Crisis and Citizens’ Trust in the European Central Bank — Panel Data Evidence for the Euro Area, 1999–2012

ABSTRACT

Throughout the crisis, citizens’ trust in the European Central Bank has significantly declined throughout the Euro area (EA-12). Although a decline in the core countries of the EA-12 has been distinct, a more pronounced decline has been taking place in the peripheral countries of the EA-12. Taking panel data and using a fixed effects DFGLS estimation for an EA-12 country sample over the time period of 1999–2012 with a total of 305 observations, this paper detects a negative and significant relationship between unemployment and trust in times of crisis. The robustness analysis of the paper confirms that this decrease in trust is strongly driven by the significant increase in unemployment rates in the four peripheral countries Spain, Ireland, Greece and Portugal.

KEY WORDS:

• Crisis
• trust
• unemployment
• European Central Bank

Acknowledgements

Felix Roth wants to thank the Verein für Socialpolitik for scheduling the special session ‘Trusting Banks in a Financial Crisis’ at its annual conference on 8 September 2010 in Kiel, which gave him the opportunity to present a preliminary version of the paper. He would like to thank the seminar participants of that session for their valuable comments. In addition, the authors would like to thank the participants of the 14^th Göttinger Workshop ‘Internationale Wirtschaftsbeziehungen’ at the University of Göttingen in February 2012, the fourth IFABS Conference in Valencia in June 2012, and the seventh Annual International Symposium on Economic Theory, Policy and Applications in Athens in July 2012. The authors are grateful to Lars Jonung, the guest editor team from the Mannheim Centre for European Social Research Jale Tosun, Anne Wetzel and Galina Zapryanova, as well as two anonymous referees for their valuable comments. Two preliminary working paper versions of this paper were published as CEPS Working Document 334 on 26 July 2010 and as CEGE Discussion Paper 124 at the University of Göttingen in May 2011.

Notes

1. Standard EB surveys include about 1000 respondents per EU country. The interviews are performed face-to-face in the home of the respondents. For each standard EB survey, new and independent samples are derived. To guarantee the polling of a representative sample of the population, the sampling design is multi-stage and random. The raw data are available on CD-ROM from Gesis ZA Data Service for Standard EBs 51-62 (Gesis 2005a, 2005b) and were received on request from Gesis ZA Data Service for Standard EBs 63-69 (Gesis 2009). Data for the Standard EBs 70-78 and Special EB 71.1 were taken from the European Commission’s (EC) tables of results (2009a, 2009b, 2009c, 2010a, 2010b, 2011a, 2011b, 2012a, 2012b). Following Jones (2009) and Ehrmann, Soudan, and Stracca (2013), the observations from the Special EB 71.1 in 1-2/2009 were taken into consideration. For a detailed reasoning, see Roth, Nowak-Lehmann D., and Otter (2013, 4). The elimination of data from EB71.1 does not modify the econometric results in any significant way (see results in row 14 in Table 3).

2. A net trust measure seemed adequate as the ‘Don’t Know’ answers varied over a wide range from 0 per cent in Greece in EB 71 to 44.6 per cent in Portugal in EB 51 with an overall mean value of 20.5 per cent. However, it should be pointed out that net trust and trust measures correlate as high as 0.92. For an equation showing how to calculate net trust, see Roth, Nowak-Lehmann D., and Otter (2013, 4).

3. Although the monthly matching methodology by Wälti (2012, 597) correlates as high as 0.99 for the variables unemployment and inflation and 0.95 for the variable growth of GDP per capita, when comparing it to a semester matching methodology, the monthly methodological approach seems to be more adequate to prevent any potential overlap between the explanatory macro-economic variables and the EB data. The exact months of polling for the EBs surveys are displayed in the legend of the x-axis in Figure 1.

4. GDP data were chain-linked, the reference year being 2005, and seasonally adjusted. Data on GDP were missing for Greece from the 2nd quarter of 2011 onwards.

5. Due to inconsistent data on population size and breaks in some country time series within the official Eurostat data, values had to be exchanged by means of interpolation whenever required.

6. Possible measurement errors from the performed interpolation seem improbable as the monthly constructed variables correlate with the semester data as high as 0.95 for growth of GDP per capita.

7. For Greece, time trend data from 2001 onwards were taken. The five countries Slovakia, Slovenia, Malta, Cyprus and Estonia were not analysed as their accession occurred only recently and thus time trend data would not have been available from 1999 onwards.

8. For reasons of adequacy population-weighted trust trends are utilized for the EA-12, EA-4 and EA-8 country sample. However, population-weighted and non-population weighted aggregates are highly correlated.

9. With the exception for Greece’s decline in trust in the EC, trust in the ECB has decreased more significantly than trust in the EC and EP in all EA-12 countries from 3-5/2008 to 11/2012. In comparison to the EC and EP, the decrease in trust in the ECB is significantly higher (one standard deviation above the mean) in particular in the three core countries Germany, Netherlands and Finland. In those three countries, the additional trust decline varies from 29 to 38 percentage points of net trust with respect to the EC and 29–32 percentage points of net trust with respect to the EP with Germany showing the largest additional decline of 38 and 32 percentage points, respectively.

10. A pre-requisite for using the DOLS approach is that the variables entering the model are non-stationary and that all the series are in a long-run relationship (cointegrated). In our case, all series are integrated of order 1, i.e. they are I(1) (and thus non-stationary, non-stationarity of inflation and growth of GDP per capita is due to non-stationarity (non-constancy) of the variance of these series) and they are cointegrated. Results for the panel unit root tests and Kao’s residual cointegration test can be obtained from the authors upon request.

11. We found first-order autocorrelation to be present.

12. FGLS is not compatible with time-fixed effects but picks up shocks and their influence over short-to-medium term periods. In addition, the potential inclusion of time dummies would not alter our results in any significant manner (see results in row 15 in Table 3), and it could be shown that time-fixed effects do not tackle the problem of autocorrelation of the error term.

13. For a detailed explanation of all steps leading from Equation (1(1) ) to Equation (2(2) ) within a similar model specification, please see Roth, Nowak-Lehmann D., and Otter (2013, 12–4).

14. The coefficients , , and are linked to the endogenous part of the explanatory variables and do not result in a t-distribution. Since we are not interested in the influence of these ‘differenced variables’ on trust, they will not be depicted.

15. In addition to the theoretical validity of differentiating a pre-crisis from a crisis period, empirically, a Chow-test showed a structural break between the pre-crisis period (3-4/1999–3-5/2008) and the crisis period (10-11/2008–11/2012). Results can be obtained from the authors upon request.

16. The insignificant relationship between unemployment and trust in the EA-8 is largely driven by the German case in which an actual decrease of the unemployment rate of 2.8 percentage points (from 3-5/2008 to 11/2012) is associated with a significant decline in net trust in the ECB of 48 percentage points (see here also Figure A2). Once excluding the German case from the EA-8 country sample, the relationship between unemployment and trust regains significance (90 per cent level) and the coefficient regains strength (–7.1).

17. This is logical as in the case of Spain trust decreased significantly during the second year of the sovereign debt crisis while its sovereign bond yields remained relatively stable.

18. Whereas our econometric analysis actually still finds a weak (90 per cent level) relationship, Wälti’s (2012) findings points towards an insignificant relationship.

19. This is in contrast to the German case where an actual reduction of the unemployment rate is associated with a significant decline in trust in the ECB. A plausible hypothesis for the German case might be that the broadening of the ECB’s mandate to assure financial stability throughout the crisis has led to a decline in trust in the ECB.

20. As the decline of trust in the ECB might be interpreted as part of a general crisis of trust in European institutions, it becomes debatable whether other trust variables, such as citizens’ trust in the EC and the EP should be included in the model specification. We excluded these variables for two reasons. First, as trust in the EC and the EP is equally determined by inflation, growth and unemployment (Roth, Nowak-Lehmann D., and Otter 2013), it is econometrically incorrect to include these trust variables in the regression, because doing so would lead not only to double counting but also to endogeneity. Second, the Durbin–Watson statistic (being around 2) did not give us reason to worry about omitted variables.

21. It should be mentioned, however, that in 1-2/2009 net trust temporarily reached a value of −1 per cent. In this instance, however, net trust recovered to a value of +14 five months later in 6-7/2009.

22. In contrast to the support for the Euro, the support for the European Union actually declined more strongly(Braun and Tausendpfund 2014).