The Central Bank Balance Sheet As a Policy Tool: Lessons From the Bank of England's Experience

While central banks have also used asset purchases for purposes other than QE (both before and after the GFC), this paper focusses on purchases made under QE programmes.

Cecchetti and Tucker (2021) provide a more detailed taxonomy of financial stability actions: lender of last resort, market maker of last resort; and provision of selective credit support. While they also note that the central bank balance sheet could in principle be used to provide emergency government financing, the Bank of England balance sheet was not used for this purpose during the period considered here.

For example, we do not consider the policy responses to more recent events including the energy price shock following Russia’s invasion of Ukraine or the temporary and targeted financial stability operations in September/October 2022 in response to problems experienced by some Liability Driven Investors.

The Special Liquidity and Funding for Lending Schemes (SLS and FLS respectively) were funded through asset swaps rather than reserves, so did not appear on the Bank of England balance sheet. They are shown for context, given the importance and scale of these policy actions in the relevant periods.

The balance sheet effects of the Term Funding Scheme (TFS) were also relatively persistent. The TFS was aimed at preserving the monetary transmission mechanism when the policy rate was cut to its lower bound, so can also be classified as a monetary policy action.

Assets and liabilities from both the Bank’s Issue Department and Banking Department have been consolidated into a single figure. QE purchases are reflected on the Bank’s balance sheet as an asset in form of a loan to the Asset Purchase Facility (APF) and these are shown as a decomposition of the loan by instrument type. The Term Funding Scheme (TFS) was transferred from the APF to the Bank’s own balance sheet in January 2019. “Other assets” includes “Ways and means”. “Other liabilities” includes notes in circulation and cash ratio deposits. The Special Liquidity and Funding for Lending Schemes (SLS and FLS respectively) were funded through asset swaps rather than reserves, so were not on balance sheet. They are nevertheless shown for context.

Dealers (part of the commercial banking system) are the only firms that can participate in the Bank’s competitive asset purchase auctions. They offer intermediation services to non-bank clients that are often the ultimate sellers of gilts. For simplicity, the dealer intermediation step in this process is omitted in the description in the text.

For example, the MPC’s 2020 remit states that “the Committee may judge it necessary to deploy unconventional policy instruments in order to set monetary policy consistent with the requirements of this remit” (HMT 2020).

An exchange of letters between the Chancellor of the Exchequer and the Governor of the Bank of England expressly acknowledges that the Bank’s QE purchases in the secondary gilt market should not directly affect the gilt issuance strategy of the Debt Management Office (DMO) in primary markets (HMT 2009).

Bhattarai and Neely (2022) highlight that while many researchers find that UMPs often had the intended effects on output and inflation, these effects were not sufficient to fully offset the effective lower bound constraint on conventional monetary policy.

The authors’ findings can be challenged on several grounds, including the potential sensitivity of results to specific modelling choices and that their sample only contained a subset of the wider QE literature (for example, excluding papers on individual channels of QE).

The observation that large scale and rapid QE occurred alongside a similarly large and rapid expansion of government debt raised some concerns of monetary financing (Economic Affairs Committee 2021). However, as explained in Sect. 2.1, several institutional safeguards ensure that QE decisions are based solely on the MPC’s judgement of the monetary policy stance required to return inflation to target.

The Treasury is responsible for the financial exposures of the CCFF and determines which firms are eligible to take part in the facility. The scheme itself is funded through the issuance of reserves, with the MPC’s agreement.

The y-axis proxies the gilt yield impact of each UK QE announcement. The measure used is the basis point change in the 10-year gilt yield in an event window following the announcement of each additional round of QE. A two-day event window was used for QE1-QE4. For the first QE1 announcement, the yield impact of both the February and March 2009 announcements are included, following Joyce et al. (2011). For QE5, a shorter event window (until 08:15 on the day following the announcement) was deemed appropriate, given the substantial noise from other Covid developments and the international policy response. The x-axis gives the percentile rank of the average bid-ask spread across 5-year, 10-year and 30-year gilts at the time of QE announcement. The percentile is given relative to the distribution from 2008–2020. So an x-axis reading of 0.5 represents a median bid-ask spread, and a reading of 1.0 represents the highest bid-ask spread observed during the period. The size of each “bubble” is proportional to the size of the QE stock surprise in each episode. This surprise measure reflects Bank staff’s best guess of the market expectation for the ultimate stock of QE prior to the announcement, based on available market intelligence and survey evidence. Only the first announcement of each new QE round is shown, rather than each subsequent extension within a given round of QE purchases. QE1 = March 2009; QE2 = October 2011; QE3 = July 2012; QE4 = August 2016; QE5 = March 2020.

More broadly, the possibility that the effects of QE are state contingent is consistent with evidence that the impact of QE has changed over time (see, for example, Greenlaw et al 2018; Krishnamurthy and Vissing-Jorgensen 2011; Kaminska and Mumtaz 2022). There is no consensus on the evidence, however: Bernanke (2020) surveys the pre-Covid literature and concludes that it is consistent with a broadly constant impact of QE across episodes.

A generalisation of this approach could therefore generate state-contingent effects of QE. To the extent that market dysfunction can be interpreted as an exogenous shift in arbitrageurs’ risk aversion, the link between QE transmission and market dysfunction is also implicit within that framework, albeit in a reduced-form way. Greenwood and Vayanos (2014) estimate time-varying risk aversion in terms of arbitrageur losses.

From a theoretical perspective, the portfolio balance channel suggests that QE works by changing the relative supplies of privately held assets (a stock concept). The empirical analysis of D’Amico and King (2013) suggests that stock effects were more pronounced and persistent than flow effects for the QE1 and QE2 programmes in the United States.

Knowledge that the central bank is likely to step in with such LOLR and MMLR operations may create moral hazard such that financial institutions under-insure themselves against the risks of a stress, an issue discussed by Bagehot (1873).

For example, the Contingent Term Repo Facility (CTRF) discussed in Sect. 3.1 was withdrawn after conditions normalised, participation declined and drawdowns were repaid.

There is a large literature documenting the countercyclical effects of unanticipated changes in short-term policy rates (see for example, Christiano et al 1999; Bernanke et al 2005; Boivin et al 2010; Ramey 2016).

When the long-term government debt stock is fixed, market clearing implies that purchases of long-term debt by the central bank (QE) determine the portfolio mix held by financial intermediaries in equilibrium (hence the form of Eq. (4)). See Harrison (2011, 2012, 2017) for further details including analysis of cases in which government debt issuance adjusts endogenously.

To the extent that it mimics the form derived from an approximation to household welfare, the loss function minimised by the policymaker captures the distortions generated by nominal frictions (sticky prices) and the portfolio frictions that give QE traction in the model. As such, it is a standard benchmark for the analysis of QE as a monetary policy tool. However, the loss function omits the potential spillover effects of QE on financial stability, broadly defined, discussed in Sect. 2.2. A richer analysis of the holistic welfare implications of state contingent QE is beyond the scope of this paper, but should be a topic for future research.

The model is a variant of a workhorse that has formed the basis of a large body of research on optimal interest rate policy (see, for example, Galí (2015) and Woodford (2003) for detailed reviews). Many of the lessons from that research therefore apply to the model we use. However, our focus is the role of QE as an additional policy instrument that can be used as a partial substitute for the short-term policy rate when it is constrained by a lower bound (Harrison 2017).

Focusing on the policy mix – the combination of short-term interest rate and QE chosen to implement the overall policy stance – is relevant because the two instruments are (partial) substitutes (Yellen 2017). This substitutability means that it may be difficult to identify the macroeconomic effects of changes in QE away from the lower bound, to the extent that the expected path of the short-term interest rate adjusts to offset the effects on the overall policy stance.

A rough quantification using 2019 nominal GDP (around £2.25 trillion) implies that headroom considerations could motivate unwinding QE by an additional £2 billion in the first year of normalisation when QE is state contingent.

See, for example, the strategies for balance sheet reduction set out by the FOMC and MPC (FOMC 2022; MPC 2021).

Estimates of the equilibrium real interest rate have continued to fall over time (see, for example, Bailey et al 2022 and Cesa-Bianchi et al 2023).

For example, the Bank of England launched a new facility (the Short Term Repo) to play this role on 1 September 2022.

In principle, an ‘ample reserves’ approach, maintaining a buffer of reserves in excess of the PMRR, may be preferable in situations where the demand curve for reserves is sharply upward sloping around the PMRR and autonomous factors can have a large impact on the supply of reserves (Afonso et al 2020). The Federal Reserve’s pre-Covid experience reducing its balance sheet presents such an example. In September 2019, the demand for securities financing increased at a time when autonomous factors sharply reduced the supply of reserves, leading to upward pressure on money market rates (Logan 2019). See also Anbil et al (2020).

Note that we focus on central bank asset purchases used to implement QE as a monetary policy tool. Other types of asset purchases may be occasionally warranted for financial stability reasons and, as discussed above, such purchases would likely be temporary with no enduring impact of the size of the balance sheet.

This approach abstracts from two factors. First, the assumption that the policy rate is fixed ignores the fact implementing QE while the policy rate is at the ELB should bring forward the expected liftoff date and raise the longer-term expected path of the policy rate relative to the counterfactual of no QE (see Chen et al 2012). Second, the assumption that the QE programme is permanent may also understate the effect of the QE operation on yields. So the values for \(\nu\) and \(\xi\) obtained by our approach could be smaller than those from a more sophisticated method.