Monetary policy decisions by the ECB and stock returns Stock returns over weeks between monetary policy decisions

(1)

Monetary policy decisions by the ECB and stock returns Stock returns over weeks between monetary policy decisions

M.Sc. Thesis in Finance and Investments Copenhagen Business School 2021

Author:

Einar Sævar Jónmundsson

Supervisor:

Christian Rix-Nielsen

(2)

Abstract

This thesis investigates the effect on the returns of the European stock market in the weeks following a monetary policy decision by the European Central Bank over the period 1999- 2019. Prior research by Cieslak et al. (2019) had shown that the equity premium in the U.S.

stock market were earned entirely in the even weeks following a Federal Open Market Committee meeting.

Two different methodologies were applied for investigating the effect; an event study based on the methodology by MacKinlay (1997) as well as the methodology that was used by Cieslak et al. (2019). The event study was performed by measuring cumulative abnormal returns and measure if they differ significantly. A parametric test and a non-parametric test were applied for measuring that effect; the standard t-test and the generalized rank test. In a similar manner to Cieslak et al. (2019), the average five-day excess return of the European stock market was studied in a way to observe a weekly pattern in returns. Furthermore, in accordance with their research the relationship between excess return and weekly periods was studied through an ordinary linear regression.

The results of this study suggest that returns of the European stock market do not follow a biweekly pattern over the weeks between monetary policy decision over the period 1999- 2019. The cumulative abnormal returns gathered in the event study did not differ significantly from zero for any of the weekly period. The results using the methodology by Cieslak et al.

(2019) did not show a sign of pattern either. The five-day average excess return showed no indication of a biweekly pattern and the beta coefficients gathered through the regression did not differ significantly from zero, suggesting no relationship between ECB monetary policy decision weekly cycles and excess return on European stocks.

(3)

List of tables

Table 1: List of events in event study. ... 12

Table 2: Day count of the week variables ... 13

Table 3: Results of event studies of returns of the FFE Index over even weeks ... 23

Table 4: Results of event studies of returns of the FFE Index over odd weeks ... 24

Table 5: Regression of daily excess returns on the Fama/French Europe Index over the 1- month U.S. T-bill rate on even weeks ... 27

Table 6: Regression of daily excess returns on the Fama/French European Index over the 1- month U.S. T-bill rate on odd weeks ... 28

Table 7: Comparison of results of event study over the period of 1999-2014 and 2015-2019 performed for even weeks ... 29

Table 8: Comparison of results of event study over the period of 1999-2014 and 2015-2019 performed for odd weeks ... 30

Table 9: Comparison of regression of daily excess returns on the Fama/French Europe Index over the T-bill rate on even weeks for the period of 1999-2014 and 2015-2019 ... 34

Table 10: Comparison of regression of daily excess returns on the Fama/French Europe Index over the T-bill rate on odd weeks for the period of 1999-2014 and 2015-2019... 36

Table 11: Regression of daily excess returns on the CRSP Index over the 1-month U.S. T-bill rate on even weeks ... 39

Table 12: Regression of daily excess returns on the CRSP Index over the 1-month U.S. T-bill rate on odd weeks ... 40

Table 13: The excess return on the Fama/French Europe index during the period of 24^th of February 1999 to 31^st of December 2019 using different trading strategies. ... 42

Table 14: Comparison of the excess return on the Fama/French Europe index using different investment strategies between the periods of 1999-2014 and 2015-2019. ... 43

(6)

List of figures

Figure 1 – The timeline of an event study ... 14 Figure 2: The average five-day excess returns of the FFE index over the T-bill rate ... 26 Figure 3: The average five-day excess returns of the FFE index over the T-bill rate during the period of 1999-2014 ... 31 Figure 4: The average five-day excess returns of the FFE index over the T-bill rate during the period of 2015-2019 ... 33 Figure 5: The average five-day excess returns of the CRSP index over the T-bill rate during the period of 1999-2019 ... 38

(7)

1

1 Introduction

The origin of money is much debated. While some historians date it back to ancient China it is commonly believed in the Western world that the first use of coins as a medium of exchange was in ancient Lydia in the 8^th century BC. Before money came along products were simply traded one for another. With time, however, traders began holding stock of a different product than the one being traded which was more easily stored and divisible, who they used for payment. As precious metals were both easily stored as well as divisible those became the medium of exchange. Later on, governments began realizing it was easier to pay soldiers in a general purchasing power and from there the use of coins as currency started (Bordo, 2010).

It is traditionally said that money has three functions: medium of exchange, store of value, and a unit of account. It is a medium of exchange meaning that it is accepted in exchange for goods that are sold and bought. Money is a store of value as it can be used to trade current goods for goods in the future. Money is also a unit of account as all values are denominated in the unit of money. The amount of money in circulation is controlled most often by a nation’s central bank (Williamson, 2018).

There has been a lot of changes in which the central bank system works since the first central bank was established in Sweden in 1668, where its main function was borrowing and lending to the government of Sweden. As international payments rose in the seventeenth century so too did the need for an international payment system and with the founding of the Bank of England in 1694, central banking officially began. The central banks, in addition to buying government debt, handled cashless international payments for the ease of international trade. With time central banks evolved to ensuring monetary stability and over the last two centuries they have expanded into regulating the value of the national currency, financing the government, and acting as a “lender of last resort”. The way in which a central bank manages the money supply and interest rates to maintain price stability, employment, and economic growth is called monetary policy (Mishkin, 2019).

Monetary policy has seen a lot of changes throughout the years. The earliest use of monetary policy is believed to be debasement, which is when a government would call in coins to melt them and mix with cheaper metals in order to alter the quality and or weight of the coins. The

(8)

2 conduct of monetary policy has evolved considerably where today a central bank uses various tools and measures to maintain the money supply and interest rates (Bordo, 2010).

The effect of monetary policy on stock returns has been thoroughly studied on the days of monetary policy changes. There has however been less focus on equity returns over the full cycle of days between monetary policy meetings.

In a research by Anna Cieslak, Adair Morse, and Annette Vissing-Jorgensen, which was published in The Journal of Finance in 2019, they studied the excess returns of the CRSP value weighted index over 3-month Treasury bill rates over the full day cycle between Federal Open Market Committee (FOMC) meetings over the period of 1994-2016. They studied the excess returns of the CRSP index in weekly patterns from the week before a FOMC meeting up to and including the sixth week after a meeting. Their results showed that the average excess returns followed a biweekly pattern where the equity premium of the CRSP value weighted index was exclusively earned during week 0, 2, 4, and 6, which they referred to as even weeks.

This thesis will investigate the returns of the European stock market in weekly patterns following a monetary policy decision (MPD) by the European Central Bank (ECB). The index used as a substitute for the European stock market is the Fama/French European index (FFE) taken from the website of Kenneth R. French. Two studies will be performed using two different methodologies.

First, the return of the European stock market index will be investigated over a MPD cycle through an event study approach based on the methodology by MacKinlay (1997) to see if the returns of the FFE differ from the returns predicted using the market model, which is an estimation method for returns. The difference between real returns and the predicted returns, referred to as abnormal returns, will be tested using a parametric t-test and a non- parametric generalized ranking test to measure if the real returns differ significantly from the returns predicted by the market model.

Second, the excess return of the FFE index will be investigated using the methodology Cieslak et al. (2019) used in their paper. In their study they graphed the average five-day excess return of the CRSP index in the weeks following a FOMC meeting and analyzed the graph for a biweekly pattern. They additionally calculated the relationship between the excess return of the index and weekly periods through an ordinary linear regression (OLS) using a dummy

(9)

3 variable per each week. Both methods will be applied in this thesis for studying the returns of the European stock market over the weeks between monetary policy decisions by the ECB.

1.1 Research question

The research question this thesis aims to answer is the following: “Do returns of the FFE index follow a biweekly pattern in the weeks after a monetary policy decision, in a similar manner as Cieslak et al. have shown the CRSP index does over a FOMC meeting cycle”. As Cieslak et al. showed evidence of a positive relationship between the CRSP index and even weeks, additional focus will be set on even weeks. The question will be answered by studying the returns during even weeks to see if they differ from their predicted returns. Four hypotheses are put forth to see if returns differ during even weeks. Cumulative average abnormal returns (𝐶𝐴𝑅̅̅̅̅̅̅) will be measured for week 0, 2, 4, and 6. The predicted returns of the FFE index will be estimated and subtracted by the 𝐶𝐴𝑅̅̅̅̅̅̅ to see if the abnormal returns differ from zero as:

𝐻₀: 𝐶𝐴𝑅̅̅̅̅̅̅(𝑤𝑒𝑒𝑘0) = 0 𝐻₁: 𝐶𝐴𝑅̅̅̅̅̅̅(𝑤𝑒𝑒𝑘2) = 0 𝐻₂: 𝐶𝐴𝑅̅̅̅̅̅̅(𝑤𝑒𝑒𝑘4) = 0 𝐻₃: 𝐶𝐴𝑅̅̅̅̅̅̅(𝑤𝑒𝑒𝑘6) = 0

Furthermore, the excess return of the FFE index will be analyzed for weekly patterns by looking at the average five-day excess return and the relationship between excess returns and weekly cycles.

(10)

4

2 Literature Overview

2.1 Monetary Policy

The act of monetary policy is in the hands of a nation’s central bank and is used to achieve six basic goals: price stability, high employment, economic growth, stability of financial markets, interest rate stability and stability in foreign markets. The way in which a central bank affects the economy is through the money supply and interest rates (Mishkin, 2019).

2.1.1 Conventional monetary policy tools

The tools which a central bank uses during normal times go by the name of conventional monetary policy tools. There are three types of conventional tools: open market operations, discount lending, and reserve requirements which a central bank uses to achieve their goals.

Open market operations revolve around buying and selling government bonds to control the money supply and short-term interest rates. They fall into two categories: Dynamic open market operations and defensive open market operations. Dynamic open market operations are meant to change the level of reserves and the monetary base whereas defensive open market operations are used to offset changes in other factors that affect reserves and the monetary base (Mishkin, 2019).

Discount lending refers to the rate which the central bank sets on loans to other banks. Most central banks base their rates on a similar method. The Fed sets three rates which are the primary rate, secondary rate, and the seasonal rate. The primary rate is usually set higher than the prevalent key interest rate. The key interest rate is the target rate at which banks borrow from each other. The primary rate is set higher than the key interest rate so that banks borrow from each other instead of borrowing from the central bank which motivates banks to continuously monitor each other for credit risk. Secondary rate is an interest rate set on loans to banks in financial trouble and is usually set 50 basis points above the primary rate.

Seasonal credit is a rate given to banks that are in areas with seasonal fluctuations in their business. The need for such a rate has though been debated by the Fed (Mishkin, 2019).

Finally, the central bank sets reserve requirements for banks to control the money supply in their nation. An increase in reserve requirements lowers the amount of deposits that can be lent out by banks which lowers the money supply. This leads to an increase in the demand for

(11)

5 reserves resulting in a higher key interest rate. The opposite happens when a central bank reduces the reserve requirement resulting in a fall in the key interest rate (Mishkin, 2019).

2.1.2 Unconventional monetary policy tools

During a distress in the economy a central bank uses unconventional tools to influence the economy. There are four tools a central bank uses: liquidity provision, asset purchases, forward guidance and negative interest rates on bank deposits at the central bank. Liquidity provision and asset purchases lead to an increase in the balance sheet of the central bank, which is most often referred to as quantitative easing. Forward guidance refers to when a central bank makes a verbal assurance about its future expectations and policies as a way to calm the markets. Negative interest rates on deposits in the central bank is a way to stimulate the economy by encouraging banks to lend out deposits they were keeping at the central bank (Mishkin, 2019).

2.2 Monetary Policy and Equity Prices

The main goals of monetary policy are price stability, maximum employment, and economic growth by which central banks use the beforementioned instruments to achieve. However, the effect of those instruments used to achieve those goals is indirect, as the most direct and immediate effect is on the financial markets and amidst them the equity market. Monetary policy affects the financial markets through changes in the value of private portfolios, changes in the cost of capital as well as other means (Bernanke & Kuttner, 2005).

According to the Gordon growth model (GGM) the value of a stock today is 𝑃₀ which is determined by the value of all future cash flows paid through dividends 𝐷₁

𝑃₀ = 𝐷₀∗ (1 + 𝑔)

(𝑘_𝑒− 𝑔) = 𝐷₁ (𝑘_𝑒− 𝑔)

where 𝑘_𝑒 is the equity discount rate and 𝑔 the expected growth in dividend payments. Based on the GGM, monetary policy affects equity prices in two ways. As interest rates are lowered the returns on bonds are reduced. Consequently, investors are willing to accept a lower rate of return 𝑘_𝑒 in equity as an alternative asset. Lower interest rates stimulate the economy as well increasing the expected growth in dividends 𝑔. The 𝑘_𝑒 reduction and the increase in 𝑔

(12)

6 both lower the denominator in the GMM and the stock price 𝑃₀ increases. With an increase in interest rates the opposite happens resulting in reduction in equity prices (Mishkin, 2019).

According to the efficient market hypothesis (EMH) all information are fully reflected in the prices of assets. The EMH is often associated with Eugene Fama who published a paper in 1969 where he reviewed the theoretical and empirical evidence behind the EMH. In his paper, Fama investigated the EMH where he grouped information into three subsets on how “all relevant information” are defined. The weak form defines all information as historical prices of a stock. The semi-strong form defines all information as historic prices and all publicly information. Finally, the strong form defines all information as any relevant information, public or private. His research indicated that stock prices show evidence of weak and semi- strong form of the EMH. If the EMH holds true, then the effect of the monetary policy decisions should immediately be reflected in relevant stock prices.

2.3 European Central Bank

The European Central Bank (ECB) is the central bank of the euro area, which comprises of countries that have adopted the euro as their primary currency. It has been responsible for conducting monetary policy for the euro area since the euro area was established on January 1^st, 1999. At the beginning the euro area consisted of 11 countries. Today there are a total of 19 member states in the euro area. The ECB together with the national banks of each member country form the Eurosystem which is the central banking system of the euro area (European Central Bank, 2021).

2.3.1 The Governing Council

The Governing Council is the main decision-making body of the ECB. The council consists of six members which form the Executive Board, in addition to governors of the central bank of each country in the euro area. The responsibilities of the Governing Council are threefold:

➢ to adopt the guidelines and take the decisions necessary to ensure the performance of the tasks entrusted to the ECB and the Eurosystem;

➢ to formulate monetary policy for the euro area. This includes decisions relating to monetary objectives, key interest rates, the supply of reserves in the Eurosystem, and the establishment of guidelines for the implementation of those decisions.

(13)

7

➢ in the context of the ECB’s new responsibilities related to banking supervision, to adopt decisions relating to the general framework under which supervisory decisions are taken, and to adopt the complete draft decisions proposed by the Supervisory Board under the non-objection procedure (European Central Bank, 2021).

2.3.2 ECB’s monetary policy decisions

Before the year of 2015, The Governing Council usually made a monetary decision every two weeks. In 2015 changes were made and now The Governing Council usually holds meetings every two weeks where decisions are taken relating to payment systems, financial stability, statistics, banknotes, legal affairs, banking supervision. Every six weeks the Governing Council makes monetary policy decision for the euro area. The monetary policy decisions are published on the day of the decision in a press release at 13:45 CET. Following the release there is a press conference at 14:30 CET where the decision is explained by the ECB president and questions by journalists answered. Four weeks after each monetary policy decisions the account of the meeting is released for transparency (European Central Bank, 2021).

2.3.2.1 Benchmark rates

The main instrument the Governing Council uses are interest rates to influence financing conditions. The rates are much the same as the rates described in chapter 2.1.1. There are set three key interest rates: interest rates for main refinancing operations, deposit facility, and marginal lending facility. The interest rate for main refinancing operations sets the value at which operations banks can borrow from the Eurosystem against a collateral. The deposit facility rate is the rate at which banks may use to make overnight deposits with the Eurosystem, which is set lower than the main refinancing rate. The marginal lending facility rate is set above the main refinancing operations rate and offers banks credit overnight from the Eurosystem. The three interest rates form a corridor to which banks lend to each other by setting a lower and upper bound with the deposit facility and marginal lending facility rates (European Central Bank, 2021).

2.3.2.2 Other monetary policy measures

Following the financial crisis in 2007 the ECB introduced several non-standard monetary measures to combat the effects of the financial crisis similar to those described in chapter 2.1.2. They offered unlimited credit to banks at a fixed rate, purchased debt securities,

(14)

8 lowering the interest rate on deposit facility to a negative, purchasing assets through an asset purchase program among other measures (European Central Bank, 2021).

2.3.3 ECB’s monetary policy objective

The primary objective of the monetary policy by the ECB is to maintain price stability which the ECB’s Governing Council defines as “a year-on-year increase in the Harmonized Index of Consumer Prices (HICP) for the euro are of below 2%”, which they adopted in 1998. The HICP is the uniform methodology used by EU Member States to measure change in prices of consumer goods and services. In 2003 the Governing Council further clarified their goal of price stability by setting the aim to maintain inflation rates close to, but below 2% over the medium term (European Central Bank, 2021).

In order to maintain price stability, the ECB uses an approach to analyze economic developments which they refer to as the two-pillar approach. The two pillars are economic analysis and monetary analysis which the Governing Council use to assess the risk to price stability and the monetary policy decisions (European Central Bank, 2021).

Economic analysis is used to evaluate price developments in short to medium-term determinants. It does so by analyzing real activity and financial conditions through supply and demands in goods, services, and factor markets. By looking at overall output developments, labor market conditions, various price and cost indicators, fiscal policy, and the balance of payments for the euro area, the ECB is able to make out price developments and derive expectations of the financial markets. With those values the ECB can with greater ability project inflation fluctuations over the short to medium-term (European Central Bank, 2021).

The ECB uses monetary analysis for evaluation of a longer-term horizon. Through assessing monetary and credit developments and linking them with the short to medium-term indications gathered from the economic analysis the ECB estimates future inflation and economic growth. To achieve their goal, they use various instruments which are constantly under assessment and revaluation. By studying developments of monetary aggregates supported by model-based analyses the ECB is able to obtain medium and long-term indicators from the monetary data (European Central Bank, 2021).

(15)

9 2.4 Previous studies

Many studies have been done on the effect monetary policy has on financial markets. Most of them have been focused on the U.S. market. However, in recent years the focus on the European market has increased.

The first study to investigate the effects of monetary policy on markets dates back to 1989, when Timothy Cook and Thomas Hahn studied the effect that changes in the Federal funds target rate had on market interest rates over the period of 1974-1979. Their conclusion was that changes in the Fed funds target rate resulted in large movements in short-term interest rates and smaller but significant movements in medium and long-term rates. A 1% increase in the Fed funds rate led to an increase of 55 basis point in the three-month T-bill rate whereas it only led to a 10-basis point increase in the 30-year bond yield. Using the same methodology as Cook and Hahn, Roley and Sellon (1995) showed that bond rates only increased by four basis points as a result of a 1% increase in the Fed funds rate over the period of 1987-1995.

Back in 2001, Kenneth N. Kuttner studied the relationship between monetary policy actions in the U.S. and market interest rates. He sorted between anticipated and unanticipated monetary policy changes using data from the futures market of the Federal funds, a method which has been implemented in much research on monetary policy decisions since then. His results showed that whereas anticipated changes had only a minor effect on market interest rates, the effect of unanticipated changes were large and highly significant.

Ben Bernanke and Kenneth Kuttner (2005) studied the effect of unexpected policy changes on the stock market as well as investigating the reasons for the effect on the market. They distinguished between anticipated and unanticipated policy actions which they based on the methodology used by Kuttner (2001). They concluded that the CRSP value-weighted index had a 1-day increase of roughly 1% when the target funds rate was reduced unexpectedly by 25 basis points. If the funds rate changes were anticipated there was little to no reaction by the market. Based on their research the effect on the stock market came not from the monetary policy‘s effect on the real interest rate but rather its effect on expected future dividends or on expected future excess returns.

In a more recent study Lucca and Moench (2015) studied the effect of monetary policy decisions made by the FOMC on average excess returns of US equities. They showed that US

(16)

10 stock returns showed large average excess returns in anticipation of scheduled FOMC meetings. They used intraday data for their research and concluded that stock returns showed on average 50 basis point increase over the period of 2 pm, on the day before an announcement, to 2 pm at FOMC announcement day.

Haitsma et al. (2016) investigated the effect of policy changes by the ECB on the EURO STOXX 50 index over the period of 1999-2015. They followed the event study method used by Bernanke and Kuttner (2005). They distinguished between the impact of conventional and unconventional monetary policy decisions and concluded that unconventional monetary policy surprises had the greatest effect on stock prices. Their result also indicated that past losers and value stocks were more affected by monetary policy surprises.

Altavilla et al. (2019) mapped the ECB policy communication into yield curve changes. They sorted the effect of information on monetary policy changes as Target, Timing, Forward Guidance and Quantitative Easing (QE) surprises during a monetary policy decision. Target were effects from changes in policy rates, Timing was policy expectation in the short-term whereas Forward Guidance was policy expectation over the medium-term. Finally, the QE effect was based on surprises relating to QE by the ECB. Their conclusion was that Target surprises only affect the yield curve on the short-term end whereas the Timing, Forward Guidance, and QE all affect the yield curve on the long-term. Additionally, they found out that changes in the yield curve were almost fully explained by the Target, Forward Guidance, and QE surprise factors.

(17)

11

3 Data and Methodology

This chapter describes the data and methodologies used in this study. The chapter begins with a description of which data was used for the research. Afterwards comes a description of an event study approach and the methodology carried out in the event study. The chapter concludes with a description of the average five-day excess return and the OLS methodology that was used in the study Cieslak et al. (2019) which will be applied for investigating the relationship between European stock returns and ECB monetary policy cycles.

3.1 Data selection

The period chosen for the study is from 1999, when the euro was introduced as an accounting currency, until the end of the year 2019. Because of the unusual market circumstances in 2020 as a consequence of the COVID-19 pandemic, that year was not included in the research.

The index used to represent the European stock market is the Fama/French European (FFE) index, which was gathered from the website of Kenneth R. French. The index is value- weighted and includes dividends and capital gains. The returns for the index are reported in U.S. dollars and are not continuously compounded. The one-month U.S. Treasury bill rate was used for calculating market excess returns in accordance with a research on international stock returns by Fama and French (2012). The T-bill rate was retrieved from the website of Kenneth R. French.

According to Park (2004), who wrote a guide on using event studies in a multi-country setting, the index which is used to represent the market should not include the market being investigated. Therefore, the MSCI ACWI ex EMU Index is used as a substitute for the market where the Economic and Monetary Union is excluded. The historical index price was gathered from the MSCI website. Yang et. al. (1985) and Chen et. al. (1986) showed that the choice between a value-weighted and an equal-weighted global index does not create a significant difference. The returns of the index were reported in U.S. dollars.

The MSCI ACWI ex EMU Index is value-weighted and composed of mid and large-cap stocks across 13 of 23 developed countries and 27 emerging countries and covers approximately 85% of the global equity outside of Europe. The index was launched on 1^st of January, 2001

(18)

12 so the data prior to that date are back-tested data. Running the tests from 2001-2019 showed no significant differences in the results (MSCI Inc., 2021).

Monetary policy decision dates were taken from the website of the ECB. The list of events used for the event study is shown in Table 1. There were a total of 265 monetary policy decisions made between 1999-2019. The period between decisions reached two weeks on 195 occasions and it reached four weeks on 46 occasions. Only once during the investigated period did the time between monetary policy decisions reach 6 weeks. Events before 15^th of December 1999 were excluded from the event study. The time period needed for estimation of normal returns was set to 250 days and the estimation window on events before 15^th of December did not extend over 250 days, excluding the first 19 monetary policy decisions.

Table 1: List of events in event study.

Total monetary policy decisions (MPD) 265 Occasion when:

Two weeks between MPD 195

Four weeks between MPD 46

Six weeks between MPD 1

-Excluded events from event study 19

In order to be in compliance with the methodology used by Cieslak et al. (2019), days that go beyond the sixth week are removed from the data in the replication study. This led to a removal of one day from the dataset. Days before the first monetary policy decision, on March 4^th, 1999, were removed as well.

3.2 Methodology

The methodology for the two studies performed are described in the following two subchapters. The calculations are done using STATASE 15, with the event study being performed using the Stata estudy module. Weeks and cycles are defined in the same way in both studies. Each monetary policy decision is defined as a 34-day cycle running from the day

(19)

13 before a monetary policy decision is made until 33 days after the decision. Table 2 shows the days included in each week period where the day of a monetary policy decision is day 0.

Weekends are excluded from the data; therefore 10 days represent two calendar weeks.

Returns during holidays are set to zero.

Table 2: Day count of the week variables where a monetary policy decision is set as day 0.

Week Day count

Week 0 -1 to 3

Week 1 4 to 8

Week 2 9 to 13

Week 3 14 to 18 Week 4 19 to 23 Week 5 24 to 28 Week 6 29 to 33

3.3 Event study

The history of event studies is long with the first published event study believed to be dated back to 1933 when James C. Dolley investigated the effects of stock splits on stock prices. The methodology of event studies has been improved since 1933, but the methodology used today is essentially the same as a methodology introduced in the late 1960s in studies by Ball and Brown (1968) and Fama, Fisher, Jensen and Roll (1969) (MacKinlay, 1997).

The purpose of an event study is to measure abnormal returns as a result of a certain event.

Abnormal returns are the difference between actual observed returns and returns that are predicted by a model. The predicted returns are subtracted by the actual returns observed and the difference between the two is considered to be the effect of the event in question.

The methodology used for this study was based on the paper by A. Craig MacKinlay, which was published in the Journal of Economic Literature in 1997.

3.3.1 Identifying event

The procedure of an event study is first to define an event. The event window is often constructed such that it includes more days than the actual event to capture effects after the stock market closes. In this case an event window covers all working days over a week. The

(20)

14 length of the estimation window is set as 250 days as recommended by MacKinlay (1997) and ends five days before the event.

Figure 1 shows the timeline of an event study where the event date is defined as 𝜏 = 0 with the event window represented as 𝜏 = 𝑇₁+ 1 to 𝜏 = 𝑇₂. The period of which normal returns of the stocks are estimated is defined as 𝜏 = 𝑇₀ + 1 to 𝜏 = 𝑇₁. The length of the event and estimation window is thus 𝐿₂ = 𝑇₂− 𝑇₁ and 𝐿₁ = 𝑇₁− 𝑇₀ respectively.

Figure 1 – The timeline of an event study. Picture taken from a paper by MacKinlay (1997) and edited.

3.3.2 Measuring normal returns

There are numerous approaches available for calculating normal returns and these can be split up into two categories: statistical and economic. Statistical models do not depend on any economic arguments but are purely based on statistical assumptions of the movement of returns. In statistical models the assumption is made that returns follow a jointly multivariate normal distribution where they are independently and identically distributed through time.

Economic models are based on assumptions relating to the behavior of investors and not solely on statistical assumptions (MacKinlay, 1997).

3.3.2.1 Constant Mean Return Model

The constant mean return model is a statistical model which is perhaps the simplest model.

Brown and Warner (1980, 1985) have however shown that the model often yields results similar to that of more advanced models when event dates are not clustered. The return of security i is assumed to be constant over time such that

𝑅_𝑖𝑡 = 𝜇_𝑖+ 𝑒_𝑖𝑡

𝐸(𝑒_𝑖𝑡) = 0 𝑣𝑎𝑟(𝑒_𝑖𝑡) = 𝜎_𝑒²_𝑖𝑡

where 𝜇_𝑖 is the mean return of security I and 𝑒_𝑖𝑡 the error term for time period t. The expected value of the error term is zero with variance 𝜎_𝑒²_𝑖𝑡 (MacKinlay, 1997).

(21)

15 Brown and Warner (1980, 1985) find that the constant mean return model outperforms many more complicated models as there is more estimation error when using the more complicated models.

3.3.2.2 Market model

The most common approach for estimating normal performance of returns is the market model. The market model is a statistical model which estimates the normal return in relation to the return of the market portfolio. An assumption is made that returns follow a joint normal distribution. The normal return of security 𝑖 is calculated as

𝑅_𝑖𝑡 = 𝛼_𝑖 + 𝛽_𝑖𝑅_𝑚𝑡+ 𝜀_𝑖𝑡 𝐸(𝜀_𝑖𝑡 = 0)

𝑣𝑎𝑟(𝜀_𝑖𝑡)= 𝜎_𝜀²_𝑖

where 𝑅_𝑖𝑡 is the return of security 𝑖 at time 𝑡. 𝑅_𝑚𝑡 is the market return at period 𝑡 with zero mean disturbance term of 𝜀_𝑖𝑡. The model parameters 𝛼_𝑖, 𝛽_𝑖 and 𝜎_𝜀²_𝑖 are estimated using the ordinary least squares (OLS) method. The OLS estimators of the parameters are:

𝛽̂_𝑖 =∑^𝑇_𝜏=𝑇¹ (𝑅_𝑖𝜏− 𝜇̂_𝑖)(𝑅_𝑚𝜏 − 𝜇̂_𝑚)

0+1

∑^𝑇_𝜏=𝑇¹ (𝑅_𝑚𝜏− 𝜇̂_𝑚)²

0+1

𝛼̂_𝑖 = 𝜇̂_𝑖− 𝛽̂_𝑖𝜇̂_𝑚

𝜎̂_𝜀²_𝑖 = 1

𝐿₁− 2 ∑ (

𝑇1

𝜏=𝑇0+1

𝑅_𝑖𝜏− 𝛼̂_𝑖 − 𝛽̂_𝑖𝑅_𝑚𝜏)²

where

𝜇̂_𝑖 = 1

𝐿₁ ∑ 𝑅_𝑖𝜏

𝑇1

𝜏=𝑇0+1

𝜇̂_𝑚 = 1

𝐿₁ ∑ 𝑅_𝑚𝜏.

𝑇₁

𝜏=𝑇₀+1

3.3.2.3 Capital Asset Pricing Model (CAPM)

The CAPM is an economic model where the expected return of a security is determined by its covariance with the market portfolio:

(22)

16 𝐸(𝑅_𝑖) = 𝑅_𝑓+ 𝛽_𝑖[𝐸(𝑅_𝑚) − 𝑅_𝑓]

𝛽_𝑖 =𝐶𝑜𝑣(𝑅_𝑖^𝑒, 𝑅_𝑚^𝑒) 𝑉𝑎𝑟(𝑅_𝑚^𝑒)

where the return of security I is calculated by the how risky the stock is estimated through 𝛽_𝑖 in relation to the market risk premium [𝐸(𝑅_𝑚) − 𝑅_𝑓] (Brooks, 2014).

The use of the CAPM models was popular in event studies in the 1970s. Since then, there have been deviations discovered in CAPM which cast doubt on the restrictions that CAPM has on the market model. Because of this, CAPM is rarely used since the restrictions can be relaxed by using the market model instead (MacKinlay, 1997).

3.3.2.4 The Fama French three-factor model

In a paper published in 1993, Eugene Fama and Kenneth French propose that firm size and book to market value of equity ratio have a strong influence on returns in addition to the systematic risk factor in the CAPM such that:

𝑅_𝑖𝑡 − 𝑅_𝐹𝑡 = 𝑎_𝑖 + 𝑏_𝑖(𝑅_𝑀𝑡− 𝑅_𝐹𝑡) + 𝑠_𝑖𝑆𝑀𝐵_𝑡+ ℎ_𝑖𝐻𝑀𝐿_𝑡+ 𝑒_𝑖𝑡

where 𝑅_𝑖𝑡 is the return of security i and 𝑅_𝐹𝑡is assumed to be a risk-free asset. 𝑆𝑀𝐵_𝑡 is the excess return of small cap companies over large cap companies at period t. 𝐻𝑀𝐿_𝑡 is the excess return of stocks with high book-to-price ratio over low book-to-price stocks. 𝑎_𝑖 is a constant and 𝑏_𝑖, 𝑠_𝑖, and ℎ_𝑖 are the factor exposures.

3.3.2.5 The Carhart model

Carhart proposed an additional factor to the Fama French three-factor model in a paper published in 1997. In his paper Carhart adds a momentum factor based of a research done by Jegadeesh and Titman (1993). In their research they showed that there were significant positive returns generated by buying stocks that have performed well and selling stocks that have performed poorly with a holding period of 3 to 12 months. By including the momentum factor to the Fama French three-factor model the model is extended to:

𝑅_𝑖𝑡 − 𝑅_𝐹𝑡 = 𝑎_𝑖 + 𝑏_𝑖(𝑅_𝑀𝑡− 𝑅_𝐹𝑡) + 𝑠_𝑖𝑆𝑀𝐵_𝑡+ ℎ_𝑖𝐻𝑀𝐿_𝑡+ 𝑝_𝑖𝑃𝑅1𝑌𝑅_𝑡+ 𝑒_𝑖𝑡 where 𝑃𝑅1𝑌𝑅_𝑡 is the one-year momentum factor and 𝑝_𝑖 the factor exposure.

(23)

17 3.3.2.6 The Fama French five-factor model

Fama and French published a paper in 2012 where they proposed two additional factors to their three-factor model for estimating returns. The two factors are profitability and investment factors, and the model extends to:

𝑅_𝑖𝑡− 𝑅_𝐹𝑡= 𝑎_𝑖 + 𝑏_𝑖(𝑅_𝑀𝑡− 𝑅_𝐹𝑡) + 𝑠_𝑖𝑆𝑀𝐵_𝑡+ ℎ_𝑖𝐻𝑀𝐿_𝑡+ 𝑟_𝑖𝑅𝑀𝑉_𝑡+ 𝑐_𝑖𝐶𝑀𝐴_𝑡+ 𝑒_𝑖𝑡. The 𝑅𝑀𝑉_𝑡 is the difference between returns on diversified stock portfolios with robust and weak profitability. The 𝐶𝑀𝐴_𝑡 factor is the difference between returns on diversified stock portfolios of low and high investment firms. 𝑟_𝑖 and 𝑐_𝑖 are factor exposures.

3.3.2.7 Choice of model

Brown and Warner (1985) examined through simulation procedures how daily stock returns affect event study methodologies. The results from their study reinforced their conclusion of their previous work with monthly data (Brown and Warner, 1980), that the OLS market model and using standard parametric tests are well-specified under variety of conditions and generally present few difficulties in the context of event study methodologies.

Kothari and Warner (2007) talk about in their paper that risk adjustments for short-horizon event studies are straightforward and typically unimportant, whereas for long-horizon event studies the appropriate adjustment for risk is critical for calculating abnormal returns of a security. In their paper they mentioned that while the definition of long-horizon event studies is arbitrary, it generally applies to event windows of 1 year or more.

Mackinlay (1997) argues in his paper that the gains from implying models with additional factors are limited. His reasoning is that the marginal explanatory power of additional factors to the market factor are small and the reduction in the variance of abnormal returns are therefore small. Binder (1998) concludes in his paper on event study methodologies that the market model works well as a measure of the benchmark rate of return. Campbell et. al.

(2009) note in their research that the single-index market model is popular among researchers for estimating normal returns in a multi-country setting.

Based on this information, the market model was chosen for estimating normal returns in this study and normal returns predicted according to chapter 3.3.2.2.

(24)

18 3.3.3 Measuring abnormal returns

Abnormal return is the difference between actual returns and the predicted normal returns.

Abnormal returns for the event window 𝐿₂ are noted as 𝐴𝑅_𝑖𝜏, 𝜏 = 𝑇₁+ 1, . . . , 𝑇₂. The sample of abnormal returns is

𝐴𝑅_𝑖𝜏 = 𝑅_𝑖𝜏− 𝛼̂_𝑖 − 𝛽̂_𝑖𝑅_𝑚𝜏.

where 𝑅_𝑖𝑡 is the return of security 𝑖 at time 𝑡 and 𝑅_𝑚𝑡 is the market return at period 𝑡, 𝛼̂_𝑖 and 𝛽̂_𝑖 are parameters that are estimated using the market model. Abnormal returns are aggregated across time and across securities in order to make an interpretation about the returns using the cumulative abnormal return (CAR). In this study there is no event clustering so the abnormal returns for security 𝑖 can be aggregated across day 𝜏₁ to day 𝜏₂ as:

𝐶𝐴𝑅_𝑖(𝜏₁, 𝜏₂) = ∑^𝜏_𝜏=𝜏² ₁𝐴𝑅_𝑖𝜏.

Abnormal returns are aggregated across separate securities 𝑖 for period 𝜏 as:

𝐴𝑅_𝜏

̅̅̅̅̅ = 1

𝑁∑ 𝐴𝑅_𝑖𝜏

𝑁

𝑖=1

Average abnormal returns are aggregated across the event window as:

𝐶𝐴𝑅̅̅̅̅̅̅(𝜏₁, 𝜏₂) = ∑ 𝐴𝑅̅̅̅̅̅_𝜏

𝜏₂

𝜏=𝜏₁

with a variance of:

𝑣𝑎𝑟(𝐶𝐴𝑅̅̅̅̅̅̅(𝜏₁, 𝜏₂)) = ∑ 𝑣𝑎𝑟(𝐴𝑅̅̅̅̅̅_𝜏

𝜏2

𝜏=𝜏1

)

where variance with a large estimation window 𝐿₁ is calculated as:

𝑣𝑎𝑟(𝐴𝑅̅̅̅̅̅) =_𝜏 1

𝑁²∑ 𝜎_𝜀²_𝑖

𝑁

𝑖=1

.

As events do not overlap, the covariance term is zero and it can be inferred that:

𝐶𝐴𝑅̅̅̅̅̅̅(𝜏₁, 𝜏₂)~𝑁[0, 𝑣𝑎𝑟(𝐶𝐴𝑅̅̅̅̅̅̅(𝜏₁, 𝜏₂))]

(25)

19 3.3.4 Testing the null hypothesis

To test the null hypothesis that an event has no effect on the mean or the variance of returns, an estimator for 𝜎_𝜀²_𝑖 must be used since the true value is unknown. The 𝜎̂_𝜀²_𝑖 measure from the market model regression is used to calculate 𝑣𝑎𝑟(𝐴𝑅̅̅̅̅̅) and the hypothesis: _𝜏

𝐻₀: 𝐶𝐴𝑅̅̅̅̅̅̅ = 0 can be tested through:

𝜃₁ = 𝐶𝐴𝑅̅̅̅̅̅̅(𝜏₁, 𝜏₂)

√𝑣𝑎𝑟(𝐶𝐴𝑅̅̅̅̅̅̅(𝜏₁, 𝜏₂))

~𝑁(0,1)

(MacKinlay, 1997).

3.3.5 Non-parametric test

The test above assumes that stock returns are normally distributed. Studies have however shown that stock returns do not follow a normal distribution (Fama (1976), Brown and Warner (1985) etc.). To combat this problem non-parametric tests, which make no assumption about the distribution of stock returns, have been used. MacKinlay (1997) mentions two popular nonparametric tests, the sign test and the rank test. Prior research has shown that the effectiveness of the two tests is diminished when they are extended to multiple day tests of cumulative abnormal returns (Cowan (1992), Kolari and Pynnönen (2010)).

The generalized rank (GRANK) test is a non-parametric test proposed by Kolari and Pynnönen (2011) as an improvement on the rank test introduced by Corrado (1989) and Corrado and Zivney (1992). In their method the estimation period returns and the cumulated return on the event period are used for parametric testing, whereas in the previous methods all the combined estimation and event period returns were used in the ranking. The number of observations in the testing are therefore one plus the length of the estimation window which makes the test less sensitive to the length of the event window compared to previous rank tests (Kolari & Pynnönen, 2011).

𝑈_𝑖0 denotes the demeaned standardized abnormal rank of the cumulative abnormal return where under the null hypothesis of no mean effect:

(26)

20 𝐸[𝑈_𝑖0] = 0

for all 𝑖 = 1, … , 𝑛. The null hypothesis of no mean effect:

𝐻₀: 𝜇_𝜏 = 0

where the expected cumulative abnormal return over period of length 𝜏 is 𝜇_𝜏 = 𝐸[𝐶𝐴𝑅_𝜏].

The generalized rank t-statistic (GRANK-T) for testing 𝐻₀ is defined as:

𝑡_{𝑔𝑟𝑎𝑛𝑘} = 𝑍( 𝑇 − 2 𝑇 − 1 − 𝑍²)^1/2 where

𝑍 =𝑈̅₀ 𝑆_𝑈̅

with

𝑆_𝑈̅ = √1 𝑇∑𝑛_𝑡

𝑛 𝑈̅_𝑡²

𝑡∈𝑇

𝑈̅_𝑡 = 1

𝑛_𝑡∑ 𝑈_𝑖𝑡

𝑛_𝑡

𝑖=1

𝑈_𝑖𝑡 is the demeaned standardized abnormal ranks of the generalized standardized abnormal returns calculated as:

𝑈_𝑖𝑡 = 𝑅𝑎𝑛𝑘(𝐺𝑆𝐴𝑅_𝑖𝑡)

𝑇 + 1 − 1/2 where 𝐺𝑆𝐴𝑅_𝑖𝑡 is computed as:

𝐺𝑆𝐴𝑅_𝑖𝑡 = { 𝑆𝐶𝐴𝑅_𝑖^∗, 𝑓𝑜𝑟 𝑡₁+ 1 ≤ 𝑡 ≤ 𝑡₁+ 𝜏 𝑆𝐴𝑅_𝑖𝑡, 𝑓𝑜𝑟 𝑡 = 𝑇₀+ 1, … , 𝑡₁, 𝑡₁ + 𝜏 + 1, … , 𝑇₂

where 𝑆𝐶𝐴𝑅_𝑖^∗ is the restandardized SCAR, proposed by Boehmer et al. (1991) to account for event induced volatiltity. 𝑆𝐶𝐴𝑅_𝑖^∗ is calculated as:

𝑆𝐶𝐴𝑅_𝑖^∗ =𝑆𝐶𝐴𝑅_𝑖𝜏 𝑆_{𝑆𝐶𝐴𝑅,𝜏} and

(27)

21 𝑆_{𝑆𝐶𝐴𝑅,𝜏} = √ 1

𝑛 − 1∑(𝑆𝐶𝐴𝑅_𝑖𝜏− 𝑆𝐶𝐴𝑅̅̅̅̅̅̅̅_𝑖𝜏)²

𝑛

𝑖=1

(Kolari & Pynnönen, 2011).

3.4 Replication study

This chapter describes the methodology used for analyzing the biweekly excess return of the Fama/French European index over the 1-month U.S. T-bill rate, following a MPD by the ECB.

The methodology is based on the methodology used by Cieslak, Morse and Vissing-Jorgensen (2019).

Excess returns are defined as the return of the FFE index subtracted by daily 1-month U.S.

Treasury bill rate:

𝐸𝑥𝑐𝑒𝑠𝑠 𝑟𝑒𝑡𝑢𝑟𝑛 = 𝑅_𝐹𝐹𝐸− 𝑅_{𝑇.𝑏𝑖𝑙𝑙} and the excess returns for period 𝑡₀ to 𝑡 are calculated as:

𝐸𝑥𝑐𝑒𝑠𝑠 𝑟𝑒𝑡𝑢𝑟𝑛_𝑡= (1 + 𝑅_𝐹𝐹𝐸

𝑡0)∗ … ∗ (1 + 𝑅_𝐹𝐹𝐸_𝑡) − (1 + 𝑅_{𝑇.𝑏𝑖𝑙𝑙}

𝑡0) ∗ … ∗ (1 + 𝑅_{𝑇.𝑏𝑖𝑙𝑙}_𝑡).

The five-day excess return for the FFE index is thus calculated as:

𝐸𝑥𝑐𝑒𝑠𝑠 𝑟𝑒𝑡𝑢𝑟𝑛_𝑡₅ =

(1 + 𝑅_𝐹𝐹𝐸_𝑡0) ∗ … ∗ (1 + 𝑅_𝐹𝐹𝐸_𝑡+4) − (1 + 𝑅_{𝑇.𝑏𝑖𝑙𝑙}_𝑡0) ∗ … ∗ (1 + 𝑅_{𝑇.𝑏𝑖𝑙𝑙}_𝑡+4) and the average of the excess returns is calculated for cycle 𝑖 as:

𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝑒𝑥𝑐𝑒𝑠𝑠 𝑟𝑒𝑡𝑢𝑟𝑛_𝑖 =∑^𝑛_𝑡=1𝐸𝑥𝑐𝑒𝑠𝑠 𝑟𝑒𝑡𝑢𝑟𝑛_𝑡

𝑁 .

3.4.1 Regression

The relationship between excess returns and weekly cycles is estimated through regression with robust variance estimates. Estimation of variance refers to the estimated standard errors as they are a subset of the estimated variance-covariance matrix. Regression with robust variance estimates is also referred to as Huber-White, the sandwich estimator of variance, and other names. It is used to combat heteroskedasticity in the error terms, and for the robust variance estimator 𝑉̂(𝐵̂) it is calculated as:

(28)

22 𝑉̂(𝐵̂) = (𝑋′𝑋)⁻¹(∑ 𝑒̂𝑥_𝑗² _𝑗′𝑥_𝑗

𝑛

𝑗=1

)(𝑋′𝑋)⁻¹

where 𝑋 is 𝑛 𝑥 𝑘 matric of covariates and 𝑒̂_𝑗² the square residual of the residual 𝑒̂ = 𝑦_𝑗 _𝑗− 𝑥_𝑗𝛽̂

where 𝑦 is an 𝑛 𝑥 1 vector representing the dependent variable and 𝑥_𝑗 the 𝑗th row of 𝑋 (StataCorp, 2019).

3.5 Returns of different investment strategies

The following formulas are used in order to compare the returns of different investment strategies over the investigated period. The returns 𝑅_{𝑡,𝑡+Δ𝑡} for period 𝑡 to Δ𝑡 are calculated as:

𝑟_{𝑡,𝑡+Δ𝑡} = 𝑃_𝑡+Δ𝑡− 𝑃_𝑡 𝑃_𝑡

and the returns for period 𝑡 to 𝑛Δ𝑡 as:

𝑟_{𝑡,𝑡+𝑛Δ𝑡} = (1 + 𝑟_{𝑡,𝑡+Δ𝑡}) ∗ (1 + 𝑟_{𝑡+Δt,𝑡+2Δ𝑡}) ∗ … ∗ (1 + 𝑟𝑡+(𝑛−1)Δ𝑡,𝑡+𝑛Δt) − 1 (Munk, 2018)

In order the evaluate the risk-return tradeoff of each trading strategy the Sharpe ratio is calculated. The ratio was first introduced by William Sharpe in 1966 from whom the ratio got its name. The ratio is calculated by dividing the risk premium by the standard deviation of the excess return and represents the additional amount of return received per unit of risk:

𝑆𝑅 = 𝑅𝑖𝑠𝑘 𝑝𝑟𝑒𝑚𝑖𝑢𝑚 𝑆𝐷 𝑜𝑓 𝑒𝑥𝑐𝑒𝑠𝑠 𝑟𝑒𝑡𝑢𝑟𝑛 (Bodie, Kane, & Marcus, 2014).

(29)

23

4 Results

This chapter shows the results of the studies performed on biweekly returns of the European stock market in relation to monetary policy meetings made by the ECB. Weekends are excluded from the data and returns during holidays are set to zero. The chapter is divided in two subchapters which show the results of the event study and the replication study based on the methodology used in the study by Cieslak et.al (2019).

4.1 Event study

The results of the event studies performed for each even week period are shown in Table 3.

In the first column the results are shown for the abnormal during the week of a monetary policy decision. The second and third column show the results from abnormal returns of week 2 and 4. The fourth column shows the effect on week 6.

During week 0, abnormal returns are 4.1 basis points over predicted returns with a t-statistic of 0.36 and GRANK-T of 0.23. The CAR̅̅̅̅̅̅ for week 2 is negative by 6.6 basis points. The t-statistic for the weekly period is -0.54 and the GRANK-T is -0.90. The CAR̅̅̅̅̅̅ for week 4 is a negative value as well, with negative abnormal returns of around 20 basis points. The t-test statistic is -0.93 and the GRANK-T has a value of -0.62. There was only one incident where the period between monetary policy decisions reached the sixth week. In that case the CAR̅̅̅̅̅̅ over the week were 0.623%. The t-statistic and the GRANK-T was the same, 0.48, for the weekly period.

Table 3: Results of event studies of returns of the FFE Index over even weeks. * indicates significance at 10% level ** indicates significance at 5% level *** indicates significance at 1% level.

Week 0 Week 2 Week 4 Week 6

CAR̅̅̅̅̅̅ _0.041% _-0.066% _-0.204% _0.623%

N 244 194 46 1

t-test (p-value)

0.36 (0.72)

-0.54 (0.59)

-0.93 (0.35)

0.48 (0.63) GRANK-T

(p-value)

0.23 (0.82)

-0.90 (0.37)

-0.62 (0.54)

0.48 (0.63)

The results of event studies performed for the returns during odd weeks are displayed in Table 4. The first column shows the result for week -1; the week before a monetary policy decision.

(30)

24 The CAR̅̅̅̅̅̅ for week -1 is -0.031% with a t-test statistic of -0.27 and GRANK-T of -0.76. The cumulative abnormal returns on week 1 are not significantly different from zero either. The CAR̅̅̅̅̅̅ over the period is 0.030% with a t-statistic of 0.25 and really low GRANK-T value of 0.04.

The t-statistic for week 3 is higher, 0.72, with a CAR̅̅̅̅̅̅ of 0.117% and a GRANK-T of 1.14. For week 5 the CAR̅̅̅̅̅̅ goes up considerably to 0.427% with a relatively high t-statistic of 1.48 The p-value is 0.14 meaning the CAR̅̅̅̅̅̅ is not significantly different from zero at the 10% critical value. The GRANK-T for week 5 is 0.67.

Table 4: Results of event studies of returns of the FFE Index over odd weeks. * indicates significance at 10% level ** indicates significance at 5% level *** indicates significance at 1% level.

Week -1 Week 1 Week 3 Week 5

CAR̅̅̅̅̅̅ _-0.031% _0.030% _0.117% _0.427%

N 246 207 98 23

t- test (p-value)

-0.27 (0.79)

0.25 (0.80)

0.72 (0.47)

1.48 (0.14) GRANK-T

(p-value)

-0.76 (0.45)

0.04 (0.97)

1.14 (0.25)

0.67 (0.51)

The event study for even weeks and odd weeks does not show a sign of a pattern in weekly returns following a monetary policy decision. Cumulative abnormal returns are negative in week -1 and go to a positive value on week 0 and week 1 but drop to a negative value on week 2. The abnormal return rises back to a positive value, going down to a negative value on week 4 and rising again to a positive abnormal return on week 5 and 6. However, none of the CAR̅̅̅̅̅̅

values are significantly different from zero or differ significantly from the mean.

4.2 Replication

The following chapter presents the results from the study using the methodology Cieslak, Morse and Vissing-Jorgensen used in their research on even weeks during a FOMC cycle. The day count starts at February 24^th, 1999, one week before the first monetary policy decision by the ECB. As in the study by Cieslak et al. (2019), days which go beyond week 6 are removed from the data which removes one day from the dataset. However, if the day would have been included the results would be almost identical.

(31)

25 4.2.1 Five-day average excess return

The five-day average excess return is the five-day return of the European index, reported in U.S. dollars, over the five-day return of the 1-month U.S. T-bill rate. Returns are shown in Figure 2 where date 0 is the day of a monetary policy decision. Weekends are excluded from the graph such that 10 days on the graph represent two weeks following a monetary policy decision. In the case of a holiday the returns are set to zero. The graph shows the five-day cumulative excess return of the FFE Index from day t to t + 4.

Figure 2 shows average five-day excess returns of the FFE Index over a monetary policy decision cycle over the period of 1999-2019. At the week before monetary policy decision (day -6 to -2) the five-day average excess returns are 0.20%. During week 0 (day -1 to 3) returns are 0.08%, dropping slightly down to 0.01% at week 1 (day 4 to 8). Average excess returns are even lower during week 2 (day 9 to 13) at -0.09%. At week 3 (day 14 to 18) there is a substantial increase of 70 basis points in returns, where they go to 0.61% over week 3 which lowers throughout the week, ending at a five-day average return of 0.19% over week 4 (day 19 to 23). Returns are similar over week 5 (day 24 to 28) at 0.15% Going past week 6 (day 29 to 33) there are a lot of fluctuations going forward as there is only one case where the time between monetary policy decision goes past week 5. Looking at Figure 2 there does not seem to be aclear weekly pattern observable for the five-day average excess return.

(32)

26

Figure 2: The average five-day excess returns of the FFE index over the T-bill rate. The vertical axis shows the five-day excess returns. The horizontal axis shows days relative to MPD where day 0 is the day of MPD, weekends are excluded. Five-day excess returns are calculated using Equation 20. Week -1 covers days -6 to -2, Week 0 covers days -1 to 3, week 1 covers days 4 to 8, week 2 covers days 9 to 13, week 3 covers days 14 to 18, week 4 covers days 19 to 23, week 5 covers days 24 to 28, week 6 covers days 29 to 33.

4.2.2 Linear regression

Listed in Table 5 are the results of regressions of the Fama/French Europe index on a dummy variable set as 1 during even weeks. The beta coefficient is displayed with the t-statistics robust to heteroskedasticity in parentheses. As Table 5 shows, none of the dummy week variables show significant relationship with the index. The two highest values are when the dummy is set as 1 during week 2 and week 4 where they show a negative relationship between the variables of -0.05 and -0.06, respectively. The values are however not significant at the 10% critical value indicating there is no relationship between even weeks and daily excess returns of the European index.

-2.500%

-2.000%

-1.500%

-1.000%

-0.500%

0.000%

0.500%

1.000%

-6 -1 4 9 14 19 24 29

Five-day excess return

Days since a monetary policy decision by the ECB

Monetary policy decisions by the ECB and stock returns Stock returns over weeks between monetary policy decisions