University of Groningen Financial markets: market Information, investment strategies and spillovers Dreher, Ferdinand Torin

(1)

Financial markets: market Information, investment strategies and spillovers

Dreher, Ferdinand Torin

IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document version below.

Document Version

Publisher's PDF, also known as Version of record

Publication date: 2019

Link to publication in University of Groningen/UMCG research database

Citation for published version (APA):

Dreher, F. T. (2019). Financial markets: market Information, investment strategies and spillovers. University of Groningen, SOM research school.

Copyright

Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).

Take-down policy

If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.

Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum.

(2)

531888-L-bw-Dreher-SOM 531888-L-bw-Dreher-SOM 531888-L-bw-Dreher-SOM 531888-L-bw-Dreher-SOM Processed on: 4-6-2019 Processed on: 4-6-2019 Processed on: 4-6-2019

Processed on: 4-6-2019 PDF page: 93PDF page: 93PDF page: 93PDF page: 93

Re-examining Financial

Market Spillovers at the

Lower Bound

4.1 Introduction

In response to the global financial crisis, central banks across the world lowered short-term interest rates – their conventional monetary policy (CMP) tool. In many cases, this resulted in central banks reaching the zero lower bound. Without their main policy tool available, central banks therefore introduced a range of unconventional monetary policy (UMP) measures, such as quantitative easing and forward guidance (see Blinder et al., 2017). The implementation of these unconventional monetary policy measures by all G4 central banks (the Federal Reserve, the European Central Bank, the Bank of England, and the Bank of Japan) in turn has triggered a large literature examining the announcement effects of UMP on financial markets.

Within this literature, authors have focused primarily on the effects of UMP on domestic government bond yields, stock prices, and exchange rates. The consensus is that announcements of unconventional policies had significant effects on domestic markets by lowering longer-term yields, depreciating currencies, and strengthening equity markets. Others have gone beyond effects on domestic financial markets and assess the spill-over effects into other

(3)

currency areas. This has been motivated by the sizeable eﬀects felt in both developed and emerging economies of unconventional monetary policies of central banks in advanced countries. As to spill-over eﬀects, most studies report that the impact of the Federal Reserve on most other economies is strong, while the spillovers of other major central banks onto world markets display more heterogeneity and generally are weaker.

Our contribution is to study multiple asset classes within and across currency areas jointly and thus to provide a comprehensive comparison of spillovers of policies pursued by major central banks. We closely follow the methodology of Ehrmann et al. (2011), henceforth referred to as EFR, that uses heteroskedasticity in financial markets to identify shocks across multiple asset classes. This approach requires weaker assumptions than the commonly used event study approach. In this way, we are also able to differentiate between direct international spillovers within asset classes, and total spillovers that also account for spillovers via a third market. We are interested in financial market linkages at the lower bound in general, and in particular how unconventional monetary spillovers differ from those of conventional policies. Our paper is related to the work by EFR and Hayo and Niehof (2013) in that it studies spillovers comprehensively using heteroskedasticity for identification, but differs in that we specifically look at spillovers of unconventional monetary policy. Importantly, rather than focussing on a formal generalisation of Rigobon and Sack (2004) to the multi-country and multi-asset case as in Hayo and Niehof (2013), we instead rely on the econometric approach of EFR. The analysis of Wright (2012) has in common with our work the VAR set-up using heteroskedasticity for identification but focuses only on interest rates from government and corporate bonds, while Ehrmann and Fratzscher (2017) and Bayoumi and Bui (2012) focus on individual asset types. Rogers et al. (2014) use heteroskedasticity in a basic unilateral one-market framework as an alternative to their monetary policy surprise measure, albeit without macro controls and common shocks. By employing a richer econometric set-up than most existing studies that generally focus on one central bank or one market at a time, we can reassess existing results on spillover effects. A particularly interesting comparison can be made to the time-varying approach by Belke and Dubova (2018).

(4)

We include shadow rates as an alternative measure for monetary policy and control for a large number of national and global macroeconomic shocks. This allows us to draw ﬁner conclusions about the interconnectedness amongst diﬀerent international asset markets within and across-countries.

Our contribution to the literature is threefold. Firstly, we are, to our knowledge, the first to analyse spillovers of global unconventional monetary policy in a comprehensive framework that models all major central banks. This is no trivial nor easy task, especially given the specifics of monetary policy during this period and the dimensionality of studying multiple assets and countries together. In doing so, we can show contemporaneous spillovers across markets and asset classes. This enables us to compare the policies of all central banks together and make statements about their relevance for world markets. While our focus lies on the spillovers of unconventional monetary policy, we are interested also in whether linkages between financial markets change at the lower bound.

Second, we employ a method of identiﬁcation which has been increasingly employed in the literature as an alternative for identifying monetary policy shocks, but which has rarely been used for studying monetary policy in a comprehensive analysis of multiple ﬁnancial markets.

Third, we contribute to the debate on the spillover eﬀects of unconventional monetary policy by incorporating the recent unconventional policies of G4 central banks. Given that these central banks in part have recently moved away from the zero lower bound or are in the process of tapering, we include the complete period during which these banks have been jointly in zero lower bound territory and make statements about unconventional policies in general rather than about speciﬁc instruments.

Our findings broadly confirm global spillovers in conventional and unconventional policies. In particular, US monetary policy had significant effects on global markets. In addition, we show that unconventional policy spillovers within short-term interest rates and longer-term government bond

yields typically rise in magnitude vis-`a-vis conventional policy spillovers.

Besides focussing on the period of monetary policy at the zero lower bound, we diverge from EFR in both our choice of economies (adding the UK) and assets (dropping exchange rates). Further diﬀerences also arise from the ﬁner

(5)

details of the empirical set-up, with results often sensitive to the choice of model speciﬁcation, thus requiring a careful discussion of the merits of the model set-up we select.

From a policy perspective, the study of financial linkages is relevant for several reasons. While spillovers from unconventional monetary policy have been documented, spillovers of conventional policies may differ from those of policies pursued at the zero lower bound. This relates to both the magnitude and the relevance of spillovers via third markets. Understanding the monetary policy transmission mechanism at the lower bound is relevant for each monetary authority individually to implement policies in the most effective way. When unconventional monetary policy targets specific markets to ensure their role in the transmission mechanism of monetary policy, spillovers into other asset markets could cause severe disruptions to the real economy. To the extent that spillovers create volatility in other markets and economies, there is a case for international policy coordination.

The remainder of the chapter is organised as follows. Section 4.2 provides an overview of existing research into spillovers of unconventional monetary policy. Section 4.3 discusses the methodological set-up adopted from EFR. Section 4.4 discusses the data, Section 4.5 discusses the results for our model, and Section 4.6 provides robustness checks. Section 4.7 concludes.

4.2 Literature

4.2.1 Unconventional policies in the G4 economies

Because of the severity and uniqueness of the financial and economic crisis experienced on a worldwide level from 2007 onwards, central banks in the G4 countries adopted a variety of measures in response. As defined by the International Monetary Fund (2013), unconventional policy can be seen to cover all policies that go beyond short-term interest rate targets and aim either at providing monetary easing beyond the lower interest rate bound or at restoring the functionality of specific asset markets and the transmission mechanism of monetary policy. With each central bank driven by different domestic conditions and in some cases different mandates and regulatory frameworks, we provide a brief overview of the measures undertaken (see

(6)

Blinder et al., 2017, for a detailed discussion of global changes in monetary policy and the (changed) perception of its role due to the crisis). Kuttner (2018) and Dell’Ariccia et al. (2018) provide a detailed dicussion of measures and channels for the US and the remaining three economies, respectively. Nonetheless, it is possible to broadly categorise measures into: i) interest rates being reduced to their lower bound, ii) asset purchases, and iii) forward guidance. A large majority of policy measures in all three areas were explicitly targeted at ﬁxing credit markets and reducing medium to long-term rates to stimulate real activity (Bauer and Neely, 2014) and bringing persistently low inﬂation rates back to their target.

In the United States, the Federal Reserve responded aggressively to liquidity problems of financial institutions. Between September 2007 and December 2008, the Fed lowered the interest rate from 5.25% to almost zero. To restore liquidity, it acted as the lender of last resort to financial institutions. Three rounds of quantitative easing (QE), QE1 from 2008 to 2010, QE2 from 2010 to 2011 and QE3 from 2012, included the purchase of bank debt (QE1), mortgage-backed securities (QE1, QE3), and Treasury securities (QE1, QE2). The Federal Reserve began tapering these policies in 2013, with purchases ending in 2014. The first federal funds rate hikes took place in late 2015 and 2016.

The European Central Bank included in its measures the switch to fixed rate tenders with full allotment, the expansion of collateral accepted in refinancing operations, the maturity extension for refinancing operations, as well as the activation of currency swap lines (see Hartmann and Smets, 2018, for an extensive discussion). In addition, a range of (expanded) asset purchasing programmes was launched between 2009 and 2016. Throughout this time, key interest rates on marginal refinancing operations and the standing facilities were lowered to their effective lower bound. At the outbreak of the financial crises, the measures broadly targeted the supply of liquidity to the interbank market. Economies troubled by the sovereign debt crisis have been supported with policies as well, notably by the Outright Monetary Transaction (OMT), the implementation of Mario Draghi’s famous ‘whatever it takes’ speech. Ever since the aim has been to keep inflation in line with the

(7)

target as well as encouraging lending to ﬁrms and households and to restore the monetary policy transmission mechanism.

The Bank of England introduced quantitative easing in March 2009, at which point the bank rate had reached 0.5%. The assets were predominantly government bonds purchased from the private sector and, to a small extent, corporate bonds. The initial volume of 165bn GBP in September 2009 rose to 375bn in July 2012 through a series of increases. A further 70bn in August 2016 were added following uncertainty arising from Brexit. Aside from an additional rate drop to 0.25% in 2016 in reaction to Brexit, the policy rate was kept unchanged until the ﬁrst hike in November 2017.

Facing deflationary pressures, the Bank of Japan had already adopted a policy of Quantitative Easing from 2001 to 2006. The key interest rate has not been above 0.5% at any point since the late 1990s. This was followed by the comprehensive monetary easing (CE) programme in 2010 due to the global financial crisis. The asset purchasing commitments were twice raised in 2011. In 2013, this was followed with the Qualitative and Quantitative Easing (QQE) programme, an asset purchase programme that entailed doubling the monetary base to avoid deflation and target an inflation rate of 2%. This bond buying programme was further expanded in 2014. In 2016, the QQE policy was complemented by a policy of yield curve control and inflation rate overshooting. Yield curve control aims at targeting short and long-term interest rates through market operations to lower real interest rates that support economic activity and prices. This commitment to inflation rate overshooting means that the expansion of the monetary base continues until the year on year inflation exceeds the 2 percent target and stays above it in a stable manner.

In sum, all central banks had interest rates at the lower bound and implemented asset purchase programmes. A further common feature is the use of forward guidance to affect medium- and long-term expectations about the stance of the central bank. These measures all constitute unconventional policies at the effective lower bound. The US economy left the effective lower bound first, with the Bank of England following soon after, while the European Central Bank and Bank of Japan are currently still operating within the confinements of this policy boundary.

(8)

4.2.2 Transmission

There are multiple channels through which monetary policy can affect domestic and international financial markets. While assets with high liquidity and low credit risk, such as US treasuries or German bunds, will for example be affected primarily through signalling and portfolio balance effects, other markets will be affected also by the credit channel, liquidity channel or potentially other channels. Although the channels cannot be separated by our methodological framework, we include a brief discussion of these channels.

Generally, the effect of asset purchases is to push up the asset price and lower the yield due to the significant additional demand. Those selling assets to the central bank will invest the money received in higher yielding assets instead. This can be corporate bonds, equities, treasuries, or currencies. In turn, their price and yield will be affected as well. In the case of corporate bonds or shares, the firm is provided with cheaper financing which encourages spending and investment. In the case of banks, they can finance new loans which encourages spending and investment to generate real effects.

The (international) signalling channel operates through the expectations of market participants and private agents more generally. How agents anticipate the economy to be in the future (based on the central banks own assessment, sometimes referred to as the confidence channel) and how committed they expect the monetary authority to be to the monetary policy target or a given path of the policy rate will affect their economic decisions. This is similar for both domestic and foreign markets. Especially in the case of core economies (like the countries in our study) this can be assumed to matter internationally too. Foreign market investors’ reactions will depend also on whether the local central bank is deemed likely to follow the core country central bank with a similar policy, as the regular domestic transmission channels will then further drive up prices (Bernhard and Ebner, 2017). When a policy statement signals a lower and flatter path of the policy rate, government bond yields should fall domestically and internationally. When the communication suggests a worse economic outlook, the outcome is the same since the term premium

(9)

will be narrowed.1 _{Equity prices will be negatively aﬀected by a negative}

economic outlook through expected earnings. In the case of exchange rates, in particular safe haven currencies will experience appreciation pressure.

Whenever purchases by a central bank lead to a rebalancing of investors’ portfolios, this is referred to as the (international) portfolio rebalancing channel (Krishnamurthy and Vissing-Jorgensen, 2011). Because different assets are not perfect substitutes for one another, for reasons of liquidity, risk, maturity, preferred habitat, returns and the like, a significant decline in the availability of a particular asset (due to central bank purchases) pushes their price up and yield down. This leads investors to switch to alternative assets with sufficient similarity. This can, however, include amongst other things riskier assets, depending on the characteristics valued by the investor. Generally speaking, since different assets are not perfect substitutes, the characteristics of alternative assets invested in will differ and likely reach issuers of securities not directly targeted by the asset purchasing programme. As a result, substitution for an asset in one country can lead to a yield decline also in another country as well as a yield decline for alternative domestic assets, in other words international and domestic portfolio rebalancing.

Central bank purchases can also be transmitted through the (international) liquidity channel. Quantitative easing brings down the liquidity premium, the premium investors demanded for an asset that is traded in relatively small volumes, i.e. in a less liquid market. An investor selling assets to the central bank will invest his new liquidity elsewhere in a proﬁtable way. Not only will prices of other assets be pushed up, but also the expected risk-free rate will be lowered. Increased investment in riskier assets will bring down risk premia, raising all assets prices. This reaction is not conﬁned to national markets, although currency risk remains relevant for investors (Bernhard and Ebner, 2017).

An additional potential channel of unconventional monetary policy is the (international) bank lending (credit) channel (Bernanke and Blinder, 1988, 1992). In the case of unconventional monetary policy, the expansion of the central banks’ balance sheets leads to domestic interest rate decreases and an

1_{As Bernhard and Ebner (2017) mention, corporate bonds and equities can also be}

aﬀected in an oﬀsetting way by a worse economic outlook, since the default risk and corresponding premium may rise.

(10)

increase in money supply. Both of these factors encourage banks to increase lending, domestically and across borders. The cross-border (international) lending means that foreign banks experience an increase in the supply of funds, which in turn can lead to more lending in the foreign country, as in

the traditional bank lending channel (Gr¨ab and ˙Zochowski, 2017).

Monetary policy measures can also affect asset prices through an exchange rate channel, since monetary expansion should lead to a depreciation of the currency (Bernhard and Ebner, 2017). Internationally, the correspondingly appreciated exchange rate in foreign countries implies more restrictive monetary conditions that lead to lower rates on short-term government (and corporate) bonds. In the case of equities, the exchange rate effect changes competitiveness of firms with international business dealings. For domestic equities, the higher competitiveness should raise share prices, while international equities could experience a fall in share prices due to lower competitiveness. Changing exchange rates can also affect all assets in general in a potentially offsetting way. A depreciation in the UMP-implementing country makes international assets more expensive than domestic ones, triggering relatively lower international prices and higher yields. If the domestic depreciation is expected to continue in the future, then international assets become a source of valuation gains from appreciation, thus raising their price and lowering their yield (Bernhard and Ebner, 2017).

The disentanglement of different channels and their relative importance is methodologically complicated and requires tailoring a specific model to a very specific policy and asset being analysed (Bernhard and Ebner, 2017). This is not the interest of our research. Instead we are interested in assessing whether there are fundamental differences in the linkages between asset markets in different monetary policy environments. Within the period of unconventional policies, we are not interested in differentiating between different unconventional measures, or their announcements and implementations (nor are we able to do so). Instead we wish to make statements on the general linkages between markets. The discussion of transmission channels in this context is relevant for providing potential qualitative explanations to differences between the periods of conventional and unconventional policy.

(11)

4.2.3 Identiﬁcation

Central to any study of monetary policy effects on asset markets is the identification of shocks to monetary policy. The issue of identification is complicated primarily by the simultaneity of effects from monetary policy to asset prices and vice versa. Each direction of effect merits its own research and displays statistical relevance (Rigobon and Sack, 2003, 2004). In addition to the endogeneity occurring from simultaneity, omitted variable bias and measurement problems come into play. Additionally, the study of spillovers, especially within markets and across countries, is a challenge due to the strong co-movement of prices.

The most commonly used solution to this identification problem has been the event study methodology (Kuttner, 2001; Bernanke and Kuttner, 2005). In it the identification comes from making the measurement window around the change in the policy instrument very small. The policy instrument is assumed to only experience surprise changes on event days, e.g. when the monetary policy committee meets, or when other major speeches or announcements are made. The underlying assumption is that innovations to the system of simultaneous equations are driven primarily by the policy shock (and neither a shock to the asset price itself nor other omitted variable shocks, since the window has been defined sufficiently narrowly). The narrow measurement window reflects the notion that news is very rapidly incorporated into asset prices. Importantly, such news refers only to the surprise element of the monetary policy announced.

A less restrictive alternative methodology uses the heteroskedasticity com-monly found in financial markets. Developed by Sentana and Fiorentini (2001) and Rigobon (2003), and first applied in the context of monetary policy by Rigobon and Sack (2004), identification through heteroskedasticity uses the stylised fact of time-varying structural shocks to form multiple regimes with differing levels of volatility. This allows for a greater number of moments to estimate the unknowns in the system of simultaneous equations. The identifying assumption is one of only higher relative variance of monetary policy on relevant dates, rather than infinitely higher variance required in event studies. This methodology is discussed in greater detail in Section 4.3.

(12)

De Santis and Zimic (2018) introduce (relative) magnitude restrictions based on the assumption that the domestic instantaneous effect of a shock is, on average, larger in absolute value than its spillover effect. They argue in favour of their approach due to its ability to generate different coefficients of the contemporaneous relationship across volatility regimes. In addition, the imposition of zero and sign restrictions necessary when using heteroskedasticity for identification is argued to be complicated by opposite forces such as flight-to-safety, flight-to-liquidity and fire sales. As an example, they provide the alternative sign implied by flight-to-safety or flight-to-liquidity vs. fire sales in the market for yields. Importantly, however, their approach only looks at the yields market, assuming that monetary policy rates do not react contemporaneously to sovereign yield shocks.

4.2.4 Contribution to literature

Frameworks for studying spillovers

From a theoretical point of view, our paper relates to three broad directions of research. The first broadly describes the development of empirical frameworks to jointly study across-market and cross-country spillovers. The literature often looks at linkages across asset classes within a currency area, which are generally well understood. This includes the interaction between short rates and equities in both directions of causality, but also significant correlations between equities and bond yields. The literature on cross-border financial market linkages (and monetary policy spillovers in particular) has often discussed individual asset prices in isolation (Bayoumi and Bui, 2012; Chinn and Frankel, 2003; Dees et al., 2007). Only in the last few years have studies focussed on frameworks combining multiple securities and multiple currency areas. Given the level of financial integration and evidence for monetary spillovers in the existing literature, analysis within a framework of multiple markets and countries is of importance. This is emphasized also by Georgiadis (2017), who shows that multilateral models are more accurate than bilateral models in indentifying spillovers, especially when higher-order spillovers and spillbacks become important and direct bilateral transmission channels are less important. Following on from the first applications of heteroskedasticity to the identification of monetary policy and asset price shocks with one country

(13)

and two securities by Rigobon and Sack (2003, 2004), the first generalised model with two countries and multiple securities using heteroskedasticity for the identification of monetary spillovers comes from Craine and Martin (2008). They find evidence for monetary surprise spillovers from the US to Australian yields and equity returns. The aim of Hayo and Niehof (2013) is to provide a formal proof of the heteroskedasticity identification methodology generalised to the multi-country and multi-market case. They document significant international spillovers among major central banks, with the Federal Reserve having the largest impact in their sample (encompassing the G4 central banks and the Bank of Canada), but these authors do not differentiate between the early years of non-standard monetary policy and the preceding conventional policy period. Their different heteroskedastic regimes are policy days and non-policy days. EFR present another extension of the traditional one country simultaneous equation set-up to show economically and statistically significant international within-market and cross-market spillovers between the US and the euro area in a VAR model, that is guided by data-driven regimes. It is this set-up by EFR that we take to the data from the unconventional policies period due to its ability to comprehensively include the currency areas of interest to us for measuring spillovers in all directions. This identification finds application also in studies on contagion within bond markets (Bayoumi and Bui, 2012; Ehrmann and Fratzscher, 2017), equity markets (Bayoumi and Bui, 2012), corporate bonds (Arai, 2017; Raskin, 2013), across different interest rates within the US (Wright, 2012), and from monetary policy to stock markets (Haitsma et al., 2016) but not to cross-country and cross-market linkages jointly. Other studies have used global VARs to show the interdependence of global asset markets (Beirne and Gieck, 2014), or generalized forecast error variance decompositions for identification (Belke and Dubova, 2018; Louzis, 2013).

Domestic UMP eﬀects

A second related strand of literature provides evidence for unconventional policies having domestic financial market effects like those of conventional monetary policy periods. The most notable difference between the two periods lies in the effect on bond yields (for an overview of estimates for the bond

(14)

market see surveys by Dell’Ariccia et al., 2018; Kuttner, 2018). Unconventional policies primarily lower the long end of the yield curve, due to forward guidance and shorter-term yields already being close to zero. As the most influential central bank on a global level, domestic effects of the Federal Reserve’s unconventional policies have received most attention, with the main emphasis on the significant and broadly found yield reduction in the market for government and corporate bonds (e.g. Wright, 2012). The negative effect on US long-term bond yields found to arise from expansionary unconventional monetary policy is robust to the methodology and confirmed in event studies (Gagnon et al., 2011; Krishnamurthy and Vissing-Jorgensen, 2011) and term structure models (Bauer and Neely, 2014). Exchange rates and equities are both significantly affected (Rogers et al., 2014), with the home currency depreciating and equity markets strengthening as a result of monetary policy easing. Needless to say, the impact of policies by the ECB (e.g. Falagiarda and Reitz, 2015), the Bank of England (e.g. Joyce et al., 2011) and the Bank of Japan (Hosono and Isobe, 2014) has been analysed as well, with a similar focus and broadly similar conclusions. The Bank of England (BOE) and ECB policies show a similarly significant domestic effect. Where included in the analysis, domestic corporate bond yields were also lowered by monetary policy (Chen et al., 2016; Rogers et al., 2014).

UMP spillovers

Finally, our research relates to studies on the international dimensions of

non-standard monetary policy.2 _{In some cases, the literature has built upon the}

methods used in studies on spillovers of conventional monetary policy, although an additional issue that arises is measurement of the monetary policy stance due to the zero lower bound. As a consequence, the event study methodology has been predominantly used. Often, the monetary policy surprise has also been proxied with bond yield changes at a higher maturity. In particular policies such as the Outright Monetary Transaction (OMT) programme of the ECB and forward guidance in general only allow for the study of announcement eﬀects. Importantly, both announcements and operations are,

2_{For a discussion of spillovers in conventional policy times see the literature review in}

(15)

where measurable, highly interesting and potentially economically relevant since they operate through multiple channels. The sample used in most papers has been with respect to both other major economies with unconventional policies (as in Rogers et al., 2014) or emerging markets experiencing eﬀects from major foreign central banks (Bowman et al., 2015; Falagiarda et al., 2015).

Most studies focus on the Federal Reserve as the most relevant policymaker, zooming in on the effects of its unconventional policies on long-term bond yields. This is shown in particular for emerging markets (Bowman et al., 2015; Fic, 2013; Fratzscher et al., 2017). Due to the short end of the yield curve approaching the lower bound, the effects are found more strongly for the long end of the curve (Chen et al., 2016; Gilchrist et al., 2016). Long-term asset purchasing programmes, when taking into account expectations to extract surprises, led to global interest rate declines as well as currency depreciations (Glick and Leduc, 2012; Neely, 2015), with only insignificant effects on global equity markets (Glick and Leduc, 2012). This finding is echoed for yields and exchange rates in 40 emerging and advanced economies but contrasts with the significant impact on stocks found in Gagnon et al. (2017). There exists also significant evidence for changes in portfolio flows (Fratzscher et al., 2017; Lim and Mohapatra, 2016). The dominance of US bond and equity markets in an international comparison of spillovers is obtained in multiple studies (Bayoumi and Bui, 2012; Belke and Dubova, 2018), with richer settings hinting at international spillovers across assets as well (Belke and Dubova, 2018). Therefore, there is a broad consensus that the Federal Reserve’s unconventional policies affected not only the US, but also global markets.

Unconventional measures used by the European Central Bank have also been studied in detail, most notably in event study settings (Fratzscher

et al., 2016; Georgiadis and Gr¨ab, 2016). The Expanded Asset Purchasing

Programme announcement depreciated the euro, raised equity prices globally and shifted portfolio ﬂows away from emerging markets to advanced economies. Fratzscher et al. (2016) use deviations of Securities Market Program (SMP) purchases from a reaction function (proxying the markets expectation of weekly purchases) to show that the SMP transactions (and announcements)

(16)

boosted global equities and depreciated the Euro. While in emerging countries bond markets were not affected, advanced economies saw yield rises consistent with safe haven flows unwinding. The OMT announcement had similar effects, except for a slight appreciation of the Euro. The implementation of Long-Term Refinancing Operations (both Supplementary and Very Long-Long-Term) boosted equity markets and lowered yields (Fratzscher et al., 2016). Rogers et al. (2014) carry out a broad comparison across the four main central banks which includes all measures (including minor changes such as allotment rules). Along with Fratzscher et al. (2016), their findings see a far greater role for Federal Reserve policies in a global setting than for the policies of the ECB. Besides the inclusion in studies by Glick and Leduc (2012) and Rogers et al. (2014), the role of the Bank of England and the Bank of Japan has largely been neglected in the context of global spillovers. This may be due to an assumption that the Federal Reserve and the European Central Bank are the only internationally relevant policymakers. Consequently, the effects of policies of the BoE and the BoJ have not been researched much beyond their comparison to Federal Reserve and ECB policies (Chen et al., 2016; Rogers et al., 2014). Claus et al. (2016) conduct a two country analysis with the US and Japan, finding significant spillovers in both directions, but changing transmission compared to conventional policy periods. The study of Rogers et al. (2014) is an event study using the change in yields on bond futures around narrow time windows to proxy for policy easing/tightening in the

absence of short-term interest rate variability.3 Using this pass-through to

bond yields to estimate effects on other markets, they show that the impact of the Federal Reserve dominates. This is the case for all markets (bonds, stocks, and currencies), across and within economies, and generally applicable at the daily and intraday level. However, both policies of the BoE and the ECB also have some spillovers to foreign bond markets and exchange rates. Specifically, the ECB policies affect all currencies (against the dollar), but otherwise only moves UK bond yields with expansionary policies surprisingly raising British yields. Expansionary policies by the Bank of England lower US, Japanese and European yields, while their exchange rate spillovers are weaker. Bank

3_{For the case of the ECB, this is approximated by the change in spread between German}

and Italian yields. For the Federal Reserve, the ﬁrst principal component of 2 year, 5 year, 10 year and 30 year futures is used.

(17)

of Japan policies do not display any spillover eﬀects at all. The asymmetry of US policies spilling over to other countries, but not or weaker vice versa,

has been documented in studies on conventional policy periods (G¨urkaynak

et al., 2005). Beyond the traditional explanation of greater relevance of the US for the world economy, Rogers et al. (2014) link the asymmetry in spillovers of unconventional policies to the diﬀering nature of unconventional policy programs. Asset purchasing programmes (and long term reﬁnancing operation announcements for the ECB) generally seem to have greater pass-through than other types of measures.

A further subset of the literature looks at the potentially time-varying nature of asset linkages. Wu (2016) ﬁnds that while US corporate bonds are highly responsive to domestic monetary policy, measured as passthrough to Treasury yields, throughout, stock markets are less responsive to monetary policy between 2008 and 2010. Short-term government bonds are less sensitive at the lower bound, while long-term government bonds are more sensitive to monetary policy at the lower bound. The changes in sensitivity observed for bond markets occur sharply in 2008. Bond markets have been found to stay highly connected through time, however, with a decline between 2008 and 2012, a period when greater shocks were experienced in Europe but markets were more fragmented (De Santis and Zimic, 2018). Belke and Dubova (2018) show that linkages across global markets do not remain constant over time and that global asset markets indeed become more integrated. The sharpest change identiﬁed in their rolling estimations is at the onset of unconventional policies.

4.3 Methodology

According to the efficient market hypothesis, asset prices should reflect all expectations about the future, such that announcement days of new policies – when the information provided to the market changes and expectations are updated – are most relevant for the study of asset prices. This theory is underpinned by the large literature on the price discovery process in different asset prices related to news and fundamentals, as well as monetary policy announcements (see amongst others Andersen et al., 2007). For the analysis

(18)

of announcement effects, most of the literature on unconventional monetary policy effects follows Bernanke and Kuttner (2005) in using event studies to analyse the effects of monetary policy shocks. The weakness associated with the event study approach is the assumptions it relies on. Unconventional policies are often discussed in the media and thus anticipated to a certain extent beforehand. By setting a typically narrow time window around an announcement within which the change in asset prices is measured, an event study tries to avoid the influence of other macroeconomic news affecting the measurement of the true effect of monetary policy on asset prices, and to circumvent the issue of potential endogeneity arising from asset prices to monetary policy. For this, the event study approach assumes that the policy variable is the only variable that is shocked, or more specifically, that the variance of the policy shock is infinitely greater than the asset price shock to remove the bias from estimating it with ordinary least squares (OLS). As a result, the estimated coefficient is as close to its real value as possible when we assume that the monetary policy shock becomes infinitely large in the limit relative to the variance of all other shocks. This makes OLS estimates consistent and means that the bias stemming from any correlation between asset price shocks and the monetary policy measure goes to zero. As Rogers et al. (2014) argue, monetary policy has become more complicated since the financial crisis, with markets possibly needing longer to interpret more conditional and lengthier statements by policymakers. If the window of measurement is extended, the risk of other contaminating events rises.

Identification through heteroskedasticity, in contrast, only requires greater (as opposed to infinite) volatility of the monetary policy variable on days when policy changes are announced. This is particularly relevant for unconventional policies as the surprise content of policies has varied more than in conventional times. Identification based on heteroskedasticity only requires the policy shock variance to rise and the variance of the remaining shocks to remain constant. The main assumption is that all other shocks to our structural-form model must be uncorrelated. That structural parameters are stable across different regimes is also assumed in an event study approach. As such the methodology allows for a data-driven identification with weaker assumptions than those required for an event study.

(19)

Figure 4.1: Methodological approach

Note: This ﬁgure outlines the sequential approach to selecting an appropriate set-up

in which to model spillovers within financial markets. This begins with choices about economies and asset types included and also includes choices relating to sample windows, details on the VAR model, selection of regimes for identification, and restrictions placed on the estimation. Our choices are in bold, model specifications that are different in EFR are underlined. CMP refers to the conventional monetary policy period; UMP refers to the unconventional monetary policy period.

4.3.1 Vector Autoregressive model

As Rigobon and Sack (2004, p. 1560, footnote 8) mention, their identiﬁcation approach, when expressed in matrix notation, can also be expressed as a VAR including all asset prices and then focussing on reduced-form residuals instead of the changes in asset prices. This means we use the covariance matrices of reduced-form residuals, rather than the covariance matrix of the endogenous variables themselves. EFR expand this set-up to spillovers across currency areas (adding the euro area alongside the United States in their study) and markets (adding bond yields and the exchange rate to short rates

(20)

and stocks). We rely on their approach to analyse the big central banks, and

brieﬂy summarise the approach below.4Figure 4.1 highlights the steps involved

in building the model we use, beginning with economies and assets included, the samples, and details of the VAR model, regime selection and restrictions placed on the model. The econometric set-up builds on a structural-form VAR model

Ayt= θ + Π(L)y_t−1+ Ψ(L)zt+ ut, (4.1)

where yt is a vector of the endogenous variables, Π(L) represents the lagged

eﬀects of these variables, Ψ(L) is the eﬀect of exogenous variables, and

ut ∼ (0, Σ) has the diagonal Σ. zt controls for common shocks, which are

explained in more detail below.

In the benchmark model of EFR, the vector of endogenous variables

contains 7 asset prices: short-term rates rt, long-term government bond yields

bt, stock market returns stand exchange rate returns etfor two currency areas.

A thus contains the parameters of interest, the contemporaneous spillovers, in the oﬀ-diagonal elements of the 7 x 7 matrix

Ayt= ⎛ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎝ 1 α12 α13 β14 β15 β16 γ17 α21 1 α23 β24 β25 β26 γ27 α31 α32 1 β34 β35 β36 γ37 β41 β42 β43 1 α45 α46 γ47 β51 β52 β53 α54 1 α56 γ57 β61 β62 β63 α64 α65 1 γ67 γ71 γ72 γ73 γ74 γ75 γ76 1 ⎞ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎠ ⎛ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎝ rUSt bUS_t sUSt r_tEA bEAt sEAt et ⎞ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎠ ,

where α denotes domestic spillovers across diﬀerent markets (within a currency area), β denotes cross-country spillovers (within and across assets) and γ denotes exchange rate spillovers. Since matrix A is not diagonal, we cannot

4_{Bayoumi and Bui (2012) conduct an analysis on within-market spillovers for the G4}

economies for both long term bonds and equities. The econometric set-up is very similar to EFR, albeit without controls, but analyses the two asset classes separately, motivated by the ﬁnding of EFR that inter-market eﬀects are strongest.

(21)

estimate the true parameters of the structural model with standard methods.

Instead, assuming that A−1 exists, the reduced-form model is estimated with

OLS.

yt= A−1θ + A−1Π(L)y_t−1+ A−1Ψ(L)zt+ A−1ut, (4.2)

yt= C0+ B0(L)y_t−1+ B1(L)zt+ t. (4.3)

where t ∼ (0, Ω). It can easily be seen that the residuals of the reduced

form t are related to those of the structural form utthrough matrix A. The

structural coeﬃcients in A show the direct linkages between markets, while

the reduced-form model (and thus matrix A−1) captures both direct and

indirect links between markets (including those that occur via other asset

prices, i.e. third markets). A−1 therefore, shows overall spillovers. As EFR and

De Santis and Zimic (2018) point out, the approach requires the imposition of additional restrictions on the parameter matrix based on economic intuition

to guarantee the right rotation of matrix A.5 _{Matrix A has (N} _{∗ N) − N}

oﬀ-diagonal unknowns, unless further restrictions are imposed. In addition, there are N unknowns from the covariance matrix of structural form residuals Σ.

In total, this makes N2 _{unknowns. Each sample covariance matrix generates}

(N∗N +N)/2 moments. Two regimes give two sample covariance matrices and

thus twice as many moments (N2_{+ N ). This only adds a further N unknowns}

(the diagonal elements in the regime specific structural shock covariance matrix). Thus, two regimes are already sufficient for identification.

The methodological approach is to estimate the reduced-form VAR with

suﬃcient lags to remove serial correlation from the residuals.6 _The

reduced-form residuals thus contain only contemporaneous eﬀects and can be used to deﬁne heteroskedastic regimes. For each individual asset, a high variance regime is based on either 10, 15 or 20 two-day rolling windows of residuals. We use a regime for estimation if more than 15 observations in the sample

5_{In such systems of simultaneous equations, there are multiple}_{A matrices that produce}

the same pattern of second moments, i.e. the exact same covariance matrices, making them empirically indistinguishable. Since the A matrices themselves differ, the intuitive implications from the coefficients are totally different. Similar to the demand and supply example provided in Rigobon (2003), the idea of imposing sign restrictions is to ensure the right rotation, without imposing any restriction on the coefficient size.

6_{We estimate the model in GAUSS and are grateful to Roberto Rigobon and coauthors}

(22)

exhibit high relative variance for residuals of that asset,7 _{more speciﬁcally}

when the relative variance of one or several assets residuals exceeds the full sample mean plus one standard deviation. We use two-day returns due to the diﬀerent time zones of the diﬀerent currency areas and non-overlapping hours of market operation.

The sample covariance matrix for each regime is estimated for the periods speciﬁed in each regime. Using a subset consisting of only the single-asset high variance (i.e. conditional volatility) regimes and the regime with low variance for all assets, the parameters of interest are estimated in a GMM procedure. For proper identiﬁcation, we need the equations to be linearly independent, which is why only regimes with a single asset in high variance (where the variance of all other assets remains stable) are used. The following minimum distance function (which works like a loss function) is minimised for each regime i

min gg with g = AΣiA− Ωi,

where the covariance matrix of structural shocks Σi to be estimated for each

regime is diagonal (because structural shocks are by assumption uncorrelated

with each other), A includes restrictions and Ωi is the (sample)

variance-covariance matrix of the reduced-form residuals estimated in each regime i.

Matrix A is assumed to be identical across regimes, while Σi and Ωiare regime

speciﬁc.8_{The minimum distance function is a simple extension of the fact that}

reduced-form residuals and structural shocks are linked via t= A−1ut. For

the signiﬁcance of parameter estimates, we use block-bootstrapping with 200 repetitions. Block bootstrapping, unlike simple case resampling, deals with correlation in the data or errors (i.e. heteroskedasticity) and can thus replicate the correlation in the data, while simple resampling cannot. Instead, blocks of data are resampled (one regime is one block). For each of the regimes, the estimated regime-speciﬁc (sample) covariance matrices are used to create new data with the same covariance structure in each bootstrap replication. For

7_{We test a variety of rolling windows to ensure that the high variance regime for each}

respective variable indeed yields a high variance for that variable relative to other regimes. The same testing is applied with regard to the limit of one standard deviation – in some cases higher limits of up to two standard deviations provide a clearer distinction between a variable’s variance across regimes.

(23)

each draw, coeﬃcients are estimated by again minimising the moments with the given restrictions (Ehrmann et al., 2011). As in EFR, we adopt a variance decomposition including endogenous variables and shocks, as well as lags and exogenous variables. This takes the following form:

var(yt)=[I− A−1Π(L)][Ψ(L)var(zt)ΨT(L)+var(ut)][I− A−1Π(L)]T.

4.3.2 Restrictions

We adopt sign restrictions to the parameter matrices of the structural and reduced form identical to those in EFR, which are motivated by existing literature and increase the speed of convergence for the estimation on the one hand and ensure the correct rotation of matrices. The restrictions matrices of the 3 market-3 country set-up are also found in Figures 4.2 and 4.3. In line

with EFR, the restrictions are placed on matrices A and A−1, rather than on

impulse response functions (as in Faust, 1998) which we are not interested in here. Whenever we find that estimation results yield parameters hitting the boundaries, we are able to revise these restrictions, also to allow for different effects between conventional and unconventional monetary policy periods. This, however, is generally not necessary. The results with instances of binding restrictions are not fundamentally changed by dropping the restrictions. The restrictions we impose stem from the interpretation given to the money market rates as an indicator of market expectations about monetary policy in the short to medium term, long-term government bond yields as inflation expectations, stock indices as explained by domestic demand or supply (productivity) shocks, and exchange rate changes as reflecting changes in relative demand.

In domestic terms, there are three main restrictions on the matrix of direct spillovers A taken from EFR. Firstly, if a rise in bond yields reﬂects an inﬂationary shock, then the market will expect a monetary tightening and short rates will rise (α12, α45, α78). Secondly, if short rates rise due to monetary policy, then the discount value rises, thereby lowering the demand for goods and services, such that equity prices decline (α31, α64, α97). Thirdly, higher longer-term bond yields lower equity prices (α32, α65, α98). These

(24)

In international terms, there are three further assumptions relating to matrix A. Firstly, this includes assuming there are no direct spillover effects across borders from one market of one country to a different market of a different country. These are represented as zeros in the A matrix. For example, there is no direct spillover from US bond markets to EA stock markets; these only occur indirectly via within-country or within-market spillovers (here: US stock market and EA bond market). Secondly, spillovers between one bond market and another, and one short rate market and another, should be positive, based on the financial openness of markets that leads to arbitrage and interest rate transmission (β14, β17, β25, β28, β47, β52, β58, β71, β74, β82, β85). Thirdly, spillovers within market and across countries should not be amplified (all non-zero β parameters).

On the reduced-form matrix, we impose the restriction that overall

spill-overs from bond to stock markets are negative (ρ32, ρ65, ρ98).9 Overall

spill-overs from bonds to money markets are positive (ρ12, ρ45, ρ78), while overall money market spillovers to stock markets are also negative (ρ31, ρ64, ρ97). Regarding exchange rates, long (government bond) rates are assumed to lead to a portfolio shift into assets of that country, and thus appreciate the currency.

While these restrictions are motivated by established findings in the liter-ature, they refer to conventional monetary policy periods. In unconventional policy times, the implications of monetary policy transmission times may point to other restrictions. Following our earlier discussion in Section 4.2, we let the transmission channels and results debated in the literature guide our restrictions regarding the effect of monetary policy. In conventional times, the short rate (on the interbank market) reflects the monetary policy stance, while in unconventional times, when the monetary policy stance is also reflected in additional policy instruments, this could fall to a shadow short rate. As argued earlier with respect to asset purchasing programmes, the long-term

9_{Since the form parameters are retrieved as the inverse of matrix A, the}

reduced-form matrix is dependent on the restrictions in the structural reduced-form matrix. In addition, the restrictions on the reduced-form matrix are not actually binding in the same way they are for structural form parameters. Instead, the empirical set-up adds penalties to the minimum

distance minimisation. It does so when a parameter inA−1 lies beyond the boundary and

thereby increases the distance, leading the estimation to minimise the distance as much as possible. As a result, reduced-form parameters can still exceed the boundary in some cases.

(25)

government bond yield is the variable most notably aﬀected by monetary policy easing at the lower bound. Running a model with the regular money market rate therefore does not make sense in unconventional times. Long-term bond yields on the other hand may accurately represent the monetary policy stance. While shadow rates go a long way in dealing with the lack of inform-ation contained in the short rate at the lower bound, a model with shadow short-term rates and the long-term government bond yields in unconventional policy times could suﬀer from potential collinearity due to the nature of how shadow short rates are constructed based on a term structure model that includes longer-term yields. In this regard, the maturity of the government bond yields is very important. We therefore back up our benchmark model that includes both shadow short rates and bond yields, with models that drop one of these assets.

Based on these considerations and findings in the literature, we implement three restrictions on the effect of short rates in the system of equations. Firstly, since default and liquidity premia should be negligible for the government bonds considered in our sample (where the euro area is proxied for with German bonds), we can expect signalling and portfolio rebalancing to be the main drivers behind bond yields in conventional times. As mentioned, the inclusion of both short and long rates in unconventional times is potentially problematic. Secondly, therefore, monetary policy (i.e. short rates in conventional times; shadow short rates or long rates in unconventional times) easing should boost equities domestically, assuming that in particular signalling is interpreted with regard to the path of the monetary policy rate rather than the economic outlook. Thirdly, exchange rates, where included in our model, should depreciate in a country experiencing significant monetary easing.

A summary of all these initial restrictions is seen in Figures 4.2 and 4.3.

4.4 Data and model

We use daily data on short-term money market rates (3-month maturity, in the case of the US we use the 3-month Treasury bill) and long-term government bond yields (10-year maturity, German rates for the euro area), broad stock

(26)

Figure 4 .2: S tructural-form restrictions ⎛ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎝ 1 α12 < 0 α13 − 1 <β 14 < 00 0 − 1 <β 14 < 00 0 α21 1 α23 0 − 1 <β 25 < 00 0 − 1 <β 25 < 00 α31 > 0 α32 > 01 0 0 − 1 <β 36 < 10 0 − 1 <β 36 < 1 − 1 <β 41 < 00 0 1 α45 < 0 α46 − 1 <β 47 < 00 0 0 − 1 <β 52 < 00 α54 1 α56 0 − 1 <β 58 < 00 00 − 1 <β 63 < 1 α64 > 0 α65 > 01 0 0 − 1 <β 69 < 1 − 1 <β 71 < 00 0 − 1 <β 74 < 00 0 1 α78 < 0 α79 0 − 1 <β 82 < 00 0 − 1 <β 85 < 00 α87 1 α89 00 − 1 <β 93 < 10 0 − 1 <β 96 < 1 α97 > 0 α98 > 01 ⎞ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎠ Figure 4 .3: R educed-form restrictions ⎛ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎜ ⎝ ρ11 > 0 ρ12 > 0 ρ13 0 <ρ 14 < 1 ρ15 ρ16 0 <ρ 17 < 1 ρ18 ρ19 ρ21 ρ22 > 0 ρ23 ρ24 0 <ρ 25 < 1 ρ26 ρ27 0 <ρ 28 < 1 ρ29 ρ31 < 0 ρ32 < 0 ρ33 > 0 ρ34 ρ35 0 <ρ 36 < 1 ρ37 ρ38 0 <ρ 39 < 1 0 <ρ 41 < 1 ρ42 ρ43 ρ44 > 0 ρ45 > 0 ρ46 0 <ρ 47 < 1 ρ48 ρ49 ρ51 0 <ρ 52 < 1 ρ53 ρ54 ρ55 > 0 ρ56 ρ57 0 <ρ 58 < 1 ρ59 ρ61 ρ62 0 <ρ 63 < 1 ρ64 < 0 ρ65 < 0 ρ66 > 0 ρ67 ρ68 0 <ρ 69 < 1 0 <ρ 71 < 1 ρ72 ρ73 0 <ρ 74 < 1 ρ75 ρ76 ρ77 > 0 ρ78 > 0 ρ79 ρ81 0 <ρ 82 < 1 ρ83 ρ84 0 <ρ 85 < 1 ρ86 ρ87 ρ88 > 0 ρ89 ρ91 ρ92 0 <ρ 93 < 1 ρ94 ρ95 0 <ρ 96 < 1 ρ97 < 0 ρ98 < 0 ρ99 > 0 ⎞ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎟ ⎠

(27)

indices (S&P 500, S&P EURO, Nikkei 225, FTSE 100), exchange rates (which are sourced from Bloomberg and Datastream) as well as a set of controls. All returns are expressed in basis points for comparability. Due to the dimensions of matrix A, we estimate diﬀerent models, a benchmark model and perform multiple robustness tests. Our conventional policy period runs from 1995 to

2008.10_{We begin our zero lower bound sample in December 2008, at the time}

when the four major central banks eﬀectively arrived at the zero lower bound and end our sample in May 2016.

Following Andersen et al. (2003, 2007) and EFR, we control for a broad range of macroeconomic shocks, as orthogonality of structural form shocks is a necessary condition for correct identification. We use survey data from Money Market Services (MMS) International, collected from Haver, as this allows for the construction of a news component for macroeconomic variables. It is constructed as the (standardised) difference between the unrevised announcement and the median of survey expectations. In line with EFR, we pay attention to covering a range of indicators relating in particular to the business cycle, and also include oil prices (Bloomberg data) and allow for a common (global) shock. Whenever there are multiple news shocks (controls) within a (two day) window, these are summed. An overview of the included macro shock controls in provided in Table 4.1. For brevity we do not report results for these controls and refer to the literature on announcement effects of macroeconomic news (e.g. Andersen et al., 2003, 2007). There are gaps in the data for the early years in our sample, so that variables were chosen with regard to availability and a broad coverage of macroeconomic relevance.

4.4.1 Model variations

As discussed in EFR, Asian markets in particular are relevant for the analysis of global ﬁnancial linkages, but are hard to include for dimensionality reasons due to the multitude of markets. Additionally, in the context of global monetary policymaking during the crisis, there is a special case to be made for Japan

10_{The choice of how to split the sample into conventional policy times and unconventional}

policy times is naturally somewhat ambiguous. Bayoumi and Bui (2012), for example, end their conventional policy sample in 2009 to obtain stable relationships for conventional policy times. This also applies to the end of unconventional policy times, i.e. when the United States left the lower bound.

(28)

Table 4.1: Macro controls

MMS Name Mnemonic Freq. 1995–₂₀₀₈ 2009–₂₀₁₆

US Nonfarm Private Payroll Employment ADP M 117 96

Consumer Conﬁdence CNCF M 264 96

Consumer Price Index CPM M 264 96

Employees on Nonfarm Payrolls ED M 265 97

Manufacturing Payrolls EMD M 60 80

Unemployment Rate EUR M 264 96

Real GDP, Advance GPAA Q 85 29

Real GDP, Final GPFA Q 87 31

Real GDP, Preliminary GPPA Q 84 29

Housing Starts HST M 260 93

Industrial Production IPM M 262 94

Composite Index of Leading Indicators LIM M 261 93

Producer Price Index PPM M 264 96

Trade Balance, Goods and Services TB M 263 95

JP Consumer Conﬁdence CNCF M 117 75

Coincident Composite Index CNP M 140 82

Unemployment Rate EUR M 146 89

Real GDP, Q/Q %change GPPQ Q 28 11

Producer Price Index IPPM M 117 76

Active Job Opening to Application Ratio JOS M 144 86

Leading Composite Index LIP M 142 84

Money Supply M2, Y/Y %change M2CY M 118 72

CPI, Y/Y %change NCPY M 143 86

Retail Sales, Y/Y %change RSRY M 122 72

Trade Balance, Goods and Services TBGN M 109 69

EU Prelim. Labour Costs, Y/Y %change AEPY M 109 12

Consumer Conﬁdence CNCF M 181 96

Harmonised CPI, M/M %change CPHM M 215 95

Business Condition Index ESI M 213 94

Unemployment Rate EUR Q 52 27

Real GDP, Preliminary GPPY M 176 86

Business Climate Index IBC M 195 86

Money Supply M3, 3M moving average M3MA M 212 93

Producer Price Index PPM M 202 86

Retail Sales, M/M %change RSRM M 189 83

Trade Balance, Goods and Services TBE M 197 86

DE Unemployment Rate EUR M 120 63

UK Prelim. Labour Costs, Y/Y %change AEPY M 254 94

Consumer Credit CNCR M 248 85

Consumer Price Index CPM M 149 93

Unemploym. Claim. Count, M/M change EUD M 259 93

Prel. Money Supply M4, M/M %change M4PM M 196 28

Average House Prices, Y/Y %change PHHM M 126 85

Net Outp. Manuf. Prod., Y/Y %change PPOY M 204 94

Retail Sales, M/M %change RSRM M 262 94

Trade Balance, Goods and Services TBG M 255 87

Note: The table reports the macro control names, their code in the MMS dataset, the

frequency of release (Monthly or Quarterly) and the number of appearances in the two subsamples. All series are available with a speciﬁc release date. We use the unrevised data releases. Unless speciﬁed otherwise, data is in levels.