• No results found

The effect of Central Banks' balance sheet expansions on long-term yields : Japan, UK and the US

N/A
N/A
Protected

Academic year: 2021

Share "The effect of Central Banks' balance sheet expansions on long-term yields : Japan, UK and the US"

Copied!
50
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

The Effect of Central Banks’ Balance Sheet

Expansions on Long-term Yields: Japan, UK and

the US

Name:

Eline Keemink

Student Number:

10802843

Supervisor:

Cenkhan Sahin

Date of submission:

26-06-2018

Study program:

BSc. Economics & Business

Faculty:

Economics & Business

(2)

2 Statement of Originality

This document is written by Eline Keemink, who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is 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.

Abstract

This study evaluates whether the effects of QE on the balance sheets have a significant effect on the long-term yields, and if this effect is different between Japan, the UK and the US. The main purpose of QE is lowering the long-term yields, which in turn stimulates consumptions and investments. Long-term yields are affected by expected short-term rates, but also by macro-economic indicators such as the unemployment gap or the inflation rate. QE itself can be carried out in different forms, but leads in every form to an expansion of the central bank’s balance sheet. Based on a linear regression with data containing all QE measures since the start of this century are taken for Japan, the United Kingdom and the United States, it can be concluded that in all countries the size of the balance sheet significantly affects the long-term yields. The results show a significant effect of the size of the balance sheet, often passed through by the signalling channel, and indicates significant differences of the effects between the countries. The main contribution to the existing literature is that there has never been looked at the effects of expanding the balance sheet, ‘pure’ qualitative easing, on the long-term yields. Also the effects have not been tested. More research however is needed to see if these effects will also play when the balance sheets start decreasing again.

Keywords: Quantitative easing, Central Bank balance sheet, balance sheet expansion, unconventional policy, long-term yields

(3)

3 Table of Contents Abstract... 2 List of Abbreviations ... 4 1. Introduction ... 5 2. Literature Review ... 8

2.1 Conventional and Unconventional Monetary Policies ... 8

2.2 Transmission Channels ... 9

2.2.1 The Portfolio Balance Model ... 9

2.2.2 The Signalling Channel ... 11

2.3 Evidence on the Effectiveness of QE ... 12

2.4 Unconventional Monetary Policies in Japan, the UK and the US ... 13

2.4.2 Japan ... 14

2.4.3 United Kingdom ... 18

2.4.4 United States ... 20

3. Data and Methodology ... 23

3.1 Data ... 23 3.1.1 The Countries ... 23 3.1.2 Dependent Variable ... 23 3.1.3 Independent Variables ... 23 3.2 Methodology ... 25 3.2.1 The Regressions ... 25 3.2.2 Econometric Concerns ... 26 4. Results ... 27

4.1 Results per Country ... 27

4.1.1 Japan ... 27

4.1.2 United Kingdom ... 30

4.1.3 United States ... 32

4.2 Dummy Regression ... 34

5. Conclusion and Discussion ... 35

References ... 37 Appendix 1 ... 43 Appendix 2 ... 43 Appendix 3 ... 44 Appendix 4 ... 46 Appendix 5 ... 47 Appendix 6 ... 49

(4)

4

List of Abbreviations

There are several abbreviations used throughout this research. Though the term will be fully written the first time it is introduced, here is an overview of the abbreviations used.

APF Asset Purchase Facility BOE Bank of England BOJ Bank of Japan

CAB Current Account Balance

CME Comprehensive Monetary Easing CP Commercial Paper

QQE Quantitative and Qualitative Easing ECB European Central Bank

FED Federal Reserve

FOMC Federal Open Market Committee JGB Japanese Government Bond LSAP Large-Scale Asset Purchase MBS Mortgage-Backed Securities MPC Monetary Policy Committee QE Quantitative Easing

(5)

5

1. Introduction

Where conventional monetary policy manages liquidity in the money markets and pursues price stability by steering the level of key interest rate, this sometimes is not enough to get an economy out of a recession and thus conventional policies prove ineffective (Joyce, Miles, Scott, and Vayanos, 2012).

After the fall of the Lehman’s brothers in 2008 and the recession that followed, many central banks implemented unconventional policy measures, of which the most popular is the ‘quantitative easing’ policy (later in this article referred to as QE) (Belke, Gros, and Osowski, 2017). Consequently, the balance sheet sizes of central banks expanded considerably (Borio and Disyatat, 2010; Shiratsuka, 2010).

The first form of quantitative easing was already implemented by the Bank of Japan in 2001, when it announced it would start targeting its’ current account balances instead of the interest rate to, which used to be the main operating target of BOJ, to higher inflation and economic growth (BOJ, March 2001). The target was a 1 trillion-yen increase, starting from 4 trillion yen in February 2001. In March 2013, balance sheet assets had increased to 164.8 trillion yen (BOJ). The Fed followed the example of Japan in September 2008. In the statement of the Federal Open Market Committee (FOMC) in March 18th 2009 (Fed, 2009), they announced increasing the size of the Federal Reserve’s balance sheet by purchasing among others $750 billion of agency mortgage-backed securities (MBS). Its balance sheet increased from $869 billion in August 2007 to $3843 billion in November 2013 (Federal Reserve). The Bank of England started implementing quantitative easing beginning 2009 by announcing the Asset Purchase Facility (APF). In March 2009, the BOE would increase its central bank reserves by purchasing £75 billion of assets (Lyonnet and Werner, 2012). At the end of 2012, total assets of the BoE were worth more than £400.000 million, more than twice the size it had in March 2009 (BOE).

The goal of quantitative easing is reducing the long-term yields to boost economic activity, as described by e.g. Krishnamurthy and Vissing-Jorgensen (2011), Thornton (2014), and Belke et al., (2017). William Dudley stated in his speech in 2010 that purchasing long-duration assets lowers the long-term interest rates, which will boost economic activity. 1

Since the first actual implementation of quantitative easing in 2001, extensive research has been done about the effects of these measures on the state of the economy,

1 Dudley, William C. 2010. "The Outlook, Policy Choices and Our Mandate." Remarks at the Society of American Business Editors and Writers Fall Conference, City University of New York. On

(6)

6 yields and their overall effectiveness. The goal of this research is to add insights of the effects of balance sheet expansions on the long-term yields. While numerous studies have tried to estimate these effects, those effects have never been compared between different countries. It is interesting to see whether those effects are different between countries, because although the measures that were taken in the different countries are fundamentally different (Shiratsuka, 2010), all measures have led to significant increases in the central bank’s balance sheet sizes in Japan, the UK and the US (Joyce et al., 2012). Central banks based their decisions on monetary policy on the experiences of the BOJ earlier that century (BOJ, 2017). (Ueda (2012a) argues that there is no evidence that the increase of balance sheet sizes lowered the long-term yields, but later that year in a different article states that the effects of expanding the balance sheets on the long-term yields is an unanswered question (Ueda, 2012c). Shiratsuka (2010) argues that the responses of the central banks show “more similarities than differences” (Shiratsuka, 2010, pp. 83), and Gambacorta, Hofmann, and Peersman (2014), who use a panel data regression to estimate cross-country effects, state that there are no significant differences in the macroeconomic effects between countries.

Though it is argued that the impact of an increase in the central bank’s balance sheet, through which QE works (Joyce et al, 2012) has significant effects on the long-term yields, it is unclear if this effect is similar in different countries. The actual impact of QE on long-term yields by itself is also still under debate. For example, Thornton (2014) uses an econometric regression used in previous studies but concludes in his paper that QE did not significantly lower long-term yields. Also, Belke et al., (2017), who test the effect on long-term interest rates by looking at the trans-Atlantic interest rate relationship, conclude that QE has been ineffective. Gagnon, Raskin, Remache, and Sack (2011) however concluded that the large-scale asset purchases (LSAPs) conducted by the Federal Reserve succeeded in reducing the term premium, using an event study analysis and a regression estimation for long-term yields. Thus, I follow Breedon, Chadha, and Waters (2012) in their opinion that the effects of QE on long-term yields remain highly controversial.

Furthermore, since unconventional measures are still on-going it is interesting to look if the effects are still similar and significant with more recent data to similar researches. Japan, for example, has been implementing unconventional measures since the beginning of this century and announced again on the 21st of September 2016 to introduce “QQE with

(7)

7 Yield Curve Control2” and to purchase Japanese government bonds (JGBs) with an annual increase of about 80 trillion yen (BOJ, 2016).

In this paper the effects of a balance sheet expansion implemented by Central Banks on the long-term yields will be estimated. While Ueda (2012a, c) argues that the expansion of the balance sheet in general had no influence on the long-term yields3, it is undeniable that though unconventional policies differed per country, balance sheets increased substantially and therefore it is interesting to look at a general effect of those balance sheet increases on the long-term yield. The regression used by Gagnon et al. (2011) will be slightly modified in this paper and used to the measure the effects on long-term yields. Instead of only focusing on the United States, this research will be expanded to Japan and the United Kingdom to see if the regressions will give similar results for these different economies.

The paper is structured in the following way. In the first section the focus will be on unconventional policy measures and the way they have been implemented in the different economies. The second section will focus on existing literature on the effects of long-term yields. In the third section the data and methodology used will be described. In the fourth section, the regressions for Japan, the UK and US will be executed, and the effects of central bank expansions on the long-term yields will be estimated. The fifth section concludes the results of this study.

I find that, using the regression method for each country specifically and for the countries in general by using ‘dummy variables’, the effects of balance sheet expansions have a significant effect on the long-term yields, and that those effects are significantly different between the countries. This contributes to the literature by filling in the gap that Ueda (2012a, c) mentions in the existing literature, on whether QE (defined as increasing the size of the central bank’s balance sheets by Bernanke, Reinhart, and Sack (2004)) significantly lower the long-term yields. It also contributes in the sense that it is unclear from existing literature whether the effects of balance sheet expansions will be the same for the US as for Japan.

2 QQE stands for “Quantitative and Qualitative Easing”.

3However later that year mentions “that the effectiveness of pure quantitative easing remains an open question”

(8)

8

2. Literature Review

The definition of what quantitative easing exactly is, how it has been implemented and affected monetary variables is heavily varied over different researches (Borio and Disyatat, 2010). In this section, the definition of unconventional policies will be defined by naming various relevant researches, as well as the channels through which unconventional policies are expected to affect the long-term yields and the implementations and effects of QE in different countries.

2.1 Conventional and Unconventional Monetary Policies

Conventional monetary policy is defined by using short-term nominal interest rates (policy rates), which can influence the economy by their effects on other asset prices . Therefore, changing the policy rates can influence the output and employment levels (Fawley and Neely, 2013). Conventional monetary policy is implemented without the need to change the size of the central banks’ balance sheet significantly (Borio and Disyatat, 2010). However, due to the existence of a zero-lower bound (ZLB)4 on the interest rates, focusing on the short-term interest rate can constrain monetary policies (Girardin and Moussa, 2011; Chung, Laforte, Reifschneider, and Williams, 2012).

Borio and Disyatat (2010) distinguish unconventional policies by the fact that central banks use their balance sheets in order to affect market prices and interest rates with a maturity longer than 12 months. When using unconventional measures, banks are not constrained by the ZLB (Chung et al., 2012; Joyce et al., 2012). Unconventional monetary policy can also be defined even more broadly, as a “residual category: any other type of intervention by the central bank that does not depend for its operation on changing the risk-free nominal interest rate now or in the future” (Sheedy, 2017, pp. 130).

Unconventional monetary policies can be implemented in different forms and are therefore difficult to define, as the unconventional monetary policy “is defined by what it is not rather than what it is” (Joyce et al., 2012, pp. 272). Examples are the use of negative interest rates, attempts to influence longer-term interest rates or expansions of the central banks’ balance sheet (Joyce et al., 2012). Ueda (2012b) classifies unconventional measures that can be used when the interest rate is near the zero-lower bound into three types: forward

4 Keynes (1936) is the first to mention the zero-lower bound for interest rates, due to the fact that it individuals can also keep cash. Due to costs for holding and protecting currency, interest rates can be slightly under 0 (examples are Denmark (2012), Sweden (2015, Switzerland (2015)). However, we follow Fawley & Neely (2013) in using zero as an approximation for the lower bound on interest rates.

(9)

9 guidance, LSAPs, and QE. QE focuses on expanding the balance sheet by traditional assets, while LSAPs can be a mix of both traditional and non-traditional assets. Thornton (2014) argues that QE and LSAPs are the same, but that the Fed used the term LSAPs for its QE policies. Ugai (2007) characterizes QE as a policy approach consisting of three features: (i) targets for the amount of bank reserves, (ii) a commitment, conditioned on e.g. the inflation rate, to maintain these levels in the future, and (iii) increased government bonds purchases.

2.2 Transmission Channels

The motivation for QE consists of two parts (Palley, 2015). The first is that if the central bank buys government bonds, it supplies money for the deficit. Second is that the central banks help to bring down interest rates on bonds which it is buying. Therefore, the long-term rates of government yields can be pushed downwards through the purchases of longer-term assets (Palley, 2015; Kobayashi et al., 2006). By reducing the longer-term yields, consumptions and investments will be stimulated (Thornton, 2014).

QE is said to affect long-term yields through different channels, but two main channels are recognized (Belke et al., 2017). These main channels are the signaling channel and the portfolio balance channel (Borio and Disyatat, 2010). Also, Fawley and Neely (2013) recognize that asset purchases can lower long-term real interest rate through the signaling channel and the portfolio balance channel. There is a general consensus that unconventional measures have had an effect on bond yields, however it is debatable which of the transmission channels was more important for these effects as it is hard to separate them (Borio and Zabai, 2016). The monetarist approach states that the focus is on the portfolio-rebalancing channel (Girardin, and Moussa, 2011). Also, Gagnon et al., (2011), emphasize the importance of this channel relative to the signaling channel, as well as the main monetary policymakers (Joyce, Liu, and Tonks, 2017). Christensen and Rudebusch (2012) however argue that the importance of each channel can be different in every country and may depend on institutional structures and communication policies of the central banks.

2.2.1 The Portfolio Balance Model

The portfolio balance model follows the preferred-habitat theories of Modigliani and Sutch (1966) (as cited in Joyce et al., 2012). When the central bank purchases bonds, the relative supply of assets held by the private sector is changed, and equilibrium effects start changing relative yields (Christensen and Rudebusch, 2012). Because of the purchases, there is a

(10)

10 smaller supply of bond assets which drives up the price. Investors will now tend to replace these bonds by relatively cheaper bonds of similar quality, which drives up the prices of those debts as well (Bauer and Neely, 2014). Therefore, announcements by the central bank push up the bond prices of the bonds bought and the prices of similar but imperfect substitutes, which pushes down the term premiums and thus the bond yields, both for the national as international substitutes (Christensen and Rudebusch, 2012; Bauer and Neely, 2014). This increases the demand for riskier assets, in particular corporate equity and real estate (Priftis and Vogel, 2016). Thus, the portfolio balance effects work in two ways (Fratzscher et al., 2018). First, the purchases affect the yields by an increased demand, which drives up the bond price and thus lowers yield premia and yields. Second, market participants who sold their bonds will rebalance their portfolio with riskier assets (Joyce et al., 2017).

Among others, policymakers in the US (Bernanke, 2010; Yellen, 2011) and in the UK have put an emphasis on the portfolio balance channel (Joyce et al., 2017). It has been highlighted by the monetary policymakers as the main channel through which QE policies work (Michael et al., 2017). Also in the majority of empirical studies, the importance of the portfolio balance model has been stressed, finding evidence that asset purchases mainly affect the bond yields and asset prices because they reduce the risk premiums through this channel and concluding that preferred habitat theories hold (Doh, 2010; Gagnon et al., 2011; D’Amico and King, 2013). Breedon et al., (2012), who estimate the effects of QE directly on bond yields through a macro-finance yield curve, conclude that their “evidence is contributing growing consensus that QE is, indeed, effective in terms of influencing longer-term bond yields through a portfolio balance effect” (Breedon, Chadha, and Waters, 2012, pp. 727).

However, the channel is still under debate, as Woodford (2012) argues that portfolio-balance effects “do not exist in a modern, general-equilibrium theory of asset prices” (Woodford, 2012, pp. 61) and finds his evidence in first reviewing theoretical arguments, after which he turns to the evidence of studies regarding the effectiveness. Thornton (2014) provides a reduced-form methodology of research done by Gagnon et al. (2011) and Krishnamurthy and Vissing-Jorgensen (2011), using alternative debt measures and checking for common trends in their time series. After accounting for the common trend, he concludes that there is no time series evidence for any effects on the long-term yields through the portfolio balance channel.

(11)

11 2.2.2 The Signalling Channel

The signalling channel implies that announcements of the central bank can, by signalling a path for future policy interest rates, affect long-term interest rates (Gagnon et al, 2011; Woodford, 2012; Bauer and Neely, 2014). The monetary policymaker signals that it expects policy rates to remain low for a longer period of time, this can be done by for example revealing its projections on the economic outlook. By undertaking QE, the policymaker demonstrates its commitment to these objectives, which helps maintaining credibility (Bowdler and Radia, 2012). Therefore, the signaling channel is linked to the average expected level of short-term interest rates over the duration of the bond (Christensen and Rudebusch, 2012). Longer-term yields are lowered because the central bank signals that the short-term future policy rates will be lower than was expected by the market (Fratzscher et al., 2018). Bauer and Rudebusch (2014) emphasize the importance of the signalling channel, providing evidence for its significance by modelling the declines in yields and separating changes in risk premia and expected future short-term rates.

Krishnamurthy and Vissing-Jorgenson (2011) argue that the signalling channel, as well as the portfolio balance channel, significantly lowered all bond yields but emphasize that the signalling channel was relatively more important (as cited in Joyce et al., 2012). Also Christensen and Rudebusch (2012) find reductions in the bond yields using an event study, their results stress the importance of the signalling channel in the transmission of QE. Bauer and Neely (2014) test the importance of both the signalling channel and the portfolio balance channel. To test this, they use term structure models to evaluate the importance of LSAP channels in the total effect of the Fed’s asset purchases on bond yields. They find that in the US, the signalling channel contributed for between 45% and 90% to the total drop in long-term yields, and conclude that here the signalling channel was more important than the portfolio balance channel. In Japan however, the response in short-term future rate expectations was weak and insignificant, so Bauer and Neely (2014) conclude that the signalling channel was insignificant in Japan. Therefore, we follow Christensen and Rudebusch (2012) in their conclusion that the relative importance of portfolio balance channels and signalling can be different among countries.

(12)

12

2.3 Evidence on the Effectiveness of QE

Effectiveness of the QE is measured by the impact it has had on long-term yields, as this is stated as the goal in numerous studies (e.g. Krishnamurthy and Vissing-Jorgensen, 2011; Thornton, 2014; Belke, Gros, and Osowski, 2017), but also by the policymakers of the central banks themselves in the US and UK (Joyce et al., 2017). Therefore, in this article I follow the definition of the goal of QE of Belke et al. (2017), who states that the main purpose of QE policies is lowering the long-term interest rates.

The studies can be divided roughly into two sorts: the event studies and studies using econometric model regressions (Martin and Milas, 2012). Examples of empirical studies using the event study method are Bernanke et al., (2004), Gagnon et al. (2011), Krishnamurthy and Vissing-Jorgenson (2011), Christensen and Rudebusch (2012), Bauer and Rudebusch (2014), Neely (2015) and Fratzscher et al. (2018). In general, these event studies conclude that QE has significantly lowered long-term yields5 (Thornton, 2017). Also, Borio and Disyatat (2010) show evidence that LSAPs have lowered long-term yields but argue that these effects are possibly not lasting: “they seem to be subject to ‘diminishing returns’, once the surprise factor wears off” (Borio and Disyatat, 2010, pp. 86). Wright (2012) concludes the same, stating that “monetary policy shocks are estimated to have effects on both long-term Treasury and corporate bond yields that are generally statistically significant, although the effects fade fairly fast over the subsequent months” (Wright, 2012, pp. 464).

D’Amico and King (2013) deviate from the event studies. They find that purchases significantly lowered bond yields and that asset purchases can reduce longer-term interest rates.6 Also, Doh (2010), who uses a preferred-habitat model which was developed by Vayanos and Vila (2009), finds that LSAPs can decrease the term premia of long-term bond yields. The general view of the literature is that asset purchases lower yields and longer-term interest rates (Joyce et al., 2012).

5 Further details about the studies which are important for the effects in Japan, the UK and the US will be mentioned in section 2.4, which focusses on each country specifically.

6 they analyze changes in yields of bonds that were purchased by the Fed, and changes in yields of bonds with similar maturities but were not purchased by the Fed.

(13)

13

2.4 Unconventional Monetary Policies in Japan, the UK and the US

The Bank of Japan was the first implementing QE in 2001. Prior to this the BOJ had a zero-interest policy. In 2007, the Bank of England and the Federal Reserve also started cutting their policy rates towards the zero bound (Fawley and Neely, 2013), which is how they started their expansionary policy (Bowdler and Radia, 2012). Figure 1 illustrates the main policy rates of the BOE, BOJ and Fed and shows how all the interest rates approached the ZLB after 2007.

Figure 1: Policy Rates in Japan, the UK and the US

Note: the main policy rates for the BoE, BOJ, and Fed are, respectively, the official Bank rate, the uncollateralized overnight call rate, and the federal funds target rate.

Source: FED, BOJ, BOE.

Just as happened before in Japan, cutting the policy rates now also was not expansionary enough in the UK and the US. After the fall of the Lehman Brothers, the Fed and BoE implemented QE in 2008 and 2009 respectively (Fawley and Neely, 2013). However, they argue that QE programs were fundamentally different in the UK and in the US than in Japan, as they focused more on purchasing bonds. Japan, on the contrary, focuses more on the bank market (Fawley and Neely, 2013). The Bank of Japan is thus the only central bank that specifically targets its bank reserves (Borio and Zabai, 2016). Though the programs and countries are fundamentally different, the policies have all led to significant increases in the balance sheet sizes of the central banks (Joyce et al., 2012). The aim of the three central banks is the same: all central banks purchased long-term assets, reducing the

-1.00 0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00 8.00 19 96- 10-01 19 98- 03-01 19 99- 08-01 20 01-01 20 02- 06-01 20 03- 11-01 20 05- 04-01 20 06- 09-01 20 08- 02-01 20 09- 07-01 20 10- 12-01 20 12- 05-01 20 13- 10-01 20 15- 03-01 20 16- 08-01 20 18- 01-01 FED BOJ BOE

(14)

14 amount of long-term assets held by the public and thus decreasing the long-term yields to stimulate the economy (Fawley and Neely, 2013).

Now, the different methods of implementing unconventional policies and the effects of empirical researches regarding these methods will be worked out for each country. The focus will be mainly of actions and announcements which had expansionary effects on the central banks’ balance sheets, though for completeness of information also some details will be given on unconventional operations which did not alter the size of the balance sheet.7

2.4.2 Japan

2.4.2.1 Overview of Unconventional Measures

The Bank of Japan started in April 1999 with what Ueda (2012a) describe as the first type of measures available when an economy is at the ZLB, the zero-interest rate policy (ZIRP). The interest rate had already been zero (or even lower) in the period before, but now the BOJ committed to holding this interest rate “until deflationary concerns were dispelled” (as cited in Ueda, 2012a, pp. 4). The ZIRP was lifted again in August 2000. Following the recession in 2001 however, the call rate, which was increased to 0.25 in August 2000, was again reduced to 0.15 in February 2001. As this was not enough to stop deflation (Woodford, 2012), BOJ adopted its first implementation of QE8 in March 2001.

Their first QE policy consisted of 3 pillars (Oda and Ueda, 2007; Ueda, 2012c): (i) the current account balance (CAB) was used as the main operating target by the BOJ; (ii) the BOJ committed to supplying liquidity “until the rate of change of the core CPI (nationwide, excluding perishables,) becomes zero percent or higher on a sustained basis” (Oda and Ueda, 2007, pp. 307); (iii) the BOJ increased its amount of JHB purchases from time to time to inject liquidity. The QE policy implemented in Japan, in March 2001, replaced the overnight interest rate target by the quantity targets for the CAB (Woodford, 2012). Borio and Zabai (2016) argue that the Bank of Japan is the only central bank that specifically targets bank reserves. The target was set on 5 trillion yen on the CAB, which meant an increase of around 1 trillion yen (BOJ announcement, 2001). Setting this target was expected to drive the call rate from 0.15 percent to 0 (Fawley and Neely, 2013). Between 2001 and 2004, the BOJ expanded its target on the CAB around 10 times, and targets were mainly achieved by through the purchases of Japanese government bonds (JGBs). It was pre-announced by the

7 For example on ‘Operation Twist’, implemented by the Fed in 2013, in which purchases of long-term bonds and sales of short-term assets were of equal amounts.

(15)

15 BOJ what amount of JGBs would be purchased each month. The purchases started at 0.4 trillion yet in March 2001 and grew to 1.2 trillion yen in May 2004 (Oda and Ueda, 2007). Figure 2 shows the purchases of JGBs by the BOJ over roughly the past 20 years. The first quantitative easing period was terminated in March 2006, when the CPI was above zero (Woodford, 2012)9. The CAB decreased by at least two-thirds (Ueda, 2012b).

Figure 2: BOJ purchases of JGBs (billions of Japanese Yen)

Source: Datastream

Figure 3: Coupon Purchases of the BOJ

Source: Datastream

Note: RA are CP purchases bought under repurchase agreements.

9 Woodford (2012) refers by the first QE policy in Japan, which took place between 2001 and 2006, to the pure QE policy. 0 2000 4000 6000 8000 10000 12000 14000 01 /01/ 1998 01 /02/ 1999 01 /03/ 2000 01 /04/ 2001 01 /05/ 2002 01 /06/ 2003 01 /07/ 2004 01 /08/ 2005 01 /09/ 2006 01 /10/ 2007 01 /11/ 2008 01 /12/ 2009 01 /01/ 2011 01 /02/ 2012 01 /03/ 2013 01 /04/ 2014 01 /05/ 2015 01 /06/ 2016 01 /07/ 2017 Purchases of JGBs 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 01 /01/ 1998 01 /02/ 1999 01 /03/ 2000 01 /04/ 2001 01 /05/ 2002 01 /06/ 2003 01 /07/ 2004 01 /08/ 2005 01 /09/ 2006 01 /10/ 2007 01 /11/ 2008 01 /12/ 2009 01 /01/ 2011 01 /02/ 2012 01 /03/ 2013 01 /04/ 2014 01 /05/ 2015 01 /06/ 2016 01 /07/ 2017 CP Purchases CP Purchases - RA

(16)

16 In 2008, following the world financial and economic crisis, Japan had to implement another QE policy (Ueda, 2012c). By the end of 2008, the BOJ re-introduced its ZIRP (Kawai, 2015). Since 2009, four major monetary programs have been announced by the BOJ. The QE program, stopped in 2006, was continued in October 2010 (Kawai, 2015) and consisted of asset purchases of around 35 trillion yen with the goal to lower the long-term interest rates (Pelizzon et al., 2017). It is also referred to as the Comprehensive Monetary Easing (CME) policy and was a response to the deflation (Berkmen, 2012). The CME included the expansions of commercial paper (CP) purchases (Ueda, 2012c), which is visualized in figure 3.

In april 2013, governor Kuroda increased the QE program and announced purchasing long-term government bonds, which was called the QQE policy (Iwatsubo and Taishi, 2017). The bond purchase program of JGBs was almost doubled to 50 trillion yen per year. Since October 2014, the QQE program was expanded (QQE2) and the bond purchase program was further increased to 80 trillion yen, which was around 30 trillion yen more than the previous amount. The goal was to “decrease interest rates across the entire yield curve” (Pelizzon et al., 2017, pp. 11). At the end of January 2016, QQE with a negative interest rate was introduced by the BOJ. The bank continued purchasing JGBs, with an increased annual amount of 80 trillion yen. In June 2017, the BOJ’s holdings of JGBs reached an amount of 437 trillion, corresponding to approximately 81% of Japan’s GDP (Pelizzon et al., 2017). Figure 4 shows how much the stock of JGBs has grown since the beginning of this century.

Figure 4: JGB holdings by the BoJ

Source: Datastream 0 500000 1000000 1500000 2000000 2500000 3000000 3500000 4000000 4500000 01 /07/ 1996 01 /08/ 1997 01 /09/ 1998 01 /10/ 1999 01 /11/ 2000 01 /12/ 2001 01 /01/ 2003 01 /02/ 2004 01 /03/ 2005 01 /04/ 2006 01 /05/ 2007 01 /06/ 2008 01 /07/ 2009 01 /08/ 2010 01 /09/ 2011 01 /10/ 2012 01 /11/ 2013 01 /12/ 2014 01 /01/ 2016 01 /02/ 2017 JGBs stock

(17)

17

2.4.2.2 Empirical Studies on QE in Japan

Bernanke et al. (2004) conducted an event study analysis and estimated a VAR (vector autoregression) to model the term structure of interest rates, to see whether the QE policy in Japan had significant effects. They find that the bond yields were lower during the QE period in Japan than would have been predicted, which suggests that QE was effective. However, Oda and Ueda (2007) do not succeed in finding any significant effects of JGB purchases by the BOJ on either future short rates or risk premiums. Ugai (2007) looks at the entire QE period from 2001 to 2006, dividing effects of balance sheet expansions into the portfolio balance effect and the signalling effect, and concludes that expanding the CAB, through the signalling effect, lowered the long-term interest rates. On the contrary, Honda, Kuroki, and Tachibana (2007), using a VAR methodology, find that the portfolio rebalancing effects dominate the first QE period. They also find that expanding the BOJ’s CAB had significant effects on the long-term interest rates. Kobayashi et al. (2006) conclude with an event study analysis (dates between 2001 and 2004) that QE has been successful in lowering the long-term rates. Ueda (2012b) divides the QE policy 2001-2006 into three types, and using this typology looks for evidence that longer-term yields were significantly lowered. He argues that there is no evidence that only expanding the balance sheet10 has any effects on the long-term yields. Berkmen (2012) states that most papers found evidence on reduced yields by QE, Woodford (2012) argues that those reductions measured in the long-term bond yields cannot be attributed only to the QE policy, as this policy was accompanied by strong forward guidance.

More recent research on the effects of monetary policy after the 2007-2008 world crisis is done by among others Lam (2011). Using an event-study approach, the author finds that the expansionary measures had a significant impact on decreasing the bond-yields. Also Kawai (2015) comes to this conclusion and states that long-term yields were lowered by QQE. Pelizzon et al. (2017) find that the large purchases of JGBs, starting in 2010, lead to a lower bond yield on JGBs. Bauer and Neely (2014) are one of the few researchers looking at transmission channels after 2010, they conclude that both transmission channels barely had any effect on the bond yields in Japan.

10

(18)

18 2.4.3 United Kingdom

2.4.2.1 Overview of Unconventional Measures

The BOE asset purchases can be divided roughly into two periods, the first one being from March 2009 until February 2010, after which was announced that the QE policy was ended. The second round of purchases started in October 2012 (Fawley and Neely, 2013). The last period of QE (QE3) started in August 2016.

The Bank of England started implementing its QE policy officially on March 5th, 2009 (Lyonett and Werner, 2012). However, its balance sheet had already increased significantly in September 2008, after the fall of Lehman Brothers, under the Special Liquidity Scheme (Breedon et al., 2012). In beginning 2009, the Asset Purchase Facility (APF) was set up and in March that year, the Bank of England aimed to purchase an amount of £75 billion the next three months through this facility (Breedon et al., 2012). By then the Bank rate had also decreased to 0.5% (BoE, 2010), which meant the interest rate was reaching its ZLB, but this had not been enough to increase economic activity (Christensen and Rudebusch, 2012). The APF target was increased multiple times, to £125 billion on the 7th of May, £175 billion on August 6th and again on November 5th, 2009 to £200 billion. In February 2010 the Bank of England decided to temporarily suspend its QE policy. The size of the Bank’s balance sheet, proportional to GDP, was by then three times as big as before the beginning of the crisis in 2007 (Joyce, Lasaosa, Stevens, and Tong, 2011). The most purchases done were gilt purchases (namely £198 billion of the £200 billion purchases; Joyce et al., 2011) and those are visualized in figure 5.

The Bank of England re-introduced QE (also called QE2) in October 2011 (Breedon et al., 2012). It consisted again of £75 billion in government bonds purchases. QE3 was announced in February 2012 and consisted of another £50 billion in asset purchases (Martin and Milas, 2012), which was again raised with £50 billion in July 2012 (Lyonnet and Werner, 2012).

The most recent introduction of QE was in August 2016, when the Monetary Policy Committee (MPC) decided to expand its QE porgramme by an addition £70 billion, of which £60 billion were gilts. The total of the purchases was by then £445 billion. The interest rate was again cut to 0.25%, which was lower than the base rate cut in March 2009 (which was 0.5%) (Haldane, Roberts-Sklar, Wieledak, Young, 2016).

(19)

19

Figure 5: Gilt Purchases by the BoE

Source: Datastream.

The expansion in the BoE’s balance sheet is mainly caused by the purchases of Gilt Yields (Fawley and Neely, 2013), therefore figure 5 gives a good overview of the expansions of the balance sheet during the QE period in the UK. All 3 periods are visible: the start of the QE in March 2009, the continuation with QE2 in October 2011 and the most recent period of QE, QE3 in August 2016.

2.4.2.2 Empirical Studies on QE in the UK

Meier (2009) tests the first four months of the QE policy in the UK using an event study analysis and concludes that the UK gilt yields were lowered because of the QE. Also Christensen and Rudebusch (2012) find lower long-term yields in the UK by using an event study. They conclude that these declines were mainly because of reductions in the term premium and do not find significant evidence on a signaling channel. Joyce et al. (2011) find that the asset purchases lowered the long-term government bond yields, mainly through a portfolio balance channel. The medium to long-term gilt yields were found to be almost 100 basis points lower because of QE, based on an event study and survey data. Breedon et al. (2012) test the QE programme in 2009 and 2010 using a term structure model and conclude that the QE significantly lowered long-term government bond yields through the portfolio balance channel. They also comment on the event studies done by Meier (2009) and Joyce et al. (2011), by citing Caglar, Chadha, Warren and Watters (2011): “event study methodology may have overestimated the effects because of the dominant, possible exaggerated, impact of the first rather than the subsequent six announcements” (as cited in Breedon, Chadha, and

0 50000 100000 150000 200000 250000 300000 350000 400000 450000 500000 01 /03/ 2009 01 /09/ 2009 01 /03/ 2010 01 /09/ 2010 01 /03/ 2011 01 /09/ 2011 01 /03/ 2012 01 /09/ 2012 01 /03/ 2013 01 /09/ 2013 01 /03/ 2014 01 /09/ 2014 01 /03/ 2015 01 /09/ 2015 01 /03/ 2016 01 /09/ 2016 01 /03/ 2017 01 /09/ 2017 01 /03/ 2018 Gilt Purchases

(20)

20 Waters, 2012, pp. 704). The channel which the emphasis is on in the UK is the portfolio balance channel (Joyce et al., 2017), which was also what the MPC announced (Bowdler and Radia, 2012).

2.4.4 United States

2.4.3.1 Overview of Unconventional Measures

As was seen in the sections summarizing the QE in Japan and the United Kingdom, prior to the implementation of QE in the United States was the lowering of the policy rate (the short-term federal funds rate) towards the ZLB (Doh, 2010; Wright, 2012). As the policy rate could not be lowered anymore to stimulate the economy further, the FOMC started implementing unconventional policies. While the official term is QE, the Fed used the term LSAPs for this (Thornton, 2014). The aim of the Fed by implementing those LSAPs was to support

mortgage markets (Song and Zhu, 2018) and to lower longer-term yields to stimulate investments and consumption (Thornton, 2014). The Federal Reserve focused on the portfolio balance channel as the main transmission mechanism for LSAPs (Wright, 2012).

The first LSAPs started in November 2008, and the Fed mainly concentrated on the longer-term MBS (Nelson, 2013). The purchases were highly increased in March 2009, when the Fed announced th purchase an additional $100 billion in agency debt, $750 billion in MBS and $300 billion in Treasury securities (Belke et al., 2017). These purchases were completed in March 2010 (Nelson, 2013), and was later named QE1 (Shogbouy and Steeley, 2017).

As the financial crisis worsened, The Fed announced QE2 in November 2010, which consisted of the purchases of $600 billion in longer-term Treasury securities and lasted until July 2011 (Song and Zhu, 2018). In August 2010, principal payments from agency debts and MBS were re-invested in longer-term Treasury securities for an amount of $180 billion (Nelson, 2013). All these programs resulted in a significant increase in the Fed’s balance sheet size (Woodford, 2012). The purchases which started in 2008 and continued until early 2010 summed up to an amount of $1.725 trillion (Gagnon et al., 2011).

In September 2011 the Maturity Extension Program (MEP) was launched, in which the Fed purchases Treasuries with 6-10 year maturities and sold an equal amount of Treasures with short-term maturities11 (Fratzscher et al., 2018). QE3, another around of LSAP, was announced in September 2012, under which the Fed purchased an amount of MBS and

(21)

21 Treasury bonds conditional on the labour market. These purchases amounted to $40 billion per month in MBS, starting September 2012, and $45 billion per month in Treasury bonds, starting December 2012. These purchases were gradually decreased and stopped between 2013 and 2014 (Fratzscher et al., 2018). Figure 6 visualizes the purchases of agency debts and MBS which were carried out by the Fed. Overall the balance sheet of the Fed increased with approximately $3.5 trillion (Belke et al., 2017).

Figure 6: Credit Easing Policies carried out by the FED

Source: Datastream

2.4.3.2 Empirical Studies on QE in the US

The ultimate goal of the Fed was to reduce medium and long-term US interest rates to stimulate real economic activity (Bauer and Neely, 2014). Bernanke et al. (2004) state that especially in the United States, large-scale asset purchases are able to affect the target yields. This conclusion is supported by various other studies. Doh (2010) uses a model developed by Vayanos and Vila (2009) and concludes that large-scale purchases of assets can decrease the long-term interest rates. He also states that asset purchases on a large scale are more effectively than other policies when the policy rate is near the ZLB. Wright (2012) uses a VAR model to test the effects of the policy shocks on different longer-term interest rates. He states that the effects are significant, but that the effects diminish quite fast in the months after. Neely (2015) argues that the specification carried out by Wright (2012) underestimated the lasting of the effects.

0 200000 400000 600000 800000 1000000 1200000 1400000 1600000 1800000 2000000 01 /10/ 2008 01 /06/ 2009 01 /02/ 2010 01 /10/ 2010 01 /06/ 2011 01 /02/ 2012 01 /10/ 2012 01 /06/ 2013 01 /02/ 2014 01 /10/ 2014 01 /06/ 2015 01 /02/ 2016 01 /10/ 2016 01 /06/ 2017 01 /02/ 2018

(22)

22 Krishnamurthy and Vissing-Jorgensen (2011) use an event study analyses for QE1 and QE2 and find that there are significant declines on the long-term assets that were purchased by the Fed. Gagnon et al. (2011) carry out an event study as well as a time series method to investigate the effects of QE1, both show significant effects on the long-term yields, and emphasize the portfolio balance channel. Other event studies are carried out by among others Christensen and Rudebusch (2012), who follow Gagnon et al. (2011) in their approach and estimate similar effects on long-term yields, which are caused mainly by lower future short-term rates. Another often-mentioned research is one of D’Amico and King (2013), who carry out a panel-data approach to estimate the effects on long-term yields. Their results conclude that LSAPs are effective and that there is evidence on a portfolio balance channel.

Fratzscher et al., (2018) use a different approach to test for effects, namely a regression including dummy variables for the QE announcements, and come to the same conclusion as D’Amico and King (2013), for both the effects of QE on long-term yields as on the existence of a portfolio balance channel. Gagnon (2016) concludes that studies in the US show that QE can lower bond yields significantly. He estimates that, based on various other studies, QE programs in the US reduced the 10-year yield by 1.2 percentage points (Gagnon, 2016, pp.4). Belke et al. (2017) however, state the opposite as they find no evidence of QE1 on long-term yields when looking at trans-international relationships, and conclude that QE has been ineffective.

The transmission channel of QE, of which the portfolio balance channel is highlighted among others Gagnon et al. (2011), is still under debate as Bauer and Neely (2014) conclude that LSAPs had a significant effect on international bond yields through the signaling channel. Also, Bauer and Rudebusch (2014) find a significant signaling channel for QE1, by looking at dynamic term structure models. Thornton (2014) argues that the finding of Gagnon et al. (2011) is caused by a common trend in data and concludes that there is no evidence of a portfolio balance effect in the US, nor a support for effectiveness of QE on the long-term yields. He also argues that event studies cannot be used as a proof for the effectiveness on QE (Thornton, 2017), which makes the effect of QE on long-term yields in the US an unsolved question.

(23)

23

3. Data and Methodology

In this section I will explain the dataset which I used, including the dependent and independent variables and the chosen timeframe. The methodology will be explained in section 3.2, which includes among others the regression model used and econometric considerations when executing the regressions.

3.1 Data

3.1.1 The Countries

With Japan being the first one to implement QE, and the BOE and Fed being the most mentioned countries in studies on QE policies, those three countries are chosen and compared.

3.1.2 Dependent Variable

The dependent variable will be the long-term yield, as the various studies mentioned in the previous section suggest that lowering the long-term yields is the goal of balance sheet policies, and the main reason it was carried out by the central banks. To estimate the long-term yields, I will use the 10-year government bond yields. The data comes from Datastream. By using the long-term yield as the independent variable, I follow the argumentation of Borio and Zabai (2016), who state that “the behaviour of government bond yields is the most telling example. Such yields reflect the combined influence of central banks and market participants” (Borio, Zabai, 2016, pp. 21). Furthermore, as mentioned in Belke et al. (2017), the most studies focus on the effects on long-term yields, which makes it easy to compare estimates.

3.1.3 Independent Variables

The variables used in the regression are the unemployment gap, the rate of inflation (CPI), the inflation disagreement and a QE proxy. To capture uncertainty on economic fundamentals in the regression, the unemployment gap, rate of inflation (CPI) and the long-run inflation disagreement are included following the argumentation of Gagnon et al. (2011).

The unemployment gap

The unemployment gap is measured by the actual rate of unemployment, minus the non-accelerating inflation rate of unemployment (NAIRU). Though there are ongoing discussions on whether the NAIRU is a useful estimator, I follow the argumentation of Ball and Mankiw (2002) who argue that it is a “synonym for the natural rate of unemployment” (Ball, Mankiw,

(24)

24 2002, pp. 115). As the NAIRU is calculated for all three countries in are dataset, data can be retrieved from Datastream. The same goes for the actual rate of unemployment. The NAIRU is estimated yearly, but as the differences between the estimations are quite small and the NAIRU changes slowly over time, I use the yearly estimated NAIRU on a quartile basis.

Inflation

Core CPI inflation is retrieved from Datastream for all three countries and is one of the measures for economic uncertainty, that may affect the long-term yields.

Long-run inflation disagreement

As a proxy for expected inflations, the 5-year forward inflation-linked swap rates are used. Here I deviate from Gagnon et al. (2011), as they use the five-to-ten-year-ahead inflation expectations, which are reported by the Michigan Survey of Consumers. The reason for this is that these kinds of surveys are not done in Japan or in the UK, at least not in the sense that they can be used in the regression analysis. Each country however has data available on the 5-year forward swap rates, which is also used as the estimated inflation rate in the ECB and the Fed.

Size of the central bank’s balance sheet over GDP

As described by Priftis and Vogel (2016), QE is a monetary policy in which the central bank’s balance sheet is increased, with the aim of lowering long-term yields. Also, Aksoy and Basso (2013) state that banking asset growth correlates with lower risk premia. Therefore, I take the size of the central bank’s balance sheet over GDP as a proxy for QE. The sizes of the balance sheets in Japan, the UK and the US are visualized in figure 7. In the graph, it is clearly visible that the balance sheets increased significantly in all three countries. Though some remarks have been made on the effects of a central bank balance sheet expansion in previous research (Ueda 2012a, b), the effects of the expansion in a balance sheet have, to my knowledge, never been estimated. As many studies state, and what also the definition is that we follow, is that QE is a policy in which the central bank’s balance sheet is significantly expanded (Bernanke and Reinhart, 2004; Borio and Disyatat, 2010).

(25)

25

Figure 7: Expansions in the CB balance sheets

Source: BOJ, BOE, Federal Reserve

Timeframe

Data was not available for the same time period of all three countries. For the Bank of England, for example, quarterly data on balance sheet sizes were only available between 2006 and 2016. However, to make the regression as inclusive as possible, all available quarterly data was taken for each country. For Japan, this means a period between 1998 and 2017. The data for the Bank of England is available between 2006 and 2016 and for the US between 2003 and 2017. Though the time frames are different, every dataset for the country contains all QE policy movements executed in that country. All data has been retrieved from the Federal Reserve Bank of St. Louis, the Bank of Japan, the Bank of England and Datastream.

3.2 Methodology

3.2.1 The Regressions

As mentioned before, most studies that find significant effects of QE on long-term yields follow the event study approach (Priftis and Vogel, 2016). Examples of event studies are among others Gagnon et al. (2011), Krishnamurthy and Vissing-Jorgensen (2011), Bauer and Rudebusch (2014) and Wright (2012). Also Thornton (2017) concludes that all event studies

0.000% 10.000% 20.000% 30.000% 40.000% 50.000% 60.000% 70.000% 80.000% 90.000% 100.000% 19 98q 2 19 99q 1 19 99q 4 20 00q 3 20 01q 2 20 02q 1 20 02q 4 20 03q 3 20 04q 2 20 05q 1 20 05q 4 20 06q 3 20 07q 2 20 08q 1 20 08q 4 20 09q 3 20 10q 2 20 11q 1 20 11q 4 20 12q 3 20 13q 2 20 14q 1 20 14q 4 20 15q 3 20 16q 2 20 17q 1 20 17q 4 Japan UK US

(26)

26 agree on significant effects on the long-term yields, but he states that event studies “cannot and should not be cited as evidence of QE’s effectiveness” (Thornton, 2017, pp.69).

However, also other approaches on the effects have been done, for example by Gagnon et al.12 (2011) and D’Amico and King (2013). The model I will use is based on the regression analysis executed by Gagnon et al. (2011), to estimate the effects of an expansion in the balance sheet. The regression analysis mainly captures the portfolio balance effects of QE (Gagnon, 2016). Though opinions in empirical studies are mixed, the portfolio balance effect seems to be significant, at least in the US (Doh, 2010; Gagnon et al., 2011; Breedon et al., (2012); D’Amico and King, 2013). Bauer and Neely (2014) argue that in Japan no portfolio balance effects are visible. Thus, by using this regression, it will become clear whether portfolio balance effects exist and if there are significant differences between the three countries.

The linear regression is as follows: 𝑡𝑡𝑝𝑝𝑡𝑡10 = 𝛽𝛽0+ 𝛽𝛽1∗𝐶𝐶𝐵𝐵𝐺𝐺𝐺𝐺𝐺𝐺𝑖𝑖𝑡𝑡

𝑖𝑖𝑡𝑡+ 𝛽𝛽2∗ 𝑈𝑈𝑈𝑈𝑈𝑈𝑝𝑝𝑖𝑖𝑡𝑡+ 𝛽𝛽3∗ 𝐶𝐶𝐺𝐺𝐼𝐼𝑖𝑖𝑡𝑡+ 𝛽𝛽4∗ 𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑖𝑈𝑈𝑈𝑈𝑟𝑟𝑖𝑖𝑡𝑡

To be able to compare the different results for the 𝛽𝛽4 estimations of the different countries, I use a dummy regression. The size of the central bank’s balance sheet over in Japan and the UK are taken as the dummy variables, while the US is used as the reference country.13 The regression then becomes:

𝑡𝑡𝑝𝑝𝑡𝑡10 = 𝛽𝛽0+ 𝛽𝛽1∗ 𝑈𝑈𝑈𝑈𝑈𝑈𝑝𝑝𝑡𝑡+ 𝛽𝛽2∗ 𝐶𝐶𝐺𝐺𝐼𝐼𝑡𝑡+ 𝛽𝛽4∗𝐺𝐺𝐺𝐺𝐺𝐺𝐶𝐶𝐵𝐵𝑡𝑡

𝑡𝑡∗ 𝐽𝐽𝐺𝐺𝐽𝐽 + 𝛽𝛽5∗

𝐶𝐶𝐵𝐵𝑡𝑡

𝐺𝐺𝐺𝐺𝐺𝐺𝑡𝑡∗ 𝑈𝑈𝑈𝑈

𝛽𝛽4 and 𝛽𝛽5 are the dummy variables, for which JPN = 1 if the QE proxy corresponds to Japan,

otherwise 0, and UK=1 if the QE proxy corresponds to the UK, otherwise 0. 3.2.2 Econometric Concerns

Problematic can be the presence of heteroskedastic errors in a regression analysis. This can be solved by using robust standard errors. To see if one needs to use robust standard errors (which allow for heteroscedasticity), one can perform a test on heteroscedasticity. The test used in this research is the Breusch-Pagan test, which gives a 𝜒𝜒2 value. All tests on heteroscedasticity can be found in appendix 4. In the results it will be clear that robust

12 In their research, they use both an event study as well as a regression analysis to estimate the effects of QE on long-term yields.

(27)

27 standard errors must be used with the dummy regression and could14 be used with the regression for Japan. As no harm is done by using heteroskedastic robust standard errors if the errors are homoscedastic (Stock and Watson, 2015), all regression results with robust standard errors are added in the appendix.

As the R2 tends to increase in general when an extra variable is added from the regression (Stock and Watson, 2015), the adjusted R2 will be added to the regression tables to see if the additional regressors tend to contribute in explaining the long-term yields. The adjusted R2 is a modified version of the R2 but does not necessarily increase when an extra variable is added to the regression. It is only available for regressions which do not assume robust standard errors.

A check for multicollinearity is also done in each regression. As long there is no perfect multicollinearity (a collinearity of 1), there are no problems for the OLS estimation theory (Stock and Watson, 2015). To prevent a ‘dummy variable trap’, which is, as described by Stock and Watson: “A problem caused by including a full set of binary variables in a regression together with a constant regression, leading to perfect multicollinearity” (Stock and Watson, 2015, pp. 819), one of the three countries is used as the ‘reference’ country. As this regression is based on the model of Gagnon et al. (2011), who use the model for the United States, the US is chosen as the reference country. I use a test on the variance inflation factor (VIF) to see if there is any problematic multicollinearity in the regressions. The results of these tests are shown in the appendix. There are problems of multicollinearity in the regressions for Japan and the US.

4. Results

4.1 Results per Country

4.1.1 Japan

The estimates of the regression results in Japan are shown in Table 1. All the estimated coefficients are significant on a 10% significance level, and most results show a higher significance level of 1%. As the unemployment gap and the CPI are both highly significant in regression (3), but less significant in regression (4), this might be due a correlation with the inflationary disagreements.

(28)

28

Table 1: Regression Results for Japan

(1) (2) (3) (4) VARIABLES Regression 1 Regression 2 Regression 3 Regression 4 CBGDP -0.0230*** -0.0279*** -0.0285*** -0.00899*** (0.00123) (0.00151) (0.00141) (0.00183) unemploymentgap -0.239*** -0.343*** 0.101* (0.0533) (0.0593) (0.0558) CPIinflation -0.0973*** -0.0406** (0.0222) (0.0169) infldisagreements 0.769*** (0.0679) Constant 0.0189*** 0.0215*** 0.0222*** 0.00843*** (0.000536) (0.000694) (0.000686) (0.00129) Observations 79 79 79 79 R-squared 0.794 0.827 0.846 0.939

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

When looking at correlations in Japan (appendix 2 table 13) one can see that the QE proxy is highly correlated with the unemployment gap and the expected inflation. This is confirmed by the test of the variance inflation factor (VIF) which gives the following output. The high VIF outcome for CBGDP (the QE proxy) shows that there is a possible problematic multicollinearity.

Table 2: VIF Estimations Japan

VIF 1/VIF CBGDP 8.27 0.1210 unemploymentgap 5.26 0.1900 infldisagreements 4.1 0.2438 CPIinflation 1.59 0.6283 Mean VIF 4.81

If the QE proxy is removed, this improves to the following results, as shown in Table 3. Here there is no problematic multicollinearity anymore, however this makes it is impossible to

(29)

29 conclude anything on the effect of the CB size on the long-term yields as the CB size is no longer included in the regression.

Table 3: VIF Estimations excluding QE proxy

VIF 1/VIF unemploymentgap 1.49 0.6707 CPIinflation 1.46 0.6842 infldisagreements 1.03 0.9671

Mean VIF 1.33

Another way to improve the multicollinearity problem, would be to check if the problem was resolved when removing the expected inflation. These results are shown in table 4. This also removes the problematic multicollinearity but makes it still possible to see the effect of the QE policy on the long-term yields. Therefore, for looking at the effects on long-term yields, I look at regression (3) of Table 1.

Table 4: VIF Estimations excluding Expected Inflation

VIF 1/VIF unemploymentgap 2.73 0.3669

CBGDP 2.08 0.4799

CPIinflation 1.49 0.6720

Mean VIF 2.10

I conclude that the size of the central bank’s balance sheet has a significant negative effect on the long-term yields. There seems to be proof for a signalling channel in the Japan, as the QE proxy is highly correlated with the expected inflation – which was measured by using short-term future rates. All other terms in regression (3) are significant, which means they can explain the movements in the long-term yields.

(30)

30 4.1.2 United Kingdom

Table 5 shows the outcomes for the regression in the UK. The regression outcomes for the UK are quite different than in Japan. The effect of the size of the balance sheet is highly significant in every regression, but none of the other variables seems to have a significant effect on the long-term yields.

Table 5: Regression Estimations for the UK

(1) (2) (3) (4)

VARIABLES Regression 1 Regression 2 Regression 3 Regression 4

CBGDP -0.165*** -0.165*** -0.164*** -0.164*** (0.00988) (0.00999) (0.0132) (0.0133) unemploymentgap 0.0503 0.0444 0.0454 (0.0777) (0.0966) (0.0978) CPIinflation 0.0104 0.00787 (0.0975) (0.0990) infldisagreements -0.0461 (0.154) Constant 0.0570*** 0.0579*** 0.0574*** 0.0589*** (0.00169) (0.00218) (0.00488) (0.00690) Observations 43 43 43 43 R-squared 0.871 0.873 0.873 0.873 Adjusted R-squared 0.868 0.866 0.863 0.86

Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Again, a test for multicollinearity is done to see if there can be any problems with highly correlated variables. In the table of correlations, there seems to be no issue. The VIF test is used to check these results, of which the results are shown in Table 6. The results of the VIF test tell that there is no problematic multicollinearity in the regression on the UK.

Table 6: VIF Estimations for the UK

VIF 1/VIF CPIinflation 2.12 0.4728 CBGDP 1.71 0.5837 unemploymentgap 1.52 0.6578 infldisagreements 1.01 0.9919 Mean VIF 1.59

(31)

31 There is a significant effect of the QE project on the long-term yields, and it is substantially higher than the effect of the QE proxy in Japan. I therefore conclude that the size of the balance sheet in the UK has a significant negative effect on the long-term yields. There seems to be no specific evidence for a portfolio balance channel or a signalling channel.

What is remarkable, is that every other variable has no significant effects on the long-term yields, as they are commonly used in regression estimations for the long-long-term yields. A reason for this might be that the UK is a relatively small, open economy. It could be that bond-yields depend more on international fluctuations in the variables. If the CPI is highly volatile in the UK, but not with for example the OECD countries, this could explain why the national CPI has no significant effect on the long-term yields. I show the quarterly CPI for the OECD countries and the UK in figure 8. It indeed indicates that inflation in the UK is a lot more volatile than the for all OECD countries. This could explain why the results are not significant in the regression. Further research on these effects I leave to future studies.

Figure 8: CPI for the OECD and UK.

Source: Datastream, OECD. 0.0000% 1.0000% 2.0000% 3.0000% 4.0000% 5.0000% 6.0000% 7.0000% 8.0000% 9.0000% 1 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 81 85 89 93 97 10 1 10 5 10 9 OECD UK

(32)

32 4.1.3 United States

Table 7 shows the variable estimations for the United States. Except for the CPI, all variables seem to significantly affect the long-term yields. Again, I look for any problems there might be with multicollinearity. The correlations between the QE proxy and the expected inflation seems to be quite high (- 0.8420), a VIF test is used to see whether this causes any problems. The results are shown in Table 8. The VIF test indeed shows that there is a high multicollinearity between the QE proxy and the expected inflation. As was the case with Japan, I check if removing either the QE proxy or the expected inflation improves the results. In table 9, the QE proxy is removed which improves the results, however again makes it impossible to say something about the effects of the CB balance sheet on the long-term yields. When removing the expected inflation from the regression instead of the QE proxy, results are improved as well and there is no high multicollinearity that influences the regression results, this is shown in table 10. Therefore, to look at the effects of the size of the balance sheet I use regression (3).

Table 7: Regression Estimations for the US

(1) (2) (3) (4)

VARIABLES Regression 1 Regression 2 Regression 3 Regression 4

CBGDP -0.123*** -0.120*** -0.124*** -0.0404*** (0.00829) (0.00842) (0.00993) (0.0113) unemploymentgap -0.0669 -0.0767* -0.144*** (0.0430) (0.0451) (0.0301) CPIinflation -0.0445 -0.0145 (0.0594) (0.0385) infldisagreements 0.552*** (0.0620) Constant 0.0501*** 0.0505*** 0.0522*** 0.0168*** (0.00138) (0.00138) (0.00259) (0.00431) Observations 60 60 60 60 R-squared 0.791 0.799 0.801 0.919 Adjusted R-squared 0.787 0.792 0.791 0.913

Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

The p-value of the QE proxy is significant at a 1% level, which proves that there is a negative effect on the long-term yields because of the size of the balance sheet. This would mean that if the balance sheet increases, the long-term yields decrease. Inflation, however,

(33)

33 has no significant effect on the long-term yields. This is different than what Gagnon et al. (2011) estimate from their regression, who get significant effects at the 1% level of the CPI on the 10-year Term Premium. The unemployment gap is not significant in regression (2) but is significant at a 10% level in regression (3) and highly significant in regression (4). While my results show a negative effect of the unemployment gap on the 10-year yields, Gagnon et al. (2011) note positive effects of the unemployment gap.

Table 8: VIF Estimations for the US

VIF 1/VIF CBGDP 4.56 0.2195 infldisagreements 3.77 0.2656 CPIinflation 1.6 0.6236 unemploymentgap 1.23 0.8118 Mean VIF 2.79

Table 9: VIF Estimations for the US excluding QE

VIF 1/VIF CPIinflation 1.39 0.7211 infldisagreements 1.20 0.8299 unemploymentgap 1.17 0.8552

Mean VIF 1.25

Table 10: VIF Estimations for the US excluding Expected Inflations

VIF 1/VIF CPIinflation 1.59 0.6284

CBGDP 1.46 0.6860

unemploymentgap 1.15 0.8660

(34)

34

4.2 Dummy Regression

Now, the results for the regression which includes dummy coefficients for countries are shown. As there was a significant result on the test for heteroscedasticity (see appendix 4) robust standard errors are used in the regression.

Table 11: Dummy Regression Estimations

(1) VARIABLES Regression 1 unemploymentgap -0.315*** (0.0510) CPIinflation 0.119** (0.0532) infldisagreements 0.434*** (0.0574) jpncb2_b4 -0.0239*** (0.00302) ukcb2_b4 -0.0588*** (0.0101) Constant 0.0173*** (0.00162) Observations 182 R-squared 0.824

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

In the dummy regression, every variable is significant with at least the 5% level on the long-term yields, and all except for the CPI are significant at the 1% level. There seems to be no high correlation between some of the variables, however I use a VIF-test to be certain of this. The results are shown in table 12. As there are no high VIF values, I conclude that there is no case of problematic multicollinearity in the dummy regression.

Table 12: VIF Estimations on Dummy Regression

VIF 1/VIF infldisagreements 3.18 0.3148 Dummy(JPN) 2.47 0.4047 CPIinflation 1.98 0.5051 Dummy(UK) 1.93 0.5178 unemploymentgap 1.89 0.5288 Mean VIF 2.29

(35)

35 To test whether the dummy regression coefficients for the Central Bank assets over GDP are significantly different from each other an F-test is used. A result of the F-statistic 𝐹𝐹(2, 176) = 34.09 comes out, which is significant at a 1% significance level. The results of this test show that the regression coefficients on the size of the central bank compared to its GDP differ significantly across the countries, both in Japan and in the UK the long-term yields are more affected by the size of the CB balance sheets. Thus, the dummy coefficients show that, ceteris paribus, the long-term yield is lower if the size of the balance sheet over its GDP is bigger in the UK and in Japan than it is in the US.

The correlation value between the QE proxy in Japan and the expected inflation rate is quite high (appendix 2). This suggests that there seems to be a signalling channel in Japan, which is also confirmed by the country regression for Japan.

In general, expanding the size of the balance sheet is significantly effective in lowering the long-term yields. The effects in the UK and Japan are also significantly different from the effects in the US. The value of the QE proxy in Japan is quite low compared to the UK and the US.

5. Conclusion and Discussion

This study finds that an expansion in the size of the central bank’s balance sheet significantly lowers the long-term interest rates on government bonds. Therefore, it follows the results of among others Borio and Disyatat (2010), Gagnon et al. (2011), Joyce et al. (2012), Aksoy and Basso (2014), and Fratzscher et al. (2018).

By using a regression instead of an event study analysis, the argumentations of Thornton (2017) are considered to be able to make conclusions on the effects of QE. Also, there is no chance of overestimating the effects of QE (Breedon et al., 2012). The model used in this study however , also has its own limitations, one being the fact that it does not consider international interest rates on the long-term yields. The UK is a small open economy, and as the effects of CPI and the unemployment gap were not significant in their effects, it could be argued that this is because its yields are more dependent on international macroeconomic uncertainty variables.

Referenties

GERELATEERDE DOCUMENTEN

Het krediete- quivalent volgens de opslagmethode wordt weer­ gegeven door de bovenste V-vormige lijn, die wordt gevormd door de gezamenlijke (niet negatieve)

[21] P. Sidorov, Unique representations of real numbers in non-integer bases, Math. Komornik, Expansions in noninteger bases, Integers 11B, Paper No. Li, Hausdorff dimension of

The interest rate curve is decomposed into underlying risk factors using Principal Component Analysis (PCA).. Simulation of these risk factors is used to compute the VaR of three

In this section we identify all matching intervals for the symmetric doubling maps by linking them to the maximal quadratic intervals of α-continued fraction maps.... The

[r]

Als we kijken naar de gebieden in ons land waar jaarrond- begrazing wordt toegepast, dan blijkt volgens de Alterra- database van begraasde terreinen het gemiddelde areaal niet meer

Voor het onderzoek naar de invloed van vochttekort tijdens de teelt op het verschijnsel van zachte vruchten in de afzetketen is in 2006 een aparte proef opgezet.. Het verloop en

kosten_tandarts hoe veel kosten heeft respondent afgelopen jaar gehad voor de