The effect of quantitative easing on EUR/USD exchange rate

The effect of Quantitative Easing on the

EUR/USD exchange rate.

Master's Thesis, academic year 2016-2017

Faculty of Economics and Business

Universiteit van Amsterdam

Christina Haseth, 10141936

1

Statement of Originality

This document is written by Christina Haseth who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it. The faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

2

Abstract

Since the global financial crisis central banks have taken extraordinary measures to aid the economy’s recovery. One of these measures is known as Quantitative Easing. This thesis analyzes whether Quantitative Easing has affected changes in the EUR/USD exchange rate through its transmission channels. The empirical analysis describes a multiple ordinary least squares including a dummy variable and interaction term. Five independent variables represent each transmission channel and four

independent variables control for other factors effecting changes in the exchange rate. Results suggest that QE has contributed to changes in the exchange rate through its money channel, however the empirical model used does not capture anticipating behavior of agents and therefore the results are a bit uncertain.

3

Table of Contents

Introduction ... 4

1. The importance of exchange rates for economic activity ... 6

2. Understanding Quantitative easing ... 8

2.1 Quantitative easing: an introduction ... 8

2.2 The transmission channels ... 9

2.3 Quantitative easing at the ECB ... 12

2.4 Limits and risks of QE ... 12

3. Literature review ... 15

3.1 Empirical evidence... 15

4. Methodology and dataset ... 17

4.1 The hypothesis ... 17

4.2 Data description ... 18

4.3 The empirical model ... 20

5. Results ... 21

5.1 Checking the data ... 22

5.2 Regression results ... 22

5.3 Robustness check ... 25

5.3.1 Additional interaction term for each control variable ... 25

5.3.2 Portfolio investments instead of capital account ... 26

5.3.3 Adjusted time period for dummy variables... 26

5.4 Comparison of results ... 27

6. Conclusion ... 28

7. References ... 29

4

Introduction

With the initiation of liquidity operations by the European Central Bank (ECB), August 2007 marked the first phase of the global financial crisis (GFC) (Lane, 2012). According to Lane (2012) the GFC entered a more intense phase when Lehman Brothers collapsed in September 2008, affecting both Europe in early 2009 as much as the United States in late 2008. The GFC and its aftermath pose many challenges for conducting monetary policy (Joyce, Miles, Scott and Vayanos, 2012). According to Joyce et al (2012) the challenge for central banks is to aid the economy in its recovery to reach the point where conventional monetary policy and macroprudential tools can jointly achieve price and financial stability. Until then, extraordinary measures must be taken.

These extraordinary measures are known as unconventional monetary policy and may comprise three elements: 1. large scale liquidity support to banks, 2. forward guidance of ultra-low policy rates over extended policy horizons, 3. large-scale financial market interventions, particularly huge asset purchases (Pattipeilohy, Van den End, Tabbae, Frost and De Haan, 2013). These unconventional monetary policy measures were taken as more conventional measures had largely lost their potency (Pattipeilohy et al, 2013). This thesis focuses on the third tool known as quantitative easing (QE).

QE is a policy in which market interest rates are reduced differentially at different maturities, lowering them at the maturities that affect investment and household decisions. The aim is to stimulate spending by increasing broad money holdings, pushing up asset prices, creating wealth, and lowering borrowing costs (Hughes-Hallett, 2017). Consequently, due to these unconventional measures, central bank balance sheets have expanded substantially. In advanced economies, central bank assets now exceed 20% of GDP. Moreover, unconventional monetary policies have led to significant changes in terms of balance sheet composition (Pattipeilohy et al, 2013). Figure one shows the assets of the Eurosystem over time.

The combined assets of national central banks across the euro area has more than quadrupled in less than 15 years and nearly doubled in the last two years. Considering the substantial increase in central banks’ balance sheet it is of great importance to analyze what potential consequences and impact this could have on economic activity and financial markets.

5 Figure 1: Assets1 _{of the Eurosystem national central banks (NCBs) and the ECB held at year-end vis-à-vis}

third parties.

Source: ECB: https://www.ecb.europa.eu/pub/annual/balance/html/index.en.html

An extensive literature has already investigated the impact of traditional interest rate movements on real activity and inflation. However, little is known about the macroeconomic effects and pass-through of non-standard policies and how they differ from conventional interest rate changes. A better understanding of the transmission mechanism and impact on economic activity is therefore essential for policymakers (Peersman, 2014). This thesis contributes to the growing literature on unconventional monetary policy by identifying the transmission channels of the ECB’s QE program and analyzing their impact on the EUR/USD exchange rate. Identifying and correctly quantifying the main transmission channels of QE, may lead to a clearer understanding of the impact QE has on the EUR/USD exchange rate. This may reveal useful information about the central bank credibility and effectiveness on employing QE strategies, and have great policy implications (Kenourgios, Papadamou, Dimitriou, 2015). The impact on the

exchange rate is chosen since a value of a currency is a vital tool to aid economic activity. The research question is therefore constructed as follows: Has QE contributed to the EUR/USD depreciation through its transmission channels?

1 _{The assets encompass from bottom to top: gold and gold receivables, claims on non-euro and euro area residents}

denominated foreign currency, claims on non-euro area residents denominated in euro, lending to euro area credit institutions related to monetary policy operations denominated in euro, other claims on euro area credit institutions denominated in euro, securities of euro area residents denominated in euro, general government debt denominated in euro and other assets.

6 To answer the research question five transmission channels are quantified using monthly data, interaction terms are also created to isolate the additional impact of the QE program on the change in the EUR/USD exchange rate. The data is split in two periods, one period ranging from January 1999 to April 2009 prior to the first QE program was announced and the second period, May 2009 until March 2017, from the first announcement of the implementation of QE until the most recent published data.

The remaining thesis is structured as follows: Chapter one describes the importance of exchange rates for economic activity, followed by a more detailed explanation of QE, the ECB’s QE program and the limits and risks of continuing with QE in chapter two. Chapter three summarizes past literature on the effects of QE, followed by a description of the dataset and methodology used in chapter four. Chapter five presents and explains the results obtained, the thesis ends with a conclusion in chapter six.

Chapter 1: The importance of exchange rates for economic activity

““Parity” was the word of the week in foreign exchange markets as the euro slid sharply against the dollar” wrote the Financial times in March 2015. The question was not if but when the euro would drop below parity. The last time the euro fell below 1:1 was in 2000, a year after its creation (Financial times, 2015). Nonetheless, if the depreciation does not become precipitous and provoke widespread financial market volatility, there is little reason to be concerned. On the contrary, it suggests the ECB has

credibility in pursuing its stated goal of raising growth and preventing deflation (Financial times, 2015). Recently, the euro has been improving by strengthening versus the dollar and pound as investors become relatively less optimistic about the US and UK economies (Financial Times, 2017). This section therefore explains how exchange rates can contribute to economic growth and shows why a depreciation is

expected to boost economic activity.

There is significant divide between policy-makers and economists regarding the impact of foreign exchange policies on growth. Laymen and politicians are often intimately convinced that a lower exchange rate spurs growth. Whereas economists, are generally skeptical that the relative price of two currencies may be a fundamental driver of growth over the long-run. They argue that the contribution to growth is difficult to disentangle (Habib, Mileva and Stracca, 2016). The most intuitive relationship between exchange rates and economic activity is through its effect on the demand for exports and imports. A depreciation of the domestic currency makes exports relatively cheaper and therefore more competitive abroad, on the contrary imports become more expensive and less competitive domestically and thereby making domestically produced goods more popular (Kohler, Manalo and Perera, 2014). According to Kohler, Manalo and Perera (2014) the first effect is likely to dominate the second effect and therefore contribute to economic growth, known as competitive devaluation. The latter is not the only

7 theory on how exchange rates boost growth, another well-known channel through which exchange rates boost economic activity is through its effect on capital flows. If the relative attractiveness of domestic and foreign investment is affected by the depreciation it can increase economic activity (Kohler, Manalo and Perera, 2014). With continuing globalization and financial integration capital flows now account for a larger share of cross-border transactions (Hau and Rey, 2004). When the domestic currency depreciates, it reduces that country’s wages and production costs relative to those of the foreign counterpart. Holding everything else equal, the country with the depreciated currency is said to have a competitive advantage or enhanced attractiveness as a location for receiving productivity capacity investments (Goldberg, 2009). According to Goldberg (2009) through this “relative wage” channel, the exchange rate depreciation improves the overall rate of return to foreigners contemplating an overseas investment project in this country and therefore boosting economic activity. The literature also suggests a correlation between real exchange rate and GDP growth. Several studies show a positive relationship between a weak exchange rate and growth (Habib, Mileva and Stracca, 2016). For example, Hausmann, Pritchett and Rodrik (2005) find a correlation between growth accelerations and increases in investment, trade and real exchange rate depreciations. However, Habib, Mileva and Stracca (2017) do conclude that the exchange rate matters for growth in developing economies but matters substantially less in advanced ones, which is in line with previous research. A country’s currency can therefore be used as a vital tool to boost economic activity. Exchange rates are driven by several mechanisms. The earliest model of exchange rates is the monetary model, it assumes that the current rate is determined by current fundamental economic variables, such as money supply and output levels. Combined with market expectations of future exchange rates, the model yields the value of the current exchange rate (Hopper, 1997). The model focusses on the demand and supply of a currency. If there is too much currency, then there is too much of it chasing too few goods. If nothing else changes, prices tend to rise, causing inflation and a fall in the value of the currency (The balance, 2017). Other economic fundamentals in the monetary model are the level of real output and futures variables that determine the market’s expectations. When output increases, while holding everything else constant, a fall in the average price level is expected and produces an appreciation of the currency. Furthermore, if the market’s expectation of the future exchange rate were to change, the current exchange rate would move in the same direction (Hopper, 1997). An extension of this model described as the portfolio balance model says the supply and demand for foreign and domestic assets, such as bonds and money, determine the exchange rate. Whenever aggregate economic conditions change, such as a change in domestic or foreign interest rates, agents adjust their portfolios to a new equilibrium through which they influence the exchange rate (Mussa, 1981).

8

Chapter 2: Understanding Quantitative easing

The first two sections explain what QE is and through which transmission channels it possibly affects the economy. Section three and four describe the QE program implemented by the ECB and limitations and risks of implementing such programs, respectively.

2.1: Quantitative easing: an introduction

Pre-crisis consensus on the role of monetary policy was containing inflation and use monetary policy tools to mop up the aftermath of the burst of a bubble rather than using it to tackle its build-up, since it is far from clear whether asset bubbles can be identified ex ante (Joyce, Miles, Scott and Vayanos, 2012). Joyce et al (2012) argue that this perspective has been highly challenged since the GFC. Financial stability and the ability to contribute to economic recovery have now become of great focus for conventional monetary policy but the latter comes with two great challenges and leads to the consideration of other policy forms (Joyce et al, 2012).

The first problem is that of the Zero Lower Bound (ZLB) on nominal interest rates (Joyce et al, 2012). Wieland (2010) argues that as long as savers have the option to hold cash, a zero-interest-bearing asset, a nominal interest rate of zero constitutes an important obstacle for central banks. To boost aggregate demand central banks would like to reduce the real interest rate, however since the nominal interest is stuck at its ZLB it may not be able to accomplish this objective with its conventional tools since the nominal interest can not be lowered beneath zero (Wieland, 2010). A second problem that led to the consideration of other forms of monetary policy is the disruption of the financial system itself (Joyce et al, 2012). Due to the severity of the recent crisis and its aftermath, many banks and borrowers were called into question. The outcome was a break in the usually reliable relationship between changes in official interest rates and market interest rates, and additionally the fear was that banks were holding onto funds to improve their viability rather than on-lending to the private sector, requiring some central banks to

intervene with the direct provision of credit (Joyce et al, 2012). Conventional monetary policy could simply not work, and unconventional tools were needed.

Unconventional monetary policy is generally described as policies that directly target cost and availability of external finance to banks, households and non-financial companies and therefore can take many forms. The sources of finance are in the form of central bank liquidity, loans, fixed-income securities or equity. The most common forms involve expansion of central banks’ balance sheets and attempts at influencing non-standard interest rates (Bini Smaghi, 2009). As mentioned before, this thesis focuses on the

9 The 1990s were tough years for the Japanese economy, they saw a bursting of a real estate bubble and dealt with deflationary pressures (Joyce et al, 2012). As Japanese money markets rates declined toward zero, the Bank of Japan (BoJ) became the first to venture into unfamiliar territory described as QE. The BoJ announced that it would increase its balance sheet and purchase assets directly, including outright purchases of government bonds (Wieland, 2010). This shifted the focus from uncollateralized overnight call rate to targeting quantity variables, hence Quantitative Easing. The BoJ purchased government securities from the banking sector thereby boosting the level of cash reserves the banks held in the system. The aim was targeting a high enough level of reserves which eventually would spill over into lending into the broader economy, helping drive asset prices up and remove deflationary forces (Joyce et al, 2012).

Since then, the Federal Reserve (Fed), European Central Bank (ECB) and the Bank of England (BoE) have all adopted similar policies that have also led to considerable increases in their balance sheets, even though there are significant differences both amongst themselves and with Japan in terms of how they have implemented QE and other unconventional policies (Joyce et al, 2012). The main difference between the different asset purchase programs is that the Fed, BoE and BoJ targeted substantial amounts of assets purchases whereas the ECB mainly focused on easing their monetary policy stance to elastically supply long-term loans (Kenourgios, Papadamou, Dimitriou, 2015). The ECB faced different problems than the other central banks which explains their particular response. Stresses within the euro area especially in 2011 and 2012 lead to a very substantial outflow of euro deposits from banks in some of the peripheral countries and into banks in other euro area countries (Joyce et al, 2012). According to Joyce et al (2012) this caused major imbalances within the euro area banking system. The ECB therefore had to alleviate the funding difficulties by easing credit conditions (Joyce et al, 2012).

Conventional monetary policy was firmly based on theoretical work, however, since unconventional policies is a response to circumstance rather than driven by intellectual developments, there is no clear explanation on how it impacts the economy itself (Joyce et al, 2012). The next section therefore explains possible transmission channels through which unconventional monetary policy such as QE can affect the economy and exchange rates.

2.2: The transmission channels

Bini Smaghi (2009) explains in a lecture at the International Center for Monetary and Banking Studies how prior to the GFC monetary policy focused on setting a target for the overnight interest rate in the interbank money market and adjusted the supply of central bank money to the target through open market operations. Monetary policy was then simply seen as the decision on key interest rates (Joyce et al, 2012).

10 The purpose of QE is the same as setting key interest rates, however it directly operates on different segments of the yield curve, asset prices and exchange rates through various transmission channels (Chen, Cúrdia and Ferrero, 2011). The transmission channels identified are:

1. Confidence channel 2. Policy signaling channel 3. Portfolio rebalancing channel 4. Market liquidity channel 5. Money channel

Figure two summarizes the different transmission channels followed by an explanation of each channel. Figure 2: QE transmission channels

Source: Quarterly Bulletin 2011 Q3

“ECB quantitative easing drives confidence for now” reads the Financial Times headline on April 1st,

2015. The first transmission channel is described as the confidence channel: if enough people believe QE works it works (FT, 2015). To revive aggregate demand, businesses and consumers must believe in the prospect of rising incomes and profits, low borrowing costs and sufficient liquidity before they will invest or consume and lead the way towards a recovery (Hughes-Hallett, 2017). By taking decisive action, the ECB restores confidence in the financial system, and encourages investment and spending directly. The latter ultimately affects yields and asset prices by reducing risk premium (Fratzscher, Lo Duca and Straub, 2014).

Eggerston and Woodford (2003) conclude that the key for monetary policy conducted at the ZLB is by managing expectation regarding the future conduct of policy, described as the signaling channel. For instance, bond purchases will cause private agents to expect future interest rates to be lower since a central bank’s bond holdings will be higher if it delivers lower short-term rates in the future (Clouse, Henderson, Orphanides, Small and Tinsley, 2000). This channel affects all bond market interest rates,

11 since lower future rates, via the expectations hypothesis, can be expected to affect all interest rates

(Krishnamurthy and Vissing-Jorgensen, 2011).

The third channel through which asset prices and exchange rates are effected is the portfolio rebalancing

channel. When a central bank purchases government bonds it reduces the amount of duration risk in the

hands of investors and potentially lowering the term premium (Koijen, Koulischer, Nguyen and Yogo, 2016). As investors respond to the lower term premium they substitute to other assets, leading to portfolio rebalancing and induce a price effect on various assets (Fratzscher, Lo Duca and Straub, 2014). As agents rebalance their portfolios the exchange rate changes if rebalancing takes place outside of the domestic country.

The next transmission channel reviewed is the liquidity channel. Central banks increase liquidity in the banking system by purchasing long-term securities and issuing bank reserves under QE. Krishnamurthy and Vissing-Jorgensen (2011) argue that the increased liquidity decreases the liquidity premium on the most liquid bonds and therefore raises yields relative to other less liquid assets. On the contrary, Joyce et al (2012), assert that by purchasing relatively long-term securities, central banks reduce the privately held stock. With less duration risk to hold in aggregate, a lower premium is required. This tends to reduce liquidity premium for all long-term securities which leads to an increase in asset prices. The rise in asset prices and reduced yields make it easier for companies and households to raise funds and therefore easing credit conditions (Joyce et al, 2012).

QE, like helicopter money involves money creation by injecting liquidity in the banking system and as the money market equilibrium shows monetary expansions reduce real and nominal interest rates (Dwivedi, 2005). The latter is known as the money channel.

The channels mentioned above ease credit conditions by lowering interest rates at different maturities. The lowered borrowing costs increase consumer’s likeliness to buy things and businesses are more likely to invest in new equipment, software or buildings and therefore increase economic activity. A lower interest rate however, offers lenders in an economy a lower return relative to other countries. Therefore, the reduced interest rates lead to capital outflow and cause the euro to depreciate. The impact is however mitigated if additional factors serve to drive the currency in the opposite direction. Additionally,

increasing the number of euros circulating in the economy would intuitively lead to a decrease in its value. Exports increase as they become cheaper and imports decrease as they become more expensive, in turn GDP rises (Mathai, 2012). This mechanism is clearly seen in the euro area economy since 2014; the euro depreciated, aided by a secular appreciation of the US dollar. Which is likely to be the cause of the upturn in growth in the euro area in 2015-2016 (Hughes-Hallet, 2017). The current objective is to analyze

12 whether QE has indeed contributed to the depreciation of the euro against the dollar through the

mechanisms described above.

2.3: Quantitative easing at the ECB

In this section, the quantitative easing program implemented by the ECB since 2009 is briefly introduced. Table one in the appendix summarizes the important announcement days and each program. The covered bond purchase program (CPBB1) was announced on May 7th _{2009 by the Governing Council of the ECB.}

The program was implemented from July 2009 until June 2010 and consisted of purchases of a nominal value of EUR 60 billion. The second covered bond purchase program (CPBB2) was announced on

October 10th _{2011, accordingly an amount of EUR 16.418 billion was purchased between November 2011}

and October 2012 (ECB press releases). On May 10th _{2010 the Governing Council announced the}

Securities Market Program (SMP), the latter was implemented from May 2010 until September 2012 and was intended to restore financial market liquidity (ECB monthly bulletin, 2010). At its peak, the SMP portfolio was around EUR 210 billion (Koijen, Koulischer, Nguyen and Yogo, 2016). The asset-backed securities purchase program (ABSPP) and third covered bond purchase program (CBPP3) were

announced on September 4th _{2014 and implemented in November 2014 and October 2014 respectively.}

On January 22nd _{2015 the Governing Council announced its newest addition to the QE program, the}

expanded asset purchase program (APP) to fulfill its price stability mandate. The program encompasses the ABSPP and CBPP3, a public-sector purchase program (PSPP) and corporate sector purchase program (CSPP). Combined monthly purchases were EUR 60 billion until March 2016 (ECB press release, 2015). The purchases were then increased to EUR 80 billion until March 2017 and reduced to EUR 60 billion from April 2017 until the end of December 2017 (ECB Introductory statement, 2017). In the latest press conference on June 8th _{2017 held by Mario Draghi, it was suggested that the asset purchases could extend}

further into 2018: “…we confirm that our net asset purchases, at the current monthly pace of €60 billion, are intended to run until the end of December 2017, or beyond, if necessary, and in any case until the Governing Council sees a sustained adjustment in the path of inflation consistent with its inflation aim.”, despite announcing that the risks to the growth outlook now look "broadly balanced" (ECB Press

conference, 2017).

2.4: Limits and risks of QE

Further extending the QE program into 2018 might seem like the only way to obtain the inflation target, but it could also be too much of a good thing. This section explains the limits and risks of QE.

13 Three major limits and risks can be identified:

1. Threats to economic performance and financial stability 2. Transmission failures and global cycles

3. Wealth inequalities and reduced savings

The first limit in extending the ECB’s QE program is that its impact on economic performance may fall short of popular expectations. Evidence shows rather modest economic improvements and could suggest asset purchases not being enough to create significant recovery. QE turns out to be more effective in preventing a tough situation from getting worse rather than generating substantive recovery. A serious risk would then be the potential contraction of gains in economic performance caused by QE fatigue and therefore reduce the probability of success (Hughes-Hallett, 2017). Additionally, prolonged

accommodative monetary policies could also pose some challenges to financial institutions and lead to adverse consequences for financial stability (Claeys and Leandro, 2016). In search for higher yields to match their long-term liabilities, pension funds and insurance companies are forced to make riskier investments (Hughes-Hallett, 2017). Claeys and Leandro (2016) argue that if risk-taking becomes excessive and goes beyond what is socially, desirable it might contribute to future financial instability. Also, investors might look for more riskier investments abroad and hence transferring risk to and easing monetary conditions in the recipient economies (Hughes-Hallet, 2017). Increases in asset prices

disconnected from fundamentals are also a potential consequence of QE and pose a threat to financial stability (Claeys and Leandro, 2016). Speculation on higher asset prices creates a serious risk of an asset bubble creation, especially in housing (Hughes-Hallet, 2017). The possibility of an asset price bubble is the form of instability that causes the greatest concern according to Hughes-Hallet (2017). However, there is little sign of an asset bubble creation in the euro area level but some regions and sectors do show signs of a possible bubble emerging, the German housing market for example. Yet ECB policies must be directed at the euro area rather than specific regions (Hughes-Hallet, 2017). Bank regulation, stricter supervision and market pressure could eliminate the potential threat to financial instability (Claeys and Leandro, 2016). Claeys and Darvas (2015) conclude that the benefits of unconventional monetary policy measures such as QE outweigh their potential risks to financial stability.

The second potential fallback of QE is failure in its transmission channels. Lower interest rates at

different maturities is not a guarantee for boosting investment and consumption. Investors and consumers may prefer to save and pay off past debts as protection against future recessions. If the transmission channels don’t function, then QE would be ineffective (Claeys and Leandro, 2016). Transmission

channels may fail due to risk aversion or incomplete pass-through and cause the target variables not to be boosted as they should be. In recent years there has also been a rise in financial integration world-wide.

14 Consequently, assets have developed similar characteristics in prices or yields. That means that credit flows in different economies show similar pro-cyclical patterns and volatilities, given free capital flows (Claeys and Leandro, 2016). Synchronized global financial cycles constrain domestic monetary policies and therefore increase the risks of a fall short of expectations, inflation and instability if the QE program is extended. Luckily, there are ways to deal with this problem by targeting capital controls, policies undertaken to restrain the drivers of the world financial cycle, macroprudential policies to restrain cyclical increases of credit and leverage in recipient economies, domestic policies to weaken the transmission of excess credit/leverage using financial regulation or to weaken the transmission of world financial cycle by slowing down the monetary transition mechanism (Claeys and Leandro, 2016).

A long-term trend in rising income and wealth inequality is recently observed in many advanced economies. This is primarily due to deep structural changes, globalization, demographics, institutional and political changes and, particularly, changes to fiscal, educational and labor institutions (Claeys and Leandro, 2016). Though there are some concerns QE is expected to change the distribution of income and wealth, as lower interest rates and abundant liquidity will benefit investors, banks, firms, mortgage holders more than savers, employees or pensioners (Hughes-Hallet, 2017). Through increases in financial asset prices, central bank asset purchases could increase inequality between wealthy and poor or young and old. Increases in equities or government and corporate bonds will tend to favor the rich who usually own them (Claeys and Leandro, 2016). According to Hughes-Hallet (2017) the two big influences on savings and consumption however have been the loss of jobs, reduced earnings and expectations of inflation. Empirical research suggests that asset purchase programs tend to boost inflation, output and employment. So, compared to what would have happened without the implementation of QE, most people are better off (Claeys and Leandro, 2016).

For the euro area, the risks lie within the divergent monetary policies it will have to live, meaning more volatile exchange rates, especially versus the dollar, and consequent financial changes. Trade imbalances are likely to become larger and the QE response weaker (Hughes-Hallet, 2017). Hughes-Hallet (2017) also addresses the limit of QE within the euro area, arguing that the exchange rate gains will accrue to the more competitive economies. It will mainly benefit trade in countries that have a large export sector such as Germany, which are not the ones currently needing assistance (Financial Times, 2015).

The main reason why according to Fazi (2015) the QE program has had limited impact on the economy is due to the lack of coordination between fiscal and monetary policy. In the long term, the best proposition is to implement structural reforms to make the recovery self-sustaining, a long-term proposition that imposes a dilemma, for some, between short run costs and long run benefits (Hugh-Hallet, 2017). For now, it is the combination of the fiscal and monetary policy that can strengthen the transmission

15 mechanisms. Expansionary policies can offset the effects of deleveraging, inflation, tighter bank

regulation and the risk of financial instability and therefore lower the probability of additional risks if QE were to be extended (Hughes-Hallet, 2017).

Chapter 3: Literature review

The following chapter describes the empirical evidence on the effects of QE. There are two main strands of literature on the effects of QE, one describes the macroeconomic effects of QE and the other focuses on the effects on financial markets. A general summary is given on the latter two, followed by a more specific strand of literature which focuses on the euro area and the EUR/USD exchange rate.

3.1: Empirical evidence

The effects of conventional monetary policy shocks have well been documented in the literature (Schenkelberg and Watzka, 2011). Evidence on monetary policy shocks on exchange rates date back to 1993 when Eichenbaum and Evans (1993) present empirical evidence of sharp and persistent depreciation of the US dollar following an expansionary monetary policy. Rosa (2010) investigates the impact of US monetary policy on the level of exchange rates and finds that both policy decisions and communication have economically large and highly significant effects. A cut in the federal funds target is on average associated with a depreciation of the exchange value of the dollar against foreign currencies (Rosa, 2010). Overall, there is a broad consensus that expansionary conventional monetary policy shocks negatively effects interest rates however, only sluggishly and temporarily. Which is in line with macroeconomic theory on how monetary policy shocks affect the real economy during normal times (Schenkelberg and Watzka, 2011).

Empirical literature on unconventional monetary policy however is still growing and there is no clear consensus on what the effects and their magnitude are. The macroeconomic effects of QE are hard to quantify since the lags may be long and variable and there are consequently a host of additional factors that need to be controlled for (Joyce et al, 2012). Nonetheless a growing literature has begun to describe the macroeconomic effects of QE. Meinusch and Tillmann (2014) show that QE shocks in the US have a significant impact on real and nominal interest rates and financial conditions, however real activity is less affected. On the contrary using VAR models Weale and Wieladek (2016) and Gambacorta, Hofmann and Peersman (2014) show that an exogenous increase in central bank balance sheets at the ZLB leads to a temporary increase in economic activity and consumer prices. However, Gambacorta, Hofmann and Peersman (2014) conclude that the impact on the price level is weaker and less persistent than the effects of conventional monetary policy. Chen, Cúrdia and Ferrero (2011) also find relatively moderate effects on macroeconomic variables such as GDP growth and inflation.

16 On the contrary the effects on financial markets are usually more amenable to direct event studies (Joyce et al, 2012). The following literature focuses on the effects on medium and long-term interest rates. Krishnamurthy and Vissing-Jorgensen (2011) find using an event-study methodology a large and significant drop in nominal interest rates on long-term safe assets but the impact on the interest rate is reduced when focusing on less safe assets. Gagnon, Raskin, Remache and Sack (2011) present evidence that the Fed’s asset purchases led to economically meaningful and long-lasting reductions in longer-term interest rates on a range of securities, including securities that were not included in the purchase

programs. These reductions in interest rates primarily reflect lower risk premiums. Although the precise estimates differ across studies, there is a broad consensus in the literature that QE had economically significant effects, at least on government yields. However, the debate on the transmission channels linking asset purchases and asset prices continues (Joyce et al, 2012).

Driffill (2016) analyzed unconventional monetary policy in the Euro area and concluded that QE is presumably contributing to narrowing the spread of bond yields across the Euro area. Also, it appears that QE has modest but positive effects on national income, employment and inflation while the dangers associated with it appear to be remote (Driffill, 2016). Peersman (2011) examines the macroeconomic effects of traditional interest rate innovations and unconventional monetary policy actions in the Euro area economy, he finds that a policy action which raises the size of the central bank balance sheet, such as QE, has a hump-shaped effect on economic activity and a permanent impact on consumer prices. However, the pass-through is more sluggish (Peersman, 2011). Whereas a rise in the balance sheet of the euro area is passed on to bank lending via a decline in interest rate spreads of banks, the spreads increase

significantly after a fall in the policy rate (Peersman, 2011). Koijen, Koulischer, Nguyen and Yogo (2016) consider the effect of quantitative easing both on asset prices and portfolio holdings of investors. They find possible reduction in the duration mismatch in the euro area but no large portfolio shifts towards other assets such as corporate bonds or equities in the euro area. Using a difference-in-difference model they also find a decline in bond yields by -13bp on average.

As mentioned before empirical literature on the effects of QE are still limited, especially that on the effect on the exchange rate. Studies that have focused on QE effects on exchange rates mostly find a

depreciation in the home currency. Glick and Leduc (2013) examine the effects of unconventional monetary policy announcements on the value of the dollar against the currencies of major US trading partners using high-frequency intraday data. They find that surprise monetary policy easing

announcement since the crisis began, have significant negative effects on the value of the dollar which are comparable to the changes prior to the crisis under conventional monetary policy. Neely (2012) finds evidence of a significant drop in the US dollar after the first program of LSAP. Another study by Driffill

17 (2016) concludes that after the announcement of the extended APP in 2014 the euro dropped significantly and declined dramatically more in 2015 when the plans for the program were clearly set out. Kenourgios, Papadamou, Dimitriou (2015) study the effects of QE announcements by the ECB, BoJ and BoE on exchange rate dynamics by applying a univariate APARCH (1,1). They find a delayed negative response of the euro accompanied with increased variability before and after the ECB announcements. The latter suggests an unclear signal of future monetary policy actions for investors, which can be explained by the price stabilization policy followed by the ECB (Kenourgios, Papadamou, Dimitriou, 2015). For BoJ’s and BoE’s QE announcements Kenourgios, Papadamou and Dimitriou (2015) find a more direct and

significant reduction on their currencies without producing increased volatility. These findings highlight the increased credibility and effectiveness of these central banks’ monetary easing policies and support the existence of a signaling channel in the foreign exchange market (Kenourgios, Papadamou, Dimitriou, 2015). On the contrary Ueda (2012), fails to find evidence for a weakened yen after the introduction of unconventional monetary policy in Japan. However, evidence also shows that in response to lower yields in the euro area foreign investors rebalance their portfolio towards more attractive investment

opportunities outside of the euro area (Koijen, Koulischer, Nguyen and Yogo, 2016). The latter implies that a depreciation in the euro would be expected since capital is flowing out of the euro area to more profitable investments abroad. The question remains whether this rebalancing is happening due to the QE program or for other unforeseen reasons.

The literature described above mostly use event studies or try to quantify QE by using central bank’s balance sheets. This paper however, focusses on quantifying the transmission channels of QE rather than quantifying QE itself. It therefore contributes to the literature by analyzing through which channels QE has had a possible effect on the depreciation of the euro.

Chapter 4: Methodology and dataset

This chapter describes the methodology used to answer the research question. First, the hypotheses are introduced, followed by a description of the data and the last section describes the empirical strategy and the econometric procedure undertaken to answer the hypotheses.

4.1 The hypothesis:

The research question is answered by a hypothesis. Based on previous literature, the hypothesis is stated as follows: Quantitative easing had a significant impact on the change in the EUR/USD exchange rate through its transmission channels.

18

4.2 Data description:

This section aims to describe and summarize the dataset which is used in the empirical analysis. Data is collected for the euro area and the time period covered in the dataset ranges from January 1999 to March 2017 with a monthly frequency, consisting of data on the euro area. Two time periods are identified: From January 1999 to April 2009, before the start of the first QE program and from May 2009 to March 2017, the start of the first QE program until the most recent published data. The five transmission channels, the independent variables, are described in chapter 2 and quantified as follow;

1. The confidence channel is measured by the Consumer Confidence Index obtained from the OECD.

2. The signaling channel is measured by taking the prices of Euribor future contracts obtained from Datastream. Since daily prices of Euribor future contracts are obtained from Datastream, the monthly average is calculated.

3. The rebalancing channel is measured by the capital account obtained from Datastream. 4. The liquidity channel is measured by taking prices at different maturities of the most liquid

assets; German bonds at 30-year maturity are obtained from Investing.com.

5. The money channel is measured by M1 obtained from the OECD website. M1 includes currency i.e. banknotes and coins, plus overnight deposits.

The transmission channels of QE are not the only drivers of changes in the exchange rate. Therefore, control variables are added that may also drive the change in exchange rates, such as certain economic conditions like GDP and the inflation rate. GDP is measured by Industrial production and for the inflation variable the HICP inflation is used, data for these two variables are obtained from the OECD and Eurostat respectively. Additionally, daily Euro Stoxx 50 Volatility, VSTOXX, indices are converted to monthly data and added to the regression to control for market volatility. Since a QE program was also

implemented in the US during this time period it is essential to also consider the effects the Fed had on the EUR/USD exchange rate. The same transmission channels also hold for the relationship between the Fed’s QE program and the EUR/USD exchange rate. If the Fed’s QE program was more successful than the ECB’s, it is expected that the euro would instead appreciate and thus have the opposite effect on the exchange rate. This could explain why the euro has not been declining continuously. To control for the QE program implemented by the Fed, the difference between the 3-month Euribor and the interbank rates for the US are also taken. The latter rates are obtained from Datastream.

The dependent variable is the monthly percentage change in the EUR/USD exchange rate and described as the amount of dollars per euro. Thus, a decrease in the exchange rate implies that the dollar has become

19 stronger against the euro. Exchange rates are obtained from Eurostat and plotted in the graph below. The graph shows the EUR/USD exchange rate since January 1999. The EUR/USD exchange rate has been quite volatile since its implementation, falling below parity a year after its creation in 1999 and reaching a peak of 1.6 right before the crisis. Since the implementation of QE in May 2009 the EUR/USD exchange rate has been especially volatile but an overall declining trend can be seen, especially during the past two years. This is in line with economic theory, however the empirical analysis will also have to prove it. Graph 1: Historical EUR/USD exchange rate

Source: Eurostat

Since the start of the ECB’s QE program, three main announcement days regarding the QE program are identified:

1. May 7th 2009: The ECB announces its purchase of euro-denominated covered bonds issued in the euro area.

2. January 22nd 2015: The ECB announces an expanded asset purchase program, which adds the purchase of sovereign bonds to its existing private sector asset purchase program.

3. January 19th 2017: The ECB confirms that it will continue to make purchases under the asset purchase program.

Source: ECB press releases. If the exchange rate is analyzed from 2015, at the start of the extended asset purchase program, it shows

the euro having undergone major declines. After the announcement of the expanded asset purchase program the euro dropped significantly against the dollar, see graph 2. The drop can be explained by the

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 Jan -99 N o v-99 Se p -00 Ju l-01 May-02 Mar -03 Jan -04 N o v-04 Se p -05 Ju l-06 May-07 Mar -08 Jan -09 N o v-09 Se p -10 Ju l-11 May-12 Mar -13 Jan -14 N o v-14 Se p -15 Ju l-16 May-17

Exchange rate

Exchange rate

20 transmission channels described in section 2. The euro also dropped significantly around November 2016. Part of this decline could be explained by Trump winning the elections, since his policy proposals are expected to lead to an appreciation of the dollar (Financial Times). However, after the January 2017 announcement the euro didn’t decline, the Trump momentum has apparently reversed and the euro rose against the dollar despite the announcement of furthering the QE program. This could partly be explained by weak economic data for the US and doubts over Trump’s ability to finalize his economic agenda and move it to Congress (CNN Money, 2017). The latter implies that other factors play a role in the

movement of exchange rates and if these factors dominate the effects of the QE program exchange rates could move in the opposite direction than the one expected. Therefore, the empirical model controls for some additional drivers of the exchange rate.

Graph 2: The EUR/USD exchange rate from January 2015

Source: Eurostat

4.3 The empirical model:

Since QE programs are very unusual and undertaken in times of severe economic difficulties and are only implemented when conventional tools cannot be used, the investigation of the effects of QE are limited (C. Martin and C. Milas, 2012). The problem lies within quantifying the effects of QE, since there is no well-defined policy instrument whose variation indicates the ECB’s policy stance and which is easily observable (A. Meinusch and P. Tillmann, 2014). Therefore, instead of quantifying QE itself,

transmission channels are identified and used to analyze through which of these channels QE has possibly effected the exchange rate. The method used is a simple ordinary least square (OLS) regression including a dummy variable to identify the period during which the QE program is in place. Lagged values are used

1 1.02 1.04 1.06 1.08 1.1 1.12 1.14 1.16 1.18

Exchange rate

Exchange rate

21 for the independent variables, to control for reverse causality and for the simple fact that monetary policy actions work with lags. The regression model is restated below:

[ 1 ] 𝑆

𝑡

̇ = 𝛼 + 𝜑

1

30𝑌

̇

𝑡−1

+ 𝜑

2

30𝑌

̇

𝑡

𝐷

𝑄𝐸

+ 𝛾

1

𝑀̇

𝑡−1

+ 𝛾

2

𝑀̇

𝑡

𝐷

𝑄𝐸

+ 𝛽

1

𝐶𝐴̇

𝑡−1

+ 𝛽

2

𝐶𝐴̇

𝑡

𝐷

𝑄𝐸

+

𝜃

₁

𝐶𝐶𝐼

̇

_𝑡−1

+ 𝜃

₂

𝐶𝐶𝐼

̇

_𝑡

𝐷

_𝑄𝐸

+ ∅

₁

𝐹̇

_𝑡−2

+ ∅

₂

𝐹̇

_𝑡

𝐷

_𝑄𝐸

+ Ω

₁

(𝑟

𝑡−1𝐸𝑈

− 𝑟

𝑡−1𝑈𝑆

) + Ω

2

(𝑟

𝑡𝐸𝑈

− 𝑟

𝑡𝑈𝑆

)𝐷

𝑄𝐸

+

𝜗

₁

𝑉𝑆𝑇𝑂𝑋𝑋

̇

_𝑡−1

+ 𝜔

₁

𝐻𝐼𝐶𝑃

̇

_𝑡−1

+ 𝜔

₂

𝐼𝑃̇

_𝑡−2

+ 𝜀

The dependent variable, 𝑆𝑡̇ , represents the monthly percentage change of the EUR/USD exchange rate.

The independent variables that represent the transmission channels are: • 30𝑌̇ _{; monthly percentage changes of 30-year German bond prices,} • 𝑀̇; monthly percentage changes of the money supply,

• 𝐶𝐴̇; monthly percentage changes of the capital account,

• 𝐶𝐶𝐼̇ ; monthly percentage changes of the consumer confidence index, • 𝐹̇; monthly percentage changes of LIFFE 3-month Euribor settlement price.

The regressors, 𝐼𝑃̇𝑡−1, 𝐻𝐼𝐶𝑃̇ 𝑡−1 , represent the control variables, industrial production and inflation

respectively. Since the onset of the crisis in 2007 the EUR/USD exchange rate has been quite volatile, which can be expected since there was a great amount of uncertainty. Therefore, the variable 𝑉𝑆𝑇𝑂𝑋𝑋̇ 𝑡−1

is added to control for market volatility. It is measured byVSTOXX Indices which are based on EURO STOXX 50 real-time options prices and are designed to reflect the market expectations of near-term up to long-term volatility by measuring the square root of the implied variance across all options of a given time to expiration (Stoxx.com). All control variables are measured in monthly percentage changes. The latter variables might be correlated to the main explanatory variable thus violating the first assumption of OLS regression if they are omitted. The regressor, 𝑟_𝑡−1𝐸𝑈 _{− 𝑟}

𝑡−1𝑈𝑆, represents the percentage difference in

interbank interest rates between the euro area and US. An interaction term is added to analyze whether the difference had an additional impact during the time the ECB’s QE program was implemented. The remaining variable is, 𝜀, and exhibits the residual.

The regression above is called an interacted regression model because it encompasses an interaction term between a dummy variable and a continuous variable. The dummy variable,

𝐷

_𝑄𝐸 , takes a value of one if a variable is observed from May 2009 when the first QE program was implemented, and zero if

otherwise. The interaction terms allow for different effects of the transmission channels on the percentage change in the exchange rate by altering the slope during the QE period. For example, if 𝛽2 is statistically

22 account on the exchange rate through its portfolio rebalancing channel, the total effect is then measured by the coefficients, 𝛽1 𝑝𝑙𝑢𝑠 𝛽2, given that 𝛽1 is also statistically significant.

Section 5: Results

In this chapter, the results of regression model one are presented. First a brief explanation is given on the steps taken to obtain the results, then the main regression results are given followed by robustness checks to see how the results change. If the coefficients are indeed robust and stable, model one could be

interpreted as a valid model.

5.1: Checking the data

The first step is to identify the data as time series data. This is done using the tsset command in Stata. A variable mydate is created which represents the month and year. Then the tsset command is used to identify the latter variable as time series.

A crucial assumption in time series data is that the data are stationary, in other words, the mean, variance and autocorrelation structure do not change over time. If data is not stationary the empirical results do not have a meaningful interpretation. The latter can be tested by an augmented Dickey-Fuller test. It tests whether a variable follows a unit-root process, the null hypothesis is that the variable contains a unit root, and the alternative is that the variable was generated by a stationary process. The dfuller command in Stata performs this test. It can be concluded that all variables are stationary and therefore the results have a meaningful interpretation.

The last step taken before the regression is run is creating the interaction terms between the dummy variable and each transmission channel. Interaction terms are created by generating new variables using the generate command. Six interaction terms are created including the interaction term between the dummy variable and the percentage change in the spread between the euro and US interbank rate.

5.2: Regression results

After running the regression of model one, with 216 observations a R-squared of 0.190 is found. R-square is a statistical measure of how close the data are to the fitted regression line. A R-squared of 19.0 percent means that 19.0 percent of the change in the EUR/USD exchange rate can be explained by the

independent variables used in model one. In table two of appendix two the results are presented. In the regression of model one, homoscedastic variance of the residuals is automatically assumed. The latter implies there should be no pattern between the residuals and the predicted values, if the variance of the residuals is not constant then the variance is known to be heteroscedastic. For the results to be

23 meaningful the residuals should be homoscedastic. One test on heteroscedasticity is the White’s test, this test is done on model one. White’s test obtains a p-value of 0.9783, not rejecting the null hypothesis of homoscedastic variance of the residuals and thus no evidence is found for heteroscedasticity.

Prior to analyzing the results, the vif command is used to check for multicollinearity2_{. A VIF value higher}

than ten may be of great concern for the regression results. In appendix two table three the VIF values for model one are presented. None of the VIF values are above ten, therefore there is no sign of

multicollinearity in the data.

Additionally, normality of the residuals is important to ensure that the p-values are valid, if the residuals are not normally distributed then the p-values will not be binding and consequently the model will not be accurate. The residuals are plotted in graph three in appendix two. The histogram shows the residuals being normally distributed with a mean of roughly zero. The outcomes of the tests described above conclude that the results obtained from model one can be analyzed and correctly interpreted.

Liquidity channel: The coefficient L1.y30, equals 0.00424. At a significance level of one and five

percent this coefficient is not significant, p-value equals 0.754. The interaction term, y30qe, is also not significant with a p-value equal to 0.246. The latter implies there is no statistically significant linear dependence of changes in 30-year German bond prices on the change of the EUR/USD exchange rate and therefore no evidence is obtained of a liquidity channel effect of QE on the EUR/USD exchange rate. The liquidity channel was measured by taking prices at different maturities of the most liquid asset, 30-year German bonds. However, the latter is not the only long-term liquid asset bought by the ECB, therefore a better measure could have been a weighted average of long-term liquid assets bought by the ECB and thus explain the non-significance of the coefficients.

Money channel: Both the money supply and interaction term are significant at the five percent level with

p-values of 0.009. This implies that the money channel has been effective in influencing the exchange rate during the QE period. The money supply coefficient obtained is -0.587. The latter implies an increase in the money supply two months ago leads to a decrease in the EUR/USD exchange rate leaving everything else constant. The interaction term is also negative, -1.048 which implies that the additional increase in the money supply due to the QE program has had a negative impact of -1.048% on the exchange rate leaving everything else constant. The total effect of a one percent change in the money supply on the EUR/USD exchange rate is therefore, -1.635%. The latter findings are in line with economic theory, since

2 _{Multicollinearity exists when two or more of the predictors in a regression model are moderately or highly}

correlated. The latter can result in imprecise predictions of the regression model. Using the VIF command in stata no sign of multicollinearity is found, see table 2.

24 an increase in the money supply is expected to decrease the value of the corresponding currency in the short run.

Portfolio rebalancing channel: the coefficients for the capital account is insignificant with a p-value of

0.825. On the contrary, the interaction term is significant with a p-value of 0.021. A negative coefficient of -0.000307 is obtained, implying that a positive change in the capital account effects the EUR/USD exchange rate negatively by -0.000307% because of the implementation of QE. The latter finding is the opposite of what economic theory implies. If the domestic capital account decreases, capital flows out of the country, a decrease in the domestic currency is expected. However, since this is merely a dummy variable a positive change in the capital account, since the start of QE, may not negatively effect the exchange rate if the capital account coefficient is pushing it in the opposite direction. The total effect of the capital account, β1 plus β2, could still be positive.

One could also argue that the capital account does not perfectly measure portfolio investments inflows and outflows. Therefore, as a robustness check in the next section, the portfolio rebalancing channel is measured by portfolio investments instead.

Confidence channel: the confidence coefficient 2.069 is not significant with a p-value of 0.265. The

interaction term is also not significant with a p-value of 0.356 implying that QE has not had an impact on the exchange rate through its confidence channel.

Signaling channel: the coefficient for Euribor future prices is 2.778 and highly significant with a p-value

of 0.004. The latter suggests that changes in Euribor futures prices influence the EUR/USD exchange rate. Leaving everything else constant a one percent change in Euribor future prices leads to a 2.778% change in the EUR/USD exchange rate. However, no evidence is found that the ECB’s QE program had an additional impact on the change in the EUR/USD rate through its signaling channel since the interaction term is non-significant with a p-value of 0.106.

Interbank spread: Both the spread between the Euribor and US interbank rate and its interaction term

are non-significant with p-values of 0.758 and 0.931 respectively. However, this does not necessarily mean the Fed’s monetary policy has no influence on the EUR/USD exchange rate but that no evidence is found of that the change in the spread impacted changes in the exchange rate. Perhaps the spread is not the most accurate way to control for the Fed’s monetary policy decisions.

Control variables: Results of the regression show that the control variable, HICP, is not significant with

a p-value of 0.637. The coefficient can therefore not be interpreted. The reason for the latter finding can be due to several reasons. Economic conditions have been rather volatile in the past years due to the

25 recent GFC. Economic growth is sluggish and the economy is not improving as quick as was expected. This could have affected the relationship between the control variables and the exchange rate. Other problems could be omitted variable bias or too little observations. The volatility index and industrial production variables however, are significant with a p-value of 0.006 and 0.014 respectively. The coefficient of VSTOXX is negative, -0.0270 indicating that a one percent increase in market volatility decreases the EUR/USD exchange rate by 0.027% when everything else is held constant. This is in line with economic theory since more volatility brings uncertainty and therefore negatively affect a currency. The effect is small yet significant. The coefficient for industrial production is positive with a value of 0.382 indicating that an increase in industrial production has a positive effect on the EUR/USD exchange rate of 0.382%. The sign of the coefficient is what is to be expected; a strong country, thus high industrial production, generates a strong currency.

Overall the results are a bit uncertain. It is widely known that monetary policy actions work with lags. Empirical evidence also shows that monetary policy actions affect economic conditions only after a lag that is both long and variable (Friedman, 1961). The latter implies that depending on the number of lags used, the results could change. And that for each action and variable the lag could differ. Therefore, it is important to have a good understanding and reasoning what the appropriate number of lags are. In this model, lagged values were also used to solve the problem of endogeneity, however, it is not certain if the latter has completely solved the problem and could potentially lead to a less precise model. If the problem of endogeneity is not fully solved, it could produce biased and inconsistent coefficients and explain why some coefficients are not in line with economic theory. In the case that there is still fear of endogeneity an alternative method such as two-stage least square model with instrumental variables could be a better alternative. However, finding appropriate instrument is quite difficult.

A reason why lagged values are not the best solution to the endogeneity problem is due to strategic anticipation, if agents anticipate the ECB introducing or continuing with a QE program the reaction can be even before the announcement day. Agents reacting to future monetary policy decisions leads to the exchange rate adjusting prior to the implementation of the program itself. Therefore, a model which controls for the anticipating behavior of agents would be better.

Also, there is some debate on when the QE program has started. Some might argue this was when the ECB introduced its extended asset purchase program as the latter is far larger and widespread than the programs implemented before. Therefore, a robustness check is done to see how the results change when the QE time period is alternated.

26

5.3: Robustness checks

5.3.1: Additional interaction terms for each control variable

First, three new interaction terms are added for each control variables. Because of the increased

uncertainty and volatility since the onset of the crisis, the relationship between these control variables and the EUR/USD exchange rate might have changed with time. To see how this would affect the model, model one is slightly adjusted by adding three new interaction terms:

[ 2 ] 𝑆

𝑡

̇ = 𝛼 + 𝜑

1

30𝑌

̇

𝑡−1

+ 𝜑

2

30𝑌

̇

𝑡

𝐷

𝑄𝐸

+ 𝛾

1

𝑀̇

𝑡−1

+ 𝛾

2

𝑀̇

𝑡

𝐷

𝑄𝐸

+ 𝛽

1

𝐶𝐴̇

𝑡−1

+ 𝛽

2

𝐶𝐴̇

𝑡

𝐷

𝑄𝐸

+

𝜃

1

𝐶𝐶𝐼

̇

𝑡−1

+ 𝜃

2

𝐶𝐶𝐼

̇

𝑡

𝐷

𝑄𝐸

+ ∅

1

𝐹̇

𝑡−2

+ ∅

2

𝐹̇

𝑡

𝐷

𝑄𝐸

+ Ω

1

(𝑟

𝑡−1𝐸𝑈

− 𝑟

𝑡−1𝑈𝑆

) + Ω

2

(𝑟

𝑡𝐸𝑈

− 𝑟

𝑡𝑈𝑆

)𝐷

𝑄𝐸

+

𝜗

1

𝑉𝑆𝑇𝑂𝑋𝑋

̇

𝑡−1

+ 𝜗

2

𝑉𝑆𝑇𝑂𝑋𝑋𝐷

̇

_𝑄𝐸𝑡

+ 𝜔

1

𝐻𝐼𝐶𝑃

̇

𝑡−1

+ 𝜔

2

𝐻𝐼𝐶𝑃𝐷

̇

𝑄𝐸_𝑡

+ 𝜔

4

𝐼𝑃̇

𝑡−2

+ 𝜔

3

𝐼𝑃̇

𝑡

𝐷

𝑄𝐸

+ 𝜀

After running a regression for model two, an R-squared of 0.198 is obtained. The latter implies that 19.8% percent of the change in the EUR/USD exchange rate can be explained by model two. That is slightly higher than model one. The results are presented in table four in appendix three.

Adding interaction terms for the control variables does not lead to major changes, all variables that were statistically significant in model one, are now as well. However, the R-squared has slightly improved implying that model two better describes the variation in percentage changes of the EUR/USD exchange rate. None of the control variable interaction terms are significant, suggesting that there is also no evidence that a linear relationship between changes in the control variables and changes in the exchange rate has adjusted since the crisis.

5.3.2: Portfolio investments instead of capital account

The second robustness check analyzes how the results change when portfolio investments are used instead of the capital account in model two, leaving everything else unchanged.

[ 3 ] 𝑆

𝑡

̇ = 𝛼 + 𝜑

1

30𝑌

̇

𝑡−1

+ 𝜑

2

30𝑌

̇

𝑡

𝐷

𝑄𝐸

+ 𝛾

1

𝑀̇

𝑡−1

+ 𝛾

2

𝑀̇

𝑡

𝐷

𝑄𝐸

+ 𝛽

1

𝑃𝑂𝑅𝑇𝐼𝑁𝑉

̇

𝑡−1

+

𝛽

₂

𝑃𝑂𝑅𝑇𝐼𝑁𝑉

̇

_𝑡

𝐷

_𝑄𝐸

+ 𝜃

₁

𝐶𝐶𝐼

̇

_𝑡−1

+ 𝜃

₂

𝐶𝐶𝐼

̇

_𝑡

𝐷

_𝑄𝐸

+ ∅

₁

𝐹̇

_𝑡−2

+ ∅

₂

𝐹̇

_𝑡

𝐷

_𝑄𝐸

+ Ω

₁

(𝑟

_𝑡−1𝐸𝑈

_{− 𝑟}

𝑡−1𝑈𝑆

) +

Ω

₂

(𝑟

_𝑡𝐸𝑈

_{− 𝑟}

𝑡𝑈𝑆

)𝐷

𝑄𝐸

+ 𝜗

1

𝑉𝑆𝑇𝑂𝑋𝑋

̇

𝑡−1

+ 𝜗

2

𝑉𝑆𝑇𝑂𝑋𝑋𝐷

̇

_𝑄𝐸𝑡

+ 𝜔

1

𝐻𝐼𝐶𝑃

̇

𝑡−1

+ 𝜔

2

𝐻𝐼𝐶𝑃𝐷

̇

𝑄𝐸_𝑡

+

𝜔

4

𝐼𝑃̇

𝑡−2

+ 𝜔

3

𝐼𝑃̇

𝑡

𝐷

𝑄𝐸

+ 𝜀

After running the regression of model three, with 216 observations a squared of 0.173 is found. R-square is a statistical measure of how close the data are to the fitted regression line. A R-R-squared of 17.3 percent implies that 17.3 percent of the percentage change in the EUR/USD exchange rate can be

27 explained by the independent variables used in model three. That is lower than in model one and two. In appendix three table five the results of model three are presented.

One major change is that the portfolio rebalancing interaction term becomes non-significant with a p-value of 0.622. Concluding, model two explains the percentage changes in the EUR/USD better than model three.

5.3.3: Adjusted time period for dummy variable

The last robustness check adjusts the time period in which QE is identified. It is adjusted to only focusing on the day of the announcement of the extended asset purchase program, January 22nd _{2015, until the}

most recent obtained data. Therefore, model two is used again however a new dummy variable,

𝐷

_𝑄𝐸 , is generated that takes a value of one if a variable is observed from January 2015 when the extended asset purchase program was implemented, and zero if otherwise.

After running the regression of model two, with 216 observations and an adjusted time period for the dummy variable, a R-squared of 0.184 is found. R-square is a statistical measure of how close the data are to the fitted regression line. A R-squared of 18.4 percent means that 18.4 percent of the change in the EUR/USD exchange rate can be explained by the independent variables used in model four. That is slightly lower than in the original model one and model two. The regression results obtained are presented in the table six in appendix three.

Adjusting the QE time period leads to some major changes. The money supply and capital account interaction terms become non-significant as does the control variable industrial production. On the contrary, the confidence interaction term becomes highly significant, indicating that since January 2015 consumer confidence has greatly influenced the EUR/USD exchange rate. The coefficient is negative, -13.09. This mean that a one percent increase in consumer confidence adjusts the exchange rate

negatively by 13.09% after January 2015. The latter indicates that agents greatly believe QE works. If there is a great amount of confidence in the ECB’s unconventional policy then the exchange rate reacts accordingly. Also, the market volatility indicator and the Euribor future prices look stable as they stay significant when adjusting the QE time period.

5.4: Comparison of results

This section compares the results obtained from the regressions. If the original time period for QE is considered from March 2009 until the most recent obtained data, there is clear evidence of a money channel and importance of the market volatility, industrial production and Euribor future prices for changes in the EUR/USD exchange rates. These coefficients remain significant despite the model being