The effect of macroeconomic announcements on stock
market volatility
Klaas Bonnema
∗January 14, 2016
Abstract
This paper investigates the effects of macroeconomic announcements from sev-eral countries on volatilities of sixteen stock indices around the world. The United States, Europe and China are the selected cases that will be examined in an EGARCH(1,1) model. We have taken the unexpected news component, which is equal to the expec-tation minus the actual value, as proxy for the announcements. We find significant effects for the announcements producer manager index and unemployment rate of the U.S. and consumer confidence indicatore and gross domestic product of Europe. The other results indicate a low response in volatility for most ‘news’ variables re-leased by official parties. It implies that investors are relatively resistant against surprising news.
Keywords: Stock indices, volatility, macroeconomic announcements JEL classification: E44, G14, G15
∗Studentnumber: 2032023. Supervisor: S.Drijver. I would like to thank Dr. S. Drijver for useful
1
Introduction
The volatility of stock markets is well-known to investors who are active in trading stocks. Investors cause these volatilities by showing rational and behavioral investing. Research points out that stockholders are sensitive to news and adapt their behavior to such an-nouncements. This news can be available on micro-level, like news about a specific firm on the internet or in the newspaper, but we can also take it to a macro-level, such as macroeconomic announcements of the central banks or governments. This news can re-late to all kind of forecasts, for example economic growth, consumer prices or about the employment situation. In a globalizing world it is of interest if announcements in, for example, the United States also influence volatilities of stock markets in other parts of the world. Therefore it raises the question: How large is this impact of different types of economic news on the volatility of stock markets around the globe? This paper will try to answer this question by analyzing several cases, namely the United States, Europe and China.
Why is in particular volatility of interest for an investor? First of all, volatility can be described by the uncertainty of investors about the value of a stock. A period of high volatility indicates that the investor expects the value of a stock can increase or decrease over a wider range of values compared to periods with low volatility. High volatilites could lead to positive opportunities for investors who want to invest on the short term. They speculate on a decrease of an index by selling short in stocks1. On the other hand, investors focused on the long term try to minimalize this volatility effect as much as possi-ble. Their objective is different since they posses the security all along, and thus a steady increase in the index is required, which is mostly accompanied by low volatility. This can be explained by the fact that low volatility represents low uncertainty about the current situation. The availability of macroeconomic news is connected with volatility as it may increase uncertainty during a release.
Since we are looking for an effect on the stock markets in different parts of the world, it is important to approach the availability of macroeconomic announcements as an investor. How would an investor, who wants to make a return on his investment, interpret released macro news. It is clear, that macroeconomic announcements are a forecast of the trend of the current economic situation of a country. If investors respond to this, it means they expect that, even though it happens on the other side of the world, it will impact their investments. Capturing this possible effect will lead us to a conclusion regarding the research question.
1Short selling means that an investor borrows a security and sells it for the current price. In order
From the existing literature, we can observe that most of the papers are focused on the United States and its effects on other stock markets. Reasons for this could be that the United States is up till now the largest economy and has the most important cur-rency around the globe, namely the dollar. Europe can be seen as a large economy and an important player in the world and thus has possible effects on stock markets in the world. On the other hand, China has been one of the largest upcoming economies the past decade and it is already the second largest economy in the world in terms of Gross Domestic Product (GDP)2. Therefore we want to broaden our scope by giving a more comprehensive look on other macroeconomic news releases from Europe and China in addition to the ones from the United States.
Previewing our paper, we conducted research on volatility present in sixteen different stock markets around the world. The sample length of our data varies, the announce-ments of the United States and Europe are regressed on a time period of September 2008 till October 2015. Next to that, China’s sample lasts from May 2011 till October 2015. The results are estimated by an Exponential Generalized AutoRegressive Conditonally Heteroskedastic (EGARCH) model inspired by Nelson (1991). It shows that for news of the United States, the Producer Manager Index (PMI) has a positive impact on some parts in the world. The European announcements Consumer Confidence Indicator (CCI) and GDP showed significant impacts in North and South America, with some exceptions. The CCI showed a positive impact, where the GDP news had a negative effect. For Chinese announcements, we could not draw any meaningfull conclusions with respect to the volatility of the stock markets. As robustness check we switched the stock indices for volatility indices of some of the used markets. Some results are in line with our findings, but we also found some deviations.
The paper will proceed as follow; first we will discuss the existing literature about the effects of macroeconomic news on stock markets; next the data specifications, followed by the methodology; the results and lastly the conclusions. In addition it gives limitations and recommendations for further research.
2
Literature Review
The reaction of stock markets on announcements from the United states (Flannery and Protopapadakis (2002), Hanousek, Kocenda and Kutan (2009), Hardouvelis (1987), Kim, McKenzie and Faff (2004), Nikkinen and Sahlstr¨om (2004), Nikkinen et al. (2006), An-dersen et al (2004), McQueen and Roley (1993)), Europe (Federova, Wallenius and Collan (2014)) and China (Baum, Kurov and Wolfe (2015)) is confirmed by a large number of
papers. This is a broad conclusion, since announcements can be selected and specified in many ways. Next to that, the selected models and use of other variables can give different estimates. Therefore we search for a consensus in the literature regarding these findings3.
2.1
Macroeconomic announcements released by the United States
The effects of macro news from the United States on other stock markets are thoroughly researched in this area. A reason for this may be the leading role of the American econ-omy in the world. In addition the availability and well-structured data contributes to the numerous papers on this topic.
Flannery and Protopapadakis (2002) investigate the American equity markets with an-nouncements of the country itself. They find that six of their seventeen macro vari-ables have an impact on the volatility of the value-weighted NYSE-AMEX-NASDAQ index4, which are consumer price index, producer price index, balance of trade, employ-ment/unemployment, housing starts and monetary aggregate. Kim et al. (2004) confirm the impact of the announcements consumer price index and producer price index on volatility by taking the Dow Jones index as stock market. Additionally, they find that retail sales growth and the unemployment rate have an effect on volatility. Hardouvelis (1987) also researched the American stock market, among which are the New York Stock Exchange and Standard & Poor 500, with respect to macroeconomic announcements of the United States. His results indicate that stock markets mainly react to monetary news, where companies involved in finance respond the strongest. Furthermore, only the nonmonetary variables trade deficit, the unemployment rate and personal income, show a response in the market. The reliability of this results, however, is questionable, since the estimation method5 is not convincing if one takes the stylized facts of a stock market into account. Moreover, the estimates are a direct impact instead of volatility implications.
Andersen et al. (2004) consider future markets based on U.S. equity markets as a rep-resentative for price adjustment due to economic news. They find negative results for a direct impact of the variables Consumer Price Index, Producer Price Index, Nonfarm Payroll Employment, New Home Sales and Net Exports. However, when they make a distinction between “good” and “bad” news and seperate the sample in expansion and re-cession periods, the number of significant announcements goes up and they find opposing 3In the existing literature there are also many papers specifically investigating the relation between
monetary policy announcements and the response of stock markets. Since this is beside the main question of this paper, this will not be covered.
4NYSE, AMEX and NASDAQ stand for New York Stock Exchange, American Stock Exchange and
National Association of Securities Dealers Automated Quotations respectively
5Hardouvelis (1987) regresses the announcements on the stock markets by means of an Ordinary Least
signs in the two periods. Expansion periods lead to negative signs, whereas recession peri-ods have postive coefficients. Similar to this type of research, McQueen and Roley(1993) investigate the Standard & Poor 500 for a couple of macroeconmic announcements consid-ering different states of the economy. They find that the variables industrial production, unemployment, producer price index and the money supply have a negative direct impact on the returns in the high state of the economy, which implies according to McQueen and Roley (1993) that “good” news has a negative influence. Kim et al. (2004) also make a distinction between “good” and “bad” news in their empirical research. They find a difference in signs for the producer price index and the consumer price index by seperating the interpretation of news. In their case positive news leads to more volatility and negative announcements to less.
Nikkinen and Sahlstr¨om (2004) specifically research the response of European markets to U.S. announcements and compare this with domestic news. In their paper they make use of implied volatility rather than realized volatility. A clear distinction is made here by Nikkinen and Sahlstr¨om (2004), where implied volatility stands for expectations of future volatility and realized volatility for actual market reactions. Their findings indi-cate a significant impact of U.S. announcements on the European markets, even though the markets in the sample differ in size and foreign ownership. The interest rate and employment report are the variables of impact. Additionally, Hanousek et al. (2009) investigate emerging European markets, like Poland, Hungary and the Czech Republic, with respect to American news. Their research is focussed on markets with a different degree of foreign ownership of stocks in order to see if there is a different effect of news. Their findings indicate that a high degree of foreign investors is more prone to macroeco-nomic announcements from the United States. In addition they point out that the effect is larger for multiple news compared to a single announcement.
De paper of Nikkinen et al. (2006) is one of the few that focuses on global stock market reactions. Where other researchers investigate specific markets or areas, Nikkinen et al. (2006) take 35 different countries spread around the world as their sample. The argument for research on global stock markets is to see if these markets show signs of worldwide in-tegration. Nikkinen et al. (2006) convert the seperate volatility estimates of each country to regional samples, from which they calculate the effect on six compiled regions in the world6. Especially in the G7 region, significant positive effects on volatility are found for the announcements consumer price index, employment cost index, employment situation and NAPM, which stands for National Association of Purchasing Management.
6For mathematical details on the calculations we refer you to the paper of Nikkinen et al. (2006). The
2.2
Macroeconomic announcements released by Europe
The positive impact on volatility in the stock market is confirmed by Federova et al. (2014). On the other hand, some papers find no result at all, like Nikkinen and Sahlstr¨om (2004). A note should be made on the definition European news, since Europe exists of multiple countries and therefore announcements can be viewed on country level or European level. Here is also a discrepancy in the existing literature, where writers use different countries as proxy for European macroeconomic announcements. Further on, for each paper the macro news will be defined.
Federova et al. (2014) research the effect of European macro news on the following stock markets, namely Colombia, Indonesia, Vietnam, Egypt, Turkey and South Africa. The macro news considers the whole European Union, where they make use of the announce-ments like consumer price index, gross domestic product and retail sales. Following the EGARCH model proposed by Nelson (1991), their findings confirm the hypothesized pos-itive effect on the volatility in the market. In addition they made a distinction between positive and negative news, where negative news has a large impact on the returns.
As before mentioned, Nikkinen and Sahlstr¨om (2004) have researched the effects of do-mestic and American announcements on European markets. The results indicate that European news has no effect on the uncertainty in the domestic markets. The announce-ments of consumer price index, employment report and producer price index are in this case defined as local news.
2.3
Macroeconomic announcements released by China
The literature about the effects of macroeconomic news from China is not covered ex-tensively in the existing literature. Reasons for this are, for example, the fact that the booming economy of China is recent, and therefore less covered by researchers. Another issue is the reliability of the data coming from China , which will be explained in more detail in the data section later on.
of the world economy.
2.4
Discussion
The main question left for this section, is what the existing literature can tell us about the expectations regarding our research topic. As we have seen, many papers have shown significant results of the impact of macro announcements on the volatility of stock mar-kets. Important drawbacks are the differences in the selected macroeconomic variables, and thus the comparability of the literature. Some select a few variables, where others in-clude almost every variable available. However, we can see a significant trend in the most important variables, which turn out to have positive relations with the volatility in the stock market. Therefore we can argue that the effect of stock markets will be non-zero, which will be specified in our hypthesis later on. What sign this impact will have is not clear from the literature. In general, one could say that an unexpected deviation in an announcement leads to increased uncertainty, thus a positive sign. However, we should keep in mind that the announcements can be good or bad for the economy, which can give complications for the estimated coefficients.
The methodology of the previous mentioned literature shows that there exists a twospalt. One half investigates the direct impact by means of an event study method using high frequency data and a rather short time span7, where the other half uses volatility mod-els of Engle (1982), Bollerslev(1986) and Nmod-elson(1991). In the latter, reasearch can be conducted on the effects on volatility rather than the immidiate response of asset prices. Therefore our preference and starting point will be volatility models. Taking a more closer look on the use of these models, we can observe that Flannery and Protopapadakis (2002), Kim et al. (2004), Hanousek et al. (2009), Nikkinen and Sahlstr¨om (2004) and Nikki-nen et al. (2006) make use of Generalized Autoregressive Conditionally Heteroskedastic (GARCH) models. Federova et al. (2014) are an exception as they make use of an EGARCH model. From a theoretical perspective, the EGARCH model is closer to re-ality as it includes an asymmetry term, which implies that negative news of the same magnitude as positive can have a different impact. This reasoning started out by the paper of Black (1976), who found asymmetries in the stock market returns. Therefore an EGARCH model would be theoretically a more realistic approach. Later on, we will justify the choiche of our model by statistical tests.
To summarize the literature section, we have seen that there has been plenty of research on the subject macroeconomic announcements and their effects on stock markets. The documented effects vary from local to global reactions. Next to this, most of the research 7In many papers, the time span is chosen from 5 minutes before an announcement up to 10 minutes
has been done on American announcements, and on a smaller scale European and Chi-nese news. This paper will extend the literature in a few ways. First, it will analyze all three cases on global scale and make a comparison with respect to worldwide integra-tion and usefullness for investors. Second for Chinese announcements, it will focus on volatility, which is not covered before. Lastly, we will employ an EGARCH model for our estimations, where most researchers in the literature hold on to the GARCH model of Bollersev(1986).
3
Data
The data is divided into three samples, as we can make a distinction between the United States, Europe and China. The time span of the United States and Europe is seven years, from the 29th of September 2008 till the 20th of October 2015. The sample of China is a bit shorter, namely from the 15th of March 2011 till the 20th of October 2015. The reason for this is the shortage of data for the announcements of China. In the next subsections we will discuss the variables used in detail.
3.1
Stock markets
Table 1: Stock Markets
Region Country Stock index Trading Hours (GMT)
North America United States New York Stock Exchange composite 14:30 - 21:00 Dow Jones Industrial Average 14:30 - 21:00
SP 500 Index 14:30 - 21:00
Canada SP/TSX 60 composite 14:30 - 21:00
South America Brazil Brazil Bovespa Stock Index 13:00 - 20:00
Chile Santiago Index general 12:30 - 20:00
Mexico IPC 14:30 - 21:00
Europe United Kingdom FTSE all share 08:00 - 16:30
France CAC 40 08:00 - 16:30
Germany DAX 08:00 - 16:30
Switzerland Swiss Market Index 08:00 - 16:30
Pacific Australia Australia ASX 200 00:00 - 06:00
Asia China Shanghai SE Composite Index 01:30 - 07:00
Hong Kong Hang Seng 01:30 - 08:00
Japan Nikkei 225 00:00 - 06:00
Taiwan Taiwan TSEC 50 Index 01:00 - 05:30
Notes: GMT stands for Greenwich Mean Time.
The sample consists of daily data obtained from the database DataStream. For the regression we used daily returns calculated as log linearized first differences of the closing value of each day, which is displayed in the following formula8:
ri,t = 100 ∗ (log(pricei,t) − log(pricei,t−1)) (1)
Stock markets are characterized by several stylized facts which are concluded from previ-ous empirical research. This is an important factor to take into account before we start estimating the variables. According to Brooks (2008) financial time series deal with excess kurtosis and to some extent skewness. Additionally, it is said that there is a so-called leverage effect present in financial data, which implies that volatility increases more when bad news of the same magnitude arrives compared to positive news. This fact is com-parable to the findings of Black (1976) and Nelson (1991). At last, volatility clustering is visible if one analyzes a sample of, for example in our case, stock indices. This means that large returns follow on large returns and small returns follow small returns. First of all, we will have a look at figure 1, which depicts the daily returns of the Standard & 8The Augmented Dickey-Fuller test for the stock markets indicated that all markets had a unit root in
Poor’s 500 index from September 2008 till October 2015.
Figure 1: Daily returns S&P 500
As we can see in figure 1, volatility is most of the time clustered, which means that volatil-ity shows periods of high volatilvolatil-ity followed by periods of low volatilvolatil-ity. The first period in the graph, roughly from 2008 till halfway 2009 visualizes the high uncertainty during the financial crisis. After that, we can observe some smaller periods of high volatility which are related to the aftermath of the crisis, and other types of crises, like the sovereignty problems in Europe.
Table 2: Descriptive statistics stock markets
Stock market Mean Median Maximum Minimum Std. Dev. Skewness Kurtosis New York Stock Exchange 0.01 0.03 11.53 -10.23 1.46 -0.41 13.25 Dow Jones Industrial Average 0.02 0.03 10.51 -8.20 1.25 -0.04 14.44
SP 500 Index 0.03 0.04 10.96 -9.47 1.38 -0.33 14.15
SP/TSX 60 composite 0.01 0.04 9.37 -9.79 1.24 -0.78 14.63 Brazil Bovespa Stock Index -0.00 0.00 13.68 -12.10 1.74 0.10 11.18 Santiago Index general 0.02 0.00 11.80 -7.19 1.02 0.24 18.28
IPC 0.03 0.01 10.44 -7.27 1.25 0.24 12.35
FTSE 100 0.01 0.01 9.38 -9.27 1.26 -0.28 11.66
CAC 40 0.01 0.03 10.59 -9.47 1.57 0.01 8.73
DAX 0.03 0.06 10.80 -7.34 1.52 0.10 8.92
Swiss Market Index 0.01 0.02 10.79 -9.07 1.20 -0.28 13.31
Australia ASX 200 0.00 0.01 5.63 -8.70 1.13 -0.55 8.24
Shanghai SE Composite Index 0.02 0.00 7.01 -8.87 1.56 -0.69 7.37
Hang Seng 0.01 0.00 13.41 -13.58 1.58 0.02 14.02
Nikkei 225 0.02 0.00 13.23 -12.11 1.64 -0.57 12.40
When taking a look at the descriptive statistics in table 2, we can conclude that each stock market has signs of skewness and excess kurtosis. The excess kurtosis is defined as leptokurtosis by Brooks (2008), which implies fat tails and excess peakedness at the mean. The consequence of leptokurtosis is that the distribution of the sample is not normal. This non-normality can be statistically concluded by interpreting the Jarque-Bera statistic9 for each stock index (Jarque and Bera (1987)). In all cases we can reject the null-hypothesis of a normal distribution. We observe no severe outliers in table 2, which have to be removed from the sample due to odd behaviour. In most cases the maxima and minima are inherent to the highly volatile periods. However, one could argue that the first period is too volatilite compared to other periods of higher volatility. Therefore we will analyze in the result section samples excluding this period. Moreover, we executed the test for “ARCH effects” of Engle (1982)10. It turned out that all stock markets showed signs of autocorrelation in the variance. This is confirmed by compiling correlograms of each stock market, where the Ljung-Box statistic of Ljung and Box (1978) is significant up to 20 lags11. Since all markets are estimated in seperate equations, the correlation between the stock indices does not matter.
3.2
Announcements
According to several authors (Flannery and Protopapdakis (2002); Hardouvelis (1987); Kim et al. (2004)) we should take a look rather at the surprise component of an announce-ment instead of the news itself. The efficient market hypothesis of Fama (1970) states that prices in the stock market reflect all relevant information. Therefore expectations of investors are incorporated in the value of an asset. Only an unexpected deviation could violate this hypothesis and cause uncertainty. Therefore the announcements are defined as:
S = A − E(A)
σ (2)
Where S stands for the surprise component, A for the actual value, E(A) for the expec-tation and σ for the standard deviation of the announcement.
9A high value for the Jarque-Bera statistic rejects the null-hypothesis of a normal distribution. The
alternative hypothesis is the existence of a non-normal distribution. Table 12 in the appendix shows the Jarque-Bera statistic and its p-value.
10The test is executed with five lags in all cases. We could reject the null-hypothesis of no “ARCH”
effects for all stock indices at a 1 % significance level. The alternative hyptohesis is the existence of “ARCH” effects The results are provided in table 13 in the appendix.
11The null-hypothesis of the Ljung-box test is no serial dependence in the residuals; the alternative
Equation (2) indicates that the unexpected component of the announcement is equal to the released value minus the expected value of investors. The righthand side is divided by the standard deviation of the relevant announcement to adjust for the difference in units. We obtained the actual values from official press releases by national statistic bureaus or central banks. Following other papers (Flannery and Protopapdakis (2002); Hardouvelis (1987); Kim et al. (2004)) we took the median of polls as proxy for the expected value, in our case released by Thomson Reuters12 . For the estimation we transformed the surprise component in a dummy variable, which takes 1 on the day a surprise occurs, and 0 if there is no announcement or the expected value is equal to the actual one. To correct for the time difference between stock markets, in some cases the 1 will shift to the next trading day. In case of American and European announcements this holds for the Asian and Pacific markets. For Chinese announcements we have no adjustments, since they are the first markets that open on a trading day. We ignored in this paper the effect of holidays since most announcements are released intentionally on business days. Therefore it will not have a large impact on the results. Additionally we have to deal with the release of announcements on weekend days, which is in particular the case in China. We decided to shift the impact of the news to the next trading day, which is Mondays. We did not incorporate any other calendar effects in this paper.
12The polls are defined by the official website of Thomson Reuters as: “Reuters polls hundreds of
3.2.1 United States
Table 3: Announcements United States
Type of announcement Symbol Frequency Unit Source
Consumer Price Index YoY CPI Monthly % BLS
Producer Price Index (YoY) PPI Monthly % BLS
Unemployment Rate (MoM) UR Monthly % BLS
Industrial Production (MoM) IP Monthly % FED
Import Price Index (MoM) IPI Monthly % BLS
Export Price Index (MoM) EPI Monthly % BLS
Retail Sales (MoM) RS Monthly % CB
Leading Economic Indicators Index (MoM) LEI Monthly % TCB Consumer Sentiment Index CSI Monthly Index (0 to 100) UMSC Consumer Confidence Index CCI Monthly Index (0 to 100) CB Purchasing Manager Index PMI Monthly Index (0 to 100) ISM
Gross Domestic Product (QoQ) GDP Quarterly % BEA
Employment Cost Index ECI Quarterly % BLS
Notes: YoY stands for Year over Year rates, MoM for Month over Month rates, BLS for Bureau of Labor Statistics, FED for Federal Reserve, CB for U.S. Census Bureau, TCB for The Conference Board, UMSC for University of Michigan Surveys of Consumers, ISM for Institute of Supply Management and BEA for Bureau of Economic Analysis.
The announcements for America are selected in line with the major economic indicators provided by the Bureau of Labor Statistics13 . In addition we added a few extra variables, which are considered as important economic signs. The release of announcements in the United States is strict and has an embargo untill the release time to avoid leaking. The details are shown in table 3. Due to unavailability of data we had to express most variables in month over month rates for the United States. This is less ideal than year over year, since month over month rates are more volatile to short-term events. It only takes in account the change between two months, where year over year rates consider a whole year of events. The quarter over quarter rates are in between these two cases. Unfortunately, the two variables PPI and CCI are not suitable for the regressions due to missing observations. Next to that, we used the test of Variance Inflator Factors (VIF) to investigate multicollinearity between the dummy variables. O’Brien (2007) finds that the interpretation of this test has no statistical threshold, but relies on a rule of thumb. This rules states that if the VIF value is between 5 and 10 we have serious multicollinearity in the sample, and for values above 10 severe multicollinearity. Table 15 in the appendix presents the results of this test, and it indicates that the variables EPI and IPI have severe signs of multicollinearity. We decided to leave the variable EPI out, since it shows a strong correlation with IPI, which is due to the release on the same date. After excluding this variable, we obtained new values for the VIF test, which are presented in table 18 in the
appendix. There are no further signs of multicollinearity in the sample. As confirmation we added the correlation matrix in the appendix, namely table 19. The correlation matrix indicates the correlation on a scale from 0 to 1, where 1 is equal to perfect correlation. In the table we can see the high correlation between the EPI and IPI.
3.2.2 Europe
Table 4: Announcements Europe
Type of announcement Symbol Frequency Unit Source
Industrial Production (YoY) IP Monthly % ES
Producer Price Index (YoY) PPI Monthly % ES
Retail Sales (YoY) RS Monthly % ES
Unemployment Rate (YoY) UR Monthly % ES
Business Climate Indicator BCI Monthly Index DGEF Consumer Confidence Indicator CCI Monthly Index DGEF Economic Sentiment Indicator ESI Monthly Index DGEF Industrial Confidence Indicator ICI Monthly Index DGEF Service Confidence Indicator SCI Monthly Index DGEF Gross Domestic Product (YoY) GDP Quarterly % ES Notes: YoY stands for Year over Year rates, ES for Eurostat and DGEF for Directorate General for Economic and Financial affairs.
3.2.3 China
Table 5: Announcements China
Type of announcement Symbol Frequency Unit Source
Consumer Price Index (YoY) CPI Monthly % NBS
Producer Price Index (YoY) PPI Monthly % NBS
Retail Sales (YoY) RS Monthly % NBS
Industrial Production (YoY) IP Monthly % NBS
Imports (YoY) IMP Monthly % CC
Exports (YoY) EXP Monthly % CC
Urban Investment (YoY) UI Monthly % NBS
Money Supply M2 (YoY) M2 Monthly % PBOC
Outstanding Loan Growth (YoY) OLG Monthly % PBOC
New Yuan Loans NYL Monthly Billions Chinese yuan PBOC
External Trade Balance ETB Monthly Hundreds of millions U.S. dollar CC
Purchasing Manager Index PMI Monthly Index (0 to 100) NBS
Gross Domestic Product (YoY) GDP Quarterly % NBS
Notes: YoY stands for Year over Year rates, NBS for National Bureau of Statistics, CC for China Customs and PBOC for People’s Bank of China.
The list of Chinese announcements is based on the previous two cases and the availability of the polls of Thomson Reuters. The selection is shown in table 5. Most of the variables are defined in year over year rates, which makes them less prone to short-term volatility. Again we performed tests on multicollinearity, which can be found in table 17, 22 and 23 in the appendix. We excluded the variables RS, EXP, UI, OLG, NYL and ETB due to severe correlation with other dummy variables. The variable PPI has a VIF value that is within the threshold of 5, but near the boundry, and thus it can still give complications in the estimations. As we can see from the raw data that PPI and CPI are announced on the same date and thus only differ in terms of surprise we decided to leave this variable out for the main estimations. We will produce an extra table in the robustness checks section, to see if PPI can be included in the estimation.
4
Methodology
4.1
Model specifications
Considering the presence of conditional heteroskedasticity, we need to find a suitable model in the collection of autoregressive conditionally heteroskedastic (ARCH) models initiated by Engle (1982). Our starting point is a GARCH (1,1) model of Bollerslev (1986) accompanied by a normal error distribution. This is a widely used model in the world of empirical research, since its a rather simple and not overparameterized model. One important drawback of the GARCH (1,1) model is the assumption of symmetry. It means that the model doesn’t make a distinction between the effects of positive and negative news of the same magnitude on volatility. As mentioned before, a more realistic approach would be the inclusion of an asymmetry term, which leads us to the EGARCH model suggested by Nelson (1991). Another improvement of the EGARCH model is the relaxed assumption of the non-negativity constraint of the GARCH model. Since the con-ditional variance is measured in log14, the parameters that define the variance do not need to be positive. Even negative parameters ensure positive conditional variance. Therefore a more appropriate choice would be the EGARCH (1,1) model. We can support this decision by reporting the Akaike information criterion and the log likelihood value of the GARCH model and the EGARCH model of each estimation. The estimates are presented in table 24, 25 and 26 in the appendix. The Akaike information criterion ensures a better fit of the model if the value is minimized. Opposite to this criteria is the maximization of the log likelihood value. For each stock index we can confirm the better fit of the EGARCH (1,1) model to our data.
In addition, we need to take a closer look at the error distribution. Clearly our sample has signs of leptokurtosis and some skewness, which thus rejects the use of a normal error distribution. Although the gaussian distribution in the GARCH and EGARCH models captures more kurtosis than ordinary least squares (OLS) models (Brooks (2008)), we want to ensure selecting the right use of error terms. Other options available are the student t distribution and the generalized error distribution. As before we used the two selection criteria to select the appropriate model. In this case we compare an EGARCH (1,1) model with different error distributions. Table 27, 28 and 29 show the criteria for the estimated models, and we can conclude that the use of an EGARCH (1,1) model with a generalized error distribution is the model that fits our data best.
Following Nelson (1991) and Federova et al. (2014) we will regress our data with use of an EGARCH (1,1) model with a generalized error distribution.
14The equation of the conditional variance of the EGARCH(1,1) is given later on in this section, namely
This gives the following equations:
ri,t = µi,t+ εi,t (3)
εi,t = σi,tzi,t (4)
zi,t | Ωt−1 ∼ Ψ(0, 1, ν) (5) ln(σ2i,t) = ω + α |εi,t−1| √ σ2 i,t−1 −q2 π + γqεi,t−1 σ2 i,t−1 + βln(σi,t−12 ) + δΣDi,t (6)
Equation (3) presents the mean equation, where ri,t is the daily return of a stock index, µi,t is the constant and εi,t is the error term. Equation (4) states that the error consists of σi,t and zi,t, which are the standard deviation and standardized residuals respectively. Equation (5) indicates that the standardized residuals are computed from the previous period. Ψ is the conditional density function, which consists of a mean of zero, a variance of one and ν. ν is a vector of parameters which specifies the probability function. Equa-tion (6) indicates the condiEqua-tional variance, where ω is a constant, α is a coefficient for the ”ARCH” term, γ represents the effect of the asymmetry term, β measures the impact of the ”GARCH” term and δ gives the response of the dummy variable for the different components of macroeconomic announcements. According to Brooks (2008) the constant ω equals the long-term average value of the variance. When γ is smaller than zero, it implies that negative news has a larger impact on the variance of the daily returns than positive news.
4.2
Hypotheses
Before we turn to the results we find with the abovementioned model, we first need to determine our hypothesis regarding the research question. Considering the literature and our expectations, we obtain the following hypotheses:
Hypothesis 1: We expect that the effect of macroeconomic announcements will have an effect on the volatility of each stock index.
Statistically this implies:
H0 = 0 (7)
Hypothesis 2: We expect that the effect of macroeconomic announcements will be positive.
Statistically this implies:
H0 ≤ 0 (9)
H1 > 0 (10)
Furthermore, since the ARCH effects are proven to be present in the samples, we expect positive signs for the ”ARCH” and ”GARCH” term. We predict a significant effect for the asymmetry term, since we foresee that the impact of events are not symmetric across a time sample. This means that a negative event of the same magnitude can have a larger effect on volatility than a postive event or the other way around.
5
Results
5.1
United States
Table 5 below shows the results of the estimations of the stock markets with respect to announcements of the United States. First, we can observe a trend in significance for the announcement PMI, which is mainly significant in Europe and North and South America, except for Canada and Chile. For example, the announcement PMI has an impact of 0.318% on the volatility of the New York Stock Exchange on the announcement date. All other impacts can be interpreted in this manner. Furthermore we can observe a negative effect of the announcemount UR in the European Area, which is against the hypothesis of increasing volatility. Other announcements have a less robust impact in the world, since it shows up incidentally in the table. Next to this, we can conclude that in all cases, except for the stock exchange in Shanghai, the effect of negative news has a larger impact on volatility than positive. All asymmetry terms are significant on a 1% level. In addition, the ”ARCH” and ”GARCH” terms are highly significant, and it appears that volatility is highly persitent since the ”GARCH” term is high in all cases. Lastly the constant is in most estimations significant and has a negative sign.
only a constant regressed on the daily returns of the stock markets, which is a probable cause of this low R-squared. The Durbin-Watson statistic is an indicator of serial cor-relation between the constant and the dependent variable. In most cases, we can reject the case of serial correlation15. However, in a few cases the sample suffers slightly. A solution for this, is the inclusion of lagged variables in the mean equation of the daily stock returns. For most announcements, the figures do not change much16.
5.2
Europe
In table 6 the results are displayed for the effects of the European announcements. A significant impact of the announcement CCI is visible for the countries in North and South America, except for Brazil and Chile. For example, the impact on the New York Stock exchange can be interpreted as follow; The volatility increases with 0.502 % on an announcement day of CCI. All other coefficients in this table can be explained in the same way. In addition we can observe a negative trend for news about GDP in the countries in North America. All coefficients are significant at a 5% level at least. Furthermore, no trends can be concluded from these estimations. The ARCH and GARCH effects are significant in all regressions on a 1% level and have positive signs. It can be said that volatility is highly persistent again as the coefficient for the GARCH effect is high. As expected, the effect of the asymmetry term is negative, which means that bad European news causes a larger effect on volatility than positive news. Only the stock exchange in Shanghai is an exception, but its sign is still negative.
The estimates of the mean equation are presented in table 30 and can be interpreted as in the previous case. Again, the R-squared is in all cases below zero, and thus the model has no explanatory power. We adress the same reasons as mentioned before, namely that the mean equation solely exists of a constant. The Durbin-Watson statisics for all the stock indices show similar signs as in the case of American announcements.
15A durbin-Watson statistic of 2.00 would indicate no serial correlation. A positive deviation implies
positive autocorrelation, whereas a negative one causes negative autocorrelation.
16Since the magnitude or sign of each coefficient does not alter much, the results of including a lagged
Table 6: Estimates EGARCH(1,1) United States announcements
Stock market C ARCH ASYM GARCH CSI ECI GDP IP IPI LEI PMI RS UR CPI DW New York Stock Exchange -0.081*** 0.120*** -0.141*** 0.985*** -0.283* 0.107 -0.225 -0.061 -0.059 -0.027 0.318** -0.170 -0.176 0.184 2.13
(0.028) (0.024) (0.018) (0.003) (0.167) (0.171) (0.175) (0.194) (0.177) (0.164) (0.149) (0.197) (0.129) (0.159) Dow Jones Industrial Average -0.135*** 0.159*** -0.192*** 0.972*** -0.272 0.186 -0.143 -0.106 -0.129 0.100 0.331** 0.127 -0.175 0.148 2.17
(0.031) (0.030) (0.022) (0.005) (0.171) (0.188) (0.196) (0.217) (0.182) (0.171) (0.154) (0.203) (0.140) (0.178) SP 500 Index -0.108*** 0.143*** -0.196*** 0.975*** -0.339** 0.203 -0.174 -0.043 -0.049 -0.040 0.368** -0.103 -0.173 0.188 2.16
(0.030) (0.028) (0.022) (0.004) (0.170) (0.181) (0.190) (0.202) (0.185) (0.165) (0.152) (0.204) (0.137) (0.168) SP/TSX 60 composite -0.040* 0.080*** -0.114*** 0.990*** -0.358** 0.402** -0.161 -0.221 0.126 -0.054 0.177 -0.426** -0.254** 0.268* 2.06
(0.024) (0.019) (0.015) (0.002) (0.158) (0.163) (0.153) (0.171) (0.184) (0.151) (0.146) (0.214) (0.124) (0.146) Brazil Bovespa Stock Index -0.029 0.078*** -0.072*** 0.986*** -0.298** 0.277* 0.015 0.099 -0.006 -0.090 0.404*** -0.340* -0.369*** -0.01 2.01
(0.023) (0.016) (0.012) (0.003) (0.152) (0.155) (0.154) (0.195) (0.147) (0.148) (0.166) (0.177) (0.112) (0.150) Santiago Index general -0.134*** 0.163*** -0.068*** 0.970*** -0.081 0.107 -0.475** -0.170 -0.110 0.058 0.388** -0.245 -0.176 0.159 1.64
(0.031) (0.026) (0.014) (0.007) (0.174) (0.211) (0.185) (0.213) (0.170) (0.159) (0.180) (0.208) (0.143) (0.170) IPC -0.071*** 0.088*** -0.089*** 0.990*** -0.041 0.301* -0.044 0.044 -0.049 0.050 0.111 -0.079 -0.229* 0.121 1.79 (0.024) (0.018) (0.014) (0.002) (0.154) (0.165) (0.161) (0.201) (0.160) (0.164) (0.168) (0.187) (0.119) (0.158) FTSE 100 -0.113*** 0.126*** -0.135*** 0.979*** -0.097 -0.154 -0.177 -0.173 -0.193 0.218 0.511*** 0.183 -0.331*** 0.189 2.02 (0.026) (0.024) (0.016) (0.004) (0.157) (0.182) (0.176) (0.182) (0.159) (0.159) (0.157) (0.213) (0.128) (0.169) CAC 40 -0.063** 0.092*** -0.175*** 0.973*** -0.061 -0.266 -0.199 0.072 -0.287* 0.257 0.403*** -0.006 -0.383*** 0.095 2.05 (0.025) (0.021) (0.019) (0.005) (0.151) (0.174) (0.167) (0.185) (0.167) (0.159) (0.144) (0.214) (0.118) (0.162) DAX -0.112*** 0.128*** -0.139*** 0.977*** -0.265* -0.125 -0.076 -0.042 -0.212 0.423** 0.569*** 0.182 -0.219* 0.060 1.97 (0.027) (0.024) (0.018) (0.005) (0.158) (0.188) (0.176) (0.188) (0.171) (0.172) (0.155) (0.227) (0.131) (0.161) Swiss Market Index -0.085*** 0.130*** -0.143*** 0.969*** -0.509*** -0.168 -0.065 -0.201 -0.311 0.090 0.431*** 0.394 -0.322** -0.007 1.88
(0.030) (0.026) (0.019) (0.006) (0.173) (0.208) (0.187) (0.218) (0.195) (0.179) (0.164) (0.255) (0.157) (0.202) Australia ASX 200 -0.095*** 0.108*** -0.078*** 0.983*** 0.109 0.170 -0.198 0.013 -0.106 -0.093 0.075 0.103 -0.069 0.124 2.02
(0.024) (0.021) (0.014) (0.004) (0.150) (0.162) (0.157) (0.164) (0.143) (0.147) (0.160) (0.164) (0.121) (0.146) Shanghai SE Composite Index -0.120*** 0.184*** -0.026 0.974*** 0.065 -0.231 -0.647*** 0.015 0.273 0.099 0.014 0.050 0.076 -0.077 1.91
(0.033) (0.031) (0.019) (0.008) (0.202) (0.230) (0.213) (0.271) (0.223) (0.204) (0.214) (0.270) (0.164) (0.213) Hang Seng -0.034 0.071*** -0.058*** 0.991*** 0.028 0.105 -0.104 -0.193 -0.205 -0.062 0.210 -0.027 -0.286** 0.162 2.04
(0.024) (0.017) (0.012) (0.003) (0.166) (0.164) (0.153) (0.202) (0.160) (0.152) (0.173) (0.215) (0.125) (0.146) Nikkei 225 -0.082*** 0.143*** -0.116*** 0.969*** -0.014 -0.108 -0.414** -0.423** -0.398*** 0.041 0.052 0.443** 0.119 0.121 2.03
(0.025) (0.025) (0.016) (0.006) (0.157) (0.178) (0.176) (0.195) (0.151) (0.153) (0.156) (0.182) (0.131) (0.162) Taiwan TSEC 50 Index -0.051** 0.056*** -0.058*** 0.993*** 0.141 0.174 -0.075 0.025 -0.132 0.110 0.037 -0.038 -0.015 0.057 1.90
(0.022) (0.016) (0.011) (0.003) (0.165) (0.152) (0.172) (0.186) (0.139) (0.164) (0.176) (0.210) (0.117) (0.142)
Notes: The adjusted R-squared is in all cases -0.00. The number of observations is 1842 in all estimations. The values in brackets are the corresponding standard errors.*,**,*** indicate statistical significance at 10%, 5% and 1% level, respectively. All regressions are estimated with an General Error Distribution, except for the Nikkei 225, where a Student’s T distribution fits better. DW stands for the Durbin Watson statistic.
Table 7: Estimates EGARCH(1,1) Europe announcements
Stock market C ARCH ASYM GARCH CCI GDP IP PPI RS UR DW
New York Stock Exchange -0.081*** 0.082*** -0.159*** 0.983*** 0.502*** -0.393*** 0.107 -0.151 0.031 0.120 2.13
(0.017) (0.020) (0.017) (0.003) (0.101) (0.150) (0.149) (0.120) (0.124) (0.112)
Dow Jones Industrial Average -0.124*** 0.132*** -0.206*** 0.972*** 0.457*** -0.499*** 0.169 -0.286** 0.059 0.132 2.17
(0.021) (0.026) (0.022) (0.004) (0.125) (0.178) (0.149) (0.143) (0.138) (0.141)
SP 500 Index -0.105*** 0.107*** -0.208*** 0.975*** 0.493*** -0.486*** 0.152 -0.218 0.086 0.147 2.16
(0.020) (0.024) (0.021) (0.004) (0.117) (0.162) (0.148) (0.134) (0.132) (0.129)
SP/TSX 60 composite -0.074*** 0.082*** -0.125*** 0.988*** 0.288*** -0.336** 0.004 -0.130 -0.116 0.221 2.06
(0.015) (0.018) (0.015) (0.002) (0.097) (0.144) (0.143) (0.110) (0.121) (0.100)
Brazil Bovespa Stock Index -0.052*** 0.093*** -0.078*** 0.985*** 0.064 -0.227 -0.059 -0.014 -0.205 0.154 2.01
(0.016) (0.018) (0.012) (0.004) (0.093) (0.158) (0.141) (0.126) (0.120) (0.101)
Santiago Index general -0.151*** 0.174*** -0.067*** 0.967*** 0.245 0.017 -0.232 -0.152 0.051 -0.014 1.64
(0.022) (0.026) (0.014) (0.007) (0.156) (0.195) (0.146) (0.172) (0.164) (0.165) IPC -0.069*** 0.097*** -0.098*** 0.987*** 0.205** -0.305* 0.050 -0.179 -0.175 0.051 1.79 (0.016) (0.019) (0.013) (0.03) (0.100) (0.170) (0.140) (0.125) (0.130) (0.112) FTSE 100 -0.084*** 0.107*** -0.132*** 0.979*** 0.365*** -0.012 -0.025 -0.172 -0.134 -0.014 2.02 (0.018) (0.022) (0.014) (0.004) (0.119) (0.157) (0.141) (0.127) (0.136) (0.120) CAC 40 -0.052*** 0.102*** -0.181*** 0.971*** 0.128 -0.170 -0.128 -0.123 -0.217 0.049 2.05 (0.018) (0.022) (0.019) (0.005) (0.116) (0.158) (0.139) (0.132) (0.136) (0.122) DAX -0.081*** 0.130*** -0.137*** 0.976*** 0.072 -0.127 -0.001 -0.080 -0.240* 0.046 1.97 (0.019) (0.024) (0.018) (0.005) (0.119) (0.174) (0.148) (0.143) (0.145) (0.126)
Swiss Market Index -0.092*** 0.129*** -0.137*** 0.971*** 0.154 -0.006 0.062 -0.137 -0.271* -0.184 1.88
(0.019) (0.025) (0.018) (0.006) (0.133) (0.177) (0.159) (0.158) (0.157) (0.136)
Australia ASX 200 -0.098*** 0.106*** -0.084*** 0.981*** 0.152 0.089 0.105 -0.186 0.082 0.137 2.02
(0.018) (0.021) (0.013) (0.004) (0.105) (0.151) (0.125) (0.127) (0.124) (0.107)
Shanghai SE Composite Index -0.108*** 0.159*** -0.011 0.985*** -0.147 0.224 0.138 -0.160 0.236 -0.022 1.91
(0.024) (0.028) (0.016) (0.006) (0.134) (0.227) (0.180) (0.161) (0.165) (0.157)
Hang Seng -0.045*** 0.080*** -0.060*** 0.990*** 0.058 0.054 -0.278* -0.139 -0.007 0.210* 2.04
(0.016) (0.018) (0.012) (0.003) (0.095) (0.162) (0.151) (0.113) (0.126) (0.110)
Nikkei 225 -0.120*** 0.167*** -0.112*** 0.963*** 0.152 0.131 0.057 -0.008 0.260 -0.044 2.03
(0.022) (0.029) (0.017) (0.08) (0.150) (0.205) (0.152) (0.165) (0.161) (0.162)
Taiwan TSEC 50 Index -0.044*** 0.055*** -0.067*** 0.992*** 0.119 0.131 -0.052 0.066 -0.062 0.097 1.90
(0.013) (0.015) (0.012) (0.003) (0.080) (0.156) (0.150) (0.102) (0.112) (0.096)
Notes: The adjusted R-squared is in all cases -0.00. The number of observations is 1842 in all estimations. The values in brackets are the corresponding standard errors.*,**,*** indicate statistical significance at 10%, 5% and 1% level, respectively. All regressions are estimated with an General Error Distribution.
5.3
China
Table 7 presents the estimates of the regressions with the macroeconomic announcements from China. From the table we can only conclude a small effect on a 5% and 10% level of CPI on the stock indices in the United States. However, this effect is negative, which would imply that stock markets cool down instead of increasing uncertainty. For example, the effect can be explained as decrease in volatility of -0.400% on the New York Stock Exchange as result of a CPI announcement in China. In addition two of the three indices in the United States show a positive response in volatility towards the news about the GDP of China. Turning to the GARCH and ARCH effects, we can observe the same significance and effects as in the previous cases. Furthermore the asymmetry term is in almost all estimations significant on a 1% level significant except for the stock exchange in Australia. Also the Shanghai stock exchange is on a lower level significant, namely 10%. The constant is significant in all cases and has negative sign.
Table 8: Estimates EGARCH(1,1) China announcements
Stock market C ARCH ASYM GARCH CPI GDP IMP IP M2 PMI DW
New York Stock Exchange -0.110*** 0.112*** -0.211*** 0.963*** -0.449** 0.133 0.253 0.216 -0.159 0.164 2.10
(0.028) (0.033) (0.024) (0.007) (0.180) (0.212) (0.192) (0.199) (0.165) (0.149)
Dow Jones Industrial Average -0.141*** 0.130*** -0.269*** 0.947*** -0.395** 0.194 0.229 0.335 -0.239 0.105 2.09
(0.031) (0.036) (0.027) (0.010) (0.189) (0.227) (0.194) (0.205) (0.170) (0.167)
SP 500 Index -0.120*** 0.109*** -0.302*** 0.946*** -0.447** 0.202 0.323* 0.254 -0.214 0.151 2.09
(0.028) (0.033) (0.030) (0.009) (0.181) (0.222) (0.190) (0.210) (0.173) (0.157)
SP/TSX 60 composite -0.027 0.013 -0.161*** 0.981*** -0.226 0.190 0.036 0.025 -0.075 0.060 1.82
(0.020) (0.021) (0.019) (0.004) (0.147) (0.178) (0.178) (0.153) (0.107) (0.116)
Brazil Bovespa Stock Index -0.049** 0.071*** -0.079*** 0.978*** -0.119 0.057 -0.033 -0.049 0.072 0.343** 2.04
(0.021) (0.023) (0.014) (0.009) (0.155) (0.183) (0.160) (0.180) (0.146) (0.137)
Santiago Index general -0.155*** 0.157*** -0.104*** 0.964*** -0.111 0.349 0.169 -0.222 -0.042 0.162 1.63
(0.034) (0.037) (0.021) (0.010) (0.210) (0.246) (0.209) (0.227) (0.191) (0.186) IPC -0.064*** 0.060** -0.119*** 0.979*** -0.353** 0.127 0.117 0.203 -0.014 0.224 1.92 (0.022) (0.024) (0.018) (0.006) (0.160) (0.199) (0.167) (0.177) (0.152) (0.140) FTSE 100 -0.125*** 0.159*** -0.160*** 0.959*** -0.078 -0.112 -0.240 0.023 -0.204 0.260 1.97 (0.033) (0.036) (0.021) (0.010) (0.200) (0.241) (0.211) (0.214) (0.183) (0.171) CAC 40 -0.066** 0.118*** -0.208*** 0.958*** -0.221 0.091 -0.289 0.011 -0.119 0.213 2.01 (0.026) (0.034) (0.028) (0.010) (0.199) (0.224) (0.207) (0.196) (0.173) (0.154) DAX -0.098*** 0.149*** -0.155*** 0.965*** -0.112 0.081 -0.194 0.118 -0.243 0.239 1.91 (0.027) (0.034) (0.024) (0.010) (0.203) (0.232) (0.217) (0.208) (0.178) (0.170)
Swiss Market Index -0.153*** 0.191*** -0.161*** 0.934*** 0.039 -0.007 -0.069 -0.087 -0.063 0.031 1.81
(0.040) (0.044) (0.030) (0.016) (0.279) (0.312) (0.245) (0.276) (0.220) (0.191)
Australia ASX 200 -0.098*** 0.094*** -0.123*** 0.973*** -0.381** 0.002 0.114 0.018 0.266* 0.232 2.01
(0.024) (0.026) (0.021) (0.007) (0.177) (0.207) (0.200) (0.180) (0.144) (0.151)
Shanghai SE Composite Index -0.077*** 0.090*** 0.008 0.994*** -0.135 -0.078 0.549*** 0.249 -0.057 -0.202 1.81
(0.019) (0.024) (0.012) (0.004) (0.146) (0.203) (0.184) (0.168) (0.134) (0.131)
Hang Seng -0.080*** 0.091*** -0.086*** 0.975*** -0.349* -0.027 0.480** 0.008 0.016 0.232 1.94
(0.023) (0.028) (0.018) (0.008) (0.186) (0.241) (0.205) (0.215) (0.169) (0.176)
Nikkei 225 -0.113*** 0.187*** -0.130*** 0.918*** 0.052 -0.120 0.180 -0.038 -0.070 0.113 2.12
(0.035) (0.044) (0.028) (0.022) (0.251) (0.323) (0.257) (0.254) (0.225) (0.217)
Taiwan TSEC 50 Index -0.029 0.021 -0.114*** 0.979*** 0.003 -0.036 0.124 -0.159 0.303 0.007 1.96
(0.020) (0.021) (0.018) (0.006) (0.167) (0.209) (0.203) (0.188) (0.132) (0.142)
Notes: The adjusted R-squared is in all cases -0.00. The number of observations is 1842 in all estimations. The values in brackets are the corresponding standard errors.*,**,*** indicate statistical significance at 10%, 5% and 1% level, respectively. All regressions are estimated with an General Error
5.4
Robustness checks
5.4.1 Volatility indexA meaningful way of looking at the robustness of our results, is to check if it also holds for implied volatility, which is an expectation of the volatility over the next period. As a proxy we take the volatility indices of some of the markets we used in our previous estimations. The indices are: Volatility index of the S&P 500 (VIX), Volatility index of CAC40 (VCAC40), Volatility index of the DAX (VDAX), Volatility index of the FTSE 100 (VFTSE100), Volatility index of the ASX 200 (VASX200), Volatility index of the Nikkei 225 (VNikkei225), Volatility index of the Hang Seng Index (VHSI). The results of these estimations for each case, are presented in table 9, 10 and 11. For American news we can observe similar effects for the announcement PMI on the volatility indices in Europe. This confirms the increased uncertainty around this announcement. For the cases Europe and China we find no corresponding responses. The results we found for CCI and GDP in North and South America are not visible in the estimation with implied volatility. However, we need to stress here that we only included one index for this area, namely the volatility index of S&P 500. The negative effect of GDP is confirmed for the case Europe. That we did not find anything meaningful in this robustness check for China actually confirms our findings in the main estimation. We did not acquire any significant results there.
5.4.2 Time period
We also checked if the exclusion of the year 2008 for the samples America and Europe has any effect on the results we found. The results are not included in this paper, but instead we give an indication of the estimations we found. For the U.S. announcement we find similar significant results, except that the significance decrease in most case to a 5% or 10% level. Also for European news the significant effects in the markets of North America remain.
Table 9: Robustness check: Estimates EGARCH(1,1) United States announcements
Stock market C ARCH ASYM GARCH CSI ECI GDP IP IPI LEI PMI RS UR CPI DW
VIX 0.142*** 0.033 0.230*** 0.956*** -0.241 -0.100 -0.165 -0.052 -0.000 0.075 0.099 0.047 -0.290** 0.092 2.13 (0.038) (0.024) (0.024) (0.009) (0.156) (0.167) (0.175) (0.185) (0.187) (0.172) (0.159) (0.189) (0.131) (0.154) VCAC40 0.169** 0.176*** 0.102*** 0.916*** -0.141 -0.460* -0.505** 0.190 -0.047 0.371** 0.315* -0.103 0.216 -0.257 2.12 (0.069) (0.039) (0.025) (0.021) (0.172) (0.254) (0.234) (0.224) (0.191) (0.176) (0.177) (0.236) (0.165) (0.202) VDAX 0.031 0.071*** 0.115*** 0.961*** -0.217 -0.114 -0.035 0.356* -0.133 0.345** 0.454*** 0.088 -0.014 0.053 1.93 (0.040) (0.025) (0.021) (0.011) (0.150) (0.193) (0.177) (0.206) (0.197) (0.165) (0.171) (0.236) (0.129) (0.169) VFTSE100 0.037 0.031* 0.113*** 0.984*** -0.288* 0.078 -0.190 -0.074 -0.085 0.104 0.392** 0.079 -0.279*** 0.073 2.10 (0.029) (0.017) (0.020) (0.005) (0.158) (0.152) (0.156) (0.187) (0.147) (0.151) (0.179) (0.222) (0.107) (0.142) VASX200 0.021 0.041** 0.090*** 0.984*** 0.063 0.231 0.005 -0.053 0.229 -0.216 -0.119 0.011 -0.021 0.068 2.15 (0.025) (0.017) (0.016) (0.005) (0.158) (0.159) (0.159) (0.193) (0.158) (0.151) (0.167) (0.217) (0.113) (0.133) VNIKKEI225 0.086* 0.124*** 0.143*** 0.942*** 0.085 -0.126 -0.135 -0.227 0.116 0.076 0.035 -0.014 0.230 0.185 1.99 (0.045) (0.033) (0.027) (0.013) (0.175) (0.213) (0.211) (0.215) (0.189) (0.170) (0.181) (0.211) (0.153) (0.193)
Notes: The adjusted R-squared is in all cases -0.00. The number of observations is 1842 in all estimations. HSI is not added since it has not enough observations for this time period. The values in brackets are the corresponding standard errors.*,**,*** indicate statistical significance at 10%, 5% and 1% level, respectively. DW stands for the Durbin Watson statistic.
Table 10: Estimates EGARCH(1,1) Europe announcements
Stock market C ARCH ASYM GARCH CCI GDP IP PPI RS UR DW
VIX 0.142*** 0.045* 0.231*** 0.951*** -0.085 -0.417** 0.167 -0.229 0.036 0.004 2.13 (0.037) (0.024) (0.023) (0.009) (0.115) (0.165) (0.140) (0.145) (0.140) (0.132) VCAC40 0.085** 0.100*** 0.107*** 0.959*** -0.061 -0.061 -0.149 -0.131 0.071 0.032 2.12 (0.042) (0.029) (0.0213) (0.013) (0.130) (0.185) (0.142) (0.150) (0.155) (0.143) VDAX 0.068* 0.059** 0.119*** 0.964*** -0.000 -0.101 0.233 -0.134 -0.040 0.001 1.93 (0.035) (0.025) (0.021) (0.011) (0.109) (0.167) (0.143) (0.130) (0.139) (0.125) VFTSE100 0.043* 0.026 0.114*** 0.984*** -0.091 0.044 -0.126 -0.070 0.106 -0.019 2.10 (0.023) (0.016) (0.019) (0.005) (0.073) (0.142) (0.137) (0.099) (0.117) (0.084) VASX200 0.020 0.038** 0.094*** 0.981*** -0.037 0.076 0.163 -0.070 0.362*** -0.184* 2.15 (0.025) (0.017) (0.018) (0.006) (0.080) (0.152) (0.141) (0.112) (0.119) (0.095) VNIKKEI225 0.125*** 0.102*** 0.159*** 0.935*** 0.196 -0.075 0.115 -0.199 0.384** -0.073 1.99 (0.046) (0.031) (0.025) (0.015) (0.167) (0.225) (0.175) (0.182) (0.175) (0.187)
Notes: The adjusted R-squared is in all cases -0.00. The number of observations is 1842 in all estimations. HSI is not added since it has not enough observations for this time period. The values in brackets are the corresponding standard errors.*,**,*** indicate statistical significance at 10%, 5% and 1% level, respectively. DW stands for the Durbin Watson statistic.
Table 11: Estimates EGARCH(1,1) China announcements
Stock market C ARCH ASYM GARCH CPI IMP IP GDP M2 PMI DW
VIX 0.178** 0.080** 0.251*** 0.927*** -0.080 0.635*** 0.185 0.487** -0.149 -0.104 2.12 (0.070) (0.036) (0.031) (0.019) (0.204) (0.202) (0.235) (0.237) (0.184) (0.169) VCAC40 0.183** 0.068* 0.099*** 0.931*** -0.057 0.395* -0.057 -0.004 -0.116 0.225 2.05 (0.092) (0.037) (0.029) (0.027) (0.212) (0.214) (0.210) (0.248) (0.190) (0.173) VDAX 0.059** 0.015 0.088*** 0.977*** 0.234* 0.007 -0.220 0.220 0.013 0.029 1.96 (0.030) (0.020) (0.022) (0.008) (0.129) (0.185) (0.156) (0.198) (0.129) (0.124) VFTSE100 0.034 0.023 0.116*** 0.983*** 0.114 0.328* -0.084 -0.229 0.101 -0.021 2.06 (0.033) (0.020) (0.022) (0.008) (0.149) (0.192) (0.151) (0.215) (0.118) (0.125) VASX200 -0.00 0.071** 0.058*** 0.977*** 0.031 0.321 0.336* -0.489** 0.057 0.181 2.19 (0.043) (0.028) (0.019) (0.013) (0.196) (0.222) (0.198) (0.222) (0.149) (0.164) VNIKKEI225 0.054 0.123*** 0.114*** 0.946*** 0.232 0.319 -0.254 0.461* 0.204 0.101 2.06 (0.046) (0.039) (0.034) (0.015) (0.216) (0.226) (0.219) (0.265) (0.186) (0.189) VHSI 0.045 0.132*** 0.108*** 0.952*** -0.390* 0.352** 0.127 -0.304 0.293 -0.048 1.98 (0.044) (0.043) (0.033) (0.014) (0.215) (0.179) (0.248) (0.287) (0.191) (0.201)
Notes: The adjusted R-squared is in all cases -0.00. The number of observations is 1842 in all estimations. The values in brackets are the corresponding standard errors.*,**,*** indicate statistical significance at 10%, 5% and 1% level, respectively. DW stands for the Durbin Watson statistic.
6
Conclusion
6.1
Summary
This paper contributed in several ways to the field of volatility observed in stock mar-kets. We started with analyzing the existing literature and searching for a consensus. It appeared that research is mainly focused on announcements from the United States, and that effects on stock markets are confirmed by various papers. Next to that, most of the investigated markets were analyzed with help of GARCH models. This paper broadens the scope by researching stock markets on a global scale and taking multiple countries as source of announcements. We decided to take the United States, Europe and China as representative for the release of macroeconomic news. Furthermore, this paper diversi-fies from most of the other papers by taking an EGARCH(1,1) model with a generalized error distribution as estimation model. The main reasoning for this is the inclusion of the asymmetry term, which allows positive and negative news of the same magnitude to have different impacts on volatility. Statistical criteria supported this choiche. After estimating the results, we found several announcements supporting our hypothesis of a positive and significant effect on stock market volatility. However, we also found some puzzling results, since they turned out to be negative. In the next two sections we will first put these results in context with the literature. After that limitations and further recommendations will be given.
6.2
Conclusions
liter-ature found on announcements from the United States, we cannot support any paper nor can other papers support ours on particular announcements. Next to that, we can argue that there is some response on volatility across the world, but according to our estimates not globally. This contradicts with the paper of Nikkinen et al. (2006) who espcially for the G7 region, multiple postive effects on volatility. However, we should keep in mind that it is difficult to compare the estimates as we use a different estimation method and derivation of the surprise component, since many papers use other polls as median17.
The European announcements CCI and GDP mainly impact the indices in North Amer-ica including Mexico. The positive signifAmer-icant effect of CCI confirms our hypotheses. A simple explanation is that it indicates the trust of consumers in the economy, which is valuable information for manufactures. As we can see it increase volatility and thus cre-ates uncertainty on the market. On the other hand, GDP does not, which is an odd effect. Looking at the raw data, we can see mainly negative surprises, which in theory suggest more uncertainty in the stock markets, since GDP growth decreases. The robustness check does not support the findings on CCI, but does find some evidence for the negative effect of GDP. Federova et al. (2014) find positive effects of CPI and GDP on volatility in their research. Although we use different markets, we can argue that the effects are also present in our sample. We need to stress that the effect of GDP is opposite and therefore not logical in a theoretical sense.
For China we can observe that the effects on volatilities are not present on a global scale. We can only find a local negative effect of CPI in the American stock markets and a positive effect of IMP on Chinese markets. This gives mixed conclusions regarding our hypothesis. For the IMP announcement we should take in mind that also the com-ponent EXP is related to this response as these variables are strongly correlated. This effect could be explained by the fact that China is a large export country and therefore can be valuable information towards investors in the domestic markets. Furthermore the regressions with implied volatilities underline the positive response towards IMP of the Hang Seng index. Grasping back to literature, we can not say that the manufacturing announcements, like in the paper of Baum et al. (2015), influence the volatilities of stock markets on a global scale.
6.3
Limitations & recommendations
Our research has encountered several limitations, among which the specification of the mean equation, other possible explanatory variables, remaining “ARCH” effects, the length of the sample, the state of the economy and data reliability. First, we are aware 17An often used database is Bloomberg, which has its own polls. Those are probably slightly different
that the mean equation is not correctly specified as the r-squared value is slightly below zero in all estimations. The misspecification is related to the fact that we did not include any other explanatory variables. Although in general stock market returns are unpre-dictable and thus have not many explanatory variables, there could be regularities. For instance, there could be other calendar effects, like the January effect. It would also be a suggestion for further research to include the dummy variables in both the mean and variance equation. This could strengthen the model as it is said in the literature that there is an impact of announcements on the daily returns.
After estimating the models, we performed again the “ARCH” test of Engle(1982) and the Ljung-Box test of Ljung and Box(1987) on the estimation residuals18. It appears that some markets still have signs of serial dependence. Stock markets are inherent to this serial dependence as Brooks (2008) indicated with the stylized facts of market returns. However, futher research could try to minimize this effects as much as possible in order to capture a more precise effect on volatility.
Another improvement of the model could be the extensions of the time period. The additional years of data can result in higher robustness of the estimated coefficients. In addition it can reduce the possible effects of multicollinearity present in the data (O’Brien (2007)). In our estimations we had to drop several variables caused by multicollinearity. This can have a severe impact on the conclusions if this reduction of regressors is not justified in the theory. Since we work with dummy variables and daily data in our esti-mation, it is hard to make a distinction between them if they are released on the same date. In this paper is chosen to leave the correlated variables out as they can seriously bias the results of the estimation otherwise. A recommendation for further research could be the use of standardized values of the announcements or a higher frequency of data. A higher frequency, like 5-minute intervals, can give a closer look on the volatilities around the release time instead of the date.
Moreover, one could extend this research by investigating the difference between “good” and “bad” news. “bad” news could have, for example, a larger effect on volatility, like in the papers of Andersen et al. (2004) and Kim et al. (2004). Additionally, the state of the economy can be of interest, as news can have a different impact in a downturn or rising, which is investigated by McQueen and Roley (2003). Therefore it can be an extension by using different stock markets to seek for external validity.
At last, the reliability of Chinese data is intended to increase the next years, as the government took stricter measures since 2011 regarding the release of macroeconomic
nouncements (Baum et al. (2015). We assume this process will go on the upcoming years, and thus futher research can profit from these developments.
7
References
Andersen, T.G., Bollerslev, T., Diebold, F.X. and Vaga, C. (2007), “Real-Time Price Discovery in Stock, Bond and Foreign Exchange Markets”, Journal of International Eco-nomics, Vol. 73, pp. 251-277
Baum, C.F., Kurov, A. and Wolfe, M.H. (2015), “What do Chinese macro announce-ments tell us about the world economy”, Journal of International Money and Finance, 2015 (article in press)
Black, F. (1976), “Studies of Stock Market Volatility Changes”, Proceedings of the Amer-ican Statistical Association, Business and Economic Statistics Section, pp. 17-181
Bollerslev, T. (1986), “Generalized Autoregressive Conditional Heteroskedasticity”, Jour-nal of Economics, Vol. 31, pp. 307-327
Brooks, C. (2008), “Introductory Econometrics for Finance”, Cambridge, second edition
Dickey, D.A. and Fuller, W.A. (1979), “Distribution of the Estimators for Autoregres-sive Time Series with a Unit Root”, Journal of the American Statistical Association, Vol. 74, pp. 427-431
Engle, R.F. (1982), “Autoregressive Conditional Heteroscedasticity with Estimates of the Variance of United Kingdom Inflation”, Econometrica, Vol. 50, pp. 987-1007
Fama, E.F. (1970), “Efficient Capital Markets: A Review of Theory and Empirical Work”, The Journal of Finance, Vol. 25, pp. 383-417
Fedorova, E., Wallenius, L. and Collan, M. (2014), “The impact of euro area macroe-conomic announcements on CIVETS stock markets”, Procedia Emacroe-conomics and Finance, Vol. 15, pp. 27-37
Flannery, M.J. and Protopapadakis, A.A. (2002), “Macroeconomic Factors Do Influence Aggregate Stock Returns”, The Review of Financial Studies, Vol. 15, pp. 751-782
macroeconomic announcements in new EU markets: Evidence from intraday data”, Jour-nal of Financial Stability, Vol. 5, pp. 199-219
Hardouvelis, G.A. (1987), “Macroeconomic information and stock prices”, Journal of Economics and Business, Vol. 39, pp. 131-140
Jarque, C. M. and Bera, A. K. (1987), “A test for normality of observations and re-gression residuals”, International Statistical Review, Vol. 55, pp. 163–172
Kim, S., McKenzie, M.D. and Faff, R.W. (2004), “Macroeconomic news announcements and the role of expectations: evidence for US bond, stock and foreign exchange markets”, Journal of Multinational Financial Management, Vol 14., pp 217-232
Ljung, G.M. and Box, G.E.P. (1978), “On a measure of lack of fit in time series models”, Biometrika, Vol. 65, pp. 297-303
McQueen, G. and Roley, V.V. (1993), “Stock Prices, News, and Business Conditions”, The Review of Financial Studies, Vol. 6, pp. 683-707
Nelson, D.B. (1991),“Conditional Heteroskedasticity in Asset Returns: A New Approach”, Econometrica, Vol. 59, pp. 347-370
Nikkinen, J. and Sahlstr¨om, P. (2004), “Scheduled domestic and US macroeconomic news and stock valuation in Europe”, Journal of Multinational Financial Management, Vol. 14, pp. 201-215
Nikkinen, J., Omran, M., Sahlst¨orm, P. and ¨Aij¨o, J. (2006), “Global stock market re-actions to scheduled U.S. macroeconomic news announcements”, Global Finance Journal, Vol. 17, pp. 92-104
O’Brien, R.M. (2007), “A Caution Regarding Rules of Thumb for Variance Inflation Factors”, Quality & Quantity, Vol. 41, pp. 673-690
8
Appendix
Table 12: Statistcal test for normality: The Jarque-Bera test
Stock market JB stat P-value
New York Stock Exchange 8110.07 0.00***
Dow Jones Industrial Average 10039.01 0.00***
SP 500 Index 9574.13 0.00***
SP/TSX 60 composite 10561.67 0.00***
Brazil Bovespa Stock Index 5139.20 0.00***
Santiago Index general 12336.14 0.00***
IPC 6720.69 0.00***
FTSE 100 5777.38 0.00***
CAC 40 2516.23 0.00***
DAX 2689.54 0.00***
Swiss Market Index 8182.14 0.00***
Australia ASX 200 24116.28 0.00***
Shanghai SE Composite Index 1609.97 0.00***
Hang Seng 9324.07 0.00***
Nikkei 225 6877.57 0.00***
Taiwan TSEC 50 Index 1126.76 0.00***
Table 13: Statistcal test for the presence of ”Arch” ef-fects
Stock market F-statistic P-value
New York Stock Exchange 161.92 0.00***
Dow Jones Industrial Average 128.17 0.00***
SP 500 Index 135.76 0.00***
SP/TSX 60 composite 133.93 0.00***
Brazil Bovespa Stock Index 120.29 0.00***
Santiago Index general 71.31 0.00***
IPC 75.14 0.00***
FTSE 100 131.57 0.00***
CAC 40 80.20 0.00***
DAX 65.17 0.00***
Swiss Market Index 140.50 0.00***
Australia ASX 200 114.02 0.00***
Shanghai SE Composite Index 47.48 0.00***
Hang Seng 129.47 0.00***
Nikkei 225 168.31 0.00***
Taiwan TSEC 50 Index 45.55 0.00***
Table 14: Statistical test autocorrelation error term: Ljung-Box Q-statistic on squared residuals
1 lag 10 lag 20 lag
Stock market Q-statistic P-value Q-statistic P-value Q-statistic P-value
New York Stock Exchange 86.60 0.00*** 1744.20 0.00*** 2851.00 0.00***
Dow Jones Industrial Average 72.55 0.00*** 1530.70 0.00*** 2602.30 0.00***
SP 500 Index 79.92 0.00*** 1537.90 0.00*** 2551.80 0.00***
SP/TSX 60 composite 233.34 0.00*** 1919.00 0.00*** 3143.00 0.00***
Brazil Bovespa Stock Index 59.21 0.00*** 1496.60 0.00*** 2347.70 0.00***
Santiago Index general 150.66 0.00*** 844.19 0.00*** 921.83 0.00***
IPC 44.95 0.00*** 1066.00 0.00*** 1889.80 0.00***
FTSE 100 124.02 0.00*** 1535.90 0.00*** 2086.10 0.00***
CAC 40 74.62 0.00*** 917.44 0.00*** 1314.10 0.00***
DAX 50.51 0.00*** 827.09 0.00*** 1441.80 0.00***
Swiss Market Index 353.79 0.00*** 1401.00 0.00*** 1732.80 0.00***
Australia ASX 200 86.42 0.00*** 1592.10 0.00*** 2264.60 0.00***
Shanghai SE Composite Index 91.57 0.00*** 651.25 0.00*** 1054.70 0.00***
Hang Seng 302.54 0.00*** 1588.70 0.00*** 2279.60 0.00***
Nikkei 225 64.40 0.00*** 1766.90 0.00*** 2310.50 0.00***
Taiwan TSEC 50 Index 14.64 0.00*** 538.42 0.00*** 974.92 0.00***
Notes: *,**,*** indicate statistical significance at 10%, 5% and 1% level, respectively.
Table 15: Correlation matrix announcements United States
CCI CPI CSI ECI GDP EPI IP IPI LEI PMI PPI RS UR
