The role of financial markets during the Euro crisis : methods of impact measurement

Bachelor-Thesis

Josef Nitsch (st.nr. 10082913)

The Role of Financial Markets during the Euro Crisis

(Methods of Impact Measurement)

academic year 2012/2013

University of Amsterdam

Abstract

In this thesis I try to find evidence against the notion that financial markets driven by panic and fear caused or worsened the Euro – Crisis. With some significant results I can raise doubts on the thoroughness of these claims. Specifically I used the European Commission´s semiannually conducted surveys on the Euro-public´s satisfaction with Europe in such a way, that I obtained a measure with which to demonstrate my objective. Throughout, the thesis concerns itself with the findings of Aizenman, et al (2012), re-researches Aizenman´s evidence and looks for alternative answers to Aizenman´s proposed conclusion that the Euro-Crisis is best described by a multiple equilibrium model which once turned pessimistic renders us – the Europeans – powerless until it loses impetus. I conclude that the crisis could rather be a matter of solidarity.

Table of Contents

Introduction………

4

Literature Review………..………

5

Description of the Model ……….. 7

Endogeneity-Bias and Theory……… 8

Shortcomings of the Literature

And Research Method………

9

Reverse Causation ……….. 12

Research Setu-Up………..………… 13

The Data………

15

Replication and Comparison………

16

A Moment of Truth ………….……… 19

Conclusion………. 26

AppendixI………27

AppendixII……… …….29

AppendixIII……… ………...31

AppendixIV……… ………...32

Reference-List………. 33

3

Introduction

Since the beginning of the Euro-Debt crisis in 2009 it seems European policy-makers were always one step behind the course of events taking place in financial markets. Complying to what they described as “market pressures” they took supposedly inevitable actions to dampen the damage to the EMU (European Monetary Union), imposing harsh austerity measures on Southern-European states. In his recent paper “Panic-Driven Austerity in the Eurozone and its Implications” DeGrauwe (2012) presents evidence which point at policy-makers having been “gripped by panic and fear that erupted in the financial markets”. For the troubled Euro-countries DeGrauwe proves that investors did not price premiums according to underlying fundamentals (i.e. Debt/GDP ratio, etc.) but were instead bidding up premiums irrationally. His evidence heavily relies on the obvious declines in troubled sovereigns’ bond premiums, after the ECB’s announcement of its willingness to become a lender of last resort (OMT1-program, August 2012).

In a reaction to rumors of speculative attacks on the Eurozone – and specifically against Greece – the German regulation Authority BaFin (2010) has denied having found any evidence supporting these claims. Ten days later (March 18, 2010) however, the same agency forbade “naked CDS in government bonds” (BaFin, 2010), which are held to be the main weapons for speculative attacks.

The question motivating this thesis is to find out by means of statistical evidence whether financial markets really worsened the present crisis. DeGrauwe relies i.a. on statistical evidence presented by Aizenman, et al (2012) who draw on a sample of 50 countries over a timeframe of 2005 – 2010. Using the evidence of Aizenman et al, DeGrauwe paints a

frightening picture, in which society seems to have lost control on financial markets. The media also has contributed to this suspicion.

In this thesis I examine DeGrauwe’s evidence – which is the work of Aizenman et al. I try to see how strong this evidence is and research whether there is an alternative interpretation possible to the recent Euro-Debt crisis: I hypothesize that financial markets accounted for the specific monetary-policy constraints of Southern-European countries and contrary to

DeGrauwe’s findings, did price sovereign debt on rational analysis.

1 _{OMT is an acronym for Outright Monetary Transactions.}

4

Literature Review

DeGrauwe (2012) argues in his paper that financial markets demanded unreasonably high yields on SWEAP2- debt. His main evidence is the regression depicted in figure 1 below. Here sovereign bonds are compared by means of spreads which“are defined as the difference between each country’s 10-year government bond rate and the German10-year government bond rate.

Figure 1

DeGrauwe’s interpretation of the evidence is as follows: After the ECB announced the OMT-program in August 2012, the bonds initially priced with the highest premia declined the most. Since the ECB’s action however did not change the fundamentals of the SWEAP, the sharp declines in premia prove the initial spreads not to have been based on fundamental analysis. Furthermore, DeGrauwe (2013) argues that the austerity measures clearly had a negative effect on the SWEAP’s GDP growth, something which according to finance-theory should have been mirrored by further rising bond-premia. However premia declined instead, in spite of further GDP-deterioration, merely due to the ECB’s willingness to be a lender of last resort.

2 _{SWEAP = South-West-Euro-Area-Periphery: Greece, Ireland, Italy, Portugal and Spain. This acronym will be used}

throughout the thesis for the troubled euro-crisis countries and is originally used by Aizenman et al (2012).

5

DeGrauwe (2012, p.3) explains these events by a theory which he summarizes as follows: “Collective movements of fear and panic can have dramatic effects on spreads. These movements can drive the spreads away from underlying fundamentals very much like in the stock markets prices can be gripped by a bubble pushing them far away from underlying fundamentals.” He concludes that “panic and fear are not good guides for economic policies”, which rather ought to be conducted by rational decision-makers.

DeGrauwe quotes the work of Aizenman, et al (2012) and he refers to its findings.

Aizenman et al conduct a research which deals with CDS spreads-behavior during the crisis3. In this paper Aizenman et al develop a model based on fundamentals which according to financial-theory are considered to determine CDS-spreads. The model thus is built to produce reasonable predictions for CDS-spreads and is tested w.r.t. the Eurozone. In particular Aizenman et al compare the models’ explanatory power for the pre-crisis-period (2005-2007) with that of the crisis-period (2008-2010). Their results suggest that financial markets’ CDS-pricing is well in line with the fundamentals considering the pre-crisis period. CDS-pricing during the crisis period however seems to have permanently parted away from fundamentals w.r.t. the Eurozone. As opposed to non-Euro countries, whose risk-pricing dynamics converges to reasonable

predictability after 2009, the Eurozone stays largely unexplained by the model.

The interpretation Aizenman et al mainly give to their results is “that market-priced risk of sovereign default follows waves of contagion, overreacting and mispricing risk of sovereign default over a period of several years….The “good” (optimistic) expectational equilibrium [during the EMU’s first 10 years] quickly switched to a “bad” (pessimistic) expectational equilibrium in the SWEAP countries in 2010 as markets overreacted to fiscal deterioration and generated extraordinarily high CDS.” Not only does this interpretation resemble that of

DeGrauwe’s negative bubble-theory, but Aizenman et al explicitly refer in the same paragraph to DeGrauwe 2012.

3 _{Note, CDS spreads are highly correlated to bond-yields (or bond-premia) and accordingly can be regarded as a}

risk-measure, but they are not exactly the same. See Duffy (1998), or the “Appendix Data-sources” for more details.

Note also: CDS-spreads should not be confused with bond-spreads either, the latter being a mere comparative measure for DeGrauwe’s arguments. The next section “Description of the Model” offers more details on CDS-spreads.

6

Description of the Model

4

The model tested by Aizenman et al is defined as follows:

CDS

it

=

α

i

+

λ

t

+

θCDS

it-1

+X

it

’

β+ε

it

Next to a vector (X) containing the fundamentals, the model accounts for fixed effects w.r.t. to time (λt) and entities (αi)5. The X-vector in turn contains the following fundamentals used as predictors: ted-spread, inflation, trade/GDP, external debt and debt/taxbase6 or alternately fiscal-balance/taxbase. As mentioned above, a country’s fundamentals supposedly serve as basis for financial markets when pricing country risk. The justification for their usage in this context arises from their implications w.r.t. a country’s economic health. A short explanation with regard to each fundamental-variable included in the model, follows with a note next to each fundamental w.r.t. to their expected correlation with the dependent variable CDSit:

1.) Sovereign CDS-spreads (cds)7 are comparable to insurances against sovereign debt default. They are measured in basis points, whereby 1 basis point equals $ 1000 to ensure $ 10 million of debt.8 Also CDS-spreads are highly correlated to bond-yields (Duffy, 1998). This thesis follows the paper of Aizenman et al in that it uses data on 5year-maturity CDS when testing the model. The sign of the correlation with its lagged value can be positive and negative, depending on the perceived risk dynamics from one period to the next. 2.) The ted-spread (ted) is a measure for perceived risk during a specific period. It measures

the difference between US-treasury bills and the 3-month US$ LIBOR (London Interbank Offered Rate). While treasury bills are supposed to be risk-free, lending to banks is not. Consequently, the larger the ted-spread, the higher the perceived risk during a specific period; i.e. ted is expected to correlated positively with cds (Aizenman et al, 2012, p. 55).

4 _{Since the model of Aizenman at al plays a central role in this thesis I include a thorough description.} 5 _{(λt) and (αi) represent the respective intercepts of the fixed effects contained in the error-term.}

6 _{Aizenman et al call the Debt/Taxbase measure (or alternately Fiscal-balance/Taxbase measure) also}

“fiscal-space”.

7 _{The use of Italics here and in the following definitions refers to the labels I used for each respective variable in}

my sample. Throughout the thesis I use italics when I refer to my sample-variables (cds, ted, infl etc.) and normal letterings when referring to the variables of Aizenman et al.

8 _{Footnote Aizenman et al, 2012, p.40: “For example, if the spread is 197 basis points, meaning 197,000 USD to}

insure against 10,000,000 in sovereign debt for 10 years; 1.97% of notional amount needs to be paid each year, so 0.0197 x 10 million = $197,000 per year.”

7

3.) Inflation (infl) is known to exacerbate expected return on investment such as suggested by the Fisher equation9. The higher the inflation the lower the expected real-return on bonds of a specific country; i.e. infl. is expected to correlate positively with cds (Mishkin, 2012). 4.) Trade-Openness (trade) (measured by trade/GDP) allows a country to more efficiently use

its resources and thus diminishes country risk. A country engaged in trade is thought to specialize in the goods it most efficiently can produce (Krugman et al, 2012). Therefore

trade is expected to correlate negatively with cds.

5.) External debt (exdebt) as opposed to internal debt is denominated in foreign currency and cannot be inflated away by a country’s monetary-policy actions (Edward, 1984).

Accordingy exdebt is expected to correlate negatively with cds.

6.) Debt (debt) in general and alternately fiscal balance (fiscb) are held to be the key drivers of country risk. Sustainability of debt depends on a country’s fiscal balance. Once a country’s interest-rate payments on debt grew permanently larger than its fiscal-growth, its debt is considered to be unsustainable and default is expected to occur (Spaventa, 1987). Therefore

debt is expected to correlate positively with cds while fiscb is expected to correlate

negatively with cds-pricing.

Endogeneity-bias and Theory

Apart from fixed-effects and fundamentals, the model contains the lagged-value (CDS(it-1)) of the dependent variable (CDSit) as additional regressor. As such econometric theory assumes this variable to be endogenous since it correlates with the model’s residuals

ε(

it) which in turn

are modeled according to the equation εit = μi_ + vit, where μi_ represents the fixed-effects and vit is the idiosyncratic (non-fixed) disturbance-term (Roodman, 2006, pp.14).

Being the dependent variable from one period ago, CDS(it-1)is co-determined by its residuals

ε(

it-1) from one period ago. This bias supposedly can be overcome neither by OLS-

nor by fixed-effects estimation. In case of OLS-estimation the bias results from CDS(it-1)’s

potential correlation with the fixed-effects (μi_), which per definition are unchanged from one period to the next and thus present in the model’s contemporaneous OLS-residuals

ε(

it).10

9 _{Fisher equation: 𝑟𝑟 ≈ 𝑖𝑖 − 𝜋𝜋, where r is the real interest rate, i the nominal interest rate and π is inflation.} 10 _{This is a clear violation of the Gauss-Markov assumption that E(εit| Xit) = 0, where Xit is any regressor of the}

model in question.

8

Secondly, in case the fixed-effects estimator is applied, endogeneity-bias is assumed due to the correlation of CDS(it-1) with v(it-1), which is contained in v of the contemproaneous i

“demeaned” residual ∆ vit = vit – v (Roodman, 2006, p. 18).i 11

As such the resulting endogeneity-bias is present in both estimators, in the OLS- as well as in the fixed effects-estimator. Consequently the model is required to be estimated by the Arellano and Bond’s GMM-estimator, which is specifically developed for handling this

problem, given that the panel-data stretch only over a few periods (small T), but contain a large number of entities (large N). Aizenman et al however, apply this estimation-method labeled in their specification as “Arellano-Bond” (Appendix table C2, p.58) only in two instances from a total of 44 published regression-results. It is also only in these two instances that Aizenman et al specify the use of the lagged dependent variable in its 1st difference. In all other 42 instances Aizenman et al disclose the level of the lagged variable, labeled as y(t – 1) and explicitly accept its correlation with the residuals12. However, in instances where this correlation is negligibly small, CDS(it-1) can be treated as exogenous to its future value, thus possibly explaining the observed practice of Aizenman et al.

For the sake of conciseness I postpone the Hausman-test for exogeneity to Appendix I., but taking the results of Appendix I. into consideration I conclude that Aizenman et al indeed justifiably treat the endogeneity of CDS(it-1) as negligible. Therefore I will also assume exogeneity for CDS(it-1) throughout the rest of my thesis.

Shortcomings of the literature and Research Method

Economic theory has repeatedly pointed at the limitations of monetary policy to

countries participating in a monetary union (Gertler, 2003). Once hit by a recession the country in question is not able to inflate itself out of its declining economy. With countries largely differing in their economic competitiveness within the EMU, inflating the currency is not an option. Instead problem-stricken countries have to endure a painful process called internal

11 _{This effect is called “dynamic panel bias” by Roodman (2006).}

12 _{This is the reason why I adjusted the model in the discussion by presenting the level of the lagged dependent}

variable CDS(it-1) in the model, while Aizenman et al use its 1st difference ∆CDS(it-1) when introducing the model (p.44).

9

devaluation (Roubini, 2011). Therefore the comparison13 of Eurozone-CDS with non-Euro-CDS has to account for the specific constraints of monetary-policy which the Eurozone is subjected to. Consequently I claim that Aizenman et al as well as DeGrauwe have overlooked an important factor which financial markets likely considered for their risk-analysis. This claim is strengthened by the fact that it is exactly this constraining situation of the Eurozone, which has been alleviated when the ECB announced its OMT program. In light of this argument the declines of Euro-spreads after the OMT-announcement are apt to an interpretation which allows for reasonability of financial markets as opposed to DeGrauwe’s sentiment-theory.

Research Method

I clearly presume the model of Aizenman et al and de the theory of DeGrauwe to suffer from omitted variable bias. I therefore introduce two additional variables to the model of Aizenman et al, whose explanatory power goes around the ECB’s role during the crisis. In doing so, I intend to strengthen my hypothesis with a measure unique to the Eurozone. The two variables are: antiemu and its 1st differenced measure dantiemu14. Both variables measure European’s aversion against the union’s economic framework. I claim that financial markets accounted for public sentiment in the Eurozone when pricing country-risk. This argument is based on the fact, that European public cohesion can substantially relieve monetary-policy constraints in the form of intra-European aid- and bail-out programs. Voters’ opinions in this respect deliver the necessary basis for their policy-makers to conduct such programs. Once this basis is weakened by a change in public sentiment, the monetary constraints regain weight w.r.t. risk-assessment. As a consequence I expect such changes to be measurable and of predictive power to CDS-pricing evolution.

13 _{Aizenman et al specifically compare the SWEAP-countries to Panama, South-Africa, Malaysia, Mexico and}

Colombia, of which none is part of a monetary union.

14 _{To explain the introduction of the differenced variable dantiemu, consider the following example: it is well}

known to investors that Germany for instance always had very low Euro-Sympathy while Greece’s values were extremely unstable during the sampling period. In case the level of the sentiment is not influencing CDS-pricing then it could be the difference between today and yesterday. That is, investors knew that Germans didn’t like the EMU too much. But a relative change, even if small in absolute value, could be large enough to make investors nervous.

10

The data I use for these two variables are provided by the European Commission’s

Eurobarometer website15. On it, the European Commission has published public opinion polls conducted since the beginning of the 70ies in the Euro-Area. For my sample (which broadly covers that of Aizenman et al) I specifically used one continuous survey for all Euro-countries, “the standard Eurobarometer”, which has been conducted semiannually16.

less changes in public sentiment with Belgium experiencing the most stable values among all Euro-countries, with dissatisfaction settling at somewhat above 50%.

15 _{See Appendix, Data-sources.}

16 _{Note that for my experiments I used the “Autumn Reports” of each year.}

During the nine consecutive years of my sample (2004, 2012) approximately 1000 people were asked in each Euro-country the following question: “Please tell me if you tend to

trust in the European Union.” After

transformation I obtain the mentioned aversion-measure antiemu and its 1st difference dantiemu. Figure 3 depicts the movement of public opinion (antiemu) during the 2004 – 2012 period for the Eurozone countries included in my sample. As can be observed, the largest movements have taken place in the SWEAP countries. From an initially positive attitude towards Europe (low

antiemu values), disappointment in the

EMU is the largest in Greece (sample-id nr: 39) climbing above 80% in 2012. At the same time, non-SWEAP Euro-countries underwent fewer changes in public sentiment with Belgium and Finland experiencing the smallest changes in public dissatisfaction reaching about 50%. Responsible for these large swings are supposedly two intertwined measures taken during the Euro-crisis:

Figure 3

11

W.r.t. the SWEAP-countries the imposed austerity measures have decreased the EMU’s popularity substantially, while on the other hand dissatisfaction in non-SWEAP Euro-countries rose due to the controversial bail-out programs demanded from non-SWEAP populations.

Potential Reverse Causation

Whether the public cares about CDS-spreads or not, is important to determine potential simultaneous bias (reverse causation) of the added variables to the model. It seems reasonable to assume that the majority of the population is not directly influenced by CDS-pricing

developments. On the other hand it is equally likely that politicians – who evidently do care about financial markets – will introduce policies which in turn make the general public concerned about investors. This would weaken the introduced variables due to simultaneous bias.

However the assumption that nevertheless antiemu and dantiemu are rather accounted for by financial markets (sentiment  CDS-pricing) than reversely, CDS-pricing is accounted for by the Euro-population (rev. causation: CDS-pricing sentiment), is strengthened by the fact that financial markets are known to react immediately to news. Accordingly the added variables can be assumed consistent. To prove this, I compare the data on EURO-CDS as opposed to WORLD-CDS, whose evolutions have differed substantially in 2008, according to one of the main results of Aizenman et al.

Figure 4

EURO-CDS (Ecds) WORLD-CDS (cds)17

17 _{It’s the second peak (2011) which we’re interested in. This coincides with the Euro-crisis.}

0 50 0 10 00 15 00 20 00 25 00 2004 2005 2006 2007 2008 2009 2010 2011 2012 year Ecds Ecds_mean1 12

My data confirm this (figure 4). As can be seen EURO-CDS were almost not affected by the inception of the financial-crisis (2008) while they peaked at the height of the Euro-crisis (2011). The following figure (5) shows that the latter was synchronically paralleled by Euro-sentiment, while both measures underwent a negligible effect in 2008.

Under the presumption that financial markets have an immediate reaction time, I

conclude that the added variables don’t suffer (substantial) simultaneous bias (endogeneity) and as such have explanatory power w.r.t. CDS-pricing. Crucial hereby is the distinction between financial-crisis and Euro-crisis.

Research Set-Up

.3 .4 .5 .6 .7 .8 2004 2005 2006 2007 2008 2009 2010 2011 2012 year antiemu antiemu_mean1 -.1 0 .1 .2 .3 2004 2005 2006 2007 2008 2009 2010 2011 2012 year dantiemu dantiemu_mean1

Figure 5; antiemu (2004 – 2012) dantiemu (2004 - 2012)

As presented in figure 6 (= fig. 4b) there are two peaks in CDS-pricing evolution observable: the first occurs in 2008 and the second in 2011. Both of these peaks represent the inception of a crisis, with 2008 measuring the effect of the world-wide financial crisis and 2011 measuring the inception of the Euro-crisis. In other words figure 6 gives proof of two breaks having occurred, which is in line with the historic events. Aizenman et al

Figure 6

accordingly treat the peak in 2008 as a break in their

sample and subject their research set-up to the following procedure18:

1. The whole sample is regressed with OLS-estimates and clustered standard errors with the X-Vector containing {ted inflation trade exdebt debt year-dummies}. 2. The whole sample is regressed with OLS-estimates and clustered standard errors

with the X-Vector containing {ted inflation trade exdebt fiscb year-dummies}. 3. The whole sample is regressed with Fixed-Effects-estimators19 (no clustered

standard error) with the X-Vector containing {ted inflation trade exdebt debt year-dummies}.

4. The whole sample is regressed with Fixed-Effects-estimators (no clustered standard error) with the X-Vector containing {ted inflation trade exdebt fiscb year-dummies}.

This set-up is repeated three times accounting for the break in 2008: First over the whole sample period, secondly over the period before the break (pre-crisis) and thirdly it is followed for the period after the break (crisis-period). Thus in total there are 4 x 3 = 12 regressions published by Aizenman et al which are the basis for the regressions of this thesis.

As a starting point I intend to replicate the discussed procedure in order to obtain the respective twelve residuals. Thereby I will also discuss similarities and dissimilarities between the two samples (that of Aizenman et al and mine) and furthermore compare the results. This is done in order to re-test the power of the model, draw conclusions on the correctness of the applied specifications and finally to make assumptions about the differences between both datasets.

In order to justify the discussed procedure I ensure the occurrence of the mentioned break in my replication-data by running a Chow-test. I include a detailed discussion of this test in Appendix II. The outcome of the Chow-test confirms the practice of Aizenman et al and consequently justifies the replication of the twelve respective regressions.

After obtaining the twelve residuals I subsequently intend to regress these on the two assumedly omitted variables antiemu and dantiemu, which is the main objective of the replication-procedure.

In case the estimators yield significance, a piece of evidence is obtained that Aizenman et al as well as DeGrauwe overlook an important aspect in the Euro-crisis-development.

Consequently I regard such evidence as proof that financial markets actually did account for the

18 _{The break in 2011 is outside the sample of Aizenman et al.} 19 _{I denote these estimators also as “fe”}

14

monetary-constraints of SWEAP-countries and priced risk accordingly. Such outcomes will contradict the results of DeGrauwe as well as those of Aizenman et al to as large an extent as they are significant.

With regard to the regressions of the obtained residuals on antiemu and dantiemu I assume heteroscedasticity and I therefore use the “White”-standard-errors. Furthermore I treat estimates as “significant” up to the 10%-level in all regressions.

The Data

I try to use exactly the same sources which were used by Aizenman et al20 when replicating their regressions. While all sources except for one (“Economist Intelligence Unit”) are accessible to me via internet, the dataset unfortunately includes many missing variables.21 To make up for this inconvenience I include additional data for the period 2011-2012 as well as for the year 2004. I also add Finland to the sample which is not present in the sample of

Aizenman et al.22 By so doing my panel-sample contains 51 countries for the period 2004-2012. Aizenman et al use a sample that stretches over 50 countries during the period 2005-2010.

20 _{Sources are disclosed on page 55 in Aizenman et al, 2012.}

21 _{I received the Dataset of Aizenman et al on July 30}th_{, 2013. After comparing I conclude that Aizenman et al}

disclose “maximum sample observations” in their paper as opposed to the amount of observations available per regression.

22 _{For a full list of the countries included in my sample please refer to Appendix III.}

Variable | Obs Mean Std. Dev. Min Max ---+--- country | 0 id | 459 26 14.73566 1 51 year | 459 2008 2.584806 2004 2012 cds | 326 207.0859 392.2637 1.2 3905 CDSY1 | 293 176.7133 411.4138 0 3436.3 ---+--- ted | 459 .6674444 .5670541 .183 1.719 infl | 441 4.299682 4.044072 -4.863278 28.18746 trade | 446 89.0781 40.98703 22.1183 210.3743 exdebt | 370 116.8028 162.3066 8.407328 1129.771 fiscb | 426 -.1141696 .2672414 -1.267699 .6669243 ---+--- debt | 399 3.189987 3.058099 .2289787 25.11955 antiemu | 99 .5313131 .1089809 .35 .82 dantiemu | 88 .0263636 .0666985 -.12 .23 --- Figure 7 15

Figure 7 above depicts all the available data in my sample. For the websites offering the various data please refer to Appendix IV(Data sources) while a list with all country-names is included in Appendix III. The variables of the model are written in italics in the text as I

continue. The economic justification for using these variables in the model is outlined on page 7 of this thesis, section The Model. Also included in the discussion on page 7 is the expected sign of the correlation of each variable with the dependent variable. For interpreting the results in the following section I clearly refer to this sub-section.

Replication and comparison

1.

Whole sample regressions

Below (figure 8) I present the first four regressions23 of Aizenman et al. I type and underline Aizenman’s results printed next to my own for reference and easy distinction. As described in the previous section (p.14) I follow the procedure of Aizenman: Consequently I start with respectively two OLS-regressions interchanging debt (OLSd) and fiscb (OLSf). The

second four columns in figure 8 below concern themselves with the fixed regression command. As for OLS regressions, debt is interchangeably used with fiscb. The columns disclose the following results where “A” denotes the results of Aizenman et al and FE is an acronym for fixed effects regressions.

The Columns in this table are sorted by the type of estimators used together with the name of the interchanged variables debt and fiscb. Accordingly the acronyms in the first line of each column stands for the following regression set-ups:

Col. 1: OLSdebt col2: OLSdebt-A col3: OLSfiscb col4: OLSfiscb-A col5: FE.debt col6.: FE.debt-A - col7: FE.fiscb col8: FE.fiscb-A

Note, for all other regressions in this section I use these subdivisions labeled as shown above.

23 _{Aizenman et al, p. 48 – table 3.}

16

1 2 3 4 5 6 7 8 --- Variable | OLSd OLSdA OLSf OLSfA FEd FEdA FEf FEfA ---+--- cds | L1. | 0.2 0.3*** 0.2 0.3*** 0.13 0.3*** 0.1 0.2*** | ted | 222.5* 17.5 271.3** 43.6*** 254.4** -29.6 196.1* 64.1 infl | 41.5 28.1*** 35.6 * 29.9 50.6** 38.6*** 53.3*** 35.7*** trade | -0.1 1.2 -0.4 -12.4 -9.5 -186.7 -7.9 182.7 exdebt | 0.2 17.0*** 0.16 9.8** -.12 57.0* 0.2 -53.7 debt | 32.4 13.8*** 110.9 48.2 fiscb | -287.8** -418.7*** 16.4 -910.3*** _cons | -229.4* -87.3*** -175.8** -82.5** 342.7 -389.5 545.9 -501.7* ---+--- N | 177 300 186 300 177 300 186 300 r2 | 0.36 0.35 .36 0.35 0.30 0.19 .29 0.18 r2_a | 0.32 .31 -0.01 -.04 --- legend: * p<0.10; ** p<0.05; *** p<0.01

Starting with the positive aspects of this lineup, the R^2s have some resemblance. So do the estimated coefficients of the lagged variables’ estimators. However the overall results exhibit negligible resemblance. While 17 out of 28 of the results published by Aizenman et al in the above regressions are significant up to the 10% level, that is 61%, my results achieve 36% significance. When running the regressions without time-dummies the results differ even more. (Note, time-dummies are not disclosed in all tables presented in this section.)

Taken apart, accounting for the economic meaningfulness of my sample-estimators, most have an arguably reasonable sign (with the exception of fiscb=16.4 being positive, though its p-value is far from significant). As for the magnitude of the individual estimators I suggest informed comparisons. For example accounting for fixed effects (col. 7) fiscb in my regression and trade in Aizenman’s (col. 6 and 8) differ substantially from OLS-results (first four

columns) with up to a 156fold fixed effect (col.6; trade). Comparing both research set-ups

exdebt differs substantially between my results and those of Aizenman et al. However all these

results are not significant and thus do not lend themselves to informed inference.

Figure 8

2. Pre-crisis Period Regressions

Years 2004 – 2007 (pre-crisis)

col. | 1 2 3 4 5 6 7 8 --- Variable | OLSd OLSdA OLSf OLSfA FEd FEdA FEf FEfA

--- cds | L1. | 0.2 0.4*** 0.3** 0.5*** 0.1 0.0 0.0 0.2*** | ted | 45.5** 23.2 43.3*** 24.9*** 1.8 34.5*** 29.3 21.7*** infl | 17.3***8.7** 15.5** 7.5*** 0.2 9.8*** 3.4 6.5*** trade | -0.02 -14.0 -0.2 -24.6*** 0.7 -92.8* 0.6 -5.2 exdebt | -0.1 0.6 -0.03 1.4 0.05 -13.7 -0.2 debt | 15.5 18.3*** -16.5 123.4*** fiscb | -60.1 -202.9*** -311.8** -291.7*** _cons |-103.4 -44.1*** -42.3 -6.8 65.5 81.9 -2.1 107.4 ---+--- N | 51 150 53 150 51 150 53 150 r2 | 0.74 0.76 0.69 0.68 0.31 0.84 0.48 0.89 r2_a | 0.70 . 0.65 . -0.4 . -0.07 . --- legend: * p<0.1; ** p<0.05; *** p<0.01

As regards the pre-crisis period, again Aizenman et al disclose many more significant results than I achieve. Such as explained in footnote 22 this can be due to Aizenman et al using “maximum sample”- observations. Anyhow my sample is about a third the reported size of the compared one. The ratio of significant results for this time-period is 17 out of 28, which is 61% obtained by Aizenman et al (a result they achieve also with whole-sample regressions). My results’ significance-ratio is however three out of 28 estimated coefficients, which is 11%. Remarkably only one result (fiscb, col. 7) in my fixed effect regressions is significant.

The signs of all significant estimators are economically justifiable in both set-ups (mine and that of Aizenman et al). R^2s in the OLS regressions (col. 1 – 4) look alike while they do not when accounting for fixed effects (col. 5 – 8).

As for the magnitude of the individual estimators they again differ substantially between both samples. For example comparing trade (col. 6 and 7) the magnitude obtained by

Aizenman et al is roughly 133fold the size of my estimator. However, as before my estimator in this case is far from significant.

Figure 9

3. Crisis Period Regressions

Year 2008 – 2010,’12 (crisis period)

col. | 1 2 3 4 5 6 7 8 --- Variable | OLSd OLSdA OLSf OLSfA FEd FEdA FEf FEfA

--- cds |

L1. | 0.4*** 0.2*** 0.4*** 0.2*** 0.9*** -0.1 1.2*** -0.1 |

ted |(om) 190.6*** (om) 197.3*** (om) 150.2*** (om) 186.7*** infl | -3.0 52.2 -5.8 52.9*** -36.4*** 29.5** -26.3* 27.2*** trade | -0.2 -94.5 -0.5 -96.4 -4.0 -489.7** -1.1 -191.4 exdebt | 0.2** 28.6** 0.15** 21.5* 1.7* 189.3 0.14 33.0 debt | 20.0 14.4 299.2*** -182.1 fiscb | -237.2* -150.7 115.2 -567.4 _cons | 199.9 -82.01 98.2** -61.9 -634.8 -621.4 -130.9 -200.3 ---+--- N | 105 150 111 150 105 150 111 150 r2 | 0.45 0.45 0.46 0.45 0.70 0.58 0.60 0.61 r2_a | 0.41 . 0.41 . 0.43 . 0 .22 . --- legend: * p<0.1; ** p<0.05; *** p<0.01

As for the crisis24 period 36% of Aizenman’s results are significant while I achieve a

significance ratio of 39% up to the 10% significance-level. Note that while Aizenman’s sample ends in 2010 mine includes data for the next two years, too. However this does not alter my results substantially.25 As before I assume that Aizenman disclose “maximum sample” observations.

While the R^2 values display a good resemblance the overall estimators do not. Also, though the sign of most estimated coefficients is again compliant to economic theory, the magnitudes differ again substantially across both samples. An example is again trade when compared between col. 7 and 8, resulting in a 174fold difference.

Conclusion

Overall, the comparison between both samples is disappointing especially with regard to the significance-ratio but also w.r.t. the magnitudes of the estimated coefficients. Next to the

24 _{Again I used dummy-variables (not disclosed in the tables) and for the crisis-period I chose to leave out y08}

avoiding the dummy-trap.

25 _{These regression-results are not included in the thesis.} Figure 10

19

possibility of differing estimation-methods, as well as accounting for the “maximum

observation”-objection, an explanation could also be the use of different data-providers. Since however the same data-sources are processed in my sample as those published by Aizenman et al, the latter explanation misses some ground. Overall therefore the comparison is disappointing in that it was expected to render more similarities than observed

Nevertheless I conclude that considering the small number of significant results

obtained using my sample, there is more justification for assuming omitted variables. Therefore the tests for omitted variable bias in the subsequent section attain additional importance.

Outcomes

My replications in the previous section aim at one objective, namely to prepare the ground for the up-coming analysis. As described in the section Research Method (p.10) I expect to find evidence that Aizenman et all overlooked the SWEAP-countries monetarily constrained situation. As I argued, anti-European sentiment among the European public is expected to be an important component in the process of CDS-risk pricing. Should this be the case, then I ought to be able to prove that the sentiment-measure antiemu and dantiemu shed some light on the unexplained parts of the risk-pricing process and as such are to be regard as (one of the) omitted variables in the model of Aizenman et al. I therefore regress the twelve residuals from all the above estimations on both sentiment-measures respectively in order to check for their

significance in the model. I clearly expect a positive correlation. Additionally I expect the results for the split regressions (1st period/2nd period) to differ in significance. Logically I predict that sentiment became more important during the crisis period. As mentioned earlier I allow for significance up to the 10%-level.

The three outputs reprinted below are subdivided in four columns each with col. 1

Regression of OLS residuals on antiemu with debt included in the original

regression

col. 2

Regression of OLS residuals on antiemu with fiscb included in the original regression

col. 3

Regression of fixed effects residuals on

antiemu with debt

included in the original regression

col. 4

Regression of fixed effects residuals on

antiemu with fiscb

included in the original regression

I first present the regressions on antiemu followed by those for dantiemu. After respectively three presentations I discuss the results. I highlight significant results in gray.

1.Whole period (2004 – 2012); using antiemu

--- col.| 1 2 3 4

--- res obt with | OLSd OLSf FEd FEf ---+--- antiemu | 753.06461 816.3715 715.42858 954.65786* --- t | 1.32 1.40 1.59 1.93 P>|t| | 0.192 0.166 0.117 0.058* ---+--- N | 59 59 59 59 r2 | .06205424 .07147753 .08663858 .12757846 r2_a | .04559906 .05518766 .07061469 .11227281 legend: * p<0.1; ** p<0.05; *** p<0.01

2. First period (2004 – 2007), using antiemu

--- col.| 1 2 3 4

--- res obt with | OLSd OLSf FEd FEf ---+--- antiemu | 13.293435 50.682977 16.882822 24.041053 t | 0.15 0.78 0.36 0.43 P>|t| | 0.885 0.447 0.725 0.675 ---+--- N | 18 18 18 18 r2 | .00159857 .04142858 .00810652 .01290988 r2_a | -.06080152 -.01848213 -.05388682 -.04878325 --- legend: * p<0.1; ** p<0.05; *** p<0.01 Figure 13 21

3. Second period (2008 – 2012), using antiemu

--- col.| 1 2 3 4

--- res obt with | OLSd OLSf FEd FEf ---+--- antiemu | 906.11018 931.10189 106.86709 311.73421 t | 1.09 1.11 0.29 0.43 P>|t| | 0.284 0.277 0.775 0.67 ---+--- N | 33 33 33 33 r2 | .06175104 .06406076 .00482045 .02887796 r2_a | .03148495 .03386917 -.02728212 -.00244855 --- legend: * p<0.1; ** p<0.05; *** p<0.01

Unfortunately there is only one significant result out of twelve regressions (8.33% <10%), which since significant only at the 10% level, that is 10% of significant results are accepted to be wrongly estimated as such, renders antiemu insignificant. I conclude that the initial assumption with regard to antiemu, being an omitted variable in the model of Aizenman et al cannot be confirmed. Consequently it cannot be assumed that financial market accounted for the absolute level of Euro-cohesion when pricing risk.

I proceed with presenting the results using dantiemu.

1.Whole period (2004 – 2012) using dantiemu

--- col.| 1 2 3 4

--- res obt with | OLSd OLSf FEd FEf ---+--- dantiemu | 1061.4398** 1031.9093* 342.21594 566.28591 --- t | 2.06 1.90 0.96 1.49 P>|t| | 0.044 0.062 0.344 0.141 ---+--- N | 59 59 59 59 r2 | .05716214 .05295264 .00919157 .02081441 r2_a | .04062113 .03633777 -.00819103 .00363571 --- legend: * p<0.1; ** p<0.05; *** p<0.01 Figure 15 Figure 14 22

2.First period (2004 – 2007), using dantiemu

--- col.| 1 2 3 4

--- res obt with | OLSd OLSf FEd FEf ---+--- dantiemu | -132.84456 32.237855 18.054197 -55.760978 t t | -1.34 0.37 0.23 -0.48 P>|t | 0.198 0.714 0.817 0.639 ---+--- N | 18 18 18 18 r2 | .04224817 .00443581 .00245338 .01837979 r2_a | -.01761132 -.05778696 -.05989328 -.04297148 --- legend: * p<0.1; ** p<0.05; *** p<0.01

3.Crisis period (2008 – 20) using dantiemu

--- col.| 1 2 3 4

res obt with | OLSd OLSf FEd FEf ---+--- dantiemu | 1333.1871** 1302.5185** 42.683412 187.24281 t | 2.36 2.17 0.13 0.52 P>|t| | 0.025 0.038 0.899 0.608 ---+--- N | 33 33 33 33 r2 | .07246078 .06795227 .00041683 .00564737 r2_a | .04254016 .03788621 -.03182779 -.02642852 --- legend: * p<0.1; ** p<0.05; *** p<0.01

I conclude that differencing partly confirms my initial assumption. I obtain 4 significant estimates out of 12 regressions (33%), whereby none is significant at the 1% level; three are significant at the 5% level and one regression yields a 10% significance for the differenced

antiemu-variable.

Figure 17 Figure 16

As expected the 2nd period is more sensitive to the tested variable such as the last regression (fig. 17) shows. For the pre-crisis period no significance is displayed. The highest R^2’s is approximately 7% (fig. 17, col.1) which points towards the fact that there are more omitted variables involved than dantiemu can explain. Counter intuitively though the fixed-effects residual do not yield significance while the OLS-residuals do. Nevertheless, in order to justify the OLS-significance I regard the following assumption as necessary:

If according to the results of my thesis, the true model looks like

CDS

it

=

α

i

+

λ

t

+

θCDS

it-1

+X

it

’

β+γ∆antiemu

it

+ φ

(ϭ1μi_+ϭ2vit)

where (ϭ1μi_+ϭ2vit) = εit then

γOLS^hat =γ + φ𝑐𝑐𝑐𝑐𝑐𝑐(∆𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎.,εi,)_{𝑐𝑐𝑎𝑎𝑣𝑣(∆𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎.𝑎𝑎)} and γfe^hat = γ + η𝑐𝑐𝑐𝑐𝑐𝑐(∆∆𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎.,∆𝑐𝑐𝑎𝑎,) 𝑐𝑐𝑎𝑎𝑣𝑣(∆∆𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎.𝑎𝑎)

I assume the following interpretation of the significance of γOLS and the non-significance of γfe

1. The bias of the fe-estimator could have the opposite sign (say, γ is positive and the bias is negative) thus yielding the estimator insignificant.

2. Since φ^hat = 𝑐𝑐𝑐𝑐𝑐𝑐(𝐶𝐶𝐶𝐶𝐶𝐶𝑎𝑎,𝜀𝜀𝑎𝑎)

𝑐𝑐𝑎𝑎𝑣𝑣(𝜀𝜀𝑎𝑎) while η^hat =

𝑐𝑐𝑐𝑐𝑐𝑐(∆𝐶𝐶𝐶𝐶𝐶𝐶𝑎𝑎,∆𝑐𝑐𝑎𝑎)

𝑐𝑐𝑎𝑎𝑣𝑣(∆𝑐𝑐𝑎𝑎) the variance of εit can be assumed to be larger than the variance of ∆vit making φ < η in case the numerators’ magnitude does not change this relationship. Then the OLS-estimator is less biased. One economic explanation could be that the fixed effects have a much larger between-entity variance than the non-fixed effects’ demeaned value specifically due to the crises which occurred in the sample periods thus rendering counterintuitive results. In summery I thus assume the bias of the OLS estimator to be smaller than that of the fixed-effects-estimator. Only under this assumption can my hypothesis of omitted variable bias in the literature be supported with the evidence I present in this thesis.

With these additional assumptions I conclude to have obtained enough proof for my initial hypothesis, albeit only w.r.t. to the changes in sentiment (dantiemu). Accordingly investors actually did care about changes in public sentiment, though they did not do so for the absolute sentiment towards Europe as such. This applies throughout the crisis years and does not apply w.r.t. the pre-crisis period which is in line with my initial assumptions. I also

where η = φϭ2

conclude that my expectation with regard to the correlation (i.e. the sign) was correct. ( I expected positive correlation with the residuals, i.e. the higher the differenced anti- sentiment the more information is missing in the model of Aizenman et al.)

Interestingly, when accounting for fixed effects none of the results is significant. In order to justify the proof obtained by OLS-regression I add the assumptions outlined in the previous paragraph.

Conclusion

With this thesis I attempt to prove that financial markets were not driven by “panic and fear” when pricing default risk of Eurozone-countries. I specifically assume that the literature (DeGrauwe, 2012 and Aizenman et al, 2012) does not account for the monetary constraints Euro-countries are subjected to. Consequently the theory of “pessimistic bubbles” as well as the model of Aizenman et al suffer from omitted variable bias. To prove my hypothesis I employed to variables antiemu and dantiemu in order to measure rational decision-making in financial markets during the crisis. The logic considerations I make in my thesis provide enough ground to dismiss reversed causation and endogeneity of these to variables. The outcome of my

analysis suggests that financial markets indeed accounted for the EMU’s cohesion, which I hold to be the basis for alleviating monetary constraints in the union via aid-transfers and bail-out programs, next to role of monetary instruments of the ECB. However it is important to note, that financial markets did not account for absolute cohesion in the EMU but were rather influenced by relative changes in Euro-sentiment.

My results nevertheless cannot totally reject the theory of the literature (DeGrauwe) which assumes financial markets to have been driven by “panic and fear” during the Euro-crisis. Considering the relatively modest explanatory power of the assumed omitted variables, there consequently must have been more drivers to sovereign default-risk assessment. Therefore the “panic and fear” argument of DeGrauwe is not proven to be wrong.

However the outcomes of this thesis allow for a general point, too: since the change in Euro-sentiment is proven in this thesis to be significant in risk-assessment, which in turn is known to be decisive to the interest paid on debt – thus potentially driving countries into a debt spiral – I conclude that Europe’s future might depend more on people’s opinions than one may expect. That is, in general if optimistically changing an opinion just for the sake of optimism can reasonably assumed to be an option to men, then – accounting for the indirect influence on sovereign debt servicing- this change in attitude could well be (at least part of) the alleviation of the Euro-crisis. As such Europe’s future can be regarded to be in the hands (or the minds) of each individual European depending on the strength of his/her trust (i.e. solidarity) to his fellow European.

Appendix I.

Endogeneity Test of Lagged Dependent-Variable CDS(it-1)

Testing for endogeneity of the lagged dependent variable CDS(it-1) in my dataset I apply the Hausman-test. Next to all control-variables (i.e. the fundamentals including Dummies for each period except for one to avoid the dummy-trap), I chose as instrumenting variable the first lags of the two variables antiemu and dantiemu. This choice has been guided by the following considerations:

1. The lags of antiemu and dantiemu, i.e. antiemu(it-1) and dantiemu(it-1) are

contemporaneous to CDS(it-1). As explained in the section Potential Simultaneous Bias these variables are assumed to be influencing CDS-pricing immediately.

2. With the same token I assume the influence of antiemu and dantiemu on future CDS to be negligible, thus being adequate instruments for the lagged CDS.

3. Including more lags as instruments is therefore abandoned. Firstly it contradicts this set-up (point 2.) and secondly I want to avoid overidentification.

4. The literature (Roodman 2006, p. 15) suggests using further lags of the tested variable, i.e. CDS(i,t-2,3,etc.) as instruments. This however is not applicable to this situation due to the correlation of the lags with the “demeaned” idiosyncratic disturbance term, such as described in the section Endogeneity-bias and Theory.

Therefore I assume antiemu(it-1) and dantiemu(it-1) to be the best available instruments for the Hausman test. However testing for their significance in the reduced form yields only one significant F-test out of four tests executed. These four tests are: 1. Using the original model with fixed effects and debt 2. Exchanging debt for fiscb in the controls.3. Using OLS-estimation and debt. 4. Exchanging debt for fiscb in the controls.

As mentioned it is only in the second F-test (using the original model, i.e. fixed effects- with fiscb in the controls) of the reduced form that the which yields antiemu(it-1) and

dantiemu(it-1) to be statistically significant. (F-value=9.51 and Pv=0.0042). Therefore I

consider the second set-up as decisive for the outcome.

Note: Due to the restricted availability of both variables (# antiemu(it-1) = 88 obs. and # dantiemu(it-1) =77 obs.), all four test use only 54 observations each.

Following the discussed procedure with four tests, I present the Hausman test such as I execute it:

I.

CDS

it-1

=

α

i

+

λ

t

+

γ

1antiemu(it-1)+

γ

2∆antiemu(it-1 )

+X

it

’

β+ε*

it

II.

CDS

it

=

α

i

+

λ

t

+

φε*

it

+

θCDS

it-1

+X

it

’

β+ε

it

III. Finally I test for the significance of φ.

Significance of Reduced-Form Residuals

(2nd test assumed to be most conclusive): --- cds | Coef. Std. Err. t P>|t| [95% Conf. Interval] ---+--- res1. | -.3776779 1.105931 -0.34 0.735 -2.630386 1.87503 res2. | -.0982136 .4157554 -0.24 0.815 -.9450797 .7486525 res3. | -.6311552 .6944653 -0.91 0.385 -2.17852 .91621 res4. | -.3889811 .4314258 -0.90 0.388 -1.350258 .5722956

Conclusion Endogeneity-Testing

Due to the outcomes above – with all four tests confirming the Null-Hypothesis of exogeneity, whereby specifically test nr.2 yields a t-value of -0.24 with a p-value of 0.815 – I conclud in line with the practice of Aizenman et al, that the theoretic endogeneity of CDS(it-1) is negligible. Consequently the use of the Arellano-Bond estimator is assumed superfluous.

Interpretation

I assume the reason why the endogeneity of CDS(it-1) is rejected to lie in the

abnormality of the sample period with regard to the financial crisis and the Euro-crisis. Both events can potentially diminish the correlation between CDS(it-1) and v(it-1) (in ∆ vit = vit –

i

v w.r.t. the fixed-effects estimation) as well as the correlation between CDS(it-1) and μi_ (in

εit = μi_ + vit, w.r.t. OLS-estimation) in such a way that CDS(it-1) can be treated as exogenous.

Figure 17

Appendix II

Testing for the break

Testing for a break in the data which corresponds to the inception of the financial-crisis in 2008, I conduct a Chow-test. This is done in order to justify the replication procedure of Aizenman et al. Figure 6 of the section Research Set-Up, measuring the means of CDS during each sample-period is reproduced here for convenience. It clearly demonstrates the break in 2008.

The Couw-test is used if a break is specifically suspected at a certain point in time. It assumes the following build-up when adjusted to the Aizenman Model:

CDSit =

αi+λt+θCDSit-1+Xit’β+ γ0Dt(τ ) + γ1[Dt(τ) * Yt-1] +

+

Dt(τ) * Xit’ γ +

εit

where τ denotes the hypothesized break date and Dt(τ) = �0| 𝑡𝑡 ≤ τ_{1| 𝑡𝑡 > τ} ; that is Dt(τ) is a binary variable that equals 0 before the break date and 1 thereafter.26 Stock&Watson (2012, p.599) explain the further procedure regarding the Chow-test as follows:

26

Also Dt (τ )*Xit’γ denotes the inner product of the X-Vector with γ, which contains γi elements with i∈ {2,6} for the five controls. Dt(τ) is scalar.

0 10 00 20 00 30 00 40 00 2004 2005 2006 2007 2008 2009 2010 2011 2012 year CDS cds_mean1 Figure 18 29

“The hypothesis of a break can be tested using F-statistic that tests the hypothesis that γ0 = γ1 =

…= γm = 0 against the hypothesis that at least one of the γ’s is nonzero. This is often called a

Chow test”(Stock&Watson, 2012).

Test results with debt in the Controls27

--- L.cds | Coef. Std. Err. t P>|t| [95% Conf. Interval] ---+--- Dt | 15.14332 39.83442 0.38 0.704 -63.68181 93.96846 DT1 | 1.008543 .0099432 101.43 0.000 .9888677 1.028219 DTinfl | -18.46642 3.192917 -5.78 0.000 -24.78463 -12.14822 DTted | 29.49045 14.41574 2.05 0.043 .9643071 58.01659 DTtrade | -.0984293 .3161719 -0.31 0.756 -.7240765 .5272178 DTexdebt | .2218111 .1509297 1.47 0.144 -.0768516 .5204737 DTdebt | -15.70928 5.426352 -2.89 0.004 -26.44705 -4.971504 _cons | -146.6115 56.40773 -2.60 0.010 -258.2322 -34.99076 ---+---+--- The F-test for joint significance of all three estimators yields F(7, 127) = 1845.89 for the sample used in my thesis. This is equal to a p-value of 0.0000

Test results with fiscb in the Controls

--- L.cds | Coef. Std. Err. t P>|t| [95% Conf. Interval] ---+--- Dt | -41.60702 28.24668 -1.47 0.143 -97.47403 14.25999 DT1 | 1.00775 .0093556 107.72 0.000 .9892457 1.026253 DTinfl | -18.98705 2.95559 -6.42 0.000 -24.83269 -13.14141 DTted | 34.7661 13.7881 2.52 0.013 7.495632 62.03657 DTtrade | .1320806 .2684937 0.49 0.624 -.3989532 .6631143 DTexdebt | .141955 .1355676 1.05 0.297 -.126174 .410084 DTfiscb | 162.4271 43.82224 3.71 0.000 75.75432 249.0998 _cons | -116.9512 50.2685 -2.33 0.021 -216.3736 -17.52889 ---+---

The F-test for joint significance of all three estimators yields F (7, 134) = 2026.77 for my sample. This is equal to a p-value of 0.0000

Interpretation

Since both set ups (alternating between debt and fiscb) yield statistically significant results at far above the 1% level I conclude that also the data I use for my thesis are

characterized by the break corresponding to the inception of the financial crisis in 2008. In line with the practice of Aizenman et al I therefore specifically account for this break in my

27 _{Note: 1. I used the fixed estimator for these tests. 2. The estimates for the controls are left out in both tables.} Figure 19

Figure 20

30

regressions by running 1.whole-sample regression 2. pre-break regressions and 3. after-break regressions.28

Appendix III ……….

Country List

28 _{There is also a second break in the sample in 2011, signaling the Euro-crisis. However the data become very}

sparse in the last two years of my sample. I therefore leave this consideration to a follow-up research.

country countryid country countryid country countryid

Argentina 1 Ukraine 20 Tureky 38

Brazil 2 Venezuela 21 Greece 39

Bulgaria 3 Vietnam 22 Ireland 40

China 4 Vietnam 22 Italy 41

Colombia 5 Croatia 23 Portugal 42

Indonesia 6 Qatar 24 Spain 43

Kazakhstan 7 Australia 25 Austria 44

Lebanon 8 Chile 26 Belgium 45

Lithuania 9 Czech 27 France 46

Malaysia 10 Denmark 28 Germany 47

Morocco 11 Hungary 29 Netherlands 48

Panama 12 Iceland 30 Slovak 49

Peru 13 Israel 31 Slovenia 50

Phillipines 14 Japan 32 Finland 51

Romania 15 Korea 33 Russia 16 Mexico 34 South Africa 17 Norway 35 Thailand 18 Poland 36 Tunesia 19 Sweden 37 31

Appendix IV ……….

Data sources

Sovereign CDS spreads --- CMA Datavsion TED --- CMA Datavsion Tax base, Fiscal balance,

Public debt, Trade, Inflation --- World Bank´s WDI External Debt--- WDI, UN

Euro-Survey Result ---Eurobarometer

CMA source: Univeristy of Amsterdam

Eurobarometer: http://ec.europa.eu/public_opinion/index_en.htm

UN: http://www.unece.org/

WDI: http://data.worldbank.org/data-catalog/world-development-indicators

References

Aizenman, J., Hutchison, M., 2012. What is the risk of European sovereign debt defaults?

Fiscal space, CDS spreads and market pricing of risk, Journal of International Money

and Finance 34, (2013) p. 37 - 39.

Bafin, 8 March 2010, BaFin clarifies: So far no evidence of massive speculation against Greek

bond.;(http://www.bafin.de/SharedDocs/Veroeffentlichungen/EN/Pressemitteilung/2010 /pm_100308_cds_spekulationen_en.html), Aug. 21

De Grauwe 2012, Panic-Driven Austerity in the Eurozone and its Implications.

(http://www.voxeu.org/article/panic-driven-austerity-eurozone-and-its-implications),

Aug. 21.

De Grauwe2013, More Evidence that Financial Markets imposed excessive Austerity in the

Euro-zone, Center for European Policy Studies. ( http://www.ceps.eu/book/more-evidence-financial-markets-imposed-excessive-austerity-eurozone), Aug. 21. Duffie D., 1999; Credit swap valuation, Financial Analyst’s Journal

Edwards S., 1984; LDS’s Foreign Borrowing and Default Risk: An Empirical Investigation, The National Bureau of Economic Research. (http://www.nber.org/papers/w1172), Aug.21

Gertler et al, 2003; External Constraints on Monetary Policy and the Financial Accelarator, Journal of Money, Credit and Banking; Vol. 39, Issue 2-3, pages 295 – 330, 2007. (http://ideas.repec.org/p/nbr/nberwo/10128.htm), Aug 21.

Krugman et al, 2012; International Economics. Theory & Policy. Pearson, 9th Global Edition. F.S. Mishkin, 2012; Are Real Interest Rates Equal across Countries? An empirical

Investigation of international Parity Conditions, The Journal of The American Finance

Association. ( http://onlinelibrary.wiley.com/doi/10.1111/j.1540-6261.1984.tb04911.x/abstract), Aug. 21.

Roodman D., 2006; How to Do xtabond2. An introduction to Difference and System GMM in

Stata, Center for Global Development

Roubini N., 2011; Greece Should Default and Abandon the Euro, Roubini Global Economics.

(http://pratclif.com/economy/roubini/Greece%20Should%20Default%20and%20Aband

on%20the%20Euro%20-%20September%2015%202011.pdf), Aug 21.

Spaventa L., (1987); The Growth of Public Debt, Staff Papers – International Monetary Fund, Vol. 34, No. 2.

(http://www.jstor.org/discover/10.2307/3867139?uid=3738736&uid=2&uid=4&sid=211 02567089323), Aug. 21.

Stock J and Watson M. 2012, Introduction to Econometrics, Pearson Global Edition