The macroeconomic effects of quantitative easing in the euro area: Evidence from an estimated DSGE model

Abstract

This paper estimates an open-economy dynamic stochastic general equilibrium (DSGE) model with Bayesian techniques to provide a structural empirical evaluation of the macroeconomic effects of the European Central Bank's (ECB's) quantitative easing (QE) programme. Allowing for cross-border holding of government debt and using data on government debt stocks and yields across maturities, we identify the parameters governing portfolio rebalancing in the private sector. We rely on a methodological extension that measures the non-linear contribution of QE in shock decompositions under an occasionally binding constraint (zero lower bound). Our results suggest an average contribution of ECB QE to annual Euro Area GDP growth and CPI inflation in 2015–18 of 0.3 and 0.5 percentage points, respectively, with a maximum impact in 2016.

Introduction

In early 2015, the European Central Bank (ECB) joined the group of central banks that have implemented large-scale asset purchase programmes as unconventional policy measures. These asset purchases, also called quantitative easing (QE), have led to a strong extension of the central banks' balance sheets. The ECB's largest QE programme, announced in January 2015 (Public Sector Purchase Programme), foresaw buying €60 billion of assets a month from March 2015 to September 2016, which in sum corresponds to circa 10% of annualized euro area (EA) GDP. In December 2015, the ECB extended the programme until March 2017, and it raised the amount of monthly purchases to €80 billion starting from April 2016. In December 2016 the programme was extended and modified again, lengthening the period of asset purchases until December 2017, but at a reduced pace of €60 billion of assets a month after March 2017. In October 2017 the programme was amended once more, extending asset purchases at a reduced monthly pace of €30 billion. In December 2018, the ECB decided to stop asset purchases at the end of 2018, whereas maturing bonds will still be reinvested.

Operating close to the zero lower bound (ZLB), the ECB considered its ‘conventional’ monetary accommodation to be insufficient to address weak inflation dynamics, falling inflation expectations and sizeable economic slack in the EA. As a result, the balance sheet interventions' aim was to “achieve the price stability objective, given that interest rates have reached their lower bound” (Draghi, 2015). In practice, the ECB has purchased public sector financial assets (government debt) of longer maturity and extended liquidity (base money) to the private sector.

This paper provides a structural empirical evaluation of the macroeconomic effects of the ECB's QE programme using a two-region dynamic stochastic general equilibrium (DSGE) model for the EA and the rest of the world (RoW), estimated with Bayesian techniques. In order to capture the effect of QE policies and break Wallace's irrelevance theorem we depart from the conventional DSGE framework by introducing a specific financial friction that limits investors’ ability to arbitrage assets in their portfolio. A ‘transaction cost’ that agents have to pay when adjusting their portfolio gives rise to imperfect substitutability between assets of different maturities, and isolates the portfolio rebalancing channel of QE. The modelling assumes that investors have preferences for assets of different maturities following a ‘preferred habitat’ motive, similar to Vayanos and Vila (2009). The approach captures the idea that relative asset prices depend on the relative asset supply. Since the transaction cost renders short-term and long-term bonds imperfect substitutes, the central bank can use asset purchases to alter the relative supply of long-term versus short-term assets and, hence, bond returns, i.e. to flatten the yield curve.

In our model, central bank long-term asset purchases have effects on the real economy through the extent to which they induce private investors to rebalance the portfolio mix of short-term and long-term assets holdings. Private investors respond by holding higher amounts of corporate equity and foreign bonds, and by reducing savings. A portfolio reallocation towards equity increases the price of corporate equity (rising stock market), which reduces the financing costs of corporations and translate into stronger investment and capital accumulation. A reallocation into foreign-currency assets increases the price of foreign currency (exchange rate devaluation) and strengthens net exports. Finally, reduced savings strengthen contemporaneous consumption demand. Cross-border trade in financial assets, however, implies that non-EA residents supply a part of the long-term EA government bonds purchased by the EA, which dampens the portfolio adjustment on the side of domestic (EA) private investors.

The attention to imperfect asset substitutability follows from the contributions of Tobin (1956, 1969), which was subsequently placed in a dynamic optimizing framework by Andrés et al. (2004). From an empirical perspective, such a channel has been shown to have had a significant role in the transmission of QE policies in the US and the UK, with notable examples including work by Greenwood and Vayanos (2014), Gagnon et al. (2011), Joyce et al. (2011), Krishnamurthy and Vissing-Jorgensen (2011), and D'Amico et al. (2012). Research at the ECB has also provided evidence for the impact of unconventional monetary tools on long-term bond yields and other asset prices through portfolio reallocation. An event study by Altavilla et al. (2015) reports a 30–50 basis point (bp) decline in 10-year government bond yields, lower corporate bond spreads, higher equity prices, and euro depreciation. Andrade et al. (2016) report a decline of EA 10-year government bond yields in the range of 27–64 bp, higher equity prices and inflation expectations. De Santis (2016) finds an average decline in 10-year government bond yields by 63 bp between September 2014 and October 2015.

Recently, there has also been a growing model-based literature on the effects of QE operating through the portfolio-rebalancing channel. In their majority, these papers feature imperfect asset substitutability that originates from a transaction cost motive, that is, an asset-maturity composition decision subject to adjustment costs. As in our paper, the details of why assets of different maturities may be imperfect substitutes are not formally modelled. Instead, limited arbitrage is assumed, and its importance is explored (in our case estimated from the data) for how it matters in the transmission of QE.

In the context of US QE, Chen et al. (2012) extend a conventional DSGE model with transaction costs between short-term and long-term bonds and with the additional feature of segmented markets. For an analysis of QE in the US and the UK, the model in Falagiarda (2013) similarly adopts a portfolio adjustment cost friction that resembles the setup in this paper. De Graeve and Theodoridis (2016) too motivate imperfect asset substitutability with adjustment costs and estimate its strength in the context of US ‘Operation Twist’ policies. In turn, Ellison and Tischbirek (2014) embed the portfolio adjustment cost framework in a model with a financial sector allowing them to investigate the question of whether QE can be useful even outside a ZLB environment. Finally, Harrison (2012, 2017) studies optimal policy in a New Keynesian model extended with portfolio adjustment costs in households’ utility functions.

The introduction of imperfect substitutability is not unique in permitting QE policies to have non-neutral effects on the economy. Notably, other studies have analysed alternative extensions to the conventional DSGE model, which focus on the bank-lending channel as complementary transmission mechanism. However, the features of our model have implications for consumption and investment that are similar to those of an extension of credit. First, financial intermediaries may face a similar decision problem as Ricardian households in our set-up. In particular, when the central bank buys long-term government bonds from banks, the latter can respond by buying more equity and foreign assets, and by providing more loans to firms. Second, QE can raise the net worth of banks and extend their lending margin in the presence of capital requirements or equivalent constraints, as in, e.g. Gertler and Karadi (2013), and Carlstrom et al. (2017).

The contribution this paper makes to the literature on quantitative easing is threefold. First, the objective is to provide a structural empirical evaluation of the macroeconomic impact of ECB QE using a state-of-the-art estimated DSGE model. Other studies assessing the impact of QE in the EA rely on calibrated models (e.g., Andrade et al., 2016; Priftis and Vogel, 2016; Sahuc, 2016; Cova et al., 2015). Second, we rely on a methodological extension to existing solution algorithms that treats the non-linearity imposed by the presence of an endogenously binding ZLB in the solution of the model and the calculation of shock decompositions explicitly. The literature has so far tractably dealt with the ZLB by implementing perfect foresight simulations where the duration of the constraint is predetermined. Third, the movement of the exchange rate in response to QE affects aggregate demand, output and inflation and is a channel that is missing in most model-based studies of QE policies (e.g., Chen et al., 2012; De Graeve and Theodoridis, 2016; Gertler and Karadi, 2013), which, instead, build on closed-economy frameworks. In terms of international linkages, we even go beyond the inclusion of exchange rate responses and their impact on trade by allowing for cross-border holding of long-term government bonds, so that both domestic and foreign private investors can be counterparties of QE operations by the central bank. Combined, these innovations allow for a more encompassing empirical evaluation on the effects of ECB QE using a structural framework.

We bring the model to the data by, first, estimating a linearized version of the model in which the Taylor rule is not constrained by a lower bound. Combining data on government debt stocks, time-varying shares of cross-border holdings and yields across maturities, the estimation provides a value for the parameters governing portfolio adjustment costs in the model. The implied magnitude of these costs determines the yield spread in response to QE of a given volume and time path, such as the one announced by the ECB in January 2015, which in practice resembles an AR(2) specification.

In a second step, we use the estimated model to conduct a quantitative evaluation of QE using impulse response functions (IRFs) and non-linear shock decompositions that treat the non-linearity arising from the occasionally binding ZLB constraint explicitly, i.e. the ZLB can become endogenously binding when contractionary shocks drive the target (‘shadow’) interest rate below the lower bound. We use an algorithm by Giovannini and Ratto (2018) to quantify the impact of estimated historical shocks on observed variables, with the model solved in the piecewise-linear fashion of Guerrieri and Iacoviello (2015) to account for the non-linearity in the analysis. The algorithm provides initial conditions and smoothed variables and shocks, consistent with the occasionally binding constraint, i.e. it also estimates a sequence of binding regimes along the historical periods.

Accounting for the ZLB environment, our results show that QE increases annual EA GDP growth and CPI inflation in 2015–18 by 0.3 percentage points (pp) and 0.5 pp on average, respectively, with a maximum GDP growth and inflation impact of 0.6 pp in 2016. The non-linear environment amplifies the impact of QE compared to a symmetric linear model without binding ZLB because of two effects. First, the non-linear model does not generate a countervailing short-term policy rate response (tightening) in response to expansionary QE during the period in which the ZLB constraint binds in 2015–18. Second, QE endogenously lifts the economy away from the ZLB, so that other, contractionary shocks have less negative effects in the medium term.

The effects of QE on EA GDP and inflation in our estimated model are comparable with existing literature for EA QE and comparable in order of magnitude with results from similar exercises for QE in the US.¹ Using a version of the Gertler and Karadi (2013) model for the EA, Andrade et al. (2016) assume an AR(2) specification of the QE shock with a size equal to 11.4% of EA GDP at the peak. They find the ECB asset purchase programme (APP) to increase inflation by 40 bp and output by 1.1% at their peak, which is reached after around 2 years. In a DSGE model with shadow EONIA rate, Mouabbi and Sahuc (2019) find that EA year-on-year GDP growth and inflation would have been lower by 1.1 pp and 0.6 pp, respectively, on average in 2014–17 in the absence of unconventional monetary policies. Sahuc (2016), using the framework of Gertler and Karadi (2013), finds effects of ECB QE (9% of EA GDP) on EA real GDP growth (inflation) of 0.2 (0.1 pp) in 2015–16 for short-term rates constant in 2015, whereas keeping the policy rate unchanged for another year raises the average growth (inflation) effect in 2015–16 to 0.6 pp (0.6 pp). Cova et al. (2015) study the impact of the ECB's asset purchase programme (APP) in a multi-country DSGE model with imperfect substitutability between assets of different maturity, motivated by the differing liquidity services they provide. A QE shock corresponding to monthly purchases of 60 billion euros and lasting for 7 quarters (and subsequently being phased out) increases the level of GDP and inflation in the EA by approximately 1 pp over 2015–17.

Regarding the US, Chen et al. (2012) report GDP growth (inflation) effects of US QE, corresponding to 4 pp of GDP, of 0.1 pp (0.3 pp). De Graeve and Theodoridis (2016) report GDP growth (inflation) effects of 0.6 pp (0.3 pp) of the Federal Reserve's ‘Operation Twist’. Both studies work with scenarios in which short-term policy rates do not respond (immediately) to higher growth and inflation. Gertler and Karadi (2013) quantify the impact of US LSAP with a volume of 2.5% of GDP on output growth (inflation) to 1 pp (1.5 pp) if policy rates remain unchanged, and 0.2 pp (0.2 pp) if the standard monetary policy rule is active and partly offsets expansionary QE effects.

Finally, our paper most closely relates to the works of Alpanda and Kabaca (2019), and Kolasa and Wesolowski (2018), who investigate the spillovers of QE in an open-economy DSGE framework that allows for cross-border trade in government bonds. The two papers differ from ours in focus and methodology, however. Alpanda and Kabaca (2019) analyses the impact of US QE on RoW, whereas Kolasa and Wesolowski (2018) explores the impact of QE in a large open economy (calibrated to the UK, US, and the EA) on a smaller non-EA country (Poland). Both studies introduce the ZLB constraint as commitment by the central bank to keep interest rates fixed for a pre-determined duration, instead of determining the ZLB endogenously by the sequence of shocks hitting the economy as in our case.

The remainder of the paper proceeds as follows. Section 2 outlines the QE-specific structure of the model; Appendix A provides a full model description. Section 3 describes the model solution, estimation methodology, and discusses posterior estimates and model fit. Section 4 discusses the impact of QE in the model with occasionally binding ZLB and cross-border holding of government debt. Section 5 presents results from models without cross-border holdings of government bonds and without ZLB, for comparison. Section 6 summarizes the paper and concludes.