Revisiting Public Support for the Euro, 1999–2017: Accounting for the Crisis and the Recovery

Abstract

This contribution explores the evolution and determinants of public support for the euro since its creation in 1999 until the end of 2017, thereby covering the pre-crisis experience of the euro, the crisis years and the recent recovery. Using uniquely large macro and micro databases and applying up-to-date econometric techniques, the authors revisit the growing literature on public support for the euro. First, we find that a majority of citizens support the euro in nearly all 19 euro area member states. Second, we offer fresh evidence that economic factors are important determinants of change in the level of support for the euro: Crisis reduces support while periods of recovery from unemployment bode well for public support. This result holds for both macroeconomic and microeconomic factors. Turning to a broad set of socioeconomic variables, we find clear differences in support due to education and perceptions of economic status.

Originally published in: Felix Roth, Edgar Baake, Lars Jonung and Felicitas Nowak-Lehmann D. Revisiting Public Support for the Euro, 1999–2017: Accounting for the crisis and the recovery. Journal of Common Market Studies, Vol. 57, No. 6, 2019, pp. 1261–1273.

The authors thank three anonymous reviewers as well as Thomas Straubhaar and Silke Sturm for their valuable comments.

1 Introduction

This contribution explores the evolution and determinants of public support for the euro, using the largest up-to-date database on public opinion of the euro since its inception, available from March–April 1999 (EB 51) to November 2017 (EB88). It falls within the tradition of studies of the determinants of public support for the euro that have sprung up in recent decades (as a prominent example, see Banducci et al., 2009, Deroose et al., 2007, Hobolt & Leblond, 2014, and Hobolt & Wratil, 2015). This debate is about whether and under which circumstances the euro has been supported by citizens, in particular on the macroeconomic and microeconomic impact on public support. In line with the previous literature (see, for example, Banducci et al., 2009), we model public support for the euro at the macro- and micro-level, emphasizing the impact of economic factors. In contrast with much of the previous literature (see Hobolt & Leblond, 2014), we apply the latest econometric techniques to control for endogeneity.

Based on these specifications, we find that the euro has enjoyed support by a majority in nearly all 19 individual member states of the euro area (EA) from March–April 1999 to November 2017. Moreover, our econometric results at the macro- and micro-level find that unemployment is significantly and negatively related to public support for the euro. This result implies that the economic recovery in the EA starting in November 2013, which brought about a fall in unemployment, has increased public support.

The paper is structured as follows. Section 2 discusses the role of public support for the euro. Section 3 describes public support for the euro in the EA member states. The fourth section provides insights into the model specification, research design and data. Section 5 provides econometric results. The sixth section discusses the empirical findings in light of previous findings. The contribution ends with a short summary of our conclusions. Additional supporting information in the form of tables and figures can be found in the Appendices.

2 Public Support for the Euro

This section considers the role of public support for the Economic and Monetary Union (EMU) and the euro, as treated within various strands of the literature. First, evidence from the history of monetary unions suggests that a monetary union like EMU benefits from public support for the common currency. As long as the common currency enjoys public support, the monetary union will be able to adjust and adapt to changing circumstances (Bordo & Jonung, 2003, pp. 58, 63).

Second, the literature on the political economy in the optimum currency area approach suggests that a sustainable monetary union should feature a shared sense of common destiny (Baldwin & Wyplosz, 2019, p. 358). Such a shared sense of destiny between the partners of a monetary union is crucial to allow them to find collective solutions to common problems in times of economic strain. Public support for EMU and the euro is a prerequisite for such a sense of shared destiny. It is a vital ingredient for reconciling powerful national interests among EA governments, which have been one of the sources of the EA crisis (Frieden & Walter, 2017, p. 386).

Third, contributions within political science stress that public support for the euro is crucial for any move towards more supranational governance (Banducci et al., 2003, p. 686). Public support is necessary for European citizens to be willing to transfer power from national to European institutions (Kaltenthaler & Anderson, 2001, p. 14). This body of literature concludes that public support for EMU is crucial for its political legitimacy (Deroose et al., 2007) and hence its sustainability (Verdun, 2016, p. 306). In short, all strands of the literature note that public legitimacy matters. Therefore, widespread public support for the euro stands out as an important prerequisite for its long-term sustainability.

3 Descriptive Statistics

Figure 2.1 shows public support for the euro by the 19 member states that joined the EA between 1999 and 2017 (namely Austria, Belgium, Cyprus, Estonia, Finland, France, Germany, Greece, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, the Netherlands, Portugal, Slovenia, Slovakia and Spain – the EA-19).

Fig. 2.1

Net support for the euro in the EA-19, 1999–2017

Notes: The y-axis displays net support in percentages. As the figure depicts net support, all values above 0 indicate that a majority of the respondents support the euro. Net support measures are constructed as the number of ‘For’ responses minus ‘Against’ responses, according to the equation: Net support = (For − Against)/(For + Against + Don’t know). The dashed lines distinguish the actual physical introduction of the euro in January 2002, the start of the financial crisis in September 2008 and the start of economic recovery at the end of 2013. Average (avg.) EA-19 is population-weighted.

Figure 2.1 distinguishes between two stages in the history of the euro. The first stage covers the time from its inception until the start of the financial crisis (1999–2008). The second stage covers the time since the start of the financial crisis (October–November 2008 to November 2017). The latter is subdivided into a period of crisis (October–November 2008 to May 2013) and a period of recovery (November 2013 to November 2017).¹

Figure 2.1 shows that first, on average, a large majority of EA-19 citizens supported the euro over the 19-year period since its implementation (>30% net support). While net support declined in times of crisis by 9% points to a mean level of 31%, it has more than compensated for this drop during the recovery, with an increase of 22% points to a mean level of 53% (see Table 2.A1 in Appendix 2).

Second, since the establishment of the EA in 1999, aside from short periods in Finland and Greece in pre-crisis times and in Cyprus in crisis times, a majority of citizens in each of the 19 member states of the euro area has supported the euro. This includes continuous majority support in the largest EA economies such as Germany (with a minimum net support of 3% in November to December 2000) and Italy (with a minimum net support of 16% in November 2016) since the introduction of the euro in 1999.

Third, during the economic recovery (since November 2013), public net support for the euro has strongly increased within the EA’s periphery, in Spain and Portugal by 52% and 46% points, respectively, as well as in the EA’s core, namely Germany, by 28% points. In the majority of cases (nine of 15), the increase in public support for the euro throughout the recovery has more than compensated for the losses that accrued throughout the crisis (see Table 2.A1 in Appendix 2).²

4 Empirical Approach

4.1 Model Specification

To analyse the channels that influence public support for the euro, we adopt a model specification used by Roth et al. (2016, pp. 950–952). We estimate support for the euro as a function of unemployment, inflation, growth in real GDP per capita and the macroeconomic control variables considered important in explaining the within variation of support. Our baseline model (2.1) reads:

$$ {\mathrm{Support}}_{it}={\alpha}_i+{\beta}_1{\mathrm{Unemployment}}_{it}+{\chi}_1{\mathrm{Inflation}}_{it}+{\delta}_1{\mathrm{Growth}}_{it}+{\phi}_1{Z}_{it}+{w}_{it}, $$

(2.1)

where Support_it is the net support for the euro for country i during period t. Unemployment_it, Inflation_it, Growth_it, and Z_it are, respectively, unemployment, inflation, growth of GDP per capita and control variables deemed of potential importance, which can be lumped together in Z.³α_i represents a country-specific constant term (fixed effect), and w_it is the error term.

4.2 Research Design

Eq. (2.1) is estimated with an EA-19 country sample for the time period 1999–2017, with a total number of 38 time observations. With t = 38 and n = 19 and thus with a ratio of t/n = 2, Eq. (2.1) is estimated via a panel time-series estimation. Panel data analysis is superior to cross-section analysis as it exploits both variations over time and across cross-sections. In particular, it allows us to control for time-invariant cross-section (country) characteristics by modelling cross-section-specific intercepts. It also allows us to control for endogeneity by internal instrument techniques that require lagging the variables and to control for omitted variable bias by tackling the autocorrelation of the disturbances. In our analysis, we also apply a matching procedure between the macroeconomic variables and the Eurobarometer data (following Wälti, 2012, p. 597).

Second, to corroborate the findings between unemployment, inflation, economic growth, and support for the euro from the macro analysis, support is examined from a microeconomic point of view using 474,712 individual observations. In this case, the dependent variable is dichotomous, that is, 1 in case of support and 0 in case of no support. In this step, emphasis is put on perceptions about unemployment, inflation, and the overall economy as well as on exploring the socioeconomic characteristics of the interviewees: their gender, age, legal status, education, and employment status.

4.3 Operationalization and Data Used

Measures for public support for the euro are based upon the biannual Standard Eurobarometer (EB) surveys⁴ (European Commission, 2017) from March to April 1999 (EB51) to November 2017 (EB88), which asked respondents: ‘What is your opinion on each of the following statements? Please tell me for each statement, whether you are for it or against it. A European economic and monetary union with one single currency, the euro’. Respondents can then choose between ‘For’, ‘Against’ or ‘Don’t know’. Net support measures are constructed as described in the note to Fig. 2.1.

Data on inflation (the change in the harmonized index of consumer prices), seasonally adjusted unemployment rates, as well as seasonally and calendar adjusted data on GDP per capita (European Commission, 2013) are taken from Eurostat. A summary of the data utilized can be found in Table 2.A2 in Appendix 2.

Individual observations for support for the euro, which we obtained from the GESIS Leibniz Institute for Social Sciences, have been merged for the period 1999–2017 and include observations from EB51 (March–April 1999) to EB87 (May 2017). The merged variables include perceptions about unemployment, inflation and the overall economy and socioeconomic variables including gender, age, legal status, education and employment status. A summary of the data utilized can be found in Tables 2.A3 and 2.A4.

5 Econometric Results

5.1 Macro Analysis

We estimate Eq. (2.1) by means of dynamic ordinary least squares (DOLS), a method that permits full control for the endogeneity of the regressors (Stock & Watson, 1993; Wooldridge, 2009). To correct for autocorrelation,⁵ we apply a feasible general least squares (FGLS) procedure.⁶ Both applications lead to Eq. (2.2), representing our fixed effect dynamic feasible general least squares (FE-DFGLS) approach (the detailed steps leading from Eq. (2.1) to Eq. (2.2) are explained in Appendix 3):

$$ {\displaystyle \begin{array}{l}{\mathrm{Support}}_{it}^{\ast}={\alpha}_i+{\beta}_1{\mathrm{Unemployment}}_{it}^{\ast}+{\chi}_1{\mathrm{Inflation}}_{it}^{\ast}+{\delta}_1{\mathrm{Growth}}_{it}^{\ast}+{\phi}_1{Z}_{it}^{\ast}\\ {}+\sum \limits_{p=-1}^{p=+1}{\beta}_{2p}\Delta {\mathrm{Unemployment}}_{it-p}^{\ast}+\sum \limits_{p=-1}^{p=+1}{\chi}_{2p}\Delta {\mathrm{Inflation}}_{it-p}^{\ast}+\sum \limits_{p=-1}^{p=+1}{\delta}_{2p}\Delta {\mathrm{Growth}}_{it-p}^{\ast}\\ {}+\sum \limits_{p=-1}^{p=+1}{\phi}_{2p}\Delta {Z}_{it-p}^{\ast}+{u}_{it}\end{array}} $$

(2.2)

with α_i being the country fixed effect and Δ indicating that the variables are in first differences. On applying DFGLS, unemployment, inflation, and growth become exogenous and the coefficients β₁, χ₁, δ₁ and ϕ₁ follow a t-distribution. This property permits us to derive statistical inferences on the causal impact of unemployment, inflation, and growth. The asterisk (*) indicates that the variables have been transformed and that the error term u_it fulfills the requirements of the classical linear regression model. In addition, DFGLS estimations are very robust against the omission of other potentially relevant variables and therefore permit unbiased and consistent estimates of all right-hand side variables.

Table 2.1 shows the econometric results for Eq. (2.2) within our EA-19 country sample. When analysing the full period from March–April 1999 to November 2017 with 530 observations, we detect a highly significant negative impact of unemployment and inflation on the net support for the euro (−1.3 and −4.9, respectively). While the negative relationship between unemployment and public support for the euro is driven by the crisis-recovery period (October–November 2008 to November 2017), the negative relationship between inflation and public support for the euro is driven by both periods.⁷ More importantly, however, a sensitivity analysis of the crisis-recovery period reveals that whereas the negative relationship between unemployment and public support for the euro in the crisis-recovery period (−1.8) is strongly driven by the recovery period (−3.0), the relationship between inflation and public support becomes insignificant in times of economic recovery (see regressions 7–8 and 15–18 in Table 2.A8 in Appendix 2).⁸

Table 2.1 Unemployment, inflation, GDP per capita growth and support: fixed effect dynamic feasible general least squares estimations (aggregated level), EA-19, 1999–2017

5.2 Micro Analysis

At the micro level, we examine support for the euro by means of a probit model using individual data and account for respondents’ perceptions (PC) of unemployment, inflation and the overall economy as well as their socioeconomic characteristics. The equation for the probit model is expressed below:

$$ P\left({\mathrm{Support}}_{jit}=1\right)={\alpha}_i+\beta {\mathrm{Gender}}_{jit}+\gamma {\mathrm{Age}}_{jit}+\delta {\mathrm{Legal}\ \mathrm{Status}}_{jit}+\theta {\mathrm{Education}}_{jit}+\lambda {\mathrm{Employment}\ \mathrm{Status}}_{jit}+\phi \mathrm{Unemployment}\ {\mathrm{PC}}_{jit}+\chi \mathrm{Inflation}\ {\mathrm{PC}}_{jit}+\psi \mathrm{Economic}\ {\mathrm{PC}}_{jit}+{\eta}_t+{\varepsilon}_{jit}, $$

(2.3)

where P represents the probability with which the euro is supported. The dependent variable (Support_jit) represents the support of individual j in country i at time t and takes on 1 if the individual supports and 0 if the individual does not support the euro. Gender_jit, Age_jit, Legal Status_jit, Education_jit, and Employment Status_jit represent the gender, age, legal status, education, and employment status for individual j in country i at time t. Unemployment, Inflation, and Economic PC_jit represent the unemployment, inflation, and economic perceptions for the national economic situation or personal economic situation for individual j in country i at time t.; α_i represents the country fixed effects; η_t represents the time-fixed effects; and ε_jit represents the error term.

Regressions 1–3 in Table 2.2 list our socioeconomic background variables for the full-time sample compared with the pre-crisis and crisis-recovery period.⁹ The econometric results indicate significant negative associations for female and unemployed respondents and positive associations for married and educated respondents (aged 16–19 and 20+ years, respectively, when finishing education). The largest effect can be detected with regard to education. The probability that highly educated (20+) respondents would support the euro is around 18% points higher than those with lower education. While the pre-crisis and crisis-recovery sample results remain by and large stable, we observe a halving of the negative association for women in the crisis-recovery period¹⁰ and a complete reversal of opinion among the oldest age group, aged 65+ (a shift from −3.8 in pre-crisis times to +3.3 in the crisis-recovery period).¹¹

Table 2.2 Probit analysis (individual level), marginal effects, EA-19, 1999–2017

Regressions 4–5 incorporate the unemployment, inflation, and economic perceptions at the country and personal level for the crisis-recovery period. The two perceptions indicators, unemployment and inflation, have the expected negative effect, and the economic perceptions indicator has the expected positive effect for the national (Regression 4) as well as the personal economy (Regression 5) in the crisis-recovery period. As the estimation has utilized marginal effects, the coefficients can be interpreted in the following manner: an individual who identified the current unemployment situation of the national or their personal economy to be very/rather bad in the crisis-recovery period was around 5.6% or 6.5% points, respectively, less likely to support the euro than an individual who identified the unemployment situation of the national/their personal economy to be rather/very good.

6 Previous Empirical Results

Using the largest up-to-date dataset since the inception of the euro, from 1999 to 2017, our analysis first demonstrates that a majority of EA citizens have supported the euro in nearly each of the individual EA-19 member states. Our results are in stark contrast with those of scholars who claim to have found minority support in Italy (Guiso et al., 2016, p. 292) and Germany (Stiglitz, 2016, p. 314). However, these claims are not based on Eurobarometer data – the sole authoritative dataset for thorough research on public support for the euro across countries and over time.

Moreover, our macroeconometric results support the previous research of Roth et al. (2016, p. 953), who found a negative relationship between unemployment and support for the euro, analysing data from 2008 until 2014.¹² Extending the data up to 2017, we continue to find a negative relationship between unemployment and support for the euro. It is worth noting that the negative relationship becomes stronger in times of economic recovery. In addition, the highly significant negative relationship between inflation and support for the euro is in line with previous findings that relied on a shorter time span (Roth et al., 2016, p. 954).¹³ Extending the data up to 2017, we find that the negative relationship loses significance in times of economic recovery.

Furthermore, the findings of our macroeconomic analysis are corroborated at the micro-level. We find unemployment and inflation perceptions to be negatively related and economic perceptions to be positively related to public support for the euro in our crisis-recovery period. The patterns for our socioeconomic variables of gender, education, and employment status in the pre-crisis period are similar to previous results (Banducci et al., 2009, p. 576). Our finding that a stable pattern emerges for education, employment, and legal status when comparing the pre-crisis period with the crisis-recovery period makes a novel contribution to this literature.¹⁴

Furthermore, the halving of the negative association for women during the crisis-recovery period and the complete reversal in opinion among the oldest age group (65+) from strongly negative before the crisis towards strongly positive towards the euro during the crisis-recovery period stand out as new patterns that deserve further research.

7 Conclusions

This contribution has analysed the support for the euro for an EA-19 country sample over the 19-year period from 1999 to 2017. We reach three main conclusions. First, the euro, with few exceptions, has enjoyed majority support within each individual EA-19 member state since its introduction in 1999 until 2017. Second, our econometric results at the macro-level suggest that there is a negative and significant relationship between unemployment and public support for the euro, which is more pronounced during the recovery. The results also indicate a significant and negative relationship between inflation and public support for the euro, although this relationship was insignificant in times of recovery. Third, the findings of our micro-econometric analysis corroborate our macro-level findings. We discover a negative relationship between unemployment and inflation perceptions and public support for the euro. In addition, our results indicate that the patterns for our socioeconomic variables, including education, legal, and employment status, are stable. The largest effect is related to education; the probability for highly educated citizens (who were 20+ when finishing school) to support the euro is significantly higher than for those with lower education.

Overall, our results demonstrate that both macroeconomic and microeconomic developments are important drivers of public support for the euro. This finding generally supports previous studies on the matter.

Appendices

Appendix 1: Net Support for the Euro in the Non-EA-19, 1999–2017

Fig. 2.A1

Net support for the euro in the non-EA-19, 1999–2017

Data sources: Standard EB51-EB88.

Appendix 2: Descriptive Statistics and Test Results

Table 2.A1 Levels and changes in net support for the euro, EA-19, 2008, 2013 and 2017

Table 2.A2 Summary statistics for the macro analysis, 1999–2017

Table 2.A3 Summary statistics for the micro analysis, regressions 1–3, 1999–2017

Table 2.A4 Summary statistics for the micro analysis, regression 4–5, 2008–2017

Table 2.A5 Pesaran’s CADF panel unit root tests, EA-19 countries

Table 2.A6 Kao’s residual cointegration test, EA-19 countries

Table 2.A7 Unemployment, inflation, GDP per capita growth, change in the euro/US dollar exchange rate and support: FE-DFGLS estimations (aggregated level), EA-19, 1999–2017

Table 2.A8 Sensitivity analysis between unemployment, inflation and net support for the euro: FE-DFGLS estimations (aggregated level), 2008–2017

Table 2.A9 Correlation coefficients between unemployment, inflation and net support for the euro in the EA-19 countries, 2008–2013 and 2013–2017

Table 2.A10 Probit analysis (individual level), marginal effects, EA-19, C: 2008–2013, R: 2013–2017

Table 2.A11 Probit analysis (individual analysis), predicted probabilities, EA-19, 1999–2017

Fig. 2.A2

Unemployment and net support for the euro, EA19, 1999–2017

Data sources: Standard EB51-EB88.

Appendix 3: Transforming Eq. (2.1) into Eq. (2.2)

In the baseline model (2.1), net support for the euro is estimated as a function of unemployment, inflation, growth of GDP per capita and control variables deemed to be of potential importance:

$$ {\mathrm{Support}}_{it}={\alpha}_i+{\beta}_1{\mathrm{Unemployment}}_{it}+{\chi}_1{\mathrm{Inflation}}_{it}+{\delta}_1{\mathrm{Growth}}_{it}+{\phi}_1{Z}_{it}+{w}_{it} $$

(2.1)

where Support_it is the net support for the euro for country i during period t; Unemployment_it, Inflation_it, Growth_it, and Z_it are respectively unemployment, inflation and growth of GDP per capita and control variables deemed to be of potential importance for country i during period t. α_i depicts a country-specific constant term and w_it is the error term. As we utilize a Feasible Generalized Least Square (FGLS) estimation approach, time dummies are not included in our baseline estimation, as they are mutually exclusive with FGLS.

1.1 The Issue of Endogeneity

When running regressions such as in Eq. (2.1), one must be aware of the possibility that the right-hand side variables (unemployment, inflation and growth) might be endogenous (affected by a common event) or stand in a bi-directional relationship with support (a low level of support might lead to a self-fulfilling prophecy, speeding up and worsening an existing downturn). Therefore, we estimate the model by means of dynamic ordinary least squares (DOLS),¹⁵ a method that controls for endogeneity of the regressors (Stock & Watson, 1993; Wooldridge, 2009).¹⁶

It can be shown that by decomposing the error term and 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 baseline regression, which does not control for endogeneity and reflects a situation in which all adjustments have been made, has already been depicted in Eq. (2.1) above. Within Eq. (2.1) w_it is the iid-N error term, with the properties of the classical linear regression model. Controlling for endogeneity requires the decomposition of the error term w_it into the endogenous changes of the right-hand side variables, which are correlated with w_it(the changes in the variables) and the exogenous part of the error term υ_it; with:

$$ {\displaystyle \begin{array}{l}{\mathrm{w}}_{it}=\sum \limits_{p=-1}^{p=+1}{\beta}_{2p}\Delta {\mathrm{Unemployment}}_{it-p}+\sum \limits_{p=-1}^{p=+1}{\chi}_{2p}\Delta {\mathrm{Inflation}}_{it-p}\\ {}+\sum \limits_{p=-1}^{p=+1}{\delta}_{2p}\Delta {\mathrm{Growth}}_{it-p}+\sum \limits_{p=-1}^{p=+1}{\phi}_{2p}\Delta {Z}_{it-p}+{\upsilon}_{it}\end{array}} $$

(2.1a)

Inserting Eq. (2.1a) into Eq. (2.1) leads to the following Eq. (2.1b) in which all explanatory variables from the baseline model can be considered exogenous:

$$ {\displaystyle \begin{array}{l}{\mathrm{Support}}_{it}={\alpha}_i+{\beta}_1{\mathrm{Unemployment}}_{it}+{\chi}_1{\mathrm{Inflation}}_{it}+{\delta}_1{\mathrm{Growth}}_{it}+{\phi}_1{Z}_{it}+\\ {}\sum \limits_{p=-1}^{p=+1}{\beta}_{2p}\Delta {\mathrm{Unemployment}}_{it-p}+\sum \limits_{p=-1}^{p=+1}{\chi}_{2p}\Delta {\mathrm{Inflation}}_{it-p}+\sum \limits_{p=-1}^{p=+1}{\delta}_{2p}\Delta {\mathrm{Growth}}_{it-p}+\\ {}\sum \limits_{p=-1}^{p=+1}{\phi}_{2p}\Delta {Z}_{it-p}+{\upsilon}_{it}\end{array}} $$

(2.1b)

with α_i representing country fixed effects and Δ indicating that the variables are in first differences; the error term υ_it, Unemployment, Inflation and Growth become exogenous, and the coefficients β₁, χ₁, δ₁ and ϕ₁ follow a t-distribution. In addition, υ_it must fulfil the requirements of the classical linear regression model. Fulfilment of these properties allows us to draw statistical inferences concerning the impact of unemployment, inflation, and growth on support for the euro at the national and European level.

1.2 Omitted Variables and Autocorrelation

Having found that net support for the euro and the economic variables (unemployment, inflation, and growth) are non-stationary and cointegrated, we can be confident that omitted variables (which are lumped together in the error term) do not systematically influence our long-run relationship between support and macroeconomic variables. Omitted variables could include macroeconomic variables of potential importance, such as the change in the euro/US dollar exchange rate and the interest rate (Banducci et al., 2003, 2009; and Hobolt & Leblond, 2014), or socio-political factors such as positive attitudes towards EU membership (Banducci et al., 2009; Hobolt & Leblond, 2014), consumer confidence (Hobolt & Leblond, 2014), as well as social indicators, such as measures of income inequality and poverty rates, all of which have most likely deteriorated within the periphery countries of the EA-12.

Even though the error term is stationary [I(0)], which is a characteristic of cointegration, autocorrelation of the error terms might still be a problem that must be fixed. We do so by applying the two-step FGLS procedure. In a first step, we collect the ${\hat{\upsilon}}_{it}$ s from Eq. (2.1b), which has been estimated by means of DOLS. Thereafter, we estimate ρ₁, the first-order autocorrelation¹⁷ coefficient, via OLS based on Eq. (2.1c).

$$ {\hat{\upsilon}}_{it}={\rho}_1{\hat{\upsilon}}_{it-1}+{u}_{it}. $$

(2.1c)

Since the coefficient ρ₁ is usually unknown (as in our case), it has been estimated (giving us ${\hat{\rho}}_1$) by means of the Cochrane-Orcutt method (see Pindyck & Rubinfeld, 1991), which is an FGLS procedure. In a second step we transform all variables of Eq. (2.1b), which can be described by the following formulas (2.1d):

$$ {\displaystyle \begin{array}{l}{\mathrm{Support}}_{it}^{\ast}={\mathrm{Support}}_{it}-{\hat{\rho}}_1{\mathrm{Support}}_{it-1},\\ {}{\mathrm{Unemployment}}_{it}^{\ast}={\mathrm{Unemployment}}_{it}-{\hat{\rho}}_1{\mathrm{Unemployment}}_{it-1},\\ {}{\mathrm{Inflation}}_{it}^{\ast}={\mathrm{Inflation}}_{it}-{\hat{\rho}}_1{\mathrm{Inflation}}_{it-1},\\ {}{\mathrm{Growth}}_{it}^{\ast}={\mathrm{Growth}}_{it}-{\hat{\rho}}_1{\mathrm{Growth}}_{it-1},\\ {}{Z}_{it}^{\ast}={Z}_{it}-{\hat{\rho}}_1{Z}_{it-1}\end{array}} $$

(2.1d)

where the differences of the explanatory variables are transformed in exactly the same way as the variables in levels.

Correcting for autocorrelation in the error term via FGLS leads to Eq. (2.2):

$$ {\displaystyle \begin{array}{l}{\mathrm{Support}}_{it}^{\ast}={\alpha}_i+{\beta}_1{\mathrm{Unemployment}}_{it}^{\ast}+{\chi}_1{\mathrm{Inflation}}_{it}^{\ast}+{\delta}_1{\mathrm{Growth}}_{it}^{\ast}+{\phi}_1{Z}_{it}^{\ast}+\\ {}\sum \limits_{p=-1}^{p=+1}{\beta}_{2p}\Delta {\mathrm{Unemployment}}_{it-p}^{\ast}+\sum \limits_{p=-1}^{p=+1}{\chi}_{2p}\Delta {\mathrm{Inflation}}_{it-p}^{\ast}+\sum \limits_{p=-1}^{p=+1}{\delta}_{2p}\Delta {\mathrm{Growth}}_{it-p}^{\ast}+\\ {}\sum \limits_{p=-1}^{p=+1}{\phi}_{2p}\Delta {Z}_{it-p}^{\ast}+{u}_{it}\end{array}} $$

(2.2)

with α_i being the country fixed effect and Δ indicating that the variables are in first differences; * indicating that the variables have been transformed (purged from autoregressive processes) and that the new error term u_it(${u}_{it}={\upsilon}_{it}-{\hat{\rho}}_1{\upsilon}_{it-1}$) fulfils the requirements of the classical linear regression model (it is free from autocorrelation). Eq. (2.2), which is an improved version of Eq. (2.1b), represents the fixed effects dynamic feasible generalized least squares (FE-DFGLS) approach.