• No results found

The effect of Quantitative Easing announcement on stock indexes : an event-study and regression analysis on the Hang Seng Index and its sub-indexes

N/A
N/A
Protected

Academic year: 2021

Share "The effect of Quantitative Easing announcement on stock indexes : an event-study and regression analysis on the Hang Seng Index and its sub-indexes"

Copied!
37
0
0

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

Hele tekst

(1)

Announcement on Stock Indexes

An event-study and regression analysis on the Hang Seng Index and its Sub-Indexes.

Rense Verbeek 11014334

Bachelor’s Thesis Finance & Organisation

2017/2018

Supervisor R.C.R. van Lamoen

Abstract

Because of the global financial crisis and its aftermath, central banks implemented an unconventional monetary tool named Quantitative Easing. This study examines the relationship between a Quantitative Easing announcement1 by the FED and the stock indexes in Hong Kong. It focusses on the

Hang Seng Index, the Hang Seng Property Index and the Hang Seng Finance Index and looks at three different time intervals. An interval before, during and after the announcement. The tools used to examine this possible effect are a regression analysis including dummy variables based on time and control variables like the Citigroup Economic Surprise Index and the Volatility Index of the HSI. Furthermore, two event studies are executed, which are based on the market-model (CAPM) and on a price/earnings-ratio model. This study shows that the effect of QE on stock markets heavily depends on the methodology that is used. According to the event studies, an event effect occurred. However, the regression showed that there does not exists a significant relationship between the Indexes in Hong Kong and the QE announcement.

1 The examined announcement is the announcement by the FED of 18 December 2013 to reduce the amount of Quantitative Easing by $10 billion a month.

(2)

Statement of Originality

This document is written by Rense Verbeek who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Table of Contents

Abstract ... 1 1. Introduction ... 4 2. Literature Review ... 5 3. Implementation of QE ... 10 4. Hypothesis ... 12 5. Methodology ... 12 6. Results ... 18 7. Conclusion ... 22 8. References ... 24 9. Appendix ... 26

(4)

1. Introduction

The crisis of 2008/2009 is known as one of the biggest crises in the modern financial world. This crisis can be divided into 2 phases. One is the phase of August 2007 until August 2008. The second phase of the financial crisis started when the investment bank Lehman Brothers collapsed. This resulted in uncertainty in the financial markets. Banks were reluctant to lend money to each other. New loans fell by 47%2 in Q4 of 2008 (Ivashina & Scharfstein, 2010). To boost the economy, the FED decreased the Federal Funds Rate to the zero lower bound and started with the implementation of Quantitative Easing, which is known as an unconventional monetary policy. Only the Bank of Japan used Quantitative Easing before to prevent deflation in the early years of the 20th century. Quantitative Easing can be seen as a policy to purchase assets from the private sector and therefore increase the balance sheet of the central bank. Due to these purchases, the amount of money in the economy increases, which results in an increase in nominal spending and inflation (Joyce et al, 2011).

Quantitative Easing has already extensively been examined by Christensen & Gillan (2018), Georgiadis & Grab (2016) and Joyce et al. (2012). The majority of these papers does research on the effects of QE on the exchange rates, interest rates or the liquidity premium. Furthermore, papers like Kurihara (2006) and Jonathan & Thomas (2012) focus only on the effects QE has on the country which implemented it. However, less research has been conducted on cross-country effects. Furthermore, the sectors that were affected by the crisis were the Financial sector and the Real Estate sector. Therefore, it would be interesting to look at the relationship between these sectors and Quantitative Easing. Additionally, insights in these cross-country effects could be beneficial to (institutional) investors and policy makers in the affected country to make better decisions in the future.

On 18 December 2013, the FED announced to reduce the amount of Quantitative Easing from $85 billion to $75 billion. Furthermore, the FED announced to reduce the amount even more in the near future, if the conditions on the labour market would improve.

In this research, the effects of this announcement on three different indexes are examined. The Hang Seng Index, Hang Seng Properties and Hang Seng Finance. These Indexes3 are all leading indexes in the region and cover companies located in Hong Kong and

2 Relative to the quarter before (Q3 of 2008)

3 More information about the indexes and their components could be found on: https://www.hsi.com.hk/eng/about-us/company-profile

(5)

Mainland China. This study looks at three different periods in time. The period surrounded by the event, the period prior the event and the period after the event. This research examines two alternative methodologies that are often used in studies to examine the effect of events, namely an event study and a regression to condition the event with influences from other factors. Other papers that examine an event by a regression analysis or an event study are Agrawal & Kamakura (1995), Austin (1993) and Binder (1985).The general idea of using regression for an event study is based on Binder (1985), while the paper of Georgidas & Grab (2016) connected it with Quantitative Easing.

The structure of this study is as follow. First, the relevant literature regarding monetary policies and the efficient market hypothesis will be explained in Section 2. Section 3 contains information about the implementation of QE by the FED. The hypotheses will be stated in Section 4. Section 5 will include the methodology of the event study and the regression. In section 6, the results will be presented. The conclusion and the limitations of this research can be found in section 7.

2. Literature Review

In this section, the literature around this subject is explained. First (2.1), conventional monetary policy is explained. After that, the theory about QE is told in section 2.2. Section 2.3 contains empirical information about Quantitative Easing. Because one of the ways used in this paper to examine the effect of QE is an event study, section 2.4 will deal with how prices react to new information. The hypothesis will be explained in section 2.5.

2.1 Conventional Monetary Policy

Price stability, also known as stable and low inflation, is viewed at the most important goal of monetary policy conducted by central banks. Other possible goals of monetary policy are high employment, economic growth, interest rate stability, stability in foreign exchange markets and stability in the financial markets (Mishkin, 2013). Currently, the view on price stability is close but not higher than 2% inflation per year.

Monetary Policy is normally conducted by buying and selling short-term debt securities to target short-term interest rates. This is also known as Open Market Operations. These purchases and sales of assets lead to a change in the monetary base and the short term interest rates (Mishkin, 2013). This conventional monetary tool can potentially stimulate the

(6)

economy in two different ways. First, by decreasing the interest rate, stock prices will increases because of the negative relationship (Berk & DeMarzo, 2014). Furthermore, the lower interest rate encourages borrowing for consumption or investing. This is known as asset

price channels. The other way of stimulating the economy is through credit channels. Credit

channels reduce the effect of economic frictions that decrease the willingness to borrow. For example, low profitability expectations can lead to a situation where creditworthy borrowers do not look for seeking borrow. Low creditworthy borrowers do not stop looking for borrow. Banks know this and will be very reluctant to make loans. Problems of moral hazard and adverse selection will therefore decrease by a decreasing interest rate (Fawley & Neely, 2013).

Other monetary tools that influence the interest rates are the standing facilities and the reserve requirements. Central banks have two facilities: the deposit facility and the lending facility. The deposit facility is the option which is given to banks to deposit their funds at the central bank. The central bank has to pay an interest to the bank which is below the overnight rate. The lending facility enables banks to lend funds from the central bank at an interest rate which is above the overnight rate. The other tool to influence the interest rate is the reserve requirements ratio. This tool has a negative effect on the interest rate. This means that when the central bank decreases the ratio, banks could hold less reserves, which results in a decreasing overnight rate (Mishkin, 2013)

2.2 Unconventional Monetary Policy

As the crisis intensified, the meltdown of the global economy resulted in new challenges for central banks. Central banks believed conventional monetary tools would not be enough to meet the 2% inflation target. This was partially due to the fact that the interest rates almost touched the zero lower bound on December 16, 2008 (Fawley & Neely, 2013; Blinder, 2010). A situation where the interest rate hits the zero lower bound has come to be called “liquidity trap”. This terminology is based on Kruger (1998). The logic behind this is as follows. The real interest rates is the rate which matters. The real interest rate can be defined as (R = I – P) where R is the real interest rate, I the nominal interest rate and P the inflation (CPI). In times of crisis, central banks need to push real interest rates to values smaller than zero. However, if the nominal interest rate is equal to zero, only a small positive value of P is left over, which creates a small negative real interest rate. In other words, conventional monetary policy cannot create a strong effect anymore (Blinder, 2010). Because of this liquidity-trap, central

(7)

banks started with the implementation of an unconventional tool. In order to boost the economy, central banks expanded their balance sheets by purchasing large amounts of public and private assets (Joyce et al, 2011). The ECB bought mostly assets from the banking sector, while the Bank of England and the Federal Reserve purchased assets mostly from non-bank parties (Mishkin, 2013). These purchases were financed by central bank money and led to an increase in the balance sheets of the ECB, FED and Bank of England by respectively 150%, 200% and 300% (Mishkin,2013). This was done to increase nominal spending and thus help to achieve the 2% target. This policy is known as Quantitative Easing (QE) (Joyce et al, 2011). The policy of Quantitative Easing was different from credit easing. Because credit easing is a policy to reduce the interest rate and restore market function. However, QE takes place when the interest rate is already around the zero lower bound (Fawley, Neely, 2013).

Quantitative Easing has a few possible channels through which asset purchases may affect the economy. Three of these channels work through asset prices, like policy signalling

effects, portfolio balance effects and liquidity premia effects. Other effects are confidence effect and bank-lending effects (Joyce et al, 2011).

The policy signal effect focusses on the actions of the Federal Reserve and what kind

of new information they provide to investors about the short-term interest rates in the future. This is the most straight-forward way of how Quantitative Easing affects the economy (Christensen & Gillan, 2018). In general, these announcements may signal to investors that the central banks changed their views on future economic conditions. Because of these signals, investors will change their expectations and therefore change their behaviour (Bauer & Rudebusch, 2014). Mishkin (2013) argues that because of this change in behaviour the expectations of inflation will increase and drive up wages and prices.

Long-term yields can be divided into two different components. The risk premium and the average level of short-term risk free interest rates. The risk premium equals the expected additional return that investors want for holding an asset which has some sort of risk. The average level of short-term risk free interest rates is the return investors could get when they invest in these short-term assets. Quantitative Easing could affect the long-term interest rate by changing one of these components. Nevertheless, the FED did not use QE to signal that the future short term interest rates would remain extraordinary low. They even mentioned that it would be possible to raise the short-term interest rates if necessary (Gagnon et al, 2011). According to Gagnon et al. (2011), the primary channel through which quantitative

(8)

easing works is through the change in risk premium on assets. Because the central banks buy particular assets, the amount of securities hold by the private sector decreases. Therefore, the price of these assets will increase due to a bid up process. This will decrease the yield on those assets. Therefore, companies could cheaper raise finance from the capital markets (Mishkin, 2013). This is known as the portfolio effect. This effect is all under the assumption that securities are not perfect substitutes.

Furthermore, Central Banks increase liquidity in financial system by increasing the reserve balances and buying long term securities. Because of that, the liquidity in the market increases. This is partially because they are a significant buyer in the market and because of the fact that reserves are more liquid than long-term securities. Because of the increasing liquidity, the liquidity premium of treasuries would decrease. Therefore, QE would cause an increase in the yield on treasuries. This is known as the liquidity premium effect (Krishnamurty & Vissing-Jorgensen, 2011).

The confidence effect was first introduced by Keynes’ “state of confidence”. The two determinants of this transmission channel are the confidence of the borrowers and the confidence of the lenders which are respectively: “the confidence that borrowers have about future yields” and “the confidence that lenders have in financing borrowers” (de Bondt, 2015). By taking credible actions in order to save the financial system, the state of confidence will increase.

The bank lending effect is an effect which occurs when the central bank purchases directly assets from non-banks. This increases customer deposits. Because of this, the banking sector could lend more money than they otherwise would have. This stimulates economic activity. However, according to Joyce et al (2011), this effect has little impact because of the need to reduce the size of the banks their balance sheets.

2.3 Empirical Quantitative Easing

Quantitative Easing is examined a lot in different studies recent years. In this section, the empirical findings of these studies will be showed. First, Joyce & Lasaosa (2011) show that due to the Quantitative Easing policy of the Bank of England, the yield on the gilts rose by 100 percentage points. They suggest that the increase was (partially) due to the portfolio rebalance effect. Furthermore, the share prices in their study recovered in 2009, which could

(9)

be an example of an wider impact of QE on, for example, asset prices. Hashem Pesaran & Smith (2016) agree on this. They tested the effectiveness of the policy in the UK. The outcome of their research showed that mainly the short-term showed beneficial effects due to QE. However, they argue that the statistical power to test this is really low.

Secondly, an empirical analysis of the Bank of Japan in July 2006 leaded to the finding that the QE of Japan contributed to an accommodative financial environment. It resulted in sustainable financial market stability and therefore avoiding other recurring fears of market instability (Ugai, 2006).

Another research conducted by Meaning & Zhu (2011), showed that both the policies of the UK and the US resulted in significant decreasing yields curves. Furthermore, recent asset purchases could be effective, but the asset purchases nowadays face some limitations. They argue that the long-term government bonds are already around the zero-lower bound. Additionally, the “surprise” factor is not available anymore. This corresponds to the confidence/signalling effect mentioned in the literature review. Finally, they warn that central banks should keep an eye on the inflation rate, because the inflation would increase if the programs continue.

Finally, the paper of Georgiadis & Grab (2016), which is the paper where a part of this study is built on, found that the announcement of APP (Asset Purchase Programme) of the ECB let to a depreciation of the Euro and an increase in European and global equity prices. Furthermore, they found that only the APP announcement had an immediate effect on the exchange rate, while all the other announcements of the ECB’s unconventional monetary policy had an effect on the (global) equity prices.

2.4 New information

The capital market’s goal is to allocate the ownership of the capital stock. If this market fully reflects all the available information into the prices, this market is called efficient (Fama, 1970). This efficient market hypothesis is based on the random walk idea. Which means that changes in prices represent departures from prices in the past (Malkiel, 2003). According to this theory, three different forms of efficiency are distinguished. When prices only depends on historical prices, we speak about the weak form of the efficient market hypothesis. Another form is the semi-strong form, in which prices depends on all the information that is publicy

(10)

available. Furthermore, we can speak about the strong form if all the information, private or public, is incorporated in the prices (Fama, 1970).

This view on efficient markets was widely accepted in the 20th century. This implied that neither technical analysis nor fundamental analysis would work to select undervalued stocks. The best way to achieve returns was to select a randomly independent portfolio of individual stocks without comparable risk between the stocks (Malkiel, 2003).

At the beginning of the 21th century, the dominance of the EMH decreased. Many academics thought that stock prices are at least partially predictable based on their valuation parameters. These patterns are initial dividend yields and initial price-earnings multiples. Research showed that there exists a negative relationship between the price earnings ratios and the return of the stock. This means that when the P/E-ratio decreases, the return would increase. Furthermore, there exists a positive relationship between the dividend yields (dividend/price) and the returns. This means that when D/P-ratio increases, the returns would increase either. Besides the valuation parameters, firm characteristics may influence the predictability of stock prices as well. The strongest effect over long periods is that smaller companies outperform larger companies, this effect is known as the size effect (Malkiel, 2003).

3. Implementation of QE

The first Quantitative Easing action began in early 2008. The Federal Reserve started with buying illiquid assets. On the other hand, they sold its holdings of Treasuries. This was done to provide more liquidity to the market. This leaded to a reduce in the liquidity premiums. Secondly, the Treasury “borrowed in advance of its needs”. The excess funds were deposed at the Central Bank accounts. Because of this, the treasury enabled the FED to expand its balance sheet. After the failure of Lehman Brothers the Open Market Operations drove the interest rate to the zero-lower bound. After that, the assets of the FED increased dramatically from 907$ billion on September 3, 2008 to 2.214$ trillion on November 12, 2008. Although the amount of capital of the FED didn’t change, the leverage ratio increased from 22:1 to 53:1. The goal of this policy was to decrease the risk premiums4, which increased a lot during the panic of the crisis (Blinder, 2010). This asset purchase program, also known as QE1, was

4 The risk premium could be defined as follow: the return on an asset is R. This return is composed out of two things. The premium to carry the risk of the asset (P) and the risk-free rate (RF). This leads to the following formula: R = RF + P.

(11)

originally designed to save the overall economy with particular attention to the housing market. Over 80% of the assets which were purchased, were assets related to the housing market (Fawley and Neewly, 2013).

In the second half of 2010, a new round of QE was implemented. This was due to the fact that the recovery did not go as fast as hoped and that the inflation decreased to 1%. The FED announced that it would continue with the asset purchases. On 3 November 2010, the FED announced to purchase Treasuries for an amount of $600 billion. The reason behind this was to increase the pace of economic recovery and to increase the inflation to its target. Because of this new round, long-term interest rates lowered and the inflation increased. No significant effect happened with the asset prices, because everyone already expected this kind of policies. Therefore, it was already calculated in the prices (Fawley & Neewly, 2013).

In the summer of 2011, fears of a new recession appeared. Financial Stress Indexes were really high. To reduce the long-term interest rate relative to the short one, the FED sold for an amount of 400$ billion in short-term assets. Furthermore, they purchased long term assets for the same amount. Therefore, the monetary base didn’t change. Because of this policy, the yield curve “twisted”, which is the reason this policy is called Operation Twist (Fawley & Neewly, 2013).

The last round of QE, which is known as QE3 and started on September 2012. The FED announced that it would purchase $40 billion in MBS until “the outlook for the labor market

does not improve substantially...in a context of price stability”. Furthermore, on December

2012, the FED announced to continue buying long term assets as in Operation Twist, but that they would stop with selling short term assets. Therefore, the monetary base would increase from then on (Fawley & Neewly, 2013).

On 18 December 2013 the FED announced to reduce the amount of Quantitative Easing from $75 billion dollar to $65 billion dollar. Furthermore, they announced to reduce this amount even more in the future if the labour market improved. The labour market improved and the FED reduced the amount of QE by $10 billion each 6 weeks. On 29 October 2014, the FED announced to quit the programme.

(12)

4. Hypothesis

The objectives of this study are to examine the effects between a Quantitative Easing announcement and the returns of stock-indexes. Based on the information provided in the Literature Review, the following hypotheses could be made:

Hypothesis 1: The expectation of an upcoming announcement about QE had a negative effect on the returns of the indexes5 of interest.

Hypothesis 2: The announcement about QE had a negative effect on the returns of the indexes6 of interest.

Hypothesis 3: The period after the announcement about QE had a negative effect on the returns of the indexes7 of interest.

The reason why all the periods should have a negative return related to the announcement, is because of the following. Quantitative Easing drives up asset prices due to different channels like the confidence channel, the portfolio rebalance channel, liquidity premium channel and the signalling channel (Christensen & Gillan, 2018; Joyce et al, 2011; Gagnon et al., 2011). If an announcement to reduce the amount of QE is published, these positive effects should decrease. This would result in a negative return in the short term period after the announcement. According to the semi-strong form of the efficient market hypotheses, share prices immediately react to new publicly available information (Fama, 1970). Therefore, it should show a negative effect from the publishing date. Furthermore, investors already had expectations about an decrease in the amount of QE. Therefore, a slightly negative return should be in the expectation period as well.

5. Methodology

The methodology of this paper is divided into four sections. The first section will expand on the sample used in this study. Section 3.2 explains the event study methodology whether section 3.3 explains the regression methodology. Finally, section 3.4 will summarize and

5 Indexes of interest: Hang Seng Index, Hang Seng Finance, Hang Seng Properties. 6 Indexes of interest: Hang Seng Index, Hang Seng Finance, Hang Seng Properties. 7 Indexes of interest: Hang Seng Index, Hang Seng Finance, Hang Seng Properties.

(13)

explain all the statistical requirements which are needed. The negative effect of the announcement on the index levels of the Hang Seng Index and its Sub-Indexes can be answered by an event study and a regression. This research will use both ways. The event study makes use of the standard market model (CAPM) and a model which depends on Price/Earnings Ratios. The regression approach is the same as in the paper of Georgiadis and Grab (2016), but will contain more variables.

5.1 Sample

Event study

The sample of the event study consists of data of the Hang Seng Index (HSI), Hang Seng Properties (HSNP), Hang Seng Finance(HSNF), MSCI ACWI (ex HK), S&P500 Real Estate Index and MSCI World Financials Index. The HSI, HSNP, HSNF are the indexes of interest. The other three indexes are used as benchmarks. The MSCI ACWI is used as benchmark of the HSI. The S&P500 Real Estate Index is used as HSNP benchmark and the MSCI World Financials Index is the HSNF’s benchmark. Both the P/E ratios and the index prices are collected. The collected daily and monthly data is between the period 01/01/2008 and 01/05/2014. The data are collected from DataStream and the official Hang Seng Index website.

Regression method

The regression method uses daily data between 21/02/2012 and 01/05/2014. The data necessary to run this regression are the daily returns of the Hang Seng Index (HSI), Hang Seng Properties(HSNP) and the Hang Seng Finance(HSNF). Furthermore, the data of the Citigroup Economic Surprise Index (Asia-Pacific) is needed as control variable. The VIX of the HSI is added as control variable as well. The CESI is collected from DataStream while the VIX is collected from Bloomberg.

5.2 Event Study

The event study is computed in two different ways. It is done with the market model (CAPM) and a model which depends on the P/E-ratio’s.

(14)

Market Model (CAPM):

According to (Berk & DeMarzo, 2014) the market model is as follows:

𝑅"#$%&,( = 𝑅+,( + 𝛽.(𝑅0%#123456,( − 𝑅+,() (1)

Where 𝑅"#$%& is the return of the index. 𝑅+ is the risk-free rate and 𝑅0%#123456 is the

return of the benchmark. The subscript t is the time indicator. When this formula is computed in a regression, the following formula will be obtained:

(𝑅"#$%&,(− 𝑅+,() = 𝛼 + 𝛽.(𝑅0%#123456,( − 𝑅+,() + 𝜀",( (2)

As risk free rate, the data from the official Fama & French website8 is used.

P/E model:

The P/E model is made in the following way: 𝐿𝑛 =𝑃 𝐸"#$%&@ ( − 𝐿𝑛 = 𝑃 𝐸"#$%&@(A. = 𝛽B+ 𝛽.(𝐿𝑛 = 𝑃 𝐸C3@(− 𝐿𝑛 = 𝑃 𝐸C3@(A. + 𝜀",( (3) Where 𝐿𝑛 DE

F"#$%&G (is the natural logarithm of the P/E ratio of the index at time t and

𝐿𝑛 DE

FC3G( is the natural logarithm of the P/E ratio of the benchmark at time t. The 𝜀",( is the

error term of index i at time t.

The coefficients of both models are calculated in the estimation period. This estimation period is monthly data between 01/01/2008 and 01/12/2013. With the estimated coefficients the Abnormal Return and Cumulative Abnormal Return could be calculated with daily data. This is done in three different periods. The expectation period which is between 16/10/2013 and 26/11/2013. The event period which is between 27/11/2013 and 07/01/2014 and the post event period which is between 08/01/2014 and 18/02/2014. These periods all consists of 30 trading days.

(15)

The abnormal return (AR) follows from:

𝐴𝑅"#$%&,( = 𝑅"#$%&,( − 𝛽.(𝑅0%#123456,() (4)

The cumulative abnormal return is simply computed by cumulating the abnormal returns in a given interval.

𝐶𝐴𝑅"#$%& = J 𝐴𝑅"#$%&

KL

K.

(5)

Where T1 is the start date of the period and T2 is the end date of the period. The CAR is tested on significance. Three types of effects will be examined on every index. The expectation effect (-45<T<-16) , the event effect(-15<T<15) and the post event effect(16<T<45), where T=0 is the date of the announcement to reduce the amount of Quantitative Easing by the FED on 18 December 2013.

If the residuals of the regressions are normally distributed, a t-test may be used to test the CAR and CAAR on significance.

𝑡NOP =𝐶𝐴𝑅

𝑆NOP (6)

5.3 Regression method

The effect of QE on certain indices is already examined by Georgiadis and Grab (2016). They used the following model in their research:

𝑅",( = 𝛽B,"+ J 𝛽.,"𝑄𝐸𝐴4,( O "S. + J 𝛽L,#𝑆𝑈𝑃𝑅#,( U "S. + 𝜀",( (7)

Where 𝑅",( is the return of the index of interest at time t. The explanatory variable

(QEA) is a set of dummies which equal unity when the date belongs to the day in the interval of the different effects. The control variable in this regression is the Citigroup Economic Surprise Index. Three indexes of CESI are tested in the model (US, Asia-Pacific and China).

(16)

According to the data, the Asia-Pacific index suits best. This index is an objective and quantitative measure of data surprises (Georgiadis & Grab, 2016).

In this research, a control variable is added instead of only the CESI. The control variables which is added is the HSI VIX. Furthermore, more dummies are added to look at different kind of effects. This will lead to the following model:

𝑅",( = 𝛽B,"+ J 𝛽.,"𝑄𝐸𝐴4,( O 4S. + J 𝛽L,"𝑄𝐸𝑃𝐸𝐹EWF,( EWF EWFS. + J 𝛽X,#𝐶𝐸𝑆𝐼(𝐴𝑃)#,( U "S. + J 𝛽Z,#𝑉𝐼𝑋#,( U "S. + 𝜀",( (8)

Where 𝑅"#$%& is the return of the index of interest. QEA and QEPEF are all sets of dummies

which equal unity in different periods, respectively (-15<T<15) and (16<T<45). T=0 equals the announcement date on 18 December 2013. The periods are chosen to look at different effects. QEA is the announcement effect and QEPEF is the effect fifteen days after the announcement. The expectation effect is the effect which occurs when both the QEA and the QEPEF do not equal unity before the event period. The period before T=-15 is the expectation period. SUPR is the Citigroup Economic Surprise Index of the Asia-Pacific region, VIX is the Hang Seng Volatility Index. These extra variable is added to solve (partially) the omitted variable bias. First, this regression included the inflation (CPI) and the three month treasury rate of Hong Kong as well, but these variables added nothing to the model. Therefore, they were deleted. Because of data limitations with the VIX, the start date of the regression is 21/02/2012. The end date is 30/4/2014. This leads to 572 dates which are examined.

The CESI faces some really extreme values, therefore all the growth values of smaller than -2 and bigger than 1 are deleted. The descriptive statistics of the variables are in Table 1.

(17)

Table 1: This table looks at the descriptive statistics of the variables used in the regression. dHSI, dHSNP and

dHSNF are the return of respectively the Hang Seng Index, the Hang Seng Properties and the Hang Seng Finance. The dCESI(AP) is the change in the Citigroup Economic Surprise Index of the Asia-Pacific Region. dVIX equals the change in the Hang Seng Volatility Index.

dHSI dHSNP dHSNF dCESI(AP) dVIX

N 571 571 571 529 572 Mean -0.0000998 0.0000335 0.0000571 -0.0069495 -0.0004887 SD 0.0097187 0.0118099 0.010431 0.2446693 0.0283605 Min -0.0318925 -0.0499627 -0.0377676 -2 -0.1101871 Max 0.0308632 0.0366826 0.0374523 1 0.162963

5.4 Statistical requirements

Normality

To test for normality in the estimation period, the Kolmogorov-Smirnov Test is used. This test checks if the distribution of the variables is normally distributed in the estimation period. The hypothesis of normality is for none distribution rejected when an 5% significance level is used. The results of this test are in the Appendix (Table 1). The test of normality of the regression model are in the Appendix as well (Table 2)

Equality of Distribution in the Event Period

To check whether the distributions in the three event periods are equal to each other, the Kolmogorov-Smirnov Test is used again. The test is done for the Expectation Period with the Event Period, The Expectation Period with the Post-Event Period and the Event Period with the Post-Event Period. Furthermore, it’s done for the HSI, HSNP and the HSNF and the HSIPER, HSNPPER and the HSNFPER. The results of the test were that in every compared period, the distribution were not different. The results of the Kolmogorov-Smirnov test are shown in the Appendix (Table 3 – 18).

Serial Correlation

Serial Correlation is the correlation of residuals over time. This is also one of the assumptions of Gauss-Markov. It needs to be hold that serial correlation is not present in the model,

(18)

otherwise the estimates aren’t BLUE neither. If this is the case, the presented standard errors are smaller than they are in reality. The hypotheses of this test is as follow:

H0: No serial correlation H1: Serial correlation

The test showed that only the regression of the HSNP with the VIX and CESI(AP) of the faces some problems with serial correlation. The results of the serial correlation test could be found in the Appendix (Table 25).

Non-stationary

With the test of non-stationary a possible unit root could be found. This is a random walk which cannot be explained by the data. This unit-root could lead to situations where a high R2 is denoted, but when the data is actually uncorrelated. Furthermore, t-distributions will not perfectly follow the t-distribution, which will lead to implication with testing. This could be tested with the Augmented Dickey Fuller Test (DFuller). If the H0 could be rejected, it means that predicting is possible. In this test, the hypotheses are as follow:

H0: Not stationary H1: Stationary

The test showed that none of the regressions faced problems with serial correlation. In every test, except for the dVIX and dCESI(API) the “trend” option was used because of the upward trend in the data. The results of non-stationary test could be found in the Appendix (Table 26).

6. Results

6.1 Event Study

Market Model (CAPM)

In the estimation period, the quantitative relationship between the benchmarks and the indexes is calculated. This is done by an OLS regression. Three regressions are computed. With the coefficients, the event study could be done. The results of the regressions are stated below in Table 2:

(19)

Table 2: This table shows the regressions in the estimation period between the index of interest and the

benchmark. Three regressions are computed. One for each index of interest. The formula which was used is formula (2). The robust standard errors are posted in parentheses. *P<0.2, **P<0.1, ***P<0.05

Dependent variable:

HSI Dependent variable: HSNF Dependent variable: HSNP

(1) (2) (3)

MSCI ACWI 0.9193***

(0.1098) - -

MSCI World Financials - 0.5743***

(0.1309) -

S&P Real Estate Index - - 0.6735***

(0.203)

N 65 65 55

R2 0.6816 0.3766 0.2785

Root MSE 0.04263 0.06601 0.06711

In the table below (3), the results of the Event Study based on the Market Model are presented. It’s clear that not many CAR are statistically significant in other periods than the event period. Furthermore, the 1% significance only holds in the event period. According to the market model, there is not an effect in the Post-Event Period. In the Expectation Period, there occurred only an effect at the HSI. The sub-indexes interval. The only period where a strong negative effect occurred because of the announcement was in the event period. All the CAR are at least statistically significant at 5% or lower.

Table 3: This table shows the results of the Event Study based on the Market Model. The values of the confidence

interval are the values of the Abnormal Return, so on a 1 day interval. Because measuring the Confidence interval of the CAR.

MacKinley (1997) explains another way to test the CAR. He argues that the variance of the CAR is equal to σ^_`L = (t2 − t1 + 1) ∗ σeL, where t1 and t2 are the dates where a certain period started and ended. σ

e

L is the variance of the error term of the regression. This way is of calculating is done in the Appendix.

*P<0.1, **P<0.05

Index: HSI Index: HSNF Index: HSNP

Expectation

period Event period Post-Event period

Expectation

period Event period Post-Event period

Expectation

period Event period Post-Event period CAR -3.37% -4.48% -1.26% -0.32% -9.19% -0.08% -0.82% -9.05% -3.18% SD(CAR) 0.015 0.019 0.014 0.018 0.031 0.017 0.016 0.024 0.029 T-Score -2.2249* -2.389* -0.889 -0.1812 -2.937** -0.046 -0.512 -3.757** -1.1 CI Lower 95% (AR) -0.0236% -0.045% -0.037% -0.4389% -0.221 -0.289% -0.019% -0.024% -0.068% CI Upper 95% (AR) 0.0363% 0.004% 0.045% 0.0166% -0.181 0.311% 0.021% -0.013% 0.029%

(20)

P/E Ratio(Model)

The same procedure as with the market model needs to be done with the P/E Ratio model. Again, three regression are computed with OLS-technique. The outcomes are stated in Table 4.

Table 4: This table shows the regressions in the estimation period between the index of interest and the

benchmark. Three regressions are computed. One for each index of interest. The formula which was used is formula (3). The robust standard errors are posted in parentheses. *P<0.2, **P<0.1, ***P<0.05

Dependent variable:

HSI Dependent variable: HSNF Dependent variable: HSNP

(1) (2) (3)

HSI P/E Ratio 0.9166***

(0.0444) - - HSNF P/E Ratio - 0.9857*** (0.01892) - HSNP P/E Ratio - - 1*** (0.0095) N 68 67 60 R-Sq 0.9249 0.9722 0.9924 Root MSE 0.02075 0.01418 0.00723

The results of the event study with the P/E-Ratio model are presented in Table 5. Instead of the market model, this model argues that an Expectation effect definitely occurred. However, this is partially (except for HSNF) due to the very low standard deviations of the CAR. The HSI, HSNP did not have a large deviation from their normal return, but because of those extreme small standard deviations, the results are all significant for at least 5%. An explanation for these low standard deviations could be the extreme high R-squared in the regressions. This is due to the high positive relationship between stocks and P/E-ratios.

The effect of the announcement had also a large effect. This is similar to the market model. A probable explanation for the insignificance of the HSNF in the Event Period could be that the HSNF Index already made a loss before the announcement. Therefore, the loss due to the announcement of QE could be less strong. Similar to the market model is the Post-Event Period. This time series does not show a strong effect after the announcement, which is the same as with the CAPM. Only the HSNP shows an effect, which is the strongest effect of the HSNP in all the time periods.

(21)

Table 5: This table shows the results of the Event Study based on the P/E-Model. The values of the confidence

interval are the values of the Abnormal Return, so on a 1 day interval. Because measuring the Confidence interval of the CAR is not possible.

MacKinley (1997) explains another way to test the CAR. He argues that the variance of the CAR is equal to σ^_`L = (t2 − t1 + 1) ∗ σeL, where t1 and t2 are the dates where a certain period started and ended. σeL is the variance of the error term of the regression. This way of calculating is showed in the Appendix. *P<0.1, **P<0.05.

Index: HSI Index: HSNF Index: HSNP

Expectation

period Event period Post-Event period

Expectation

period Event period Post-Event period

Expectation

period Event period Post-Event period CAR 0.24% -0.49% 0.19% -6.65% -0.58% 0.31% 0.09% -0.17% -0.66% SD(CAR) 0.001 0.002 0.003 0.0213 0.0059 0.007 0.0003 0.0008 0.003 T-Score 1.7788* -2.43** 0.742 -3.1221** -0.97 0.453 2.5776** -2.19* -2.361* CI Lower 95% (AR) -0.03% -0.38% -0.31% -3.4% -0.55% 1.16% 0.04% -0.21% -0.49% CI Upper 95% (AR) 0.07% -0.23% -0.12% -1.88% -0.13% 1.65% 0.07% -0.16% -0.29%

Graphs with the developments of the CAR in different periods over the time for each index are in the Appendix (Graph 1-6).

6.2 Regression Analysis

With the regression method, the quantitative effect of the announcement is examined. The explanation of this regression model are stated in the Methodology part. The results of the three regressions are in the table below (6)

Table 6: This table shows the results of the regression analysis. The constant equals the period known as the

“expectations period”. The standard-errors of the coefficients are in parenthesis. 3 regressions are computed. All had the same explanatory and control variables. Only the dependent variable differed. The return of one the three indexes of interest is used as dependent variable in each regression. *P<0.20, **P<0.10, ***P<0.05.

Dependent variable:

Return HSI Dependent variable: Return HSNF Dependent variable: Return HSNP

Constant 0.0000903 (0.000367) 0.0000828 (0.0003939) 0.000588 (0.0004748) Event effect (dummy) -0.0012814 (0.0015773) -0.0019364 (0.0016722) -0.0019439 (0.0020156) Post-Event effect (dummy) 0.0000137 (0.0012444) 0.000032 (0.0016832) 0.0002405 (0.0020409) dVIX -0.18304*** (0.0136335) -0.19027** (0.0131686) -0.17228** (0.0158728) dCESI(AP) 0.002045*** (0.0000509) 0.000247*** (0.000145) 0.0003116*** (0.0001748) N 571 571 571 Adjusted R-Sq 0.2927 0.272 0.1749 Root MSE 0.0082 0.0089 0.01073

(22)

Although the models are not really strong, the outcome can tell something about the possible effect of the quantitative easing announcement on the three different indexes. The outcome of the regressions is that no significant effect occurred at each index in each period. This is contrary to the Event Study, which stated that there was a negative effect in the event period. In this regression method, the coefficient in that specific period are indeed negative, however the results aren’t significant. The coefficient that are (really) significant, are the change in the VIX (dVIX) and the change in the CESI (dCESIAP). The VIX has a negative relationship to the return, which is strange because it is against the theory of “more risk results in more return”(Berk & DeMarzo, 2014).

7. Conclusion

This paper examined the different effects of the announcement by the FED to reduce the amount of Quantitative Easing on the Hang Seng Index (HSI) and its sub-indexes. The effects which are measured are the expectation effect, the event effect and the post-event effect. This is done by two different approaches. First, the possible effects are examined by an event study and second by a regression based on the paper of Georgiadis and Grab (2016).

The event study showed that when making use of the market model (CAPM), an expectation effect occurred for the HSI. Furthermore, the announcement had significant effects on all examined indexes in the event period. Some indexes even got a cumulative abnormal return of more than -9%. No significant effect appeared in the period fifteen days after the announcement.

According to the event study which made use of the P/E-ratio, an expectation effect definitely occurred for all the indexes. The HSNF even got a cumulative abnormal return of -6.65%. Additionally, an event effect happened. Nevertheless, those effects were less strong than with the market model. In the period after the effect, a significant negative return only occurred in the Property Index.

The regression method showed that no significant effect happened during the Event Period, nor in the period around it. This is in contrast with the Event Study. However, the signs of the coefficients are similar to that of the market-model. In the Event Period, these signs are all negative. However, these signs are not significant.

(23)

The announcement had a negative effect on those indexes, yet the size of these effects could not be determined from these results. The quantitative results between the Market-Model and the P/E-ratio model vary too much in either CAR and standard deviation. The regression showed that there was not an effect in any of the periods. Therefore, according to the regression analysis, insufficient evidence is available to conclude that the announcement to reduce the amount of QE affects the Hang Seng Index and its Sub-Indexes. This could be partially explained by the fact that QE has stronger effects in the region of its implementation than it has on other countries. Since, the Event Studies showed a negative effect while the regression analysis did not, The outcome of the research clearly depends on the method used. Despite the fact that some results are significant, this study faces some limitations. Firstly, only one announcement is taken into consideration. This is due to the fact that the FED only made one official announcement to reduce the amount of QE. Additionally, the regression does not include purchase volumes. In further research, this variable could be added to examine the direct influence of the QE policy on stock markets. Next to that, other announcements and implementation dates could be examined as well.

Furthermore, due to time constraints only indexes in Hong Kong are examined. It would be beneficial to look at the effect of QE on all the major stock exchanges in the world. Additionally, more sub-indexes could be taken into consideration. This research only focusses on the general index, the real estate index and the finance index. However, some other indexes may show a significant effect on Quantitative Easing.

Besides that, by making use of the market-model (CAPM) and the P/E-model, this study assumed that these models unequivocally hold. Nevertheless, these models sometimes failed. Because of the inconsistency of these models, a regression method was added to correct for multiple effects. This increased the validity of the results of this study. However, the results of these regressions are contrary to the Event Study. Therefore, future research on this topic should be conducted using regression-analysis instead of an Event Study.

(24)

8. References

Agrawal, J., & Kamakura, W. (1995). The Economic Worth of Celebrity Endorsers: An Event Study Analysis. Journal of Marketing, 56-62.

Austin, D. (1993). An Event-Study Approach to Measuring Innovative Output: The Case of Biotechnology. The American Economic Review, 253-258.

Bauer, M. D., & Rudebusch, G. D. (2014). The Signalling Channel of Federal Reserve Bond Purchases. Federal Reserve Bank of San Fransisco, 233-289.

Berk, J., & DeMarzo, P. (2014). Corporate Finance. Essex: Pearson Education.

Binder, & J.J. (1985). On the Use of Multivariate Regression Model in Event Studies. Journal

of Accounting Research, 370-383.

Binder, J. (1985). Measuring the Effects of Regulation with Stock Price Data. The RAND

Journal of Economics, 167-183.

Blinder, A. S. (2010). How Central Should the Central Bank Be? Journal of Economic

Literature, 123-133.

Christensen, J. H., & Gillan, J. M. (2018). Does Quantitative Easing Affect Market Liquidity.

Federal Reserve Bank of San Francisco Working Paper, 1-11.

De Bondt, G. J. (2015). Confidence and Monetary Policy Transmission. European Central

Bank, 2-25.

Fama, E. F. (1970). Efficient Capital Markets: a Review of Theory and Empirical Work. Journal

of Finance, 383-417.

Fawley, B. W., & Neely, C. J. (2013). Four Stories of Quantitative Easing. Federal Reserve

Bank of St. Louis REVIEW, 51-88.

Gagnon, J., Raskin, M., Remache, J., & Sack, B. (2011). The Financial Market Effects of the Federal Reserve's Large-Scale Asset Purchases. International Journal of Central

Banking, 3-42.

Georigadis, G., & Grab, J. (2016). Global Financial Market Impact of the Announcement of the ECB's Asset Purchase Programme. Journal of Financial Stability, 257-265.

Hashem Pesaran, M., & Smith, R. (2016). Counterfactual analysis in macroeconometrics: An empirical investigation into the effects of quantitative easing. Research in Economics, 262-280.

Ivashina, V., & Scharfstein, D. (2010). Bank lending during the financial crisis of 2008. Journal

(25)

Jonathan, B., & Thomas, R. (2012). The impact of QE on the UK economy - some supportive monetarist arithmetic. Bank of England Working Paper, 3-48.

Joyce, M., & Lasaosa, A. (2011). The Financial Market Impact of Quantitative Easing in the United Kingdom. Journal of Central Banking, 113-161.

Joyce, M., Miles, D., Scott, A., & Vayanos, D. (2012). Quantitative Easing and Unconventional Monetary Policy - an Introduction. The Economic Journal, F271-F288.

Joyce, M., Tong, M., & Woods, R. (2011). The United Kingdom's Quantitative Easing Policy: Design, Operation and Impact. Bank of England Quarterly Bulletin, 200-212.

Krishnamurthy, A., & Vissing-Jorgensen, A. (2011). The Effects of Quantitative Easing on Interest Rates: Channels and Implications for Policy. National Bureau of Economic

Research, 2-46.

Kurihara, Y. (2006). The Relationship between Exchange Rate and Stock Prices during Quantitative Easing Policy in Japan. International Journal of Business, 375-386. MacKinlay, A. C. (1997). Event Studies in Economics and Finance. Journal of Economic

Literature, 13-39.

Malkiel, B. G. (2003). The Efficient Market Hypothesis and Its Critics. Journal of Economic

Perspectives, 59-82.

Matthews, K., Giuliodori, M., & Mishkin, F. S. (2013). The Economics of Money, Banking &

Financial Markets. Toronto: Pearson Education Limited.

Meaning, J., & Zhu, F. (2011). The impact of recent central bank asset purchase programs.

BIS Quarterly Review, 73-83.

Ugai, H. (2006). Effects of the Quantitative Easing Policy: A Survey of Empirical Analyses .

(26)

9. Appendix

Statistical tests & tables

Table 7 – Test of Normality with variables used in the Event Study

(27)

Tables 9 – 24 Test of Equality of Distribution between different periods

PeriodX PeriodY Variable Similar

Expectation Event HSIPER Yes

Expectation Event HSNPPER Yes

(28)

PeriodX PeriodY Variable Similar

Expectation Event HSI Yes

Expectation Event HSNP Yes

(29)

PeriodX PeriodY Variable Similar

Event Post-Event HSIPER Yes

Event Post-Event HSNPPER Yes

(30)

PeriodX PeriodY Variable Similar

Event Post-Event HSI Yes

Event Post-Event HSNP Yes

(31)

PeriodX PeriodY Variable Similar

Expectation Post-Event HSIPER Yes

Expectation Post-Event HSNPPER Yes

(32)

PeriodX PeriodY Variable Similar

Expectation Post-Event HSI Yes

Expectation Post-Event HSNP Yes

(33)

Dependent variable

Independent variable

Chi^2 P-value Serial

Correlation*

HSI HSIPER 3.447 0.0634 No

HSNP HSNPPER 1.119 0.2901 No

HSNF HSNFPER 1.705 0.1916 No

HSI MSCI ACWI 1.995 0.1578 No

HSNP S&P Real Estate 0.001 0.9810 No

HSNF MSCI World

Financials

1.739 0.1872 No

HSI VIX, CESI(AP) 0.002 0.9650 No

HSNP VIX, CESI(AP) 6.892 0.0087 Yes

HSNF VIX, CESI(AP) 0.000 0.9971 No

Table 25 – Test on Serial Correlation (*= based on 5% significance)

Variable Dfuller Statistic MacKinnon P-value Not stationary*

HSI(Monthly) -6.229 0.000 No HSIPER -5.940 0.000 No HSNP(Monthly) -7.683 0.000 No HSNPPER -7.671 0.000 No HSNF(Monthly) -5.881 0.000 No HSNFPER -6.317 0.000 No MSCI ACWI -4.856 0.0004 No

S&P Real Estate -6.930 0.000 No

MSCI Financials -5.043 0.0002 No HSI(Daily) -13.086 0.000 No HSNP(Daily) -11.966 0.000 No HSNF(Daily) -13.232 0.000 No dVIX -14.024 0.000 No dCESI(AP) -15.889 0.000 No

(34)

Graphs

Graph 1, the CAR of different indexes in the Expectation Period, based on the Market Model.

Graph 2, the CAR of different indexes in the Event Period, based on the Market Model.

-5.0000% -4.0000% -3.0000% -2.0000% -1.0000% 0.0000% 1.0000% 2.0000% 3.0000% 4.0000% -50 -45 -40 -35 -30 -25 -20 -15 -10 -5 0 Cu m ul at iv e Ab no rm al R et ur n Time

Expectation Effect

Average HSI HSNP HSNF -10.0000% -8.0000% -6.0000% -4.0000% -2.0000% 0.0000% 2.0000% 4.0000% -20 -15 -10 -5 0 5 10 15 20

Event Period

Average HSI HSNP HSNF

(35)

Graph 3, the CAR of different indexes in the Post-Event Period, based on the Market Model.

Graph 4, the CAR of different indexes in the Expectation Period, based on the P/E Model.

-8.0000% -7.0000% -6.0000% -5.0000% -4.0000% -3.0000% -2.0000% -1.0000% 0.0000% 0 5 10 15 20 25 30 35

Post-Event Period

Average HSI HSNP HSNF -7.0000% -6.0000% -5.0000% -4.0000% -3.0000% -2.0000% -1.0000% 0.0000% 1.0000% 2.0000% -35 -30 -25 -20 -15 -10 -5 0

Expectation Period

Average HSI HSNP HSNF

(36)

Graph 5, the CAR of different indexes in the Event Period, based on the P/E Model.

Graph 6, the CAR of different indexes in the Post-Event Period, based on the P/E Model.

-2.5000% -2.0000% -1.5000% -1.0000% -0.5000% 0.0000% 0.5000% 1.0000% 1.5000% -20 -15 -10 -5 0 5 10 15 20

Event Period

Average HSI HSNP HSNF -2.0000% -1.5000% -1.0000% -0.5000% 0.0000% 0.5000% 1.0000% 1.5000% 0 5 10 15 20 25 30 35

Post-Event Period

Average HSI HSNP HSNF

(37)

MacKinley T-Statistics

P/E-Model Hang Seng Index Hang Seng Finance Hang Seng Properties

Expectatio n Effect Event Effect Post-Event Effect Expectation Effect Event Effect Post-Event Effect Expectation Effect Event Effect Post-Event Effect T-Score -0.54 -1.34 -1.01 -0.95 0.26 -0.19 -0.47 -0.48 -0.66 *P<0.10 **P<0.05 Market

Model Hang Seng Index Hang Seng Finance Hang Seng Properties

Expectatio

n Effect Event Effect Post-Event Effect

Expectation

Effect Event Effect Post-Event Effect

Expectation

Effect Event Effect Post-Event Effect T-Score -0.21 -0.79 -0.41 -0.55 -1.95* -0.95 -0.47 -1.82* -1.38

*P<0.10 **P<0.05

Referenties

GERELATEERDE DOCUMENTEN

De hoeveelheid koolstof in de lichte en zware Ludox-fracties en in de fracties tussen 20 en 150 urn en &lt; 20 urn in het voorjaar in de bouwvoor van de percelen met gangbare

Closing Material Loops at various scales Reuse of resources – materials, energy, and waste Circular City Spatial Planning &amp; Neighbourhood Development Urban

[r]

We additionally expected the above described effects to be most pronounced in patients with epileptic focus or surgery in the speech-dominant hemisphere, since our primary

Data from cross-sectional studies and RCTs were analysed separately. For all quantitative analyses, data were evaluated for the proportion of patients with at least one overall

(1.) The blue arc indicates the stator current at rated load that reflects a constant apparent power where the real and reactive power can vary according to

For women, we found that this indirect effect of gender identification was also contingent upon the gender ratio (i.e., numerical dominance) in the direct work environment. That

En materia económica, los invernaderos de alta tecnología reportan mayores rendimientos en los cultivos hortícolas, alta eficiencia en el uso de insumos, baja demanda de agua, así