0
THE IMPACT OF EUROPEAN
CENTRAL BANK’S PURCHASE
PROGRAMS ON GOVERNMENT
BOND PRICES
Hotbed of asset bubbles?
BY
ZSOFIA KOKENY
MASTER THESIS
MSC FINANCE ASSET MANAGEMENT TRACK
OF
U
NIVERSITY OFA
MSTERDAM,
A
MSTERDAM BUSINESS SCHOOLJ
UNE2018
S
UPERVISOR:
R.
C.
R.
VANL
AMOENAbstract: To maintain price stability and monitor inflation, the ECB engaged in QE, through different asset purchase programs. The quantity of these purchases raise questions of the risks and undesirable side effects, such as deviations between market prices and fundamental values. The aim of this study is to analyse the degree of these divergences and identify exuberant price periods on government bond prices. The paper applies GSADF test on monthly time series to analyse five purchase programs for 10 Euro Area countries between January 2003 and September 2017. The results indicate that government bond prices, triggered by QE policies, increased in such extent that they remarkably deviate from their fundamental values. Whilst CBPP1 is responsible for short-term decline in yields, PSPP and ABSPP/CBPP3 induce government bond yields to decline in both short and long-term. Also, the GSADF tests provide some evidence that the programs are responsible for formations of asset price bubbles; even though months of exuberant behaviour are temporary. Thus, the ECB should implement QE policies with more caution in the future and adjust monetary policy in order to mitigate the discrepancy of government bond yields from their fundamental levels.
1 Statement of Originality
This document is written by Zsofia Kokeny who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
2
T
ABLE OFC
ONTENTS1 Introduction ... 4
2 Literature Review ... 7
2.1 Fundamental drivers of government bond yields ... 7
2.2 Asset Bubbles and GSADF Applications ... 9
2.3 QE’s impact on asset prices ... 10
2.4 Hypotheses ... 12
3 Data ... 13
3.1 Data selection and fundamental drivers ... 13
3.2 Descriptive Statistics ... 18
4 Methodology ... 23
4.1 Identifying fundamental yields ... 23
4.2 Identifying Asset Bubbles – The GSADF test... 25
5 Results ... 28
5.1 Fundamental drivers of government bond yields: observed and forecasted variables ... 28
5.2 Explosive behaviour in government bond yields and prices ... 32
6 Robustness checks ... 36
6.1 Fundamental drivers: adding demographic variables ... 37
6.2 Fundamental drivers: including additional forecast variables ... 40
7 Conclusion ... 43
8 Appendix ... 45
3
Figure 1: Monthly net purchases per purchase program at book value. Source of data: ECB ... 5
Figure 2: Government bond yield levels by country ... 13
Figure 3: The development of GSADF test ... 27
Figure 4: GSADF test results: exuberant periods in government bond yields/prices for CBPP1 and CBPP2 ... 35
Figure 5: GSADF test results: exuberant periods in government bond yields/prices for ABSPP/CBPP3 and PSPP ... 36
Figure 6: GSADF test results: exuberant periods in government bond yields/prices for inverse yields ... 45
Table 1: Description, expected sign and source of fundamental drivers of government bond yields . 15 Table 2: Descriptive statistics of government bond yields and observed variables per country ... 19
Table 3: Descriptive statistics of forecast variables per country ... 20
Table 4: Descriptive statistics of PSPP net volume per country at book value. ... 21
Table 5: Descriptive statistics of demographic variables per country ... 22
Table 6: Government bond yields and its fundamental drivers: observed and forecasted macroeconomic, fiscal, financial, and trade related variables ... 31
Table 7: Exuberant price periods: GSADF test results of inverse government bond yields and purchase programs by country ... 33
Table 8.b Response functions of purchase programs after adding demographic variables. Extension of Table 8.a ... 38
Table 8.A: Government bond yields and fundamental drivers including demographic variables……….40
Table 9.B: Response functions of purchase programs after including additional forecast variables. Extension of Table 9.a ... 41
Table 9.a: Government bond yields and fundamental drivers with additional forecast variables for a restricted sample………43
4
1 INTRODUCTION
The main role of the European Central Bank (ECB) is to maintain price stability in the euro area, thereby preserving the purchasing power of the single currency, supporting investment and consumption and keeping inflation rates within the ECB’s target.
In October 2014, to ease monetary and financial conditions the Governing Council of the ECB engaged in quantitative easing (QE) through an Extended Asset Purchase Programme (APP) following the Bank of Japan, Federal Reserve in the US and the Bank of England in the UK. APP embraces all purchases under which private or public sector securities are purchased and consists of four different programmes, namely the public sector purchase programmes (PSPP), corporate sector purchase programmes (CSPP), asset-backed security purchase programmes (ABSPP) and third covered bond purchase programmes (CBPP3).
The study continues and broadens the paper of Van Lamoen et al (2017). Firstly, by applying a more elaborate model to account for the fundamentals and to more accurately detect exuberant price behaviour. The regression estimation includes the real purchase volume of PSPP functions as it is reasonable to put more emphasis on the public sector because it has by far the highest purchase volume (Figure 1). Furthermore, besides restricting the analysis only on observed variables, the model is expanded with consensus data forecasts, demographical coefficients and directly models the expectation on government bond yields. Secondly, more programs are considered in this research; besides PSPP, CBPP3 and ABSPP, two terminated programmes, CBPP1 and CBPP2, are analysed. Finally, using an updated sample; the study covers over 14 years, starting from January 2003 until the third quarter of 2017 and consists of monthly observations for 10 Euro Area countries including Austria, Belgium, Finland, France, Germany, Italy, Ireland, the Netherlands, Portugal and Spain. To test if bond yields differ significantly from their fundamental values and if they experience bubble behaviour, the GSADF model by Phillips et al. (2013; 2015) is applied.
Figure 1 represents the monthly net purchases at book value of each programme between October 2014 and March 2018. In October 2014 the Eurosystem launched the first extended programme, the CBPP3. While the initial monthly net purchases were about €11bn, they dropped to €3bn in 2018; nonetheless the total net purchases reached €255.92bn in March 2018. One month after launching, in November 2014, ABSPP was introduced and targeted to support banks by easing the funding and credit conditions. Albeit CSPP (started in June 2016)
5
having relatively modest total monthly net purchases of €26.07bn, it reached €150.31bn in total. Under the extended purchase programme the most notable total monthly net purchases belong to PSPP; it amounts to €1.99 trillion in three years, starting from March 2015 until March 2018. Initially the purchases were around €50bn/month, peaking at €80bn in May 2016. Since October 2017, the Governing Council decided to gradually decrease the PSPP net purchases, at first to €50bn/month in 2017 and to €20bn/month from 2018 January.1
FIGURE 1: MONTHLY NET PURCHASES PER PURCHASE PROGRAM AT BOOK VALUE. SOURCE OF DATA: ECB The bare quantity of these purchases, the associated risk and potential misconduct makes QE a target of severe criticism. As altering the supply of money in the highly liquid Eurozone involves the hazard of audacious behaviour of investors by taking more risk to maintain profit margins, distortion of market prices, formation of asset bubbles, QE fatigue and raise in the level of inflation. (Hallett, 2017; Ewing, J. 2015). Although Hallett argues that instead of QE, the inadequate domestic policies in the euro area are responsible for the emerging instabilities, the concern around bubble behaviour is well-founded and feasible due to the abundant liquidity, easy or cheap credit and low borrowing costs (interest rates) in the Eurozone. What makes the debate even more intricate is that quantifying the associated risks of recent extension of QE and to predict the time horizon they might materialize is extremely hard. (Fiedler-et-al.,-2017).
1 CBPP1 and CBPP2 are both terminated programmes by the ECB, the former one was announced in 2009 May
and ended in 2010 June and the latter was announced in 2011 November and terminated in 2012 October. Regarding volumes, CBPP1 reached €60bn and CBPP2 achieved a more moderate amount of €16.4bn. This paper focuses on five APP programs; PSPP, CBPP1, CBPP2, CBPP3 and ABSPP.
6
Whilst Deutsche Bank described Quantitative Easing (QE) as “many asset bubbles, expropriated savers and lack of creative destruction”, recent studies found little sign of emerging asset bubbles (Schneider and Kirk, 2016; Lamoen et al.,2017; Haas and Young-Taft, 2017). However the existing literature linking APP with government bond price bubbles is currently inextensive. (Andrade et al., 2016; Altavilla et al., 2015; De Santis., 2016). The sub-questions of this paper are; Do demographic variables play a significant role in determining the fundamental yield? And, what are the fundamental drivers of government bond yields? The main research question is whether there is a distortion of government bond prices from their fundamental levels triggered by the ECB’s latest purchase programmes? If so, are these purchases responsible for the formation of asset price bubbles?
The academic relevance of this paper is to enhance the understanding of the drivers of government bond yields and how they are affected by QE in the Euro Area. As demographic ratios have not been elaborately examined for Europe (Rawdanowicz et al., 2017), the paper helps to fill the void in the field of linking demographics to QE effects in the euro area. Also, the study aims to contribute towards understanding the role of monetary policy in the formulation of asset bubbles. Upon finding supporting evidence for the hypothesis, the existing monetary policy should be adjusted to stimulate the pace of reversal and normalization of government bond prices to their fundamental levels and to adopt policies to limit the damage to the real economy. The outcome of this paper is also important for government debt policies and stakeholders of government bonds such as private investors, pension funds or Central Banks as shifts in interest rates have significant effect on saving behaviours and portfolio rebalancing. (Arshanapalli et al., 2016; van Lamoen at al., 2017).
The remainder of the paper is organized as follows. Section 2 reviews the most recent empirical literature on asset price bubbles, ECB’s impact on asset prices and the determinants of government yields. The empirical dataset and methodology applied to test the hypotheses are explained in sections 3 and 4. Section 5 reviews the empirical results and section 6 is the robustness checks. Section 7 provides a conclusion and further discussion.
7
2 LITERATURE REVIEW
2.1
FUNDAMENTAL DRIVERS OF GOVERNMENT BOND YIELDS
In order to perform GSADF statistics the fundamental value of government bonds has to be determined, even though the measurement is not straightforward. There is a great deal of empirical literature on analysing the fundamental variables of government bond yields with various results. Along with identifying bubbles, there is no general agreement among academics on what is the best method to apply. Following the hypothesis that movements in bond prices can be explained by movements of the key drivers of bond yields, the main determinants of sovereign yields have to be selected.
To determine the fundamental component for the bubble indicator estimation, Blot et al., (2017) established a model based on a variety of macroeconomic and financial coefficients such as quarterly observations for real GDP growth and disposable income, and monthly inflation, oil prices, VIX index and industrial production.
Akram and Das (2017) analysed the dynamics of government bond yields for 11 Euro zone countries. To review key drivers they applied pooled mean group method between 1997 and 2015, extended with autoregressive distributive lag method to examine individual countries. They selected short-term interest rates, inflation, industrial production and debt-to-GDP ratio as relevant variables. In the long run, either analysing quarterly or monthly datasets, for 10-year government bond yields the short term interest rates appear to be the most important driver, increasing yields by 64 basis points for quarterly datasets and 40 basis points when monthly observations were used.
On the other hand, Poghosyan, (2014) analysed both short and long-term determinants by applying panel cointegration analysis over a relatively long time period, using annual observations between 1980-2010 for a sample of 22 advanced economics. Results suggests that real GDP growth has the highest impact on yields; 45 basis point increase in response to 1 percent increase in real GDP growth. More moderate, but another highly significant long-term effect observable for debt-to-GDP ratio, specifically causing 2 basis point increase in sovereign yields. Regarding short-term results, effects of inflation is significantly negative while debt ratio, and short-term interest rates have positive impact, although in one year roughly half of the deviation adjusts to the long-term equilibrium.
Similarly, Costantini et al., (2013) applied panel cointegration model and found that ceteris
8
point increase for the full sample to 0.5 basis point rise when certain countries (Greece, Spain and Portugal) are excluded. Besides expected debt-to-GDP, they use another forecast ratio; the budget balance-to-GDP as they argue that these two fiscal drivers are the main sources from which investors make assumptions on a given country’s fiscal position(s) and associated default risk(s). Among other drivers, cumulated inflation stands out; resulting in a 20 basis point increase in the full sample model. The sample consist of 9 EMU countries between 2001 and 2011 with dependent variable of 10 year government bond spread over Germany.
Another paper that examined the determinants of long-term government bond yields in the EMU is an ECB’s working paper by Afonso and Rault (2015). The paper applies a panel of 10 EMU countries in three time periods; pre, during and post the global credit crash turned into a sovereign debit crises. Their findings indicate that drivers of sovereign spreads have changed over time, more specifically the previously insignificant macro- and fiscal variables turned into highly significant drivers. Hence the changing nature of drivers could be an explanation of the ongoing empirical debate about the fundamentals determinants of government yields. Their excessive panel estimation model includes VIX, bid-ask spread, forecasted budget balance to GDP, debt-to-GDP forecast, real exchange rate and industrial production growth.
Besides focusing on fiscal variables, recent papers examine the role of the external sector. To overcome model uncertainty, Maltritz (2012) uses Bayesian Model Averaging to examine the relation between sovereign spreads and a wide range of variables. Considering 10 EMU countries between 1999 and 2009 they concluded that one of the important drivers are budget balance to GDP and current account balance, furthermore these ratios they also serve as early warning signals.
Less extensive empirical literature is focusing on adding demographical drivers as determinants of government yields as measurement errors and lack of data availability makes it difficult to research. Overcoming limitations of dataset, Rawdanowicz et al., (2017) selected demographic variables as the third category of drivers to investigate the reasons of fall in government yield determinants for seven countries; Canada, France, Germany, Italy, Japan, UK and US. Applying country-specific error-correction models starting from 1970 they analysed dependency ratio, life expectancy, household saving and share of 40-64 population in total population. The most outstanding results belong to dependency ratio, indicating that
ceteris paribus; one basis point increase in dependency ratio results in 3.02 basis point
9
relation to bond yield. They found that demographic fluctuations such as changes in a population’s age structure can explain the persistent co-movement between real stock and nominal bond yields.
After all, in the debate of choosing the key drivers, certain realized and forecasted macroeconomic, financial, fiscal, and trade related drivers are frequently used in researches with highly significant correlation to government yields. As there is no agreement on which drivers are the best, this study aims to include several frequently used variables together with less frequent demographic ratios to examine elaborately the fundamental drivers of government bond yields. .
2.2
ASSET BUBBLES AND GSADF APPLICATIONS
While there is no general agreement among researchers on what is the best way to empirically identify asset price bubbles, one generally accepted interpretation is that bubbles arise if the price of an asset suddenly and significantly exceeds the asset’s fundamental value followed by an equally prompt collapse. Thereby the value of an asset in a bubble is very transitory; in a boom period investors tend to be overly optimistic, yet prospects of the asset’s future price are getting pessimistic when the bubble burst. (Blot et al., 2017; Contessi and Kerdnunvong, 2015). Price distortions implicated by price bubbles can cause capital misallocations, deepening financial crises, diverging the transmission of monetary policy and ultimately, put the financial stability under threat. (Blot et al., 2017).
An outstanding approach that is successfully applied in various studies to detect multiple bubbles is the Generalized Supremum Augmented Dickey Fuller (GSADF) model, developed by Phillips et al., (2013, 2015). The basis of the model is to analyse the co-movement between the price and the fundamental value of an asset to identify periods of explosive behaviour by extending sample sequence with a range of flexible windows. (Contessi and Kerdnunvong, 2015). The model has been found the best in comparisons; Arshanapalli and Nelson, (2016) reviewed economic tools to detect stock market bubbles in S&P500 index prices and only GSADF was able to detect multiple bubbles, dating also the starting and end periods. The test is widely applied to detect stock and/or housing market (Engsted et al., 2016; Huang and Shen, 2017; Huston and Spencer, 2017), exchange rate (Yang and Oxley (2017) and oil price bubbles. (Caspi et al., 2015).
10
A large set of recent empirical literature has studied the effect of QE and asset bubbles, mainly focusing on US or UK stock and housing markets. Huston and Spencer (2017) argue that there is little indication of asset price bubbles in the equity market after the Federal Reserve’s QE programme. Implementing Campbell-Schiller and GSADF models, they are analysing the US equity, corporate bond and housing markets. Even though GSADF does not find evidence of explosive behaviour in stock and housing markets, it does find a persistent bubble for Treasury bond prices between September 2011 and February 2013.
Lamoen et al., (2017) also applied Phillips’s GSADF model and the Generalized Sup Phillips&Perron (GSPP) approach to find evidence of exuberant price behaviour in the euro area. Analysing 10 euro area countries between 2003 and 2016, they conclude that government bond prices significantly deviate from their fundamental levels in all countries triggered by ECB’ purchase programmes and the results indicate an intrinsic bubble. However, the exuberant price behaviour disappears for almost all countries after using PSPP and ABSPP/CBPP3 announcements as drivers of government yields.
Apart from the well-known severe deteriorate effects, identifying asset bubbles has been a prolonged challenge for academics as there is no generally accepted metrics of a bubble. (Contessi and Kerdnunvong, 2015) Existing literature, focusing on different regions and markets proves that a noble approach, the GSADF test can successfully identify periods of exuberance in prices, indicating asset price bubbles.
2.3
QE’S IMPACT ON ASSET PRICES
There is a vast literature providing evidence that QE programmes have remarkable impact on asset prices. These impacts are all well illustrated by the study of De Santis, (2016); by analysing Bloomberg news coverage with country-panel error correction model, he found that 10-year euro area sovereign yields were reduced by 63 basis points up to 2015 October. The Netherlands had the lowest effect, a fall of 38 basis points whilst the fall in Portugal’s yield was almost triple that of The Netherlands; it declined by 106 basis points. Furthermore, a vast majority of the impact was before the announcement, between September 2014 and February 2015. Correspondingly, Altavilla et al., (2015) found that yields for 10-year maturity government bonds decline by 30 to 50 basis points and almost twice as much in higher yield countries, such as Italy or Spain. Also, the programme was announced when the financial distress was low, as a consequence of early hints in the news about the potential APP, yields
11
were already declining months before the announcement. They used event study methodology by extending a term structure model with bond supply effects focusing mainly on PSPP. Apart from focusing on the Euro area, Mamaysky (2018) analysed QE impacts on asset prices triggered by QE announcement by the Federal Reserve in the US, Bank of England for UK and ECB in the euro zone. Within each region, 6 security types were analysed by event study methodology with extended time horizons. Results show that QE is associated with immediate price reactions in equities and stocks both in country and industry level. Also, the paper reveals the importance of pre-announcement effects and that bond-type securities have faster reaction to QE; especially medium and long-term government bonds.
Contrary to previous conclusions, certain studies did not find evidence of significant impact of QE on asset prices. Blot et al (2017) analysed the impact of monetary policy shocks on asset price by examining stock and housing markets in the US between 1986 and 2016 with an extension of euro area using a restricted sample. Using Principal Component Analysis, their results suggest there is no evidence that the risk of QE would significantly increase stock prices or in other words, inflate bubbles, and while expansionary policy inflate asset price bubbles, restrictive policies are not able to deflate them. Similarly, Haas and Young-Taft (2017) studied the casual links between QE, asset overvaluation and macroeconomic performance, concluding that QE acts as a sort of phrase shift with respect of time. By applying stock-flow consistent model and various equations from Godley and Lavoie’s (2007)
Monetary Economics, they found that QE fed hardly any economic panic and in the long run the effects are almost irrelevant.
Besides diverse model selection and differences in data and time determinants, certain general conclusions can be drawn from the existing literature which found evidence that QE has impact on asset prices. At first, government bond yields are significantly decreased and as a result inverse relationship; government bond prices are significantly increased as a result of QE. Secondly, the rate of the impact of is country-specific. Finally, the importance of pre-announcement effects: the decline of yields have started before the press conference in 2015 January announcing PSPP, as investors discounted the implication of monetary policy before the actual purchases. (De Santis, 2016)
12
2.4
HYPOTHESES
Based on the evidences from existing literature and datasets, the study is focusing on answering three hypothesis, one main and two additional.
Hypothesis 1 (main): The ECB’s purchase programmes causing government bond prices to deviate from their fundamental levels and are responsible for formulating asset price bubbles.
Previous studies concluded that there are divergences between the market value and fundamental value of government bonds as a consequence of QE policies. Considering the adverse effects of QE and the volume by which the ECB engaged in purchase programmes, the study aims to provide further evidence that in the euro area, divergences in the government bond market are so salient that the market prices do not reflect the fundamental value of government bonds anymore. By applying a novel approach for detecting bubbles, the purpose of the research is to examine further the extent of the misalignment between government bond yields and fundamental levels in the euro zone.
Hypothesis 2: Forecast variables are at least as significant fundamental drivers of government bond yield as observed variables.
When investors are building a long-term portfolio they base their borrowing behaviour on forecasts and expectations, therefore forecasted variables should have at least the same significance as realized variables in relation to changes in 10-year maturity government bond yields.
Hypothesis 3: Demographic variables are fundamental drivers of government bond yields.
There is no consensus among academics on what are the key drivers of government bond yields, although after carefully examining the existing literature in the field of fundamental drivers it appears that academics are restricting drivers on macroeconomic, fiscal, financial or trade related, observed or forecasted variables. The study intended to broaden the categories of fundamental drivers and states that demographic variables, such as life expectancy, dependency ratio and population growth have significant impact on bond yields.
13
3 D
ATAThis section reveals the sources of the data, summarizes the variables and provides descriptive statistics. The study covers the period of January 2003 to September 2017 using monthly observations. The ten euro area countries which the analysis includes are Austria, Belgium, Finland, France, Germany, Ireland, Italy, the Netherlands, Portugal and Spain. Harmonized government bond yields are obtained on monthly basis to clearly show the announcement effects from DataStream. All variables are obtained or interpolated (Debt-to-GDP ratio, Population growth, Life expectancy, Dependency ratio) as monthly observations.
3.1
DATA SELECTION AND FUNDAMENTAL DRIVERS
Figure 2 represents the evolution of government bond yields for 10 euro area countries. Yields had similar trends until September 2008, when Lehman Brothers collapsed, followed by the sovereign debt crises in 2010 when the differences in yield levels across countries started to rise. From 2012, at the announcement and during the implementation of asset purchase programmes, bond yields were gradually falling for all countries. That phenomenon is especially outstanding in 2014, when CBPP3 and ABSPP were announced and early hints of PSPP started to spread. In general, after 2015 yields remain below 2% apart from Portugal.
-2 0 2 4 6 8 10 12 14 16 2003.01.01 2003.06.01 2003.11.01 2004.04.01 2004.09.01 2005.02.01 2005.07.01 2005.12.01 2006.05.01 2006.10.01 2007.03.01 2007.08.01 2008.01.01 2008.06.01 2008.11.01 2009.04.01 2009.09.01 2010.02.01 2010.07.01 2010.12.01 2011.05.01 2011.10.01 2012.03.01 2012.08.01 2013.01.01 2013.06.01 2013.11.01 2014.04.01 2014.09.01 2015.02.01 2015.07.01 2015.12.01 2016.05.01 2016.10.01 2017.03.01 2017.08.01 2018.01.01 Yi el ds (%) Date
10 yr Government Bond Yields
France Austria Portugal Ireland Finland Netherlands Germany Spain Italy Belgium
FIGURE 2: GOVERNMENT BOND YIELD LEVELS BY COUNTRY
Notes: Figure 2 illustrates the monthly fluctuations of harmonized 10-year maturity government bond yields between January 2003 and January 2018 for 10 euro area countries.
14
The research adds to the discussion of determining the key drivers of government bond yields by using an extended set of drivers. Table 1 provides a summary of the fundamental drivers, expected signs and sources. The key drivers selected in this paper can be categorized in three groups; traditional variables - which have been used in many researches before, consisting of; fiscal, macroeconomic and financial ratios. The second category includes forecast variables, as forward-looking investors and market participants base their borrowing behaviour on expected variables, therefore it is safe to assume that forecasts have a direct effect on government bond prices, thus yields. (Poghosyan, 2014). In addition, while certain observed ratios such as Current account balance or real GDP growth are only available on quarterly or yearly basis, the forecasted ratios are on monthly basis. Consequently, in the main panel data model, certain historical ratios have been replaced by forecast variables to determine the fundamental level of government bond yields. The correlation between variables have been tested in advance; as a correlation between forecasted and observed variables is possible. Although the correlation across the selected drivers is low, therefore observed and forecasted variables can be both used together in the model specifications. 2
Even though demographic variables are not widely used in similar researches, a life-cycle pattern appears to be a predictable element in government yields (Guzluklu and Morin, 2015). However, datasets for demographic drivers are limited as data collection is highly time consuming, therefore most demographic variables are not updated with the latest observations. For that reason, an extensive analysis is undertaken in section 6 with a limited sample, from January 2003 to December 2016.
Traditional variables and government bond yields are obtained mainly from Datastream on a monthly basis, bar Debt-to-GDP ratio which is obtained quarterly. Monthly real purchase volume of PSPP is gathered from the ECB in euro billions. Demographic variables are retrieved from Datastream on yearly basis and interpolated as monthly observations. Forecast variables are collected on monthly basis from Consensus Economics and due to limitation of the data source, expected budget balance is available for France, Germany and Italy while the 1-year ahead long-term government yield forecast variable is available for France, Germany, Italy, the Netherlands and Spain. Both variables are used in section 6, for an alternative
2 The correlation between forecasted government bond yield and PSPP net volume variables are significant,
therefore in section 6, the model specification excludes the net volume of PSPP functions to avoid multicollinearity.
15
analysis. Regarding government bond yields, the study uses monthly observations of harmonized 10-year maturity government bond yields.
Categories Variables Description Expected
Sign Source Traditional Drivers (Fiscal, Macroeconomic, & Financial)
Debt-to GDP ratio Gov. Debt as a percentage of GDP, quarterly
(+) Datastream
Inflation Harmonized Consumer Price Index,
% monthly
(+) Datastream
VIX Implied volatility Index, last day of
VIX close,% monthly
(+) CBOA
EONIA Euro Overnight Index Average, %
monthly
(+) Datastream
PSPP net volume Net volume at book value of PSPP, in bn., monthly
(+) ECB
Demographic Drivers
Dependency ratio Dependent/Working age population,
% yearly
(+) Datastream
Life expectancy Life expectancy at birth in years, yearly
(-) Datastream
Population grotwh Change in growth of total active populations, % quarterly
(-) Datastream
Forecast Drivers
Current Account Current account forecast in bn.,
monthly
(+/-) Consensus
Economics
Consumer Price Changes in consumer price (+) Consensus
Economics
Real GDP growth Forecasted change in real GDP,
monthly %
(+/-) Consensus
Economics
Budget Balance Forecast of budget balance (fiscal
year) in bn. monthly
(+) Consensus
Economics
Gov. Bond Yield 1 year forecast of 10-year
government bond yields, monthly %
(+) Consensus
Economics
TABLE 1: DESCRIPTION, EXPECTED SIGN AND SOURCE OF FUNDAMENTAL DRIVERS OF GOVERNMENT BOND YIELDS
16
Debt-to-GDP ratio, a well-known fiscal driver, is the general government debt as a percentage of GDP and obtained on quarterly basis. The ratio is expected to be positively related to government bond yield, because an increased debt level signals higher risk to the market participants, thus following the risk/return trade-off, which results in an increased yield level (Akram and Das, 2017; Ciocyte et al, 2016; Poghosyan, 2014).
Regarding macroeconomic drivers, inflation is one of the most common variables. A standard measure of inflation in the euro area is the Harmonized Index of Consumer Prices (HICP), on all items on a monthly basis. Inflation inflicts serious damage on the purchasing power of the bond’s cash flow, thus having a decreasing effect government bond prices. As investors require compensation for the inflation risk, a rise in inflation level is expected to increase bonds yields. (Ciocyte et al., (2016).
There are two widely used financial market variables. The first being the Euro Overnight Index (EONIA) which is the weighted average of the overnight cost of lending by euro-area banks and is a proxy for risk free rate. The second variable is the VIX index, which is a proxy for global risk-aversion and stands for the Chicago Board Options Exchange implied volatility of S&P500 index options. Both are expected to have a positive correlation, as higher risk-free rate and risk aversion should result in an increased level of bond yields.
The net volume of PSPP purchases are reaching more than €1.56 trillion in total for the 10 selected countries. As the ECB just recently presented the real net volume per country, this seems to be the first paper including volumes as a fundamental driver of government bond yields. Both in theory and as one would assume, higher net volume results in higher yields. Three demographic variables are included in the extended model estimation. Dependency ratio, which measures the percentage of dependant population or in other words; the none working age, over the percentage of the working population; which is defined as residents between age 16 to 65, on total population. It is expected to positively affect bond yields, as middle-aged population have the highest propensity to save, indicated by life-cycle hypothesis (Bean et al, 2015; Rawdonowicz et al, 2017). The ratio is higher (lower) if the dependent population increases (decreases) or the working population falls (rises) which means a decrease (increase) in household savings and wealth and an increase (decrease) in yields. While dependency ratio is expected to be positive, the rest of the demographic variables should have the opposite sign; life expectancy and population growth. As population growth
17
shows a decreasing trend in advanced economies because of falling fertility rates and increased longevity, the dependency ratio is raising, reducing savings and increasing bond yields. (Rawdonowicz et al, 2017).
Forecasts of current account and real GDP growth are added to the model, in place of historical observations. A common trade-related variable is current account ratio, which is the current account as a percentage of GDP. The expected sign of the ratio in relation to government bond yields is controversial; on one hand, it can be expected to have positive effect on bond prices as it is an indicator of competitiveness and of a country’s ability to raise funds for debt servicing. As reducing borrowing costs and sovereign risks, an increase in the ratio is expected to cause bond yields to decrease. (Giordano et al., 2012, Poghosyan, 2014). On the other hand, as pointed out by Maltritz, (2012), a surplus in current account could signal a capital flight or the country’s inability to borrow from abroad, either of which could increase government bond yields. (Giordano et al., 2012)
Changes in the consumer prices variable measures a country’s weighted average price of selected consumer products and services. It measures the monthly expected change in prices relative to the benchmark year of 2003. As changes in price is associated with the cost of living, the ratio is often used to identify periods of inflation or deflation. Therefore, just as inflation, when the ratio rises, government bond yields are expected to increase. Another macroeconomic driver is the real GDP growth, which is the expected monthly change of a country’s GDP relative to the benchmark year, 2003. There is no clear expected sign of the ratio. Slower real GDP growth is expected to have negative correlation with government bond yields. (Rawdanowicz et al, 2017; Costantini et al, 2014; van Lamoen et al, 2017). Alternatively, potential growth rate reflects economics health of a country and positive relatinship with government bond yields. (Poghosyan, 2014)
Budget balance is a popular fiscal driver which is directly linked with sovereign default risk and is forecasted monthly in euro billions. Based on empirical literature, it is assumed that when the budget balance of a country’s government decreases and/or public debt increases, government bond prices fall, hence yields increase. (Afonso and Rault, 2015; Costantini et al., 2014)
The paper analyses the 1-year forecast of government bond yields as a driver of government bond yields. Needless to say, it is expected that the government bond yields increase when the forecasted yields are increasing and vice versa.
18
3.2
DESCRIPTIVE STATISTICS
Table 2 represents the descriptive statistics (mean, standard deviation, minimum and maximum value and the number of observations) of each traditional variable and the government bond yield. While bond yield averages are similar, ranging from 2.53% in Germany to 4.96% in Portugal, there are relatively high variations in maximum and minimum observations. Unsurprisingly, Portugal and Ireland have the highest observed yields, 13.85% and 12.45% because of weak economic performance and rising concerns of heavily indebted countries in the euro zone. Meanwhile, Germany had negative yields between June and September 2016. On average, inflation, measured by HICP is relatively modest; it is in the range of 0.12 and 0.17, except for Belgium which has by far the highest mean of 0.88. The sovereign debt as a percentage of GDP is the highest for Italy (96.76%) followed by Belgium (85.03%) which indicates insufficient debt policies as countries with high debt-to-GDP signal that they might be unable to pay back their debts thus having higher risk of default. EONIA is lower than the average of government bond yields and along with VIX index, are obviously the same across all countries.
Table 3 shows the descriptive statistics of forecast variables. Expected current account balance as a percentage of GDP is the lowest for France and more Southern European countries; Spain, Italy and Portugal. Amongst these countries, Spain has the lowest ratio, coming in at -114.23%, which indicates serious current account deficits. Simultaneously, Germany has a notable surplus, 145.31% on average with highest value of 272.13%. Real GDP growth is the lowest for Portugal (0.14%) and for Italy (0.21%) whilst Ireland has the highest mean (2.07%), standard deviation, minimum and maximum values. Changes in consumer prices forecast has similar distribution across countries, on average it is between 1.40% (France) and 1.93% (Spain). Budget balance forecast, measured in euro billions is available for France, Germany and Italy, while all countries are predicted to have negative balance on average, France and Italy are predicted to have negative balance throughout the whole sample. For all countries, the mean value of the 1-year ahead government yield forecasts are higher than the actual observations. In the cases of Germany and the Netherlands, the forecasted mean is 0.45% higher than the historical values while for Italy it is remarkably accurate, the difference is only 0.08%. This phenomenon could indicate that the yields of core countries in the euro zone are tend to fail more to comply with the expectations, while for peripheral countries, such as in Italy or Spain, the unpleasant forecasts are more in line-with-the-actuals.
19
Gov. Yield (%) Inflation (%) Debt-to-GDP ratio (%) EONIA (%) VIX
Country Mean SD Min Max Obs. Mean SD Min Max Obs. Mean SD Min Max Obs. Mean SD Min Max Obs. Mean SD Min Max Obs.
AT 2.85 1.40 0.11 4.80 177 0.16 0.44 -1.4 1.4 177 63.23 5.29 52.8 69.9 177 1.27 1.44 -0.36 4.3 177 18.95 8.27 9.51 59.89 177 BE 3.05 1.38 0.15 4.85 177 0.17 1.01 -2.1 2.5 177 85.03 5.25 52.8 92.7 177 1.27 1.44 -0.36 4.3 177 18.95 8.27 9.51 59.89 177 DE 2.53 1.45 -0.15 4.56 177 0.12 0.41 -1.2 1.2 177 49.59 4.73 39.9 56.3 177 1.27 1.44 -0.36 4.3 177 18.95 8.27 9.51 59.89 177 ES 3.74 1.31 1.01 6.79 177 0.16 0.81 -2.5 2.4 177 53.40 19.67 28.1 83.6 177 1.27 1.44 -0.36 4.3 177 18.95 8.27 9.51 59.89 177 FI 2.74 1.40 0.06 4.78 177 0.13 0.35 -0.7 1.2 177 37.66 7.42 22.8 52.8 177 1.27 1.44 -0.36 4.3 177 18.95 8.27 9.51 59.89 177 FR 2.87 1.31 0.15 4.73 177 0.12 0.36 -1.1 0.9 177 66.89 11.92 49.8 84.8 177 1.27 1.44 -0.36 4.3 177 18.95 8.27 9.51 59.89 177 IE 4.04 2.27 0.40 12.45 177 0.88 0.43 -1.0 1.2 177 41.86 16.76 18.2 70.9 177 1.27 1.44 -0.36 4.3 177 18.95 8.27 9.51 59.89 177 IT 3.84 1.23 1.18 7.06 177 0.15 0.92 -2.5 2.5 177 96.76 11.78 52.8 114.1 177 1.27 1.44 -0.36 4.3 177 18.95 8.27 9.51 59.89 177 NL 2.73 1.40 0.03 4.73 177 0.13 0.58 -1.6 1.5 177 45.19 6.08 34 55.7 177 1.27 1.44 -0.36 4.3 177 18.95 8.27 9.51 59.89 177 PT 4.96 2.48 1.74 13.85 177 0.14 0.58 -1.5 2.2 177 60.82 11.15 41.6 76.3 177 1.27 1.44 -0.36 4.3 177 18.95 8.27 9.51 59.89 177
TABLE 2: DESCRIPTIVE STATISTICS OF GOVERNMENT BOND YIELDS AND OBSERVED VARIABLES PER COUNTRY
Notes: Table 2 shows the mean, standard deviation, minimum and maximum values and the number of observations for the 10 year government bond yield, inflation, debt-to-GDP ratio, EONIA and VIX index from January 2003 until September 2017 for 10 euro area countries. VIX index and EONIA are identical across countries
20
Current Account forecast (%) Real GDP growth forecast (%) Consumer Price forecast (%) Budget Balance forecast (%) Gov. Yield forecast (%)
Country Mean SD Min Max Obs. Mean SD Min Max Obs. Mean SD Min Max Obs. Mean SD Min Max Obs. Mean SD Min
Max Obs. AT 5.71 4.36 -1.94 14.63 177 1.30 1.37 -3.96 3.26 177 1.78 0.60 0.30 3.46 177 - - - - - - - - BE 3.76 5,54 -9.99 14.60 177 1.12 1.32 -3.76 2.64 177 1.77 0.86 -0.01 4.53 177 - - - - - - - - DE 145.31 65.79 38.92 272.13 177 1.19 1.74 -5.90 3.56 177 1.43 0.66 0.25 2.92 177 -37.82 42.86 -131.46 20.18 177 2.98 1.41 0.23 4.87 177 ES -33.48 40.64 -114.34 22.44 177 1.34 2.06 -3.78 3.78 177 1.93 1.40 -0.86 4.55 177 - - - 3.91 1.08 1.45 5.67 177 FI 3.65 4.69 -3.68 12.11 177 1.37 1.97 -6.66 4.65 177 1.56 0.88 -0.13 3.83 177 - - - - - - - - FR -22.73 20.05 -50.87 29.81 177 1.03 1.11 -2.90 2.48 177 1.40 0.75 0.06 3.17 177 -81.28 30.38 -161.56 -46.16 177 3.22 1.28 0.56 5.0 177 IE 0.99 7.97 -13.99 23.12 177 2.07 3.18 -8.46 5.91 177 1.51 1.69 -3.39 4.22 177 - - - - - - - - IT -12.52 31.13 -64.58 43.79 177 0.21 1.56 -5.10 1.94 177 1.69 0.95 -0.14 3.58 177 -50.10 15.05 -86.01 -29.14 177 3.92 1.03 1.34 5.70 177 NL 41.48 18.81 7.61 74.23 177 0.94 1.67 -4.73 3.07 177 1.57 0.65 0.32 2.71 177 - - - 3.18 1.38 0.29 5.02 177 PT -8.55 7.25 -23.78 1.72 177 0.14 1.81 -4.27 2.56 177 1.66 1.09 -0.79 3.46 177 - - - - - - - -
TABLE 3: DESCRIPTIVE STATISTICS OF FORECAST VARIABLES PER COUNTRY
Notes: Table 3 shows the mean, standard deviation, minimum and maximum values and the number of observations for the current account, real GDP growth, consumer price, budget balance and 1-year ahead 10-year maturity government bond yield forecasts from January 2003 to September 2017 for 10 euro area countries. Budget balance forecast is available for Germany, France and Italy and the forecast of government bond yields are for Germany, Spain, France, Italy and the Netherlands. Both variables are excluded from the main analyses and used in Section 6, for an alternative model.
21
The net volume of PSPP at book value is presented in Table 4 for each country. Based on the table it is clearly that volumes significantly differ across countries. While average purchases for Germany and France are notably higher than the rest of the sample, €13.73bn and €11.15bn, for Finland and Ireland the purchases are less than €1 billion; €0.87bn and €0.76 respectively.
PSPP net volume (in euro billions)
Country Mean Standard Deviation Minimum Maximum Observations
AT 1.54 0.31 1.06 0.21 31 BE 1.94 0.39 1.36 2.70 31 DE 13.73 2.92 9.80 19.57 31 ES 6.85 1.44 4.88 9.62 31 FI 0.87 0.25 0.20 1.37 31 FR 11.15 2.18 8.09 15.40 31 IE 0.76 0.19 0.49 1.11 31 IT 9.70 1.97 6.72 13.44 31 NL 3.07 0.66 2.17 4.36 31 PT 0.95 0.31 0.41 1.45 31
TABLE 4: DESCRIPTIVE STATISTICS OF PSPP NET VOLUME PER COUNTRY AT BOOK VALUE.
Notes: This table provides the mean, standard deviation, minimum and maximum values and the number of observations for the net volume of PSPP at book value from March 2015 until September 2017 in euro billions for 10 euro area countries. The data is obtained from ECB.
Table 5 represents demographic variables for each country. Dependency ratio and life expectancy have similar distribution across countries. The former is between 47.31% for Spain and 55.40% for France, whilst the latter has the highest volatility and lowest mean of 79.14 years in Portugal and the highest mean of 81.65 years in Italy. Regarding population growth, Portugal is the only country having a negative average of -0.06%. Meanwhile the highest value belongs to Ireland, 1.35%, which shows a significant average increase in the population, relative to the rest of the sample.
22
Life expectancy (years) Population growth (%) Dependency ratio (%)
Country Mean SD Min Max Obs. Mean SD Min Max Obs. Mean SD Min Max Obs.
AT 80.11 0.91 78.63 81.84 177 0.55 0.27 0.24 1.13 168 48.19 0.67 47.33 49.53 168 BE 79.63 1.02 78.00 81.29 177 0.56 0.11 0.40 0.75 168 54.27 6.32 51.95 81.29 168 DE 79.89 0.88 78.38 81.09 177 0.01 0.63 -1.85 0.98 168 51.10 1.04 48.92 52.34 168 ES 79.89 0.89 78.38 81.09 177 0.82 0.84 -0.33 1.85 168 47.31 2.28 44.94 51.39 168 FI 79.69 0.92 78.37 81.39 177 0.39 0.08 0.24 0.48 168 52.51 3.15 49.8 59.12 168 FR 81.08 1.01 79.12 82.67 177 0.56 0.11 0.40 0.75 168 55.40 2.26 53.54 60.05 168 IE 79.73 1.14 78.00 81.50 177 1.35 0.93 0.22 2.89 168 47.99 3.51 44.51 54,52 168 IT 81.65 0.89 79.98 83.49 177 0.43 0.34 -0.17 1.16 168 52.82 0.83 51.95 54.80 168 NL 80.17 0.97 78.49 81.71 177 0.38 0.11 0.16 0.53 168 49.85 1.97 47.76 53.75 168 PT 79.14 1.23 77.22 81.52 177 -0.06 0.31 -0.55 0.38 168 50.73 1.80 48.19 53.8 168
TABLE 5: DESCRIPTIVE STATISTICS OF DEMOGRAPHIC VARIABLES PER COUNTRY
Notes: Table 3 represents the mean, standard deviation, minimum and maximum values and the number of observations for life expectancy, population growth and dependency ratio for 10 euro area countries between January 2003 and September 2017, except from Population growth and Dependency ratio that are obtained from January 2003 until December 2016.
23
4 M
ETHODOLOGYThis section reviews the econometric models to identify fundamental drivers and exuberance or bubble behaviour in government bond yields for the 10 countries.
4.1
IDENTIFYING FUNDAMENTAL YIELDS
The paper uses panel data estimations expanded with response functions to identify the fundamental level of government bond yields. The following panel data estimation (1.a), is the basic reduced form model from which several model specifications are developed. It includes the traditional and the forecast drivers to identify the estimated yields.
is the dependent variable and stands for the government bond yield for each country i in each time period (month) t, is the Debt-to-GDP-ratio, is the inflation,
is the risk-free rate, is the volatility index, is the forecasted real GDP growth, is the forecasted current account ratio, are time fixed-effects is the country-specific unobserved heterogeneity and is the error terms, assumed that .
An alternative model (1.b), which includes demographic variables in the panel data estimations; stands for dependency ratio, is life expectancy and is the total population growth.
As it is expected to have a structural change among drivers after the sovereign debt crises in 2010, (Giordani et al., 2013; De Haan et al., 2014; van Lamoen et al., 2017) a dummy variable was created for the crises (equation 2) and to not only analyse the relation of the crisis and the government yields, the study is expanded with interaction terms of observed and
24
traditional macroeconomic, fiscal and trade related variables to analyse whether there is a structural change in drives before and after 2010 (equation 3).
Where represents the 2010 sovereign crises dummy.
To reflect the announcement effects of PSPP, CBPP3/ABSPP, CBPP2 and CBPP1, the previous panel data model is expanded with the response functions for each purchase programme in the equation (4).
To represent the effect of the announcements, separately in each purchase programme, ten dummy variables are created. These ten dummies (p=10) are covering the following different time periods: the first response (dummy 1) represents the actual plus the next month of the announcement of the purchase programme, response 2 captures the effects 2 and 3 months after the announcement, response 4 shows the effects 4 and 5 months after the announcement and so on until response 10, which records effects after 18 and 19 months after the announcement. With this method, besides capturing the short-term announcement effects, the medium and long term effects of APP are also analysed.
Where is the announcement variable and k represents the period of the
announcement at time t. In case of PSPP; which is responsible for by far the highest volume of APP, the net volume variable for each country is added to the regression
estimation.
There are various modifications of the panel data estimation, which is discussed more detailed in the robustness checks, section 6
25
4.2
IDENTIFYING ASSET BUBBLES – THE GSADF TEST
To detect asset bubbles, GSADF procedure is applied based on Phillips et al., (2013, 2015) studies. As a consequence of the inverse relationship between bond yields and prices, the inverse of government bond yields are used as a proxy for price movements. (van Lamoen et al., 2017). The method can derive a ratio between fundamental and observed yields thus to analyse the extent by which corrected time series for fundamentals diverge from observed and express explosive behaviour. GSADF is based on dividing the dataset into subsamples and recursively applying Augmented Dickey-Fuller (ADF) unit root test repeatedly. (Phillips et al., 2015) ADF test regression can be written as:
(5)
Where is the first difference of the corrected asset is price, and are fractions of the time window to show the start and end point of a subsample, where is the fundamental yield and is the observed yield level, k is the lag order and are the lagged dependent variables. The ratio of BY*/BY equals 1 if the observed yield is described perfectly by the fundamental yield. The ratio increases in case the observed yields decrease (government bond prices increase) faster than the fundamental yield and that is what GSADF test examines; the extent which the observed time series for fundamentals show explosive behaviour (van Lamoen et al., 2017). GSADF is the generalized form of a right-tailed unit root test, the backwards supremum ADF. (Phillips et al., 2011) The Sup Augment-Dickey Fuller (SADF) test is:
The difference between SADF and GSADF is represented by figure 4, the evolution of GSADF. While the former has a fixed starting point of the sample, the latter lets the starting and ending points vary. The Backward Sup ADF (BSADF), as its name suggests, uses backward expanding sample sequence with varying starting point from 0 to r2-r0 and fixed end point at r2. Comparing
26
detect bubbles is significantly increased. (Phillips et al., 2015) BSADF test statistics can be written as:
(7)
An even more evolved method, the GSADF varies both r1, the starting, and r2, the ending points.
Besides GSADF has greater statistical power by using more extensive subsamples than SADF or BSADF, this approach can analyse long time series in rapidly changing market data where there are multiple periods of exuberant behaviour expected. The usage of subsamples is a key component of detecting bubbles; as the number of extreme values would be lower when the full sample is analysed, it prevents the test from identifying boom or burst periods. With GSADF, multiple bubbles can be detected in continuous time- and date-stamping cycles as the test is able to distinguish a unit root process to a stochastic process with explosive behaviour by extending sample sequence with range of flexible windows. (Phillips et al., 2013, 2015). The test statistics of GSADF is the sup BSADF:
(8)
The alternative hypothesis of the GSADF tests stands for an explosive period, hence by rejecting the null hypothesis one can conclude that there is empirical evidence of explosive price behaviour.
Unit root
Explosive price behaviour
27
(c) The BSADF test (d) The GSADF test FIGURE 3: THE DEVELOPMENT OF GSADF TEST
Notes: The figure shows the ADF, SADF, BSADF and GSADF tests. Source: Phillips et al. (2015)
Phillip’s et al (2015) suggests a date stamping methodology to identify the beginning and ending of exuberant price behaviour. The method is based on BSADF test and relies on comparing the critical values with the BSADF statistics; the start of a bubble is defined as the point when the BSADF test result is higher than the critical value (equation 9) while the bubble ends at the point when the BSADF statistic is lower than the critical value (equation 10).
(9) (10)
Where is the 100(1- )% critical value of the SADF test based on {Tr2} observations.
There are different statistical programs which allow one to perform the GSADF test, although based on Caspi’s (2017) research, EViews simplifies the complex procedure with a user-friendly add-in, the Rtadf, which stands for right-tail augmented Dickey-Fuller test. To perform the test, firstly the add-in calculates the relevant test statistics, according to the selected test. Then the add-in runs Monte Carlo simulations to derive the corresponding exact finite sample critical values under the assumption of Gaussian innovations. The final sample critical values are generated by 2,000 random walk processes. Besides reporting GSADF test results, the software also illustrates the exact periods of exuberance by using date-stamping method based on BSADF tests. (Caspi, 2017).
28
5 RESULTS
5.1
FUNDAMENTAL DRIVERS OF GOVERNMENT BOND YIELDS: OBSERVED ANDFORECASTED VARIABLES
Table 6 serves as an overview of government bond yields and its correlation with fundamental drivers and response functions of purchase programs. Columns 1 to 5 represent the basic models, where time and fixed effects are gradually added to the model, also incorporating the sovereign debt crisis in 2010. Columns 6 to 10 represent the effects of different purchase programs on government yields.
In column 1, which stands for the starting point of the research without time and fixed effects, EONIA and Consumer price forecast are significant at 1% level and both have the expected positive sign whilst real GDP forecast is highly significant throughout the panel data estimation and have negative sign. Current account forecast is significant at 1% level and stays negative, although with a modest amount; it ranges from -0.006 to -0.002. The highly significant and positive correlation of EONIA throughout the model and the insignificant results of inflation is in agreement with the findings of van Lamoen et al (2017) although, while current account forecast stays mostly negative (with the exception of column 5), the current account ratio of van Lamoen’s results turned positive after column 1. However, the difference can arouse from the fact that this research uses current account forecast instead of historical observations as the other study did.
In the panel data regression with fixed effects (Column 2), EONIA, Consumer price and Current account forecast are slightly increased relative to column 1 and Debt-to-GDP ratio becomes highly significant with the expected positive sign and with little variation throughout the model specifications; it stays in the range of 0.027 (Column 5) to 0.033 (Column 7). Column 3 catches the effects of the sovereign debt crisis in 2010. The variable is significant at 1% level although regarding other variables, there is only a minor slope difference compared to column 2. The most significant change is, contradicting to the expectations and previous empirical findings, VIX index is highly negatively significant in columns 3-4 and 9-10. However, it has a minor negative effect, -0.0118 in column 3 and -0.0116 in column 4. To test the possibility that government bond yields and their key drivers had a structural change after the sovereign crisis, in column 4, the interaction between key drivers and the sovereign crisis dummy are estimated. In order to test
29
for the joint significance of the crisis variables and interactions with key drivers, various F-tests have been performed. When the F-test is performed for all coefficients and the crisis variable, the results show joint significance at 5% level (F-value is 2.19). When including drivers separately, real GDP forecast, Debt-to-GDP ratio, Expected current account and consumer price are all significant at 1% level (F-values are 341.78, 219.63, 5.39 and 83.94 respectively). Both real GDP and consumer price forecast change to be positive after 2010, significant at 5% and 1% level. These results provide evidence of structural changes in the interaction of government bond yields and key drivers and are in line with the findings of van Lamoen et al (2017), De Haan et al (2014) and Giordano et al., (2013).
After controlling for time fixed effects in column 5, the expected signs of the drivers remain the same as before, similarly to van Lamoen, with the exception of expected current account, that changes to be positive, and consumer price forecast, that turns into negative. However, both drivers turned back to their expected sign from column 6. The most significant change is in EONIA, which almost doubles.
In column 6 to 10, time response variables for each purchase programs are included, where 10 response dummies are created and each consist of a period of two months. These dummy variables capture the cumulative effect of government bond yields relative to the period before the announcement. Response 1 starts with the month of the announcement of the purchase program while the rest of the responses account for the following 2 months.
In column 6, after the announcement of the PSPP, monthly time effects are added to the regression. All 10 response functions are significant and are between -1.415 and -0.606. Thus these results indicate that government bond yields dropped after the announcement of PSPP. This outcome is in line with the findings of Lamoen et al, (2017) and De Santis (2016). Although, in comparison with the paper by van Lamoen et al, their model using 5 response functions with highly significant, negative and decreasing outcomes (from -2.316 to -1.677), the results with 10 responses show that the two highest effects occurred at response 1 and 2, at and right after the announcement and are also having a decreasing trend until response 6. Then the negative effect increases, from -0.606 in response 6 to -1.190 in response 10, which indicates that instead the decrease of government bond yields as a response of QE programs are fading away, the negative effects are long-lasting. That phenomenon is also present at column 7 and 8. In column 7, the net volume of PSPP purchases are included in the panel data estimation. While in general, adding
30
volumes have a meagre effect on the slope of the rest of the coefficients, current account forecast becomes insignificant. As expected, the PSPP net volume is positively related to government bond yields and the all response functions are significantly negative. However, comparing to column 6, in general, the effects are less negative.
In column 8 response functions include the announcement of CBPP3 and ABSPP. There are only modest changes in drivers, the highest variation occurs for consumer price forecast, which increases to 0.311. Again, all response functions are negatively significant from -0.595 to -1.375.
Here, the highest negative effects are in response 3 (-1.169) and 4 (-1.375), suggesting that few months were required for investors to react and government bond prices to increase (yields to decrease). The outcome of this study shows more moderate negative effects for PSPP and CBPP3/ABSP, comparing to van Lamoen’s, where the volume of the responses are doubled or sometimes tripled. Again, the results present a supporting evidence on the hypothesis that QE programs have a strong negative impact on government bond yields.
In columns 9 and 10, the effects of two terminated programs, CBPP2 and CBPP1 are analysed. Whilst the key drivers remain similar as before, the response functions are contradictory to the expectations. CBPP2 responses positively significant with a decreasing trend from 2.091 in response 1 to 0.517 in response 6. As CBPP2 is the least significant in purchase volume, the dummy variables might catch other events that caused the government yield to increase. Regarding CBPP1, response 1 to 4 are negatively correlated with government yields, in line with the previous findings. Nonetheless, responses 5 to 7 are positive, showing that in certain months the response functions are reflecting certain events that the panel estimation model is not accounting for.
All in all, observed macroeconomic, fiscal and financial variables are quite robust throughout the models although forecasted indicators are highly significant for all model specifications supporting the hypotheses that forecasted variables are as adequate drivers of government bond yields as observed variables. In general, Table 6 indicates that CBPP1 is responsible for short-term decline in yields while PSPP and ABSPP/CBPP3; having 10 highly significant response dummies, are causing government bond yields to decline both in short and long-term. Furthermore, there is evidence of fall in yields since the announcement of CBPP1 in May 2009, which shows that the initial announcement of QE already had a strong effect on government bond-yields.
