To analyse the determinants of net trust in the national government/parliament and European Commission/Parliament over a longer time horizon, the fixed-effects Dynamic Ordinary Least Squares (DOLS) or a Feasible Generalised Least Squares (FGLS) estimation method is used (see Table 10.3). There are basically three econometric issues that deserve prior discussion. One is whether and how to control for omitted variables and whether inclusion of time fixed effects is an adequate approach; the second issue is about structural breaks or whether inflation, growth and unemployment influence trust in the national and European institutions in the same way under normal economic conditions and in times of crisis. The third issue concerns the endogeneity of the explanatory variables.
Table 10.3 Trust in the national government and parliament, EU-15 country sample, controlling for endogeneity (FE-DFGLS or FE-DOLS estimation) 6.1 The Issue of Dealing with Unexpected Events and Omitted Variables
In econometric modelling, we often have to deal with unquantifiable or unobservable events (or both). In the panel data literature, it has become very common to work with fixed time dummies in order to proxy events that are identical for all countries in the sample but which change over time. In contrast to traditional panel data studies, we do not favour the use of time dummies. We have reason to believe that countries (our cross-sections) are usually affected differently by the same ‘general’ event. With respect to the 27/15 EU economies under investigation, for example, the EU’s enlargement strongly affects the neighbouring countries in Central Europe but less so the countries farther away. Moreover, the state of the world economy affects especially those countries having commercial and investment banks with considerable international exposure and/or a strong dependency on exports, and tight financial markets do more harm to countries with a housing bubble, such as Spain, Ireland and United Kingdom. By plugging in time dummies, one would mimic the same exposure to an unspecified risk in all 27 EU countries under investigation. We thus find it more appealing to control for unknown omitted variables that are country-specific and that change over time (vit) through FE-FGLS. FGLS works with transformed variables (denoted by an asterisk *). It is realistic to assume that today’s disturbances are somehow related to past values of the disturbance term wit, that is, to variables that are omitted over the entire sample period.
The new estimation equation describes the long-run co-movement of the series when all adjustments have been made. Therefore, it does not contain lagged values of the explanatory variables. Still, estimation requires the series to be non-stationary and to be in a long-run equilibrium (cointegrated). See Tables 10.A3, 10.A4, 10.A5 and 10.A6 in the Appendix for the panel unit root tests and cointegration tests performed. The series turned out to be I(1) and cointegrated.
The Equation reads as follows:
$$ {\displaystyle \begin{array}{l}{\mathrm{Trust}}_{it}^{\ast }={\alpha}_i+{\beta}_1{\mathrm{Inflation}}_{it}^{\ast }+{\chi}_1{\mathrm{Growth}}_{it}^{\ast }+{\delta}_1{\mathrm{Unemployment}}_{it}^{\ast }+{\varepsilon}_1{\mathrm{DebtperGDP}}_{it}^{\ast }+\\ {}{\phi}_1{Z}_{it}^{\ast }+{\beta}_2\Delta {\mathrm{Inflation}}_{it}^{\ast }+{\chi}_2\Delta {\mathrm{Growth}}_{it}^{\ast }+{\delta}_2\Delta {\mathrm{Unemployment}}_{it}^{\ast }+{\varepsilon}_2\Delta {\mathrm{DebtperGDP}}_{it}^{\ast }+\\ {}{\phi}_2\Delta {Z}_{it}^{\ast }+{u}_{it}\end{array}} $$
(10.2)
with Δ indicating that the variables are in first differences; and with * indicating that the variables have been transformed (purged from autoregressive processes) and that the error term uit fulfills the requirements of the classical linear regression model (it is free from autocorrelation).
$$ {\displaystyle \begin{array}{l}\mathrm{Trust}/\mathrm{Nat}/{\mathrm{Europ}}_{it}^{\ast }=\mathrm{Trust}/\mathrm{Nat}/{\mathrm{Europ}}_{it}-{\rho}_1\mathrm{Trust}/\mathrm{Nat}/{\mathrm{Europ}}_{it-1}-{\rho}_2\mathrm{Trust}/\mathrm{Nat}/{\mathrm{Europ}}_{it-2},\\ {}{\mathrm{Inflation}}_{it}^{\ast }={\mathrm{Inflation}}_{it}-{\rho}_1{\mathrm{Inflation}}_{it-1}-{\rho}_2{\mathrm{Inflation}}_{it-2}\\ {}{\mathrm{Growth}}_{it}^{\ast }={\mathrm{Growth}}_{it}-{\rho}_1{\mathrm{Growth}}_{it}-{\rho}_2{\mathrm{Growth}}_{it}\\ {}{\mathrm{Unemployment}}_{it}^{\ast }={\mathrm{Unemployment}}_{it}-{\rho}_1{\mathrm{Unemployment}}_{it-1}-{\rho}_2{\mathrm{Unemployment}}_{it-2}\\ {}{\mathrm{DebtperGDP}}_{it}^{\ast }={\mathrm{DebtperGDP}}_{it}-{\rho}_1{\mathrm{DebtperGDP}}_{it-1}-{\rho}_2{\mathrm{DebtperGDP}}_{it-2}\\ {}{Z}_{it}^{\ast }={Z}_{it}-{\rho}_1{Z}_{it-1}-{\rho}_2{Z}_{it-2}\kern0.5em \mathrm{and}\kern0.5em {u}_{it}={w}_{it}-{\rho}_1{w}_{it-1}-{\rho}_2{w}_{it-2}={w}_{it}^{\ast}\end{array}} $$
(10.3)
The transformations of the variables in first differences are generated in exactly the same way. Note that the new error term uit is free of autocorrelation and that the omitted variable problem is reduced – if not eliminated – by transforming the variables. Since the coefficient ρ1 is usually unknown (as in our case), it has been estimated by means of, for example, the Cochrane–Orcutt method (an FGLS procedure).Footnote 14 In addition, we use country-specific fixed effects in our analysis.
6.2 The Issue of Structural Break
Given that we would expect a structural break caused by the economic crisis, a test for parameter stability is indicated. The Chow test showed a structural break between the pre-crisis period (spring 1999–autumn 2007) and the crisis period (spring 2008–autumn 2010). Although we also present results for the full sample period (spring 1999–autumn 2010) in Tables 10.A7 and 10.A8 of the Appendix, the emphasis should be on the separate regressions for the pre-crisis period (column (1)) and the crisis period (column (2)). It also becomes evident that a regression over the full sample period can produce misleading results. For example, the price level seems to be of importance in the spring 1999–autumn 2010 period, but it is never significant in the subperiods (pre-crisis and crisis) (columns (1) and (2)).
6.3 The Issue of Endogeneity
When running regressions, one must be aware of the possibility that the left-hand side variables and the right-hand side variables influence each other. More specifically, the right-hand side variables (inflation, growth, unemployment and debt) might be endogenous (affected by a common event) or stand in a bidirectional relationship with trust (a low level of trust might lead to a self-fulfilling prophecy and might thus speed up and worsen an existing downturn). Therefore, we estimated the model for the pre-crisis and the crisis periods by means of DOLS, a method that controls for the endogeneity of the regressors. DOLS is also known as the ‘leads and lags’ approach proposed by Stock and Watson (1993) and described by Wooldridge (2009). It can be shown that by inserting the leads and lags of the right-hand side variables in first differences, the explanatory variables become (super-) exogenous and the regression results thus become unbiased. The coefficient estimator of the explanatory variables in levels follows a t-distribution and hence enables us to draw statistical inferences. Due to a multicollinearity problem, we included only the first differences of the explanatory variables. As we also eliminate autocorrelation (whenever necessary), the DOLS estimation turns into a DFGLS estimation. One should note that the DOLS/DFGLS estimation technique requires the series to be integrated of e.g. order 1 (I(1)) and cointegrated, i.e. to stand in a long-run relationship.Footnote 15
6.4 Regression Results
All the tables contain results for the full sample (Tables 10.A7 and 10.A8 in the Appendix) and the pre-crisis and the crisis periods. Table 10.3 shows the results for the EU-15 country sample concerning citizens’ trust in the national government and parliament. Inflation has the expected, negative impact on trust in the national government and parliament in the pre-crisis period. As inflation does not play a role in the crisis period, the full sample result in which inflation matters (Table 10.A7) is driven by the pre-crisis period. Growth is an important determinant of trust only in the pre-crisis period,Footnote 16 whereas unemployment has a significant, negative impact on trust in all sample periods (pre-crisis, crisis and full sample period). Government debt influences trust only in national parliament in the pre-crisis and full sample periods. Overall, we can conclude that the increasing/decreasing rate of unemployment is very strongly associated with citizens’ trust in the national government and parliament.
Yet once incorporating an interaction term between debt and those EU-15 countries that have strongly supported their financial industry in the crisis period, the interaction term turns out to be negative and highly significant. Citizens’ loss of trust in the national parliament and European Commission is negatively associated within those countries in which the increase of government debt was due to aid for the financial industry.Footnote 17 To determine the magnitude of a country bailing out its financial industry, data from DG Competition were used (European Commission, 2009). The data present the amount of state aid given by each of the 27 EU member states to financial services (sector j) in year 2009. As the state aid data for the financial industry for 2010 have not been published yet, we have assumed that the same list of countries should be applicable for 2010. When incorporating an interaction term between debt and those EU-15 countries that have faced significant austerity measures in the crisis period, the interaction term turns out to be negative and highly significant for trust in the European Commission and the European Parliament. The data are provided by Theodoropoulou and Watt (2011).Footnote 18
Table 10.4 shows the results for the EU-15 country sample concerning citizens’ trust in the European Commission and Parliament. The two variables inflation and government debt (and, to a somewhat lesser extent, growth) are strongly associated with trust in the European Commission and European Parliament in the full and pre-crisis samples, but not in the crisis period. The positive association between growth and trust in the European Commission and European Parliament is driven by the strong association in the aftermath of the financial crisis, where a dramatic economic downfall was accompanied by a severe drop in citizens’ trust in the two European institutions. The increase in unemployment during and after the crisis has led to a decrease in trust in both the European Commission and the European Parliament.
Table 10.4 Trust in the European Commission and European Parliament, EU-15 country sample, controlling for endogeneity (FE-DFGLS or FE-DOLS estimation) Up to now, we have only considered an EU-15 country sample for the reason that with this sample, it is possible to extend the timeframe back to 1999. We now shift our analysis to an EU-27 sample. As the EU-25 has been in place since 2004, our timeframe when estimating our EU-27 country sample focuses on the period from 2004 to 2010.Footnote 19 Thus, Table 10.5 shows the results for the EU-27 country sample concerning citizens’ trust in the national governments and parliaments.
Table 10.5 Trust in the national government and parliament, EU-27 country sample, controlling for endogeneity (FE-DFGLS or FE-DOLS estimation) It is interesting to note in the EU-27 sample that inflation only plays a role in trust in the national government or the national parliament when we look at the pre-crisis period. Growth is not of importance for trust in the national government in the EU-27, whereas growth positively impacts on trust in the national parliament in the pre-crisis period. The negative impact of unemployment on trust shows up only in the pre-crisis period. The increase in debt leads to a decrease in trust in the national parliament and national government in the crisis period. This econometric result confirms the first empirical evidence given by Roth (2011).
Table 10.6 shows the results for the EU-27 country sample concerning citizens’ trust in the European Commission and European Parliament. Regression 1 shows the results when employing the pre-crisis sample of 2004–2010. If inflation increases during good times, European institutions lose trust. A decline in growth diminishes trust during an economic crisis. Government debt must be considered harmful for trust in European institutions in both the pre-crisis and crisis periods. Unemployment reduces trust in the European Commission at all times and trust in the European Parliament in the pre-crisis period.
Table 10.6 Trust in the European Commission and European Parliament, EU-27 country sample, controlling for endogeneity (FE-DFGLS or FE-DOLS estimation)