau, Piazzesi and Schneider 2015) and euro area (Hoffmann et al 2018)), the flattening of the yield curve that tends to occur when interest rates fall to low levels lowers banks' profits from maturity transformation (i.e. lending long and funding short).^[11]
  • A likely important factor causing the divergent hedging practices between Australia and other jurisdictions is differing regulations, with APRA the only regulator requiring banks to reserve capital against interest rate risk (Waters et al 2020).
• Risk-taking by banks in response to spreads falling for other reasons could mitigate the short-term reduction in spreads, but at the risk of higher future losses. This ‘search for yield’ behaviour is a risk frequently highlighted in the literature. Consistent with theory, the empirical literature tends to find this behaviour in banks that offload risk via securitisation, have low capital levels, or are poorly supervised. Given that none of these features apply to Australian banks, it is not surprising that there is no evidence of Australian banks having increased the risk of their loan portfolios as interest rates fell (RBA 2021).

This spread compression in other jurisdictions is touted in the literature as the primary cause of lower bank profitability at low interest rates (especially in the long run, once the positive effects on losses and asset repricing fade). The broadly stable spread in Australia therefore suggests that perhaps we should be less concerned about the profitability of Australian banks at low interest rates.

Even if banks' lending spreads remain constant, their NIMs still fall as the level of interest rates fall. This mechanically results from banks having more assets than liabilities, and can be shown by rewriting the NIM identity:

In Equation (1), i_A and i_L are the average interest rates on banks' interest-bearing assets and liabilities. A, L and E are the values of banks' assets, liabilities and equity. When the NIM identity is rewritten as the lending spread plus banks' cost of debt funding multiplied by their equity ratio, it is clear that a constant spread means NIMs will fall as the level of interest rates (i_L) falls.

Interestingly, while the ELB on retail deposits is expected to reduce profitability as rates fall in some jurisdictions (see above discussion), the deposit ELB reduces the pass-through of policy rate changes to i_L. Therefore, with constant spreads, the ELB should actually help Australian banks maintain profitability at low interest rates (all else equal).

3.2.1 A potential caveat

A potential caveat to this conclusion is the observed compression in lending spreads during 2021 (Figure 4). The literature does highlight that banks' profitability is likely to be more adversely affected after an extended period of low interest rates. So even if Australian banks' were able to maintain their lending spreads while interest rates were falling, this spread compression in 2021 could be the beginning of the delayed effect predicted by the literature.

However, the delayed effect in the literature is caused by fixed-rate loans being progressively rolled over to lower rates as the yield curve flattens. Given Australian banks' interest rate risk is very well hedged, this is unlikely to be the cause of the recent spread compression in Australia. Moreover, previous analysis has attributed much of this spread compression to increased competition for fixed-rate loans resulting from cheap term funding facilitated by the RBA's suite of unconventional policies, rather than as a result of low interest rates.^[12] All that said, given the lack of definitive evidence that the recent spread compression was not caused by low rates, this compression warrants further investigation and monitoring.

3.2.2 A potential confounding factor

Banks use both debt and equity to fund their loans. So it is possible that the stability of the major banks' lending spreads (i_A – i_L) could reflect a spread compression via the usual competition channel cited in the literature being offset by an increase in the relative cost of equity or an increase in the equity share of assets. To see this more explicitly, the way banks price their loans in period t can be expressed as a time-varying mark-up $\left({\mu }_t\right)$ over their cost of funding the loan:

Using this general pricing equation, it is possible to determine the relationship between mark-ups and equity funding required for lending spreads to remain constant between periods t –1 and t:

In this framework, an increase in competition at low interest rates would reduce the mark-ups the major banks are able to charge $\left(\text{Δ}{\mu }_t<0\right)$. If lending spreads remain constant despite this increase in competition, then it must be the case that this fall in mark-ups was offset by either an increase in the relative cost of equity $\left(\text{Δ}\left(i_{E,t}-i_{L,t}\right)>0\right)$ or increase in the equity share of funding $\left(\Delta \frac{E_t}{A_t}>0\right).$

While there is no single widely accepted method of measuring the cost of equity, recent research by Cheung and Printant (2019) computes a simple average of three commonly used pricing models. Using their data, the relative cost of equity in 2019 was around the same level that is was in 2010, and significantly lower than during 2009 (Figure 6). Therefore, the relative cost of equity is not a confounding factor.

Cheung and Printant (2019) also show how the major banks' common equity Tier 1 capital ratio has increased since 2009, in response to regulatory changes – first in response to the Basel III reforms, then further changes designed to make Australia's bank capital framework ‘unquestionably strong’ (RBA 2016). So this could be a potential confounding factor. However, in the calculation of capital ratios assets are risk weighted in the denominator, whereas for determining overall funding costs it is the unweighted equity ratio that needs to be analysed. The major banks' unweighted equity ratio has been fairly stable since 2009, and has fallen recently (Figure 6, bottom panel).

In short, neither the cost nor funding share of equity appear to be confounding factors that could hide the competition effect cited in the literature.

3.2.3 A robustness check

Figure 4 is constructed using detailed information on the pricing of the major banks' various loans and sources of debt funding, and detailed information on the compositions of these banks' balance sheets. But it does not directly use information on banks' reported NIMs. Equation (1) provides a way to use banks' reported NIMs as a robustness check; if lending spreads have been broadly constant since 2009, the major banks' NIMs should be highly correlated with movements in $i_L\left(\frac{E}{A}\right)$.

Figure 7 plots the major banks' reported NIMs against a model that includes a constant and $i_L\left(\frac{E}{A}\right)$ as the sole regressor (with regressor coefficient restricted to one, consistent with Equation (1)). With an R-squared of 82 per cent, the overwhelming majority of the movements in the major banks' NIMs since 2010 can be explained by movements in $i_L\left(\frac{E}{A}\right)$, confirming the ‘broadly stable lending spread’ result in Figure 4.^[13],[14]

Footnotes

While some of this interest rate risk can be temporarily hedged, via a ‘replicating portfolio’ (see Brassil et al (2018)), these hedges at best delay the spread expansion. [5]

While banks could offset this increased risk on their balance sheet by actively managing their portfolio, if this increased risk is already ‘priced in’, then banks would still bear some cost of the increased risk.
This long-run interest rate risk differs from higher frequency interest rate volatility (i.e. the volatility of rates around neutral). The volatility of bond prices at these higher frequencies does not appear to depend on the level of rates (Simon 2015). [6]

Brassil et al (2018) also control for this endogeneity by incorporating RBA forecasts into their model. [7]

Although some recent research suggests that monetary policy can itself move the neutral rate, at least temporarily, by intertemporally shifting demand (McKay and Wieland 2021; Mian, Straub and Sufi 2021a). [8]

A small subset of the expansive international literature (including some thorough literature reviews): Albertazzi et al (2020); Alessandri and Nelson (2015); Altavilla et al (2018); Amzallag et al (2019); Arce et al (2018); Balloch, Koby and Ulate (2022); Barmeier (2022); Basten and Mariathasan (2018); Bikker and Vervliet (2017); Borio, Gambacorta and Hofmann (2017); Borio and Hofmann (2017); Brandão-Marques et al (2021); Brei et al (2019); CGFS (2018); Claessens, Coleman and Donnelly (2018); de Groot and Haas (2022); Dell'Ariccia, Laeven and Suarez (2017); Demiralp, Eisenschmidt and Vlassopoulos (2017); Eggertsson et al (2019); Heider, Saidi and Schepens (2019, 2021); Ioannidou, Ongena and Peydró (2015); Jiménez et al (2014); Junttila, Perttunen and Raatikainen (2021); Lopez, Rose and Spiegel (2020); Maddaloni and Peydró (2011); Malovaná et al (forthcoming); Martínez Pagés (2017); Onofri, Peersman and Smets (2021); Schelling and Towbin (2020); Stráský and Hwang (2019); Tenreyro (2021); Turk (2016) and Ulate (2021). [9]

While the instruments used to hedge interest rate risk incorporate the compensation institutional investors require to hold this risk, households and small businesses are likely less able to deal with this risk, and would therefore be willing to pay more to remove this risk (so the spreads between fixed lending rates and equivalent-maturity money market rates would still retain some compensation for interest rate risk). Therefore, lending spreads on new fixed-rate loans could still fall a small amount when interest rate risk is lower even though the banks' interest rate risk is fully hedged. [10]

For banks with unhedged fixed-rate assets and variable-rate liabilities, a fall in interest rates would provide a temporary boost to their interest spreads. However, this would only boost long-run profits to the extent that the rate fall was unexpected. [11]

These policies included the Term Funding Facility (TFF), yield target and bond purchase program. For example, because the cost of accessing the TFF was the same for all banks, the TFF lowered the term funding costs of the non-major banks by more than the majors' term funding costs (Black, Jackman and Schwartz 2021). Moreover, because the major banks issued fewer bonds while drawing on the TFF, it also indirectly reduced the costs of the non-major banks' and non-bank lenders' other funding sources. The TFF therefore likely contributed to greater competition for fixed-rate loans, with this increased competition being a primary driver of the 2021 spread compression (Fitzpatrick et al 2022). [12]

Because I am imposing a coefficient of one (rather than estimating the value) and the regression residuals are stationary, the spurious regression risk associated with potentially non-stationary variables is low. In any case, estimating the model in first-differences finds the constant to be insignificantly different from zero, consistent with a stable lending spread. [13]

Interestingly, this strong positive relationship does not appear to be evident in New Zealand, despite similar banking systems (Richardson 2022). [14]