1
Faculty of Economics & Business
THE EFFECTS OF QUANTITATIVE EASING ON GOVERNMENT BOND YIELDS IN
THE UK
BY
MUSTAFA UNAL
10814191
Thesis Supervisor: Cenkhan Sahin
University of Amsterdam
May-June 2018
2
Statement of Originality
This document is written by Student Mustafa Unal, 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.
3
Abstract
The Bank of England has a history of implementing Quantitative Easing during uncertain
and difficult times to revive the nominal spending in order to achieve the inflation target.
The most recent QE announcement was aimed at countering the growing uncertainty
about the Brexit process and worries about the productivity and economic growth. The
aim of this paper is to measure the effects of the most recent QE announcement on
government bond yields in the UK, and to determine if these changes in yields were caused
by a change in the term premium. In order to identify any abnormal changes in the yields,
an event study is conducted using daily data. The change in the term premium is in turn
determined by both an OLS regression using monthly data, and an event study using daily
data. The results obtained show that the QE announcement did lower the yield of the 5
and 10-year government bond over the corresponding event windows. Furthermore, there
is significant evidence to assume the QE announcement lowered the risk premium of the
20-year government bond during the month, in contrast to the results obtained for the 5
and 10-year government bond which suggest that the announcement actually raised the
term premium during the month.
4
Table of Contents
1. INTRODUCTION
5
2. LITERATURE REVIEW
6
Monetary Transmission Mechanisms (MTMs)
Quantitative easing (QE)
Transmission Mechanism QE
Related studies
3. DATA AND METHODOLOGY
12
Description of the data
Evaluating QE using an event study
The effect of QE on the term premium
4. RESULTS AND DISCUSSION
18
5. CONCLUSION
22
6. REFERENCES
23
7. APPENDIX
29
Jarque-Bera normality test
White test for Heteroskedasticity
5
1. Introduction
The financial crisis that followed the collapse of the Lehman Brothers in September 2008 forced the central banks and governments around the world to respond with a variety of measures aimed at stabilizing financial conditions and support demand. In the UK, the Bank of England’s Monetary Policy Committee reacted by cutting policy rates sharply during the last quarter of 2008, which resulted in rates nearing their zero-lower bound in March 2009. With policy rates close to zero, the Bank of England (BoE) became unable to affect the economy through conventional channels. However, the economy of the UK was still in need of an additional monetary stimulus to revive its nominal spending to achieve the 2 percent inflation target. Therefore, the BoE resorted to the use of a policy, which was both unprecedented and unconventional, known as quantitative easing (Benford et al., 2009; Kapetanios et al., 2012).
The first round of quantitative easing took place between March 2009 and January 2010. The Bank of England purchased a total of £200 billion worth of assets, of which £198 billion were medium and long-term gilts. These asset purchases represented 29 percent of the amount of outstanding gilts held by the private sector at the time and were equivalent to around 14 percent of the nominal GDP (Kapetanios et al., 2012; Steeley, 2015). But as the crises intensified and the concern over competitiveness, macroeconomic imbalances of some euro-area economies and their solvency grew, the Bank of England implemented a second round of quantitative easing to achieve its price stability objective. An additional purchase of £125 billion of gilts was initiated between October 2011 and May 2012. This round, however, was expanded after a brief pause in July 2012 by an announcement stating a purchase of another £50 billion of gilts (Joyce et al., 2012a) . At this point in time, the total assets purchased amounts £375 billion and represents nearly 25 percent of annual nominal GDP. Furthermore, the Bank of England was now holding around 35 percent of the total amount of outstanding gilts held by the private sector. Taken together with the liquidity support measures to the banking sector, the Bank of England’s balance sheet relative to GDP increased more than fivefold when compared to the pre-crisis period (Joyce et al., 2011a; Joyce et al., 2012a). Then on the 4th of August 2016, to confront the growing uncertainty about the Brexit process and worries about the productivity and economic growth, the Bank of England loosened monetary policy through an interest cut and renewed the quantitative easing program. A further extension of £60 billion in government bonds was announced, to prevent a recession (Allen and Elliott, 2016).
This paper examines the impact of these measures on the UK bond market. Given the important role it played in stabilizing the economy, the focus will be on the Bank of England’s gilt purchases. Specifically, the aim of this paper is to review how quantitative easing affected the gilt
6 markets and their yields.1 Unlike the euro-area, the UK is perceived to be less bank-centric, which makes QE interventions more effective (Gern et al., 2015) and according to Joyce et al. (2011b, p. 115) “the place where we might have expected to see the clearest direct impact of QE in the reaction of financial markets is in the UK”.
Taking these statements into consideration and the uncertainties surrounding not only the strength and timing but also the mechanism of the QE program (Meier, 2009), the second section of the paper will provide a general description of the monetary transmission mechanism and the main channels through which QE affects the economy. In section 3, the data and specific research methods used in this paper will be discussed. Section 4 examines the results of the immediate reaction of bond yields and term premium of government bonds to the Bank of England’s most recent QE announcement. Finally, section 5 concludes.
2. Literature Review
To fully comprehend the effect of quantitative easing on the financial markets, and especially on the bond market it is necessary that an extensive description of the monetary transmission mechanisms is provided. This is essential in understanding the program’s own transmission mechanism and in distinguishing the effect it has on the economy compared to its conventional counterpart. This section therefore starts with the basic theory necessary to understand quantitative easing and its corresponding channels, which will be followed by an overview of the extensive number of studies conducted in the past decade on its effect on the UK gilt market.
2.1 Monetary Transmission Mechanism
The Bank of England has a hierarchal mandate in price stability, which means that the Bank’s Monetary Policy Committee (MPC) has been given the sole responsibility for meeting the
Government’s Consumer Price Index (CPI) inflation target of 2% over the medium term (Kapetanios et al., 2012). To meet this target the Bank of England has certain tools of so-called monetary policy at its disposal. These tools of monetary policy can be divided further into two groups namely: conventional and unconventional monetary tools (Mishkin et al., 2013).
Normally, to affect the economic activity and inflation the Monetary Policy Committee decides on the level of the official rate (Bank rate) at which the Bank of England deals with the
1In this paper the terms quantitative easing and asset purchase program, gilts and UK government bonds are used interchangeably.
7 money markets. This interaction between the central bank and the money markets takes place through open market operations and standing facilities provided by the central bank. Besides the money market, a change in the official rate also affects asset prices, exchange rate, and the expectations about the future course of real activity in the economy (The Bank of England, 2005). However, as a result of the financial crisis of 2007-09 many central banks around the world resorted to the use of unconventional monetary policy tools, which were mainly introduced to supplement the conventional monetary policy tools. This was perceived to be necessary in order to revive the economy, since the conventional monetary policy tools had exhausted their stimulating effects. One of these unconventional monetary tools, which has excessively been used by the Bank of England is quantitative easing (Bernanke et al., 2004; Bowdler and Radia, 2012).
The effect a change in the official interest rate or base money has on the output and prices is known to work in two broad stages. In the first stage, these adjustments lead to changes in financial market conditions, which is reflected in general liquidity and credit conditions, market interest rates, exchange rates, and in the asset prices. Subsequently “in the second stage, the changes in the financial market conditions lead to changes in nominal spending on goods and services by households and firms” (European Central Bank, 2005, p. 43). The various channels through which changes in monetary policy can affect the economic activity and inflation is collectively referred to as the monetary transmission mechanism (European Central Bank, 2005). Mishkin (1996) splits these different channels into three main categories, which are known as the credit view channel,
traditional interest rate channel, and other asset price channels.
According to the Bank of England (2005) monetary policy changes do have a short to
medium term effect on the real activity, as these changes affect nominal spending. The magnitude of the impact on the real economy, however, depends on the flexibility of the economy and on the degree of nominal price rigidities. In the long run, monetary policy determines the general price level and will not have any effect on the real sector of the economy (European Central Bank, 2005).
2.2 Quantitative Easing (QE)
Quantitative easing is the policy of expanding the central bank’s balance sheet through large-scale purchases of public and private assets, financed with new central bank money. The aim of such a policy is to inject money into the economy and increase its liquidity to revive the nominal spending and thereby the domestically generated inflation (Benford et al., 2009).
Purchases of financial assets should lower the interest rates and devaluate the currency, which in turn should lead to an increase in money holdings, increase in asset prices, stimulate spending, and increase wealth (Gern et al., 2015; Joyce et al., 2011a). There are three main channels
8 through which quantitative easing might affect asset prices: policy signalling effect, portfolio balance effects, liquidity premia effects (Steeley, 2015). The overall effect of quantitative easing on the economy can be further broken down into two stages. In which the first stage is referred to as the impact phase and emphasizes the effect of QE on the asset prices and a second adjustment phase stage where this stimulus works through the economy (Joyce et al., 2011a).
2.3 Transmission Mechanism QE
The initial impact stage works especially through the policy signalling channel and the portfolio rebalance channel, where both are targeted at lowering the long-term interest rates. According to the literature the long-term interest rates are primarily determined and affected by: first,
expectations about the future short-term interest (policy) rates and secondly by the term premium, which is the spread of the longer-term interest rates over expected policy rates. The policy signalling channel affects the former, while the portfolio rebalance channel affects the latter (Christensen and Krogstrup, 2016; Gern et al., 2015).
Signalling channel includes anything market participants learn about the future path of monetary policy from asset purchases and can therefore be used as an instrument to influence the expectations of economic agents regarding future short-term interest rates (Gern et al., 2015; Joyce et al., 2011a). Eggertson and Woodford (2003) share that unconventional monetary policy could have a favourable effect in lowering long-term bond yields as long as such a policy is perceived to be credible and economic agents trust that the central bank will commit to keep interest rates low even after the economy has recovered. Clouse et al. (2000) further confirms that purchases of large quantity of long duration assets through QE by the central bank can signal such a commitment. Asset purchases may thus strengthen the credibility of the central bank’s intention to keep interest rates low because an earlier exit from this strategy, in other words raising interest rates, would lead to losses for the central bank. However, asset purchases could also be interpreted as a signal of how bad the state of the economy really is and seen as a necessary intervention of the central bank to keep the economy healthy (Gern et al., 2015).
For the portfolio rebalancing channel to work it is necessary that the seller of the asset does not regard money as a perfect substitute for the assets sold. This assumption implies that the seller will rebalance its portfolio after selling the assets he owns by buying other assets that he considers to be better substitutes for the excess money he now holds. This, however, only shifts the excess money to the next seller of those assets, who in turn may also attempt to rebalance its portfolio by buying further assets that match his preferences. This process will continue until asset prices have risen to the point where investors are willing to hold the overall supplies of money and assets.
9 Additionally, as the central bank starts purchasing assets and push up their prices and, since prices are inversely related to yields, lower the corresponding yields relative to those on other assets, sellers might be encouraged to switch into other types of asset in search of a higher return. Thereby raising the prices of other assets, such as equities in the process (Joyce et al., 2011 a, b). Both the monetarist as the Keynesians view defend the mechanism of this channel. The monetarist approach states that a monetary supply rise can lead to excess money holdings, which the public will try to reduce by increasing their spending. One place where this happens is the stock market. The increase in demand for equities affect their prices by increasing them. According to the Keynesian view, a fall in interest rate stemming from expansionary monetary policy makes bonds less attractive relative to equities, thereby causing the equity prices to rise (Mishkin, 1996). Finally, Christensen et al. (2012) and Joyce et al. (2011b) both emphasize the importance of this channel and consider it to be the primary channel the QE works through.
“When the operation of financial markets is impaired, asset purchases by the central bank can improve market functioning by increasing liquidity through actively encouraging trading” (Joyce et al., 2012a, p. 679). This stimulus works through the third main channel, referred to as the liquidity premia channel. When financial markets are dysfunctional, investors will demand a higher
compensation in the form of higher returns for taking on the risk of buying an illiquid asset.
Intervention of the central bank can push up the asset prices and lower the premia for illiquidity. The effect of this channel, however, is temporary and limited to the duration of the QE program. Since, the liquidity premia for gilts is generally comparatively low the effect of this channel is expected to be small and not important in practice (Bowdler and Radia., 2012; Krishnamurthy and Vissing-Jorgensen, 2011).
Lower long-term interest rates therefore not only have a direct effect on borrowing rates, but could also indirectly effect prices of other asset, such as equities (Huston and Spencer, 2016). This in turn can affect the real economy. Modigliani (1971) states in his life-cycle theory that consumption spending is determined by the lifetime resources of consumers, which consists of several factors such as: real capital, human capital, and financial wealth. Furthermore, common stocks are known to be a major component of financial wealth. Therefore, a rise in stock prices increases financial wealth, and thus the lifetime resources of consumers, causing consumption to rise. This channel is referred to as the wealth effect (Mishkin, 1996). However, a second channel through which equity prices could affect the real economy is present. This channel is known as the Tobin’s q Theory, and argues that “higher stock prices encourages firms to issue additional stock that might be used to spur investment in plant and equipment” (Huston and Spencer, 2016, p. 473). Thus, the rise in asset prices and decline in yields on these other assets not only improve financing
10 conditions, making it easier for companies to raise funds, but also increase wealth. If companies invest some of the extra funding raised on capital markets, or households consume part of that increased wealth, demand (GDP) will increase (Joyce et al., 2012b).
In the adjustment phase rising asset and consumer prices increase the demand for money balances and supply of long-term assets. Which means that the initial imbalance in asset and money markets shrinks, and the price of real assets begin to fall back. “The boost to demand therefore diminishes and the price level continues to increase but by smaller amounts” (Joyce et al., 2011a, p. 202). This process will keep on repeating until the price level has reached the point where the real asset prices, real money balances and real output are restored to their equilibrium levels. Thus, asset purchases should lead to a faster return of the economy to its equilibrium level (Joyce et al., 2011a).
2.4 Related Studies
In the past decade a number of studies have been dedicated to quantify and measure the first and second round effects of quantitative easing on the financial markets in the UK, especially on the gilt market.
The first round of QE in the UK has been deemed as a great success. Meier (2009) and Joyce et al. (2011b) found significant evidence concerning the impact the Bank of England’s QE had on gilt yields. The latter estimated this effect on medium and long-term gilt yields to be an overall decline of 100 basis points, summing up the two-day reactions of the first round QE announcements. “Caglar et al. (2011) do, however, suggest that the event study methodology may have overestimated the effects because of the dominant, possibly exaggerated, impact of the first rather than the
subsequent six announcements” (Breedon et al., 2012, p. 704). Despite this, Joyce et al. (2012a) found and confirmed that the largest gilt reaction by far was to the announcement of March 2009, in which the Bank of England communicated the start of an asset purchase program. Meier’s (2009) study was also based on the market’s reaction to news about asset purchases and suggested that the initial QE announcement reduced gilt yields by 35-60 basis points at the very least compared to where they otherwise would be. Moreover Daines et al. (2012) added that the QE announcement had a bigger impact on the yields, than the actual purchases that took place in the early stage of QE. Gagnon et al. (2011) further argue that a credible policy announcement will be factored into the market prices the moment they are made rather than when this policy is carried out, because that is the moment economic agents will form their valuations and expectations. On top of this, as
Charfeddine et al. (2018) found evidence that government bonds in the UK are gradually becoming more efficient, the research section of this paper will conduct an event study to measure the effect the most recent QE announcement had on government bond yields. Additionally, event studies
11 show that yields on one and two-year bonds fell very little, “suggesting that expectations about the future policy interest rate were not responsible for most of the decline in longer-term yields” (Gagnon et al., 2011, p. 380). Which brings us to the importance of the portfolio rebalance channel. Christensen and Rudebusch (2012) consider this channel to be the single most important channel in the UK and argue that yield declines that followed the QE announcement were entirely driven by the reduction in term premium, for which among others Breedon et al. (2012) found significant
evidence. By using macro-finance yield curve model, Breedon et al. (2012) found that QE significantly lowered government bond yields by around 50 basis points through the portfolio balance channel. Most of the empirical studies in the UK and US covering QE, have concluded that bond yields and other asset prices are mainly affected through the portfolio rebalance channel by reducing the term premia (D'Amico and King, 2010; Gagnon et al., 2011; Christensen et al., 2012; Rosa, 2012). Therefore, after measuring the effect the announcement had on total yields, a second study will be conducted. The second study covers the effect the announcement had on the term premium, which will be measured using the model put together by Gagnon et al. (2011).
The second round of the QE program, however, was not as effective as the first one. Joyce et al. (2012a, p. 685) expressed this as follow: “In contrast to the early announcements during QE1, the absolute size of the impact on gilt yields was smaller—and, indeed, on average following QE2 announcements, gilt yields actually rose rather than fell over the 2-day window”. According to McLaren et al. (2014) the smaller market impact is in line with what would be expected if the QE program was widely anticipated, and that in any case there are reasons to believe that it may take longer for the portfolio balance channel to affect the other asset prices (Bridges and Thomas, 2012). Furthermore, the choice of the window size is considered being a significant issue within event studies. “Too short, and there is a risk that the full market reaction will be missed; too long, and there is a risk that other factors may be driving the observed response” (Joyce et al., 2012b, p. F283). Moreover, Joyce et al. (2011 a, b) emphasize the impact and importance of the window size by arguing that a one-day rather than a two-day window halves the effects of UK QE announcements. This was confirmed by event studies conducted by Glick and Leduc (2011) & Meaning and Zhu (2011), where both studies elected the use of a one-day window. Their estimates of the impact of the first round of quantitative easing on the yields were closer to 50 basis points. Daines et al. (2012), however, concluded that the gilt market took varying amounts of time to process the announcements on QE, with early announcements taking two days to get fully incorporated into the yields. Finally, the relevance of the efficient market hypothesis causes disagreements among agents. According to Malkiel (2012) opponents of the market efficiency theory argue that market prices during several instances in the past could not plausibly have been set by rational investors and that
12 psychological factors must have had a dominant influence.
The first two rounds of QE carried out in the past have been informative and their effects are worth considering before implementing possible future asset purchase programs. Holding on to the results and methodologies presented in this section, the next part of this paper is dedicated to measuring the effect the most recent QE announcement had on government bond yields and their term premium in the UK.
3. Data and methodology
This section starts first with a description of the data. Thereafter, the reaction of the gilt yields that followed the announcement on QE will be evaluated. Finally, following the approach of Gagnon et al. (2011) in determining the term premium effect, an extensive description of the determinants and model will be provided.
3.1 Data
The dataset for the event study consists of only two variables with daily observations, namely: the daily data on government bond yields and the market index, which is the S&P U.K Gilt Bond Index. Following the results of Joyce et al. (2012a) and Daines et al. (2012), we assume that the QE announcement that signalled the beginning of the asset purchase program will have the largest gilt reaction and that this announcement will have a greater impact than the actual purchase itself. This allows us to use the dataset mentioned above to determine the effect of QE on the yields. This research uses data for 265 trading days, starting on 20 July 2015 and ending on 8 August 2016.2 The data of these trading days are used to calculate 265 daily percentage changes of the yield on 5,10,20-year UK Government Bonds and the S&P U.K. Gilt Bond Index. The 265 percentage changes are divided into two windows (periods): an estimation window, and an event window. The
estimation window consists of 255 trading days, i.e. the (-265, -10) time interval where 𝑇 = 0 is the event date of the corresponding QE announcement. This window length is considered to be
sufficient in order to conduct an event study using daily data (MacKinlay, 1997). The event window, however, will in turn consist of the following three sizes: 3-day event window (-1, +1), 5-day event window (-2, +2), and finally, following Daines et al. (2012) a 2-day event window (0, +1) will be presented. The data on yields are obtained from the Bank of England, and the indices for the market index from S&P.3
2 Official holidays and days with limited trading are excluded in all analyses. 3 The concerning data is retrieved from the BoE’s and S&P indices’ official webpage.
13 The dataset employed in the second part of this section includes six variables with monthly observations. It comprises macroeconomic and financial variables: the term premium, inflation uncertainty, interest rate uncertainty, inflation expectations, net public sector supply of longer-term debt securities, and finally a dummy variable covering the QE announcement. This dataset allows us to analyse the financial and economic impacts on the term premium and, as suggested by Bjørnland and Leitemo (2009), it is important to jointly consider monetary policy and financial variables when analysing monetary policy transmissions.
Data are obtained from DataStream, The Organisation for Economic Co-operation and Development (OECD), and the Office for National Statistics GB (ONS). The time span of the dataset covers the period from May 2008 to December 2017. All variables are in percentages except the net public sector supply of longer-term debt securities and the QE announcement which is included as a dummy variable. The following table summarizes the variables discussed in this section:
Table 1
An overview of the variables used.
Reaction of the term premium Reaction of the yields
Variable Proxy Variable
Term Premium Bond Yields
Inflation Uncertainty CPI Inflation Market Index Inflation Expectations Inflation Swap Rate
Interest Rate Uncertainty Volatility bond Government Net Debt
QE Announcement
Notes: The data for the yields and term premia are obtained from the site of the BoE/WSJ; Data concerning the UK Gilt Bond Index is obtained from the site of S&P Indices; Data on CPI inflation and Government net debt is acquired from ONS; Inflation Swap Rate is retrieved using DataStream.
The event study consists of daily observations, while the regression of the term premium consists of only monthly observations.
14 According to the liquidity premium theory, the yield of a long-term bond is equal to the average of the expected short-term rates plus a term that covers the interest rate risk of long-term bonds, which is referred to as the term premium. The literature defines the term premium as the spread of the longer-term interest rates over expected policy rates (Joyce et al., 2011a) and can be calculated following Kim and Orphanides (2007) who argue that under the assumption of efficient capital markets the expected future short rates can be approximated by ex post realised rates. Therefore, the term premium is calculated by subtracting the nominal yield of a short-term 2-year government bond from the long-term bond yield.
This paper follows Gagnon et al. (2011), in using Consumer Price Index (CPI) as a proxy for inflation uncertainty and realised daily volatility as a proxy for the interest rate uncertainty. According to Haubrich and Lindner (2010) the future inflation expectations of the market can be determined by looking at the inflation swaps market. Therefore, the inflation swap rate is used as a proxy for the inflation expectations.
3.2 Methodology
This section starts by decomposing the bond yield into the expected future short rates and the term premium:
𝑦
,=
1
𝑛
𝐸 (𝑟
) + 𝑇𝑃
,(1)
where yi, t is the yield of the gilt i at time t, TPi, t is the term premium of gilt i at time t and Et (rt+n) captures the summation of the n year expected future short rates. The goal of this paper is to measure the effect of QE on gilt yields and to determine if the changes in these yields were primarily caused by a change in term premium. Therefore, the first part of this section covers the research method used in measuring the effect of QE on 5,10,20-year gilt yields. The second part explains the method conducted in determining the effect of QE on the term premium.3.2.1 Evaluating QE using an event study
In order to capture the effect of the QE announcement on the yields, this paper uses a simple version of the event study methodology as has been presented by MacKinlay (1997). The aim of this event study is to assess whether the cumulative daily abnormal percentage changes in the event window are statistically significant. In this paper the daily abnormal percentage change is defined as the actual percentage change of the yield on 5,10,20-year UK government bonds minus the
15 The yield changes are estimated using the following model:
𝛥𝑅
,= 𝛼 + 𝛽 𝛥𝑅
,+ 𝜀
,(2)
where 𝛥𝑅, is the daily percentage change of gilt yield i at time t, 𝛥𝑅 , is the daily percentage
change of the S&P U.K. Gilt Bond Index, and 𝜀, is the error term of the regression. The estimated
percentage change can be interpreted as the expected change in the absence of the event, as if it did not take place.4
The impact of QE is measured and determined by the abnormal percentage changes (AR) over the relevant window around the event date (T) . The model of abnormal returns is given by:
𝐴𝑅
,= 𝛥𝑅
,− 𝑎 + 𝑏 𝛥𝑅
,(3)
where ai and bi are the estimated values of αi and βi. A t-test is conducted to assess the significance of the announcement after calculating the cumulative abnormal percentage (CAR) changes and testing its normality using the Jacque-Bera test (see Appendix A). The t-test is performed following:
𝐶𝐴𝑅 𝑡 − 𝑡𝑒𝑠𝑡 =
𝐶𝐴𝑅(−𝑡 , +𝑡 )
(𝑉𝐴𝑅(𝐶𝐴𝑅 −𝑡 , +𝑡
(4)
with 𝐶𝐴𝑅 −𝑡 , +𝑡 = ∑ (𝐴𝑅) ; 𝑉𝐴𝑅 𝐶𝐴𝑅
−𝑡
𝑤1, +𝑡
𝑤2 = 𝐿𝐸𝑊 ∗ (𝑆 )where CAR (−𝑡 , +𝑡 ) is the sum of abnormal percentage changes over the relevant window around the event date, −𝑡 and +𝑡 stand for the 𝑤 days before the announcement to 𝑤 days after. Furthermore, LEW is the total length of the corresponding event window.
Finally, there are a number of assumptions this event study is based on. The first assumption is that under the market efficiency theory the impact of the announcement will be instantly
reflected in bond yields. Therefore, making it possible to measure the market’s reaction to the announcement by observing the changes in the bond yields over the study time period. Secondly, the announcement is unforeseen. The abnormal percentage changes in the yields indicate the
4 Furthermore, it is necessary to emphasize the difference between returns and yields. Return expresses the earnings on an investment during a certain time period in the past. The return includes factors, such as capital gains, interest payments, and dividends. The yield, however, describes the income obtained on an investment, such as receiving interest from holding a government bond (Bodie et al., 2013).
16 reaction of the market to the unanticipated event. The third assumption is the absence of other disruptive effects during the event window, meaning that there is no effect besides the
announcement (Sitthipongpanich, 2011).
3.2.2 The effect of QE on the term premium
The second analysis of this paper covers the effect of the QE announcement on term premia. This effect is determined following the approach of Gagnon et al. (2011). The historical variation in the term premium is explained using factors related to: the business cycle, uncertainty about economic fundamental and the net public-sector supply of longer term debt securities. Therefore, the
explanatory variables of the regression will be particularly focused on and reflect the three types of factors noted above. These variables are: inflation uncertainty, inflation expectations, interest rate uncertainty, government net debt, and the QE announcement. Each will be discussed in turn next. Inflation uncertainty: Beechey (2007) and Wright (2008) found empirical evidence for a positive relationship between inflation uncertainty and bond risk premia. They argue that as inflation uncertainty increases, risk-averse investors will demand a compensation in the form of a higher return for the possibility of a capital loss on selling long-term bond prior to maturity. According to Fischer and Modigliani (1978) the uncertainty surrounding future price levels, will be much higher when the current inflation levels are high. Moreover, increasing uncertainty about future inflation rates will reduce the safety of nominal assets. Ball (1992) clarifies this by assuming that there are only two types of policymakers. One (X) whose sole objective is to keep inflation low, while the other (Y) cares about both inflation as well as unemployment. When inflation is low, the public expect both types of policymakers to do the same thing, which is to try keep it low. However, when inflation is high X will disinflate, while Y is unwilling to start a recession by disinflating.
Therefore, high inflation leads to uncertainty because the public does not know who is in charge. Taken all together, the parameter of this variable is expected to be positive because higher uncertainty leads to a higher risk premium.
Inflation expectations: Friedman (1968) argues that an unemployment rate exceeding its natural level will lead to inflation and concludes that there is always a temporary trade-off between unemployment and inflation. Lower unemployment levels, which are usually the result of
expansionary monetary policy, will stimulate spending both through the impact on investment of lower market interest rates and through the impact on other spending and thereby raise prices, in other words cause inflation (Mishkin et al., 2013). However, another effect could take place, rising prices as result of monetary policy could lead economic agents to expect prices to continue to increase in the future (Friedman, 1968). Which brings us to the Fisher effect: “A rise in expected
17 inflation rate will eventually cause an equal rise in the interest rate” (Krugman, Obstfeld, & Melitz, 2015, p. 452). It is unclear if the term premium causes this increase in interest rate. However, according to Adrian, Crump & Moench (2013) the term premium is countercyclical, and therefore rises when unemployment grows. Higher inflation expectations will result in a higher term premium; therefore, the parameter of this variable is expected to be positive.
Interest rate uncertainty: The parameter of this variable is expected to be positive (see inflation uncertainty).
Public-sector supply of longer-term debt securities: Haugh et al. (2009) find that the spread between long and short-term interest rates across all OECD countries is positively correlated with government indebtedness. Conway and Orr (2002) further show that the interest rate effects are non-linear and tend to be greater at higher levels of indebtedness. According to Haugh et al. (2009, p. 10) “there is also some evidence suggesting that following EMU the role of the expected fiscal deficits may be more important in explaining risk premiums incorporated in sovereign bond yields”. Therefore, the parameter of this variable is expected to be positive.
QE announcement dummy: The parameter of this variable is expected to be negative (see previous studies).
The OLS regression can be expressed as follows:
𝑇𝑃
,= 𝛼 + 𝛽 𝐶𝑃𝐼 + 𝛽 𝐼𝑆𝑅 + 𝛽 𝑉𝑂𝐿 + 𝛽 𝑃𝑁𝐷 + 𝛽 𝑄𝐸 + 𝜀
,(5)
where 𝑇𝑃, is the term premium of bond i at time t, 𝐶𝑃𝐼 is the CPI inflation, 𝐼𝑆𝑅 is the inflation
swap rate, 𝑉𝑂𝐿 is the six-month realized daily volatility of the government bond, 𝑃𝑁𝐷 is the net public-sector supply of debt securities, 𝑄𝐸 is the dummy variable capturing the announcement effect and 𝜀, is an independently and identically distributed zero-mean error term.5
An important assumption of this analysis is that the long-term debt stock variables are exogenous with respect to the term premium. This implies that as the term premium declines the government will stop issuing more long-term debts. To the extent that public-sector agencies react to term premiums, in a manner similar to private investors. That is, by buying more long-term debt when the term premium is high (Gagnon et al., 2011).
18
4. Results and discussion
4.1 The reaction of the gilt yields to the QE announcement
For each type of government bond of different maturity, the significance of the abnormal return on the event date and the cumulative abnormal return during the test window is tested using t-tests. Table 2 (Panel A) summarizes the results.
Table 2
Gilt market’s reaction to the QE announcement.
Panel A: Yield Panel B: Term Premium
(𝑇 = 0) (0, +1) (−1, +1) (−2, +2) (𝑇 = 0) (0, +1) (−1, +1) (−2, +2)
GBP5 -8.13*** -3.26*** -3.23*** -3.36*** -6.38*** -8.06*** -3.23*** -1.28
GBP10 -5.93*** -1.99** -3.42*** -2.21** -2.99*** -2.45** -3.71*** -1.05
GBP20 -2.56*** -1.24 -1.00 -1.42 -1.17 -1.76* -1.68* -0.74 Notes: This table shows the t-test values of the cumulative abnormal returns for the 5, 10, and 20-year government bonds in the UK. T = 0 measures the effect of the announcement on the yields and term premia, the same day it took place.
*** 1% significance level. ** 5% significance level. * 10% significance level
Table 2 (Panel A) shows statistically significant outcomes regarding the announcement effect on the event day as the cumulative effect within the length of the corresponding test windows. Meaning that the QE announcement not only successfully lowered 5, 10 and 20-year government bond yields on the same day the announcement took place, but also had a significant impact over the length of the test windows for the 5 and 10-year government bond yield. The results are in line with studies conducted in the past, but covering the first round of quantitative easing (Rosa, 2012; Joyce et al., 2012 a, b; Meier, 2009). The 5 and 10-year government bond being (more) significant could be explained by economic agents expecting the government to emphasize the purchase of these bonds more than others. However, like mentioned before event study analysis is sensitive to the chosen window length. Studies show that a one-day rather than a two-day window halves the effects (Joyce et al., 2011 a, b). Daines et al. (2012) states that it takes two days for early announcements to be fully incorporated into the yields, making test windows consisting of more than two days
questionable since the risk of contamination by other events is higher (Joyce et al., 2012b).
Furthermore, there are a number of strong assumptions this event study makes. First, assuming that there is no other effect besides the announcement during the event window, most likely will not
19 hold in reality. Joyce et al. (2012a) argued that during the global crisis there was regularly other relevant news around the time of the quantitative easing announcement affecting and driving bond yields in different directions. UK’ s statement about a possible exit from the EU and the referendum held several months prior to the QE announcement caused not only disagreements among the citizens in the UK, but also among EU member countries setting a series of negotiations and deals in motion.6 Since, the QE announcement in August 2016, took place middle and during the Brexit process, other factors besides this announcement could have affected the government bond yields. Secondly, the market efficiency theory under which event studies are conducted has significant opposition. According to many economic agents, market prices during several instances in the past could not plausibly have been set by rational investors. They argue that psychological factors must have had a more dominant influence and therefore consider the market efficiency theory as invalid (Malkiel, 2012).
20
4.2 The reaction of the term premium to the QE announcement
Table 3 summarizes the test statistics and p-values of the different variables that according Gagnon et al. (2011) play a role in determining the term premium of the government bond. Like the event study, this regression has been conducted for all three types of government bond of different maturities.
Table 3
The effect of QE on the term premium.
Dependent Variable:
Term Premium 𝐺𝐷𝑃 𝐺𝐷𝑃 𝐺𝐷𝑃
Regressors
Inflation Uncertainty 0.194 (0.2653) -0.120 (0.2118) 0.408 (0.3282)
Inflation Expectations 0.805 (0.2294)*** -0.266 (0.3699) -0.608 (0.5215)
Interest Rate Uncertainty -2.36 (0.4497)*** -5.866 (1.2459)*** -1.022 (2.0013)***
Government Net Debt 0.023 (0.0451) -0.045 (0.0487) 0.578 (0.0547)***
QE Dummy 0.467 (0.2017)** 0.791 (0.3056)** -0.719 (0.2388)***
Summary Statistics
SER
0.592 0.592 0.674R
2 0.392 0.309 0.884Obs
116 116 116Notes: This table shows the values of the coefficients with their corresponding standard errors for the 5, 10, and 20-year government bonds in the UK. The Std. Errors are reported in parentheses and are Newey-west Std. Errors.
*** 1% significance level. ** 5% significance level. * 10% significance level.
Table 3 shows statistically significant results for all three government bond yields. However, only the 20-year government bond has the anticipated negative value for the announcement dummy, in contrast to the 5 and 10-year government bonds who show positive values for the announcement dummy. Meaning that the QE announcement increased the term premium of the 5 and 10-year government bond yields and lowered the term premium of the 20-year government bond yield. The Brexit has not only brought down investors’ confidence to all-time low levels, but was also perceived
21 as the biggest risk to the UK’s financial system.7 Therefore, the announcement could have been seen as a signal of how bad the state of the economy really is by investors. Which resulted in investors demanding a compensation in the form of higher returns, causing the term premium of the
corresponding government bond yields to rise during the month.8 To assess the reaction of the term premium to the announcement, a third analysis is conducted by performing an event study (see Table 2 Panel B). The results show statistically significant evidence for the first three window sizes for the 5 and 10-year government bond, meaning that the term premium did decrease over the corresponding event window. In summary, the results show that the announcement did lower the term premium over the corresponding event windows, however there is evidence that shows that the term premium for the 5 and 10-year government bond increased over the month.
7 The interested reader is referred to Hodgson (2017).
8 However, there is a second explanation for the observed values of table 3. The use of monthly data could have presented a biased and, therefore invalid outcome. Section 4.1 discusses the presence of any other factor besides the announcement that possibly could have affected the movement of the yields over the relevant event window. Since, the second analysis in this paper is dealing with monthly data, the chance on results being contaminated by other factors besides the announcement is plausibly higher than the first analysis. Meaning that the regression conducted may not be capturing the real effect of the announcement on the term premium.
22
5. Conclusion
This paper evaluates the reaction of the gilt yields and term premia of the 5, 10, and 20-year gilts to the QE announcement. The key findings are as follow: First, there is statistically significant evidence that show the announcement did lower the yields of the 5 and 10-year gilt on the announcement day, as over the relevant event windows. The 20-year gilts reaction only tested significant for the announcement day. Second, the term premium using monthly data tested significant for all three gilts of different maturity, with the announcement dummy of the 5, and 10-year gilt showing a positive value . This suggest the announcement increased the term premium of the 5 and 10-year gilt. However, when daily data is used, there is significant evidence to assume the QE announcement successfully lowered the term premia over the corresponding event windows. Finally, the analysis of this paper suggests that the QE announcement did lower the yields and term premia for the 5 and 10-year gilts.
23
References
Adrian, T., Crump, R., & Moench, E. (2013). Do Treasury Term Premia Rise Around Monetary Tightenings? Federal Reserve Bank of New York, URL: http://libertystreeteconomics.newyorkfed.org/2013/04/do-treasury-term-premia-rise-around-monetary-tightenings-.html
Allen, K., & Elliott, L. (2016). Bank of England Cuts Interest Rates to 0.25% and Expands QE. The Guardian, Retrieved May 15, 2018, from https://www.theguardian.com/business/2016/aug/04/bank-of-england-cuts-uk-interest-rates
Ball, L. (1992). Why Does High Inflation Raise Inflation Uncertainty? Journal of Monetary Economics, 29(3), 371-388.
Beechey, M. (2008). Lowering the Anchor: How the Bank of England’s Inflation-Targeting Policies Have Shaped Inflation Expectations and Perceptions of Inflation Risk, 2008-44.
Benford, J., Berry, S., Nikolov, K., & Young, C. (2009). Quantitative Easing. Bank of England Quarterly Bulletin, 90-100.
Bernanke, B. S., & Reinhart, V. R. (2004). Conducting Monetary Policy at Very Low Short-Term Interest Rates. American Economic Review, 94(2), 85–90.
Bernanke, B. S., Reinhart, V. R., & Sack, B. P. (2004). Monetary Policy Alternatives at the Zero Bound: An Empirical Assessment. Brookings Papers on Economic Activity, 2004(2), 1–78.
Bjørnland, H.C., & Leitemo, K. (2009). Identifying the Interdependence Between US Monetary Policy and the Stock Market. Journal of Monetary Economics, 56(2), 275-282.
Bodie, Z., Kane, A., & Marcus, A. J. (2013). Investments (10th ed., The Irwin series in finance). Boston: McGraw-Hill.
Bowdler, C., & Radia, A. (2012). Unconventional Monetary Policy: The Assessment. Oxford Review of Economic Policy, 28(4), 603–621.
24 Breedon, F., Chadha, J. S., & Waters, A. (2012). The Financial Market Impact of UK Quantitative Easing.
Oxford Review of Economic Policy, 28(4), 702–728.
Bridges, J., & Thomas, R. (2012). The Impact of QE on the UK Economy – Some Supportive Monetarist Arithmetic. SSRN Electronic Journal.
Charfeddine, L., Khediri, K. B., Aye, G. C., & Gupta, R. (2018). Time-Varying Efficiency of Developed and Emerging Bond Markets: Evidence from Long-Spans of Historical Data. Physica A: Statistical Mechanics and Its Applications, 505, 632–647.
Christensen, J. H. E., & Krogstrup, S. (2016.). Transmission of Quantitative Easing: The Role of Central Bank Reserves. Federal Reserve Bank of San Francisco, 64.
Christensen, J. H. E., & Rudebusch, G. D. (2012). The Response of Interest Rates to US and UK Quantitative Easing*. The Economic Journal, 122(564), F385–F414.
Clouse, J., Henderson, D., Orphanides, A., Small, D., & Tinsley, P. (2000). Monetary Policy When the Nominal Short-Term Interest Rate is Zero. Federal Reserve Bank, 84.
Conway, P., & A. Orr (2002). The GIRM: A Global Interest Rate Model. Westpac Institutional Bank Occasional Paper.
Daines, M., Joyce, M. A. S., & Tong, M. (2012). Working Paper No. 466 QE and the Gilt Market: A Disaggregated Analysis, (466), 46.
D’Amico, S., & King, T.B. (2010). Flow and Stock Effects of Large-Scale Treasury Purchases. Finance Economics Discussion Series, No. 2010-52.
Edgerton, D., & Shukur, G. (1999). Testing Autocorrelation in a System Perspective Testing Autocorrelation. Econometric Reviews, 18(4), 343–386.
Eggertsson, G. B., & Woodford, M. (2003). The Zero Bound on Interest Rates and Optimal Monetary Policy. Brookings Papers on Economic Activity, 2003(1), 139–211.
25 European Central Bank. (2000, July). Monetary Policy Transmission in the Euro Area. ECB Monthly Bulletin,
43-58.
Fama, E. F. (1970). Efficient Capital Markets: A Review of Theory and Empirical Work. The Journal of Finance, 25(2), 383–417.
Fischer, S., & Modigliani, F. (1978). Towards an Understanding of the Real Effects and Costs of Inflation. Weltwirtschaftliches Archiv, 114, 810.
Friedman, M. (1968). The Role of Monetary Policy. The American Economic Review, 58(1), 1–17. Gagnon, J., Raskin, M., Remache, J., & Sack, B. (2011). The Financial Market Effects of the Federal
Reserve’s Large-Scale Asset Purchases. International Journal of Central Banking, 7(1), 41.
Gern, K.-J., Jannsen, N., Kooths, S., & Wolters, M. (2015). Quantitative Easing in the Euro Area: Transmission Channels and Risks. Intereconomics, 50(4), 206–212.
Glick, R., & Leduc, S. (2012). Central Bank Announcements of Asset Purchases and the Impact on Global Financial and Commodity Markets. Journal of International Money and Finance, 31(8), 2078–2101.
Haubrich, J.G., & Lindner, J. (2010). Inflation Swaps. Federal Reserve Bank of Cleveland, URL:
https://www.clevelandfed.org:443/newsroom and events/publications/economic trends/economic trends archives/2010 economic trends/et 20101208 inflation swaps
Haugh, D., Ollivaud, P., & Turner, D. (2009). What Drives Sovereign Risk Premiums? OECD Economics Department Working Papers, No.718, 1-24.
Hodgson, C. (2017). Brexit Has Driven UK Investor Confidence to an All-Time Low- below Levels Seen During the 2008 Crisis and Iraq War. Business insider, Retrieved May 30, 2018, from
https://www.businessinsider.nl/investor-confidence-below-crisis-wartime-levels-2017-12/?international=true&r=UK
Hunt, A., & Wheeler, B. (2018). Brexit: All You Need to Know About the UK Leaving the EU, BBC News, Retrieved May 15, 2018, from http://www.bbc.com/news/uk-politics-32810887
26 Huston, J. H., & Spencer, R. W. (2016). The Wealth Effects of Quantitative Easing. Atlantic Economic
Journal, 44(4), 471–486.
Jarque, C. M. (2011). Jarque-Bera Test: In Lovric M. (eds) International Encyclopedia of Statistical Science. Berlin, Heidelberg: Springer Berlin Heidelberg.
Joyce, M. A. S., McLaren, N., & Young, C. (2012a). Quantitative Easing in the United Kingdom: Evidence from Financial Markets on QE1 and QE2. Oxford Review of Economic Policy, 28(4), 671–701.
Joyce, M., Lasaosa, A., Stevens, I., & Tong, M. (2011b). The Financial Market Impact of Quantitative Easing. International Journal of Central Banking, 7(3), 113-61.
Joyce, M., Miles, D., Scott, A., & Vayanos, D. (2012b). Quantitative Easing and Unconventional Monetary Policy – an Introduction*. The Economic Journal, 122(564), F271–F288.
Joyce, M., Tong, M., & Woods, R. (2011a). The United Kingdom’s Quantitative Easing Policy: Design, Operation and Impact, Bank of England Quarterly Bulltion, 51(3), 200-212.
Kapetanios, G., Mumtaz, H., Stevens, I., & Theodoridis, K. (2012). Assessing the Economy-wide Effects of Quantitative Easing*. The Economic Journal, 122(564), F316–F347.
Kim, D., & Orphanides, A. (2007). The Bond Market Term Premium: What Is It, and How Can We Measure It?, BIS Quarterly Review, 14.
Krishnamurthy, A., & Vissing-Jorgensen, A. (2011). The Effects of Quantitative Easing on Interest Rates: Channels and Implications for Policy, NBER Working Paper Series, 47.
Krugman, P. R., Obstfeld. M., & Melitz. M. J. (2015). International Economics Theory and Policy. Pearson Education.
Lumley, T., Diehr, P., Emerson, S., & Chen, L. (2002). The Importance of the Normality Assumption in Large Public Health Data Sets. Annual Review of Public Health, 23(1), 151–169.
27 MacKinlay, A. C. (1997). Event Studies in Economics and Finance. Journal of Economic Literature, 35(1),
13-39.
MacKinnon, J. G., & White, H. (1985). Some Heteroskedasticity- Consistent Covariance Matrix Estimators with Improved Finite Sample Properties*. Journal of Econometrics, 29, 305-325.
Malkiel, B. G. (2003.). The Efficient Market Hypothesis and Its Critics. Journal of Economic Perspectives, 17(1), 59-82.
McLaren, N., Banerjee, R. N., & Latto, D. (2014). Using Changes in Auction Maturity Sectors to Help Identify the Impact of QE on Gilt Yields. The Economic Journal, 124(576), 453–479.
Meaning, J., & Zhu, F. (2011). The Impact of Recent Central Bank Asset Purchase Programmes. Bank of International Settlements Quarterly Review.
Meier, A. (2009). Panacea, Curse, or Nonevent? Unconventional Monetary Policy in the United Kingdom. IMF Working Papers, 09(163), 1.
Mishkin, F. S. (1996). The Channels of Monetary Transmission: Lessons for Monetary Policy. National Bureau of Economic Research, 1-28.
Mishkin, F., Matthews, K., & Giuliodori, M. (2013). The Economics of Money, Banking, and Financial Markets, Pearson Educated Limited.
Modigliani, F. (1971). Monetary Policy and Consumption. Consumer spending and Monetary Policy: The Linkages, 9-84
Rosa, C. (2012). How “Unconventional” are Large-Scale Asset Purchases? The Impact of Monetary Policy on Asset Prices. SSRN Electronic Journal.
Schwert, W. (1989). Test for Unit Roots: A Monte Carlo Investigation. Journal of Business and Economic Statistics, 7, 147-159.
Sitthipongpanich, T. (2011). Understanding the Event Study. Journal of Business Administration, 34(130), 59-68.
28 Steeley, J. M. (2015). The Side Effects of Quantitative Easing: Evidence from the UK bond Market. Journal
of International Money and Finance, 51, 303–336.
Stock, J., & Watson, Mark W. (2011). Introduction to Econometrics / (3rd ed., Addison-Wesley series in economics). Boston: Addison-Wesley.
The Bank of England. (2005). The Transmission Mechanism of Monetary Policy. The Bank of England, 1-12.
Wright, J. H. (2008). Term Premiums and Inflation Uncertainty: Empirical Evidence from an International Panel Dataset, The American Economic Review, 101(4), 49.
29
Appendix
A.1 Jarque-Bera normality test
When outcomes are normally distributed the t-test is identified as the optimal analysis by several mathematical criteria (Lumley et al., 2002). Therefore, the cumulative abnormal returns are tested for normality using the Jarque-Bera test. The Jarque-Bera test is able to identify the normality of a sample and detect deviations from the normal distribution (Jarque, 2011). The test statistic is calculated using the following model:
𝐽𝐵 = 𝑛 ∗
+
( )~ χ
22(A.1)
where n is the sample size, S is the sample skewness, and K is the sample kurtosis. The Jarque-Bera test statistic is χ2 distributed with 2 degrees of freedom. The hypotheses are 𝐻 :
𝜗 = 0
𝑎𝑔𝑎𝑖𝑛𝑠𝑡 𝐻 :
𝜗 ≠ 0 where the null corresponds to the data being normally distributed.
The test statistics and their corresponding p-values are: Table A.1
Jarque-Bera normality test.
𝐽𝐵 (−2, +2)
𝐍𝐨𝐫𝐦𝐚𝐥𝐢𝐭𝐲 𝐭𝐞𝐬𝐭GBP
5 5.80*GBP
10 4.17GBP
20 5.29* *** 1% significance level. ** 5% significance level. * 10% significance levelAccording to the Jarque-Bera test, the null hypothesis of a normally distributed CAR is rejected only at a significance level of 10% for
GBP
5 andGBP
20 .30
A.2 White test for Heteroskedasticity
It is necessary that the ordinary least squares (OLS) meets a number of assumptions for it to be able to provide an appropriate estimator of the unknown regression coefficients. One of the assumptions that need to be satisfied is that the conditional distribution of the error term given Xi has a mean of zero. Furthermore, if the variance of the conditional distribution of the error
term given X is constant and independent on x, the error term is known to be homoscedastic. In cases where heteroskedasticity is present, the standard error must be corrected before
computing t-statistics
(Stock and Watson, 2011)
. To determine whether the variance of the errors in the regression model is constant, a so-called White test is conducted. The White test regresses the squared error terms of the original regression model onto a set of regressors that consists of the original regressors along with their squares and cross products (MacKinnon and White, 1985). The basic form of the White test is shown in the following auxiliary regression:𝑒 = ϒ + ϒ 𝑋 + ⋯ + ϒ 𝑋 + ϒ
𝑋 + ⋯ + (A.2)
ϒ 𝑋 + 𝜆 𝑋 𝑋 + 𝜆 𝑋 𝑋 + ⋯ + 𝜆 𝑋 𝑋 +
𝜀
∗where e2 is the squared residuals, Xi the regressors of the original model, X2 the squared of
these regressors and
X
kX
i their cross products. The test statistic used to test the hypotheses,n𝑅 ~ χ2
P, is asymptomatically χ2 distributed with p degrees of freedom, where p is the number
of regressors in the auxiliary regression, the constant excluded . Furthermore, n stands for the sample size and 𝑅 is the coefficient of determination. The hypotheses for the White test are:
H
0: Homoskedasticity
ϒ = ϒ = 𝜆 = ⋯ = ϒ = 𝜆 = 0
H
1:Heteroskedasticity
𝐴𝑡 𝑙𝑒𝑎𝑠𝑡 𝑜𝑛𝑒 𝑜𝑓 ϒ 𝑜𝑟 𝜆 ≠ 0
Because the predicted value ỷ includes all independent variables, 𝑒 = 𝛿 + 𝛿 ỷ + 𝛿 ỷ + 𝑣 could be used instead to test the heteroskedasticity of the regression. The hypotheses tested are 𝐻 : 𝛿 = 𝛿 = 0 𝑎𝑔𝑎𝑖𝑛𝑠𝑡 𝐻 : 𝛿 = 𝛿 ≠ 0
where the null corresponds to the absence of
heteroskedasticity in the error term.
This test uses the F(t) statistic,𝐹(𝑡) = , in testing31
The results of the White test for heteroskedasticity are: Table A.2
White test for heteroskedasticity.
𝜒2
𝐇𝐞𝐭𝐞𝐫𝐨𝐬𝐤𝐞𝐝𝐚𝐬𝐭𝐢𝐜𝐢𝐭𝐲 𝐭𝐞𝐬𝐭GBP
5 68.97***GBP
10 83.43***GBP
20 76.28*** *** 1% significance level. ** 5% significance level. * 10% significance levelAccording to the White test, the null hypothesis of homoscedastic errors is rejected, at 1%
significance level. When heteroskedasticity is present, robust standard errors tend to be
more trustworthy. The use of robust standard errors does not change the coefficient of the
estimates, but the test statistics will give you a reasonable accurate p value (Stock and
Watson, 2011).
A.3 Breusch-Godfrey test for Autocorrelation
In time series data, the value Xt in one period may be, and typically is correlated with its value
Xt-1 in the next period. Similarly, this correlation between periods could also occur among the
error terms in the regression. If the regression errors are autocorrelated, the usual
(heteroskedasticity-robust) standard errors will not be valid and therefore need to be corrected before using t-statistics
(Stock and Watson, 2011)
. To determine whether autocorrelation is present in the error terms, a Breusch-Godfrey and LM-test is conducted (Edgerton and Shukur, 1999).32
The auxiliary regression used is defined as follow:
𝑒 = 𝜌 𝑒
+ 𝜌 𝑒
+ ⋯ + 𝜌 𝑒
+
𝜀
∗(A. 3)
where
𝑒
is the residuals and𝑒
is defined as its value k periods ago, better known as the kthlagged value. The test statistic, n𝑅 ~ χ2k, is asymptomatically χ2 distributed with k degrees of
freedom, where k is the number of lagged error terms. Furthermore, n stands for the sample size and 𝑅 is the coefficient of determination. The hypotheses tested are 𝐻 :
𝜌 = 0
against
𝐻 :𝜌 ≠ 0 where the null corresponds to the absence of autocorrelation in the
residuals. As suggested by
Schwert (1989), the number of lagged error terms is determined by the following rules of thumb: 𝑘 = 12 ∗ , where τ is the number of observations. The test statistics and their corresponding p-values are:Table A.3
Breusch-Godfrey test for autocorrelation.
𝜒2
𝐀𝐮𝐭𝐨𝐜𝐨𝐫𝐫𝐞𝐥𝐚𝐭𝐢𝐨𝐧 𝐭𝐞𝐬𝐭
GBP
5 68.93***GBP
10 80.43***GBP
20 84.09***Notes: The number of lagged error terms included is equal to 2. *** 1% significance level.
** 5% significance level. * 10% significance level
According to the
Breusch-Godfrey and LM-test, the null hypothesis of no autocorrelation in
the residuals is rejected at 1% significance level. Since, both heteroskedasticity and
autocorrelation is present in our data the regular t-test outcome will be biased and
therefore not accurate. To solve this problem, this paper turns to the use of Newey-West
variance estimator. This estimator of the variance overcomes the heteroscedasticity and
33
