Online Appendix

Populism, Political Risk and the Economy:

Lessons from Italy

by Balduzzi P., Brancati E., Brianti M. and Schiantarelli F.

April 28, 2020

1 Robustness checks

This section contains a series of robustness checks for the results presented in the paper

both at a daily and a monthly frequency.

1.1 Daily frequency

At a daily frequency results remain unchanged if we allow for eight lags of the controls

instead of four. Moreover, the domestic and international results remain robust to using

either CDSITA03 or CDSITA14 as an instrument and indicator variable, denominated

in euros or dollars. Domestic results are also robust to removing from the list of selected

dates those for European elections and for the submission to the European Commission

of the draft budget because they are common to all the euro-zone countries (the baseline

results for the spillover effects – see Section 6.4 of the paper – are already derived

by excluding those dates). In addition, as a robustness exercise for both the CDS

spread and the 10-year bond yield spread relative to the Bund, we report results in

which we have been more drastic in reducing the list of dates using in constructing our

instrument. More specifically, we removed from our instrument all dates that fall in

1

a 2-sided window of seven days centered around election dates of other euro countries

(47 events in total), the Brexit referendum and other key events in the Brexit process

(32 additional events). The domestic daily results are also robust to this robustness

exercise, but to limit the length of the Online Appendix, we have only included the

results for the spillover effects. Furthermore, for all results at a daily frequency, the

estimated impulse response functions are virtually unchanged if we employ a Cholesky

identification strategy and order our instrument after the VIX and the first principal

component of euro-zone countries’ CDS spreads, and before the other financial variables.

Finally, the domestic results at a monthly frequency are invariant to using the average

of the last 5-days or the monthly average instead of the end of period observation.

1.1.1 Additional number of lags

At a daily frequency results remain unchanged if we allow for eight lags of the controls

instead of four. In this section we present both domestic and international results.

2

Figure 1: Financial variables: impulse responses at a daily fre-quency, 8 lags

Impulse response functions of financial variables to a political risk shock at a daily frequency.The solid black line is estimated via Local Projections - Instrumental Variables where theinstrument is the change in the CDS spread for the 2014-clause contract (CDSITA14) on theselected dates and the indicator variable is CDSITA14, denominated in dollars. Confidencebands are estimated with 2000 block-bootstrapped simulations. All the variables enters inthe LP-IV regressions in first differences. The estimated responses are then cumulated in thegraph above. In each regression, we control for 8 lags of the instrument and all the endogenousvariables and the present together with 7 lags of a measure of international volatility (VIX)and the first principal component of the change in the sovereign CDS spread of the 2014-clausecontract for euro countries, denominated in dollars.

3

Figure 2: Redenomination spread and quanto spread: impulse re-sponses and variance decomposition at a daily frequency, 8 lags

The first row shows impulse responses of redenomination spread and quanto spread to a politicalrisk shock at a daily frequency. The solid black line is estimated via Local Projections -Instrumental Variables where the instrument is the change in the CDS spread for the 2014-clause contract (CDSITA14) on the selected dates and the indicator variable is CDSITA14,denominated in dollars. In each regression, we control for 8 lags of the instrument and allthe endogenous variables and the present together with 7 lags of a measure of internationalvolatility (VIX) and the first principal component of the change in the sovereign CDS spreadof the 2014-clause contract for euro countries, denominated in dollars. Confidence bands areestimated with 2000 block-bootstrapped simulations. The second row shows the lower boundof the variance of redenomination spread and quanto spread explained by political risk shocks.

4

Figure 3: Spillover effects on sovereign CDS spreads for euro-zonecountries: impulse responses at a daily frequency, 8 lags

Impulse response functions of euro-zone country sovereign CDSs to a political risk shock ata daily frequency. All CDS contracts are denominated in dollars and use the 2014 clause.The solid black line is estimated via Local Projections - Instrumental Variables where theinstrument is the change in the CDS spread for the 2014-clause contract (CDSITA14) on theselected dates and the indicator variable is CDSITA14, denominated in dollars. Confidencebands are estimated with 2000 block-bootstrapped simulations. All the variables enters inthe LP-IV regressions in first differences. The estimated responses are then cumulated in thegraph above. In each regression, we control for 8 lags of the instrument and all the endogenousvariables and the present together with 7 lags of a measure of international volatility (VIX)and the first principal component of the change in the sovereign CDS spread of the 2014-clausecontract for euro countries, denominated in dollars.

5

Figure 4: Spillover effects on gov. bonds yields relative to the Bundfor euro-zone countries: impulse responses at a daily frequency, 8lags

Impulse response functions of the difference between the 10-year sovereign bond yield and the10-year bund yield of a series of euro-zone countries at a daily frequency. The solid blackline is estimated via Local Projections - Instrumental Variables where the instrument is thechange in the CDS spread for the 2014-clause contract (CDSITA14) on the selected dates andthe indicator variable is CDSITA14, denominated in dollars. Confidence bands are estimatedwith 2000 block-bootstrapped simulations. All the variables enters in the LP-IV regressionsin first differences. The estimated responses are then cumulated in the graph above. In eachregression, we control for 8 lags of the instrument and all the endogenous variables and thepresent together with 7 lags of a measure of international volatility (VIX) and the first principalcomponent of the change in the sovereign CDS spread of the 2014-clause contract for eurocountries, denominated in dollars.

1.1.2 Dollar-denominated CDSITA03 as an alternative instru-

ment

At a daily frequency, domestic and international results remain robust to using dollar-

denominated CDSITA03 as an instrument (and indicator variable).

6

Figure 5: Financial variables: impulse responses at a daily fre-quency, CDSITA03 USD as an instrument

Impulse response functions of financial variables to a political risk shock at a daily frequency.The solid black line is estimated via Local Projections - Instrumental Variables where theinstrument is the change in the CDS spread for the 2003-clause contract (CDSITA03) on theselected dates and the indicator variable is CDSITA03, denominated in dollars. Confidencebands are estimated with 2000 block-bootstrapped simulations. All the variables enters inthe LP-IV regressions in first differences. The estimated responses are then cumulated in thegraph above. In each regression, we control for 4 lags of the instrument and all the endogenousvariables and the present together with 3 lags of a measure of international volatility (VIX)and the first principal component of the change in the sovereign CDS spread of the 2014-clausecontract for euro countries, denominated in dollars.

7

Figure 6: Redenomination spread and quanto spread: impulse re-sponses and variance decomposition at a daily frequency, CDSITA03USD as an instrument

The first row shows impulse responses of redenomination spread and quanto spread to a politicalrisk shock at a daily frequency. The solid black line is estimated via Local Projections -Instrumental Variables where the instrument is the change in the CDS spread for the 2003-clause contract (CDSITA03) on the selected dates and the indicator variable is CDSITA03,denominated in dollars. In each regression, we control for 4 lags of the instrument and allthe endogenous variables and the present together with 3 lags of a measure of internationalvolatility (VIX) and the first principal component of the change in the sovereign CDS spreadof the 2014-clause contract for euro countries, denominated in dollars. Confidence bands areestimated with 2000 block-bootstrapped simulations. The second row shows the lower boundof the variance of redenomination spread and quanto spread explained by political risk shocks.

8

Figure 7: Spillover effects on sovereign CDS spreads for euro-zonecountries: impulse responses at a daily frequency, CDSITA03 USDas an instrument

Impulse response functions of euro-zone country sovereign CDSs to a political risk shock ata daily frequency. All CDS contracts are denominated in dollars and use the 2014 clause.The solid black line is estimated via Local Projections - Instrumental Variables where theinstrument is the change in the CDS spread for the 2003-clause contract (CDSITA03) on theselected dates and the indicator variable is CDSITA03, denominated in dollars. Confidencebands are estimated with 2000 block-bootstrapped simulations. All the variables enters inthe LP-IV regressions in first differences. The estimated responses are then cumulated in thegraph above. In each regression, we control for 4 lags of the instrument and all the endogenousvariables and the present together with 3 lags of a measure of international volatility (VIX)and the first principal component of the change in the sovereign CDS spread of the 2014-clausecontract for euro countries, denominated in dollars.

9

Figure 8: Spillover effects on gov. bonds yields relative to the Bundfor euro-zone countries: impulse responses at a daily frequency,CDSITA03 USD as an instrument

Impulse response functions of the difference between the 10-year sovereign bond yield and the10-year bund yield of a series of euro-zone countries at a daily frequency. The solid blackline is estimated via Local Projections - Instrumental Variables where the instrument is thechange in the CDS spread for the 2003-clause contract (CDSITA03) on the selected dates andthe indicator variable is CDSITA03, denominated in dollars. Confidence bands are estimatedwith 2000 block-bootstrapped simulations. All the variables enters in the LP-IV regressionsin first differences. The estimated responses are then cumulated in the graph above. In eachregression, we control for 4 lags of the instrument and all the endogenous variables and thepresent together with 3 lags of a measure of international volatility (VIX) and the first principalcomponent of the change in the sovereign CDS spread of the 2014-clause contract for eurocountries, denominated in dollars.

1.1.3 Euro-denominated CDSITA14 as an alternative instru-

ment

At a daily frequency, domestic and international results remain robust to using euro-

denominated CDSITA14 as an instrument (and indicator variable).

10

Figure 9: Financial variables: impulse responses at a daily fre-quency, CDSITA14 EURO as an instrument

Impulse response functions of financial variables to a political risk shock at a daily frequency.The solid black line is estimated via Local Projections - Instrumental Variables where theinstrument is the change in the CDS spread for the 2014-clause contract (CDSITA14) on theselected dates and the indicator variable is CDSITA14, denominated in euros. Confidencebands are estimated with 2000 block-bootstrapped simulations. All the variables enters inthe LP-IV regressions in first differences. The estimated responses are then cumulated in thegraph above. In each regression, we control for 4 lags of the instrument and all the endogenousvariables and the present together with 3 lags of a measure of international volatility (VIX)and the first principal component of the change in the sovereign CDS spread of the 2014-clausecontract for euro countries, denominated in dollars.

11

Figure 10: Redenomination spread and quanto spread: impulse re-sponses and variance decomposition at a daily frequency, CDSITA14EURO as an instrument

The first row shows impulse responses of redenomination spread and quanto spread to a politicalrisk shock at a daily frequency. The solid black line is estimated via Local Projections -Instrumental Variables where the instrument is the change in the CDS spread for the 2014-clause contract (CDSITA14) on the selected dates and the indicator variable is CDSITA14,denominated in euros. In each regression, we control for 4 lags of the instrument and all theendogenous variables and the present together with 3 lags of a measure of international volatility(VIX) and the first principal component of the change in the sovereign CDS spread of the 2014-clause contract for euro countries, denominated in dollars. Confidence bands are estimated with2000 block-bootstrapped simulations. The second row shows the lower bound of the varianceof redenomination spread and quanto spread explained by political risk shocks.

12

Figure 11: Spillover effects on sovereign CDS spreads for euro-zone countries: impulse responses at a daily frequency, CDSITA14EURO as an instrument

Impulse response functions of euro-zone country sovereign CDSs to a political risk shock ata daily frequency. All CDS contracts are denominated in dollars and use the 2014 clause.The solid black line is estimated via Local Projections - Instrumental Variables where theinstrument is the change in the CDS spread for the 2014-clause contract (CDSITA14) on theselected dates and the indicator variable is CDSITA14, denominated in euros. Confidence bandsare estimated with 2000 block-bootstrapped simulations. All the variables enters in the LP-IVregressions in first differences. The estimated responses are then cumulated in the graph above.In each regression, we control for 4 lags of the instrument and all the endogenous variables andthe present together with 3 lags of a measure of international volatility (VIX) and the firstprincipal component of the change in the sovereign CDS spread of the 2014-clause contract foreuro countries, denominated in dollars.

13

Figure 12: Spillover effects on gov. bonds yields relative to the Bundfor euro-zone countries: impulse responses at a daily frequency,CDSITA14 EURO as an instrument

Impulse response functions of the difference between the 10-year sovereign bond yield and the10-year bund yield of a series of euro-zone countries at a daily frequency. The solid black line isestimated via Local Projections - Instrumental Variables where the instrument is the change inthe CDS spread for the 2014-clause contract (CDSITA14) on the selected dates and the indicatorvariable is CDSITA14, denominated in euros. Confidence bands are estimated with 2000 block-bootstrapped simulations. All the variables enters in the LP-IV regressions in first differences.The estimated responses are then cumulated in the graph above. In each regression, we controlfor 4 lags of the instrument and all the endogenous variables and the present together with3 lags of a measure of international volatility (VIX) and the first principal component of thechange in the sovereign CDS spread of the 2014-clause contract for euro countries, denominatedin dollars.

1.1.4 Euro-denominated CDSITA03 as an alternative instru-

ment

At a daily frequency, domestic and international results remain robust to using euro-

denominated CDSITA03 as an instrument (and indicator variable).

14

Figure 13: Financial variables: impulse responses at a daily fre-quency, CDSITA03 EURO as an instrument

Impulse response functions of financial variables to a political risk shock at a daily frequency.The solid black line is estimated via Local Projections - Instrumental Variables where theinstrument is the change in the CDS spread for the 2003-clause contract (CDSITA03) on theselected dates and the indicator variable is CDSITA03, denominated in euros. Confidencebands are estimated with 2000 block-bootstrapped simulations. All the variables enters inthe LP-IV regressions in first differences. The estimated responses are then cumulated in thegraph above. In each regression, we control for 4 lags of the instrument and all the endogenousvariables and the present together with 3 lags of a measure of international volatility (VIX)and the first principal component of the change in the sovereign CDS spread of the 2014-clausecontract for euro countries, denominated in dollars.

15

Figure 14: Redenomination spread and quanto spread: impulse re-sponses and variance decomposition at a daily frequency, CDSITA03EURO as an instrument

The first row shows impulse responses of redenomination spread and quanto spread to a politicalrisk shock at a daily frequency. The solid black line is estimated via Local Projections -Instrumental Variables where the instrument is the change in the CDS spread for the 2003-clause contract (CDSITA03) on the selected dates and the indicator variable is CDSITA03,denominated in euros. In each regression, we control for 4 lags of the instrument and allthe endogenous variables and the present together with 3 lags of a measure of internationalvolatility (VIX) and the first principal component of the change in the sovereign CDS spreadof the 2014-clause contract for euro countries, denominated in dollars. Confidence bands areestimated with 2000 block-bootstrapped simulations. The second row shows the lower boundof the variance of redenomination spread and quanto spread explained by political risk shocks.

16

Figure 15: Spillover effects on sovereign CDS spreads for euro-zone countries: impulse responses at a daily frequency, CDSITA03EURO as an instrument

Impulse response functions of euro-zone country sovereign CDSs to a political risk shock ata daily frequency. All CDS contracts are denominated in dollars and use the 2014 clause.The solid black line is estimated via Local Projections - Instrumental Variables where theinstrument is the change in the CDS spread for the 2003-clause contract (CDSITA03) on theselected dates and the indicator variable is CDSITA03, denominated in euros. Confidence bandsare estimated with 2000 block-bootstrapped simulations. All the variables enters in the LP-IVregressions in first differences. The estimated responses are then cumulated in the graph above.In each regression, we control for 4 lags of the instrument and all the endogenous variables andthe present together with 3 lags of a measure of international volatility (VIX) and the firstprincipal component of the change in the sovereign CDS spread of the 2014-clause contract foreuro countries, denominated in dollars.

17

Figure 16: Spillover effects on gov. bonds yields relative to the Bundfor euro-zone countries: impulse responses at a daily frequency,CDSITA03 EURO as an instrument

Impulse response functions of the difference between the 10-year sovereign bond yield and the10-year bund yield of a series of euro-zone countries at a daily frequency. The solid blackline is estimated via Local Projections - Instrumental Variables where the instrument is thechange in the CDS spread for the 2003-clause contract (CDSITA03) on the selected dates andthe indicator variable is CDSITA03, denominated in euros. Confidence bands are estimatedwith 2000 block-bootstrapped simulations. All the variables enters in the LP-IV regressionsin first differences. The estimated responses are then cumulated in the graph above. In eachregression, we control for 4 lags of the instrument and all the endogenous variables and thepresent together with 3 lags of a measure of international volatility (VIX) and the first principalcomponent of the change in the sovereign CDS spread of the 2014-clause contract for eurocountries, denominated in dollars.

1.1.5 Experimenting with the selection of dates

Domestic results are also robust to removing from the list of selected domestic dates

those overlapping European elections or the dates of submission of the draft budget

to the European Commission. Because the deadline is common to all the euro-zone

countries, the submission of the draft budget may be shared by other European coun-

tries. In addition, international results are very similar if we further reduce the list

of selected dates in constructing our instrument. More specifically, we removed from

our instrument all dates that fall in a 2-sided window of seven days centered around

election dates of other euro countries (47 events in total), the Brexit referendum and

other key events in the Brexit process (32 additional events). This is also true for the

18

domestic daily results for Italy. To limit the length of the Online Appendix we have

only included this robustness exercise for the spillover effects.

Figure 17: Financial variables: impulse responses at a daily fre-quency, excluding common EU dates

Impulse response functions of financial variables to a political risk shock at a daily frequency.The solid black line is estimated via Local Projections - Instrumental Variables where theinstrument is the change in the CDS spread for the 2014-clause contract (CDSITA14) onthe selected dates minus eight dates common to the ones of other euro-zone countries andthe indicator variable is CDSITA14, denominated in dollars. Confidence bands are estimatedwith 2000 block-bootstrapped simulations. All the variables enters in the LP-IV regressionsin first differences. The estimated responses are then cumulated in the graph above. Ineach regression, we control for 4 lags of the instrument and all the endogenous variables andthe present together with 3 lags of a measure of international volatility (VIX) and the firstprincipal component of the change in the sovereign CDS spread of the 2014-clause contractfor euro countries, denominated in dollars.

19

Figure 18: Redenomination spread and quanto spread: impulse re-sponses and variance decomposition at a daily frequency, excludingcommon EU dates

The first row shows impulse responses of redenomination spread and quanto spread to a po-litical risk shock at a daily frequency. The solid black line is estimated via Local Projections- Instrumental Variables where the instrument is the change in the CDS spread for the 2014-clause contract (CDSITA14) on the selected dates minus eight dates common to the ones ofother euro-zone countries and the indicator variable is CDSITA14, denominated in dollars. Ineach regression, we control for 4 lags of the instrument and all the endogenous variables and thepresent together with 3 lags of a measure of international volatility (VIX) and the first principalcomponent of the change in the sovereign CDS spread of the 2014-clause contract for euro coun-tries, denominated in dollars. Confidence bands are estimated with 2000 block-bootstrappedsimulations. The second row shows the lower bound of the variance of redenomination spreadand quanto spread explained by political risk shocks.

20

Figure 19: Spillover effects on sovereign CDS spreads for euro-zonecountries: impulse responses at a daily frequency, excluding thosethat are close to political dates for other European countries

Impulse response functions of euro-zone country sovereign CDSs to a political risk shock at adaily frequency. All CDS contracts are denominated in dollars and use the 2014 clause. Thesolid black line is estimated via Local Projections - Instrumental Variables where the instrumentis the change in the CDS spread for the 2014-clause contract (CDSITA14) on the selected datesminus 15 dates that fall within a week-long 2-sided window around political dates for Europeancountries. The indicator variable is CDSITA14, denominated in dollars. Confidence bands areestimated with 2000 block-bootstrapped simulations. All the variables enters in the LP-IVregressions in first differences. The estimated responses are then cumulated in the graph above.In each regression, we control for 4 lags of the instrument and all the endogenous variables andthe present together with 3 lags of a measure of international volatility (VIX) and the firstprincipal component of the change in the sovereign CDS spread of the 2014-clause contract foreuro countries, denominated in dollars.

21

Figure 20: Spillover effects on gov. bonds yields relative to the Bundfor euro-zone countries: impulse responses at a daily frequency,excluding those that are close to political dates for other Europeancountries

Impulse response functions of the difference between the 10-year sovereign bond yield and the10-year bund yield of a series of euro-zone countries at a daily frequency. The solid black line isestimated via Local Projections - Instrumental Variables where the instrument is the change inthe CDS spread for the 2014-clause contract (CDSITA14) on the selected dates minus 15 datesthat fall within a week-long 2-sided window around political dates for European countries andthe indicator variable is CDSITA14, denominated in dollars. Confidence bands are estimatedwith 2000 block-bootstrapped simulations. All the variables enters in the LP-IV regressionsin first differences. The estimated responses are then cumulated in the graph above. In eachregression, we control for 4 lags of the instrument and all the endogenous variables and thepresent together with 3 lags of a measure of international volatility (VIX) and the first principalcomponent of the change in the sovereign CDS spread of the 2014-clause contract for eurocountries, denominated in dollars.

1.1.6 Cholesky identification

At a daily frequency, the estimated impulse response functions obtained with LP-IV are

similar to those obtained by putting our instrument after the VIX and the first principal

component of euro-zone countries’ CDS spreads, and before the other financial variables,

and using a Cholesky identification strategy.

22

Figure 21: Financial variables: impulse responses at a daily fre-quency, Cholesky identification

Impulse response functions of domestic financial variables to a political risk shock at a dailyfrequency. The solid black line is estimated via Cholesky where the order is: (i) the VIX, (ii)the first principal component of the change in the sovereign dollar-denominated CDS spread ofthe 2014-clause contract for euro countries, (iii) our instrument (the change in the CDS spreadfor the 2014-clause contract on the selected dates), (iv) the indicator variable (CDSITA14,denominated in dollars), and (v) the above endogenous variables. In the reduced-form VARwe control for 4 lags. Confidence bands are estimated with 2000 bootstrapped simulations. Allthe variables, except for the instrument enters, in the VAR in first differences. The estimatedresponses are then cumulated in the graph above.

23

Figure 22: Redenomination spread and quanto spread: impulse re-sponses and variance decomposition at a daily frequency, Choleskyidentification

Impulse response functions of redenomination spread and quanto spread to a political riskshock at a daily frequency. The solid black line is estimated via Cholesky where the order is:(i) the VIX, (ii) the first principal component of the change in the sovereign dollar-denominatedCDS spread of the 2014-clause contract for euro countries, (iii) our instrument (the change inthe CDS spread for the 2014-clause contract on the selected dates), (iv) the indicator variable(CDSITA14, denominated in dollars), and (v) the above endogenous variables. In the reduced-form VAR we control for 4 lags. Confidence bands are estimated with 2000 bootstrappedsimulations. All the variables, except for the instrument enters, in the VAR in first differences.The estimated responses are then cumulated in the graph above.

24

Figure 23: Spillover effects on sovereign CDS spreads for euro-zone countries: impulse responses at a daily frequency, Choleskyidentification

Impulse response functions of euro-zone country sovereign CDSs to a political risk shock at adaily frequency. All CDS contracts are denominated in dollars and use the 2014 clause. Thesolid black line is estimated via Cholesky where the order is: (i) the VIX, (ii) the first principalcomponent of the change in the sovereign dollar-denominated CDS spread of the 2014-clausecontract for euro countries, (iii) our instrument (the change in the CDS spread for the 2014-clause contract on the selected dates), (iv) the indicator variable (CDSITA14, denominated indollars), and (v) the above endogenous variables. In the reduced-form VAR we control for 4lags. Confidence bands are estimated with 2000 bootstrapped simulations. All the variables,except for the instrument enters, in the VAR in first differences. The estimated responses arethen cumulated in the graph above.

25

Figure 24: Spillover effects on gov. bonds yields relative to theBund for euro-zone countries: impulse responses at a daily fre-quency, Cholesky identification

Impulse response functions of the difference between the 10-year sovereign bond yield andthe 10-year bund yield of a series of euro-zone countries at a daily frequency. The solidblack line is estimated via Cholesky where the order is: (i) the VIX, (ii) the first principalcomponent of the change in the sovereign dollar-denominated CDS spread of the 2014-clausecontract for euro countries, (iii) our instrument (the change in the CDS spread for the 2014-clause contract on the selected dates), (iv) the indicator variable (CDSITA14, denominated indollars), and (v) the above endogenous variables. In the reduced-form VAR we control for 4lags. Confidence bands are estimated with 2000 bootstrapped simulations. All the variables,except for the instrument enters, in the VAR in first differences. The estimated responses arethen cumulated in the graph above.

26

1.2 Monthly frequency

The domestic results at a monthly frequency are invariant to using the average of

observations for the last 5-days of the month or the monthly average, instead of the

end of the month observation for the endogenous variables.

Figure 25: Financial variables: impulse responses at a monthlyfrequency, mean of last 5 observations

Impulse response functions of financial variables to a political risk shock at a monthly fre-quency. The solid black line is estimated via Local Projections - Instrumental Variables wherethe instrument is the change in the CDS spread for the 2003-clause contract (CDSITA03) onthe selected dates and the indicator variable is CDSITA03, denominated in dollars. Confi-dence bands are estimated with 2000 block-bootstrapped simulations. All the variables entersin the LP-IV regressions in first differences. The estimated responses are then cumulated inthe graph above.

27

Figure 26: Redenomination spread and quanto spread: impulse re-sponses and variance decomposition at a monthly frequency, meanof last 5 observations

First row shows impulse responses of redenomination spread and quanto spread to a politicalrisk shock at a monthly frequency. The solid black line is estimated via Local Projections - In-strumental Variables where the instrument is the change in the CDS spread for the 2003-clausecontract (CDSITA03) on the selected dates and the indicator variable is CDSITA03, denom-inated in dollars. Confidence bands are estimated with 2000 block-bootstrapped simulations.Second row shows the lower bound of the variance of redenomination spread and quanto spreadexplained by political risk shocks.

28

Figure 27: Financial variables: impulse responses at a monthlyfrequency, mean across the month

Impulse response functions of financial variables to a political risk shock at a monthly fre-quency. The solid black line is estimated via Local Projections - Instrumental Variables wherethe instrument is the change in the CDS spread for the 2003-clause contract (CDSITA03) onthe selected dates and the indicator variable is CDSITA03, denominated in dollars. Confi-dence bands are estimated with 2000 block-bootstrapped simulations. All the variables entersin the LP-IV regressions in first differences. The estimated responses are then cumulated inthe graph above.

29

Figure 28: Redenomination spread and quanto spread: impulse re-sponses and variance decomposition at a monthly frequency, meanacross the month

First row shows impulse responses of redenomination spread and quanto spread to a politicalrisk shock at a monthly frequency. The solid black line is estimated via Local Projections - In-strumental Variables where the instrument is the change in the CDS spread for the 2003-clausecontract (CDSITA03) on the selected dates and the indicator variable is CDSITA03, denom-inated in dollars. Confidence bands are estimated with 2000 block-bootstrapped simulations.Second row shows the lower bound of the variance of redenomination spread and quanto spreadexplained by political risk shocks.

30

2 Placebo test

In this section, we conduct a standard placebo test in which we apply our IV-LP

procedure to a randomly selected set of dates equal in number to those include in our

own original set. We then repeat this procedure 2000 times and present the 2.5th

(5th) and 97.5 (95th) percentile for the impulse response functions obtained using the

change of the CDS spread on the randomly selected dates as an external instrument

in the same local projection context. The solid black line is the median. Both 90th

and 95th confidence intervals include the zero at all horizons of the impulse response

functions for all the variables, with one exception. The exception is the response of

the change in the spread of the sovereign CDS on impact which is significant at the

10% level but not at the 5%. Note however, that the CDS is our indicator variable and

by construction its coefficient on impact is normalized to be one and basically we are

regressing the change in the CDS spread on itself on a subset of dates. Therefore this

finding is neither surprising nor worrisome. In sum, the placebo test suggests that our

results are not driven by background shocks we do not control for.

31

Figure 29: Financial variables: impulse responses at a daily fre-quency; placebo

Impulse response functions of financial variables to a political risk shock at a daily frequency.The solid black line is estimated via Local Projections - Instrumental Variables where theinstrument is the change in the CDS spread for the 2014-clause contract (CDSITA14) onrandom dates and the indicator variable is CDSITA14, denominated in dollars. The estimatedresponses are then cumulated in the graph above. In each regression, we control for 4 lagsof the instrument and all the endogenous variables and the present together with 3 lags ofa measure of international volatility (VIX) and the first principal component of the changein the sovereign CDS spread of the 2014-clause contract for euro countries, denominated indollars. The exercise is repeated 2000 times in order to select confidence interval and pointestimate (median).

32

Figure 30: Financial variables: impulse responses at a monthlyfrequency; placebo

Impulse response functions of financial variables to a political risk shock at a monthly frequency.The solid black line is estimated via Local Projections - Instrumental Variables where theinstrument is the change in the CDS spread for the 2003-clause contract (CDSITA03) onrandom dates and the indicator variable is CDSITA03, denominated in dollars. The estimatedresponses are then cumulated in the graph above. The exercise is repeated 2000 times in orderto select confidence interval and point estimate (median).

33

Figure 31: Redenomination spread and quanto spread: impulseresponses and variance decomposition at a daily frequency; placebo

The first row shows impulse responses of redenomination spread and quanto spread to a politicalrisk shock at a daily frequency. The solid black line is estimated via Local Projections - In-strumental Variables where the instrument is the change in the CDS spread for the 2014-clausecontract (CDSITA14) on random dates and the indicator variable is CDSITA14, denominatedin dollars. We control for 4 lags of the instrument and all the endogenous variables and thepresent together with 3 lags of a measure of international volatility (VIX) and the first prin-cipal component of the change in the sovereign CDS spread of the 2014-clause contract foreuro countries, denominated in dollars. The exercise is repeated 2000 times in order to selectconfidence interval and point estimate (median).

34

Figure 32: Redenomination spread and quanto spread: impulseresponses at a monthly frequency; placebo

First row shows impulse responses of redenomination spread and quanto spread to a politicalrisk shock at a monthly frequency. The solid black line is estimated via Local Projections - In-strumental Variables where the instrument is the change in the CDS spread for the 2003-clausecontract (CDSITA03) on random dates and the indicator variable is CDSITA03, denominatedin dollars. The exercise is repeated 2000 times in order to select confidence interval and pointestimate (median).

35

Figure 33: Spillover effects on sovereign CDS spreads for euro-zonecountries: impulse responses at a daily frequency; placebo

Impulse response functions of euro-zone country sovereign CDSs to a political risk shock at adaily frequency. All CDS contracts are denominated in dollars and use the 2014 clause. Thesolid black line is estimated via Local Projections - Instrumental Variables where the instrumentis the change in the CDS spread for the 2014-clause contract (CDSITA14) on random datesand the indicator variable is CDSITA14, denominated in dollars. All the variables enters inthe LP-IV regressions in first differences. The estimated responses are then cumulated in thegraph above. In each regression, we control for 4 lags of the instrument and all the endogenousvariables and the present together with 3 lags of a measure of international volatility (VIX)and the first principal component of the change in the sovereign CDS spread of the 2014-clausecontract for euro countries, denominated in dollars. The exercise is repeated 2000 times inorder to select confidence interval and point estimate (median).

36

Figure 34: Spillover effects on gov. bonds yields relative to theBund for euro-zone countries: impulse responses at a daily frequency;placebo

Impulse response functions of the difference between the 10-year sovereign bond yield and the10-year bund yield of a series of euro-zone countries at a daily frequency. The solid blackline is estimated via Local Projections - Instrumental Variables where the instrument is thechange in the CDS spread for the 2014-clause contract (CDSITA14) on random dates and theindicator variable is CDSITA14, denominated in dollars. All the variables enters in the LP-IVregressions in first differences. The estimated responses are then cumulated in the graph above.In each regression, we control for 4 lags of the instrument and all the endogenous variablesand the present together with 3 lags of a measure of international volatility (VIX) and the firstprincipal component of the change in the sovereign CDS spread of the 2014-clause contract foreuro countries, denominated in dollars. The exercise is repeated 2000 times in order to selectconfidence interval and point estimate (median).

37

Figure 35: Real variables: impulse responses at a monthly fre-quency; placebo

Impulse response functions of real variables to a political risk shock at a monthly frequency. Thesolid black line is estimated via Local Projections - Instrumental Variables where the instrumentis the change in the CDS spread for the 2003-clause contract (CDSITA03) on random datesand the indicator variable is CDSITA03, denominated in dollars. The endogenous variablesare the log-transformation of the Purchasing Manager Index of the manufacturing sector (PMIManufacturing), the log-difference between the Italian PMI Manufacturing and the Global PMIManufacturing, the level of the Composite Leading Indicator from OECD database (OECDCLI), and the log-trasformation of a survey of firms’ confidence (Firm Confidence). For moreinformation on the sources and interoperability of those variables see Appendix A. Results areshown using different detrending techniques: (i) BP Filter is the High Pass filter removingperiodicities above 24 frequencies; (ii) Quadratic Trend is a standard time quadratic trend; (iii)Level is variables without being treated and controlling for the past value of the dependentvariable in each regression. The exercise is repeated 2000 times in order to select confidenceinterval and point estimate (median).

38

3 Other

Figure 36: Distribution of the difference between impact and 4thday response coefficient

Distribution of the difference between the impact and the 4th day response of the dollar-denominated CDSITA14 constructed using 2000 block-bootstrap replications.

39

Table 2: First stage regressions

Dependent variable ∆CDSITA14(1) (2) (3) (4)

Instrument(CDSITA14) 0.732*** 1.113*** 0.723***(0.0908) (0.295) (0.116)

Instrument(CDSITA03) -0.570 0.544***(0.428) (0.172)

Instrument(CDSITA14-CDSITA03) 1.113***(0.295)

Instrument(FTSE MIB) 0.000224(0.00594)

Instrument(IV FTSE MIB) 0.000302(0.00126)

Additional controls Yes Yes Yes Yes# obs. 1219 1219 1219 1219R2 0.524 0.533 0.533 0.539F-test main 65.0 14.2 34.1 38.9F-test additional (combined) – 1.77 – 0.03

OLS estimates for first stage regressions. The dependent variable is the daily changein CDS spreads on Italian sovereign bond under the 2014-clause (columns 1 to 4) or2003 clause (columns 5 and 6, used for the construction of the monthly instrument).Instrument(X) is the instrument constructed as the daily change in the reference seriesX around our selected dates (see section 3 for details). Untabulated additional controlsin each regression include: contemporaneous values and 3 lags of the VIX index andof PC∆CDS14 (first principal component of the change in the CDS spreads for eurocountries, excluding Greece and Italy, plus the UK); 4 lags of each instrument used; 4lags of the dependent variable; 4 lags of the FTSE MIB and its implied volatility; 4lags of the BTP-BUND spread at 5 and 10 years, 4 lags of banks’ CDS spread index.“F-test main” reports the F test of the joint significance of the instrument used ineach specification: Instrument(CDSITA14) in columns 1, 2, and 4; the combination ofInstrument(CDSITA14-CDSITA03) and Instrument(CDSITA03) in column 3; Instru-ment(CDSITA03) in columns 5 and 6. “F-test additional (combined)” reports the F-teston the joint significance of the additional instruments tested: Instrument(CDSITA03)in column 3; the combination of Instrument(FTSE MIB) and Instrument(IV FTSEMIB) in columns 4 and 6. Standard error in parentheses. *, **, *** indicate statisticalsignificance at the 10%, 5%, and 1%, respectively.

40

Tab

le1:

Inte

rnat

ional

com

par

ison

Rea

lG

DP

Nom

inal

GD

PD

ebt

Su

rplu

sP

rim

ary

Su

rplu

sIn

tere

stP

aym

ent

Mu

ltif

act

or

Pro

du

ct.

All

(2002-2018)

Ger

many

1.3

02.5

970.0

-1.0

81.1

22.2

10.6

2Ir

elan

d4.6

96.2

764.1

-4.7

4-2

.56

2.1

81.0

3G

reec

e-0

.12

1.3

0141.5

-6.9

0-2

.27

4.6

3-0

.72

Sp

ain

1.5

23.2

968.4

-4.0

8-1

.72

2.3

6-0

.34

Fra

nce

1.2

32.5

581.3

-3.9

5-1

.51

2.4

4-0

.11

Italy

0.1

11.8

2119.4

-3.0

91.4

44.5

2-0

.04

Port

ugal

0.5

52.4

699.5

-5.2

6-1

.72

3.5

50.3

4

Pre-Crisis

(2002-2007)

Ger

many

1.2

92.3

864.3

-2.6

00.2

22.8

20.8

2Ir

elan

d5.2

98.3

527.1

0.9

72.0

71.1

00.9

8G

reec

e4.0

47.3

6103.9

-6.9

0-2

.08

4.8

21.1

4S

pain

3.3

67.4

043.6

0.7

32.7

01.9

7-0

.51

Fra

nce

1.8

73.9

664.5

-3.2

0-0

.43

2.7

70.0

2It

aly

0.9

93.6

3105.7

-3.1

01.6

74.7

7-0

.16

Port

ugal

1.0

84.3

768.3

-4.7

3-1

.98

2.7

50.6

5

Crisis

(2008-2012)

Ger

many

0.7

71.9

576.4

-1.7

20.8

02.5

20.2

1Ir

elan

d-1

.42

-2.2

584.2

-14.8

-12.0

2.7

41.0

8G

reec

e-5

.31

-3.7

5142.8

-11.1

-5.4

25.7

2-3

.62

Sp

ain

-1.2

8-0

.81

61.9

-9.1

6-7

.02

2.1

4-0

.82

Fra

nce

0.3

71.5

083.1

-5.5

2-2

.88

2.6

4-0

.43

Italy

-1.3

60.1

5117.6

-3.6

81.0

04.6

8-0

.18

Port

ugal

-1.3

6-0

.80

101.4

-7.7

8-4

.14

3.6

4-0

.35

Post-Crisis

(2013-2018)

Ger

many

1.7

63.3

570.5

0.9

72.3

01.3

30.8

1Ir

elan

d9.1

711.3

84.4

-2.1

00.6

82.7

81.0

8G

reec

e0.0

5-0

.54

178.0

-3.3

70.1

73.5

3-0

.04

Sp

ain

2.0

12.6

198.5

-4.6

5-1

.72

2.9

30.2

4F

ran

ce1.2

92.0

196.5

-3.4

0-1

.45

1.9

50.0

7It

aly

0.4

51.4

0134.5

-2.5

81.5

74.1

50.3

1P

ort

ugal

1.6

13.2

6129.2

-3.7

00.5

74.2

70.6

0

Rea

lG

DP

isth

egro

wth

rate

of

GD

Pat

chain

lin

ked

pri

ces

(2010),

Nom

inal

GD

Pis

the

gro

wth

rate

of

GD

Pat

curr

ent

pri

ces,

Deb

tis

the

gover

nm

ent

deb

tto

GD

Pra

tio,

Su

rplu

sis

the

tota

lgover

nm

ent

surp

lus

(defi

cit

ifn

egati

ve)

toG

DP

rati

o,

Pri

mary

Surp

lus

isth

egover

nm

ent

pri

mary

surp

lus

(bef

ore

inte

rest

exp

an

ses,

defi

cit

ifn

egati

ve)

toG

DP

rati

o,

Inte

rest

Paym

ent

isth

era

tio

bet

wee

nin

tere

stp

ayed

on

deb

tan

dG

DP

,M

ult

ifact

or

Pro

du

ct.

isth

ean

nu

al

gro

wth

rate

of

mu

ltif

act

or

pro

du

ctiv

ity.

Th

efo

ur

pan

els

rep

ort

aver

ages

of

yea

rly

data

on

the

per

iod

of

inte

rest

spec

ified

inth

efi

rst

colu

mn

.A

uth

ors

’ca

lcu

lati

on

on

Eu

rost

at

an

dO

EC

Dd

ata

.

41

Figure 37: Financial variables: impulse responses and forecast errorvariance decomposition at a monthly frequency using Implied Volatil-ity FTSE as an instrument

First two rows show impulse response functions of financial variables to a political risk shock at amonthly frequency. The solid black line is estimated via Local Projections - Instrumental Variableswhere the instrument is the log-change in the implied volatility of the FTSE on the selected datesand the indicator variable is the implied volatility of the FTSE. Confidence bands are estimatedwith 2000 block-bootstrapped simulations. All the variables enters in the LP-IV regressions in firstdifferences. The estimated responses are then cumulated in the graph above. Third and fourthshow the respective forecast error variance decomposition.

42

Figure 38: Financial variables: impulse responses and forecast errorvariance decomposition at a monthly frequency using FTSE as an in-strument

First two rows show impulse response functions of financial variables to a political risk shockat a monthly frequency. The solid black line is estimated via Local Projections - InstrumentalVariables where the instrument is the log-change in the FTSE on the selected dates and the indicatorvariable is the FTSE. Confidence bands are estimated with 2000 block-bootstrapped simulations.All the variables enters in the LP-IV regressions in first differences. The estimated responses arethen cumulated in the graph above. Third and fourth show the respective forecast error variancedecomposition.

43

Figure 39: Real variables: impulse responses at a monthly frequencyusing Implied Volatility FTSE as an instrument

Impulse response functions of real variables to a political risk shock at a monthly frequency. Thesolid black line is estimated via Local Projections–Instrumental Variables where the instrumentis the log-change of Implied Volatility of the FTSE on the selected dates and the indicatorvariable is Implied Volatility of the FTSE. The endogenous variables are the log-transformationof the Purchasing Manager Index of the manufacturing sector (PMI Manufacturing), the log-difference between the Italian PMI Manufacturing and the Global PMI Manufacturing, thelevel of the Composite Leading Indicator from OECD database (OECD CLI), and the log-trasformation of a survey of firms’ confidence (Firm Confidence). For the sources and definitionsof those variables see Appendix A. In each regression, we control for one lag of the endogenousvariable under consideration and one lag of the instrument. Results are shown using differentdetrending techniques: (i) BP Filter is the High Pass filter removing periodicities above 24frequencies; (ii) Quadratic Trend is a standard time quadratic trend; (iii) Level is variableswithout being treated and controlling for the past value of the dependent variable in eachregression. Confidence bands are estimated with 2000 block-bootstrapped simulations.

44

Figure 40: Real variables: impulse responses at a monthly frequencyusing FTSE as an instrument

Impulse response functions of real variables to a political risk shock at a monthly frequency.The solid black line is estimated via Local Projections–Instrumental Variables where the in-strument is the log-change of the FTSE on the selected dates and the indicator variable isthe FTSE. The endogenous variables are the log-transformation of the Purchasing ManagerIndex of the manufacturing sector (PMI Manufacturing), the log-difference between the ItalianPMI Manufacturing and the Global PMI Manufacturing, the level of the Composite LeadingIndicator from OECD database (OECD CLI), and the log-trasformation of a survey of firms’confidence (Firm Confidence). For the sources and definitions of those variables see AppendixA. In each regression, we control for one lag of the endogenous variable under considerationand one lag of the instrument. Results are shown using different detrending techniques: (i) BPFilter is the High Pass filter removing periodicities above 24 frequencies; (ii) Quadratic Trendis a standard time quadratic trend; (iii) Level is variables without being treated and controllingfor the past value of the dependent variable in each regression. Confidence bands are estimatedwith 2000 block-bootstrapped simulations.

45