University of Amsterdam, Amsterdam Business School MSc Finance Quantitative Finance
Master Thesis
The Effects of Political Uncertainty on Default Risk:
Research on the determinants of credit default swaps in the European Financial
markets
Abstract
Financial markets seem to react to political events. Hence, political uncertainty that arises from such events impacts companies in the financial sector. Since the uncertainty contains systematic risk and can therefore not be diversified by hedging, investors seek ways to protect themselves against these risks. Investing in credit default swaps (CDS) provides this protection. We would therefore expect an effect of political uncertainty on CDS spreads. However, existing literature tends to ignore the effects of political uncertainty. This study fills this gap in the literature by testing the effects of political uncertainty on the (change in) the probability of default of 67 individual European firms, as measured by their credit default swap (CDS) spreads. The period January 2009 – May 2018 is examined. In addition, model extensions control for the effects of an exogenous political shock (the Brexit) and country-specific effects. The main contribution of this study is that political uncertainty has a significant effect on the change in default risk, which is robust to different model specifications. Hence, political uncertainty plays a rather important role in the determination of default risk and should be considered in further research in this field.
Keywords: Credit Default Swaps, Default Risk, Political Uncertainty, Brexit Referendum, European Financial Market
Author: Van der Brug, N.
Student number: 10650830
Thesis supervisor: Ladika, T.
Statement of Originality
This document is written by Student Natasja van der Brug who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Table of Contents
List of Tables ... iii
List of Figures ... iv
1. Introduction ... 1
2. Literature review and predictions ... 4
2.1 Political uncertainty and financial markets ... 4
2.2 Measuring political uncertainty ... 5
2.3 The Brexit referendum – a political shock ... 6
2.4 Default risk and its determinants ... 7
2.5 Country-specific analysis in Europe ... 9
3. Data and approach ... 11
3.1 The general approach ... 11
3.2 Probability of default ... 12
3.3 Political uncertainty ... 12
3.4 Three traditional variables ... 13
3.5 Macroeconomic uncertainty ... 15
3.6 The political shock ... 17
3.7 Robustness check ... 18
4. Results ... 20
4.1 Determinants of the change in default risk ... 20
4.2 Political shock on the European financial markets ... 21
4.3 Country-specific behavior after a political shock in Europe ... 25
5. Robustness check ... 29
5.1 Determinants of the change in default risk ... 29
5.2 Political shock on the financial markets ... 30
6. Discussion and conclusion ... 34
Appendix ... 36
List of Tables
Table 1 Determinants of credit default swap spread growth rates over the whole sample period.
p. 21
Table 2 Determinants of credit default swap spreads in Europe, controlled for an exogenous political shock.
p. 23
Table 3 Determinants of credit default swap spreads for subsamples only United Kingdom’ versus ‘without United Kingdom’, controlled for an exogenous political shock.
p. 27
Table 4 Determinants of credit default swap spreads for subsamples peripheral- versus core countries, controlled for an exogenous political shock.
p. 28
Table 5 Determinants of credit default swap spread growth rates over the whole sample period.
p. 30
Table 6 Determinants of credit default swap spread growth rates in Europe, controlled for an exogenous political shock.
p. 32
Table A1 Country-frequency. p. 36
Table A2 Variable definitions. p. 37
Table A3 Dickey-Fuller test for unit root. p. 39
Table A4 Evidence of stochastic trend in ‘ten-year treasury yield’ variable. p. 39
Table A5 Levels- and differences regression analyses summary statistics. p. 40
Table A6 Unstandardized level regression variables summary statistics. p. 40
Table A7 Levels regression approach variable correlations. p. 41
Table A8 Differences regression approach variable correlations. p. 41
Table A9 Determinants of credit default swap spread growth rates subsamples ‘only United Kingdom’ versus ‘without United Kingdom’, controlled for an exogenous political shock.
p. 42
Table A10 Determinants of credit default swap spread growth rates for subsamples peripheral- versus core countries, controlled for an exogenous political shock.
List of Figures
Figure 1 European Economic Policy Uncertainty Index. p.22
Figure A1 The distribution of the countries as used in the country analysis in the last step of the approach.
p. 36
Figure A2 S&P Europe 350. p. 38
Figure A3 Financial macroeconomic uncertainty sigma series p. 38
Figure A4 Total European Gross Domestic Product (GDP). p. 38
1. Introduction
It is a well-established fact that political developments and financial markets are intertwined. Recent examples are the election of United States (US) president Trump, the Brexit referendum and the formation of a new Italian government in May 2018 consisting of two allegedly populist parties. All three events generated much activity at the stock markets and nervous reactions of financial markets. In the case of Trump’s election, stock markets went up, while the other events lead to drops in stock markets. Even though political events clearly have an impact on financial markets, research into the financial sector tends to pay little attention to the effects of political uncertainty. This study contributes to the literature by explicitly testing the effects of political uncertainty on the financial markets.
The main reason why political events can cause shocks to financial markets is that the corresponding political risks are not fully diversifiable (Pástor & Veronesi, 2013). The political uncertainty that arises from political events contains systematic risks, and can therefore not be diversified by hedging. This uncertainty can result from different factors, in which both election results and uncertainties about the policies of new incumbent governments are seen as the two main factors. Mei and Guo (2004) investigated the impact of election cycles on nine financial crises. They discovered that eight out of these nine crises happened during periods of political election and transition. Next to this, the level of political instability and its relationship with default causes doubt on the knowledge about economic policy restructuring (Citron & Nickelsburg, 1987). Thus, the political uncertainty that arises from both elections and around government policy decisions impact financial markets.
However, political uncertainty not only affects profits and cash flows. Research by Balkan (1992) already stressed the importance of including quantified proxies of political events in the assessment of overall country risk. Namely, increase in political risk appears to drive up sovereign credit risk (Belke, Dubova and Osowski, 2018). Hence, also political uncertainty that arises from stability of a country itself can have rather big influence on the sovereign ratings of that country.
While it is thus clear that political events affect aggregate (country level) economic developments as well as stock market fluctuations, existing literature at the level of individual firms tends to ignore the effects of political uncertainty and tends to focus solely on more traditional variables, such as the leverage ratio, volatility and the treasury bond yield. Yet, Pástor and Veronesi (2013) argue that day after day, asset prices seem to react to political news, creating and destroying tremendous amounts of dollars of market value around the world. So, political uncertainty appears to clearly affect financial markets. In this study, I focus on the effects of political uncertainty on the (change in) the probability of default of individual firms, as measured by their credit default swap
(CDS) spreads. As will be argued below in more detail, the CDS spread is a valid measure of the risks of investments in firms. The guiding hypothesis in this study is that political uncertainty increases the risks of investments and will thus have a positive impact on the CDS spreads.
To test this, a period needs to be studied in which there is sufficient variation in political uncertainty. This study focuses on 67 financial sector companies based in 12 European countries over the period January 2009 – May 2018. It is based on monthly data. Pástor and Veronesi (2013) have demonstrated that political uncertainty has become a major concern in Europe since the financial crisis of 2008 resulted in a sovereign debt crisis. Yet, over the nine years, there is a lot of variation in political uncertainty, so that the data have enough leverage to test for the effects. In order to assess whether firms based in different countries were affected in different ways, this study will also focus on country-specific effects.
Next to investigating the general effect of political uncertainty on the change in probability of default, political events per se can be seen as exogenous shocks to the financial markets. Additionally, it is thus important to investigate how the determinants of default risk are affected by such political shocks to the financial markets. During the previously specified sample period, the United Kingdom (UK) voted on leaving the European Union. As the United Kingdom is the first country ever to leave the European Union there is no experience in this field, resulting in rather big amount of uncertainty followed by the vote. Therefore, this study pays particular attention to the shock caused by the outcome of the Brexit referendum, to assess whether this shock exerted an additional effect on top of the effect caused by the increased political uncertainty.
This paper contributes to the extensive empirical literature on change in probability of default and its determinants. It proposes a model in which default risk is analyzed on political uncertainty, while controlling for ‘traditional’ variables at the level of individual firms. In addition, it will pay extra attention to the biggest exogenous political shock on the European financial markets since 2009. Also, the paper contributes by distinguishing between different European regions: the UK-only, core- and peripheral country effects.
The first step in this study is to test whether all control variables in the model (taken from existing literature) exert the expected effects on the CDS spread. As it turns out all determinants exert the expected impact on the CDS spreads, except the overall macroeconomic uncertainty. In line with the predictions the effects of leverage ratio, volatility growth, financial macroeconomic uncertainty are positive, while the ten year treasury yield has a negative effect. In a second step, political uncertainty is added to this model. It turns out to have a highly significant (p<.01) positive relationship with the CDS spreads.
In a third step, a dummy variable is added, which distinguishes the period before and after the Brexit referendum. When this dummy variable is added to the previous model, it has a significant
negative additional effect on CDS spreads. This model controls for the political uncertainty which increased as a results of the Brexit. The negative effect of the Brexit thus shows that financial companies reacted less strongly to the Brexit than one would have expected on the basis of the increased political uncertainty.
In addition, interactions of all determinants with the political shock (Brexit) are added. The results show that most variables have the same effect before as after the Brexit. Yet, the effect of political uncertainty is significantly weaker after the Brexit than before, even though it is still positive. A possible explanation is that the financial sector companies expected the British government to negotiate some arrangement with the European Union that would not be very harmful to the financial sector. This could explain both the weaker effect of political uncertainty after the Brexit as well as its negative main effect.
Finally, two sets of analyses are conducted on subsamples of countries. In the first set of analyses the ‘United Kingdom’ is compared to all other countries in the data set. In the second set of analyses ‘core’- versus ‘peripheral’ countries are compared . The results of both analyses roughly match the earlier findings in the sense that political uncertainty has a robustly significant positive effect on CDS spread in all regions. Yet, also differences emerge. Contrary to expectations, the change in the default risk of financial companies was decreased more in the ‘UK’ subsample after the Brexit than in other European countries. In addition, also financial companies in peripheral countries seem to suffer less in terms of default risk followed by the Brexit relative to companies in core countries. These findings suggest that investors expect financial companies in core countries to be affected more than companies in peripheral countries. Hence, different European countries experience different changes in default risk due to a political shock. The degree of impact depends on the expectations of investors.
Overall, results show that the impact of political uncertainty on the probability of default stayed highly significant in all model extensions. Therefore, the main conclusion of my study is that political uncertainty plays a rather important role in the determination of default risk, and should be considered in further research in this field.
The paper is structured as follows. The second section reviews the existing literature and discusses its implications for the particular research questions of this study, followed by a section about the data and approach employed. Section 4 describes the results, supported by a robustness check in Section 5. In the last section the results are discussed, and the conclusion is drawn.
2. Literature review and predictions 2.1 Political uncertainty and financial markets
The politics and finance view of financial development was a popular topic investigated at the end of the 20th century and beginning of the 21st. It indicates that powerful governments tend to be
incompatible with financial development (Beck, Demirgüç-Kunt & Levine, 2001). So, for financial markets to function properly, limitations on government discretion are required, as in democratic institutions for instance. This statement is supported by Jensen (2008), who showed that political risk is reduced by democratic regimes.
Since politics and financial markets are intertwined, it is not surprising that political uncertainty can affect stock prices and cash flows. Recent outcomes of elections and referenda such as the election of Trump and the Brexit demonstrate the impact of this relationship. In a more systematic way, Mei and Guo (2004) investigated elections in global scale as the source of political uncertainty. They discovered that eight out of these nine financial crises happened during periods of elections and government transitions. The political uncertainty that arises in both circumstances contains systematic risk, and can therefore not be diversified by hedging. For this reason, investors require a risk premium. In a somewhat more recent study about political elections, Julio and Yook (2012) find that political uncertainty leads individual firms to reduce investment expenditures. According to standard economic theory, reduces in investments are directly linked to lower output, thereby reducing employment and individual earnings. Moreover, political uncertainty reduces growth in policy-sensitive sectors as defense, finance, healthcare, and construction. These sectors are important enough to matter at the aggregate level (Baker, Bloom & Davis, 2016). Thus, political processes affect real economic outcomes in general.
Pástor and Veronesi (2013) demonstrated that political uncertainty has become a major concern in Europe since the financial crisis of 2008 resulted in a sovereign debt crisis. They propose a model in which asset pricing implications of political uncertainty are studied, in which the mean of firms profitability is affected by current government policies. Their study assumes that government policy decisions are motivated by both economic and non-economic objectives: policy decision can maximize investors’ welfare but it also takes into account political costs associated with adopting a new policy. The latter are uncertain, which is the source of the political uncertainty in the model proposed by Pástor and Veronesi (2013).
Hence, there are different factors that drive political uncertainty for which investors require a political risk premium. For instance, Pástor and Veronesi (2013) state that investors require a risk premium for uncertainty about outcomes that are purely political events, despite the fact that these political shocks are uncorrelated to economic shocks. They find that during strong economic
conditions political risk premiums are small (see also: Jensen 2008). Thus, in stable democratic regimes, political uncertainty is small and the required risk premium is low.
While it seems plausible that financial markets have always responded to political developments, it may nowadays be more important than previously since financial markets and national economies are increasingly globally intertwined, so that developments in one country easily spread over to the economy of other countries. A recent example is the 2007 US housing bubble, which resulted in a global financial credit crisis. The mortgage backed securities that funded this bubble, was mispriced and the ABX securitization index failed to reveal the growing credit risk (Wachter, 2018). Not only the US suffered the consequences. Also in Europe, the financial crisis of 2008 dramatically increased the cost of market funding for both banks and non-financial firms (Gilchrist & Mojon, 2018). Moreover, according to them the Eurozone has become “the epicenter of world financial stress” (p. 120) after the financial crisis of 2008 escalated into a sovereign debt crisis that began in 2010. Boissel and Derrien and Ors & Thesmar (2017) even called it “the Eurozone crisis of 2008-2012” in their paper.
This has two implications for this study. The first pertains to the time frame, which is the period January 2009 – May 2018. It covers a period starting one year before the sovereign debt crisis of 2010 in Europe until 2018 when the EU’s bail-out program for Greece ended. It thus covers a period in which we will see much variation in political uncertainty. The second implication is that it is important to focus on the financial market, partially because developments in the financial markets have huge implications for the rest of the economy. As uncertainty in general increases risk, political uncertainty is expected to increase the risks of investments and thus will have a positive impact on the change in default risk. The main hypothesis to be tested in this study is therefore:
Hypothesis 1: Political uncertainty increases the risks of investments and thus will have a
positive impact on the CDS spreads.
2.2 Measuring political uncertainty
Finding an adequate measure of political uncertainty has proven to be difficult by previous research conducted in this field, as the uncertainty is driven by many factors. In general several structural- and incidental characteristics influence political uncertainty. The political regime of a country, a structural aspect, impacts overall political risk. While characteristics such as government-type, political events and decisions, are examples of incidental aspects.
The political uncertainty measure in this paper is based on the study of Baker et al. (2016), who developed a new index of economic policy uncertainty (EPU) based on newspaper coverage frequency. In their paper, they solely focused on the United States. However, as their method gained
popularity the authors decided to share their EPU database online for public usage. The measure has now been extended to newspapers from the five European countries with the largest economies (France, the United Kingdom, Germany, Italy and Spain). As discussed in more detail in the methods section, it is based on the frequency of words that indicate economic uncertainties. As argued by Baker et al. (2016), these keywords capture the uncertainty about who will make the economic policy decisions, what actions will be taken and when. The newspapers included are high quality ones, so that it seems plausible that their coverage is a valid representation of the degree of economic uncertainty. A possible drawback is that there is no separate index for each of the countries involved. Yet, due to the global interconnectedness of markets and especially those within the European Union, it seems plausible to assume that the over-time variation is more important than the variation between countries. Moreover, the models to be tested are fixed effects models, which focus on over time variation only.
2.3 The Brexit referendum – a political shock
On Friday the 29th of March 2019 at 11pm the United Kingdom is scheduled to depart from the European Union. Moreover, the United Kingdom is the first country ever to leave the European Union. Hence, there is a lot of uncertainty surrounding the event. The central variable in this study, political uncertainty, might well pick up the effect of the Brexit as this effect is expected to be indirect (through political uncertainty). Yet, because of the exceptional nature of the event, additional analyses will be conducted to assess whether the Brexit had an effect over and beyond the effect already captured by political uncertainty.
The scheduled Brexit undoubtedly impacts the financial markets, and a ‘lose-lose’ situation can be argued (Belke et al., 2018). There are different theories about the consequences of such an event. First, one might look at the foreign investments and trade prospects of the United Kingdom after the Brexit. A recent empirical study by Dhingra, Ottaviano, Sampson and Van Reenen (2016a) concluded that leaving the European Union will reduce the Foreign Direct Investment by around 22% percentage points, which could imply a 3.4% percentage point decrease of real incomes in the United Kingdom. Moreover, the European Union is the United Kingdom’s largest trade partner and it is not known how this relationship is going to change after Brexit. Dhingra and Ottaviano and Sampson, & Van Reenen (2016b) state that this lack of clarity surrounding the trade costs will increase the overall uncertainty.
Second, not only the United Kingdom itself is affected by the Brexit. The expectation is that the Brexit will harm the economies of all EU-countries. A drop in the gross domestic product (GDP) in all European Union countries is expected, which by 2030 would result in an overall European
scenario of an over 5% lower GDP relative to the expected GDP if the Brexit would not happen (Kierzenkowski, Pain, Rusticelli & Zwart, 2016).
Third, Belke, Dubova and Osowski (2018) argue that increase in policy uncertainty itself can affect the economy. Namely, this increase can effect sectors as defense, finance, healthcare, and construction, all important enough to matter at the aggregate level (Baker et al., 2016). Concluded can be that political events influence the overall political uncertainty measure throughout and should not to be overlooked.
2.4 Default risk and its determinants
Political uncertainty typically leads to option value effects as a result of a ‘wait-and-see attitude’ towards investments-type decisions (Belke, Dubova and Osowski, 2018). Standard economic theory states that, in general, investors are risk-averse. They use options to protect themselves against price and volatility fluctuations. In line with this theory, Kelly and Pástor and Veronesi (2016) found that option prices are higher before a political event. Since the uncertainty around a political event can be large, investors take into account the fluctuations in the default risk of individual firms resulting from political uncertainty.
As government bonds become more collateral when a sovereign crisis arises, it would be a logical choice to investigate what drives the change in government bond yield spreads (Boissel et al., 2017). However, usage of credit default swaps (CDS’s) has a significant advantage over the usage of government bonds. There are several arguments to substantiate this, but most importantly CDS’s provide information about firm-level default risk, while government bond yields do not. Also, while CDS spreads are economically comparable to bond yield spread, nevertheless they do not require a benchmark risk-free yield curve and thus do not carry any noise in the determination of this risk-free yield (Ericsson, Jacobs & Oviedo-Helfenberger 2009). In addition, credit default swap rates represent somewhat “fresher” price information than bond yield spreads (Berndt, Douglas, Duffie, Ferguson & Schranz, 2008). Thus, these rates are less likely to be affected by market illiquidity than government bond yield spreads.
Credit default swaps are used by investors as a protection instrument against fluctuations in firm-level default risk. The method has become increasingly popular (Blanco, Brennan & Marsh, 2005), since the swap allows for trading of default risk just like options allow for trading of market risk. Hence, via the swaps one can separate default risk from other sources of uncertainty. Therefore, investigating the determinants of the quoted CDS spreads can provide a useful insight into the sources of default risk (Ericsson et al., 2009).
To investigate the effects of political uncertainty on default risk, this study will therefore focus on the determinants of CDS spread. CDS spread is defined by the price of the debt claim, its
contractual cash flow, and the risk-free rate (Collin-Dufresne, Goldstein & Martin, 2001). Therefore, earlier research has explained CDS spread by three variables, which this study will control for (Gali, Shapir, Amiram & Ben-Zion, 2014; Ericsson et al., 2009; Collin-Dufresne et al., 2001). The first control variable is the leverage ratio. Standard economic theory predicts that a firm will default when the value of its liabilities exceeds the value of its assets (Black & Scholes, 1973; Merton, 1974). Thus, the more leveraged the firm, the higher the probability of default (Ericsson et al. 2009).
Hypothesis 2a: Leverage ratio increases the firm-level (change in) default risk.
Second, changes in equity volatility should be taken into account as they have strong explanatory power in the pricing of credit default swaps (Zhang, Zhou & Zhu, 2005; Tang & Yan, 2016; Kiesel, Kolaric & Schiereck, 2016). The logic behind this is that options are, similar to credit default swaps, used as a protection instrument by investors. Option values increase with volatility, hence credit default swap spreads increase with volatility (Collin-Dufresne et al., 2001).
Hypothesis 2b: Equity volatility increases the firm-level (change in) default risk.
Third, the risk-free rate should be controlled for due to the fact that this benchmark yield varies inversely with corporate yield spreads (Campbell & Taksler, 2003). Also, the level of riskless and the probability of default should have a negative relationship since they contradict each other (Ericsson et al., 2009).
Hypothesis 2c: The risk-free rate decreases the firm-level (change in) default risk.
Uncertainty increases CDS spreads. So, it is important to include different types of uncertainty in the model. According to Tang and Yan (2016) changes in macroeconomic conditions and firm-level fundamentals are important determinants of credit default swaps spread changes. Research by Baum, Chakraborty and Liu (2010) shows that this macroeconomic uncertainty plays an important role in determining both the level and changes of the firm’s leverage. Next to this, macroeconomic uncertainty also affect stock returns and credit risk significantly (Chang, Chen, Gupta & Nguyen, 2015; Beaulieu, Cosset & Essaddam, 2006; Liu & Shu, 2017; Arnold & Vrugt, 2008). Moreover, these significant effects are found after exogenous political shocks on the financial markets. Hence, the model proposed in this paper will control for macroeconomic uncertainty.
Three macroeconomic uncertainty proxies are researched by Baum and Wan (2010); the conditional variance of the GDP growth rate, the index of industrial production and the returns on
the Standard and Poor’s 500 Composite Index. The first measure is constructed to reflect overall macroeconomic uncertainty. The second measure focuses on industrial activity and the last measure on financial market uncertainty. Baum and Wan find strong evidence supporting the fact that macroeconomic uncertainty is an important determinant of credit default swaps.
Since uncertainty increases credit default swap spreads, the change in overall probability of default increases when uncertainty increases. Hence, all uncertainty factors are expected to have a positive relationship with the change in probability of default.
Hypothesis 2d: Financial macroeconomic uncertainty increases the firm-level (change in)
default risk.
Hypothesis 2e: Overall macroeconomic uncertainty increases the firm-level (change in)
default risk.
Furthermore, Section 2.3 explains the impact of a political shock. The effect of a political shock will partially manifest itself through the overall effect of political uncertainty. However, because of the exceptional nature of the event, additional analyses will be conducted to assess whether the Brexit had an effect over and beyond the effect already captured by political uncertainty. Expected is that when controlled for both the political uncertainty- and political shock proxy separately, all political effects that belong to the Brexit are covered by the political uncertainty measure. So, the political effect of the shock is expected to be captured in full by the political uncertainty measure. In addition, the effects of a political shock on the other determinants is investigated. However, no specific expectations can be formulated. In the case of the main effect of Brexit, I expect support for the null-hypothesis.
Hypothesis 3: There is no effect of a political shock on firm-level (change in) default risk, as
this effect is captured by the specified political uncertainty measure.
2.5 Country-specific analysis in Europe
Bloom (2014) states that uncertainty varies heavily across countries, with higher (macro) uncertainty for developing countries. One can state that some countries in Europe might be less developed than other countries. Fontana and Scheicher (2016) provide an example for this. They distinguish between ‘‘core” (AT, BE, DE, FR and NL) and ‘‘peripheral” (GR, IE, IT, PT and SP) countries in Europe. Lately a shift to the core countries has been seen because of the “flight-to-quality” bond trading activity, as a higher bond liquidity reduces both the bond and credit default yield spread (Fontana & Scheicher, 2016). Research by Boissel et al. (2017) further documents this effect. They find empirical evidence of
a strong correlation between repurchase agreement (repo) rates and credit default swap spread. As the repo market for risky bonds if often illiquid and the agreements are usually very short, CDS provide an easy way for investors to short the credit risk (Blanco et al., 2005). This relation is concentrated in the countries that were affected the most in Europe by the recent crises, namely GIIPS countries (Greece, Ireland, Italy, Portugal and Spain). The same relation was not found for other Eurozone countries (Boissel et al., 2017).
Again, political uncertainty plays a central role in these country-specific effects. Moreover, the highest level of political uncertainty measured by the European EPU index was the Brexit vote. The policy uncertainty induced by the Brexit vote resulted in huge spillovers to financial markets all over Europe. Moreover, the concern about the development of the relationship between the United Kingdom and the European Union still causes instability on the financial markets (Belke et al., 2018). One of the main reasons for this effect is that the European Union a central role plays in regulation of its members, so all member-countries will be hit. Expected is however that the GIIPS/peripheral countries will experience a relatively higher impact, due to the fact that their financial stability is in general lower than core countries. Additionally, this effect is strengthened by the fact that the scheduled Brexit raises fears that not only the United Kingdom but other (peripheral) countries will be leaving as well.
Nonetheless, not only the effect of core- versus peripheral countries should be investigated. Namely, expected is that the departure of the United Kingdom is going to effect the United Kingdom the most (Dhingra et al., 2016a); Kierzenkowski et al., 2016; Belke et al., 2018; Bouoiyour & Selmi, 2018).
Hypothesis 4a: The Brexit referendum affects the financial sector in the United Kingdom
more in terms of firm-level (change in) default risk relative to all other European Union countries.
Hypothesis 4b: The Brexit referendum affects the financial sector of peripheral countries
more in terms of firm-level (change in) default risk, when compared to core countries in Europe.
3. Data and approach
The dataset used in this study is obtained from different sources and databases. It contains monthly data on 67 financial sector companies in Europe from January 2009 – May 2018. From the previous section it is clear that next to data on credit default swap spreads, data on the leverage ratio, volatility, 10-year treasury yield, macroeconomic- and political uncertainty is required. This section will start with explaining the approach and main dataset, followed by specification of all the determinants in order of importance and the regressions that will be analyzed. Furthermore, all variable definitions are specified in appendix Table A2.
3.1 The general approach
The analyses in this study are conducted by employing a linear panel regression model on the determinants of the CDS spreads (Ericsson et al. 2009; Belke et al., 2018; Collin-Dufresne et al., 2001; Campbell & Taksler, 2003; Zhang et al., 2005). Existing research established a choice between levels- and differences regressions . Since this paper not only focuses on political uncertainty in general, but also on the effect of a political shock, the main results will be based on level panel regressions. Level regressions report values at a certain point in time, while differences focus on how fast a variable has grown over a certain period. Thus level regressions report better on the effects that emerge for a shock at a certain point in time than (first) differences regressions. However, to check for robustness of the results, also results on the difference regression approach are presented in Section 5.
As the level values of variables can differ in terms of rather high amounts, all level variables are standardized after construction. Also after construction all variables are checked for non-stationarity, by plotting the variables and eyeballing. When a stochastic trend is expected, a Dickey-Fuller unit root test is performed. This test implies regressing the first difference of the variable on itself
∆𝑌𝑡 = 𝛽0+ 𝛿𝑌𝑡−1+ 𝑢𝑡
where ∆𝑌𝑡 is the growth rate of the specific variable in month t. The corresponding null hypothesis and Dickey-Fuller statistic (t ratio) are
𝐻0 ∶ 𝛿 = 0 (𝐷𝐹) 𝑠𝑡𝑎𝑡𝑖𝑠𝑡𝑖𝑐 = 𝛿̂
𝑠𝛿
where “acceptance” of 𝐻0 implies evidence of a stochastic trend, i.e. non-stationarity.
In addition, given that firm-specific return volatility is able to explain about one-third of the variation in bond spreads, the level regression model is estimated with correcting for firm-level
fixed-effects (Baum & Wan, 2010; Campbell & Taksler, 2003). Also, for both the levels and the differences approach, the standard errors are adjusted for clustering at the firm level.
In order to assess whether there are differences between regions in the impact of political uncertainty on CDS spreads, a two-step approach is employed. Within the first step, the general effect of political uncertainty and other determinants on CDS spreads of individual firms are estimated. In the next step, the sample is split in subsamples for a country- and region-specific analyses.
After merging, the total dataset consists of 67 financial sector companies in Europe, corresponding to 7,571 monthly observations. The dataset includes the following countries; Denmark, France, Germany, Ireland, Italy, the Netherlands, Norway, Portugal, Spain, Sweden, Switzerland and the United Kingdom. Exact country frequency is reported in appendix Table A1. Also, the last step of the approach requires a distinction between UK-only, “core”- and “peripheral” countries. The “core ” countries are defined as: Denmark, Germany, France, Ireland, the Netherlands, Norway, Sweden, Switzerland, and the “peripheral” countries: Spain, Italy, Portugal. The exact subsample distribution is presented in appendix Figure A1.
3.2 Probability of default
The data used for the dependent variable, credit default swaps spread, is obtained from the Thomson Reuters CDS database on Datastream. Firm-level 5-year credit default swap spread mid quotes are used because they are the most liquid relative to other credit default swaps.
Immediately after collecting the data, the spread mid quote is winsorized at the 1% level, to account for outliers. As the study will focus on the level regression analysis, all variables are standardized. Next to this, since the results are checked for robustness by a differences analysis, the variable is also converted into a growth factor. All differences are calculated on basis of the following specification
∆𝑆𝑖,𝑡 =
𝑆𝑖,𝑡− 𝑆𝑖,𝑡−1
𝑆𝑖,𝑡−1
where 𝑆𝑖,𝑡 denotes the credit default swap spread in month t for firm i.
3.3 Political uncertainty
Data on the political uncertainty measure based on the existing literature provided by Baker et al. (2016). This dataset can be easily collected from their Economic Policy Uncertainty (EPU) database which can be found online. The corresponding EPU index measure is plotted against the sample
period, which should give clear insight in which political event can be considered as the biggest exogenous political shock on the European financial markets since 2009.
Baker et al. (2016) state that this measure is based on newspaper articles regarding policy uncertainty, in which the terms ‘uncertain’, ‘uncertainty’, ‘economic’, ‘economy’ and some other policy-relevant terms in the native language of the newspapers are counted. The following newspapers are used in the European Economic Policy Uncertainty (EPU) database: Le Monde and Le Figaro for France, Handelsblatt and Frankfurter Allgemeine Zeitung for Germany, Corriere Della Sera and La Repubblica for Italy, El Mundo and El Pais for Spain, and The Times of London and Financial Times for the United Kingdom (Baker et al, 2016). After the obtaining this raw data, Baker et al. (2016) conduct a few steps to obtain the final European EPU index. First of all, after the initial draw, each newspaper-level monthly series is standardized to unit standard deviation prior to 2011. Second, it is equally averaged across all 10 previously mentioned newspapers. Third, prior to 2011 the index is normalized to a mean of 100.
Again, the variable is standardized and converted into a growth rate for the level and difference regression analysis respectively.
3.4 Three traditional variables
The first step in the approach is to investigate the relationship of the determinants on credit default swap spreads. The study includes three ‘traditional’ variables, based on earlier research. This paper will follow the construction approach of Ericsson et al. (2009).
Data on the leverage and equity volatility of the companies is obtained from the Equities database on Datastream. To merge companies from the Equities database and the Thomson Reuter’s CDS database into one dataset a hand-match approach is used, which provided an initial sample of 75 firms and 8,475 observations. Afterwards, all illiquid companies were excluded. A company that reported a zero percentage point return two months after each other was seen as illiquid. Next to this, also companies that defaulted or demerged during the sample period were been excluded. This left the final dataset with 67 financial sector companies corresponding to 7,571 monthly observations.
Leverage ratio
The measure is based on quarterly firm-specific data on leverage and equity values. It is constructed by the following specification
𝐵𝑜𝑜𝑘 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑑𝑒𝑏𝑡 + 𝐵𝑜𝑜𝑘 𝑣𝑎𝑙𝑢𝑒 𝑜𝑓 𝑝𝑟𝑒𝑓𝑒𝑟𝑟𝑒𝑑 𝑒𝑞𝑢𝑖𝑡𝑦
The quarterly values are linearly interpolated to obtain monthly observations for the leverage ratio 𝐿𝐸𝑉𝑖,𝑡 in month t for firm i. Also, the leverage ratio is standardized and converted into a growth rate.
Equity volatility
The equity volatility is estimated using exponentially weighted moving average model (EWMA model) of RiskMetrics. First, the error term is estimated using the following mean-equation specification
𝑟𝑖,𝑡 = 𝜇𝑖+ 𝜀𝑖,𝑡 , 𝜀𝑖,𝑡~𝑊𝑁(0, 𝜎𝑖,𝑡2).
where 𝑟𝑖,𝑡 denotes the log-return in month t for firm i, 𝜇𝑖 denotes average return of firm i over the sample period (Jan 2009 – May 2018), 𝜀𝑖,𝑡 denotes a white noise error term in month t for firm i and 𝜎𝑖,𝑡2 denotes the variance of the error term in month t for firm i.
Second, the variance of the error term is estimated using an EWMA model, specified as
𝜎𝑖,𝑡2 = (1 − 𝜆)𝜀𝑖,𝑡−12 + 𝜆𝜎𝑖,𝑡−12
where 𝜆 = 0.94 restricted by the EWMA model of RiskMetrics; the other parameters are defined as above. To initialize the equity volatility series, the sample standard deviation is used, i.e. the first standard deviation for the series corresponding to firm i is equal to the sample standard deviation of returns of firm i.
Treasury bond yield
Existing literature on determinants of credit default swaps mainly focused on the United States. However, Europe does not consists of multiple countries, which makes the decision about certain variables to include less straightforward, such as which treasury bond yield to use. For this reason data on the 10-year treasury yield is obtained from the European Central Bank, that presents a manufactured spot rate measure based on all Euro area government bonds whose rating is triple A. The choice of only including triple A government bonds depend on the fact that this treasury bond yield is a measure for the risk-free rate.
After collecting the 10-year treasury yield (𝑟𝑡10) from the European Central Bank, the variable is standardized for the level analysis, and converted into a difference measure. As with all level variables after construction, the 10-year treasury yield is checked for non-stationarity. However, when performing the Dickey-Fuller test for unit root, the previously specified null hypothesis in Section 3.1 is not rejected, and there is evidence of a stochastic trend (see appendix Table A3).
Theory states that if one of the independent variables is non-stationary and a linear relationship is estimated, this automatically implies that the dependent variable is non-stationary as
well which would result in incorrect estimations of the model. Therefore, the stationary first difference measure of the 10-year treasury yield is used in the level regression analysis. This statement is supported in appendix Table A4.
3.5 Macroeconomic uncertainty
Next to the traditional variables, data on two macroeconomic factors is added to the dataset. Both measures are constructed based on the research of Baum and Wan (2010), so the data obtained depends on their definitions. Three macroeconomic uncertainty proxies are researched by Baum and Wan (2010), however since this study solely focuses on the financial sector, only the first and the third uncertainty proxies are included in this study: the overall and financial macroeconomic uncertainty.
First, financial market uncertainty data is based on the S&P Europe 350 and obtained from the S&P Dow Jones Indices. Second of all, overall uncertainty is based on the total Gross Domestic Product of Europe. Just as previously specified 10-year treasury yield, this measure is obtained from the European Central Bank database. As the latter represents quarterly data, the measure is linearly interpolated to obtain monthly observations. Both datasets are plotted into histograms, presented in appendix Figure A2 and A4 respectively.
Baum and Wan (2010) stipulate in their approach that the use of the Generalized Auto Regressive Conditional Heteroscedasticity (GARCH) model to compute the two proxies for macroeconomic uncertainty is more appropriate compared to alternative proxies. This paper will follow their believe because general theory established that the GARCH model, which is a generalization of the specified RiskMetrics model, allows for more flexibility relative to other proxies.
However, since the GARCH model is symmetric both a negative as positive shock have the same effect on the estimated implied conditional variance. This will therefore not take into account that a negative effect of a shock can have bigger implication on the conditional variance, than a positive shock of the same size. This so called “leverage effect” should be taken into account (Black, 1976). One extension of the model that takes this news (‘shock’) impact into account is the GJR-GARCH model. Osmari (2017) compared different type of implied volatility models to each other in terms of estimating correct Value at Risk (VaR) forecasts, under which EWMA model, a GARCH-normal, GARCH-Student’s-t , GJR-GARCH-normal and GJR-GARCH-Student’s-t model. He found that the GJR-GARCH-Student’s t approach perform competitively accurate in estimating these VaR forecasts.
Next to the finding of Osmari (2017), both the data collected on the macroeconomic measures provides argumentation on usage of the and GJR-GARCH model. Namely, data on which the financial macroeconomic uncertainty measures is based is skewed to the left, presented in
appendix Figure A2. Hence, a negative news shock has a higher impact than a positive shock of the same size. In addition, as shown in appendix Figure A3, no distribution can be fitted on the data obtained to construct the overall market measure. The latter does however have some outliers to the left.
Therefore, concluded that both measures can be estimated using a GJR-GARCH-Skewed Student’s-t model. Thus this model is estimated twice, once based on S&P Europe 350 data (financial market macroeconomic uncertainty) and once for data on the Gross Domestic Product in Europe (overall macroeconomic uncertainty).
First, the returns (𝑅𝑡+1) should satisfy
𝐸𝑡(𝑅𝑡+1) = 0 𝑉𝑎𝑟𝑡(𝑅𝑡+1) = 𝜎𝑡+12 thus
𝑅𝑡+1 = 𝜇 + 𝜀𝑡+1 , 𝜀𝑡+1~𝑆𝐾𝑇(0, 𝜎𝑖,𝑡2, 𝜔)
where 𝑅𝑡+1 denotes the log-return in month t, 𝜇 denotes average return, 𝜀𝑡 denotes the error term in month t and 𝜎𝑡2 denotes the variance of the error term in month t. Second, the conditional variance of the error term is estimated using a GJR-GARCH-Skewed Student’s-t model, specified as
𝜎𝑡+12 = 𝜔 + 𝛼𝑅𝑡2+ 𝛾𝐼𝑡𝑅2𝑡 + 𝛽𝜎𝑡2, 𝐼𝑡 = ǁ(𝑅𝑡< 0)
where ǁ(𝐴) = 1 if 𝐴 is true, and zero otherwise; and 𝜔 denote the mean. The other parameters are defined as above.
Table A5 in the appendix reports the variable statistics for respectively the levels- and differences analyses. All variables are described in Table A2. As stated in Section 3.4 the (first) difference measure of the 10-year treasury yield is used in the level regression analysis. Next to this, appendix Table A6 reports the level variable statics before standardizing and Tables A7 and A8 present the correlations for both the levels- and differences approach variables respectively.
Based on these specified variables, the first level panel regressions with firm-fixed effects can be estimated. Specified by the following models
(1) 𝑆𝑖,𝑡= α𝑖+ β1i,t𝐿𝐸𝑉𝑖,𝑡+ 𝜀𝑖,𝑡 (2) 𝑆𝑖,𝑡= α𝑖+ β1i,t𝑉𝑂𝐿𝑖,𝑡+ 𝜀𝑖,𝑡 (3) 𝑆𝑖,𝑡= αi+ β1i,t∆𝑟𝑡10+ 𝜀𝑖,𝑡
(4) 𝑆𝑖,𝑡= α𝑖+ β1i,t𝐿𝐸𝑉𝑖,𝑡+ β2i,t∆𝑉𝑂𝐿𝑖,𝑡+ β3i,t𝑟𝑡10+ 𝜀𝑖,𝑡
(5) 𝑆𝑖,𝑡= αi+ β1i,t𝐿𝐸𝑉𝑖,𝑡+ β2i,t∆𝑉𝑂𝐿𝑖,𝑡+ β3i,t𝑟𝑡10+ β4i,t𝐸𝑃𝑈𝑡+ 𝜀𝑖,𝑡
where 𝑆𝑖,𝑡 denotes the CDS spread in month t for firm i, α𝑖 denotes average return of firm i, 𝐿𝐸𝑉𝑖,𝑡 the leverage ratio in month t for firm i, 𝑉𝑂𝐿𝑖,𝑡 denotes the equity volatility in month t for firm i, ∆𝑟𝑡10 denotes the growth in ten-year treasury bond yield in month t, 𝐹𝐼𝑁𝑡 denotes the financial macroeconomic uncertainty in month t, 𝑂𝑉𝑡denotes the overall macroeconomic uncertainty in month t, 𝐸𝑃𝑈𝑡 denotes the political uncertainty in month t, and 𝜀𝑖,𝑡 denotes an error term in month t for firm i.
3.6 The political shock
The last step of the approach is to include the effect of a political shock (the Brexit referendum) into the regression. Also, country-specific interaction with the Brexit referendum is controlled for. A number of dummies are constructed; dummy variable ‘Brexit’ has a value of 1 for the sample period July 2016 – May 2018, otherwise it had a value of 0; dummy variable ‘UK only’ has a value of 1 for the sample period July 2016 – May 2018 conditional on the fact that the firm is traded on the UK market; dummy variable ‘without UK’ has a value of 1 for the sample period July 2016 – May 2018 conditional on the fact that the firm is traded on all financial markets except that of the United Kingdom; dummy variable ‘core’ has a value of 1 for the sample period July 2016 – May 2018 conditional on the fact that the firm is traded on financial markets in Denmark, Germany, France, Ireland, the Netherlands, Norway, Sweden, Switzerland; dummy variable ‘peripheral’ has a value of 1 for the sample period July 2016 – May 2018 conditional on the fact that the firm is traded on financial markets in Spain, Italy, Portugal.
Hence, the following level panel differences regressions with firm-fixed effects can be estimated
(7) 𝑆𝑖,𝑡= αi+ β1i,t𝐿𝐸𝑉𝑖,𝑡+ β2i,t𝑉𝑂𝐿𝑖,𝑡+ β3i,t∆𝑟𝑡10+ β4i,t𝐹𝐼𝑁𝑡+ β5i,t𝑂𝑉𝑡+ β6i,t𝐸𝑃𝑈𝑡+ β7i,t𝐷𝑈𝑀𝑀𝑌𝑖,𝑡+ 𝜀𝑖,𝑡
(8) 𝑆𝑖,𝑡= αi+ β1i,t𝐿𝐸𝑉𝑖,𝑡+ β2i,t𝑉𝑂𝐿𝑖,𝑡+ β3i,t∆𝑟𝑡10+ β4i,t𝐹𝐼𝑁𝑡+ β5i,t𝑂𝑉𝑡+ β6i,t𝐸𝑃𝑈𝑡+ β7i,t𝐷𝑈𝑀𝑀𝑌𝑖,𝑡+ β8i,t𝐸𝑃𝑈𝑡∗ 𝐷𝑈𝑀𝑀𝑌𝑖,𝑡+ 𝜀𝑖,𝑡
(9) 𝑆𝑖,𝑡= αi+ β1i,t𝐿𝐸𝑉𝑖,𝑡+ β2i,t𝑉𝑂𝐿𝑖,𝑡+ β3i,t∆𝑟𝑡10+ β4i,t𝐹𝐼𝑁𝑡+ β5i,t𝑂𝑉𝑡+ β6i,t𝐸𝑃𝑈𝑡+ β7i,t𝐷𝑈𝑀𝑀𝑌𝑖,𝑡+ β8i,t𝐿𝐸𝑉𝑖,𝑡∗ 𝐷𝑈𝑀𝑀𝑌𝑖,𝑡+ β9i,t𝑉𝑂𝐿𝑖,𝑡∗ 𝐷𝑈𝑀𝑀𝑌𝑖,𝑡+ β10i,t∆𝑟𝑡10∗
𝐷𝑈𝑀𝑀𝑌𝑖,𝑡+ β11i,t𝐹𝐼𝑁𝑡∗ 𝐷𝑈𝑀𝑀𝑌𝑖,𝑡+ β12i,t𝑂𝑉𝑡∗ 𝐷𝑈𝑀𝑀𝑌𝑖,𝑡+ β13i,t𝐸𝑃𝑈𝑡∗ 𝐷𝑈𝑀𝑀𝑌𝑖,𝑡 + 𝜀𝑖,𝑡
where 𝐷𝑈𝑀𝑀𝑌𝑖,𝑡 denotes the earlier specified dummies ‘Brexit’, ‘UK only’, ‘without UK’, ‘core’ and ‘peripheral’ in month t for firm i. The other parameters are defined as above.
3.7 Robustness check
Robustness of the results is first of all checked via an event study of the Brexit effect on the stock prices of the sample firms. Second of all, the results of the level regression approach are checked by the panel estimation of the differences approach.
Under the efficient market hypothesis (EMH), stock prices should consist of all relevant information available. Hence, at the time of disclosure of new information the stock prices should immediately react. Therefore, in financial markets an event study can be used to measure the impact of a specific event on the value of a firm (MacKinlay, 1997). In addition to estimating the overall effect of political uncertainty, this study will assess whether the Brexit referendum exerted an additional shock to the financial markets. Since, the referendum took place in June 2016, a long-horizon study (>5 year) cannot be performed. Thus, a short long-horizon event study is performed on the financial firms in the sample, designed to decompose ‘normal return’ on the daily stock prices from ‘abnormal return’.
𝑅𝑖𝑡 = 𝑁𝑅𝑖𝑡+ 𝐴𝑅𝑖𝑡
where normal return is defined as the return expected if the event did not take place; abnormal return is defined as the actual ex post return of the stocks over the ‘event window’ minus the normal return.
The event date is day of the referendum, thus the 23rd of June 2016. It is established that for
a short-horizon event study, usually one month before and after the event is investigated. Hence, the event window consist of observations from the 23rd of May 2016 – 23rd of July 2016. To correctly
estimate the parameters of the model for the normal returns, a subset of the data is necessary before the event. Also, this ‘estimation window’ and event window cannot meet each other. Therefore, the ‘estimation window’ will consist of daily stock price data from 140 days to 60 days before the event; from the 4th of February to the 24th of April.
As stated in the third section, the Fama-French (1993) model cannot be used to estimate the normal returns due to data availability. Although the Capital Asset Pricing Model (CAPM) is, next to the Fama-French (1993) model, the best specifications, it is not wildly used in event studies. This is due to questionable restrictions imposed by the CAPM on the market model (Castro, 2017; MacKinlay, 1997). To avoid this potential for sensitivity the market model is in this study used to estimate the normal return.
𝑁𝑅𝑖𝑡 = 𝛼𝑖+ 𝛽𝑖𝑅𝑚𝑡+ 𝜀𝑖𝑡
where 𝑁𝑅𝑖𝑡 denotes the predicted normal return for firm i on day t; 𝛼𝑖 and 𝛽𝑖 denote unknown parameters to be estimated for each firm i; 𝑅𝑚𝑡 denotes the return of the market (S&P 350 Europe); 𝜀𝑖𝑡 denotes the error term for firm i on day t.
Ordinary Leased Squares (OLS) panel regression is used to estimate the normal returns on observations in the estimation window. Abnormal returns can now be calculated by the following
𝐴𝑅𝑖𝑡 = 𝜀̂𝑖𝑡 = 𝑅𝑖𝑡 − 𝛼̂ − 𝛽𝑖 ̂ 𝑅𝑖 𝑚𝑡
Following the event study set up by MacKinley (1997), a aggregated test on the abnormal returns is computed. First, for each firm the abnormal returns are summed (CAR), averaged (AAR) and cumulated (CAAR)
𝐶𝐴𝑅𝑖 = ∑ 𝐴𝑅𝑖𝑡 𝑡2 𝑡=𝑡1 𝐴𝐴𝑅𝑡 = 1 𝑁∑ 𝐴𝑅𝑖𝑡 𝑁 𝑖=1 𝐶𝐴𝐴𝑅 = ∑ 𝐴𝐴𝑅𝑡 𝑡2 𝑡=𝑡1
where t1 denotes the 23rd of May 2016 and t2 denotes the 23rd of July 2016.
The joint test is conducted on cumulative average abnormal returns, with corresponding null hypothesis and t statistic
𝐻0= 𝐸(𝐶𝐴𝑅𝑖) = 0 𝑇𝑆2 = √𝑁
𝐶𝐴𝐴𝑅
𝑠 ~ 𝑁(0,1)
In the second step of the robustness check, the results of the standardized levels regressions are checked by performing exactly the same regressions on the first differences of the variables. The earlier specified regressions are estimated, only in this case constructed as follows
(10) ∆𝑆𝑖,𝑡 = α𝑖+ β1i,t ∆𝐿𝐸𝑉𝑖,𝑡+ β2i,t∆𝑉𝑂𝐿𝑖,𝑡+ β3i,t∆𝑟𝑡10+ 𝜀𝑖,𝑡
where ∆𝑆𝑖,𝑡 denotes the growth in CDS spread in month t for firm i, α𝑖 denotes average return of firm i, ∆𝐿𝐸𝑉𝑖,𝑡 the growth in leverage ratio in month t for firm i, ∆𝑉𝑂𝐿𝑖,𝑡 denotes the growth in equity volatility in month t for firm i, ∆𝑟𝑡10 denotes the growth in ten-year treasury bond yield in month t.
4. Results
In this section the following guiding question is examined; does political uncertainty have a positive impact on the credit default swap spreads? First, the focus is on the determinants of the change in default risk in financial markets in Europe. Specifically, the importance of adding political uncertainty to the more traditional models is examined. Second, analyses are presented to explore is how the determinants react when an exogenous political shock hits the financial market. Finally, subsample analyses are conducted to distinguish country-specific effects in Europe. Detailed information on all the variables can be found in appendix Table A2. In addition, since all variables are standardized to mean 0 and standard deviation 1, the corresponding tables report standardized coefficients. The latter can be read in the following way; an increase of 1 standard deviation in leverage ratio indicates an 1.224 standard deviation increase in credit default swap spread, ceteris paribus.
4.1 Determinants of the change in default risk
The first step consists of three panel level-panel regressions with firm-fixed effects on the determinants of credit default swap spreads. The results are reported in Table 1. Coefficients are reported with standard errors in parentheses, all standard errors are adjusted for clustering at the firm-level. Bold face is used to illustrate significance at the 1% level or lower.
Three model extensions are estimated; the traditional model by itself or extended with either solely the political- or all uncertainty determinants. This approach is constructed for a clear overview of how the determinants change when different sets of variables are included.
Model 1 is the base model. Traditional variables leverage ratio and volatility are expected to have a positive impact on the CDS spreads, while the ten-year treasury yield growth is expected to have a negative relationship with the spreads. By intuition, the level of riskless and default risk have a paradoxical relationship. Hence, a negative relationship is predicted.
In line with the predictions the both the leverage ratio and volatility report a positive significant sign, which does not change when the political uncertainty measure is added in Model 2. As predicted, the political uncertainty has a positive impact on the change in default risk.
The model is extended with all uncertainty measures in Model 3. Financial macroeconomic uncertainty reports, as expected, a highly significant positive relationship with the CDS spread at the 1% significance level. However, not in line with the expectations are the significant signs of the ten-year yield growth and the overall uncertainty reported in regression 3. Both findings are counterintuitive, namely increases in the level of riskless should by intuition decrease the level of risk. While increase of the overall macroeconomic uncertainty should increase the level of risk.
Additionally, important to note is the clear relationship of CDS spreads with the political uncertainty measure throughout both models. Preliminary conclusions thus indicate that hypothesis
1, 2a, 2b, 2e are supported. Nevertheless, the period investigated corresponds to a rather large amount of variation in political uncertainty. Since, amongst others, political shocks contribute to the level of overall political uncertainty, the findings presented in Table 1 do not provide sufficient evidence to generate assumptions about the behavior of the determinants. Controlling for this will be elaborated in the next section.
Table 1. Determinants of credit default swap spreads over the whole sample period.
The table reports standardized coefficients from level-panel regressions with firm-fixed effects on the determinants of credit default swap spreads (change in probability of default). Columns 1 reports on the traditional variables model. This traditional model is extended with political uncertainty (Column 2) and all uncertainty determinants (Column 3). Included in the sample are 67 financial sector firms in Europe, with observations from January 2009 - May 2018. All variables are defined in Table A2. Standard errors (in parentheses) are adjusted for clustering at the firm level. *** denotes significance at the 1% level (bold face); ** denotes significance at the 5% level; * denotes significance at the 10% level.
Dependent variable: credit default swap spreads
(1) (2) (3) Traditional model Political uncertainty All determinants Leverage ratio 1.224*** (0.345) 1.224*** (0.339) 1.148*** (0.345) Volatility 0.243*** (0.055) 0.265*** (0.053) 0.264*** (0.052) Ten-year treasury yield growth
(standardized) 0.003 (0.003) 0.003 (0.003) 0.010*** (0.003) Financial uncertainty 0.042*** (0.010) Overall uncertainty -0.051*** (0.011) Political uncertainty 0.071*** (0.017) 0.065*** (0.015) Constant 0.012*** (0.002) 0.010*** (0.002) 0.002 (0.003)
Firm fixed effects Yes Yes Yes
Number of observations 6,780 6,780 6,658
Number of firms 67 67 67
Adj. R-squared 0.227 0.237 0.244
4.2 Political shock on the European financial markets
Section 2 implies that different political events that affected the European financial markets since the beginning of 2009. To show this the news index of Economic Policy Uncertainty (EPU) based on newspaper coverage frequency is plotted, presented in Figure 1. Three clear spikes in the level of uncertainty are displayed, corresponding to (1) the European debt crisis, where in May 2011The European Union agreed to a rescue deal for Portugal worth €78 billion followed by a second bailout loan offer for Greece worth €130 billion in October 2011; (2) the Brexit referendum in June 2016; (3) a major election year in Europe in 2017 with elections in the Netherlands, France, Germany and Italy.
0 50 100 150 200 250 300 350 400 450 500 ja n -09 ju l-0 9 ja n -10 ju l-1 0 ja n -11 ju l-1 1 ja n -12 ju l-1 2 ja n -13 ju l-1 3 ja n -14 ju l-1 4 ja n -15 ju l-1 5 ja n -16 ju l-1 6 ja n -17 ju l-1 7 ja n -18 Eu ro p e an N e ws In d e x Date
However, the figure confirms that the Brexit referendum, on the 23th of June 2016, had the highest impact on the political policy uncertainty in Europe since 2009. The Brexit vote is thus used in this paper as a proxy of an exogenous shock on the financial markets, to further investigate how determinants of default risk change.
Figure 1. European Economic Policy Uncertainty Index. The figure plots the European News Index measure from Baker et al.
(2016) corresponding to 113 monthly observations from the period January 2009 to May 2018. In addition, the upwards sloping trend is plotted. The variable statistics are reported in Table A5, and the variable definitions in Table A2.
Continuing, the previous reported model is extended with this political shock proxy. The results of panel regression are reported in Table 2. Again, standardized coefficients are reported with standard errors in parentheses, all standard errors are adjusted for clustering at the firm-level. Bold face is used to illustrate significance at the 1% level or lower.
The first two regressions reported correspond to the last ones reported in Table 1. They denote the model extension controlling for political uncertainty in general and are included in this table for comparison. The regressions in Column 3 and 4 denote the model extension solely controlling for the effect of the Brexit, followed by four model extensions with interaction effects between the Brexit and other predictors of the CDS spread. As stated in the second section, the effect of a political shock is part of the overall effect of political uncertainty. Hence, expected is that the political effect of the shock is entirely captured by the political uncertainty measure, when both the political uncertainty- and political shock proxy are controlled for.
In addition, expected is that a political shock would increase the change in default risk for affected companies. However, results in Column 3 and 4 of Table 2 show that, counterintuitive, the Brexit variable and the CDS spreads are negatively correlated. Evidently, the findings suggests that
the financial sector expected the risks of the Brexit referendum to be much smaller than one would expect on the basis of the increased political uncertainty.
Table 2. Determinants of credit default swap spreads in Europe, controlled for an exogenous political shock.
The table reports standardized coefficients of level-panel regressions with firm-fixed effects on the determinants of credit
default swap spreads (change in probability of default). Columns 1 and 2 denote coefficients of the models controlling for political uncertainty in general. In contrast, Column 3 and 4 report the models controlling for a political event (Brexit referendum). Followed by the interaction effect reported in Columns 5-8. Included in the sample are 67 financial sector firms in Europe, with observations from January 2009 - May 2018. All variables are defined in Table A2. Standard errors (in parentheses) are adjusted for clustering at the firm level. *** denotes significance at the 1% level (bold face); ** denotes significance at the 5% level; * denotes significance at the 10% level.
Dependent variable: credit default swap spreads
(1) (2) (3) (4) (5) (6) (7) (8)
Political uncertainty Political shock Interaction political uncertainty and political shock Leverage ratio 1.224*** (0.339) 1.148*** (0.345) 1.187*** (0.345) 1.117*** (0.350) 1.119*** (0.319) 1.140*** (0.313) 1.048*** (0.323) 1.072*** (0.361) Volatility 0.265*** (0.053) 0.264*** (0.052) 0.226*** (0.051) 0.230*** (0.050) 0.257*** (0.046) 0.255*** (0.045) 0.261*** (0.045) 0.259*** (0.044) Ten-year treasury yield
growth (standardized) 0.003 (0.003) 0.010*** (0.003) 0.002 (0.004) 0.010*** (0.003) 0.004 (0.004) 0.029*** (0.006) 0.003 (0.005) 0.023*** (0.006) Financial uncertainty sigma 0.042*** (0.010) 0.048*** (0.007) -0.011 (0.014) -0.011 (0.015)
Overall uncertainty sigma -0.051***
(0.011) -0.059*** (0.013) -0.024*** (0.007) -0.026*** (0.007) Political uncertainty 0.071*** (0.007) 0.065*** (0.015) 0.226*** (0.048) 0.227*** (0.048) 0.239*** (0.048) 0.238*** (0.048) Brexit -0.153** (0.063) -0.132** (0.063) -0.248** (0.095) -0.322** (0.031) -0.271** (0.106) -0.328*** (0.099) Leverage ratio *Brexit -0.093* (0.050) -0.099* (0.050) Volatility *Brexit -0.214 (0.132) -0.224 (0.135) Ten-year treasury growth
*Brexit -0.032*** (0.007) -0.030*** (0.005) Financial uncertainty *Brexit 0.003 (0.034) Overall uncertainty *Brexit 0.049*** (0.013) Political uncertainty *Brexit -0.193*** (0.044) -0.190*** (0.045) -0.195*** (0.039) -0.200*** (0.044) Constant 0.010*** (0.002) 0.002 (0.003) 0.036*** (0.009) 0.026*** (0.009) 0.088*** (0.018) 0.088*** (0.008) 0.081*** (0.018) 0.081*** (0.017)
Firm fixed effects Yes Yes Yes Yes Yes Yes Yes Yes
Number of observations 6,780 6,658 6,780 6,658 6,780 6,780 6,658 6,658
Number of firms 67 67 67 67 67 67 67 67
Adj. R-squared 0.237 0.244 0.233 0.240 0.276 0.286 0.284 0.296
Corresponding regressions 5-8 show the effects of the political shock on the credit default swap spreads via different sets of determinants. In line with previous finding, leverage ratio and volatility meet their expectations. However, the effects of the ten-year treasury yield and the overall uncertainty measure are much weaker after the Brexit than before. Although financial uncertainty first reported a positive impact at the 1% level, this effect completely disappeared when adding the interaction effect. Apparently financial macroeconomic uncertainty does not drive the default risk
