ECOLE POLYTECHNIQUE ON THE SELF-FULFILLING PROPHECY OF CHANGES IN SOVEREIGN RATINGS. Cahier n Ingmar SCHUMACHER.

Transcription

1 ECOLE POLYTECHNIQUE CENTRE NATIONAL DE LA RECHERCHE SCIENTIFIQUE ON THE SELF-FULFILLING PROPHECY OF CHANGES IN SOVEREIGN RATINGS Ingmar SCHUMACHER hal , verion 1-9 Feb 2012 February 2012 Cahier n DEPARTEMENT D'ECONOMIE Route de Saclay PALAISEAU CEDEX (33) mailto:chantal.poujouly@polytechnique.edu

2 On the Self-Fulfilling Prophecy of Change in Sovereign Rating Ingmar Schumacher Ecole Polytechnique Pari hal , verion 1-9 Feb 2012 February 3, 2012 Abtract We empirically invetigate the dynamic interaction between overeign rating and the macroeconomic environment. We ue a Panel VAR on annual data for European countrie from Our reult provide evidence for a ignificant two-way interaction between the macroeconomic environment and change in overeign rating. Thu, rating change are able to exacerbate a country boom-but cycle. Keyword: overeign rating; Panel VAR; elf-fulfilling prophecy. JEL claification: C33; H6. Ecole Polytechnique Pari, ingmar.chumacher@polytechnique.edu. I am grateful for comment to Gaton Giordana, Thoma Mathä and Eric Strobl. 1

3 1 Introduction The recent change in overeign rating have received coniderable attention from policy maker and reearcher alike. Often, change in rating are believed to induce a elf-fulfilling prophecy. For example, fear mounted that hal , verion 1-9 Feb 2012 government which are going through a period of crii would be additionally adverely affected by rating downgrade. A a conequence, downgrade would induce a elf-fulfilling prophecy of intability. Evidence i mounting that ugget ignificant effect from change in rating to macroeconomic condition, epecially bond and tock price a well a default (e.g. Cantor and Packer 1996, Kaminky and Schmukler 2002, Reinhart 2002, Brook, Faff, Hillier and Hillier 2004, Ferreira and Gama 2007), but alo from macroeconomic condition to change in rating (e.g. Afono, Gome and Rother 2011, Hilcher and Nobuch 2010, Mellio and Paget-Blanc 2006). However, the tudie jut cited focu on either the effect of rating on macroeconomic variable or the other way around. 1 In contrat, a tudy of the elf-fulfilling prophecy would require an integrated framework, allowing for two-way feedback between change in rating and change in macroeconomic condition. Our contribution in thi article i, thu, to tudy thee feedback within a Panel VAR framework. There have recently been ome contribution that quetion whether change in rating are able to exacerbate a country boom-but cycle, ince they find that change to rating were mainly reaction to new (Mora 2006). Our panel VAR analyi allow u to invetigate thi quetion more fully. In particular, the main reult of our tudy i that we find a ignificant two-way interaction between our macroeconomic variable and change in overeign rating, uggeting that rating are, indeed, able to exacerbate a country boom-but cycle. 1 Cantor and Packer (1996) tudy both but not in a dynamic, interactive way. 2

4 2 Data and Methodology Our data conit of conumer entiment (CCI), Gro Dometic Product (GDP), Government Deficit (GD) and Population, all of which come from Eurotat, while the data on the rating i taken from Fitch Complete Sovereign Rating Hitory and Bloomberg. We focu on European countrie ince their data i fully harmonized and thereby comparion i facilitated and meaningful. The rating data come from Fitch, S&P a well a Moody and it i the overeign long-term rating. We recode the rating in a numerical form, ranging hal , verion 1-9 Feb 2012 from 0 for DD to 22 for AAA for Fitch, from 0 for D to 22 for AAA in the cae of S&P, and from 0 for C to 22 for Aaa in the cae of Moody. We take the average of the three rating for each country at each point in time in order to obtain a balanced picture. In cae there are everal change in a overeign rating within one year we weigh each rating by the number of day that the rating wa active during that year. We then calculate the growth rate of the rating (in percentage), denoted by g(r), in order to obtain a variable that better fit within the econometric approach of the Panel VAR (i.e. i eentially unbounded). A overeign rating ha been related to it probability of default and economic oundne (Reinhart 2002). Thu, change in rating hould drive invetor expectation on their potential return and houehold expectation on their future income. Additionally, change in rating affect a overeign cot of financing it budget deficit (Brook et al. 2004). In conequence, we alo expect rating change to impact a overeign deficit. The variable d(cci) give the change in the harmonized conumer entiment index. A decribed in the background document of the European Commiion, (European Commiion 2007), the CCI i the arithmetic average of the balance (in percentage point) of the anwer to the quetion on the financial ituation of houehold, the general economic ituation, unemployment expec- 3

5 tation (with inverted ign) and aving, all over the next 12 month. Thu, it i a forward-looking index of the houehold perception on the development of their financial ituation. With thi variable we capture the expectation of the houehold in our ample. We anticipate that change in their overeign rating hould impact their expectation poitively. A the main indicator for the current economic ituation we ue the growth rate of GDP per capita. GDP i meaured in market price in Million of Euro. We calculate the growth rate of GDP per capita in percentage term and denote hal , verion 1-9 Feb 2012 it by g(gdppc). We expect a poitive impact from g(gdppc) on a overeign rating, but a negative impact from change in a overeign rating on it GDP growth. The government deficit i the deficit of the general government and meaured a total expenditure minu revenue. We calculate it relative to GDP in order to minimize cale effect 2 and denote change in thi variable a d(gd/gdp). Alo, thi provide u with information on the ize of the budget deficit relative to that of the national economy. Thi i the only way in which one can quantify whether a deficit i actually large. In line with the recent obervation, we expect rating downgrade to follow hock to d(gd/gdp). Furthermore, given previou reult in the literature we anticipate that government deficit decreae following rating upgrade. Baed on thi data, our ample conit of 26 European countrie, namely Autria, Belgium, Bulgaria, Cypru, Czech Republic, Denmark, Etonia, Finland, France, Germany, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, Netherland, Poland, Romania, Slovakia, Slovenia, Spain, Sweden and United Kingdom. 3 2 The reult tay eentially unchanged if we ue the per capita deficit. 3 In the Robutne Appendix we exclude thoe countrie that did not have a rating change during the period of tudy. We how that thi doe not lead to qualitative change to the reult. 4

6 Thi give u an unbalanced dataet coniting of at maximum 363 countryyear obervation ranging from 1986 to The ue of the annual data hould minimize potential anticipation effect of fical policy change (Ramey, 2006) and help u in avoiding puriou reult due to cyclical effect. The ummary tatitic are provided in Table 1, and the correlation in Table 2. hal , verion 1-9 Feb 2012 Table 1: Summary tatitic Variable Mean Std. Dev. Min. Max. d(gd/gdp) g(gdppc) d(cci) g(r) Looking at the correlation in Table 2 reveal all are ignificant at the 5% ignificance level, apart from the correlation between d(cci) and d(gd/gdp) and g(gdppc) and d(gd/gdp), which are uncorrelated. Change in conumer entiment are poitively correlated with per capita GDP growth (0.21). Change in a overeign rating are negatively correlated with change in it budget deficit (-0.119) but poitively with change in conumer entiment (0.247). Table 2: Cro-correlation table Variable d(gd/gdp) g(gdppc) d(cci) g(gdppc) (0.986) d(cci) (0.419) (0.000) g(r) (0.024) (0.000) (0.000) A our etimation trategy we reort to a Panel Autoregreive Regreion (PVAR) with two lag. 4 Since we expect all variable to be at leat weakly endogenou we reort to the reduced-form VAR approach a thi avoid impo- 4 The Robutne Appendix how that the number of lag doe not influence the reult. 5

7 ing a detailed tructural model. Furthermore, the VAR approach allow u to identify the dynamic effect of our variable, which we argued in the previou ection to be important for undertanding the full interaction between rating change and macroeconomic variable. It furthermore allow u to iolate the individual effect of each variable via orthogonalized impule repone, which we decompoe baed on the Choleky decompoition (ee e.g. Hamilton 1994). We etimate the model itelf via ytem GMM baed on the STATA routine provided by Inea Love (ee Love and Zicchino 2006). hal , verion 1-9 Feb 2012 Firtly, we time demean the erie, 5, which control for time-pecific effect. Secondly, we helmert tranform the variable, which i a forward mean-differencing of the variable in order to take away fixed effect without introducing erial correlation. We chooe the ordering {d(gd/gdp), g(gdppc), d(cci), g(r)}. Due to the Choleky decompoition, a variable i allowed to react in the ame period to all variable ordered before it, but doe not contemporaneouly react to any of the variable ordered after it. 6 Our ordering i baed on the view that the government deficit impact GDP directly (e.g. Ramey 2011), and that the rating are reponding to macroeconomic condition only contemporaneouly (e.g. Mora 2006). Thu, we align ourelve with the reult in Mora (2006), namely that rating react to new, and thereby et the tage againt a contemporaneou feedback from rating to macroeconomic variable. If we, even in thi cae, find evidence in favor of a two-way relationhip, then thi would provide the tronget upport for the elf-fulfilling prophecy. 5 Thi i done via calculating the average of each variable at each point in time, and then ubtracting thee from the actual variable. 6 In the Robutne Appendix we dicu that the reult are not qualitatively affected by the ordering. 6

8 3 Reult The reult for the variance decompoition are hown in Table 3, while the impule repone reult are preented in Figure 1. The impule repone ue 5% confidence band generated by Monte Carlo imulation with 1000 replication. Overall, we find ignificant dynamic interaction between change in countrie rating and their macroeconomic environment, providing upport for the elf-fulfilling prophecy. 7 hal , verion 1-9 Feb 2012 Table 3: Variance decompoition Equation d(gd/gdp) g(gdppc) d(cci) g(r) d(gd/gdp) g(gdppc) d(cci) g(r) In particular, our reult how that 84% of the variance in overeign rating change can be attributed to an own hock, while the ret of the variance i explained by change in a overeign government deficit (2.4%), by per capita GDP growth (3.5%) and by change to conumer entiment (10.07%). Thu, though in line with the previou literature on the utainability of government finance (e.g. Afono et al. 2011), we alo find a relevant role for per capita GDP growth and for expectation. > Figure 1 about here < The impule repone in Figure 1 how ignificant feedback from the macroe- 7 In the Robutne Appendix we preent the complete robutne tudie. We tudied tationarity, all variable were tationary at any lag length with or without trend. We excluded all thoe countrie that did not have a rating change during our period of tudy (Autria, France, Germany, Luxembourg, Netherland, United Kingdom), without a change to the reult. We changed the ordering of the variable and varied the lag tructure and included additional control. We ued the variable without controlling for time-pecific effect and we alo did not helmert tranform them. All of thee did not affect the main reult of our analyi. 7

9 conomic variable to change in the rating. The effect of change in a country GDP growth rate i hort-run and impact a country rating only contemporaneouly. A one percentage point increae in a country GDP growth rate increae the growth rate of that country rating by 0.4 percentage point. Change to conumer entiment lead to a tatitically ignificant and long-term increae in a overeign rating. Here we find that a one tandard deviation increae in conumer entiment raie a country growth rate of it rating by roughly 0.5 percentage point. Though marginally tatitically inignificant, hal , verion 1-9 Feb 2012 we find a overeign rating growth will be reduced following a one tandard deviation increae in that country government deficit-to-gdp ratio. 8 The effect from change in rating to a country macroeconomic environment are alo non-negligible. The reult indicate that change in rating can explain 0.66% of the variance in the government deficit, 3.75% of the change in conumer entiment, and up to 11.79% of the variance it the country per capita GDP growth rate. Our impule repone reult how that both per capita GDP growth and change in conumer entiment are ignificantly poitively related to change in rating. We find that the effect of rating change on per capita GDP growth work it way through change in conumer entiment. A one percentage point increae in a country growth rate of it rating increae per capita GDP growth by roughly 1.1 percentage point after two year, while it conumer confidence increae by approximately 1.2 point within one year. A country government deficit marginally increae after an increae in it rating. We find that a one percentage point increae in a country growth rate of it rating increae a country deficit-to-gdp ratio by roughly 0.15 point. Thi effect arie after two year but it i hort-term. Hence, we conclude that there i evidence 8 In our Robutne Appendix we how that thi effect may become tatitically ignificantly different from zero depending on the time period, the ample or the tranformation we apply to the variable. 8

10 that change in rating may induce countrie to take a more utainable fical poition. 4 Concluion In thi article we find evidence for the elf-fulfilling prophecy caued by change in overeign rating, uggeting that rating are, indeed, able to exacerbate a country boom-but cycle. Thu, rating eem to have a imilar impact a marking-to-market of balance heet. While marking-to-market of balance hal , verion 1-9 Feb 2012 heet may lead to fire ale and additional round of feedback between aet ale and aet price (Plantin, Sapra and Shin 2008), thereby potentially rendering an otherwie ound intitution illiquid, rating change may exacerbate a overeign boom-but cycle by two-way feedback between it rating and it macroeconomic condition. The obviou advantage of overeign rating i that they provide debt holder and invetor with an idea about the probability of a overeign default. Another advantage i that rating downgrade will place preure on government to addre tructural problem that otherwie might get potponed and potentially reult in larger cot than thoe incurred by immediately tackling the problem. The diadvantage, a we have hown, arie from the fact that change in rating can induce a downward piral and eentially aggravate exiting problem. To find the welfare trade-off between the advantage and the diadvantage a well a potential policy olution hould prove to be a fruitful future reearch agenda. 9

11 Reference Afono, A., P. Gome, and P. Rother, Short-and long-run determinant of overeign debt credit rating, International Journal of Finance & Economic, 2011, 16 (1), Brook, R., R.W. Faff, D. Hillier, and J. Hillier, The national market impact of overeign rating change, Journal of Banking & Finance, 2004, 28 (1), hal , verion 1-9 Feb 2012 Cantor, R. and F. Packer, Determinant and impact of overeign credit rating, Federal Reerve Bank of New York New York, European Commiion, The joint harmonied EU programme of buine and conumer urvey: Uer Guide, European Commiion Directorate - General for Economic and Financial Affair, Ferreira, M.A. and P.M. Gama, Doe overeign debt rating new pill over to international tock market?, Journal of Banking & Finance, 2007, 31 (10), Hamilton, J.D., Time erie analyi, Vol. 2, Cambridge Univ Pre, Hilcher, J. and Y. Nobuch, Determinant of Sovereign Rik: Macroeconomic Fundamental and the Pricing of Sovereign Debt*, Review of Finance, 2010, 14 (2), Kaminky, G. and S.L. Schmukler, Emerging market intability: do overeign rating affect country rik and tock return?, The World Bank Economic Review, 2002, 16 (2), Love, I. and L. Zicchino, Financial development and dynamic invetment behavior: Evidence from panel VAR, The Quarterly Review of Economic and Finance, 2006, 46 (2),

12 Mellio, C. and E. Paget-Blanc, Which factor determine overeign credit rating?, The European Journal of Finance, 2006, 12 (4), Mora, N., Sovereign credit rating: guilty beyond reaonable doubt?, Journal of Banking & Finance, 2006, 30 (7), Plantin, G., H. Sapra, and H.S. Shin, Marking-to-Market: Panacea or Pandora Box?, Journal of Accounting Reearch, 2008, 46 (2), Ramey, V.A., Can Government Purchae Stimulate the Economy?, Journal hal , verion 1-9 Feb 2012 of Economic Literature, 2011, 49 (3), Reinhart, C.M., Default, Currency Crie, and Sovereign Credit Rating, World Bank Economic Review, 2002, 16 (2),

13 hal , verion 1-9 Feb 2012 Figure 1: Impule repone rep. of d(gd/gdp) to d(gd/gdp) rep. of d(gd/gdp) to g(gdppc) rep. of d(gd/gdp) to d(cci) rep. of g(gdppc) to d(gd/gdp) rep. of g(gdppc) to g(gdppc) rep. of g(gdppc) to d(cci) rep. of d(cci) to d(gd/gdp) rep. of d(cci) to g(gdppc) rep. of d(cci) to d(cci) rep. of g(r) to d(gd/gdp) rep. of g(r) to g(gdppc) rep. of g(r) to d(cci) rep. of d(gd/gdp) to g(r) rep. of g(gdppc) to g(r) rep. of d(cci) to g(r) rep. of g(r) to g(r) 12

14 ROBUSTNESS APPENDIX to On the Self-Fulfilling Prophecy of Change in Sovereign Rating Ingmar Schumacher hal , verion 1-9 Feb 2012 Ecole Polytechnique Pari February 3, 2012 Abtract We empirically invetigate the dynamic interaction between overeign rating and the macroeconomic environment. We ue a Panel VAR on annual data for European countrie from Our reult provide evidence for a ignificant two-way interaction between the macroeconomic environment and change in overeign rating. Thu, rating change are able to exacerbate a country boom-but cycle. Keyword: overeign rating; Panel VAR; elf-fulfilling prophecy. JEL claification: C33; H6. 1

15 Thi i the robutne appendix to the article entitled On the Self-Fulfilling Prophecy of Change in Sovereign Rating. In Table 1 we preent panel data unit root tet baed on the Fiher tet (ee Baltagi (2005). Table 1 how that all variable are tationary. hal , verion 1-9 Feb 2012 Table 1: Fiher-type unit root tet 1 lag d(gd/gdp) g(gdppc) d(cci) g(r) Invere chi-quared p-value Mod. inv. chi-quared p-value lag and trend d(gd/gdp) g(gdppc) d(cci) g(r) Invere chi-quared p-value Mod. inv. chi-quared p-value H0: All panel contain unit root. Reult hold with any number of lag. We checked whether our reult were not driven by the lat crii. Thu, we dropped the obervation from 2008 onward and re-ran the analyi baed on thi ub-ample. The reult, hown in 1, are qualitatively the ame. However, the repone the growth rate in rating to the growth of GDP per capita i now more pronounced and i tatitically ignificantly different from zero from one year after the hock to GDP per capita until ix year after the hock. In addition, the growth rate of the rating i now tatitically ignificantly affected by a hock in the deficit-to-gdp ratio. A an additional robutne check we varied the lag tructure. In Figure 2 we preent the model with one lag, while in Figure 3 we preent the reult with three lag. Again, we find no remarkable change to our reult above. The aumed ordering tend to be important for the impule repone due to the Choleky decompoition. Thu, we inpected the robutne of the reult with alternative ordering of the variable, ome of which we preent in Figure 4 and 5. The ordering in Figure 4 i {g(r), d(cci), d(gd/gdp), g(gdppc)}, while the one in Figure 5 i {d(cci), g(r), g(gdppc), d(gd/gdp)}. We find that the only relevant difference arie in the repone of the growth rate in the rating to that in GDP per capita. Baically, if the growth in GDP per capita cannot contemporaneouly affect the growth rate in the rating, then the rating will be unaffected by a hock to GDP per capita. Apart from thi we find no important qualitative difference in the reult, indicating that the ordering ha no ignificant impact on our analyi. 2

16 We then contrained the ample to only include thoe countrie that had change in their rating during the period of obervation. The impule repone for that cae are hown in Figure 6. Our reult are unchanged and the analyi i fully robut to the excluion of thoe countrie. A an additional robutne analyi we include further variable. Omitted variable bia i one way in which reult in a VAR could be biaed. We, thu, include two additional control, the inflation rate and change in the balance of payment relative to GDP. The impule repone reult are hown in Figure 7. The inflation rate ha been found to be one determinant of rating (ee e.g. Cantor and Packer 1996). We find a negative relationhip between HICP and change in rating, and thu confirm the reult in Cantor and Packer (1996). The variable balance of hal , verion 1-9 Feb 2012 payment account for outflow or inflow of good and capital. While we find that there i a two-way relationhip between balance of payment and rating, we alo oberve that our previou reult remain unchanged. A a further robutne analyi we cut all variable at the 1% and 99% tail. In thi way we analyze whether potential outlier may drive our reult. Thi reduce the ample by 39 obervation. The reult are preented in Figure 8. We find that now the growth rate of the rating repond not only negatively but alo tatitically ignificantly different from zero to a hock in the deficit-to-gdp ratio. In contrat, while the repone of the growth rate in rating to per capita GDP growth i till poitive, it i not tatitically different from zero any longer. A final robutne exercie we do not time de-mean the erie in order to ee whether our reult may be robut without taking care of time-pecific effect. The impule repone of thi are hown in Figure 9. Our reult continue to hold. In addition, we find a much tronger tatitically ignificant relationhip between rating and the deficit-to-gdp ratio. Specifically, we find that a hock to rating growth now decreae the deficit-to-gdp ratio, while an increae in overeign deficit-to-gdp ratio reduce it rating. It i poible that thi tatitically tronger reult i driven by the repone of conumer entiment and per capita GDP, which now react negatively and tatitically ignificantly different from zero to a hock in it overeign deficit-to-gdp ratio. 1 In Figure 10 we preent the impule repone of our model without time de-meaning and without uing fixed effect. The repone of the growth rate in rating to a hock in the deficit-to-gdp ratio i now even tronger. It i poible that thi i driven by the tronger repone of the per capita GDP growth to a hock in the deficit-to-gdp ratio. Since thi effect i abent when one time de-mean the data, then we conclude that thi reult may be driven by time-pecific effect. 1 The negative impact of the deficit-to-gdp ratio on GDP growth ha alo been found in Brückner and Pappa (2010) and Jueen and Linnemann (2012). 3

17 Reference Baltagi, B.H., Econometric analyi of panel data, Wiley, Brückner, M. and E. Pappa, Fical expanion affect unemployment, but they may increae it, Centre for Economic Policy Reearch, Cantor, R. and F. Packer, Determinant and impact of overeign credit rating, Federal Reerve Bank of New York New York, Jueen, F. and L. Linnemann, Government pending and unemployment in the OECD: Evidence from an annual panel VAR, TU Dortmund Univerity, hal , verion 1-9 Feb

18 hal , verion 1-9 Feb 2012 Figure 1: Robutne of impule repone: Excluding recent crii (i.e. T < 2008) rep. of d(gd/gdp) to d(gd/gdp) rep. of d(gd/gdp) to g(gdppc) rep. of d(gd/gdp) to d(cci) rep. of d(gd/gdp) to g(r) rep. of g(gdppc) to d(gd/gdp) rep. of g(gdppc) to g(gdppc) rep. of g(gdppc) to d(cci) rep. of g(gdppc) to g(r) rep. of d(cci) to d(gd/gdp) rep. of d(cci) to g(gdppc) rep. of d(cci) to d(cci) rep. of d(cci) to g(r) rep. of g(r) to d(gd/gdp) rep. of g(r) to g(gdppc) rep. of g(r) to d(cci) rep. of g(r) to g(r) 5

19 hal , verion 1-9 Feb 2012 Figure 2: Robutne of impule repone: Changing lag (lag = 1) rep. of d(gd/gdp) to d(gd/gdp) rep. of d(gd/gdp) to g(gdppc) rep. of d(gd/gdp) to d(cci) rep. of g(gdppc) to d(gd/gdp) rep. of g(gdppc) to g(gdppc) rep. of g(gdppc) to d(cci) rep. of d(cci) to d(gd/gdp) rep. of d(cci) to g(gdppc) rep. of d(cci) to d(cci) rep. of g(r) to d(gd/gdp) rep. of g(r) to g(gdppc) rep. of g(r) to d(cci) rep. of d(gd/gdp) to g(r) rep. of g(gdppc) to g(r) rep. of d(cci) to g(r) rep. of g(r) to g(r) 6

20 hal , verion 1-9 Feb 2012 Figure 3: Robutne of impule repone: Changing lag (lag = 3) rep. of d(gd/gdp) to d(gd/gdp) rep. of d(gd/gdp) to g(gdppc) rep. of d(gd/gdp) to d(cci) rep. of g(gdppc) to d(gd/gdp) rep. of g(gdppc) to g(gdppc) rep. of g(gdppc) to d(cci) rep. of d(cci) to d(gd/gdp) rep. of d(cci) to g(gdppc) rep. of d(cci) to d(cci) rep. of g(r) to d(gd/gdp) rep. of g(r) to g(gdppc) rep. of g(r) to d(cci) rep. of d(gd/gdp) to g(r) rep. of g(gdppc) to g(r) rep. of d(cci) to g(r) rep. of g(r) to g(r) 7

21 hal , verion 1-9 Feb 2012 Figure 4: Robutne of impule repone: Order { g(r), d(cci), d(gd/gdp), g(gdppc) } rep. of g(r) to g(r) rep. of g(r) to d(cci) rep. of g(r) to d(gd/gdp) rep. of g(r) to g(gdppc) rep. of d(cci) to g(r) rep. of d(cci) to d(cci) rep. of d(cci) to d(gd/gdp) rep. of d(cci) to g(gdppc) rep. of d(gd/gdp) to g(r) rep. of d(gd/gdp) to d(cci) rep. of d(gd/gdp) to d(gd/gdp) rep. of d(gd/gdp) to g(gdppc) rep. of g(gdppc) to g(r) rep. of g(gdppc) to d(cci) rep. of g(gdppc) to d(gd/gdp) rep. of g(gdppc) to g(gdppc) 8

22 hal , verion 1-9 Feb 2012 Figure 5: Robutne of impule repone: Order { d(cci), g(r), g(gdppc), d(gd/gdp) } rep. of d(cci) to d(cci) rep. of d(cci) to g(r) rep. of d(cci) to g(gdppc) rep. of d(cci) to d(gd/gdp) rep. of g(r) to d(cci) rep. of g(r) to g(r) rep. of g(r) to g(gdppc) rep. of g(r) to d(gd/gdp) rep. of g(gdppc) to d(cci) rep. of g(gdppc) to g(r) rep. of g(gdppc) to g(gdppc) rep. of g(gdppc) to d(gd/gdp) rep. of d(gd/gdp) to d(cci) rep. of d(gd/gdp) to g(r) rep. of d(gd/gdp) to g(gdppc) rep. of d(gd/gdp) to d(gd/gdp) 9

23 hal , verion 1-9 Feb 2012 Figure 6: Robutne of impule repone: Excluding countrie with no rating change rep. of d(gd/gdp) to d(gd/gdp) rep. of d(gd/gdp) to g(gdppc) rep. of d(gd/gdp) to d(cci) rep. of d(gd/gdp) to g(r) rep. of g(gdppc) to d(gd/gdp) rep. of g(gdppc) to g(gdppc) rep. of g(gdppc) to d(cci) rep. of g(gdppc) to g(r) rep. of d(cci) to d(gd/gdp) rep. of d(cci) to g(gdppc) rep. of d(cci) to d(cci) rep. of d(cci) to g(r) rep. of g(r) to d(gd/gdp) rep. of g(r) to g(gdppc) rep. of g(r) to d(cci) rep. of g(r) to g(r) 10

24 rep. of d(gd/gdp) to d(gd/gdp) rep. of g(gdppc) to d(gd/gdp) rep. of d(bop/gdp) to d(gd/gdp) rep. of HICP to d(gd/gdp) rep. of d(cci) to d(gd/gdp) rep. of g(r) to d(gd/gdp) hal , verion 1-9 Feb 2012 Figure 7: Robutne of impule repone: Additional control (p 5) d(bop/gdp) d(bop/gdp) (p 95) d(bop/gdp) (p 5) HICP HICP (p 95) HICP rep. of d(gd/gdp) to g(gdppc) rep. of d(gd/gdp) to d(bop/gdp) (p 5) d(bop/gdp) d(bop/gdp) (p 95) d(bop/gdp) rep. of d(gd/gdp) to HICP (p 5) HICP HICP (p 95) HICP rep. of d(gd/gdp) to d(cci) rep. of g(gdppc) to g(gdppc) rep. of g(gdppc) to d(bop/gdp) (p 5) d(bop/gdp) d(bop/gdp) (p 95) d(bop/gdp) rep. of g(gdppc) to HICP (p 5) HICP HICP (p 95) HICP rep. of g(gdppc) to d(cci) rep. of d(bop/gdp) to g(gdppc) rep. of d(bop/gdp) to d(bop/gdp) (p 5) d(bop/gdp) d(bop/gdp) (p 95) d(bop/gdp) rep. of d(bop/gdp) to HICP (p 5) HICP HICP (p 95) HICP rep. of d(bop/gdp) to d(cci) rep. of HICP to g(gdppc) rep. of HICP to d(bop/gdp) (p 5) d(bop/gdp) d(bop/gdp) (p 95) d(bop/gdp) rep. of HICP to HICP (p 5) HICP HICP (p 95) HICP rep. of HICP to d(cci) rep. of d(cci) to g(gdppc) rep. of d(cci) to d(bop/gdp) (p 5) d(bop/gdp) d(bop/gdp) (p 95) d(bop/gdp) rep. of d(cci) to HICP (p 5) HICP HICP (p 95) HICP rep. of d(cci) to d(cci) rep. of g(r) to g(gdppc) rep. of g(r) to d(bop/gdp) rep. of g(r) to HICP rep. of g(r) to d(cci) rep. of d(gd/gdp) to g(r) rep. of g(gdppc) to g(r) rep. of d(bop/gdp) to g(r) rep. of HICP to g(r) rep. of d(cci) to g(r) rep. of g(r) to g(r) 11

25 hal , verion 1-9 Feb 2012 Figure 8: Robutne of impule repone: Cutting 1% and 99% tail rep. of d(gd/gdp) to d(gd/gdp) rep. of d(gd/gdp) to g(gdppc) rep. of d(gd/gdp) to d(cci) rep. of g(gdppc) to d(gd/gdp) rep. of g(gdppc) to g(gdppc) rep. of g(gdppc) to d(cci) rep. of d(cci) to d(gd/gdp) rep. of d(cci) to g(gdppc) rep. of d(cci) to d(cci) rep. of g(r) to d(gd/gdp) rep. of g(r) to g(gdppc) rep. of g(r) to d(cci) rep. of d(gd/gdp) to g(r) rep. of g(gdppc) to g(r) rep. of d(cci) to g(r) rep. of g(r) to g(r) 12

26 hal , verion 1-9 Feb 2012 Figure 9: Robutne of impule repone: Without time de-meaning rep. of d(gd/gdp) to d(gd/gdp) rep. of d(gd/gdp) to g(gdppc) rep. of d(gd/gdp) to d(cci) rep. of g(gdppc) to d(gd/gdp) rep. of g(gdppc) to g(gdppc) rep. of g(gdppc) to d(cci) rep. of d(cci) to d(gd/gdp) rep. of d(cci) to g(gdppc) rep. of d(cci) to d(cci) rep. of g(r) to d(gd/gdp) rep. of g(r) to g(gdppc) rep. of g(r) to d(cci) rep. of d(gd/gdp) to g(r) rep. of g(gdppc) to g(r) rep. of d(cci) to g(r) rep. of g(r) to g(r) 13

27 hal , verion 1-9 Feb 2012 Figure 10: Robutne of impule repone: Without time de-meaning and without fixed effect rep. of d(gd/gdp) to d(gd/gdp) rep. of d(gd/gdp) to g(gdppc) rep. of d(gd/gdp) to d(cci) rep. of d(gd/gdp) to g(r) rep. of g(gdppc) to d(gd/gdp) rep. of g(gdppc) to g(gdppc) rep. of g(gdppc) to d(cci) rep. of g(gdppc) to g(r) rep. of d(cci) to d(gd/gdp) rep. of d(cci) to g(gdppc) rep. of d(cci) to d(cci) rep. of d(cci) to g(r) rep. of g(r) to d(gd/gdp) rep. of g(r) to g(gdppc) rep. of g(r) to d(cci) rep. of g(r) to g(r) 14