Master Thesis
International Economics and Business
The impact of geopolitical risk on equity markets
2
Abstract
This paper explains the effects of geopolitical risks on equity markets by establishing a novel geopolitical risk index from Bloomberg News on a daily basis. Geopolitical risk influences the economy and financial markets for which it is important to understand this relation in detail.
The news-based index established in this paper is based upon the geopolitical risk index from Caldara and Iacoviello (2018). Novel to the index from Caldara and Iacoviello (2018) is the higher frequency and establishing the index from Bloomberg News instead of physical newspapers. Bloomberg News is especially designed for financial markets and publishes news directly, whereas physical newspapers generally publish the news from yesterday. This index is then used as an independent variable within regressions for which the different equity markets are the dependent variables.
Equity markets in general react negative to geopolitical risk and different sub-sectors react differently towards geopolitical risk. Equity indices concerning oil and tobacco have a positive relation with geopolitical risk but equity indices concerning financial services and
banks have a negative relation to geopolitical risk. These results are robust to various checks.
Comparing the index established in this paper with the index from Caldara and Iacoviello (2018) leads to the conclusion that having a higher frequency index which is specifically designed for financial markets finds a significant effect of geopolitical risk on equity markets which is not the case with the index from Caldara and Iacoviello (2018).
3
Index
1. Introduction ... 4
2. Literature review ... 6
2.1. Uncertainty and the economy ... 7
2.2. Uncertainty and financial markets ... 8
2.3. Geopolitical risk... 9
2.3.1. Measuring geopolitical risk ... 9
2.3.2. Geopolitical risk and the economy ... 10
2.4. Geopolitical risk and financial markets ... 10
3. Data and methods ... 12
3.1. The construction of the GPRB index ... 12
3.1.1 Description of the Bloomberg News ... 13
3.1.2 Comparison GPRB and GPR ... 15
3.2. The financial markets analysed ... 16
3.3. Control variables ... 17
3.4. The estimated regressions ... 19
4. Results ... 23
4.1. Results of the estimated equations ... 23
4.1.1. Results per equity index ... 26
4.2. VIX ... 30
4.3. Robustness analysis ... 31
4.3.1. Splitting the regression ... 31
4.3.2. Outlier in 2003... 33
4.4. Comparing results with GPR index ... 34
5. Conclusion ... 35
References ... 37
4
1. Introduction
Geopolitical risks have not disappeared as might have been expected with the end of the Cold War. Terrorist attacks, the changing foreign policy of the USA, and the increased migration inflows in Europe are just three factors that raise concerns for geopolitical risks.
These concerns are relevant for the economy and financial markets. For example, in a survey by the US polling company Gallup, three quarters of a thousand investors polled identified geopolitical developments and their economic impact as a potential threat to financial investments (Business Wire, 2017).
Caldara and Iacoviello (2018) constructed a geopolitical risk index and used it in an analysis that concludes that geopolitical risks adversely affects economic output and equity prices. Middeldorp et al. (2018) use the same index to estimate that geopolitical risks also negatively impact the Dutch GDP growth.
Caldara and Iacoviello’s (2018) index is constructed by counting newspaper articles with terms related to geopolitical risk. The frequency of articles containing words like “nuclear war” or “terrorist threat” are divided by the total number of articles. A similar newspaper based method is used by Saiz and Simonsohn (2013), who study corruption in American cities, Baker et al. (2016) who focus on policy uncertainty, and Husted et al. (2017) who investigate the effect of monetary policy uncertainty. Yet, Caldara and Iacoviello (2018) are the first ones using this newspaper method to study geopolitical risk.
The aim of this paper is to provide an improved understanding of the effects that geopolitical risks have on financial markets, particularly equity markets. This is done by constructing a geopolitical risk index similar to that of Caldara and Iacoviello (2018) but at a daily frequency using Bloomberg News data. Caldara and Iacoviello (2018) use a monthly index for their analysis, but also constructed a daily index. A higher frequency is useful when studying financial markets because they respond to news immediately, which is also advised by Caldara and Iacoviello (2018: 14). The index of Caldara and Iacoviello (2018) is constructed by using physical newspapers. However, when studying financial markets, physical newspapers are not well suited as they historically report news the next morning. Bloomberg News is a real-time news service especially designed for financial markets and widely used by traders and other financial market participants. For which the question researched is:
5 This paper adds to the existing literature on the assessment of uncertainty and the effects it has on the economy, like Baker et al. (2016), Gupta et al. (2018), Husted et al. (2017), and specifically on the effects of geopolitical risk on financial markets, like Caldara and Iacoviello (2018) and Kollias et al. (2013). The latter, Kollias et al. (2013), investigated the relation between war and terrorism and the oil price-stock index. The method used by Kollias et al. (2013) is different to that of Caldara and Iacoviello (2018), as they do not construct an index but study the covariance between stock and oil returns.
Caldara and Iacoviello (2018) used monthly data to look at equity markets. Their main specification uses a vector autoregressive (VAR) economic variable with the S&P 500 and a few sub-sectors. They found that the S&P 500 declines less than 3 percent after a shock that is equal to the average of the 9 largest increases in their geopolitical risk index. They also looked at a global headline equity index and that of 17 countries. While the impact on the global index is significant, they found large variations in the strength of the impact on national equity markets, with several statistically insignificant.
This paper uses daily data for a global equity market index with numerous subsectors. It also examines the impact of geopolitical risk on equity market volatility (the VIX) and the impact on a Dutch equity market headline index and subsectors. The main contribution of this paper is that when estimating the regressions in this research with the index of Caldara and Iacoviello (2018), results in the fact that their index does not lead to significant changes in equity markets, whereas the index established in this paper does.
The headline equity markets studied in this paper are depicted in Figure 1.1, together with the geopolitical risk index constructed from Bloomberg News, GPRB, and the geopolitical risk index of Caldara and Iacoviello (2018), hereafter referred to as the GPR. When the geopolitical risk indices are at their highest peak, during March 2003, the equity indices decline sharply. The next large drop in the equity indices is during the financial crisis, but this was not a geopolitical event and, as such, there is no increase in the geopolitical risk index. This paper finds that there is a relation between the headline equity market indices and the GPRB. There is also a close relation between the GPRB and the GPR.
6 that respond positively. For the world subsectors, some equity indices react positively and others react negatively to geopolitical risk. Indices like oil and gas, oil and gas producers and
tobacco react positively to geopolitical risk. Alternative energy, construction and materials, financials, banks, life insurance and financial services are indices that have a negative
reaction towards geopolitical risk. These results are robust to various checks.
Figure 1.1. Movement of GPRB, GPR and Dutch and worldwide equity indices. GPRB and the GPR are normalized to 100 in 1986. Source: Caldara and Iacoviello (2018)1, Macrobond
The rest of this paper will be structured in the following way. First in the literature review, the theoretical background is discussed. Then in the data and method section, the GPRB index is explained in greater detail and its relation to the GPR index, together with an extensive description of the equity indices analysed and the control variables used. Finally, it will elaborate on how the regressions are estimated. In the results section the outcomes of the estimated regressions, the robustness analysis and the comparison with the index from Caldara and Iacoviello (2018) are shown. The result section will be followed up with the conclusion in which the research question is answered, the limitations of this research are examined, and ideas for future research will be given.
2. Literature review
7 economy in general, where after it will get more specific towards the topic researched in this paper. At the end of this chapter the hypothesis that follows from the reviewed literature, and that will be tested in this paper, will be stated.
2.1. Uncertainty and the economy
Uncertainty has an influence on the economy, which will be demonstrated in the empirical literature later in this section, but first the intuition behind this story should be explained. The mechanism through which uncertainty exerts its effect on the economy can be found in Bernanke (1983). He explained that an optimizing investor has to make the decision on what kind of investments he wants to make and what the right time of the investment will be. In the case of an increase in uncertainty, it will be more beneficial to delay the investment decision because there might be better information available in the future. Investment will then be postponed and there is a downward pressure on the economy. This is why uncertainty has an effect on the economy in general.
The negative effect that uncertainty has on the economy, has been empirically investigated by various authors. Bloom (2009) built a model that can simulate a macro uncertainty shock, like the OPEC I oil-price shock or the assassination of John F. Kennedy, by using firm-level data. He found that an increase in uncertainty decreases aggregate investment and employment, because with an increased uncertainty, firms pause their investment and hiring. This is in line with the Bernanke’s (1983) explanation on why uncertainty has an effect on the economy. Bloom (2009) additionally discovered that an uncertainty shock also has consequences on the productivity growth, as the halt in activity stops the reallocation between different entities.
Gupta et al. (2018) researched the impact of domestic and of US economic policy uncertainty on interest rates, prices and output growth in the Eurozone, and whether there is a difference of impact of uncertainty during recessions or expansions. They found that uncertainty shocks during recessions have a relatively greater negative impact on output growth in the Euro area, than uncertainty shocks during expansions.
8 (2018). The main finding of Baker et al. (2016) is that economic policy uncertainty increases stock price volatility, reduces employment and investment in sectors that are sensitive to economic policy like healthcare, defence and finance. At the macro level it in general decreases investment, employment and output.
Besides economic policy uncertainty, there are also other types of uncertainty that seem to be of influence on the economy. Like the effect of monetary policy uncertainty, that is researched by Husted et al. (2017). The authors constructed an index about monetary policy uncertainty, the MPU, which was done by using the same method as Caldara and Iacoviello (2018) and Baker et al (2016), but then focussed on the perceived public uncertainty about central bank policy. The MPU index uses only American newspapers and focusses on the actions of the Federal Reserve Bank. They showed that an increased monetary policy uncertainty raises credit spreads and lowers economic output.
2.2. Uncertainty and financial markets
Understanding the effect that uncertainty has on the economy, the effect that uncertainty has on financial markets can be analysed. Before reviewing the empirical papers, again the theory behind the relation between uncertainty and financial markets will be explained first.
The effect that uncertainty has on asset prices can be analysed by understanding how asset prices are determined. The pricing of an asset can be explained by the expected future return of the asset (Bank of England, 2018). However, risk averse investors want to be compensated for uncertainty about future returns and demand a risk premium (Berk, DeMarzo and Harford, 2012; and Bank of England, 2018).
Therefore, prices change because of a change in future expected returns, a change in uncertainty about these returns or because of a change in risk aversion. In principle, an increase in geopolitical risks can lead to lower expected returns, higher uncertainty and possibly even higher risk aversion, all of which would lower the price of risky assets such as equities. As only the final price is observed, it is not measured what exactly moves the price of an asset.
9 That uncertainty influences financial markets is also found by Gilchrist et al. (2014). Gilchrist et al. (2014) researched the interaction between financial conditions and uncertainty. Using firm-level stock market data on a high frequency level, they constructed a proxy for uncertainty and discovered that uncertainty is an important determinant of the credit spreads on a firm’s outstanding bonds.
2.3. Geopolitical risk
As already indicated before, there are different types of uncertainty that have an effect on economic performance: geopolitical risk, economic uncertainty and domestic policy uncertainty (Carney, 2016). The focus of this paper will be on geopolitical risk as defined by Caldara and Iacoviello (2018). They defined geopolitical risk as “the risk associated with wars, terrorist acts, and tensions between states that affect the normal and peaceful course of international relations” (2018: 2). The risk of such events as well as their actual realisation are both captured in this definition. These kinds of conflicts may involve armed attacks, but that is not necessary. An oil embargo, for example, also falls under their definition. As the Caldara and Iacoviello (2018) paper is of high relevance for this study, their paper will be discussed in more detail than the other papers reviewed in this section.
Geopolitical risk is by definition different from the other types of uncertainty described in the first section of this chapter. The macro uncertainty shocks investigated by Bloom (2009), diverge from geopolitical risk as it takes events like the assassination of John F. Kennedy into account, which is not considered to affect the “peaceful course of international relations” (Caldara and Iacoviello, 2018:2). Economic policy uncertainty, as is researched by Gupta et al. (2018) and Baker et al. (2016), include circumstances like the failure of Lehman Brothers, which led directly or indirectly to the most recent financial crisis. Again, this event does not have sufficient influence on the relationship between nations to be included in the geopolitical risk measure. The same goes for monetary policy uncertainty, analysed by Husted et al. (2017), for which the actions of the Federal Reserve Bank are important.
2.3.1. Measuring geopolitical risk
10 articles in newspapers that concern geopolitical risk. They used newspapers from Canada, the United States and the United Kingdom, including the Financial Times and The New York Times. Caldara and Iacoviello (2018) state that their GPR can be viewed as a measure of global geopolitical risk or as a measure of geopolitical risk that is mostly relevant from a North-American and British perspective.
Besides making a general index that captures geopolitical risk, Caldara and Iacoviello (2018) also established two sub-indices. One that measures the actual event, like a war or a terroristic attack and one that measures the threat to geopolitical risk. The search terms for acts include words like “terroristic attack” or “air strike” and the terms for threats involve words like “geopolitical uncertainty” or “risk of war”. An entire list of these search terms can be found in Appendix 1.
In order to understand whether their constructed GPR index accurately captures geopolitical risk, Caldara and Iacoviello (2018) conducted an audit. As mentioned in their paper, automated text-searches can raise concerns about bias and accuracy. To see whether the articles were rightfully selected, 16,000 of the selected articles by their index where human read. Additionally, comparisons with external proxies were done. Both confirmed that their index accurately captures geopolitical risk.
2.3.2. Geopolitical risk and the economy
Caldara and Iacoviello (2018) used their index to learn the effects of geopolitical risk on the US economy, the real economy outside the US, stock returns, capital flows and equity markets. They found that high geopolitical risk decreases real activity, it lowers stock returns, depresses stock prices, and capital flows move away from emerging economies to developed economies. They additionally stated that the actual event, like the war or the terroristic attack, is not the only factor of influence. The expectation of such an event, already influences financial markets and business cycles. Threats of geopolitical events produce the largest increase in the level of uncertainty and therefore have the biggest impact (Caldara and Iacoviello, 2018).
2.4. Geopolitical risk and financial markets
After having reviewed that uncertainty has an effect on the economy and financial markets, and understanding the concept of geopolitical risk as is defined by Caldara and Iacoviello (2018), the effects of geopolitical risks on financial markets can be analysed.
11 episodes make energy and equity markets tremble and wanted to understand this in more detail. Kollias et al. (2013) researched the effect that war and terrorism have on the covariance between four stock markets, FTSE100, CAC40, DAX and S&P500, and oil prices, by using a dummy variable in the case of a terroristic attack or a war. War decreases the covariance between stock and oil returns, but terrorist attacks only decrease the covariance between the DAX and the CAC40 and oil returns. The intuition behind it is that in the event of a war in countries like Iraq, oil prices increase because of future supply-concerns and stock returns decrease.
Kollias et al. (2013) found no significant effect of terrorist attacks on the covariance between the FTSE100 and S&P500. The reason for this might be that these markets are less capable in absorbing the impact of an attack. Actual events therefore might have different effects on different financial markets.
Caldara and Iacoviello (2018) also researched the relation between geopolitical risk and financial markets, especially for equity markets as had been highlighted in the former section. They found that in general equity indices decline after a geopolitical shock2.
In general, it can be stated that uncertainty has a negative impact on the macro economy, like economic growth (Baker et al., 2016, Bloom, 2009, Caldara and Iacoviello, 2018, Gupta et al., 2018 and Husted et al., 2017), as well as on financial markets (Caldara and Iacoviello, 2018, Gilchrist et al., 2014 and Kollias et al., 2013). It is expected to find the same negative relation between geopolitical risk and certain equity markets when constructing the GPRB on a daily basis.
However, not all equity indices react the same to geopolitical risk. Sectors like oil, are expected to be positive related to geopolitical risk (Kollias et al., 2013) whereas other sectors, like finance, are expected to be negative related to geopolitical risk (Baker et al., 2016). Therefore, the hypotheses will be the following:
Hypothesis 1: An increase in geopolitical risk has a negative influence on equity markets
Hypothesis 1a: An increase in geopolitical risk will increase the equity markets positively influenced by geopolitical risk
12
Hypothesis 1b: An increase in geopolitical risk will decrease the equity markets negatively influenced by geopolitical risk
Finally, this paper also researches the effect that geopolitical risk has on Dutch equity markets in excess of the world. As the Netherlands is highly financially integrated within the world, it is not expected that the Dutch equities will behave differently towards geopolitical risk compared to the world. For which the hypothesis is:
Hypothesis 2: The Dutch equity indices will not behave differently than the world equity indices
3. Data and methods
In this chapter will be explained how the GPRB index is constructed, what the GPRB index measures, how it should be interpreted and how the GPRB index is related to the GPR index. The financial markets studied will be described in detail within this chapter, as well as the control variables used. Finally, the regressions are explained, as well as how the regressions are estimated.
3.1. The construction of the GPRB index
The geopolitical risk index composed from Bloomberg News, the GPRB, was constructed with the same searching terms as the GPR. These search terms are available in Appendix 1. The GPRB was constructed from Bloomberg News, as Bloomberg News is the news that investors read and react upon. Bloomberg News is accessible via the Bloomberg Terminal, and the news is covered by around 2,400 experienced journalists located all over the world (Bloomberg, 2018).
13 order to control for the fact that on some days there is more news. This counting and dividing was done on a monthly and daily basis.3
Additionally, as well as has been done by Caldara and Iacoviello (2018), two sub-indices were made: one for threats and one for acts. To establish the sub-index for threats, the search terms 1 until 9 were grouped together and divided by the total number of published articles in that specific time period. The same methodology is applicable for the act sub-index, but then the terms 10 until 13 where added together. These sub-indices were also made on a daily and monthly basis.
When researching effects on financial markets by using a variable that is established from news, there is one issue that will be encountered. Financial markets are closed during the weekend, but news is published all week long. News will be published on Saturdays and Sundays when financial markets are closed for which the news that comes out on a Saturday or on a Sunday will have an effect on the financial market on Monday. In to order correct for this problem, the results of the search terms on Saturday and Sunday are added together with the results on Monday. Afterwards, these results are then divided by the total amount of news that has been published on Saturday, Sunday and Monday. This is relevant when estimating the daily regressions. The same method has been applied to the sub-indices.
Controlling for the weekend in this way could lead to a small bias. It could be argued that since there is no other financial news published during the weekend, there could be relatively more geopolitical news during the weekend than during the week. However, at first it is doubtful whether this really is the case, second if it happens it would be an exception and third, if there is relatively more news published this could lead to a higher perception of geopolitical risk for which an increased index is actually justified, for which the potential bias should be small.
Another option could be to leave the weekend out of the analysis, however then the initial problem is encountered again. Therefore, using this method of adding Saturday and Sunday to Monday to take the weekend into account is the preferred method.
3.1.1 Description of the Bloomberg News
Bloomberg News starts at 11/23/1985, and the observation for this thesis ends at 03/20/2018. Figure 3.1 shows the GPRB on a monthly basis. The spikes correlate with the indicated labels.
3Formula: number of articles mentioning search terms Appendix 1 during a specific day or month / total articles
14 Within the Bloomberg Terminal, the news behind the spikes can be shown. Having a look at the articles that cause the index to spike provides information on what the index measures.
The spike in January 1991 corresponds to the Gulf war. Titles on the news are for example “War in the Gulf” and “Confrontation in the Gulf”. The terroristic attack at 9/11 is the reason for the spike in September 2001. News articles like “Bush vows to hunt down terrorists” and “Bush says Bin Laden cannot hide”, can be found during this spike. The highest point of the index is in 2003, during the Iraq war. After 2003, some spikes happen in 2004, 2005 and 2006. In March and April 2004, it is still the worries for Iraq and an increase in the terrorist threat that happen after the Madrid bombing on 03/11/2004, that cause an increase in the index. In March 2011, the Arabian spring causes the index to spike, with many news articles mentioning the country Libya. The spike in September 2014 is due to the escalation of the IS, as well is the spike in 2015 with the terroristic attacks in Paris.
Figure 3.1. Geopolitical risk index from Bloomberg News
15 3.1.2 Comparison GPRB and GPR
The GPR index was first only available on a monthly basis, however by the time writing this paper Caldara and Iacoviello (2018) published a daily GPR. Their daily GPR starts at 01/01/1985 and ends at the 11/30/2017. The daily index constructed in this paper differs with theirs in that it is up-to-date, it is divided into sub-indices, it controls for news published during the weekend, it is based on a real-time news source instead of newspapers and the news source is designed especially for financial markets.
Besides the differences, the GPR and the GPRB should measure the same as they are both constructed using the same search terms. The similarity between both indices can be seen in Figure 3.2. The GPR and the GPRB are normalized to 100 in 1986, because the GPR index has a greater magnitude, and indeed the spikes occur at the same moment in time. Further, note that this normalization of the data is only done to compare the two indices graphically. When using the indices to calculate to correlation, or when using the GPRB in the regressions, the data is not normalized.
The only visible difference between the two indices is the magnitude. The GPR index has a higher magnitude than the GPRB index at some points. This could be the case because the newspapers used to establish the GPR index publish more on geopolitical risk than the Bloomberg News. Which does not imply that the GPR is a better index to use for understanding the relation between geopolitical risk and financial markets. Bloomberg News is still more focussed on financial markets than the newspapers used to construct the GPR and physical newspapers publish news the next day.
Similarity between the indices is important as Caldara and Iacoviello (2018) show that their index is a valuable measure of geopolitical risk by performing audits. A close relation between these two indices would mean that indeed the GPRB is a valuable measure of geopolitical risk as well. Some similarity is already shown in Figure 3.2, yet also the correlation between the two indices is measured.
The correlation between the two indices on a monthly basis is 0.829706 and the daily indices 0.609164. A reason for a lower correlation between the daily indices could be that by design the GPR index could capture the news from yesterday. When there is a peak in the GPRB index, this peak could be visible the next day in the GPR index. Hence, the lower correlation.
16 the GPR. These correlations and the graphs of these indices can be found in Appendix 3 and Appendix 4.
The correlations between the two indices is sufficiently high to be able to conclude that indeed the two indices measure largely the same thing. That the indices are not completely correlated shows that there are still some differences within these indices, which is an indication that constructing a new index makes sense.
Figure 3.2. Comparison GPR and GPRBM on a monthly basis, 1986 = 100
3.2. The financial markets analysed
The financial markets analysed for this paper are equity markets on a world level4 and Dutch5
equity markets. The reason why equity markets are analysed and no other financial markets, is because intuitively equity markets should be more sensitive to geopolitical risk than other financial markets, for example bond markets. Bond markets are more likely to be considered safe markets as for the low probability of governments going bankrupt. Equity markets will also have a more heterogeneous reaction towards geopolitical risk, as not all the equity
4 Equity indices for: oil & gas, oil & gas producers, oil equipment, services & distribution, alternative energy,
basic materials, chemicals, forestry & paper, industrial metals & mining, mining, industrials, construction & materials, aerospace & defence, general industrials, electronic & electrical equipment, industrial engineering, industrial transportation, support services, consumer goods, automobiles & parts, beverages, food producers, household goods & home construction, leisure goods, personal goods, tobacco, health care, health care equipment & services, pharmaceuticals & biotechnology, consumer services, food & drug retailers, general retailers, media, travel & leisure, telecommunications, fixed line telecommunications, mobile telecommunications, utilities, electricity, gas water & multi-utilities, financials, banks, non-life insurance, life insurance, real estate investment & services, real estate investment trusts, financial services, technology, software & computer services, technology hardware & equipment
5 Equity indices for: basic materials, chemicals, industrial metals & mining, industrials, construction & materials,
17 markets will react in the same way. Finally, equity markets have a close connection with the real economy.
Besides these equity indices, also the VIX will be investigated. The VIX index is based on options of the S&P 500 index, an American stock market index, and it serves as a benchmark to determine market expectations of future volatility (CBOE, 2018).
The availability of the headline indices starts from January 1999 for the Dutch headline index and from March 2000 for the world headline index. For the sub-indices, data starts at January 2006 and data on the VIX is available from February 1990. It would be optimal to have data from the start of the GPRB index, however the headline equity indices still encompass the major events of 9/11 and the uncertainty around the Iraq war. After 2006 there are the risks around the terroristic attacks and the Arabian spring, for which the data available for the equity indices should be sufficient to estimate effects of geopolitical risk on financial markets.
The data is derived from Macrobond, a database for economic and financial variables. The index used is the euro’s closing price return of the FTSE index, the Financial Times Stock Exchange index. When analysing the Dutch equities, the FTSE index is used as well, instead of for example the Amsterdam Exchange index, AEX. For the Dutch equity markets the AEX could be a suitable index to analyse. However, as for consistency reasons, the FTSE index is also used when researching the Dutch equity markets. Moreover, having the same index for all equities makes it possible to estimate the effects of the Dutch equity markets relative to the corresponding world equity markets. This will be explained in more detail in Chapter 4.
3.3.Control variables
It could be argued that control variables are not necessary because this information is already in the market as is said by the efficient market hypothesis. According to the efficient market hypothesis, information is directly integrated into the prices of securities, without delay, and therefore all information that is known is fully reflected into the prices (Malkiel, 2003). This would make control variables unnecessary. However, as control variables in this research would reduce the noise within the regressions and the efficient market hypothesis has received various criticism, the control variables described in the next paragraphs are used.
18 Policy Uncertainty6 (EPU) index from Baker et al. (2016) is used as a control variable.
Caldara and Iacoviello (2018) too use this index as a control variable, which is another argument in favour of using the EPU as a control variable. The index is retrieved from Macrobond and is available on a daily and a monthly basis. EPU on a daily basis is only available for the US, but will be used within all the regressions, with the assumption that other world markets also react on the US economy. The EPU has a starting date of 01/01/1985 and is updated on a daily basis. For which the end date of the EPU will be the same as the end date of the GPRB. On a monthly basis, the EPU index is available for specific countries, also for the Netherlands. The EPU indices for the Netherlands and for the US will be used when running the monthly regressions.
Another control variable is the economic surprise index of Citigroup. The economic surprise index, the ESI, measures how often economic data performs better or worse than forecasted and the size of this difference with the forecasting. High-impact and recent data get a higher weight in this index (Mackintosch, 2011). As this paper aims to measure the effect of geopolitical risk, the economic surprise index is a useful control variable because it measures the quality of economic forecasts and therefore eliminates the effect that economic forecast has on financial markets. The index is retrieved from Macrobond, available on a daily and monthly basis, starts at 01/01/2003 and ends at the same time as the GPRB. As with the EPU index, also the economic surprise index is available for various regions. Contrary to the EPU index, the ESI is available for different regions on the three different frequencies. The surprise index for the Euro Area and the United States will be used for the Dutch equity markets. The developed markets and emerging markets index will be used when estimating the regressions on a world level.
Table 3.1 shows a selection of the correlations between the EPU and the ESI. The correlations between the different economic surprise indices (ESI) are not shown here, because there will not be two ESI in one regression, with the exception for the Euro area and US surprise index and the developed markets and emerging markets surprise index. The correlation between these two variables is 0.186970 and 0.492696 respectively, for which they can safely be used within the same regression. The correlation between the variables in Table 3.1 is always <0.1 in absolute value, indicating that multicollinearity between the variables is not a concern.
19
Table 3.1. Correlation between control variables as how they are incorporated into the regressions
Table 3.2. Summary statistics variables
Act (daily) EPU (daily) GPRBD Threat (daily) US surprise index Euro area surprise index Mean 5.94E-06 97.11655 0.006370 0.005611 0.342266 8.383931 Median 2.59E-07 79.89000 0.003217 0.002650 2.300000 10.00000 Maximum 0.000416 719.0700 0.171809 0.162468 97.50000 162.5000 Minimum 0.000000 3.320000 0.000000 0.000000 -1.406.000 -1.886.000 Std. Dev. 1.97E-05 67.35986 0.010256 0.009775 42.84250 55.17382 Skewness 6.551610 2.027886 5.270601 5.645377 -0.442143 -0.458373 Kurtosis 65.39810 10.31680 47.26998 52.52824 2.834731 3.688696 Jarque-Bera 1429601. 25325.77 728283.4 907485.5 134.5086 218.5180 Probability 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000 Sum 0.050170 843457.2 53.76580 47.35970 1365.300 33443.50
Sum Sq. Dev. 3.28E-06 3940235
3
0.887678 0.806405 7319892 12140072
Observations 8440 8685 8440 8440 3989 3989
Table 3.2 shows the summary statistics of the variables. For clarity reasons the most important variables of the economic surprise index are shown.
3.4.The estimated regressions
The equity indices, which are the dependent variables, are regressed upon the GPRB index, the independent variable. This is done in a time series analysis, as each equity will have its own regressions. Panel data would be applicable if all the equity indices would have the same reaction to geopolitical risk. However, as established within the literature review section of
20 this paper, not all financial markets react the same to uncertainty, for which a times series analysis is more suitable for this research.
The estimated models are stated below. The monthly model is different from the daily model in that it does have the EPU available per country. On a daily basis, the EPU index for the US is used, as is explained before in the section of control variables. In the monthly model, for the Dutch equity markets, the EPU index for the Netherlands and the US will be used as control variables and for the world level equity markets the EPU index for the US will be used.
Further, regressions that estimate Dutch equity markets will have the surprise index for the Euro area and the United States as control variable. Equity markets on a world level will be regressed with the surprise indices for developed markets and emerging markets as control variables. The ESI variables are the same for the monthly and the daily model.
The Dutch equity markets are compared with the rest of the world (excess return relative to the world index) within the regressions, for which hypothesis 2 about whether Dutch equity markets react differently than world equity markets can be tested. This is done by subtracting the world headline equity index from the Dutch headline equity index and subtracting the corresponding world equity index from the Dutch equity index. For example, when estimating the effect of geopolitical risk on the Dutch equity constructions and
materials, the world equity construction and materials is subtracted. Also, the different equity
sectors on a world level are compared with the rest of the world, the general world equity index. This is done in order to understand whether different equity markets react differently, in excess of the headline index, to geopolitical risk.
The models are estimated by using heteroscedasticity and autocorrelation consistent (HAC) standard errors, in order to correct for potential bias of heteroscedasticity and autocorrelation within the regressions. Using HAC standard errors helps to avoid the necessity of specifying the exact nature of the auto-correlated error model (Hill et al., 2012).
21 upon their index. It showed that their index captures the geopolitical risk as they defined and not the geopolitical risk from the past that could have an influence on financial markets.
Measuring actual levels of these variables does not give the information needed, as it is the level of geopolitical risk relatively to the period before that matters. A certain level of the GPRB index alone cannot tell whether this is high or low. It has to be measured relatively to what happened before. A stable but high level of geopolitical risk would not influence equity markets, as this level of geopolitical risk is already in the market. However, a sudden increase in geopolitical risk should have an effect on the equity indices, for which it is important to measure the changes and not the actual levels. Therefore, the first difference for GPRB, the EPU and the ESI is taken.
The percentage change is taken for the equity markets as these are non-stationary. Performing the Dickey-Fuller unit root test for the equity markets results in not being able to reject the null-hypothesis of having a unit root, for which the equity markets are said to be nonstationary. It could be dangerous using a time series that is nonstationary in regressions as it can lead to results that seem to be significant, while in reality, the data is unrelated. Regressions with nonstationary variables are called to be spurious regressions (Carter Hill et
al., 2012). The issue of nonstationary series can be solved by taking the difference or the
percentage change of the variable. The percentage change is taken and not the first difference as with the GPRB index as the GPRB index can have values of zero and the equity indices do not have values of zero.
Time-series variables that are nonstationary should not be used within regressions as for the reason described above, with one exception. When both the dependent and independent variables are nonstationary, but the error term is stationary, the variables are cointegrated and they can be used within regressions. Cointegration implies that the dependent and independent variables have a related stochastic trend and as the error term is stationary, they do not deviate too much from each other (Hill et al., 2012). This is not the case within this research as the independent variables are stationary, for which the variables are not cointegrated.
Finally, all the independent variables are included with a lag as for time differences worldwide. News that comes out when a financial market is closed, will influence the financial market the next day.
22 Daily models:
%∆WorldEquityHeadlineIndext = β0 + β1∆GPRBt daily + β2∆GPRBt daily(-1) + β3∆EPUt +
β4∆EPUt (-1) + β5∆ESItemergingmarkets + β6∆ESItemergingmarkets(-1) + β7∆ESItdevelopedmarkets +
β8∆ESItdevelopedmarkets(-1) + ɛt
(%∆DutchEquityHeadlineIndext) - (%∆WorldEquityHeadlineIndext) = β0 + β1∆GPRBtdaily +
β2∆GPRBtdaily(-1) + β3∆EPUt + β4∆EPUt (-1) + β5∆ESIteuroarea + β6∆ESIteuroarea(-1) +
β7∆ESItunitedstates + β8∆ESItunitedstates(-1) + ɛt
(%∆SubEquityIndexWorldt) - (%∆WorldEquityHeadlineIndext) = β0 + β1∆GPRBtdaily +
β2∆GPRBtdaily(-1) + β3∆EPUt + β4∆EPUt (-1) + β5∆ESItemergingmarkets + β6∆ESItemergingmarkets(-1)
+ β7∆ESIdevelopedmarkets + β8∆ESIdevelopedmarkets(-1) + ɛt
(%∆SubEquityIndexDutcht) - (%∆SubEquityIndexWorldt) = β0 + β1∆GPRBtdaily +
β2∆GPRBtdaily(-1) + β3∆EPUt + β4∆EPUt(-1) + β5∆ESIteuroarea + β6∆ESIteuroarea(-1) +
β7∆ESItunitedstates + β8∆ESItunitedstates(-1) + ɛt
Monthly models:
%∆WorldEquityHeadlineIndext = β0 + β1∆GPRBtmonthly + β2∆GPRBtmonthly(-1) +
β3∆EPUtunitedstates + β4∆EPUtunitedstates(-1) + β5∆ESItemergingmarkets + β6∆ESItemergingmarkets(-1) +
β7∆ESItdevelopedmarkets + β8∆ESItdevelopedmarkets(-1) + ɛt
(%∆DutchEquityHeadlineIndext) - (%∆WorldEquityHeadlineIndext) = β0 + β1∆GPRBtmonthly
+ β2∆GPRBtmonthly(-1) + β3∆EPUtnetherlands + β4∆EPUtnetherlands(-1) + β5∆EPUtunitedstates +
β6∆EPUtunitedstates(-1) + β7∆ESIteuroarea + β8∆ESIteuroarea(-1) + β9∆ESItunitedstates +
β10∆ESItunitedstates(-1) + ɛt
(%∆SubEquityIndexWorldt) - (%∆WorldEquityHeadlineIndext) = β0 + β1∆GPRBtmonthly +
β2∆GPRBtmonthly(-1) + β3∆EPUtunitedstates + β4∆EPUtunitedstates(-1) + β5∆ESItemergingmarkets +
23
(%∆SubEquityIndexDutcht) - (%∆SubEquityIndexWorldt) = β0 + β1∆GPRBtmonthly +
β2∆GPRBtmonthly(-1) + β3∆EPUtnetherlands + β4∆EPUtnetherlands(-1) + β5∆EPUtunitedstates +
β6∆EPUtunitedstates(-1) + β7∆ESIteuroarea + β8∆ESIteuroarea(-1) + β9∆ESItunitedstates +
β10∆ESItunitedstates(-1) + ɛt
4. Results
This part of the paper will describe the results from the estimated models. First will be established whether equity markets in general react to geopolitical risk. Thereafter the different equity markets will be shown, which will give insight on which markets react the most towards geopolitical risk. After the different equity markets, the results of the VIX are discussed. The main focus of the results will be on the daily index and the results of the monthly index can be seen as an extra robustness check, whether the results hold at a different frequency. Following the results is the robustness analysis and this chapter will end with a comparison between the GPRB and the GPR index which will help discover whether the GPRB index is indeed an improved measure to estimate the impact of geopolitical risk on financial markets. Only the variable of interest of interest, the GPRB, will be demonstrated. The entire regression results will be available in Appendices 6, 7 and 8.
4.1.Results of the estimated equations
24
Table 4.1. Regression results of general equity indices
Dutch equity excess return World equity GPRB daily -5.71 (2.30)* -7.72 (2.06)** GPRB monthly 62.92 (48.44) -172.10 (47.14)** Threats daily -5.95 (2.30)* -7.91 (2.14)** Threats monthly 65.93 (50.71) -171.68 (47.07)** Acts daily -25257.49 (13634.15) -34868.64 (11518.54)* Acts monthly 443.46 (399.77) -438.68 (548.32) (SD) p>0.05* p>0.01**
Interpreting the results from Table 4.1 shows that on a daily basis the difference of the world headline equity index decreases with 7.72 percentage points when the difference of the GPRB is 1. A difference of the GPRB index by 1 means that GPRBt2 – GPRBt1 = 1, where t indicates time, implying that there is an increase in geopolitical risk. There are more articles published in GPRBt2 about geopolitical risk than there are in GPRBt1. This positive difference in the GPRB index leads to a negative percentage point change in the world headline equity index. This works in the following way, assume that x1 = [(Indext1 – Indext0)/Indext0]*100 and x2 = [(Indext2 – Indext1)/Indext1]*100. The difference in percentage points indicates that x1 exceeds x2 by 7.72 percentage points. For example, if x1 = -10% then x2 = -17.72%. An increase in geopolitical risk therefore decreases the world headline equity index7.
For the Dutch equity market this decrease in difference is 5.71 percentage points in excess of the world index when the difference in the GPRB is 1. Assume that for the Dutch headline equity market x3 = [(Indext2 – Indext1)/Indext1]*100. Then x2 is subtracted from x3 in order to calculate the excess return. As the coefficient is negative and significant, (x3 - x2) = -5.71 percentage points. Assuming that x2 = -17.72%, x3 = -23.43% for which x3 > x2. The
7For example if Index
t0 = 100 and x1 = -10% then x1 = [(Indext1 – 100)/100]*100 = -10% solve for Indext1 gives
Indext1 = 90. If then [(Indext2 – 90)/90]*100 = -17.72% solve for Indext2 gives Indext2 = 72.28, the equity index
in Indext0 = 100, Indext1 = 90 and Indext2 = 72.28 for which indeed the index decreases over time when there is
25 Dutch headline equity index reacts more negatively towards a difference of 1 in the GPRB index than the world headline equity index.
As a change of 1 in the GPRB index is not meaningful, the coefficients can also be explained through the effect of an event. One large increase in the GPRB index happens with the risk of the Iraq War in the beginning of 2003. The peaks related to this event happen during various days, however the first day that the index spikes is taken for this calculation. This day is 31/12/2002, in which the index increases with 0.076 relatively to the day before. Multiplying this with the coefficient of the world headline index gives the result of -0.598.
This implies that if there is an increase in geopolitical risk as big as the Iraq War, the headline world index would decrease with 0.59 percentage points. When calculating the effect on a monthly basis, taking the month December 2002, the index increases on a monthly basis with 0.0191 relatively to the previous month and multiplying this with the coefficient gives the result of -3.299. This indicates that on a monthly basis the headline world index decreases
with 3.29 percentage points. The reason for a larger marginal effect will be explained later in this section.
Both coefficients are significant and negative for which geopolitical risk has a significant negative effect on the general equity market at a world level. The Dutch equity market reacts even more negative to geopolitical risk relative to the world. Looking at the different equity markets in the next section will shed light on where this difference comes from, why the Dutch equity market is more sensitive to geopolitical risk compared to the rest of the world. The coefficients are significant and negative, however when calculating the marginal effect of these coefficients, the economic effect seems to be small. The most important conclusion that can be taken from these coefficients is that geopolitical risk has a significant negative effect on equity markets.
The index on a monthly basis serves as a kind of robustness check, to see whether the results hold at a different frequency. The results for the world equity index remains significant, though it might appear that there is a larger effect on equity markets at a monthly frequency. This difference in magnitude of the coefficients from the world equity index between the daily and monthly basis can be explained through examining how the index is established. If during a specific day there is a lot of geopolitical news, say 10% of all the articles is about geopolitical risk, the daily index will have a high peak during that day. Now assume that it is the only day within that month that there is a lot of geopolitical news. When
26 establishing the monthly index, this day with high geopolitical risk gets averaged out within the month wherefore the monthly index has a lower peak. This happens because the total amount of geopolitical risk related articles is divided by the total number of articles published during the month. The effect of the one day increase will smoothen when taking the total average of the month.
Going back to Appendix 3 gives a graphical explanation on this fact. In Appendix 3 can be seen that the scale of the daily GPRB index is higher than that of the monthly GPRB index. Having a higher scale indicates that the GPRB index has higher peaks. What is invisible in the graph in Appendix 3 are the points where the index moves to zero. This does happen on a daily basis during days on which no geopolitical risk related news is published. Again, these zeroes get averaged out on a monthly basis.
The movement in the equity indices, however, stay the same as the equity indices do not get averaged out on a monthly basis. Running the regression with the monthly index results in a higher coefficient because it seems like less geopolitical news is needed to move the equity indices.
Analysing the regressions with the sub-indices in Table 4.1, shows that the resemblance between the GPRB index and the sub-index threat is remarkable. This similarity could be an indication that the equity indices are mostly influenced through threats, which is in line with the findings of Caldara and Iacoviello (2018). The coefficients of the sub-index acts are on the other side extremely large and, with the exception of the daily world equity, not significant. The world equity on a daily basis reacts very heavily towards geopolitical acts, more than that it does towards threats.
4.1.1. Results per equity index
27
Table 4.2. Results using the daily GPRB
Oil & Gas Oil & Gas Producers Oil Equipment, Services & Distribution Alternative Energy Basic Materials Worl d 7.34 (3.07)* 7.67 (2.93)** 5.89 (5.27) -15.38 (6.78)* -1.63 (2.21) NL 2.55 (4.82) Chemicals Forestry & Paper Industrial Metals & Mining Mining Industrials Worl d -1.63 (1.56) -0.98 (2.68) -4.39 (3.44) 0.43 (4.66) -1.06 (0.98) NL 2.66 (3.80) -3.66 (11.24) -6.12 (4.56) Construction & Materials Aerospace & Defense General Industrials Electronic & Electrical Equipment Industrial Engineering Worl d -3.81 (1.81)* 3.10 (2.47) 1.22 (1.98) -4.31 (2.21) -4.02 (2.28) NL -4.42 (6.00) 2.57 (8.90) Industrial Transportation Support Services Consumer Goods Automobiles & Parts Beverages Worl d -2.58 (1.62) 1.89 (1.67) 0.83 (1.20) -6.01 (3.36) 2.61 (1.92) NL -7.05 (6.63) -19.79 (7.02)** -1.97 (3.59) -4.41 (4.10)
Food Producers Househol
d Goods & Home Construc- tion Leisure Goods
Personal Goods Tobacco
Worl d 3.91 (1.79)* 4.33 (2.07)* -4.55 (4.30) -0.82 (1.58) 7.26 (2.58)** NL 2.09 (4.90)
Health Care Health
Care Equip- ment & Services Pharma- ceuticals & Bio- technology Consumer Services
28 General Retailers Media Travel &
Leisure
Tele-
communications
Fixed Line Tele- communications Worl d 1.99 (2.17) 1.46 (1.69) -1.60 (1.50) 0.48 (1.79) 3.06 (2.29) NL -1.56 (3.81) 4.42 (5.16) 1.91 (5.11) Mobile Tele- communications
Utilities Electricity Gas, Water & Multi-Utilities Financials Worl d -3.25 (2.11) 4.19 (1.95)* 4.54 (2.14)* 3.63 (2.25) -5.18 (1.35)** NL -11.02 (5.71) Banks Non-Life Insurance Life Insurance Real Estate Investment & Services Real Estate Investment Trusts Worl d -7.82 (1.87)** -0.40 (1.52) -8.85 (2.51)** 1.48 (3.74) 4.52 (2.59) NL 4.44 (5.82) -13.47 (5.66)* -8.25 (5.25) Financial Services Tech- nology Software & Computer Services Technology Hardware & Equipment Worl d -3.98 (2.00)* -1.99 (1.87) -1.29 (2.19) -2.35 (2.18) NL -2.20 (5.40) -17.30 (11.44) -1.41 (5.49) (SD) p>0.05* p>0.01**
29 Staying with the oil and gas and oil and gas producers indices, their coefficients are both positive and significant. Meaning that for the oil & gas coefficient a difference of 1 of the GPRB index on a daily basis, leads to an increase of 7.34 percentage point in the percentage change of the oil and gas index and for an increase of 7.67 percentage point in the percentage change of the oil and gas producers index. Implying that when there is increased uncertainty, the equity index for oil and gas and oil and gas producers goes up. The reason for this is logical, as geopolitical risk can have an influence on the supply of oil. If a large oil producing country is on the edge of a war, and its capability of supplying oil is at risk, oil prices will go up which in turn leads to an increase in the index.
The coefficient for alternative energy is negative and significant. A difference of 1 of the GPRB index, leads to a 15.38 percentage point decrease in the percentage change of the
alternative energy index. For which if geopolitical risk goes up, the equity for alternative energy goes down. It is difficult to understand why this index is negative, as when oil prices
rise, one would expect that the substitute also gets more attractive for which this index should rise. A possible explanation could be that due to that alternative energy is very sensitive to policy. Alternative energy projects, like wind mill parks, are investments made by the governments. It could be that when there is an increase in geopolitical risk, governments opt to invest in other sectors like defence instead of alternative energy projects.
Construction and materials also have a significant negative reaction with geopolitical
risk. This is in line with the theory as an increased geopolitical risk leads to a decrease in investment (Bernanke, 1983), which has in turn an effect on construction and materials.
The variable that is not significant, but worthwhile the discussion, is the one of
aerospace and defence. This equity index is not significant, which could be contradictory to
theory as when geopolitical tensions rise, the equity for defence should rise as well. Defence is within this index combined with aerospace, which might be the reason why this coefficient is not significant. Military aerospace might profit from geopolitical tensions, consumer aerospace does not. Caldara and Iacoviello (2018) do split these indices and find indeed that defence reacts significantly positively towards geopolitical risk while aerospace does not. This might be the reason why the coefficient for this index is not significant.
Support services is the only equity index in which the Dutch equity is largely and
significantly different than the index on a world level.
30 Then the tobacco index, which is positive and significant. It may be assumed that geopolitical tensions lead to an increase in the stress and anxiety of people, for which people relieve this stress by starting to smoke more. Therefore, this index could have a positive reaction towards geopolitical risk.
Utilities and electricity also have a positive and significant reaction towards
geopolitical risk. This is in line with the expectation of the alternative energy index. An increase in the oil prices should make the alternative more attractive for which it rises, what does not happen with alternative energy but does with utilities and electricity.
Financials, banks and financial services are all sensitive to geopolitical risk. Financials, banks and financial services are generally highly leveraged. This makes them
riskier than other equities, which explains the reaction of these equities towards geopolitical risk.
Finally, the coefficient of the life insurance index will be discussed. On a world level, it is negative and significant, but the Dutch index reacts even more negatively and significant towards geopolitical risk. Life insurance companies invest a lot in the other equities in the market. When these equities decline, their profits will also decline, for which the reaction of the life insurance index can be stated to be a market effect.
The results show that the first three hypotheses stated in this paper can be accepted. In general, equity indices react negatively towards geopolitical risk and different equity markets react differently to geopolitical risk. What the exact reasons are for the equity indices to behave in a certain way, remain assumptions. What can be said is that certain equity indices behave in a certain way when it comes to geopolitical risk. The final hypothesis, that the Dutch headline index does not behave differently than the world headline index, cannot be accepted. The Dutch headline index does react significantly more towards geopolitical risk. This however, as explained before, is probably due to a composition effect.
4.2.VIX
31 The control variables used for this regression are the same as used for estimating the daily regression for world equity. Like in the other regressions, the first difference is taken for the VIX. The estimated regression for the VIX is therefore:
∆VIXt = β0 + β1∆GPRBtdaily + β2∆GPRBtdaily(-1) + β3∆EPUt + β4∆EPUt(-1) +
β5∆ESItemergingmarkets + β6∆ESItemergingmarkets(-1) + β7∆ESItdevelopedmarkets + β8∆ESItdevelopedmarkets(-1)
+ ɛ
Table 4.3. Results of the regression with the VIX index
Table 4.3 shows the results of the regression with the VIX index as dependent variable. There is a positive significant relation between the VIX and the GPRB index. Implying that part of the movement of the equity markets is indeed due to a change in uncertainty and risk aversion. Appendix 8 shows the entire regressions results.
4.3.Robustness analysis
This section looks deeper into the estimated regressions in order to see how robust the previous findings are. In the former section, there has already been some sort of robustness analysis in the form of the monthly regression results. One of the robustness checks is to split the regression in two, one before the recent financial crisis and one after the financial crisis, to see whether there is a difference between these two split regressions. Another one is checking for any breakpoint within the regression in which the regression starts to behave differently. The final robustness check concerns the outlier within the GPRB index which happens around March 2003. It is important to understand whether the results are different when excluding this outlier as concerns could be that the results are a consequence of this outlier.
4.3.1. Splitting the regression
The breakpoint between the two regressions is the start of the financial crisis, for which the fall of Lehman Brothers at 9/15/2008 is taken (The Economist, 2013). Splitting the regression in two around the financial crisis is executed as it could be a concern that investors react
VIX GPRB daily 11.54
(2.91)**
32 differently towards geopolitical risk because of the financial crisis. Breaking the regression will shed light on whether this concern is justified or not. There is no stated end point of the financial crisis, as the literature remains indecisive on when the financial crisis exactly ended. It could be argued that even ten years later the crisis is still in the minds of the investors wherefore it still can be of influence. Therefore, the regression is only split in two and not three, on before, during and after the financial crisis.
What is important to understand from the two split regressions is whether they are statistically different from each other. Therefore, the Chow breakpoint test is performed, which provides the results stated in Table 4.4 and 4.5. The Chow breakpoint test, tests with an F-test whether there are differences between these two regressions.
Table 4.4. Results of Chow breakpoint test for Dutch equity
Chow Breakpoint Test: 9/15/2008 Dutch equity
Null Hypothesis: No breaks at specified breakpoints Equation Sample: 1/03/2003 3/20/2018 F-statistic 0.792321 Prob. F(9,3950) 0.6235 Log likelihood ratio 7.156928 Prob. Chi-Square(9) 0.6208
Table 4.5. Results of Chow breakpoint test for world equity
Chow Breakpoint Test: 9/15/2008 world equity
Null Hypothesis: No breaks at specified breakpoints Equation Sample: 1/03/2003 3/20/2018 F-statistic 0.796276 Prob. F(9,3950) 0.6198 Log likelihood ratio 7.192615 Prob. Chi-Square(9) 0.6171
Table 4.4 and 4.5 indicate that for both the Dutch and world equity regressions the null hypothesis cannot be rejected for which there is no significant difference between the regressions of before and after the financial crisis. Equity markets did not significantly behave differently towards geopolitical risk during the financial crisis compared to before.
33 hypothesis of no breakpoints cannot be rejected. There is no significant breakpoint at which the regression starts to behave differently.
There is no significant difference between the regressions of before and after the start of the financial crisis, or at any other point within the estimated regression. The regression remains robust for over different time periods, for which the results can be stated to be robust.
4.3.2. Outlier in 2003
As mentioned before, it could be a concern that the previously stated coefficients of the regressions are a result of the outlier in March 2003 in the GPRB index. This outlier is visible in the graphs concerning the GPRB index in the introduction and the data and method section.
As the headline equity indices start only shortly before this outlier10, the regressions
are run from October 2004. The index has already declined from its peak in July 2003, however some margin is taken. As the sub-indices start after 2006, the outlier has no influence on the results from Table 4.2 which contains the coefficients of these sub-indices. Table 4.6 shows the coefficients of the regressions when controlling for the outlier, the entire regression results can be found in Appendix 9.
Table 4.6. Results when controlling for the outlier
Dutch equity excess return World equity GPRB daily -5.70 (2.32)* -7.63 (2.08)** GPRB monthly 76.77 (59.67) -32.50 (47.68) (SD) p>0.05* p>0.01**
The results of the daily index remain almost the same when estimating the regressions from October 2003. These daily coefficients are therefore robust and not influenced by the outlier in March 2003. This is in line with the previous findings on the Chow breakpoint test and the Quandt-Andrew breakpoint test that the regression does not start to behave differently after a certain point.
Different from the daily index, the world equity headline index on a monthly frequency has a lower magnitude and loses its significance when measuring after October 2003. However, relying on the Quandt-Andrew breakpoint test done previously, the
