• No results found

Effects of ECB monetary policy on eurozone stock market volatility

N/A
N/A
Protected

Academic year: 2021

Share "Effects of ECB monetary policy on eurozone stock market volatility"

Copied!
39
0
0

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

Hele tekst

(1)

Effects of ECB monetary policy on eurozone stock

market volatility

Piet Dijkema

*

January 8, 2021

Abstract

The effect of ECB monetary policy surprises on the VSTOXX index is estimated using a factor model, which leads to the extraction of the three most important monetary policy instruments. I decompose the VSTOXX into a risk aversion and uncertainty part. It is shown that monetary policy surprises affect risk aversion and explain changes of the VSTOXX index on monetary policy meeting days, with changes in the short-term rate and quantitative easing having the largest effect. The effects on risk aversion and the VSTOXX index are greater after the introduction of quantitative easing as a policy instrument.

Keywords: Monetary Policy, VSTOXX index, Risk aversion, Uncertainty JEL Codes: E44, E58

Supervisor: V. Chatzikonstanti, PhD

(2)

Contents

1 Introduction 3 2 Literature review 4 2.1 Stock market . . . 5 2.2 Implied volatility . . . 6 3 Methodology 8 3.1 Identification . . . 8 3.1.1 Factor analysis . . . 10 3.1.2 VSTOXX decomposition . . . 11 3.1.3 VAR model . . . 12 3.2 Data . . . 14 4 Results 16 4.1 Factor model . . . 17 4.2 Decomposition . . . 18

4.3 Monetary policy surprises . . . 21

4.4 VAR . . . 27

5 Conclusions 29

A Factor analysis 35

(3)

1

Introduction

The influence of central banks on financial markets is demonstrated in many cases. Re-cently, the European Central Bank (ECB) announced its Pandemic Emergency Purchase Programme (PEPP), which calmed down financial markets, and is the starting point of the recent stock market rally. During the beginning of the COVID-19 outbreak and the subsequent losses of the stock markets all across the world, the VSTOXX index reached its all-time high. The VSTOXX index measures the expected volatility of the Euro Stoxx 50 index, a blue-chip index for 50 sector-leading companies within the eurozone. Major spikes in the index can also been seen when the dot-com bubble burst and during the fi-nancial crisis last decade. Volatility indexes are therefore seen as an “investor fear gauge” (Whaley, 2000). Additionally, major increases in volatility and stock market downturns are observed together, as is the also seen with low volatility and rising stocks; this explains why volatility indexes are known as fear indexes.

In several studies (Christensen and Prabhala, 1998; Jiang and Tian, 2005; Corrado and Miller, 2005; Busch et al., 2011; Bekaert and Hoerova, 2014), implied volatility is found to be a better predictor of future volatility than historical volatility. This implies that volatility indexes do have a financial relevant function, e.g. Bloom (2009) uses the CBOE Volatility Index (VIX) as a measure for uncertainty. The importance of the relationship between implied volatility and monetary policy is stressed by Bekaert et al. (2013). They note that monetary policy and financial stability are interlinked, given the great amounts of liquidity central banks provide to financial markets. In addition, they show that the VIX index also contains a risk aversion component, which is found to be a good predictor of stock returns. Lastly, it is found by Bloom (2009) and Bloom et al. (2018), that economic uncertainty lowers employment. The connection between monetary policy and implied volatility may thus increase the understanding of the effects of monetary policy on the economy and financial markets.

The purpose of this thesis is to examine the effect of monetary policy decisions of the ECB on the EURO STOXX 50 Volatility (VSTOXX) index, which is the 30-day implied volatility of the EURO STOXX 50 index. Measuring the direct impact of monetary policy decisions is not straightforward, as financial markets have already incorporated any expected policy changes. However, it is possible to measure monetary policy surprises, by the intraday reaction of financial markets. This leads to the following research question:

What is the influence of monetary policy surprises of the ECB on the VSTOXX index?

(4)

The ECB started with several programmes like fixed-rate full allotment and Outright Monetary Transactions (OMT). This provided liquidity to banks and financial markets in order to restore financial stability. Another policy measures that is at first sight less clear, is forward guidance. This contains all the communication about future monetary policy changes and the economic outlook of the eurozone, to manage expectations of financial markets. These forms of non-standard monetary policy measures have become standard practice.

In order to answer the research question, I develop a factor model to extract monetary policy surprises from financial markets. Using this approach, I am able to estimate the effects of standard and non-standard monetary policy surprises on the VSTOXX index, in particular policy surprises in the short-term rate, forward guidance and quantitative easing. In addition, I decompose the VSTOXX index into a risk aversion and an un-certainty part to analyse what drives changes in expected volatility. This adds to the literature in various ways. First, the monetary policy literature is mainly focused on the impact of FOMC decisions and not on ECB monetary policy. Second, decomposition of the VSTOXX index and the effects of monetary policy has not been done before. Lastly, it contributes to the understanding of the behaviour of financial markets in relationship to monetary policy.

I find that uncertainty is the largest part of the VSTOXX index, but does not react to monetary policy surprises quickly. However, I show that risk aversion is affected by surprise changes in monetary policy, and explains the initial reaction of the VSTOXX index to policy surprises. In addition, the reaction of financial markets on monetary policy surprises is the largest after the introduction of quantitative easing in 2014.

This thesis has the following outline, section 2 gives a description of the existing literature on the influence of monetary policy on financial markets. In section 3 I discuss the research methodology and give a description of the data that is used. The results are presented in section 4, and section 5 concludes.

2

Literature review

(5)

2.1

Stock market

Most research about monetary policy is done in relationship with the reaction of financial markets, and especially stock market returns. It is shown that stock returns increase after Federal Open Market Committee (FOMC) announcements (Thorbecke, 1997; Cies-lak et al., 2019) and even exhibit a positive pre-FOMC announcement drift (Lucca and Moench, 2015). However, Ozdagli and Velikov (2020) show that stocks who are more positively correlated with expansionary monetary policy, earn lower average returns.

It is found by Bernanke and Kuttner (2005) that the positive stock market reaction is mostly explained by unexpected monetary policy measures. They stress the difficulty of measuring the direct impact of monetary policy on the stock market, as financial markets are forward looking and have already incorporated any information of possible changes in monetary policy. In order to measure unexpected changes, they use changes of Federal funds futures as a proxy for monetary policy, which is a method that is widely adopted to analyse the surprise impact of FOMC meetings on returns and volatility of stock markets, see for example Haitsma et al. (2016) and Krieger et al. (2012).

The major task in conducting research on monetary policy is to identify monetary policy shocks, certainly after the introduction of non-standard monetary policy measures1.

The availability of high-frequency data has made it possible to measure the immediate impact of monetary policy announcements on financial markets, with G¨urkaynak et al. (2005) being one of the most influential earlier studies on the subject. They introduced a factor model to let the data determine the different dimensions of monetary policy. This approach is followed by other researchers in later work on the subject (Altavilla et al., 2019; Swanson, 2020). In fact, I also follow this factor model approach. The advantage of using a factor model, is that is not necessary to make prior assumptions about the relevant financial variables that are used to measure monetary policy, the data will determine the variables that are most influenced by monetary policy changes. However, it is feasible to choose a set of variables that are relevant from an economic perspective, i.e. financial instruments that are directly influenced by monetary policy2. The specific variables that

I use in this thesis, are discussed in the methodology section below.

Monetary policy announcements do not only contain pure policy announcements, but also convey information about the economic outlook of the central bank (Nakamura and Steinsson, 2018; Kerssenfischer, 2019; Jaroci´nski and Karadi, 2020; Andrade and Ferroni,

1Standard monetary policy measures as changing the policy rate, are easily observed by changes of

different interest rate benchmarks. However, non-standard monetary policy measures as quantitative easing and forward guidance, have a more indirect effect on asset prices, and are therefore more difficult to measure.

2The selection of variables is arbitrary. However, I base the decision which instruments to use, on

(6)

2020). These are called information effects or Delphic shocks, and are measured by the reaction of financial markets around monetary policy announcements. The central bank information effects are generally identified as shocks that raise both stock prices and yields. Whereas a monetary policy shocks is seen as a negative co-movement of stock returns and the interest rate. It is found that a policy shock as defined above, lowers output and increases corporate bond spreads. Alternatively, a decline of expected inflation and interest rates, can signal bad news about future macroeconomic conditions. These implicit information contained in monetary policy announcements do fall under the instrument forward guidance, and is becoming an increasingly important monetary policy measure.

2.2

Implied volatility

The amount of literature on the relationship between monetary policy and (implied) volatility is smaller than the one on stock returns. As with most of the literature about the effects of monetary policy, most studies are conducted with respect to U.S. financial markets and FOMC monetary policy announcements. The effects of monetary policy on implied volatility are in line with the results found with respect to stock market returns, i.e. it is found that the VIX declines after FOMC meetings and the subsequent announcements (Nikkinen and Sahlstr¨om, 2004; Chen and Clements, 2007; V¨ah¨amaa and

¨

Aij¨o, 2011; Krieger et al., 2012, 2015; Fernandez-Perez et al., 2017). The research approach in these studies is simple, i.e. linear regression is used to estimate the relationship between VIX index returns and monetary policy meeting days and/or policy surprises. In addition to the effects found on implied volatility, there is evidence for a profitable trading strategy involving VIX futures on an annual basis, which is shown by Krieger et al. (2012) and Fernandez-Perez et al. (2017).

Although it is found in all studies that the VIX declines on FOMC meeting days, the methodologies differ. The research approach of Chen and Clements (2007) and Nikkinen and Sahlstr¨om (2004) is very similar. They both use dummies in their regression model to examine the changes of implied volatility on FOMC meeting days and one day before and after these meetings. It is not taken into account whether these changes come from unexpected monetary policy changes or not. V¨ah¨amaa and ¨Aij¨o (2011) basically extend the research done in these studies by using the Feds fund futures rate to examine surprise changes in monetary policy decisions. Also do they distinguish between scheduled and unscheduled FOMC meetings and is the possible effect of the monetary policy cycle taken into account.

(7)

decisions and the effects on the VIX and VDAX, which is the German equivalent of the VIX. In contrast to the results of the studies mentioned above, they find no statistical significant relationship between the VDAX and ECB Governing Council meeting days. In addition, they don’t find a trend in the magnitude and sign of VDAX changes on meetings days of the ECB, it is argued that this is due to ECB statements being more “neutral” than FOMC statements and the higher frequency of ECB meetings. However, once the surprise element of ECB monetary policy decisions are taken into account, it is found that ECB meetings are uncertainty reducing, i.e. a decline of the VDAX. This effect only holds for negative or no surprises of ECB monetary policy decisions.

The German volatility index is also topic of other studies, F¨uss et al. (2011) and Jiang et al. (2012) investigate the relation between macroeconomic news announcements and implied volatility indexes, including the VDAX. It is found that implied volatility decreases after (scheduled) news announcements. The direct effects of ECB monetary policy meetings are not taking into account. In addition to the other studies mentioned, L´opez and Esparcia (2021)3 find that the VSTOXX and VDAX index significantly decline

on ECB monetary policy meeting days. In fact, ECB meeting days are found to be the most important macroeconomic news release event which influence the VSTOXX index, among other news announcement like employment rate and consumer price index changes. Lastly, it is found by Kerssenfischer (2019) that the VSTOXX index is not significantly effected by a monetary policy news shock, as defined by an increase of the 2-year German bond yield. However, when pure policy or information effects are found, the VSTOXX index increase or decreases by 50 basis points, respectively.

It is shown by Bekaert et al. (2013) and Bekaert and Hoerova (2014) that the VIX index can be decomposed into a risk aversion and a uncertainty part, and find that a loose monetary policy decreases both components. Similarly, Kaminska and Roberts-Sklar (2018) find that realized variance is significantly effected by monetary policy surprises. In addition, there is evidence of the influence of monetary policy decisions on investor sentiment (Kurov, 2010), with the effect being greater in bear markets. However, this measure of investor sentiment is not the same as the measure of risk aversion used in this thesis and the one of Bekaert et al. (2013).

Decomposition of the VIX index into a variance risk premium and a conditional vari-ance part, leads to other economically relevant results. The varivari-ance risk premium, mea-sured as the difference between the variance of a volatility index and conditional variance, is found to be a good predictor of stock markets returns, see for example Bollerslev et al. (2009, 2011) and Chow et al. (2020). In addition, Bekaert and Hoerova (2014) show that conditional variance is a good predictor of economic activity and financial instability. A

(8)

study closely related to this thesis is the research of Gospodinov and Jamali (2015), they analyse the dynamic effect of monetary policy shocks on stock market volatility and the volatility risk premium. To estimate the effect, they develop a VAR model with monetary policy as a exogenous variable; and find a decrease in stock market returns and an increase of stock market volatility, following a monetary policy shock.

In conclusion, there is an extensive literature on the relationship between monetary policy decisions and financial markets, and stock markets in particular. The amount of literature on volatility indexes and monetary policy is much smaller, especially in the context of ECB monetary policy and the VSTOXX index. Additionaly, the results of the few studies that are conducted on the subject are mixed. To extent the literature on monetary policy, I investigate the relationship between the VSTOXX index and ECB monetary policy. In addition, I decompose the VSTOXX index, to examine the influence of monetary policy on its components; which, to my knowledge, has not been studied before.

3

Methodology

In this section I describe the methodology and data of my thesis. Specifically, subsection 3.1 starts with a description of the time window that is used and how it is used to measure monetary policy. Thereafter, it is discussed how the influence of monetary policy on the VSTOXX index is analysed. Lastly, a description of the data is given in subsection 3.2.

3.1

Identification

In order to measure monetary policy, I use a time window consisting of changes in financial instruments around ECB Governing Council monetary policy decisions. The observed changes of financial instruments serve as a proxy for monetary policy surprises. The rationale behind this, is that it is assumed that all prior information and expectations about monetary policy are already reflected in the prices of financial instruments. Thus, only unexpected changes of monetary policy should influence financial markets.

The Governing Council takes its monetary policy decision every six weeks4. On

meet-ing days, monetary policy decisions are made public at 13.45 CET in a press release. The press release does not contain any motivation of the decisions made. However, a press conference at 14.30 CET discusses the monetary policy decisions announced. Dur-ing the press conference, the ECB President explains the monetary policy decisions in an introductory statement, whereafter journalists can ask questions. The time window I use

(9)

starts at 13.25 CET and ends on 15.50 CET, and is constructed such that it captures the reaction, i.e. the change, of specific financial instruments on both the press release and the press conference about the monetary policy decision.

The usage of an intra-day window to measure monetary policy is widely used in the literature, e.g. Andrade and Ferroni (2020), Altavilla et al. (2019) and Jaroci´nski and Karadi (2020). A short intra-day window has the advantage over a daily window, because it allows for better isolation of the effects from monetary policy, as financial markets are also influenced by other types of information, like macroeconomic or political news. As is shown by G¨urkaynak et al. (2005), interest rate changes are accounted for immediately by financial markets. This is confirmed by Altavilla et al. (2019) and Kerssenfischer (2019), in addition they conclude that unconventional monetary policy measures like quantitative easing and forward guidance, take more time to be processed by financial markets, though short enough to be isolated by a short intra-day window. Thus, a short time window is sufficient to capture monetary policy surprises.

The financial instruments that are chosen, are assumed to be the variables that are most influenced by monetary policy decisions of the ECB. The variables that are used in this research, are overnight index swap (OIS) rates and sovereign bond yields. Specifically, I use 1-month, 3-month, 6-month, 1-year and 2-year OIS rates, and the 5-year and 10-year sovereign bond yields of Germany, Italy and Spain. OIS rates and other short-term interest rates like the Federal Funds rate, are widely used in the literature to measure monetary policy changes (Altavilla et al., 2019; Bernanke and Kuttner, 2005; Jaroci´nski and Karadi, 2020). This is due to the fact that the short term interest rate is an important monetary policy instrument.

Another major policy instrument of central banks is forward guidance, which contains all the communication regarding future monetary policy, based on the outlook of inflation. As forward guidance is aimed at the future, longer-term OIS rates are also included in the factor model. Long-term OIS rates are not available for the full period, therefore I use the 5 and 10-year German yields as alternative variables. It is shown that German yields closely resemble OIS rates and therefore used as a proxy for OIS rates in the literature Altavilla et al. (2019); Kerssenfischer (2019).

The third major monetary policy instrument of the ECB is the Asset Purchase Pro-gramme (APP), which consists of four subproPro-grammes aimed at the purchase of public and private debt. From the four different programmes, the public sector purchase pro-gramme (PSPP) is the most important5. The purchase of government bonds influences

the yield of those bonds, which is one of the aims of the PSPP. In order to are able to

5As of December 2018, assets purchased via PSPP make up for around 90 percent of the total

(10)

measure this kind monetary policy, I have included 5 and 10-year government bond yields of Italy and Spain in the dataset. The usage of this set of variables is a new approach to measure monetary policy. As discussed above, usually only OIS rates are used in the context of ECB monetary policy. However, these yields do not accurately represent yield changes of countries that do have high debts. As the APP is an important part of ECB monetary policy, it is important to include yield changes of a country as Italy to measure ECB monetary policy. The details of the time window and the data are discussed in subsection 3.2.

3.1.1 Factor analysis

To extract monetary policy surprises from the data, I follow the approach used by Altavilla et al. (2019), which is build on G¨urkaynak et al. (2005) and Swanson (2020)6. The

approach is based on factor analysis, which is a method to explain a set of variables by a smaller amount of factors. The collected variables are represented in a T × n matrix X, where T is equal to ECB Governing Council monetary policy decision dates and n corresponds to the number of observed variables. The following equation represents the matrix notation of the factor model,

X = F Λ + ε, (1)

where F is a T × k matrix of latent factors (k < n), Λ a k × n matrix of factor loadings, which represent the correlation between the factors and the variables used in the model. ε denotes a T × n matrix of white noise disturbances. The number of factors tell how many dimensions underlay monetary policy decisions, that influence the financial instruments represented in matrix X. If no factors are being found that explain matrix X, then this would imply that the observations can be described by just a white noise process7. Pre-vious studies like the ones mentioned above that use factor analysis to extract monetary policy surprises, do find that at least one factor is underlying monetary policy decisions that explains changes in financial instruments, depending on the time period that is used. To estimate the matrix of factors F and the matrix of factor loadings Λ, it is necessary to determine the number of factors k first. I use Horn’s parallel analysis (Horn, 1965) to decide on the number of factors to be extracted. The results of the factor model can be found in section 4.1 below.

The factors F estimated in equation (1) can be used as predictors in a regression

6The study of Swanson was already publicy available in 2017, but only recently published in The

Journal of Monetary Economics.

7A white noise process has no discernible structure. This means that is a random process, with mean

(11)

model to estimate the influence of monetary policy on the VSTOXX index. The following regression model can be estimated via OLS and is represented by,

∆V ST OXXt= α + β1F1t+ ... + βkFkt+ εt. (2)

Where F is a vector of factor scores (T × 1), the number of factors is equal to k. ∆V ST OXXt is the percentage change of the VSTOXX index on days that ECB

Govern-ing Council decided on monetary policy, i.e. ln(V ST OXXt/V ST OXXt−1) × 100.

3.1.2 VSTOXX decomposition

The EURO STOXX 50 Volatility (VSTOXX) index reflects the 30-day8 expected volatility

of the EURO STOXX 50 Index. This is done by interpolation or extrapolation of two sub-indices. These sub-indices measure the implied variance of all options on the EURO STOXX 50 for a given time to expiration. Thus, every sub-index has a different time to expiry. With this method, the main index does have a fixed time to maturity and is measured as the square root of implied variance. The methodology to calculate the sub-indices is based on a non-arbitrage approach and also used to price variance swaps, implying that the VSTOXX index is a risk-neutral expectation of volatility.

It is shown by Bekaert and Hoerova (2014), that the VIX index9 can be decomposed

into conditional variance and a variance risk premium part. As the VSTOXX index is similar to the VIX index, i.e. it is also measured as the square root of implied variance of a range of options and is model-free, this will also hold for the VSTOXX index. In the literature the variance risk premium is seen as a proxy for risk aversion, whereas conditional variance is seen as a proxy for uncertainty, see Bollerslev et al. (2011), Bekaert et al. (2013) and Bekaert and Hoerova (2014). Additionally, it is shown by Bollerslev et al. (2009), Drechsler and Yaron (2011) and others, that the variance risk premium is a good predictor of stock market returns. These findings make it interesting to examine the impact of monetary policy surprises on the components of the VSTOXX index.

The variance risk premium is generally defined as the difference between conditional variance and a risk-neutral expectation of variance, see also Carr and Wu (2009), Boller-slev et al. (2009) and Bekaert and Hoerova (2014). Using the approach by Bekaert and Hoerova (2014), I measure conditional variance as the expected monthly variance of the EURO STOXX 50 index. Whereas the risk-neutral expectation of variance is equal to the squared VSTOXX index. The variance risk premium (VRP) is the residual of the

8There are different VSTOXX indices, ranging from 30 to 360 days. However I use the main VSTOXX

index (V2TX) that reflects 30-day expected volatility, expressed in annualized terms.

9The Cboe Volatility Index (VIX Index) is the U.S. equivalent of the VSTOXX index. It measures

(12)

squared VSTOXX index and the conditional variance and is computed as follows, V RPt= V ST OXXt2− Et[RV

(22)

t+1 ]. (3)

The value of the squared VSTOXX can easily be derived from the data, and is equal to V ST OXX2/12, where V ST OXX is the closing price on time t. It becomes clear from

the equation that the estimation of the variance risk premium, depends on an accurate forecast of next month realized variance of the EURO STOXX 50 index, Et[RV

(22)

t+1 ], where

month is defined as 22 trading days.

For the estimation of RVt(22), I use a model suggested by Bekaert and Hoerova (2014). The model has the following structure,

RVt(22)= α + β1V ST OXXt−222 + β2RV (22) t−22+ β3RV (5) t−22+ β4RV (1) t−22+ εt, (4)

where RVt(22) is defined as the sum of daily realized variance RV over the past 22 trading days10, i.e. RV(22)

t =

P22

j=1RVt−j+1. All returns computed are expressed in

percent-age form. The variable V ST OXX2

t−22 is the monthly squared VSTOXX index level, i.e.

V ST OXX2/12, and also expressed in percentage levels. The other independent

vari-ables RVt−22(22), RVt−22(5) , RVt−22(1) , are respectively the past monthly, weekly and daily realized variance of the EURO STOXX 50 index.

As described above, the conditional variance (Et[RV (22)

t+1 ]) and variance risk premium

(V RPt) are respectively a proxy for uncertainty and risk aversion. To further examine

the impact of ECB Governing Council monetary policy decisions on the VSTOXX index, I construct two regression models to estimate the impact of monetary policy decisions on risk aversion and uncertainty. The following two equations represent the regression models,

∆RAt= α + β1F1t+ ... + βkFkt+ εt, (5)

∆U Ct = α + β1F1t+ ... + βkFkt+ εt. (6)

Where ∆RAt and ∆U Ctcorrespond to the changes of the variance risk premium and

un-certainty, on the days that the ECB Governing council takes its monetary policy decision. Fkt represents the factors scores estimated in equation 1.

3.1.3 VAR model

In order to determine the dynamic impact of monetary policy surprises, I construct a daily VAR. To identify the monetary policy surprises I use external instruments, based

10Bekaert and Hoerova (2014) use 5-minute intraday returns to compute the daily variance. However,

(13)

on Stock and Watson (2012), Mertens and Ravn (2013), Gertler and Karadi (2015) and Altavilla et al. (2019). The external instruments I use are the monetary policy factors obtained via the factor model in equation 1. The general structural form of the VAR is given by the following equation,

AYt= p

X

j=1

CjYt−j+ t. (7)

The reduced-form VAR model is obtained by multiplying each side by A−1 and is repre-sented by the following equation,

Yt= p

X

j=1

BjYt−j+ ut, (8)

where Yt is a vector consisting of a monetary policy indicator, the log of the VSTOXX

index, the log of risk aversion and the log of uncertainty. ut = St denotes the reduced

form shock. In addition, coefficient matrix Bj = A−1Cj and S = A−1. The

variance-covariance matrix P is represented as follows, E[utu

0

t] = E[SS

0

] =X. (9)

The methodology for identification of monetary policy surprises via external instru-ments works as follows. Let Zt be a vector of an instrument and let qt be a vector of

structural shocks, which are different to the monetary policy shocks pt. The instruments must satisfy the following assumptions,

E[Zt p0

t ] = φ, E[Zt q0

t ] = 0. (10)

The reduced-form residuals ut are obtained via OLS regression of the reduced-form VAR

model. The residuals can be split into residuals of the monetary policy indicator upt and residuals of the other variables uqt. In addition, sq is the reaction of uqt to a unit increase in the monetary policy shock pt, similarly sp is the response of upt to the monetary policy shock. To get an estimate of sq and sp, I first regress up

t on Zt. The remaining

reduced-form residuals uqt are then regressed on the fitted value of upt. The estimated coefficient is equal to s

q

sp. From the variance-covariance matrix an estimate for s

p is obtained by using

the fitted values upt and uqt, consequently an estimate of sq can then be obtained. With

(14)

policy surprises using the following equation, Yt= p X j=1 BjYt−j+ spt, (11)

where s is the column in S that corresponds to the response of the reduced-form residuals utto the monetary policy shock pt. As the aim is to estimate the dynamic effects of all the

monetary policy surprises separately, I instrument the residuals with one factor at a time. A more detailed description of the external instruments approach is found in Gertler and Karadi (2015).

3.2

Data

For data collection, I make use of the Euro Area Monetary Policy Event-Study Database (EA-MPD)11 provided by Altavilla et al. (2019), which is publicly available and regularly

updated. This database reports changes of overnight index swap rates, sovereign yields, exchanges rates and stock index prices around ECB Governing Council monetary policy decisions noted in basis points. The EA-MPD consists of three different time windows; a press release window, a press conference window and a monetary event window.

I only make use of the monetary event window, as this contains both the press release and the press conference window. The monetary event window is constructed for each variable, and is equivalent to the change between the median quote from the window 13.25 - 13.35 CET and the median quote from the window 15.40 - 15.50 CET on days that monetary policy decision are made by the ECB Governing Council. As I study the relationship between ECB monetary policy and the VSTOXX index, I extend the database with daily data of the VSTOXX (V2TX) index. Additionaly, I collect daily data of the EURO STOXX 50 to calculate the daily squared variances described above. The data is retrieved from the Thomson Reuters EIKON database.

The time period I use covers the period from 3 January 2002 to 4 June 2020. The number of ECB Governing Council meetings observed in this period is 201. The selection of the time period is based on the following points. From this point on, data of the 10-year yield of Italian bond is available. It is also marked as the beginning of the usage of the Euro as currency. Additionally, it is only the second meeting of the Governing Council after the meeting interval was changed from every two weeks to every month. This change of meeting interval has probably as a result that the impact of monetary policy on financial markets is larger, as the build up period is longer.

In table 1 shown below, descriptive statistics of the EA-MPD are presented. As is

(15)

Table 1: Descriptive statistics Observations Mean change Median change Standard deviation Minimum value Maximum value 1-month OIS 201 0.12 0.00 3.00 -20.10 14.20 3-month OIS 201 0.08 0.00 2.97 -11.20 16.15 6-month OIS 201 0.10 0.00 3.36 -14.00 16.60 1-year OIS 201 0.01 -0.15 4.14 -17.60 20.30 2-year OIS 201 -0.20 -0.25 4.57 -22.80 18.70 Germany 5-year 201 -0.18 -0.25 4.37 -19.75 15.30 Germany 10-year 201 -0.06 -0.30 3.25 -12.70 15.89 Italy 5-year 201 -0.25 -0.70 6.91 -22.15 44.00 Italy 10-year 201 -0.04 -0.50 6.33 -19.55 45.60 Spain 5-year 201 -0.51 -0.40 5.02 -21.25 19.85 Spain 10-year 201 -0.15 -0.10 5.00 -16.00 29.30

Descriptive statistics of variables used from the EA-MPD database. The values, except for the number of observations, are given in basis points.

shown in the table, only the sovereign yields and the 2-year OIS rate, do have a negative mean and median; however, these values are small. With respect to the sovereign yields, it seems that the longer the maturity, the smaller the minimum value. Such a trend cannot been seen with regard to the maximum value of the observations. Notable is that the Italian sovereign yields do have the largest maximum value, largely exceeding the maximum values of the other variables.

Table 2: Descriptive statistics of the VSTOXX index

Observations Mean value Median

value Standard deviation Minimum value Maximum value VSTOXX 4689 23.42 20.82 10.23 10.68 87.51

Some relevant statistic about the VSTOXX index are shown above in table 2. The value of the VSTOXX index represents the expected volatility of the EURO STOXX 50 index quoted in annualized percentage levels. As one can see, the mean value of the VS-TOXX is equal to 23.42, with minimum and maximum values of respectively 10.68 and 87.51. The latter value was reached on 16 October 2008, as a result of a financial market meltdown resulting from the financial crisis of 2007-2008. Looking at figure 1 below, an-other spike can been seen on 16 March 2020, when the uncertainty about the COVID-19 pandemic was at its high. The most recent spike of the VSTOXX index quickly disap-peared when the ECB Governing Council announced its pandemic emergency purchase programme (PEPP) on 18 March 2020. Another period in time when the VSTOXX index stood high, was during burst of the dot-com bubble in 2002.

(16)

Figure 1: VSTOXX index

Time-series plot of the VSTOXX index. The period spanned is from 3 January 2002 till 4 June 2020, the y-axis represents the closing day index level.

debt crisis and the corona crisis, some large spikes in changes of both OIS rates and sovereign yields can been seen. Since 2009, when the ECB started with non-standard monetary policy measures such as the public sector purchase programme, the variation of sovereign yield changes starts to increase, especially for the longer maturities. On the contrary, large changes of the OIS rate occur less often when the end of the time period is approached. This can be related to the fact that the ECB changes the standing facilities rates less.

4

Results

(17)

4.1

Factor model

In this section I present the results of the factor model of equation 1. As discussed above, I use Horn’s parallel analysis (Horn, 1965) to determine the number of factors to be extracted in the model. The scree plot of the parallel analysis conducted can be found in the appendix, figure A.1. Parallel analysis indicates that there are 3 factors that explain the variation in the data due to monetary policy surprises. Factor analysis itself

Table 3: Factor loadings

FA1 FA2 FA3

OIS 1-month - 0.102 0.848 OIS 3-month 0.321 0.131 0.932 OIS 6-month 0.554 0.150 0.794 OIS 1-year 0.726 0.180 0.624 OIS 2-year 0.824 0.224 0.472 DE 5-year 0.918 0.269 0.230 DE 10-year 0.831 0.285 -IT 5-year 0.203 0.930 0.196 IT 10-year - 0.942 0.117 ES 5-year 0.403 0.849 0.153 ES 10-year 0.234 0.939 -Proportion variance 0.311 0.331 0.269 Cumulative variance 0.311 0.642 0.911

This table representes the rotated factor loadings estimated in equation 1. Proportion variance shows the variance of the variables that is explained by each factor.

does not give a structural interpretation to the factors. To make interpretation easier, I rotate the factors using varimax, which maximizes the sum of the variance of the squared factor loadings and is an orthogonal rotation method, meaning that the factors are not correlated with each other. To give an economic interpretation of the factors, I plot the time series of the factors12 and check which variables have the highest loadings on the

estimated factors. The rotated factor loadings can be found in table 3 above.

It can be seen from table 3 that the short-term OIS rates do have the highest loadings on factor 3. From the time series plot, it becomes apparent that factor 3 is most relevant between 2009 and 2013, which is the period in which the policy rate is most adjusted by the ECB. Combining the factor loading and time series, leads to the interpretation of factor 3 as the short-term interest rate. However, it should be noted that this factor does not cover the factor short-term rate perfectly, as it has the highest loading on the 3-month OIS rate. Thus, this factor also contains market expectations of changes in the short-term rate over the next few months. This is probably caused by the relatively wide

(18)

time window around policy announcements that I use. It is found by Altavilla et al. (2019) that the factor loadings on short-term OIS rates differ depending on the time window that is used to measure monetary policy surprises13.

The 5 and 10-year German yield and longer term OIS rates do have the highest loadings on factor 1. The plot of the corresponding factor scores is noisy. However, as this factor is correlated with longer term maturities, factor 1 is interpreted as forward guidance; because this monetary policy instrument is used to steer expectations about future monetary policy and the economic outlook.

The second factor explains the largest proportion of the variance, and is most cor-related with the 5 and 10-year yield of Italy and Spain. The time-series plot of this factor indicates that this factor becomes most relevant after 2011. This coincides with the sovereign debt crisis in the eurozone and the start of the Securities Market Programme (SMP); this programme was aimed at the purchase of sovereign bonds of high debt coun-tries like Italy and Spain. This explains the high factor loadings on the long-term yields of Italy and Spain. The quantitate easing (QE) measures of the ECB formally started as of 2014, so this does not cover that factor completely. However, for simplicity I interpret this factor as QE, as the SMP is also focused on providing liquidity to debt securities markets.

To make interpretation of the effects of monetary policy surprises more intuitive, I have normalised the factor scores, which are used to estimate the impact of monetary policy in section 4.3. This leads to the short-term rate, forward guidance and QE factors having unit effect on the 3-month OIS rate, 2-year OIS rate and the 10-year Italian yield, respectively. Table A.1 in the appendix shows the effects of monetary policy surprises on the intra-day change of the variables used and shows the unit effect of the factors on the above mentioned variables.

4.2

Decomposition

As discussed in section 3.1.2 above, the VSTOXX index can be decomposed into a risk aversion and an uncertainty part. Below in figure 2 the time series of the uncertainty proxy are plotted. The values are estimated via the following equation,

RVt(22)= 6.937 (1.204)+0.612(0.045)V ST OXX 2 t−22+0.034 (0.040)RV (22) t−22+0.091 (0.041)RV (5) t−22+0.012 (0.018)RV (1) t−22+εt, (12)

with White’s heteroskedasticity consistent standard errors shown in parentheses. The pattern of the time series closely resembles the plot of the VSTOXX index shown above,

13Altavilla et al. (2019) use a press release and a press conference window. They find 1 factor in the

(19)

indicating that a large part of the VSTOXX index consists of a factor uncertainty; which is to expect, as the VSTOXX index is a measure of expected volatility. The periods with the largest uncertainty, coincide with the dot-com bubble, the financial crisis of 2008 and the recent outbreak of the coronavirus. As discussed in the data section above, these spikes can also been seen in the time series plot of the VSTOXX index. The most apparent

Figure 2: Uncertainty

Time series plot of conditional variance as a proxy for uncertainty, estimated via equation 12.

difference between the values of uncertainty and the VSTOXX index, is that the value of uncertainty is much larger. This is due to the fact that the uncertainty measure is equal to the past monthly variance. On the contrary, the VSTOXX index is a measure of volatility, which is the square root of the variance.

The difference between the squared VSTOXX index14 and the uncertainty measure is

the variance risk premium and is considered as a proxy for risk aversion15. The time series

of the values are shown in figure 3 below. As can been seen in the figure, risk aversion is negative in some periods. This can be explained by the fact that expected realized variance16 lags the VSTOXX index, as most of the weight in equation 12 is given to the

squared VSTOXX, and to realized monthly and weekly variance. Therefore the actual VSTOXX index responds much quicker to economic and monetary news. Large changes in risk aversion coincide with spikes of the VSTOXX index and uncertainty. However, the average value of risk aversion is small and moves around zero. This confirms that in most

14A plot of the time series can ben found in the appendix, figure B.3 15See section 3.1.2 for details.

(20)

periods, uncertainty is the most important component of the VSTOXX index. Another explanation for increases in risk aversion is that investors dislike volatility, leading to higher risk aversion. As the VSTOXX index is tradeable, it can be used to hedge away increases in stock market volatility.

Figure 3: Risk aversion

Time series plot the variance risk premium as a proxy of risk aversion. The values are estimated by equation 3.

Table 4: Regression result of VSTOXX decomposition

VSTOXX α 13.64∗∗∗ (0.44) Risk aversion 0.16∗∗∗ (0.01) Uncertainty 0.18∗∗∗ (0.01) R-squared 0.94 Observations 4644

This table shows the influence of levels of the risk aversion and uncertainty part on the VSTOXX index. The equations are estimated via OLS and standard errors corrected for 44 Newey West lags are presented in parentheses; ∗ ∗ ∗, ∗∗ and ∗ denote statistical significance at the 1%, 5% and 10% levels, respectively.

(21)

same influence on the VSTOXX index, with a 0.16 and 0.18 increase of the VSTOXX per unit change, respectively.

4.3

Monetary policy surprises

This section presents the reaction of a set of financial instruments to monetary policy surprises on ECB Governing Council meeting days. Some descriptive statistics of the VSTOXX index, risk aversion and uncertainty are presented in table 5. An interesting result is that the mean and median change of the VSTOXX index is negative on ECB meeting days, as is also the case for uncertainty and risk aversion. However, changes of the VSTOXX index are expressed in percentage levels, therefore these changes are much larger relatively. L´opez and Esparcia (2021) also find that the VSTOXX index declines on ECB monetary policy meeting days, with an average decline of -1.85%17. Summarizing,

the descriptive statistics give an indication that the VSTOXX index responds negatively to ECB monetary policy decisions, meaning that monetary policy decisions on average lower expected volatility.

Table 5: Descriptive statistics

VSTOXX Risk aversion Uncertainty

Mean change -0.74 1.66 -0.24

Median change -1.56 -0.30 -0.58

Minimum -17.97 -41.59 -30.52

Maximum 28.54 187.44 45.55

Observations 199 199 199

This table represents the mean and median change of the VSTOXX index, risk aversion and uncertainty on ECB meeting days. The results of the VSTOXX index are in percentage levels.

To further examine the influence of ECB monetary policy on expected volatility, I use equation 2, 5 and 6 presented above. These equations regress daily changes of the VSTOXX index, risk aversion and uncertainty on the factor scores, which represent the short-term rate, forward guidance and QE. I have also included the regression results on the Euro Stoxx 50 index daily returns, as the VSTOXX index represents the expected volatility of the Euro Stoxx 50 and is therefore closely related to stock returns. In addition, adding the Euro Stoxx 50 as a variable makes it possible to check whether the results are in line with the monetary policy literature. The results of these regressions are presented in table 6 below.

As becomes clear from the results is that QE has an important effect on the VSTOXX index and the level of risk aversion. The factor QE is correlated to, among other things,

(22)

Table 6: Regression results of monetary policy surprises

%∆VSTOXX ∆Risk aversion ∆Uncertainty %∆ESTOXX

α −0.74 (0.47) 1.67 (1.36) −0.24(0.57) −0.17 (0.12) Short-term rate −0.01 (0.29) 1.09 (1.54) (0.27)0.07 −0.04(0.09) Forward guidance −0.07 (0.13) −0.34 (0.54) −0.07 (0.12) 0.06 (0.04) QE 0.38∗∗∗ (0.13) 1.26 ∗ (0.67) −0.08 (0.10) −0.14∗∗∗ (0.04) Observations 199 199 199 199 R-squared 0.11 0.16 0.01 0.23

This table shows the results of equation 2, 5 and 6 and the effects of monetary policy on the Euro Stoxx 50 index. It represents the effects of monetary policy surprises on the respective variables on ECB meeting days. The equations are estimated via OLS and White’s heteroskedasticity consistent standard errors are presented in parentheses; ∗ ∗ ∗, ∗∗ and ∗ denote statistical significance at the 1%, 5% and 10% levels, respectively.

the yields of high sovereign debt countries. Meaning that a positive QE surprise leads to an increase of the long-term yields of Italy and Spain. Increasing yields of high debt countries is a bad sign of the state of the economy in the respective countries, and raises uncertainty about whether debt is going to be paid back. This explains why a positive QE surprise increases expected volatility and risk aversion among investors. Likewise, it lowers the return of the Euro Stoxx 50, as investors respond negatively to rising sovereign yields of high debt countries. In addition, this confirms that increasing volatility goes often together with negative returns.

None of the regressors has an statistically significant impact on changes in uncertainty. Which is explained by the fact that expected uncertainty is only influenced by past vari-ance18, meaning that it is to expected that the variable uncertainty does not respond quickly to new information. Similarly, it is found that the factor forward guidance does not have a significant effect on the variables. This is also a finding of Kerssenfischer (2019), he finds that when forward guidance is defined as changes in the 2-year German yield, the VSTOXX and Euro Stoxx 50 index are not affected by this monetary policy instrument19.

To check whether the influence of monetary policy surprises is different over time, I have made two subsamples. The first sample represents the period before quantitative easing was officially introduced, and the second sample the period thereafter. The pre-QE

18See equation 12.

19Kerssenfischer (2019) finds that changes of the 2-year German bond yield does not have an effect

(23)

sample spans from January 2002 to December 2013, whereas the QE sample covers the time period from January 2014 to June 2020. The results of the pre-QE and QE samples can be found in table 7 and 8, respectively. The factor loadings and the factor scores are defined as before, meaning that the interpretation of the factors has not changed. What has changed, is the exclusion of the QE factor in the pre-QE sample, as it is assumed that quantitative easing measures were not taken in this period.

Table 7: Regression results of monetary policy surprises in the pre-QE sample

%∆VSTOXX ∆Risk aversion ∆Uncertainty %∆ESTOXX

α 0.17 (0.58) (1.68)2.28 −0.72 (0.65) −0.23 (0.15) Short-term rate −0.19 (0.34) −0.07 (1.67) 0.02 (0.31) (0.11)0.06 Forward guidance 0.03 (0.13) (0.40)0.11 (0.13)0.01 (0.04)0.05 Observations 143 143 143 143 R-squared 0.01 0.00 0.01 0.02

This table shows the effect of the short-term rate and forward guidance surprises in the pre-QE period. This is the period from 01/2002 - 12/2013. It represents the effects of monetary policy surprises on the respective variables on ECB meeting days. The equations are estimated via OLS and White’s het-eroskedasticity consistent standard errors are presented in parentheses; ∗ ∗ ∗, ∗∗ and ∗ denote statistical significance at the 1%, 5% and 10% levels, respectively.

The regression results show that none of the monetary policy surprises have a signif-icant impact on the variables in the pre-QE sample. The R-squared values are low for all variables, indicating that none of the monetary policy surprises explain the variance of the variables. In addition, the coefficient signs have changed for almost all variables, indicating that monetary policy surprises should have an effect on the variables in the QE sample, as monetary policy surprises do explain part of the variance in the full sample.

The results from the QE sample greatly differ from the pre-QE sample. The R-squared is much higher, meaning that monetary policy surprises do explain the variability of expected volatility, risk aversion and stock returns. As was already found in the regression results of the full sample, the variable uncertainty is not influenced by monetary policy surprises. However, for the other variables it turns out that unexpected ECB policy decisions do explain a large part of the changes in the VSTOXX index, risk aversion and the Euro Stoxx 50. For example, the effect of QE surprises is larger in the QE period compared to the full sample, which is to expected, as it is assumed that the factor QE is only relevant in this period.

(24)

It is found that a negative policy rate surprise20 decreases expected volatility and risk aversion. This can be explained by the theory that a loose monetary policy leads to lower borrowing costs and less savings, leading to more investment and consumption, which in turn increases economic growth. As such, lowering the policy rate lowers risk aversion and consequently the VSTOXX index. Additionally, a decrease of the short-term rate has a positive effect on stock prices, as it increases discounted dividends by decreasing the discount rate and higher expected dividends due to larger projected economic growth.

Table 8: Regression results of monetary policy surprises in the QE sample

%∆VSTOXX ∆Risk aversion ∆Uncertainty %∆ESTOXX

α −3.06∗∗∗ (0.67) −1.37 (2.11) 1.12 (0.98) (0.15)0.15 Short-term rate 0.70∗ (0.36) 5.60 ∗∗ (2.41) −0.12 (0.49) −0.49∗∗∗ (0.13) Forward guidance −0.32 (0.27) −1.37 (1.31) −0.62 (0.40) −0.03 (0.07) QE 0.52∗∗∗ (0.11) 1.65 ∗∗ (0.62) (0.17)0.00 −0.16∗∗∗ (0.03) Observations 56 56 56 56 R-squared 0.47 0.60 0.03 0.69

This table shows the effect of the short-term rate, forward guidance and quantitative easing surprises in the QE period. This is the period from 01/2014 - 6/2020. It represents the effects of monetary policy surprises on the respective variables on ECB meeting days. The equations are estimated via OLS and White’s heteroskedasticity consistent standard errors are presented in parentheses; ∗ ∗ ∗, ∗∗ and ∗ denote statistical significance at the 1%, 5% and 10% levels, respectively.

From the results it becomes clear that changes in risk aversion are better explained by monetary policy surprises than changes uncertainty. In addition, the signs of the risk aversion coefficients have the same magnitude compared to the VSTOXX index. This indicates that changes in expected volatility, are due to changes in risk aversion and not uncertainty in the first place. This finding is in line with Bekaert et al. (2013), they also find that risk aversion is more affected by the monetary policy surprises than uncertainty is. As already explained above, the definition of uncertainty probably leads to the fact that it is not affected by monetary policy surprises on announcement days.

As is also found in the other samples, surprises in forward guidance do not have a significant effect on the VSTOXX index, risk aversion, uncertainty and the stock market. This could be explained by the rationale that investors do have the same economic outlook as the ECB Governing Council. Meaning that large surprises in the economic outlook and

20For the interpretation of the short-term rate surprises, I use a negative surprise as an example, as

(25)

future monetary policy communicated by the ECB do not occur. Another explanation is the different reaction of financial markets to changes in the economic outlook (Delphic shock) and changes in future monetary policy (Odyssean shock) contained in central bank communication about monetary policy. As is shown by Kerssenfischer (2019) and Jaroci´nski and Karadi (2020) rising yields may not be interpreted equally. According to standard economic theory, interest rate raises would imply contractionary monetary policy, leading to a lowering of expected growth and inflation and therefore lower stock prices. However, it is shown that rising yields co-move with higher stock prices if central bank announcements signal an improved economic outlook. The intuition behind the definition of both shocks is that they both rise the yield curve, but have a different effect on the stock market. An improved economic outlook by the central bank implies higher economic growth and normally higher inflation, leading to a potential increase in interest rates as central banks try to keep inflation low. These effects may be offsetting each other, leading to insignificance and different signs of the forward guidance coefficients across the samples. In addition, the finding that forward guidance, defined as changes in longer-term OIS rates, does not affect financial markets, is in line with findings of the literature (Kerssenfischer, 2019; Andrade and Ferroni, 2020).

To check whether Delphic and Odyssean shocks have different impacts on financial markets, I replace the forward guidance factor by measures of Delphic and Odyssean shocks. To identify the shocks, I use the changes of the 2-year OIS rate and the Euro Stoxx 50 index in the monetary event window of the EA-MPD, which is the same window used in the factor model. A Delphic shock is measured as an increase in the 2-year OIS rate and a positive reaction of the stock market, whereas an Odyssean shock is identified as an increase in the 2-year OIS rate and a negative reaction of the stock market. The shocks are normalised, such that a unit increase in a shock is equivalent to a unit increase in the 2-year OIS rate, which is the same normalisation as in the previous results. As in the above discussed regressions, I first estimate the effects on the full period and thereafter estimate the impact in the pre-QE and QE period. The results are presented in table 9 below.

The replacement of the forward guidance factor by Delphic and Odyssean shocks does not lead to large changes in the results. In fact, almost all estimated effects are not signif-icant. Additionally, the combined effect of Delphic and Odyssean shocks is approximately of the same magnitude as the effect of surprises in forward guidance estimated in the other regressions. However, it turns out that an Odyssean shock does decrease the VSTOXX index significantly in the QE sample.

(26)

Table 9: Effect of Delphic and Odyssean shocks as replacement of the forward guidance factor.

%∆VSTOXX ∆Risk aversion ∆Uncertainty %∆ESTOXX

α −0.89 (0.52) 1.14 (1.66) (0.71)0.20 −0.24∗ (0.14) Short-term rate −0.04 (0.30) 0.85 (1.64) (0.28)0.27 −0.06 (0.10) Delphic 0.15 (0.30) (0.60)0.04 −0.03 (0.19) 0.05 (0.06) Odyssean 0.06 (0.19) (0.91)0.47 −0.40∗ (0.22) 0.03 (0.06) QE 0.38∗∗∗ (0.14) (0.73)1.20 −0.03 (0.10) −0.15∗∗∗ (0.04) Observations 199 199 199 199 R-squared 0.11 0.16 0.02 0.22 α −0.21 (0.60) 0.77 (1.46) −0.33 (0.79) −0.27 (0.17) Short-term rate −0.28 (0.33) −0.67 (1.51) 0.21 (0.32) (0.11)0.07 Delphic 0.27 (0.44) −0.07 (0.65) 0.17 (0.25) (0.06)0.10 Odyssean 0.18 (0.19) (0.85)1.13 −0.34 (0.23) −0.01 (0.06) Observations 143 143 143 143 R-squared 0.02 0.03 0.02 0.02 α −2.62∗∗∗ (0.76) −0.71 (2.21) 1.52 (1.17) (0.16)0.19 Short-term rate 1.05∗∗ (0.41) 6.63 ∗∗ (3.18) (0.72)0.44 −0.46 ∗∗∗ (0.15) Delphic −0.01 (0.34) −0.12 (0.90) −0.38 (0.43) 0.00 (0.10) Odyssean −0.89∗∗ (0.35) −1.75 (2.68) −0.75 (0.64) −0.08 (0.13) QE 0.65∗∗∗ (0.14) 1.92 ∗∗ (0.78) (0.25)0.08 −0.15∗∗∗ (0.05) Observations 56 56 56 56 R-squared 0.49 0.59 0.02 0.69

This table shows the effect of the short-term rate, Delphic shocks, Odyssean shocks and quantitative easing surprises in the full sample, pre-QE sample and QE sample. It represents the effects of monetary policy surprises on the respective variables on ECB meeting days. The equations are estimated via OLS and White’s heteroskedasticity consistent standard errors are presented in parentheses; ∗ ∗ ∗, ∗∗ and ∗ denote statistical significance at the 1%, 5% and 10% levels, respectively.

(27)

of uncertainty that is used.

4.4

VAR

To analyse the dynamic impact of monetary policy surprises I estimate the daily VAR model described in section 3.1.3, and identify the monetary policy shocks via external instruments. The endogenous variables in the model are the log of the VSTOXX index, the log of risk aversion, the log of uncertainty and the 2-year OIS rate as a monetary policy indicator. The number of lags used in the model is 38 and is based on the Aikake information criterion (AIC). The impulse responses of the variables to monetary policy surprises are presented in figure 4 below.

Figure 4: VAR

(a) Short-term interest rate (b) Forward guidance

(c) Quantitative easing

Impulse responses of the variables to monetary policy shocks. The VSTOXX index is represented by the blue line, risk aversion by the red line, uncertainty by the orange line and the 2-year OIS rate by the green line. The horizon represents the number of days, the response is in percent.

(28)

is, however, positive. The earlier finding that changes in the VSTOXX index can be attributed to changes in risk aversion, is confirmed by the VAR model, especially in the forward guidance and QE case. To check whether the results are different in the pre-QE and QE period, I split the sample, and estimate the same VAR model. The results are presented below in figure 5.

Figure 5: VAR pre-QE and QE sample

(a) Short-term interest rate (pre-QE) (b) Forward guidance (pre-QE)

(c) Short-term interest rate (QE) (d) Forward guidance (QE)

(e) QE (QE)

Impulse responses of the variables to monetary policy shocks in the pre-QE and QE sample. The VSTOXX index is represented by the blue line, risk aversion by the red line, uncertainty by the orange line and the 2-year OIS rate by the green line. The horizon represents the number of days, the response is in percent.

(29)

more in the QE period.

Overall, monetary policy surprises do affect the VSTOXX index, risk aversion and uncertainty, but the effect is not persistent over time. In addition, it can be concluded that risk aversion is the main driver of changes in the VSTOXX index at the moment of a monetary policy surprise. However, uncertainty reacts slower to policy shocks, but moves eventually in the same direction as the VSTOXX index. Implying that the finding21that uncertainty is the largest part of the VSTOXX index is not rejected.

5

Conclusions

In this thesis I have estimated the effect of monetary policy surprises on the VSTOXX index, which reflects the 30-day expected volatility of the Euro Stoxx 50 index by in-vestors. First, I have estimated a factor model, to extract monetary policy surprises from the reaction of financial markets around monetary policy announcements of the ECB. The factors that are found represent the short-term interest rate, forward guidance and quantitative easing. To analyse the impact on the VSTOXX, I decompose the index into a risk aversion and an uncertainty part. This is done by estimating the next month vari-ance of the Euro Stoxx 50 index. The estimated conditional varivari-ance is used as a proxy for uncertainty. The difference between the variance of the VSTOXX and uncertainty is called that variance risk premium and used as a proxy for risk aversion.

To estimate the effect of monetary policy surprises I regress the factors on daily changes of VSTOXX index, risk aversion, uncertainty and the Euro Stoxx 50 index on monetary policy announcement days. This is done for the period of January 2002 to June 2020, and for samples of the pre-QE and QE period to check whether the results are different. I find that the QE factor affects the VSTOXX index, risk aversion and the Euro Stoxx 50, the results are more clear when I divide the sample. It turns out that the monetary policy factors do not affect the initial reaction of financial markets in the pre-QE period. However, monetary policy surprises do explain changes in the variables in the QE period. Only the factor forward guidance does not affect financial markets significantly in this period.

It turns out that uncertainty is not affected by monetary policy surprises initially, but reacts to policy surprises over time. This could be explained by the definition of uncertainty, leading to the fact that is does not respond quickly to changes in monetary policy. However, I show that uncertainty largely determines the pattern of the VSTOXX index. On the contrary, it is found that risk aversion reacts quickly to monetary policy surprises. This leads to the conclusion that changes of the VSTOXX index at monetary

(30)

policy announcement days are due to changes in risk aversion and not uncertainty, which is in line with the literature Bekaert et al. (2013). This finding is of interest, as it is found in the literature that risk aversion is a good predictor of stock market returns (Bollerslev et al., 2009; Bekaert and Hoerova, 2014).

Overall, I conclude that monetary policy surprises of the ECB do affect the VSTOXX index, with quantitative easing measures have the largest affect. In addition, monetary policy surprises have a greater effect on financial markets in recent years. As risk aversion is much more affected by changes in the monetary policy stance than uncertainty, I argue that changes in the VSTOXX index are related to changes in the risk aversion of investors. The results presented above do, however, have some limitations. I have used closing day Euro Stoxx 50 index data, to compute the daily variance in equation 4. However, using intraday data which is more common in the literature, improves the accuracy of the analysis. This leads to more precise measures of uncertainty and risk aversion. In addition, changing the definition of uncertainty may also lead to other results. Next, I have assumed that ECB monetary policy decisions only do have an impact during the announcement of Governing Council decisions and the press conference, and that there is no other relevant monetary policy news from the ECB. However, there are events in which the ECB does reveal or explain its monetary policy, the ”whatever it takes” speech from Mario Draghi is a good example of this.

(31)

References

Altavilla, C., Brugnolini, L., G¨urkaynak, R.S., Motto, R., 2019. Measuring euro area monetary policy. Journal of Monetary Economics 108, 162–179.

Andrade, P., Ferroni, F., 2020. Delphic and Odyssean monetary policy shocks: Evidence from the euro area. Journal of Monetary Economics .

Bekaert, G., Hoerova, M., 2014. The VIX, the variance premium and stock market volatil-ity. Journal of Econometrics 183, 181–192.

Bekaert, G., Hoerova, M., Duca, M.L., 2013. Risk, uncertainty and monetary policy. Journal of Monetary Economics 60, 771–788.

Bernanke, B.S., Kuttner, K.N., 2005. What Explains the Stock Market’s Reaction to Federal Reserve Policy? Journal of Finance 60, 1221–1257.

Bloom, N., 2009. The impact of uncertainty shocks. Econometrica 77, 623–685.

Bloom, N., Floetotto, M., Jaimovich, N., Saporta-Eksten, I., Terry, S.J., 2018. Really uncertain business cycles. Econometrica 86, 1031–1065.

Bollerslev, T., Gibson, M., Zhou, H., 2011. Dynamic estimation of volatility risk pre-mia and investor risk aversion from option-implied and realized volatilities. Journal of Econometrics 160, 235–245.

Bollerslev, T., Tauchen, G., Zhou, H., 2009. Expected stock returns and variance risk premia. Review of Financial Studies 22, 4463–4492.

Busch, T., Christensen, B.J., Nielsen, M.Ø., 2011. The role of implied volatility in forecast-ing future realized volatility and jumps in foreign exchange, stock, and bond markets. Journal of Econometrics 160, 48–57.

Carr, P., Wu, L., 2009. Variance risk premiums. Review of Financial Studies 22, 1311– 1341.

Chen, E.T., Clements, A., 2007. S&P 500 implied volatility and monetary policy an-nouncements. Finance Research Letters 4, 227–232.

Chow, K.V., Jiang, W., Li, B., Li, J., 2020. Decomposing the VIX: Implications for the predictability of stock returns. Financial Review 55, 645–668.

(32)

Cieslak, A., Morse, A., Vissing-Jorgensen, A., 2019. Stock returns over the FOMC cycle. Journal of Finance 74, 2201–2248.

Corrado, C.J., Miller, T.W., 2005. The forecast quality of CBOE implied volatility in-dexes. Journal of Futures Markets 25, 339–373.

Drechsler, I., Yaron, A., 2011. What’s vol got to do with it. Review of Financial Studies 24, 1–45.

Fernandez-Perez, A., Frijns, B., Tourani-Rad, A., 2017. When no news is good news–The decrease in investor fear after the FOMC announcement. Journal of Empirical Finance 41, 187–199.

F¨uss, R., Mager, F., Wohlenberg, H., Zhao, L., 2011. The impact of macroeconomic announcements on implied volatility. Applied Financial Economics 21, 1571–1580. Gertler, M., Karadi, P., 2015. Monetary policy surprises, credit costs, and economic

activity. American Economic Journal: Macroeconomics 7, 44–76.

Gospodinov, N., Jamali, I., 2015. The response of stock market volatility to futures-based measures of monetary policy shocks. International Review of Economics & Finance 37, 42–54.

G¨urkaynak, R.F., Sack, B., Swanson, E.T., 2005. Do Actions Speak Louder Than Words? The Response of Asset Prices to Monetary Policy Actions and Statements. International Journal of Central Banking 1, 55–93.

Haitsma, R., Unalmis, D., de Haan, J., 2016. The impact of the ECB’s conventional and unconventional monetary policies on stock markets. Journal of Macroeconomics 48, 101–116.

Horn, J.L., 1965. A rationale and test for the number of factors in factor analysis. Psy-chometrika 30, 179–185.

Jaroci´nski, M., Karadi, P., 2020. Deconstructing monetary policy surprises—the role of information shocks. American Economic Journal: Macroeconomics 12, 1–43.

Jiang, G.J., Konstantinidi, E., Skiadopoulos, G., 2012. Volatility spillovers and the effect of news announcements. Journal of Banking & Finance 36, 2260–2273.

(33)

Kaminska, I., Roberts-Sklar, M., 2018. Volatility in equity markets and monetary policy rate uncertainty. Journal of Empirical Finance 45, 68–83.

Kerssenfischer, M., 2019. Information effects of euro area monetary policy: New evidence from high-frequency futures data. Discussion Paper Deutsche Bundesbank (7).

Krieger, K., Mauck, N., Chen, D., 2012. VIX changes and derivative returns on FOMC meeting days. Financial Markets and Portfolio Management 26, 315–331.

Krieger, K., Mauck, N., Vazquez, J., 2015. Comparing US and European market volatil-ity responses to interest rate policy announcements. International Review of Financial Analysis 39, 127–136.

Kurov, A., 2010. Investor sentiment and the stock market’s reaction to monetary policy. Journal of Banking & Finance 34, 139–149.

L´opez, R., Esparcia, C., 2021. Analysis of the performance of volatility-based trading strategies on scheduled news announcement days: An international equity market per-spective. International Review of Economics & Finance 71, 32–54.

Lucca, D.O., Moench, E., 2015. The pre-FOMC announcement drift. Journal of Finance 70, 329–371.

Mertens, K., Ravn, M.O., 2013. The dynamic effects of personal and corporate income tax changes in the United States. American Economic Review 103, 1212–47.

Nakamura, E., Steinsson, J., 2018. High-frequency identification of monetary non-neutrality: the information effect. The Quarterly Journal of Economics 133(3), 1283– 1330.

Nikkinen, J., Sahlstr¨om, P., 2004. Impact of the federal open market committee’s meetings and scheduled macroeconomic news on stock market uncertainty. International Review of Financial Analysis 13, 1–12.

Ozdagli, A., Velikov, M., 2020. Show me the money: the monetary policy risk premium. Journal of Financial Economics 135, 320–339.

Stock, J.H., Watson, M.W., 2012. Disentangling the Channels of the 2007-2009 Recession. Technical report, National Bureau of Economic Research.

(34)

Thorbecke, W., 1997. On stock market returns and monetary policy. Journal of Finance 52, 635–654.

V¨ah¨amaa, S., ¨Aij¨o, J., 2011. The Fed’s policy decisions and implied volatility. Journal of Futures Markets 31, 995–1009.

(35)

A

Factor analysis

Figure A.1: Scree plot of Horn’s parallel analysis.

Figure A.2: Time-series plot of the factor scores estimated in equation 1.

(a) Factor 1 (b) Factor 2

(36)

Table A.1: Intraday reaction of the variables used to monetary policy surprises

α Short-term rate Forward guidance QE Observations R-squared

OIS 1M 0.12 (0.11) 0.92 ∗∗∗ (0.10) (0.06)0.00 0.05 ∗∗∗ (0.02) 201 0.74 OIS 3M 0.08∗∗∗ (0.01) 1.00 ∗∗∗ (0.01) 0.25 ∗∗∗ (0.00) 0.06 ∗∗∗ (0.00) 201 0.99 OIS 6M 0.10∗∗ (0.04) 0.97 ∗∗∗ (0.03) 0.49 ∗∗∗ (0.02) 0.08 ∗∗∗ (0.01) 201 0.97 OIS 1Y 0.01 (0.06) 0.93 ∗∗∗ (0.04) 0.79 ∗∗∗ (0.03) 0.12 ∗∗∗ (0.02) 201 0.95 OIS 2Y −0.20∗∗∗ (0.06) 0.76∗∗∗ (0.03) 1.00 ∗∗∗ (0.03) 0.17 ∗∗∗ (0.02) 201 0.97 DE 5Y −0.18∗∗∗ (0.05) 0.37∗∗∗ (0.03) 1.04 ∗∗∗ (0.02) 0.19 ∗∗∗ (0.01) 201 0.97 DE 10Y −0.06 (0.12) 0.05 (0.04) 0.68 ∗∗∗ (0.06) 0.15 ∗∗∗ (0.04) 201 0.74 IT 5Y −0.25∗∗ (0.11) 0.44∗∗∗ (0.06) 0.41 ∗∗∗ (0.05) 1.07 ∗∗∗ (0.05) 201 0.94 IT 10Y −0.04 (0.12) 0.28∗∗∗ (0.08) 0.15 ∗∗∗ (0.05) 1.00 ∗∗∗ (0.07) 201 0.92 ES 5Y −0.51∗∗∗ (0.11) 0.27∗∗∗ (0.06) 0.54 ∗∗∗ (0.04) 0.71 ∗∗∗ (0.04) 201 0.91 ES 10Y −0.15∗ (0.09) 0.11∗ (0.06) 0.28 ∗∗∗ (0.04) 0.78 ∗∗∗ (0.05) 201 0.93

(37)

B

Additional plots

Figure B.1: Time-series plot of all OIS rates used. It represents the change in basis points during the monetary event window.

(a) 1-month OIS (b) 3-month OIS

(c) 6-month OIS (d) 1-year OIS

(38)

Figure B.2: Time-series plot of all sovereign yields used. It represents the change in basis points during the monetary event window.

(a) 5-year yield Germany (b) 10-year yield Germany

(c) 5-year yield Italy (d) 10-year yield Italy

(39)

Referenties

GERELATEERDE DOCUMENTEN

The non-normal incidence of thin-film guided, in-plane unguided optical waves on straight, possibly composite slab waveguide facets is considered.. The quasi-analytical,

In February 2018, building on the already existing rules on on-road testing of CAVs, the Adopted Regulatory Text for Driverless Testing Regulations (Title 12, Division 1, Chapter

Hypothesis 5: Spouse ’s partner role salience moderates the negative relationship between spouse ’s career role salience and employee ’s willingness to accept an

For instance, limits on the speed and nature of feedback during learning, as well as limits on the validity of the cues experts derive from their environments, prohibit

Findings indicated that students were able to collectively advance the community’s discourse as they built on each others’ ideas, generated theories, questions and

Hence, I explain these insignificant results with other plausible reasons; The SRISK measure is not suitable to capture UMP shocks; There exist a long run causality between UMP

The variables are: TRR is the Taylor rule residual, FDCAR is the first difference capital adequacy ratio – tier 1, ircr is the interaction variable between Taylor rule residual

We end the section with a short discussion of our hypotheses. First of all, we believe that pre-crisis the conventional policy surprise will yield less significant results