The effects of unconventional monetary policy
on the real economy in the US and euro area
Master Thesis
Supervised by dr. G.H. Kuper
Michiel Koopmans
∗June 5, 2020
Abstract
This paper examines the impact of unconventional monetary policy measures on the real economy by estimating a Bayesian vector autoregression (BVAR) with monthly data from the United States and the euro area. The sample period covers the period when central banks implemented non-traditional measures in an attempt to boost economic recovery following the global financial crisis. The findings indicate that an exogenous balance sheet innovation is more effective in the US than in the euro area. For the US both output and prices react positively to an expansionary balance sheet shock, while in the case of the euro area only the price level rises. The results also suggest that the relevance of certain transmission channels between the regions differ. The portfolio rebalancing channel seems to be activated in the US, while there is also suggestive evidence that in both regions the uncertainty channel was operative.
Keywords: Unconventional monetary policy; Bayesian VAR (BVAR); balance sheet policies
JEL Codes: C32;E52
1
Introduction
In the wake of the financial crisis, many central banks limited the adverse effects of the crisis by cutting interest rates to historically low levels. The rationale for lowering the rates was to provide stimulus to the real economy, however this conventional monetary policy response reached its limits and lost its effectiveness as policy rates reached the zero lower bound (ZLB). Further stimulus was required to recover from the global financial crisis (GFC) and central banks had to come up with innovative measures. As a consequence, central banks resorted to unconventional monetary policy (UMP) measures in order to restore economic and financial stability. Unconventional policy instruments, such as quantitative easing and forward guidance, replaced the conventional instrument, the short-term interest rate, as main monetary policy tool. There is a broad consensus in the literature that a decline in short-term interest rates results in a temporary rise in economic activity and a persistent increase in the price level. In contrast, the literature on the impact of unconventional monetary policy on the real economy is relatively limited. Early studies examining the effects of unconventional monetary policies focused on the impact on financial markets. Moreover, the studies initially focused predominantly on the US. In recent years the literature on the macroeconomic effects of non-standard monetary policies has been growing. This is also the case for the literature focusing on the euro area, especially after the larger-scale asset purchase programme by the ECB starting in 2015. In general, the findings in the literature indicate that the non-traditional monetary policy measures were effective in stimulating economic activity and boosting the price level (Boeckx, Dossche, & Peersman,2017;
Gambacorta, Hofmann, & Peersman,2014;Weale & Wieladek,2016). However, there exists no
substantial agreement in the literature on the magnitude and persistence of these measures on the macroeconomy. Furthermore, studies focusing on the transmission channels of unconventional often have conflicting findings. So, there is a gap in the literature regarding the macroeconomic effects of unconventional monetary policy measures, while it is of pivotal importance for policy makers, especially for central bankers, to get a clearer understanding of non-standard policies’ impact on the real economy.
This paper examines the effects of unconventional monetary policy at the zero lower bound on the real economy in both the euro area and the United States. More specifically, it investigates whether the balance sheet policies of the European Central Bank (ECB) and US Federal Reserve (Fed) are transmitted to macroeconomic factors, such as output and prices. Additionally, this study discusses and intends to identify important transmission channels of unconventional monetary policy. I will also compare the results of the eurozone and the US and discuss the similarities and differences between the impact of the non-traditional monetary policy on the broader economy.
monetary policy measures on these two advanced economies will be discussed. The lion share of studies comparing the effects of unconventional monetary policy measures focus either on the US or the euro area. Finally, I will consider the transmission channels and try to identify the main channels.
The 6-variable VAR is estimated using a Bayesian approach. To examine the effect on the macroeconomic variables, I explore the dynamics of real economic variables in response to an increase in the assets on the balance sheet of the central bank. Furthermore, the transmission channels are analysed by adding a relevant variable from a list of variables to the benchmark model as seventh variable. The impulse response functions indicate that output and prices increase in the US after an unexpected, exogenous innovation to central bank assets. For the euro area, prices increase after an increase of the volume of assets on the central bank’s balance sheet, while output displays a negative response. Regarding the transmission channels, the results indicate that in the US the portfolio rebalancing channel was operative, while in both countries the uncertainty channel appears to be activated.
The remainder of this paper is structured as follows. Section 2 gives a description of conventional and unconventional monetary policy. Moreover, it elaborates upon the non-traditional monetary policies adopted by the Fed and ECB after the global financial crisis. Section 3 provides a literature overview and describes the transmission channels. Section 4 introduces the empirical approach. The data is described in section 5. Section 6 present the main results, while in section 7 some additional robustness check are performed. Finally, section 8 concludes.
2
Conventional and unconventional monetary policy
A distinction can be made between conventional and unconventional policy measures. This section gives a brief recap of the conventional monetary policy mechanism and provides a general overview of the main unconventional monetary policy measures - balance sheet policies and forward guidance. In addition, this section describes some non-standard policies adopted by the Fed and ECB in response to the ineffectiveness of conventional monetary policy instruments to further counter the worsened economic and financial conditions in the aftermath of the crisis.
2.1
Conventional monetary policy
operations. Open market operations influence liquidity on the money market. For example, the Fed conducted open market operations by purchasing and selling securities to match the interest rate with the targeted level associated with their desired inflation rate. If the actual interest rate level exceeds (is lower than) the desired level, central banks can reduce (increase) the interest rate by buying (selling) government securities from (to) banks. In the case of the eurozone, banks can manage their liquidity by using the marginal lending facility, which allows commercial banks to obtain overnight liquidity from the central bank, usually for above interest rates, while the deposit facility enables commercial banks to store liquidity at the central bank for lower interest rates than the money market rate (Pattipeilohy, Van Den End, Tabbae, Frost,
& De Haan, 2013). Thus, these two facilities represent a floor (deposit facility) and ceiling
(marginal lending facility).The ECB can affect the money market interest rate, if the rate is located inside the corridor, by increasing or reducing the liquidity provision, which will lead to higher and lower market interest rates, respectively.
Short-term interest rates cannot decrease further when the policy rate is stuck at the zero lower bound1 and the mechanism only holds when the interest rate is in the positive territory of the yield curve.
Monetary policy actions transmit to the broad economy through the interest rate channel. Suppose the rate of inflation is too high, then the central bank can choose to increase the interest rate by selling securities. The higher nominal short-term interest rate is expected to remain high for a prolonged period and should thus increase the nominal long-term interest rate. Under the assumption of sticky prices, movements in the nominal rates will result in movement in real rates. Consequently, the real cost of borrowing has increased, discouraging investments by firms and durable good purchases by households. Ultimately, the reduction the level of inflation is achieved by ensuring a slowdown in economic activity.
2.2
Unconventional monetary policy
2.2.1 Balance sheet policies
Several types of balance sheet policies can be adopted in order to influence financial markets and the real economy. Borio and Zabai (2016) point out various balance sheet policies and their differences. This section will focus on quantitative easing (QE) and credit easing (CE) policies. Quantitative easing could be defined as policies that lead to extraordinary expansions in the size of central bank liabilities, which consists of currency and reserves, and hence increase the monetary base without targeting specific interest rates and/or markets (Fawley, Neely, et al.,
2013). Credit easing policies, which also could include quantitative easing, are designed to fix the market functioning and improve credit conditions by lowering specific interest rates (Fawley
1It should be noted that in recent years policy rates turned slightly negative in some countries and
et al., 2013). The quantitative easing programs in the United States and eurozone are not classified as pure quantitative easing programs, following the definition of Fawley et al.(2013), since these programs mainly focused on purchasing assets and providing lending programs and not merely raised the reserve base to a targeted level. As a result, the composition of loans and securities on the asset side of the central bank’s balance sheet was not randomly changed due to quantitative easing, which would typically occur in a pure quantitative easing program. The aim of quantitative easing is to boost spending and investment by increasing the liquidity in the economy. For example, purchasing long-term government securities (or other financial assets) from secondary markets by issuing new reserves results in higher prices for these assets and lower yields, which reduces the cost of borrowing for companies and households and should stimulate spending and investment. Furthermore, if a central bank purchases assets from a non-bank entity, the central non-bank deposits the money in an entity’s account at a commercial non-bank, increasing the liabilities (new deposit) and assets (reserves at the central bank) on the balance sheet of a commercial bank (Benford, Berry, Nikolov, Young, & Robson,2009). It is expected that the increase in deposits on a bank’s balance sheet will induce banks to increase their lending to the private sector, which ultimately results in a boost in spending and investment. So, these targeted quantitative easing policies also contribute to credit easing conditions. The transmission to the real economy will be explained more extensively in section 3.2.
By embarking on credit policies, central banks focus on segments of private debt markets and influence the exposure to these segments by modifying counterparty terms or buying private sector assets (Borio & Zabai,2016). The goal of credit easing is to affect the financial conditions in the private sector in such as way that liquidity is available to households and companies.
2.2.2 Forward guidance
The goal of adopting forward guidance at the zero lower bound is to lower the long-term nominal interest rate, which is more relevant for economic decisions by agents than the short-term nominal interest rate. In order to reduce long-term rates central banks have to manage expectations of the public. In particular, lower rates can be achieved by influencing the public’s expectation that short-term rates remain at the zero lower bound for a prolonged period and take away uncertainty concerning the short rates’ expected path in the future (Charbonneau & Rennison,
2015). The objective of targeting the long-term end of the yield curve is to flatten the curve and bring down longer-term interest rates. Reducing uncertainty enhances the public’s ability to forecast the short- and long-term and lowers the variability of interest rates, which could drive the risk premium downwards.
Filardo and Hofmann(2014) point out the necessity that the public perceives that the central
forward guidance in the intended way.
A distinction can be made between three forms of forward guidance: Qualitative; time-contingent; state-contingent (Filardo & Hofmann, 2014). Qualitative forward guidance statements do not provide explicit information on the path of the policy rates and the anticipated period of time. Time-contingent forward guidance specifies the envisaged time period while state-contingent forward guidance explicitly ties its guidance to the state of the economy. Time-contingent guidance publicly commits the policymaker to a particular course of action for a specified period, but also faces the time inconsistency problem. In certain situations, central banks could have an incentive to renege on their commitment during the specified time period. If the public anticipates that under some circumstances a central bank might renege on its committed policy course, policies might not have the intended impact. The importance of credible commitment is shown by Werning (2011). He finds that a promise, perceived as credible by the public, to keep the interest rate at the zero lower bound for a longer duration than economic conditions suggest can provide a stimulus to aggregate demand.
2.2.3 United States
The financial system of the United States is generally considered as a market-based system. This implies that the main channel through which non-financial corporations raise funds is the capital market (i.e stock and bond markets). Financial intermediaries play a relatively less prominent role in providing the required finance. Data shows that during the period 2001-2011 the share of non-bank financing to the private sector was 80%, while bank-based financing only accounted for 20%. After the federal funds rate reached its lower bound, the Fed pursued several rounds of quantitative easing and announced a maturity extension program. Furthermore, the Fed used forward guidance- communications by the Federal Open Market Committee (FOMC)- about the future short rates to influence the longer-term U.S. interest. The first large scale asset purchase program (LSAP)- QE1- amounted to 1.725 trillion U.S. dollar and resulted in a huge increase in the monetary base. This program included purchases of government-sponsored enterprise (GSE) debt, mortgage-backed securities and long-term securities on 25 November 2008, and additional, unexpected purchases on the March 18, 2009 (Bhattarai & Neely, 2016). Additionally, the Fed executed a second round of quantitative easing (QE2) from November 2010 to June 2011, which comprised the purchase of 600 billion U.S. dollar in U.S. Treasuries. This program was aimed at further lowering the long-term interest rate and increasing the inflation rate to levels more in line with the price stability objective of the Fed’s dual mandate (Bhattarai & Neely,
of Treasuries with longer maturities, the Treasury purchase included in QE3 was not sterilised, i.e. not offset by a sale of short-term Treasuries. Ultimately, the Fed discontinued QE3 in October 2014. In contrast to the ECB, the Fed directly started with conducting large-scale asset purchase programs once the policy rate approached the zero lower bound during the crisis period. In addition, the actions of the Fed to combat the adverse effect of the crisis were more aggressive compared to the measures introduced by the more conservative ECB. Figure 1 displays that the FED introduced the open-ended, monthly asset purchase programme – QE3- in 2012 and that the Fed started with tapering its purchases in the later stages of the program before it was concluded in October 2014. From 2014-2017, the Fed rolled over maturing securities in order to keep the balance sheet at a constant level. After having pushed the federal funds rate to close to zero in response to the crisis and maintaining unprecedented low rates for more than seven years, the Fed started to raise the federal funds rate in December 2015. After keeping the balance sheet at reasonable constant level for several years, the Fed announced in 2017 that they would implement programmes aimed at normalising the balance sheet (i.e. shrinking the balance sheet). As part of the balance sheet normalisation policies, the Fed decided in September 2017 to discontinue rolling over maturing securities and this consequently led to a reduction in the size of the balance sheet.
The non-standard monetary policy conducted by the Fed has contributed to the recovery of the US economy after the crisis. The evidence suggests that the asset purchase programmes, especially the (announcement of) the first asset purchase programme, lowered the long-term yields considerably and reduced the term spreads. Ultimately, the favourable financial conditions should transmit to the real economy. The general consensus in the literature is that the non-standard policies stimulated economic activity and was successful in preventing large declines in the interest rate.
The forward guidance statements also changed during the post-crisis period, moving from qualitative and unclear statements to calendar-based statements and finally to more explicit statements (Kuttner, 2018). Although the Fed already provided the market with relevant information in normal times, on which people could form expectations regarding the future policy expectations, the forward guidance policy was more explicit about the future path of the interest rate.
2.2.4 Euro zone
Figure 1: Economic indicators of the US and euro area
(a) Real GDP (b) Inflation
(c) Policy rates (d) Total central bank assets
Source: FRED, SDW ECB
Note: Plot of some economic series for the US (blue) and euro area (red). For (a) the index is
2008Q1=100 and for (d) the index is 2008M1=100.
liquidity measures. For instance, the fixed rate full allotment tender replaced the variable rate tender. The fixed rate tender procedure with full allotment provided unlimited access to central bank liquidity at a fixed rate (main refinancing rate), however it required adequate collateral (Pattipeilohy et al., 2013). Moreover, the list of assets eligible as collateral was temporarily expanded, which in combination with the fixed rate full allotment policy stimulated access to credit. Other non-standard measures adopted by the ECB consist of multiple security purchase programs, such as covered bonds purchase programs (CBPP1 and CBPP2) and securities markets programs (SMP), intended to fix disfunctions in the financial markets caused by the global financial crisis and the sovereign debt crisis (Gambetti & Musso,
asset purchase program (APP), which comprised the private sector assets purchases (CBPP3 and asset-backed securities purchase programme) and a new public securities purchase programme (Andrade, Breckenfelder, De Fiore, Karadi, & Tristani,2016). Under the APP the monthly purchase of public and private sector securities would amount to 60 billion euro and the monthly amount increased to 80 billion euro in March 2016. After 2016 the ECB decided to reduce the amount of the purchases several times, before ceasing its purchases of assets in December 2018. In 2019 the ECB initially intended to maintain the size of the total asset purchases by completely reinvesting the collected amount of principal payments of maturing securities, before announcing another purchase program in September. Contrary to situation in the US after the discontinuation of QE3, the policy rates in the euro area remained unchanged. The timing of the unconventional policies by the ECB could be described in three phases: (i) supporting the banking sector (September 2008- end of 2009); (ii) restoring the transmission mechanism of monetary policy to the real economy and the impaired market functioning during the sovereign debt crisis (2010- late 2012); (iii) implementing a range of more drastic measures, such as large-scale asset purchases, more explicit forward guidance and negative interest rates (from 2013 onwards) (Dell’Ariccia, Rabanal, & Sandri,2018). In short, initially the ECB focused on the providing liquidity the banking sector, intending to ensure that the adverse effects on credit flow to the real economy would be alleviated, while in the later stages the actions were more aggressive and could be classified as increasingly “unconventional” (especially regarding the introduction of asset purchase programmes).
Overall, the non-standard policy instruments appear to have prevented a complete meltdown of the financial sector after the global financial crisis. In most cases, findings in the literature indicate that the implementation of unconventional measures has a positive effect on output and inflation (see e.g. Boeckx et al.,2017;Gambacorta et al.,2014) and financial markets (see
Dell’Ariccia et al., 2018 for discussion and overview of the findings in the literature). During
the period following the financial crisis bank lending rates decreased and there was positive economic growth. However, in comparison with the US, the economic recovery in the eurozone was sluggish in the euro area and the economic slump was protracted. The inflation rate in the euro area showed a positive trend but remained low in the post-crisis period.
3
Literature review
3.1
Empirical evidence
Literature focusing on the impact of unconventional monetary policy on the real economy is relatively limited compared to literature investigating the effects on financial markets. Most studies that have been conducted examine the effects of unconventional monetary policy on financial conditions and evidence on the effectiveness of these policies on the real economy is more extensive for the US compared to the euro area (Borio & Zabai,2016). Studies analysing the impact of unconventional monetary policy on financial markets generally find that unconventional measures put downward pressure on long-term interest rates and raise asset prices. For example, Gagnon, Raskin, Remache, Sack, et al. (2011), Hamilton and Wu (2012)
and d’Amico, English, López-Salido, and Nelson (2012) find evidence that quantitative easing
programs conducted by the Fed reduced long-term yields and compressed term spreads in the US. It is of pivotal importance to know whether financial effects are being transmitted to the real economy and which channels are responsible for the transmission. The transmission channels will be discussed in section 3.2.
Table 1 summarises studies on the effect of unconventional monetary policy on real GDP and prices, which are the main real economic factors. In general, most studies show that the implementation of unconventional measures stimulated economic activity and increased the price level. However, the magnitude of GDP and price responses appears to depend on several factors, such as the countries or areas investigated and economic modelling decisions made by the authors. This section elaborates on the main findings in the literature and the methodology applied by the authors.
Empirical studies face a problem in selecting an appropriate proxy for unconventional monetary policy that measures monetary policy changes when the short-term interest rate is constrained by the zero lower bound. One approach, taken by Boeckx et al. (2017),
Gambacorta et al. (2014) and Burriel and Galesi (2018) among others, is to use central bank
balance sheet changes as a measure of non-traditional monetary policy. The motivation behind using the balance sheet as proxy hinges on the idea that the aim of the expansion of central banks’ balance sheets was to provide liquidity to the liquidity-constrained financial sectors and support to the ailing economies in the wake of the 2008 financial crisis. The underlying assumption is that balance sheet changes reflect unconventional monetary policy. Most countries also experienced a noticeable expansion of the monetary base. However, Gambacorta
et al. (2014) argue that central bank assets are a better measure for unconventional monetary
Table 1: Summary of empirical findings of the impact of unconventional monetary policy measures on the real economy
Study Sample Methodology Real GDP Prices Notes
Gambacorta et al.(2014) 2008M1-2011M6 (CA,euro area, JP, NO, CH, US) Panel VAR Shock: increase in CB balance sheet +1.3%-3.2% +0.2%-1.5% Weale and Wieladek(2016) 2009M3-2014M5 (UK,US) Bayesian VAR Shock: asset purchase US:+0.58% UK:+0.25% US:+0.62% UK:+0.32% Schenkelberg and Watzka(2013) 1995M3-2010M9 (Japan) BVAR Shock: CB reserves Counterfactual: subtract QE-shock +0.07% Industrial production: +0.4% Boeckx et al. (2017) 2007M1-2014M12 (euro zone) BVAR Shock: increase in CB balance sheet +0.1 pp +0.09 pp Increase volume bank lending Gambetti and Musso(2017) 2009Q3-2016Q4(euro zone) TVP-VAR Shock: APP announcement +0.02-0.18 pp 0.06-0.36pp Burriel and Galesi(2018) 2010M1-2015M9(euro zone) GVAR Shock: balance sheet +4 bps +3 bps Baumeister and
Benati(2013) US,UK US+0.9% +1%
potential limitation of using the balance sheet as indicator for unconventional monetary policy is that some policies are announced in advance and hence anticipated by the actors in the real economy. For example, Andrade et al. (2016) show that long-term bond yields reduced on announcement in response to the ECB’s announcement about their asset purchasing program (APP). An additional limitation is that the characteristics of non-standard monetary measures implemented by central banks could change over time (Burriel & Galesi,2018).
An alternative indicator for unconventional monetary used in the literature (see Wu and
Xia (2016),Krippner (2012) and Damjanović and Masten (2016) among others) is the shadow
period.
In order to assess the effect of non-traditional monetary policy instruments on the macro economy, studies usually either conduct a VAR analysis or simulate a calibrated dynamic stochastic general equilibrium (DSGE) model and look at the counterfactual outcome(s). The VAR-based analyses generally apply a shock to the balance sheet or shadow short rate, depending on which unconventional policy measure is selected, and look at the responses of output and prices to an exogenous shock. Counterfactual analyses focus on changes in macroeconomic variables if the purchasing program did not occur or altered in size (both relative to the observed situation with a purchasing program). The drawback of structural models, such as DSGE models, is that these models do not take structural breaks into account and require strong assumptions, i.e. policy makers have perfect foresight and shocks time invariant (Chung, Laforte, Reifschneider, & Williams, 2012). H. Chen, Cúrdia, and Ferrero
(2012) estimate a DSGE model with bond market segmentation and try to estimate the effect of a large scale asset purchases in the US by simulating a counterfactual without asset purchases. They found that the asset purchase program modestly stimulated GDP growth (0.13 bp) and inflation (0.03 bp) with a lasting effect on GDP (after 6 years 0.07% above simulated level without asset purchases).
Initially the majority of the literature focused on the effectiveness of unconventional monetary policy in influencing the macro economy in the US and usually examined the effects of quantitative easing on monetary aggregates. Weale and Wieladek(2016) assess the effect of unconventional monetary policy on the two main macroeconomic variables- output and price level- from 2009 to 2014. They employ a BVAR estimation and use four identification schemes, ranging from Cholesky ordering to combining zero and sign restrictions, to identify shocks in asset purchases. The average2 maximum impact in response to a one percent asset shock on output and prices in the US is 0.58% and 0.62% across all identification schemes, respectively. After replacing assets by the long rate, which yields the term spread in a case without policy rate changes, the authors obtained quantitatively similar findings to the benchmark model.
Haldane, Roberts-Sklar, Young, and Wieladek (2016) decide to use the same identification
scheme and extend the analysis by examining a broader time period (2009-2015). They find that the average response of both output and price was 0.63%, which are slightly higher estimates compared those obtained by Weale and Wieladek(2016).
Baumeister and Benati (2013) take a different approach by tracing the responses on
macroeconomic variables to a spread shock instead of a balance sheet shock and identify the shock by using a combination of zero and sign restrictions. Moreover, they relax restrictions on the evolution of parameters over time by using a time-varying parameters VAR instead of the usual VAR models, in which parameters are constant over time or only have very limited variability. Their counterfactual simulations indicate that adverse effects on output have been
2The maximum impact of the Cholesky decomposition,which is the most restrictive scheme, puts
mitigated and the rate of inflation was higher than in a situation without unconventional monetary policy actions.
The literature on the impact of unconventional monetary in the euro area has been growing in recent years. Gambacorta et al. (2014) model a SVAR for eight advanced economies and qualitatively there appear to be reasonable similarities in the results at the individual country level. In the euro area, the response of output to a balance sheet shock shows a less persistent response compared to the US, while the effect on prices seems to be stronger in the euro area. The underlying economic structure and financial fundamentals of the US and euro area are not identical and this is likely to contribute to slightly different results. Similarly,Boeckx et al.(2017) examine the impact in the euro zone at the aggregate and country level and observe differences in the output response pattern of individual euro area countries. The authors attribute these differences to impaired transmission from the financial sector to the real economy. In particular, they mention dissimilarities in solvency of banks across the core and periphery countries as a factor affecting the transmission. Damjanović and Masten (2016), measuring unconventional monetary by the shadow short rate, have comparable findings that the transmission to the real economy is weaker in the periphery countries. In the euro area, Boeckx et al. (2017) find that unconventional monetary policy innovations boost economic activity and lead to a rise in the price level. The results of Damjanović and Masten (2016) also indicate that prior to 2014 the stimulus by the ECB had a weak effect on the macroeconomy. Qualitatively the output results are similar to a conventional monetary policy shock while prices more or less follow the response of output in the unconventional case (Boeckx et al., 2017). This is in contrast with the sluggish response of prices found in the case of conventional expansionary monetary policy.
Haldane et al. (2016) study the impact of quantitative easing programs for several countries.
They apply four different identification schemes and only find one statistically significant real variable. Their explanation for this finding is that the ECB’s reasons for expanding the balance sheet was related to providing liquidity, while the central bank’s objective in the areas displaying significant results (including the US) was to loosen monetary policy. According to the authors, the purpose and execution are central elements in successfully affecting the broader economy.
Wieladek and Garcia Pascual(2016) estimate a BVAR with the same four identification schemes
but they consider a different sample period (2012-2016). In contrast to the results obtained by
Haldane et al. (2016), they find significant positive results on output and prices, ranging from
0.07-0.15% and 0.05-0.1% respectively3, following a one percent shock in asset purchases. The contrast between the findings indicates that the results are sensitive to the sample period under consideration.
3.2
Transmission channels
In the conventional, pre-crisis monetary models the perfect markets assumption was usually made. After the crisis, the literature identified several potential transmission channels of balance
sheet policies and departed from the frictionless financial sector assumption. The effectiveness of the transmission from financial markets to the real economy is reliant on imperfections in financial markets. In other words, frictions in the financial markets impair the transmission of balance sheet policies to the real economy while in a frictionless world these policies do not affect the real variables. Under the assumption of no market frictions, complete markets, completely rational and infinitely-lived economic agents and no arbitrage opportunities, quantitative easing does not exert effect on the real economy, since short-term and long-term interest rate remain unchanged (Eggertsson & Woodford, 2003). Haldane et al. (2016) discuss six transmission channels of unconventional monetary policy which have been identified in the literature. This section provides an overview of some of these channels.
The portfolio rebalancing channel posits that assets are viewed as imperfect substitutes by investors and bond holders in the private sector. For example, some investors might have a preferred habitat for bonds with a particular maturity (Vayanos & Vila,2009). By purchasing assets central banks can alter portfolios in the private sector and hence has the ability to influence the price and yield of these particular assets. If a central bank decides to purchase long-term assets from secondary markets, the price of those assets increases due to a reduction in supply and the yield decreases as a result of the inverse relation between prices and yields. With lower returns on the investors’ current assets in the portfolio, they could get higher returns by rebalancing their portfolio with other assets. This will also drive up the price of those other (riskier) assets and ultimately lower the yield of those other assets. Long-term rates decrease as asset prices increase and the effects spill over to other long-term assets. With lower rates the cost of financing for both investors and banks is reduced, providing stimulus to aggregate spending.
Unconventional monetary policy can also influence the supply of credit and lending conditions (credit easing channel). Andrade et al. (2016) discuss the rebalancing of bank assets along the duration dimension. They claim that as a result of asset purchasing programs bank’s longer-term and more risky assets (with higher liquidity or duration risk) are replaced with short-longer-term central bank reserves, which are deemed to be safe. As a consequence, banks are exposed to less risk which enables them to lower lending rates. Furthermore, banks are looking for higher returns as a result of the compression in the term spread, leading to a rise in risky loans. The increase in the availability of loans stimulates aggregate spending, since the provision of liquidity to liquidity-constrained economic agents enables them to invest in profitable investment opportunities. The claims by Andrade et al. (2016) indicate that the credit easing channel could also operate through the portfolio rebalancing mechanism. Evidence for this channel is found for the euro zone (see Altavilla, Carboni, & Motto, 2015 and the US (see d’Amico et
al., 2012;Weale & Wieladek, 2016. Related to a reduction of risks on the central bank balance
banks’ balance sheets. The combination of stronger capitalisation and expansions in the balance sheets enable banks to increase the volume of loans to the private sector.
The signalling channel influences market expectations of the future policy rate path. Forward guidance can directly signal the future policy rate path. For example, if the central banks signals that it will keep the short-term interest rate at the effective lower bound for a prolonged period of time, agents’ expectations will be revised downwards (reduction expectation component long-term rates) (Gambetti & Musso, 2017). The effectiveness of the signal about the future policy rate depends on the clearness of the communication (i.e. to what degree the imperfectly informed public understands the intended policy and its implications) and whether central banks commit by acting in accordance with their signal. For instance, central banks can show commitment by buying assets after providing a signal that future policy rates will remain low (Bowdler & Radia, 2012). The signalling channel reduces the term premium component of the interest rate and mitigates volatility in inflation expectations, both putting downward pressure on the long-term real interest rate (Wieladek &
Garcia Pascual, 2016). Long-term real rates influence contemporary investment and
consumption decisions and lower rates, which reduce the borrowing costs, encourage investment and promote economic activity. Bauer and Rudebusch (2014) conclude that signalling played an important role in lowering the interest rate during QE1 in the United States, while Altavilla et al. (2015) argue that in the euro zone signalling, besides other channels, contributed to the decline in long-term yields.
According to the liquidity premium channel, central bank balance sheet operations can improve liquidity when markets are dysfunctional, giving rise to high liquidity premia
(Krishnamurthy & Vissing-Jorgensen,2011). To increase liquidity, illiquid assets are purchased
(long-term securities) with a liquid asset (central bank reserves). Consequently, yields of the most liquid bonds will increase relative to less liquid bonds. So, central banks have the power to increase liquidity and reduce liquidity premia.
The exchange rate channel implies that the nominal exchange rate should depreciate in response to asset purchases by a central bank. A depreciation should lead to a stimulative effect on exports and a reduction in imports, improving the trade balance. Changes in future interest rates and risk premia expectations of agents can affect the exchange rate (Inoue & Rossi,2019).
Inoue and Rossi (2019) argue that in periods, in which central banks resort to unconventional
monetary policy, both short- and longer-term rates are able to alter the exchange rate.
4
Empirical Methodology
This section provides information regarding the econometric model and identification used to examine the impact of unconventional monetary policy on the real economy. I give a brief overview of the VAR model and the strategy used to identify unconventional monetary policy shock, i.e. unanticipated innovations in the size of central banks’ balance sheets.
4.1
VAR model
Since the seminal work of Sims (1980), VAR models have become a popular instrument in macroeconomic research. Traditionally, the effectiveness of conventional monetary policy has been investigated extensively in the literature by using structural VARs (see e.g. Christiano,
Eichenbaum, & Evans, 1999; Bernanke & Blinder, 1992. In the period when policymakers
resorted to non-traditional monetary policy, structural VARs remained one of the main tools to analyse the macroeconomic implications of monetary policy operations. In VAR models, each endogenous variable in regressed on its own lags and the lags of all other variables included in the system. In this paper, the reduced-form VAR specification is given by the following representation:
Yt= α + A(L)Yt+ ut (1)
ut∼ N (0, Σ) (2)
where Y is a vector consisting of the following six endogenous variables in the benchmark specification: the log of seasonally adjusted consumer price index (CPI); the log of seasonally adjusted real GDP; the log of central bank total assets; long-term interest rate (yield on 10-year government bond); the log of stock market index; a measure of financial volatility. The term A(L) represents a matrix polynomial in lag operator L, α is a vector of constants, while utis a
vector of reduced-from disturbances with mean zero and variance-covariance matrix E(utu
0
t) =
Σ. Identification of a structural VAR model requires additional identifying restrictions, since orthogonal, structural economic shocks (t) have to be disentangled from reduced-form shocks
(ut). The structural form of the VAR can be obtained by pre-multiplying equation (1) by matrix A and defining µ = Aα, B(L) = AA(L) and Aut= t:
AYt= µ + B(L)Yt+ t (3)
The relationship between the structural and reduced-form disturbances can be expressed by using a matrix A, which specifies the contemporaneous relations between variables. Ultimately, the relationship between the shocks boils down to Aut = t. In this paper unconventional
specifically, I opt to impose a Minnesota prior (see Doan, Litterman, & Sims,1984;Litterman,
1986) on the reduced-form coefficients and hence assume that the variables follow a random walk, which is frequently the case for macroeconomic variables. Bayesian vector autoregression (BVAR) models alleviate the problem of over-parameterisation, which is common issue when using macroeconomic data in combination with a short sample size. For example, a VAR model comprising six endogenous variables (n = 6), a constant, and four lags (p = 4) requires 150 coefficients for the complete model (n(np + 1)). In each individual equation 25 coefficients (np + 1) coefficients have to be estimated. With many coefficients to be estimated relative to the sample size, the application of a Bayesian approach improves the quality and reliability of interference compared to a classical VAR. The Minnesota prior assumes that the residual-variance matrix Σ is fixed prior to the posterior sampling. Additionally, it is assumed that Σ is replaced by an estimate, ˆΣ. Another convenient assumption is that Σ is diagonal, which imposes independence between coefficient of separate equations. I decide to set the diagonal of the variance-covariance matrix equal to the diagonal elements from VAR estimated by OLS. By, construction, the non-diagonal elements are set to zero. The Minnesota prior, which is shown in equation (4), is based on the normal prior distribution.
β ∼ N (β
M, VM) (4)
where β
M and VM denote the prior mean and prior variance, respectively. By selecting the prior
mean and prior variance, the researcher puts a structure on the Minnesota prior in accordance with the beliefs regarding the empirical context. The rationale behind the Minnesota prior is to shrink the model towards a random walk by choosing hyperparameters, which govern prior variance of the coefficients and consequently the dynamics of the model. In empirical work the prior mean β
M is usually set to zero in the case of growth rates data, such as economic growth
and the inflation rate, while for data in levels the prior mean is usually set (close) to one. This reflects the belief that the levels data, such as output and consumption, are highly persistent and that a random walk specification is appropriate. I choose a prior mean of 1 and 1 for the US and euro area, respectively. This incorporates the belief that the data is highly persistent and taking into account that all data is included in (log) levels. The prior variance is given by the following equation:
VlM,i,j = λ1 lλ3 2 for i = j λ1λ2σi lλ3σj 2 for i 6= j (5)
where l is the considered lag, σi2 and σj2 denote the residual variances for variables i and j estimated by OLS, and λ1 (own lag tightness), λ2 (tightness lags other variables) and λ3 (lag
1, respectively, in the BVAR model for the US. The values of the hyperparameters selected for the Minnesota prior used in the eurozone model are 0.1, 0.6 and 1. The selection of the hyperparameters are guided by the avoidance of explosive dynamics of the impulse response functions. In addition, these hyperparameters are combinations of some frequently used values assigned to the hyperparameters the literature. The number of lags are selected using the optimal number of lags indicated by results of the likelihood ratio test. For the US model four lags are included, while for the euro area the number of lags is six. The estimated models satisfy the stability condition, since all roots lay inside the inverse unit root circle.
4.2
Identification of the model
Restrictions are required to identify the VAR model and need to be imposed on the structural model. In order to analyse the effect of an isolated innovation structural shocks are assumed to be orthogonal. The recursive ordered restrictions are imposed on the contemporary relations matrix A. In the ordering of the variables in the Cholesky decomposition I follow the scheme
of Weale and Wieladek (2016). However, this paper uses total assets on the central bank
balance sheet instead of asset purchases announcements and augments the scheme with a sixth variable- a volatility indicator. This scheme was selected since it incorporates the notion of slow-moving macroeconomic variables and fast-slow-moving financial variables. The identification scheme is given in table 2. Restrictions are solely imposed on the contemporaneous effects between the endogenous variables and the relations between variables remains unrestricted in future periods. According to the identification scheme, macroeconomic variables react to monetary policy and financial shocks with a lag, under the assumption of sticky prices and lagged response of output. In the monetary policy literature output and prices are usually ordered first in accordance with the nominal rigidity theory (i.e. persistence in output and prices inertia in response to a monetary policy shock (Christiano, Eichenbaum, & Evans, 2005). Furthermore, the identification assumes that monetary policy variables do not react contemporaneously to innovation in the financial sector and stock market, while monetary policy does react to shocks to macroeconomics economic variables on impact. Finally, financial and stock marked based variables- stock market index and VIX- respond instantly to innovation in the real sector and monetary policy shocks (Bekaert, Hoerova, & Duca,2013). So, it is assumed that the variables ordered first are the slow-moving variables, while the variables ordered later react faster to shocks.
5
Data
The sample spans the period from 2008M1 until 2015M1 for the US and from 2007M1 until 2014M12 for the euro area, since during this period central banks employed unconventional monetary policy instruments. The start date for the euro area is inspired by other studies, such
Table 2: Identification scheme 1: Cholesky decomposition Shock
Log CPI Log Log Long-term rate Log VIX
Real GDP CB assets Stock Prices
Log CPI 1 0 0 0 0 0
Log Real GDP x 1 0 0 0 0
Log CB assets x x 1 0 0 0
Long-term rate x x x 1 0 0
Log Stock Prices x x x x 1 0
VIX x x x x x 1
Note: This table displays the restrictions imposed by the recursive identification scheme. A "x" denotes that the response of the variable (row) to the shock (column) is unrestricted, "1" indicates that the response of to the shock is one and "0" indicates that there is no response.
for the US is based on the start date of Gambacorta et al. (2014). The sample period for the US and eurozone ends in January 2015 and December 2014, respectively, for two, region-specific reasons. First, the ECB announced the Expanded Asset Purchase Program (EAPP) in January and hence the shock can no longer be classified as unanticipated, since the purchases are announced multiple months in advance (Boeckx et al.,2017). Second, during 2015 interest rates in the US started to rise and hence the constraint of the ZLB was less relevant than in the period when the fed funds rate was close to zero. The period under consideration is short and therefore monthly data is used, which is a common approach in the empirical literature analysing the effect of unconventional monetary policy. The pre-crisis period was not included in the sample, because the validity of the results is questionable in normal economic times with higher policy rates and the results may only be relevant in periods of financial stress and policy rates close to the ZLB (Boeckx et al.,2017). The data set contains macroeconomic, financial, and monetary variables. A detailed overview of the US data is included in table A1 in appendix A, while Table A2 gives a detailed description of the EU data. The real economy is represented by the usual variables- output and prices. Real GDP is a proxy for output and the quarterly GDP data is Chow-Lin interpolated into monthly data based on monthly industrial production data
4 Prices are proxied by the consumer price index. The indicator of unconventional monetary
policy is the total amount of central bank assets, which is a component of the central bank balance sheet. The 10-year yield on government bonds is included to capture the transmission of unconventional monetary policy through the long-term interest rates. Long-term interest rates
4An overview of the Chow-Lin interpolated GDP series is given in B3 in appendix B. The technique
influence the spending and investment decisions of businesses and households, which ultimate affect the macroeconomy. The inclusion of stock market indices is based on the idea that purchasing long maturity bonds and other assets is likely to result in a shift towards other assets, which has an effect on equity prices (portfolio rebalancing channel). The volatility index (VIX) and the Composite Indicator of Systematic Stress (CISS) ofHolló, Kremer, and Lo Duca
(2012) proxy for financial turmoil and economic uncertainty in the US and eurozone, respectively, during the post-crisis period. The inclusion of a financial stress proxy is essential in disentangling exogenous balance sheet shocks from the endogenous response of balance sheet policies to the financial and economic turmoil caused by the financial crisis (Gambacorta et al.,2014). All the US time series are retrieved from the Federal Reserve Economic Data (FRED) database with the exception of the data on the stock market index and exchange rate and bank lending data, which are obtained from Datastream and BIS, respectively. The main source of the euro area data is the ECB’ statistical data warehouse (SDW). All variables except for the uncertainty proxy and the long-term interest rate are transformed into natural logarithms. The Wu and Xia (2016) shadow-short rate data is collected from one of the author’s website5, while the rate computed by
(Krippner,2016) is obtained from the Reserve Bank of New Zealand’s website. Some additional
variables will be included one at the time (as a seventh variable), similar to the approach of
(Gambetti & Musso, 2017), to examine the activation of some transmission channels besides
the channels activated by the inclusion of stock prices and the long-term interest rate. The nominal effective exchange rate will be included to analyse whether the exchange rate channel is operational, while volume of credit to the non-financial sector and bank lending rate will act as seventh variable to capture the transmission from the financial sector through the credit (easing) channel to the real economy.
A visual representation of the time series included in the benchmark specification for the US and euro area is given in figure B1 and figure B2 in appendix B, respectively. One can clearly observe that the global financial crisis is included in the sample period. There appears to be a considerable decline in real GDP, prices, and the stock price index, while at the same time there is an enormous increase in the volatility index. In the US, the series real equity price index, real GDP and consumer price index have an upward trend again after the crisis. The time series of the euro area display an additional period of large volatility, namely the European sovereign debt crisis. So, the sample period is characterised by some events associated with high levels of volatility and uncertainty.
In contrast to the eurozone, the policy rate in the US - the effective Federal Funds rate-, which is plotted in figure 2 along with two shadow rate measures, quickly drops toward zero and remains more or less at the ZLB until the end of the sample period. One convenient property of the shadow rates is that these rates can take values below zero. The large increase in the total amount of central bank assets coincides with the policy rate approaching the ZLB. Therefore, one can observe that the policy rate was no longer the main instrument of the Fed, who shifted
Figure 2: Evolution of policy rate and shadow short rate over time
United States euro area
towards unconventional monetary policy, primarily quantitative easing. Overall, the US long-term interest rate decreased over time while also showing some fluctuations.
The stationarity of the series is analysed by conducting Augmented-Dickey-Fuller (ADF) tests. The test results of the benchmark specification and transmission channel data for the US and euro area are displayed in table B1. In the case of the US, all variables are integrated of order one except for the bank loan rate and perhaps the VIX, which is just significant at the five percent level. At the ten percent level real equity prices could also be considered in the group of variables integrated of order zero. Regarding the eurozone, all variables are integrated of order one. Outcomes of the Johansen integration test indicate that there are one or four and one to four cointegration relationships for the US and the euro area, respectively,according to table B1 and B3. However, alterations of the parameters of the Johansen integration test (not displayed in this paper) indicate that there could be more integration relations. The number of cointegration equations indicated by the trace and eigenvalue tests depends on the trend assumption and lag length selected. There are three option to proceed on the basis of the stationarity and cointegration test results: (i) first-differencing the variables; (ii) estimate a VECM, which explicitly models the long-term behaviour of the cointegration equations; (iii) estimation in (log) levels and allowing for implicit cointegration relationships in the data (Sims,
Stock, & Watson, 1990). Given the relatively short size sample period and the presence of
(potentially) multiple cointegration relationships in the data, the VAR will be estimated in (log) levels, which yields consistent estimators according toSims et al.(1990).
6
Results
are discussed.
6.1
Benchmark model
Figure 3 and 4 display the impulse responses for an unconventional monetary policy shock for the US and euro area, respectively. The impulse responses, which show the dynamics of the endogenous variables over time due to innovations in central bank assets, are obtained by estimating a VAR using Bayesian methods. In particular, the prior used is the Minnesota prior throughout this section. The time horizon of the impulse response functions is 40 months. The solid blue lines represent the posterior median response.6 Following Boeckx et al. (2017) and
Boeckx, de Sola Perea, and Peersman (2020) all the variables expressed in logarithms are
multiplied by 100. This is done in order to make sure that one can interpret the standard deviation shock of logarithmic transformed on other variables, expressed in logarithms, in percentages.
The Bayesian VAR results provide an indication that the characteristics of the shock differ between the US and the euro area. On impact, the magnitude of the balance sheet shock in the US and the eurozone is around 4%. In accordance with the findings ofWeale and Wieladek
(2016) and Gambacorta et al. (2014), the responses of output and prices illustrate that the unorthodox policy measures are effective in stimulating the real economy in the US.
Looking at the euro area, prices increase as a result of a positive innovation to the volume of assets on the ECB’s balance sheet, while the response of output is negative for the first two years. After two years the response is slightly positive over the remaining time horizon, constantly around 0.01%. The results found byHaldane et al.(2016) also indicate that real GDP decreases on impact for the euro area, however their results are not statistically significant. However, the results in figure 4 regarding the response of output to a balance sheet shock for the euro area contrast with the findings of Gambacorta et al.(2014) and Boeckx et al.(2017). These studies find that output responds positive the balance sheet shock for at least 12 months. The positive response of prices in the eurozone is quantitatively reasonably in line with the existing literature except for the lack of a decaying effect over time once the peak has been reached.
The volatility proxy of both the euro area and US experiences dynamics characterised by an initial decrease in the first few months and a decay of the impact of the shock as time passes. In both the US and euro area, the response of the long-term interest rate demonstrates a negative effect in the first few months, followed by a period of a raising rates, which will eventually experience a decrease and remain negative. Real equity prices for the US and euro area appear to respond positively and negatively, respectively, throughout the time horizon.
All in all, prices show a persistent increase in both areas, while also the dynamics of the long-term rate in the US are to a certain extent akin to the response of the long-term rate in the
6It should be noted that the software used for the estimation, Eviews, do not allow to incorporate
euro area following a balance sheet shock. In contrast, the effects on output exhibit noticeable differences between the regions. Section 6.3 will elaborate upon the similarities and differences between the impact of an unconventional monetary shock.
Figure 3: BVAR- Results benchmark specification for the US
CPI -.2 -.1 .0 .1 .2 5 10 15 20 25 30 35 40 Real GDP -.2 -.1 .0 .1 .2 5 10 15 20 25 30 35 40 Total CB assets -1 0 1 2 3 4 5 10 15 20 25 30 35 40 Long-term rate -.04 -.02 .00 .02 .04 5 10 15 20 25 30 35 40
Real Equity Price
-2 -1 0 1 2 5 10 15 20 25 30 35 40 VIX -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 5 10 15 20 25 30 35 40
Note: This figure shows the response of all endogenous variables included in the benchmark specification to an unexpected one standard error central bank assets shock. The blue line represents the posterior median responses. The time horizon is 40 months. CPI, real GDP, central bank assets and real equity, which are all included as logs, are multiplied by 100. The variables are recursively ordered. Sample period under consideration is 2008M1-2015M1. Hyperparameters: µ1= 1; λ1= 0.1; λ2= 0.9; λ3= 1.
6.2
Transmission channels
Figure 4: BVAR- Results benchmark specification for the euro area CPI -.2 -.1 .0 .1 .2 5 10 15 20 25 30 35 40 Real GDP -.10 -.05 .00 .05 .10 5 10 15 20 25 30 35 40 Total CB assets -1 0 1 2 3 4 5 5 10 15 20 25 30 35 40 Long-term rate -.02 -.01 .00 .01 .02 5 10 15 20 25 30 35 40
Real Equity Price
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 5 10 15 20 25 30 35 40 CISS -.010 -.005 .000 .005 .010 5 10 15 20 25 30 35 40
Note: This figure shows the response of all endogenous variables included in the benchmark specification to an unexpected one standard error central bank assets shock. The blue line represents the posterior median responses. The time horizon is 40 months. CPI, real GDP, central bank assets and real equity, which are all included as logs, are multiplied by 100. The variables are recursively ordered. Sample period under consideration is 2007M1-2014M12. Hyperparameters: µ1= 1; λ1= 0.1; λ2= 0.6; λ3= 1.
channel and the uncertainty channel.
First, to assess whether the credit easing channel was activated, I add, in turn, a lending rate and volume of bank loans. To be more precise, the bank loan volumes are loans to the non-financial private sector, i.e. sum of loans to non-financial corporations and households. In theory, one could expect that an increase in the central bank assets would lead to an increase in the bank loan volumes, because asset purchases are generally financed by issuing new reserves, which are in possession of the banking system. This puts downward pressure on the interest rate on reserves and banks, faced by the lower rates, are incentivised to expand their lending
(Weale & Wieladek, 2016). Figure C1 and C5 in appendix C show the results of benchmark
specification extended with the bank loans volume for the US and euro area, respectively. The bank loan volume shows a negative response to a central bank assets innovation in the euro area and a positive response in the US. However, a caveat is that I am not able to comment on whether the median responses are included in the Bayesian credible sets. Looking at another potential transmitter of non-traditional monetary policy through the credit easing channel, the bank lending rate, figure C6 appendix C presents that there appears negative effect on the bank lending rate in the eurozone. In the case of the US, figure C2 displays that in the first few months the bank lending rate declined and after ten months the response was positive. Overall, the results provide a contradictory impression about the credit channel. On the one hand, the positive (negative) response of volume of bank loans in the US (euro area) to the balance sheet shock indicates that the credit channel was activated in the US. On the other hand, the bank lending rates ultimately increased, after initially displaying a negative response (decreased) in the US (euro area). This observation provides an indication that the credit channel was active in the euro area but was only temporarily activated in the US.
Second, I added the nominal exchange rate to the benchmark model to examine whether the exchange rate channel was operational. The impulse response functions are presented in figure C3 and C7 in appendix C . The path of the nominal exchange rate after the central bank assets shock puts downward pressure on the exchange rate for both the US and euro area. Over time the US exchange rate appreciates, while the central bank balance sheet shock has a lasting negative effect on the EU nominal exchange rate.
Finally, I decide to simultaneously add all three variables to the benchmark specification. The results can be found in figure C4 and C8 in appendix C . A lower value for the tightness hyperparameter, λ1, is selected, following the suggestion by Bańbura, Giannone, and Reichlin
separately to become the seventh variables). Most noticeably is the positive response of the bank lending rate in the eurozone after some time. Compared to the benchmark specification discussed in section 6.1 the magnitude of output and prices decreases considerably in the nine-variable model.
6.3
Comparison between the US and eurozone regarding the
responses to a central bank assets shock
In general, prices in both the US and eurozone react positively to an assets shock. On the contrary, the other real economic variable, output, does not show similar responses. According to the impulse response functions of the Bayesian VAR in section 6.1 output tends to increase in the US after an innovation to the central bank balance sheet, while the response in the euro area typically has a negative response. Moreover, the non-traditional policies seem to be more effective in affecting the macroeconomy in the US than in the euro area. Not only do the policies have a positive effect on both macroeconomic variables in the US, the non-standard monetary policy measures appear to also exert a stronger effect on the real economic variables. This is in agreement with the common observation in the literature that unconventional policy measures were more successful in the US in recovering the economy than in the EU.Q. Chen, Lombardi,
Ross, and Zhu(2017) also observe that the effects are stronger in the US than in the euro area.
They suggest that the credit and confidence channel were important channel in achieving the strong effects. Moreover, they attribute the strong effect to the type of assets bought- sovereign and private assets- during the asset purchase programmes.
In the benchmark model in section 6.1 some variables representing transmission channels were already included in the model. I will elaborate upon the finding of the channels related to certain variables and compare the results between the US and euro area based on my results and the finding in the literature. Additionally, I will just touch upon the transmission channels discussed in section 6.2.
The benchmark results indicate that variables related to the portfolio rebalancing channel, real equity prices and long-term interest rate, appear to respond positively to a central bank asset shock in the US, while the opposite occurs in the euro zone. Concerning the US, these results are in conformity with the findings of Weale and Wieladek (2016), who claim that the in the US the asset purchases (and announcement of purchases) result in lower long-term yields and have a positive effect on stock prices, indicating the activation and relevance of the portfolio rebalancing channel. With respect to the euro area, Wieladek and Garcia Pascual (2016) find that an asset purchase announcement has similar results on the long-term yield as the findings by
Weale and Wieladek(2016), while they find no convincing evidence for a positive impact on real
share prices. These stock price findings are in contrast with the results ofGambetti and Musso
(2017). Gambetti and Musso (2017) find that an asset purchase (announcement) innovation
and Tillmann(2016) shows that unconventional monetary policy had a moderate positive effect on stock prices and a mild negative impact on long-term yields. Finally, Fratzscher, Lo Duca,
and Straub (2018) find that QE1 affected stock prices (long-term yields) in the US positively
(negatively). Overall, the portfolio rebalancing channels appears to be operating successfully in the US, while the evidence for the euro area is mixed.
Another benchmark model variable, VIX, is a proxy for the uncertainty channel. Weale and
Wieladek(2016) stress the importance of the uncertainty channel in the US in transmitting the
QE shock, while Wieladek and Garcia Pascual (2016) find that the uncertainty channel is not operative in the eurozone. Gambacorta et al. (2014) find that in both the US and euro area innovations in the balance sheet reduce uncertainty. Their estimated indicates that the impact was stronger and more persistent in the US. The estimated results in the current study indicate that both the euro area and the US follow similar dynamics, indicating that in both regions the uncertainty channel was operative.
The variables related to the credit channel, bank lending rate and volume of loans, demonstrate in section 6.2 that the dynamics of the bank lending rate and volume of loans after an asset purchase shock differs. Gambetti and Musso (2017) find that in the euro area the credit channel was activated. They claim that in the short run this was mainly achieved by the decline in the lending rate and thereafter via the increase in the volume of loans. For the
US, Feldkircher and Huber (2018) find less convincing evidence that the credit channel is
activated in the US. So, the results in section 6.1 show a different response for the bank loan volume compared to Gambetti and Musso (2017). Regarding the exchange rate channel,
Gambetti and Musso(2017) find that the asset purchases result in a depreciation.
This might have to do with the fact that the US has a market-based financial system and the euro area has a bank-based system. The initial focus on stabilising the banking in the euro area and delaying the implementation of large-scale asset purchase programmes might have contributed to the stronger recovery from the crisis and more effective unconventional in the US. For example, QE1 was the most effective large-scale asset purchase programme, mainly because it was unanticipated. For the euro area the large asset purchase programmes were already anticipated.
7
Robustness
Other robustness exercises are changing the lag length, sample period, obtain identification by using sign restrictions, selecting a looser prior and opting to use a different prior- the Sims-Zha prior. The figures of the robustness exercises for the US and euro area are included in appendix D.
7.1
Changes in variables measurement
First, the measure used for prices does affect the estimated impact in both regions. When using the core CPI, the magnitude of the positive response is smaller compared to the benchmark case. Results also differ when using the GDP deflator instead of the consumer price index. The effect of the shock decays more in the later stages for the US compared to the benchmark situation, while for the euro area the response turns negative during the first year instead of positive.
Second, replacing real GDP by industrial production yields qualitatively similar results in the US. The only difference is that the reaction of output is larger, which was also found by
Gambacorta et al. (2014). Looking at the euro area, there is a difference compared to the
benchmark case. The output response to a central bank assets shock is now negative after two years instead of marginally positive.
Third, the results are to some extent robust to using the monetary base as shock variable instead of assets purchase. The results of the macroeconomic variables are qualitatively and quantitatively remarkably similar to benchmark case for the US. The most prominent alteration in the responses is that in decline in the long-term interest rate is larger and remains negative throughout the time horizon.
As a final robustness check of the measurement of the variables, the shadow short rates are substituted for central bank assets. The US responses appear to be qualitatively more or less robust to theWu and Xia(2016) orKrippner (2012) shadow short rate, while the results in the eurozone are sensitive to including the shadow short rate instead of central bank assets. Most strikingly, the responses of real GDP, the equity price and long-term interest rate switch signs when using either shadow short rate instead of total central bank assets.
7.2
Sign-restricted VAR
A sign-restricted VAR with four lags is estimated to check whether the results are robust to an alternative identification method. Bayesian methods are used in the computation of the impulse response function. The sign-restricted VAR is based on the procedure developed by
Uhlig(2005). I refer toUhlig(2005) for an elaborate discussion of estimation procedure and the
and there is a sudden decrease during the first year. In the first months after the shock some variables do not display a response in line with theory. Figure D7 contains some evidence that output rises and its response to an asset shock reaches its peak effect of about 0.2% after seven months. Overall, the variables do display the expected sign over time, but in the period 4-11 months after the shock some variables do show the opposite sign to what one would expect.
Focusing on the euro zone, similar conclusions can be drawn about the significance of the responses. Real GDP reacts predominantly positively throughout all 40 months and prices fluctuate around zero, although the results are not statistically significant. The lack of significance might be caused by the aftermath of the global financial crisis and the presence of the sovereign debt crisis during the sample period. Finally, the results indicate that in both the euro area and US the response of real equity prices is positive for the first two months. This is partly by construction, since the sign restriction imposed for the first period is positive.
7.3
Changes in lag length, sample period, hyperparameters or
prior
The benchmark results for the US are qualitatively insensitive to selecting 2 or 12 lags instead of 4. The same conclusions can be drawn from to changing the lag length to 2 and 12 instead of 6 in the euro area, except for the response of the long-term rate in the model with 2 lags. In addition, the results for both regions are insensitive to changing the sample period. The changes in the sample period for each region is specified in the notes of the corresponding figures. Similarly, choosing looser hyperparameters does not exert an enormous effect on the impulse response function for both regions. The figures can be found in appendix D with the corresponding hyperparameters specified in the figure notes. Finally, I opt to change prior from the Minnesota prior to the Sims-Zha prior. I refer to the paper of Sims and Zha(1998) for an explanation of the prior, its properties and the specific hyperparameters. Overall, the results are reasonably similar when using the Sims Zha prior. The main differences could be observed on impact and in the first couple of months (see appendix D figures D12 and D24).
