The impact of size, composition and duration of the central bank balance sheet on inflation expectations and market prices

Abstract

We analyse the effects of announcements of changes in the Eurosystem’s balance sheet size, duration and composition on inflation expectations, the exchange rate, the 10-year euro area government bond yield and stock returns, using local projections. We explicitly take into account interaction effects between the three balance sheet dimensions. We provide evidence for the duration extraction channel of monetary policy transmission as we find that the bond yield is sensitive to the combined impact of shocks to balance sheet size and duration. The exchange rate is also affected by a joint size-duration shock. Moreover, the bond yield and exchange rate are sensitive to the joint effect of changes in size and composition. The results indicate that interactions between balance sheet dimensions matter.

KEYWORDS:

• Central banks and their policies
• monetary policy

JEL CLASSIFICATIONS:

• E58
• E52

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 We focus on inflation expectations as we expect a more contemporaneous response of expectations than of realised inflation to announcement shocks to the balance sheet.

2 The Eurosystem had launched the first covered bonds purchase programme in 2009 and the second one in 2011. The asset-backed securities purchase programme and the third covered bond purchase programme were introduced in October 2014.

3 Note that duration can also be extracted without changes in the balance sheet size if the central bank sells short-term and buys long-term bonds.

4 We will focus in particular on the duration risk channel as a form of portfolio rebalancing, and not on other variants of portfolio rebalancing such as the local supply or the asset valuation channel. For the duration risk channel, asset purchases depress yields of all securities of a given duration. In contrast, in the local supply approach, only securities purchased would be affected and not-purchased securities would remain unaffected (cf. Bauer and Rudebusch 2014 for a discussion and Joyce et al. 2011, for evidence for the UK). For a further discussion see e.g. Cahill et al. (2013) or Andrade et al. (2016).

5 They further provide support for the exchange rate channel, the inflation re-anchoring channel and the credit channel.

6 The paper states that the duration risk and the local supply channel together account for about -9 basis points impact on yields (per $100 billion purchase) and account to equal extents for this result.

7 The first and second large scale asset purchases, the MBS reinvestment programme into Treasury securities, the maturity extension programme and agency MBS reinvestment programme and the second maturity extension programme.

8 They argue in the UK transmission worked mainly through the risk taking channel.

9 They also note that portfolio rebalancing towards foreign assets via the exchange rate channel was limited.

10 As discussed in the previous section, our sample is 2008:4 to 2017:6, while VP (2017) focus on 2007:1 to 2014:12.

11 These measures included liquidity management measures (fixed rate full allotment, widening of the collateral base, lengthening of maturities for refinancing operations and the provision of liquidity in foreign currencies) and the covered bond purchase programme (cf. Trichet 2009 as referred to in Lenza, Reichlin, and Pill 2010).

12 To note, our concentration measure assesses concentration from an instrument perspective, defining concentration in terms of the relative share of assets associated with each of the different monetary policy instruments of total monetary policy-related asset holdings, not in terms of underlying transaction modality. Thus, we only treat the extension of credit against a sovereign bond as collateral differently from a purchase of a sovereign bond in so far as the transactions are related to different monetary policy instruments. While this lies beyond the scope of this paper, a concentration measure further distinguishing transaction modalities could complement our analysis.

13 See e.g. De Santis and Holm-Hadulla (2017) for an event study that analyses implementation (‘flow’) effects on bond yields.

14 The cumulative sum of asset purchase announcements is constructed by Hesse, Hofmann, and Weber (2017) by adding up the announced purchases of assets in the asset purchase programmes at the announcement date.

15 See Appendix A for the announcement and implementation dates. For the LTROs this approach is applied for the 1, 3 and 4 years refinancing operations. The cumulative sum of the uptake in the LTROs is taken over the announced horizon of the programme; thereafter the non-cumulative amounts are taken. The cumulative amount is not known ex-ante since the uptake in the LTROs is determined by the demand of banks. Our approach implicitly assumes that market expectations on the uptake upon announcement of the LTROs are similar to the actual uptake.

16 For instance the horizon over which assets are purchased in the asset purchasing programmes other than the APP (like the SMP and the first covered bond purchasing programmes) was not announced ex ante. If the horizon is not announced ex-ante – for instance because it is presented as an open-ended programme - we assume that it is equal to the implementation period. Furthermore, by not calculating the present value of the expected amounts of the programmes we implicitly assume that the discount factor is zero percent. This assumption is motivated by the high uncertainty about the time preference of investors, which is basically unknown.

17 Compared to VP (2017) we restrict the asset classes included in the composition metric further to only cover those asset classes with direct links to monetary policy, thus e.g. excluding emergency liquidity assistance-related assets and asset related to the Agreement on Net Financial Assets (ANFA). Thus, the index is used in its most restricted form and could be further extended in future research, e.g. to include non-monetary policy-related assets and foreign exchange reserves.

18 The three Covered Bond Purchase programmes and purchases of government bonds vs bonds issued by supranational organisations under the PSPP are all included separately in the HI metric to reflect the metric’s broadest configuration.

19 The index does not reflect Outright Monetary Transactions as assets on the ECB’s portfolio linked to them are zero throughout the period of analysis.

20 The duration measure excludes MROs and asset-backed securities; the latter for data availability reasons.

21 The duration, measured as the remaining weighted average maturity (WAM) of LTROs is proxied by assuming that the maturity of the LTRO decreases linearly over the period in which the operation is outstanding. The duration of purchased assets is based on the observed remaining weighted average maturity (WAM) of the asset classes.

22 At the start of the rise (i.e. in April 2011) these two asset classes accounted for 34% of total assets, while amounting to 44% at the peak in April 2012.

23 In 2010 the 6 months and 1 year LTROs expired, while the amount of long-term assets purchased under the CBPP and SMP expanded. This explains why in 2010 duration increases, while size remains quite stable.

24 Variance inflation factors for the coefficients of the balance sheet variables for size, duration and composition regressed on inflation expectations in one single equation model remain below critical levels (which by rule of thumb is around 5).

25 As a robustness check we also included a dummy variable to control for announcement effects (dummy being 1 in a month in which a new (or adjusted) conventional or unconventional monetary policy measure was announced by the ECB and 0 otherwise). The dummy variable is not significant in any of the model estimations, so we left it out from the baseline regression. Another reason for doing this is that market prices are assumed to capture announcement effects as well.

26 The term spread is proxied by the difference between the 10 years euro area bond yield and the sum of real GDP growth and inflation, both expected 5–10 years ahead (source Consensus Economics). The range of credit spreads is calculated by the standard deviation of the spread on financial and non-financial corporate BBB bonds, covered bonds, asset backed securities and an index of government bonds of Greece, Ireland, Italy, Portugal and Spain (spread with regard to German government bond yield with comparable maturity, own calculation based on data from Datastream).

27 See Introductory statement to the press conference (with Q&A) of 3 April 2014.

28 The interaction of size with duration has an insignificant effect on inflation expectations (Figure 6, row iv., column A. Note that the effect is briefly significant after 10 months, but we notice that local projections become less reliable at longer horizons).

29 Note that the impulse response functions in Figure 6, row ii depict an increase in concentration, therefore the signs of the effect on bond yields and the exchange rate are positive.

30 The BVAR estimations were conducted with the BEAR toolbox, developed by Dieppe, Legrand, and Van Roye (2016).

31 Note that the impulse response functions in Figure c, row iv. depict an increase in concentration, therefore the signs of the effect on bond yields and the exchange rate are positive.