The risk-taking channel of currency appreciation against the Euro

(1)

The Risk-Taking Channel of

Currency Appreciation against the Euro

Master Thesis, academic year 2015-2016 Faculty of Economics and Business Universiteit van Amsterdam Antonius Tim Bramm, 11084669 Supervisor: Alex J. Clymo, Ph.D. Amsterdam, August 2016

(2)

Statement of Originality

This document is written by the Student Antonius Tim Bramm who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Abstract

Currency appreciation against the Euro seems to follow the theoretical model of Hofmann et al. (2016). By analyzing a set of nine countries which are not part of the Eurozone and have a significant share of Euro denominated debt I show that an increase in the bilateral exchange rate (against the Euro) leads to lower sovereign bond yields. Moreover, after splitting the panel into high and low Euro-debt share countries, the findings demonstrate that having a low share of foreign debt mutes the risk-taking channel and vice versa. Finally, the trade-weighted effective exchange rate that is unrelated to the Euro does not have a significant impact on a country’s Credit Default Swap (CDS) spread if it has a large share of Euro-denominated debt, which is also in line with their model.

JEL codes: G12, G15, G23 

Keywords: bond spread, CDS spread, credit risk, exchange rate.

Acknowledgements: I would like to thank my supervisor Dr. Alex Clymo, my family, my girlfriend Celina, and my friends.

(4)

List of tables

i.

Table 1: Local currency sovereign bond spreads

ii. Table 2: Local currency sovereign bond spreads with vSTOXX iii. Table 3: Local currency sovereign bond spreads with vDAX-NEW iv. Table 4: Local currency sovereign bond spreads with vFTSE

v. Table 5: Sovereign CDS spreads (Euro denominated CDS) vi. Table 6: Sovereign CDS spreads (Dollar denominated CDS) vii. Table 7: High Euro-debt share countries: Sovereign CDS spreads viii. Table 8: Low Euro-debt share countries: Sovereign CDS spreads

ix. Table 9: Sovereign CDS spread (Euro denominated CDS) extra lag for Romanian BER (i.e. BER_t-2)

x. Table 10: Sovereign CDS spread (Euro denominated CDS) no lag for Romanian BER (i.e. BER_t)

xi. Table 11: Sovereign CDS spread (Euro denominated CDS) without Hungary xii. Table 12: High Euro-debt share countries without Hungary

xiii. Table 13: Sovereign CDS spread (Euro denominated CDS) without Dollar CDS but including Norway

xiv. Table 14: Sovereign CDS spread (Euro denominated CDS) without any Dollar CDS xv. Table 15: Summary statistics of Panel Data with Bond Yields as dependent variable: xvi. Table 16: Summary Statistics of Panel Data with CDS spreads as dependent variable: xvii. Table 17: Correlation matrix of High-Euro debt group:

xviii. Table 18: Correlation Matrix of Low-Euro debt group: xix. Table 19: Correlation Matrix of Combined Dataset:

xx. Table 20: Correlation Matrix of Bond Yield based regression: xxi. Table 21: VIF Test for Low-Euro Debt Group

xxii. Table 22: VIF Test for combined CDS Dataset xxiii. Table 23: VIF Test for High-Euro Debt Group xxiv. Table 24: VIF Test for Bond yield Dataset

xxv. Table 25: List of countries’ Euro debt share

(6)

1 Introduction

Are bilateral currency appreciations good or bad for a country’s financial situation? It turns out that the answer is more complicated than the renowned, but alluring, Mundell-Fleming model makes many students think. Exchange rates are an important part of economics, not only because of their retributive nature. A recent example for their gravity can be seen in the actions of central banks after the financial crisis in 2008 and the consecutive sovereign debt crises in Europe. For monetary policymakers these crises meant using both conventional monetary policy tools (i.e. setting the policy rate to the adequate level) and, since the zero lower bound was hit in the recent crisis, also using unconventional monetary policy tools like quantitative easing (QE) or credit easing (CE). For both kinds of monetary policy, a change in the domestic exchange rate is one of the most powerful effects monetary policy can cause and thus use to steer the economy. Nonetheless, exchange rate movements and especially its causes, as well as effects, are not yet fully understood. This thesis aims to take a step towards gaining a better understanding of exchange rate fluctuations and thus improve future monetary policy reactions.

Following Hofmann, Shim, and Shin (2016) (HSS), I conduct an analysis of the effect of the risk-taking channel of currency fluctuations on sovereign yields of government bonds which aims to help deepen our knowledge and understanding of exchange rate fluctuations. Thus, this thesis has the goal to enhance future monetary policy measures by providing insights into one specific, relatively new channel, which was uncovered by Bruno and Shin in (Bruno & Shin, 2015a) and (Bruno & Shin, 2015b) and is in opposition to the well-known net exports channel of currency appreciation (see Mundell-Fleming model (Mundell (1963), Fleming (1962)), as HSS demonstrate in their paper for emerging markets economies (EME) against the US-Dollar. Testing if this antagonistic effect also holds true for exchange rates against the Euro is the key question of this thesis. Hence, this paper aims to understand whether depreciations are expansionary (as per the net-exports channel) or contractionary (as per the risk-taking channel).

The risk-taking channel of currency appreciation itself is a combination of a currency mismatch and a subsequent increase or decrease in the credit supply. For example, a Bulgarian company which sells most of its products or services in Bulgaria has taken up a significant amount of its debt in Euro. If the Bulgarian Lev appreciates against the Euro, the company’s balance sheet improves since liabilities fall relative to assets, and thus creditors (banks) can supply additional loans. The higher amount of investments introduces a decline

(7)

in the risk premium for sovereign bonds since the tail risks (these are risks with a very low probability of realization) of global investors are reduced. Conversely, the risk-taking channel works in the opposite direction if the Bulgarian Lev depreciates against the Euro.

I approach this issue by using the following nine countries, which are not part of the Eurozone (this is a necessary condition to ensure having balance sheet mismatches in the first place): Bulgaria, Croatia, Czech Republic, Denmark, Hungary, Norway, Romania, Sweden, and the UK. The choice is based on the share of Euro-denominated outstanding amounts of issued securities by non-financial corporations (ESA 95 classification). The data sets are from the BIS and can be found in the appendix in Table 25.

Similar to HSS I run a set of monthly country fixed-effects panel regressions, to identify the effect that a change in the nominal exchange rate has on sovereign financial conditions. Thus, I include local currency bond yields, and Euro denominated Credit Default Swaps (CDS) spreads as key dependent variables. Furthermore, I employ three different exchange rates as the main explanatory variables which enable me to distinguish the net-export channel from the risk-taking channel. A positive coefficient for the exchange rates indicates that for a 1% appreciation of the domestic currency against the Euro, the spread between the bond yields (or the CDS spreads) increases. In other words, a positive coefficient indicates that an appreciation has an adverse effect on a country’s financial condition, which would be in line with the net-export channel. However, If the coefficient is negative this supports the existence of the risk-taking channel.

The results show that there is a stark difference in the estimated effects of a change in the bilateral exchange rate on an economy’s financial condition for countries with at least 55% of outstanding debt securities denominated in Euro and those with less than that. This indicates that the underlying mechanism is a balance-sheet mismatch in the first place, which then activates the risk-taking channel. Hence, the higher the countries share of Euro debt is, the more pronounced is the positive effect of an appreciation of the domestic currency against the Euro and vice versa. In that sense, my findings support the three key policy implications that HSS suggest: Limit the currency mismatch in the first place, slow down foreign currency borrowing during an economic boom, and increase the uptake of foreign currency loans during economic downturns.

(8)

The rest of the paper is organized as follows. Section 2 gives an overview of the related literature. Section 3 reviews the underlying theoretical model. Section 4 describes the methodology, the data, the results, and the robustness checks. The final section concludes.

2 Literature Review

This thesis is related to several different parts of the economic literature. First, it adds evidence to the broader macroeconomic impact of currency appreciation/depreciation. There are two contrary arguments: currency appreciation is beneficial for the domestic economy or it is not. Based on the net exports channel in the Mundell-Fleming model, a currency appreciation lowers net exports and thus decreases output. Krugman even argues that a "sudden stop" (of capital inflows) be expansionary under floating exchange rates since the currency depreciates, which would increase net exports (Krugman, 2013). However, Blanchard et al. (2015) use an extended Mundell-Fleming model with two classes of assets (bonds and “non-bonds”) to show that currency appreciation can be expansionary and provide empirical evidence for 19 EME which supports their theoretical model. Furthermore, Plantin and Shin (2016) develop a global game model with floating exchange rates and are able to show that the unique equilibrium replicates the empirical evidence of currency appreciation being expansionary. My results point in the same direction and can, therefore, serve as additional empirical evidence for the positive effects of currency appreciations.

Second, this paper is linked to the literature on global monetary policy spillovers as these are often transmitted via exchange rate appreciations/depreciations and changes in the tail-risk of default. After all, even with a floating exchange rate regime monetary policy shocks from advanced economies (AE) do affect global financial conditions (Rey, 2013). Rey’s findings have important implications since they lead away from the famous "trilemma" of monetary policy, also known as "impossible trinity." It states, that of the three possible objectives, which are: “autonomous monetary policy," "free capital mobility” and a “fixed exchange rate regime," only two can be achieved simultaneously. The attempt of achieving all three objectives at once will lead into a crisis (e.g. Thailand 1994-1996). Hence, Rey argues that the "trilemma" be rather a "dilemma" between independent monetary policy and managing the capital account because even countries with a flexible exchange rate regime are not immune to foreign monetary policy shocks (Rey, 2014).

On the other hand, Obstfeld (2015) argues that the exchange rate regime matters, especially in the short-term since it enables policymakers to, e.g. counter higher international

(9)

interest rates which would have a contractionary effect on domestic output. Furthermore, a flexible exchange rate mechanism serves as protections against currency speculators (which is not only a concern for EME as George Soros showed in 1992 with the British Pound Sterling). However, he points out that monetary independence itself is not improving the available options per se but rather increases the amount of available tools, which becomes more and more important for policymakers when dealing with a deeply connected global financial system. As Tinbergen showed in 1965, each objective (e.g. keeping inflation below but close to two percent) needs a separate instrument (the policy rate), but only if the target needs to be hit exactly. In case there are multiple objectives, e.g. inflation targeting, minimizing unemployment, maximizing output and/or ensuring financial stability, central banks might run out of instruments. Then, the best option is to maximize some kind of utility function subject to several constraints, i.e. finding a trade-off that balances the different objectives out (Fischer, 2010). Adding an open economy setting to this, will lead to even less efficient instruments due to international leakage (see Schoenmaker’s financial trilemma1 (Schoenmaker, 2013)). Thus, solving the monetary trilemma by using flexible exchange rates leads to the financial trilemma. Both, Rey and Obstfeld conclude that if macroprudential policies are not able to address these trade-off problems (which are caused by gross capital flows), one has to take a closer look at the costs and benefits of capital controls. Regarding this discussion, my findings indicate that one should differentiate between countries where corporates have a high level of foreign currency debt and such with a low level or eventually diversified foreign debt position. It seems as if countries that belong to the former group are much stronger affected by foreign monetary policy due to balance sheet effects. Therefore, they could exploit the benefits of capital controls more as they are facing more the dilemma instead of the trilemma.

Another aspect of monetary spillovers is how domestic monetary policy influences the feedback effect of currency appreciations. Hofmann and Takáts (2015) provide evidence of an amplifying effect from lower domestic short-term interest rates in response to appreciation pressures when analyzing the spillovers from US short-term and long-term interest rates on EME and smaller advanced economies. Here again, while an appreciation may have negative effects on net exports and output, being able to quantify the influence of the risk-taking channel on the financial and fiscal position of an economy could help policy makers in their assessment process.

1_{The financial trilemma consists of national control over financial policies, financial integration with the}

global market, and financial stability. Again, only two of the three can be achieved simultaneously.

(10)

Third, this thesis is strongly connected to the strand of literature, which investigates the relation between exchange rates and sovereign risk. Important work in this area has been done by Della Corte et al. (2015) who find that decreasing sovereign risk leads to an appreciation of the currency against the US dollar. Combining this insight with the paper of HSS about the effect of the risk-taking channel of currency appreciation on sovereign yields enables us to explain self-reinforcing feedback loops. At least for cases that are influenced by the US-Dollar. Consequently, this paper improves our knowledge about possible feedback loops caused by changes in both the exchange rate and sovereign risk concerning the Euro.

Fourth, the papers from Bruno and Shin (2015a, 2015b) and Cerutti et al. (2014) have concentrated on the risk-taking channel in the flows of the banking sector. Sobrun and Turner (2015) as well as Feyen et al. (2015) focus on the bond markets and how it is affected by the risk-taking channel, whereas Morais et al. (2015) look at firm- and issuance data in a micro-empirical study setting to provide new insights about the connection between credit supply changes and financial conditions. Baskaya et al. show in a comprehensive study that the VIX and banking inflows have a great impact on the cost of borrowing and the growth rate of domestic loans in Turkey. As mentioned above, I focus on the effects of the risk-taking channel in a more European setting and set the spotlight there on the behavior of sovereign bond investors that are controlled by a tail-risk rule and currency mismatches on enterprises’ balance sheets. This subcategory of an asset-liability mismatch has already been studied by Céspedes et al. (2004) and Krugman (1999), with models that focus on the relaxing effects of currency mismatches on companies’ borrowing constraint.

However, as HSS, this paper aims to disclose the consequences of movements in the tail-risk measure of sovereign bond investors. Thereby induced credit supply fluctuations can explain major portfolio shifts for a limited change in the probability of default.

Most of the literature named above has focused on the exchange rates towards the US dollar or the effect of US policy and less so regarding the Euro, due to the dominant role of the US dollar in global financial matters. Thus, shedding light on the bilateral exchange rates against the Euro, the nominal effective exchange rates, and the implied effects from their movements on sovereign financial conditions is the key difference of this thesis to other papers and this author's main contribution.

(11)

3 Theoretical Model

In order to make sense of the following empirical work in the next section, I sketch out HSS’ proposed theoretical model of global portfolio adjustment in sovereign bonds in the following section. Importantly, instead of the US-Dollar I use the Euro in my econometric analysis and instead of EME, I focus on the nine non-Eurozone countries mentioned above. HSS chose this theoretical setup as it is able to explain the empirical finding that it is the supply of credit, through portfolio flows, which compresses the sovereign yields.

The model consists of four main aspects. Firstly, companies from non-Eurozone economies borrow money from banks denominated in Euro rather than in their domestic currency. Thus, there exists a currency mismatch on their balance sheets (assuming that it is not hedged). It is this mismatch which causes a vulnerability towards changes in the bilateral exchange rate against the Euro. For instance, a depreciation directly increases their probability of default as the relative price of their debt increases. Through tax income or (partly) government owned companies this also affects the governments’ fiscal positions negatively.

Secondly, corporate credit risk is based on the Vasicek (2002) model, which is a generalization of Merton (1974). Crucially, the default probability of the borrowers goes down when the local currency appreciates. Thus, the credit environment is directly affected by a change in the exchange rates. Nonetheless, idiosyncratic shocks can be circumvented by the banks through diversification over many different borrowers.

Thirdly, the credit supply model builds upon a Value-at-Risk constraint for banks and is a basic version of Bruno and Shin’s (2015a) model since it assumes that Euro funding costs are constant rather than determined by the interbank market. The ratio of notional liabilities to notional assets controls the credit supply and is increasing in the exchange rate. Hence, the credit supply (in Euro) rises when the domestic currency appreciates against the Euro, which leads to more investments by the corporate sector.

Fourthly, global sovereign bonds investors (who provide loans to many different countries in either Euro or the local currency) see a lower probability of government default when the number of investments in a country rises. Notwithstanding, they are limited by an economic capital constraint, i.e. a tail-risk measure that constrains their total non-Eurozone sovereign exposures (similar to the Value-at-Risk constraint for the banks). The risk-neutral fund managers find themselves in a position where their binding economic capital constraint forces them to supply more credit when the domestic currency appreciates since this reduces the risk of default, thereby inducing a reduction of sovereign bonds’ yields.

(12)

Concluding, a bilateral exchange rate appreciation of the domestic currency leads to more investments, higher output, increase in tax income, a better fiscal position of the government, a smaller probability of default, more engagement of sovereign bond investors and finally to lower yields on sovereign bonds.

Based on the above theoretical model, I first employ sovereign bond yields (BY), and afterward CDS spreads as the dependent variable. These two financial indicators allow me to test empirically if an appreciation/depreciation of the domestic currency against the Euro has an influence on sovereign yields and if so also quantifying it. Basically, I only test if the first aspect of the theory (a change in the exchange rate, given a currency mismatch on corporates’ balance sheets) causes a change in sovereign bond yields (the last step of the fourth aspect).

Concerning the general movement of sovereign bond yields I use the German 5-year sovereign bonds (denominated in Euro) as a reference point since Germany is considered the safest country in the Eurozone and thereby usually pays the lowest yields on its outstanding debt. That is to say, I employ the spread of the non-Eurozone 5-year sovereign bonds over the 5-year German sovereign bond as the dependent variable. Consequently, the same method applies for the CDS spreads.

Catching the effect of a currency appreciation/depreciation is mathematically equivalent to a change in the exchange rate. Hence, this is my explanatory variable. That being said, I distinguish between three kinds of exchange rates. First of all, I make use of a change in the bilateral exchange rate (BER), e.g. British Pound to Euro or Bulgarian Lev to Euro. Following the theory, only an appreciation (i.e. a positive change in the bilateral exchange rate) should cause a lower spread between Germany and the respective country. The reason for that being the effect of the risk-taking channel.

Secondly, I utilize a change in the nominal effective exchange rate (NEER), i.e. the weighted exchange rate average of a currency against 61 economies. A change in the NEER presents the standard textbook trade-channel effect. Thus, an appreciation of the NEER should have a negative effect on the spread, which means the spread should increase.

The third kind of exchange rate is the orthogonalized change of the NEER, which is the unexplained part (i.e. residual) of a regression from a change in the NEER on a change in the BER. Put differently, it is the change of an exchange rate against all other 61 currencies excluding one, in my case the change of the exchange rate against the Euro. Following HSS I also refer to it as the wedge (between a change in the NEER and BER). The wedge helps in

(13)

clarifying which channel is actually at work, as it filters out the effects of the trade channel which may be overlapped by the influence of the risk-taking channel.

Concluding, these three different exchange rates enable me to discriminate between changes in the BER and NEER when both change simultaneously since they should have adverse effects based on the theoretical model of HSS and the Mundell-Fleming model. The next section explains how they are combined in the different regressions.

4 Empirical Work

4.1 Methodology

To test the theoretical model of HSS, I apply several specifications of a monthly fixed-effects panel regression. In doing so, my focus lies on regressing price based indicators of non-Eurozone sovereign bond market conditions (!) on the log change in the exchange rate (Δ#$) and a group of control variables (%), where I follow HSS’ setup:

!_&,( = *_&+ ,Δ#$_&,(-.+ Γ%_&,(-.+ 0_&,(

Price based indicators are the change in the spread of domestic currency sovereign bond yields over the corresponding German bond yields2_{as well as the change in the spread} of Euro-denominated CDS spreads over the German CDS spread.

Differentiating between the risk-taking channel and the net exports channel is crucial and done by running the regression with five different specifications which are later on represented in the different columns 1) – 5): 1) with the bilateral exchange rate towards the Euro (BER); 2) with the nominal effective exchange rate (NEER); 3) with both the BER and NEER; 4) with the wedge between the NEER and the BER, which is the part of the change in the NEER unrelated to a change in the BER (i.e. the residuals of regressing the change in NEER on the change in BER for each country separately); and 5) with both the BER and the wedge. Hence, column 5 represents the most precise results as it clearly filters out the effects of a change in the NEER from a change in the BER. Importantly, an appreciation of the domestic currency against the Euro reflects an increase in the exchange rate for all the regressions of this paper.

The control variables are the log change in the VIX, the log change in the German consumer price index (CPI), the log change in German industrial production (IP), the change

(14)

in the German short-term interest rate (IR) (3-month money market rate), and the CPI, IP and IR for each of the nine non-Eurozone countries. The fixed effect regression is run with Huber-White robust standard errors.

4.2 Data

The data is mostly taken from the Datastream database. Only a few control variables were not available for the whole period and needed additional data from national statistical offices or national central banks. Table 26 in the appendix describes this in a more detailed manner. The two left-hand-side variables, i.e. the dependent variables, for sovereign bond market conditions are the change in local currency sovereign bond yields and the change in sovereign CDS spreads. The explanatory variables are three different kinds of exchange rates: the change in the BER, the change in the NEER, and orthogonalized change in NEER from the BER. As already stated in Section 4.1, the additional control variables are the log change in the VIX, the log change in the German consumer price index (CPI), the log change in German industrial production (IP), the change in the German short-term interest rate (IR) (3-month money market rate), and the CPI, IP and IR for each of the nine non-Eurozone countries. They are added to control for effects which could simultaneously affect the countries’ financial condition and their exchange rates. Hence, they allow to further distinguish between common factors and the independent effect of a change in the exchange rate (HSS).

Since all variables are either in first-differences or even first logarithmized and then put in first-differences, the data is stationary and has no unit roots. Additionally, it also removes all trends. Logarithmising also helps against having influential outliers. Lastly, all explanatory variables are lagged by one period to tackle the endogeneity problem of reverse causality. As Della Corte et al. (2015) show, a lower CDS spread leads to an appreciation of the bilateral exchange rate. Therefore, all right-hand-side variables are lagged by one month.

The starting date for the bond based regression, as well as the CDS based regression, was set to January 2008 for the former and May 2008 for the later. Both data sets end on April 2016 (included). Since all the data needs to be monthly data, fiscal indicator variables which are mostly published only quarterly are not included. For the bond yields, the exact day of the month varies between countries. The CDS prices are always from either the 13th or 14th_{day of the month. The bilateral exchange rates are mostly end of month rates besides} Romanian rates, which are from the 13th_day.

Nominal effective exchange rates are taken from the BIS database and are monthly averages (broad index). Because real exchange rate movements are mostly driven by nominal exchange rate movements, this paper follows HSS and focuses on nominal exchange rates.

(15)

The NEER, CPI and industrial production (IP) data is normalized to 2010 = 100. Furthermore, I seasonally adjusted the CPI data to take care of seasonal effects; the IP data was already seasonally adjusted. The VIX data is directly from the CBOE website (daily), and I calculated the monthly averages of this data. VSTOXX, VDAX-NEW, and VFTSE are monthly averages taken from Datastream.

Table 15 in the appendix gives the summary statistics of the data used in the bond-yield-based regression. The sample consists of 891 observations, based on nine countries and 99 months. Taking first-differences and first-lag reduces the number of observations to 873 for all control variables and most of the explanatory variables. Since the wedge between a change in the NEER which is unrelated to a change in the BER is calculated separately for each country and then lagged by one period, the number of observations is 882. The domestic interest rate has only 845 observations, due to missing reports (22 out of these 28 missing data points are in Hungary). Eight out of these 28 data points seem to be missing due to the change from ESA 95 to ESA 2010 in November/December 2014 in Bulgaria, Czech Republic, Hungary, and Romania. Thus, they fulfill the criteria of missing completely at random (MCAR), i.e. the reason why the data is missing is independent of the observable and unobservable variables. For that reason, these missing data points can be ignored. However, the other 20 missing values might pose a problem. They may be qualifying for missing at random (MAR) or missing not at random (MNAR). Section 4.3 covers this issue. Overall, column four (standard deviation) of the summary statistics table clearly demonstrates the high within variation which is necessary and beneficial for a fixed effect estimation. The correlation matrix in Table 20 shows low values for most of the variables, only the German industrial production variable being stronger correlated with German short-term interest rates (0.47). All in all, one can conclude that there is no perfect multicollinearity in this data set. After all, the VIF test (Table 24) gives a mean value of 1.48 which is far below the critical value of 10. Also, the Wooldridge test for autocorrelation of the errors in panel data does not reject the null hypothesis of no first-order autocorrelation with a p-value of 0.42.

The second regression uses the change in CDS spreads instead of bond yields as the dependent variable. Table 16 in the appendix illustrates the summary statistics of the data utilized in the CDS based regression. This time, the number of observations is lower (855), stemming from nine countries and a 95 months' period. Taking first-differences and lagging by one period lead to the final number of 837 data points, except for the domestic interest rate which has 810. As before, most of the variance is greater within each country rather than

(16)

between the countries. Table 19 of the appendix depicts the correlation matrix, with all correlation coefficients being similar to the ones from the bond yields based regression. The correlation between CDS and BER is slightly higher than for the bond yields and BER. Vice versa, the correlation with the NEER is marginally lower. Performing the VIF test (appendix Table 22) provides us with a mean value of 1.49. Thus, there is no sign of perfect multicollinearity. Moreover, regarding the serial correlation of the errors the null hypothesis of no first-order autocorrelation can not be rejected with a p-value of 0.22.

The selection of the nine countries was based on their relative share of outstanding Euro-denominated debt of non-financial corporates compared to the total outstanding amount of debt of non-financial corporates. As pointed out in Section 3 above, the effect of the risk-taking channel is more emphasized and only exists, when there is a currency mismatch that leads to a balance sheet effect of a currency appreciation/depreciation in the first place. The respective relative shares of debt are accessible in Table 25 in the appendix.

4.3 Results

The following section presents the main results of my analysis. In order to study the effect of the risk-taking channel of currency appreciations on sovereign yields, a panel data set is analyzed by employing a fixed effect regression. Notably, I use Huber-White robust standard errors instead of the normally utilized cluster-robust standard errors. This comes at the cost of not being immune to autocorrelation in the residuals. But as already mention in Section 4.2, there is no sign for autocorrelation in the two most important regression setups.

4.3.1 Local currency sovereign bond spreads

First, I employ the change in the spread of the five-year local currency sovereign bond yield over the five-year German sovereign bond yield as the dependent variable. The results are represented in Table 1. The BER, the NEER, and the wedge are all significant except for the BER in column three. Looking at the first specification (which is represented in the first column), one sees that just as the theoretical model of HSS predicts, an appreciation of the domestic currency against the Euro lowers the spread, i.e. is beneficial for the non-Eurozone countries. The effect is significant on the 5%-level and can be interpreted as follows: a 1% appreciation of the non-Eurozone currency lowers the domestic currency bond spreads by almost 1.9 basis points. However, unlike to what the model proposes a 1% appreciation of the nominal effective exchange rate leads to a 4.3 basis points smaller spread at a 1%-level of significance. Since the NEER is thought of as being mostly influenced by the trade channel,

(17)

this is at least a surprising result. Furthermore, it remains the same for the combined regression of the BER and NEER. The BER becomes even insignificant. Precisely because of a result as this there are two additional specifications. They help to disentangle the risk-taking channel effect and the trade channel effect.

Turning to column four and five, the results are closer to what one would expect. A significant negative effect of a bilateral currency appreciation and a minuscule negative influence of the orthogonalized change in the NEER (0.04 basis points), though highly significant at the 1%-level.

Throughout all five different setups, the VIX remains significant at the 1%-level, with a positive average effect on the spread of about 0.22 basis points for 1% increase in the VIX (which means the spread is growing when the VIX goes up).

In addition, both the domestic and the German interest rates have a significant influence on the bond spread. Nevertheless, while the domestic short-term interest rate has a positive impact on the bond spread, the coefficient of the German short-term interest rate is negative and has roughly twice the influence than the domestic one in absolute terms (21 bp vs. 12 bp). However, this does not imply that higher German interest rates would lower the bond spread per se. The negative coefficient in Table 1 is the residual effect after controlling for the following three mechanisms: Firstly, a tighter German monetary policy, i.e. a higher ECB policy rate would lead to a depreciation of non-Eurozone countries’ bilateral exchange rate. Furthermore, direct interest rate spillovers would cause higher domestic interest rates (Hofmann & Takáts, 2015). Finally, the VIX would go up if monetary policy is tightened as Bekaert et al. (2013) have shown for the US monetary policy. Hence, a presumably weaker but similar effect would be caused by a German/ECB monetary tightening. Each of those three mechanisms has a positive effect, i.e. the spread would increase.

It may be noted that both the domestic and the German CPI, as well as IP changes, are insignificant in this bond-based regression setting.

As a robustness check, I rerun all five regressions with the vSTOXX, vDAX-NEW, and vFTSE instead of the VIX. Tables 2 – 4 in the appendix show the results. The vSTOXX is insignificant in all five setups; the vFTSE is only weakly significant at the 10%-level in one out of the five specifications, and the vDAX is significant at the 5%-level for the wedge based regression and at the 10%-level for the combined setup with the BER and the wedge. All in all, the other estimated coefficients slightly increase and mostly do not change in their respective level of significance, but the adjusted R-squared drops.

(18)

Table 1: Local currency sovereign bond spreads

Dependent variable: change in 5 year LC spread over German sovereign bonds

(1) (2) (3) (4) (5) ΔlnBERt-1 -1.895** -0.145 -1.984** (0.045) (0.885) (0.039) ΔlnNEER t-1 -4.322*** -4.228*** (0.000) (0.000) orthogΔNEER t-1 -0.0381*** -0.0393*** (0.001) (0.000) ΔlnVIX t-1 0.227*** 0.207*** 0.206** 0.252*** 0.219*** (0.007) (0.009) (0.011) (0.001) (0.007) ΔlnCPI t-1 -1.642 -2.316 -2.274 -2.647 -2.133 (0.658) (0.524) (0.535) (0.471) (0.562) ΔlnIP t-1 0.283 0.266 0.264 0.304 0.263 (0.575) (0.589) (0.591) (0.535) (0.589) ΔIR t-1 0.122** 0.120** 0.121** 0.117* 0.125** (0.046) (0.045) (0.046) (0.071) (0.046) ΔlnCPI-GER t-1 -1.176 0.146 0.113 -0.0523 -0.0972 (0.743) (0.966) (0.974) (0.988) (0.978) ΔlnIP-GER t-1 -1.321 -0.948 -0.955 -1.030 -1.011 (0.160) (0.309) (0.313) (0.287) (0.285) ΔIR-GER t-1 -0.204* -0.205* -0.203* -0.250** -0.213** (0.052) (0.050) (0.050) (0.020) (0.040) Observations 844 844 844 844 844 Adjusted R-squared 0.043 0.062 0.061 0.047 0.055 Countries 9 9 9 9 9

This table reports monthly country fixed-effects panel regressions; the dependent variable is the change in the spread of the 5-year local currency sovereign bond yield over the corresponding German bond yield. ΔBER is the log change in the bilateral exchange rate against the Euro; positive ΔBER is an appreciation of the non-Eurozone currency. ΔNEER is the log change in the nominal effective exchange rate, and orthogΔNEER is the residual from the regression of ΔNEER on ΔBER. ΔVIX is the log change in the VIX index, ΔCPI is the log change in domestic CPI, ΔIP is the log change in domestic industrial production, ΔIR is the change in domestic 3-month money market rate, ΔCPI-GER is the log change in German CPI, ΔIP-GER is the log change in German industrial production, and ΔIR-GER is the change in German 3-month money market rate. p-values in parentheses are calculated based on Huber-White robust standard errors; *** p<0.01, ** p<0.05, * p<0.1

Interestingly, the vDAX comes closest to the VIX and European based vSTOXX is the most far off. Thus, the VIX remains as the first-best choice in the above and the following regressions as a measure for investors’ risk-aversion and the expected stock market volatility.

(19)

The results of Table 1 are a decent start especially the results in column 5, but the adjusted R-square is relatively low, the coefficient of the NEER is contrary to what theory predicts and having the local currency bond spreads as dependent variable does not allow to discriminate successfully between a change in the risk premium and conceivably dissimilarities in the expected path of short-term interest rates. Therefore, the next paragraph uses sovereign CDS spreads as the dependent variable to single out the effect of a shift in the risk premium.

4.3.2 Spread of sovereign CDS spreads

Sovereign CDS spreads are known for being a good gauge of sovereign credit risk since they represent the cost of insuring against a sovereign default. Unlike to sovereign bond yields, they are not directly affected by the expected interest rates. However, they are still not a perfect indicator for local currency credit risk because they are based on either Euro or US-Dollar denominated instruments. Nonetheless, they seem to be a better fit than the local currency sovereign bonds in estimating the influence of the risk-taking channel of currency appreciations.

Table 5 lists the results of the five different regressions with the dependent variable being the change in the spread of non-Eurozone year CDS spread over the German five-year CDS spread. The coefficient for the BER increases in its estimated effect and its significance, which represents a swing towards HSS theoretical predictions. Meanwhile, the NEER is still negative and highly significant at the 1%-level for the second column and significant at the 10%-level for the combined regression with the BER in column three. A small positive hint is that the p-value of the BER coefficient is now lower than the one of the NEER in column three. Column five represents another step towards what HSS have postulated. The BER is highly significant, and a 1% bilateral exchange rate appreciation now decreases the spread by on average 3.2 basis points. However, a change in the NEER which is unrelated to a change in the BER is still slightly negative and significant at the 5%-level.

The VIX remains very significant but its coefficients increased by about 75% compared to Table 1. Presumably illustrating the more direct affect which expected market volatility has on CDS than it has on sovereign bonds itself. The interest rates turned insignificant, whereas the German CPI and German industrial production are now significant at the 5%-level for all five different setups. A 1% increase in German CPI causes a 7.15 basis point lower spread and a +1% change in the German IP gives a 2.5 basis point smaller spread. This is a sign of how economically dependent non-Eurozone countries are from

(20)

Germany and resembles Hofmann et al. results for EME dependence on the US. Finally, the adjusted R-squared more than doubled to 0.12.

So far I have based my calculations on Euro-denominated CDS spreads. However, they have the disadvantage of a lower overall liquidity and eventually less accurate prices compared to US-Dollar denominated CDS. Therefore, Table 6 displays the same regression as before but with US-Dollar denominated CDS spreads as the dependent variable. Column one and two show only a small adjustment, the BER coefficient increases by 0.35 bp and the NEER coefficient rises by 0.12 bp Column three is more interesting: the BER has reached a 1% significance and a higher coefficient (in absolute terms) than the NEER, which is also about to become insignificant with a p-value of 0.098. Then, looking at columns four and five, one sees an insignificant effect of the wedge, where I cannot exclude a value above zero combined with a very significant BER estimate of a 3.6 bp decrease for a 1% appreciation. The VIX is unchanged, so are all domestic indicators. Like before, a change in neither the German interest rate nor in the domestic one is statistically significant in any of the five different setups. A change in the German CPI gained another 2 bp in absolute terms and remained significant. For all that being said, the adjusted R-squared is lower than with Euro-denominated CDS spreads.

(21)

Table 5: Sovereign CDS spreads (Euro denominated CDS)

Dependent variable: change in the spread of 5 year CDS spreads (€)

(1) (2) (3) (4) (5) ΔlnBER t-1 -3.066*** -2.222** -3.235*** (0.000) (0.014) (0.000) ΔlnNEER t-1 -3.498*** -2.098* (0.001) (0.060) orthogΔNEER t-1 -0.0239* -0.0288** (0.096) (0.042) ΔlnVIX t-1 0.393*** 0.402*** 0.382*** 0.441*** 0.389*** (0.000) (0.000) (0.000) (0.000) (0.000) ΔlnCPI t-1 5.706 4.768 5.398 4.635 5.402 (0.177) (0.271) (0.204) (0.283) (0.199) ΔlnIP t-1 -1.021 -1.019 -1.035* -0.950 -0.996 (0.101) (0.108) (0.099) (0.134) (0.114) ΔIR t-1 0.0809 0.0638 0.0775 0.0548 0.0760 (0.231) (0.337) (0.240) (0.415) (0.252) ΔlnCPI-GER t-1 -7.934** -6.823** -7.316** -7.137** -7.161** (0.011) (0.033) (0.019) (0.028) (0.024) ΔlnIP-GER t-1 -2.795** -2.458** -2.594** -2.580** -2.539** (0.013) (0.026) (0.019) (0.024) (0.023) ΔIR-GERt-1 0.217 0.179 0.215 0.147 0.211 (0.113) (0.182) (0.115) (0.277) (0.121) Observations 809 809 809 809 809 Adjusted R-squared 0.111 0.107 0.115 0.093 0.118 Countries 9 9 9 9 9

This table reports monthly country fixed-effects panel regressions; the dependent variable is the change in the spread of the 5-year Euro denominated CDS spread over the corresponding German CDS spread. ΔBER is the log change in the bilateral exchange rate against the Euro; positive ΔBER is an appreciation of the non-Eurozone currency. ΔNEER is the log change in the nominal effective exchange rate, and orthogΔNEER is the residual from the regression of ΔNEER on ΔBER. ΔVIX is the log change in the VIX index, ΔCPI is the log change in domestic CPI, ΔIP is the log change in domestic industrial production, ΔIR is the change in domestic 3-month money market rate, ΔCPI-GER is the log change in German CPI, ΔIP-GER is the log change in German industrial production, and ΔIR-GER is the change in German 3-month money market rate. p-values in parentheses are calculated based on Huber-White robust standard errors; *** p<0.01, ** p<0.05, * p<0.1

(22)

Table 6: Sovereign CDS spreads (Dollar denominated CDS)

Dependent variable: change in the spread of 5 year CDS spreads ($)

(1) (2) (3) (4) (5) ΔlnBER t-1 -3.436*** -2.653*** -3.589*** (0.000) (0.006) (0.000) ΔlnNEER t-1 -3.619*** -1.948* (0.001) (0.098) orthogΔNEER t-1 -0.0205 -0.0260 (0.205) (0.100) ΔlnVIX t-1 0.362*** 0.375*** 0.352*** 0.416*** 0.358*** (0.000) (0.000) (0.000) (0.000) (0.000) ΔlnCPI t-1 5.565 4.527 5.279 4.440 5.292 (0.212) (0.323) (0.237) (0.330) (0.232) ΔlnIP t-1 -0.874 -0.867 -0.887 -0.800 -0.852 (0.166) (0.178) (0.163) (0.215) (0.183) ΔIR t-1 0.0694 0.0500 0.0663 0.0416 0.0651 (0.347) (0.495) (0.359) (0.575) (0.371) ΔlnCPI-GER t-1 -9.803*** -8.640** -9.229*** -9.081** -9.107** (0.006) (0.017) (0.009) (0.012) (0.010) ΔlnIP-GER t-1 -2.573** -2.224* -2.386** -2.388** -2.343** (0.026) (0.052) (0.037) (0.043) (0.042) ΔIR-GERt-1 0.183 0.138 0.180 0.106 0.177 (0.223) (0.347) (0.228) (0.477) (0.238) Observations 809 809 809 809 809 Adjusted R-squared 0.098 0.090 0.101 0.074 0.102 Countries 9 9 9 9 9

This table reports monthly country fixed-effects panel regressions; the dependent variable is the change in the spread of the 5-year Dollar denominated CDS spread over the corresponding German CDS spread. ΔBER is the log change in the bilateral exchange rate against the Euro; positive ΔBER is an appreciation of the non-Eurozone currency. ΔNEER is the log change in the nominal effective exchange rate, and orthogΔNEER is the residual from the regression of ΔNEER on ΔBER. ΔVIX is the log change in the VIX index, ΔCPI is the log change in domestic CPI, ΔIP is the log change in domestic industrial production, ΔIR is the change in domestic 3-month money market rate, ΔCPI-GER is the log change in German CPI, ΔIP-GER is the log change in German industrial production, and ΔIR-GER is the change in German 3-month money market rate. p-values in parentheses are calculated based on Huber-White robust standard errors; *** p<0.01, ** p<0.05, * p<0.1

The results of the regressions with all countries pooled together are mixed. On the one hand, the estimated coefficient of a change in the BER has the expected sign and is about twice as high as what HSS find. On the other hand, the sign of a change in the NEER is not as predicted by the theory. Therefore, to dig deeper into the mechanisms, I now split the sample into high and low share of Euro-debt countries.

(23)

4.3.3

Splitting the panel in two

Sorted by the countries’ respective share of Euro-denominated outstanding securities issued by non-financial corporates the panel is split in two. Four countries (Bulgaria, Czech Republic, Hungary, and Romania) with a share of at least 55% of total outstanding securities combine for the “high Euro-debt share” group. The other five (Croatia, Denmark, Norway, Sweden, and the UK) represent the “low Euro-debt share” group. The resulting outputs are presented in Table 7 and Table 8. At first sight, one can see the great difference in the two distinct groups. The high Euro-debt share countries show a strong negative impact on the spread for an appreciation of their domestic currency against the Euro. A drop in the spread of up to 8.45 basis points for a 1% appreciation is more than twice as high than in the case with all nine countries above. For a change in the NEER regressed on its own I still do not find a positive impact, but in the third column, the NEER is estimated to be insignificant while having a negative sign. Column four and five display a non-significant wedge and a high and very significant BER coefficient. Accordingly, it seems as if having a higher Euro-debt share unlocks a powerful risk-taking channel influence. At the same time, an appreciation against all other currencies is less important for these countries’ fiscal and financial situation.

Besides the exchange rate variables, the estimated influence of the VIX almost doubled. This makes perfect sense since all four countries are small emerging economies and as such highly exposed to market volatility. Similarly, the growth in German IP remains as a significant variable. For all regressions that include the BER the change in the German interest rate is significant at the 5%-level. German inflation movements are hardly significant anymore.

Table 7: High Euro-debt share countries: Sovereign CDS spreads

Dependent variable: change in 5 year CDS spread (€)

(1) (2) (3) (4) (5) ΔlnBER t-1 -8.448*** -7.615*** -8.393*** (0.000) (0.000) (0.000) ΔlnNEER t-1 -5.849*** -1.673 (0.001) (0.366) orthogΔNEER t-1 -0.0212 -0.0142 (0.449) (0.590) ΔlnVIX t-1 0.595*** 0.633*** 0.586*** 0.696*** 0.591*** (0.002) (0.001) (0.002) (0.000) (0.001) ΔlnCPI t-1 6.926 5.800 6.752 5.916 6.774 (0.241) (0.351) (0.255) (0.341) (0.247)

(24)

ΔlnIP t-1 -1.330 -1.228 -1.359 -0.976 -1.299 (0.407) (0.454) (0.398) (0.555) (0.424) ΔIR t-1 0.111 0.0527 0.104 0.0545 0.107 (0.243) (0.576) (0.267) (0.566) (0.257) ΔlnCPI-GER t-1 -9.796* -9.330 -9.201 -11.86* -9.546 (0.091) (0.136) (0.118) (0.054) (0.105) ΔlnIP-GER t-1 -3.943* -4.508** -3.825* -5.262** -3.837* (0.072) (0.045) (0.079) (0.029) (0.082) ΔIR-GERt-1 0.541** 0.411 0.538** 0.349 0.532** (0.037) (0.105) (0.037) (0.180) (0.040) Observations 345 345 345 345 345 Adjusted R-squared 0.220 0.172 0.220 0.138 0.218 Countries 4 4 4 4 4

This table reports monthly country fixed-effects panel regressions of countries with a high level of Euro debt; the dependent variable is the change in the spread of the 5-year Euro denominated CDS spread over the corresponding German CDS spread. ΔBER is the log change in the bilateral exchange rate against the Euro; positive ΔBER is an appreciation of the non-Eurozone currency. ΔNEER is the log change in the nominal effective exchange rate, and orthogΔNEER is the residual from the regression of ΔNEER on ΔBER. ΔVIX is the log change in the VIX index, ΔCPI is the log change in domestic CPI, ΔIP is the log change in domestic industrial production,

ΔIR is the change in domestic 3-month money market rate, ΔCPI-GER is the log change in German CPI, Δ IP-GER is the log change in German industrial production, and ΔIR-GER is the change in German 3-month money market rate. p-values in parentheses are calculated based on Huber-White robust standard errors; *** p<0.01, ** p<0.05, * p<0.1

The low Euro-debt share countries stand in stark contrast to these results. The BER is insignificant, and it even has a positive sign in column one and three. This lack of significance can be seen as evidence of a small imprint of the risk-taking channel due to a lower share of Euro denominated debt in these economies. Hence, the risk-taking channel simply does not unfold itself because investors’ capital constraint is unchanged or at most hardly changed.

The wedge is now significant at the 5%-level when taken by itself and also for the regression with the BER, contrasting with its insignificance in Table 7. For a 1% appreciation of the NEER that is unrelated to an appreciation in the BER, the spread decreases by 0.022 bp on average. On the one hand, this is not what one would expect based on the trade channel effects. On the contrary, the estimated average coefficient is close to zero and the confidence interval also covers positive values. Also, the influence of the VIX on the spread is estimated to be smaller again. It almost drops back to the level of the bond-based regression in Table 1. Naturally, advanced economies are more robust towards economic shocks and thus also less affected by expected market volatility. Generally, the adjusted R-squared is very low (0.04) for this sample, i.e. the model cannot explain a lot of the dataset’s

(25)

variance. Vice versa for the high Euro-debt share sample, where the adjusted R-square tops out at 0.22.

Table 8: Low Euro-debt share countries: Sovereign CDS spreads

Dependent variable: change in 5 year CDS spread (€)

(1) (2) (3) (4) (5) ΔlnBER t-1 0.193 0.523 -0.0559 (0.615) (0.210) (0.876) ΔlnNEER t-1 -0.582 -0.947 (0.417) (0.238) orthogΔNEER t-1 -0.0221* -0.0223* (0.084) (0.081) ΔlnVIX t-1 0.254*** 0.244*** 0.250*** 0.255*** 0.254*** (0.004) (0.005) (0.005) (0.003) (0.003) ΔlnCPI t-1 4.281 4.405 4.096 3.993 4.024 (0.437) (0.421) (0.454) (0.461) (0.460) ΔlnIP t-1 -0.554 -0.563 -0.555 -0.555 -0.556 (0.124) (0.120) (0.125) (0.127) (0.127) ΔIR t-1 0.0319 0.0362 0.0336 0.0309 0.0314 (0.582) (0.531) (0.565) (0.588) (0.591) ΔlnCPI-GER t-1 -4.596 -4.616 -4.381 -3.886 -3.901 (0.204) (0.195) (0.220) (0.281) (0.283) ΔlnIP-GER t-1 -0.784 -0.778 -0.691 -0.610 -0.615 (0.427) (0.428) (0.482) (0.515) (0.516) ΔIR-GERt-1 0.0180 0.0244 0.0162 0.0163 0.0173 (0.891) (0.856) (0.902) (0.902) (0.895) Observations 464 464 464 464 464 Adjusted R-squared 0.035 0.036 0.035 0.045 0.043 Countries 5 5 5 5 5

This table reports monthly country fixed-effects panel regressions of countries with a low level of Euro debt; the dependent variable is the change in the spread of the 5-year Euro denominated CDS spread over the corresponding German CDS spread. ΔBER is the log change in the bilateral exchange rate against the Euro; positive ΔBER is an appreciation of the non-Eurozone currency. ΔNEER is the log change in the nominal effective exchange rate, and orthogΔNEER is the residual from the regression of ΔNEER on ΔBER. ΔVIX is the log change in the VIX index, ΔCPI is the log change in domestic CPI, ΔIP is the log change in domestic industrial production,

ΔIR is the change in domestic 3-month money market rate, ΔCPI-GER is the log change in German CPI, Δ IP-GER is the log change in German industrial production, and ΔIR-GER is the change in German 3-month money market rate. p-values in parentheses are calculated based on Huber-White robust standard errors; *** p<0.01, ** p<0.05, * p<0.1

(26)

4.4 Robustness checks

For the US-Dollar denominated CDS regression, the null hypothesis of no first-order autocorrelation is denied on the 5%-level. This does potentially lead to a downward bias in the estimated standard errors, i.e. the p-value is underestimated.

Another case of a first-order autocorrelation is the low Euro-debt share country sample. Here, the null hypothesis is denied on a 1%-level. Nevertheless, all but the VIX coefficient are insignificant even with the potential downward bias in the standard errors. Hence, autocorrelation is not a problem in this particular case.

Running the CDS-based regressions without lagged explanatory variables generates highly significant estimates for the NEER and an insignificant or less significant impact of a change in the BER. This supports the approach of taking one-period lags in the first place to overcome endogeneity problems.

Furthermore, I test if a lagging of the Romanian BER by an additional month does have an influence on the estimated coefficients. The goal of this test is to verify that although the Romanian BER is measured at the middle of each month (all the other countries BER are end-of-month rates), it is still outperforming an additional lag. Table 9 in the appendix supports the one-period lag version over a two-period lag. The estimated BER coefficient is lower (and even insignificant in column three) than in the original regression (Table 5). Also, the adjusted squared decreases. A similar, though a smaller reduction in the adjusted R-squared and the significance levels of the coefficients is found for not lagging the bilateral exchange rate, i.e. employing the BER as a coincidental variable for Romania only. Table 10 in the appendix displays the regression results. In conclusion, the one-period lag seems to have the best fit, even though it is not an end-of-the-month exchange rate.

Based on the 20 not completely randomly missing data points for the Hungarian short-term interest rate, I redo the CDS-based regression without Hungary as a basic test. Multiple imputation would be much better here, but is out of this paper’s scope. As Table 11 in the appendix reveals, the coefficient for a change in the bilateral exchange rate drops from 3.1 to 1.3 in column one and 3.2 to 1.5 in column five. At the same time, the estimated effect of a change in the German CPI increases by one basis point and its significance rises from 5% to 1%. Despite this, the adjusted R-square is again lower than in Table 5.

An identical reaction is found in the High-Euro debt country regression when dropping Hungary. To be fair, one has to say that there are only three countries left in the resulting regression, which results in only 273 observations (Table 12). Therefore, a conclusion can only be drawn carefully from this, but it seems as if Hungary has an undeniable influence in both regression setups.

(27)

Finally, the last two robustness exercises are done by excluding all US-Dollar denominated CDS rates from the Euro denominated CDS regression. For Denmark, Sweden and the UK the first six months were not available in Euro CDS. Norway drops out entirely because there are no Euro-denominated 5-year CDS spreads available. Table 13 shows slightly higher estimates, which can be explained by the now greater impact of High-Euro debt countries. This influence of High-Euro debt countries grows when one drops Norway, as is demonstrated in Table 14. Hence, besides the naturally rising leverage of countries with a significant share of Euro-denominated debts, the exclusion of all US-Dollar denominated CDS data is not causing a major shift in the estimated coefficients.

5 Conclusion

I have shown that there seems to be a link between non-Eurozone countries’ (if they have a high share of outstanding debt securities by non-financial corporates, which are denominated in Euros) bilateral exchange rate and their respective sovereign yields. Hence, they behave as proposed by the risk-taking model of HSS. These results are still not complete since there are some more obstacles which could adulterate my findings. For example, deviations in the interest rate that are already priced into forward rates. Hofmann et al. (2016) control for this by utilizing both a foreign currency bond spread and the local currency risk measure from Du and Schreger (2016). This lies out of scope for me, due to limited access to databases. Nonetheless, I believe that my findings present sufficient evidence to continue testing the influence of the risk-taking channel of currency appreciations against the Euro. For high Euro-debt share countries, my estimates show a four times greater impact on the 5-year CDS spread than Hofmann et al. (2016) estimated for EME’s currencies against the US-Dollar in their respective CDS-based regression. Explanations for this difference could be a higher share of foreign currency debt in the countries I picked than in the EME HSS included or/and the sample size. A subsample of HSS might have a similar scale.

There are several things to improve in future research. One could be to sharpen the risk measurement variable by either decomposing the VIX into two factors: uncertainty (a measure of expected stock market volatility) and the variance premium (a gauge for risk-aversion). Alternatively, instead of using the VIX as the risk measure variable, it is also possible to calculate a global CDS spread average as in Della Corte et al. (2015). Moreover, controlling for a structural break in the data could provide better insights. The monetary policy of the ECB (and other major central banks) clearly changed the financial markets and

(28)

especially the nominal interest rates. Finally, including the strength of legal institutions as a control variable to find out if a weak legal system influences companies decision to borrow more money in Euro or Dollar is another suggestion. If this variable is significant, strengthening these institutions could help preventing the risk-taking channel from being activated in the first place.

Another interesting point about the risk-taking channel is to consider the impacts on the non-Eurozone countries’ banking system, regarding the systemic risk. Even though diversification over many borrowers, i.e. companies, may help against idiosyncratic shocks, they remain vulnerable to strong currency depreciations against the Euro as this would cause multiple companies to default.

Finally, HSS suspect that especially EME with a state-owned oil and/or gas industry would profit from the initiated economics upturn. However, none of the countries in my high Euro-debt share group is known for being a resource-rich economy. Hence, taking a closer look at if resource rich countries are affected stronger, weaker or the same could be another extension of this topic.

I hope that my findings provide some guidance and new insights for policy makers. They might also help for solving the discussion about how considered major central banks should be regarding the international impacts of their decisions since there are two opposing effects of currency appreciation/depreciation and smaller economies might be able to protect themselves at least from the effects of the risk-taking channel of currency appreciation by reducing their share of clustered foreign currency debt (clustered in the sense of e.g. high share of Euro, US-Dollar or Renminbi).

(29)

6 Bibliography

Baskaya, Y. S., di Giovanni, J., Kalemli-Ozcan, S., Peydró, J.-L., & Fatih Ulu, M. (2015). Capital flows, credit cycle and investment: the role of macroprudential policy. CBRT-BIS-IMF conference on Macroprudential Policy: Effectiveness and Implementation Challenges. Istanbul.

Bekaert, G., Hoerova, M., & Lo Duca, M. (2013). Risk, uncertainty and monetary policy. Journal of Monetary Economics, 60, 771-788.

Blanchard, O., Ostry, J., Ghosh, A., & Chamon, M. (2015). Are capital inflows expansionary or contractionary? Theory, policy implications and some evidence. NBER Working Paper(21619).

Bruno, V., & Shin, H. (2015a). Cross-border banking and global liquidity. Review of Economic Studies, 82(2), 535-564.

Bruno, V., & Shin, H. (2015b). Capital flows and the risk-taking channel of monetary policy. Journal of Monetary Economics, 71, 119-132.

Cerutti, E., Claessens, S., & Ratnovski, L. (2014). Global liquidity and drivers of cross-border bank flows. IMF Working Paper, WP/14/69.

Céspedes, L. F., Chang, R., & Velasco, A. (2004). Balance sheets and exchange rate policy. American Economic Review, 94(4), 1183-1193.

Della Corte, P., Sarno, L., Schmeling, M., & Wagner, C. (2015). Exchange Rates and Sovereign Risk. unpublished manuscript.

Du, W., & Schreger, J. (2016). Local currency sovereign risk. Journal of Finance, 71, 1027-1070.

Feyen, E. H., Gosh, S. R., Kibuuka, K., & Farazi, S. (2015). Global liquidity and external bond issuance in emerging markets and developing economies. World Bank Policy Research Working Paper, WPS7363.

Fischer, S. (2010). Myths of monetary policy. Israel Economics Review, 8(2), 1-5.

Fleming, J. M. (1962). Domestic financial policies under fixed and floating exchange rates. IMF Staff Papers, 9, 369-379.

Hofmann, B., & Takáts, E. (2015, September). International monetary spillovers. BIS Quarterly Review, pp. 105-118.

Hofmann, B., Shim, I., & Shin, H. S. (2016, January). Sovereign yields and the risk-taking channel of currency appreciation. BIS Working Papers(538).

Krugman, P. (2013, Nov). Currency regimes, capital flows and crises. IMF Economic Review, 62(4), 470-493.

(30)

Krugmann, P. (1999, November). Balance Sheets, the Transfer Problem and Financial Crises. International Tax and Public Finance, 6(4), 459-472.

Merton, R. C. (1974). On the pricing of corporate debt: the risk structure of interest rates. Journal of Finance, 29, 449-470.

Morais, B., Peydró, J.-L., & Ruiz, C. (2015). The international bank lending channel of monetary policy rates and quantitative easing. World Bank Policy Research Working Paper, WPS7216.

Mundell, R. A. (1963). Capital mobility and stabilisation policy under fixed and flexible exchange rates. Canadian Journal of Economic and Political Science, 29(4), 475-485. Obstfeld, M. (2015, January). Trilemmas and trade-offs, living with financial globalisation. BIS

Working Papers, No 480.

Plantin, G., & Shin, H. (2016, January). Exchange rates and monetary spillovers. BIS Working Paper, 537.

Rey, H. (2013). Dilemma not trilemma: the global financial cycle and monetary policy independence. proceedings of the Federal Reserve Bank of Kansas City Economic Symposium at Jackson Hole.

Rey, H. (2014). International channels of transmission of monetary policy and the Mundellian trilemma. Mundell-Fleming Lecture at the 15th Jacques Polak Annual Research Conference on Cross-Border Spillovers.

Schoenmaker, D. (2013). Governance of international banking: the financial trilemma. Oxford: Oxford University Press.

Sobrun, J., & Turner, P. (2015, August). Bond markets and monetary policy dilemmas for the emerging markets. BIS Working Papers, 508.

Vasicek, O. A. (2002). The Distribution of Loan Portfolio Value. Risk. Retrieved from http://www.bankofgreece.gr/BoGAttachments/Vasicek.pdf

(31)

7 Appendix

Table 15: Summary statistics of Panel Data with Bond Yields as dependent variable:

Variable Mean Std. Dev. Min Max Observations

month overall 626 28.59343 577 675 N = 891 between 0 626 626 n = 9 within 28.59343 577 675 T = 99 country overall 5 2.583439 1 9 N = 891 between 2.738613 1 9 n = 9 within 0 5 5 T = 99 ΔBondYields overall .0027526 .3737048 -1.626 2.2384 N = 891 between .0072009 -.0119354 .0098808 n = 9 within .3736431 -1.63052 2.253088 T = 99 ΔlnBER t-1 overall -.0007829 .0184678 -.1562991 .0943508 N = 873 between .0008795 -.0020952 .0001424 n = 9 within .0184492 -.1566884 .0953637 T = 97 ΔlnNEER t-1 overall -.0009828 .0147287 -.0887504 .0569763 N = 873 between .0009252 -.0022826 .0003898 n = 9 within .0147029 -.0877465 .0578853 T = 97 orthogΔNEER t-1 overall -.0631981 1.107242 -4.587085 4.446216 N = 882 between .0407533 -.1172925 .0152527 n = 9 within 1.106575 -4.540839 4.416129 T = 98 ΔlnVIX t-1 overall -.004886 .1817697 -.3729794 .7046549 N = 873 between 0 -.004886 -.004886 n = 9 within .1817697 -.3729794 .7046549 T = 97 ΔlnCPI t-1 overall .001535 .0039272 -.027256 .02596 N = 873 between .0006128 .0006775 .0026697 n = 9 within .0038844 -.0283907 .0248253 T = 97 ΔlnIP t-1 overall -.0002302 .0317684 -.278316 .1880527 N = 872 between .0013077 -.0020078 .0024883 n = 9 within .0317444 -.2784829 .1878858 T = 96.8889 ΔIR t-1 overall -.0583432 .4386832 -3.34 5.21 N = 845 between .0139179 -.0857895 -.0383158 n = 9 within .4384843 -3.332158 5.237446 T-bar = 93.8889

(32)

ΔlnvSTOXX t-1 overall -.0010811 .2099736 -.5702827 .6460268 N = 873 between 0 -.0010811 -.0010811 n = 9 within .2099736 -.5702827 .6460268 T = 97 ΔvDAX-NEW t-1 overall -.0002208 .2002732 -.4519851 .596611 N = 873 between 2.87e-20 -.0002208 -.0002208 n = 9 within .2002732 -.4519851 .596611 T = 97 ΔlnvFTSE t-1 overall -.0044258 .2266672 -.4741418 .8292379 N = 873 between 0 -.0044258 -.0044258 n = 9 within .2266672 -.4741418 .8292379 T = 97 ΔlnIP-GER t-1 overall -.0001127 .0166913 -.0715733 .037827 N = 873 between 1.44e-20 -.0001127 -.0001127 n = 9 within .0166913 -.0715733 .037827 T = 97 ΔlnCPI-GER t-1 overall .0009633 .0038025 -.0110235 .0088668 N = 873 between 0 .0009633 .0009633 n = 9 within .0038025 -.0110235 .0088668 T = 97 ΔIR-GERt-1 overall -.0473196 .1788352 -.95 .24 N = 873 between 0 -.0473196 -.0473196 n = 9 within .1788352 -.95 .24 T = 97

Table 16: Summary Statistics of Panel Data with CDS spreads as dependent variable:

Variable Mean Std. Dev. Min Max Observations

months overall 628 27.43867 581 675 N = 855 between 0 628 628 n = 9 within 27.43867 581 675 T = 95 country overall 5.555556 3.024318 1 10 N = 855 between 3.205897 1 10 n = 9 within 0 5.555556 5.555556 T = 95 ΔCDS € overall .0021382 .3589364 -2.238 3.0683 N = 855 between .0068499 -.0071895 .0180726 n = 9 within .3588782 -2.228672 3.070622 T = 95 ΔCDS $ overall .0031111 .3724341 -2.65 2.51 N = 855 between .005821 -.0029474 .0174737 n = 9 within .3723936 -2.643941 2.495637 T = 95

The risk-taking channel of currency appreciation against the Euro