Current Accounts and Coordination of Wage Bargaining

This study provides novel evidence on the impact of labor market institutions on current account dynamics. Our results suggest that a high degree of coordination of wage bargaining has a positive effect on the current account balance over the long run. This result is not driven entirely by wage moderation induced by centralized wage setting, however. A high degree of coordination of wage bargaining is associated with a slower current account adjustment toward its long-run equilibrium. This result seems theoretically plausible; the aggregate shocks in the exporting sector are largely driven by idiosyncratic shocks and the presence of idiosyncratic shocks increases the importance of labor market flexibility. This analysis of the impact of labor market institutions on current account balances complements the existing empirical current account literature focused on macroeconomic and financial measures.


Introduction
Academic and policy debates have devoted considerable attention to global current account imbalances and intra-euro area imbalances over the past ten years. The many studies on medium-term determinants of current account balances (e.g., Chinn and Prasad, 2003;Chinn and Ito, 2007;Gruber and Kamin, 2007;Ca' Zorzi et al., 2012) concentrate on macroeconomic factors such as GDP per capita, government budget balance, or institutional variables that measure differences in financial development and political stability. A closely related strand of the literature considers determinants of the rate of current account adjustment (e.g., Chinn and Wei, 2013;Ghosh et al., 2013). These studies are largely limited to examining the role of exchange rate regimes. Given that the type and degree of wage-bargaining coordination affects macroeconomic performance (for reviews, see Flanagan, 1999;Aidt and Tzannatos, 2010), the lack of empirical studies of the role of labor market institutions in current account dynamics is striking. For example, EMU member countries, unable to use exchange rate adjustment as a policy instrument after adopting the euro, must rely increasingly on labor market institutions in economic adjustment.
We argue that factors related to labor market institutions such as the degree of wage-bargaining coordination can significantly affect both the current account balance and speed of adjustment of the current account toward its long-run equilibrium.
Collective wage bargaining may take place at firm level, by industry, or on a national scale.
Although it is not easy to classify countries by this criterion, the substantial differences across countries in the degree of coordination of wage bargaining is quite apparent. Northern European countries tend to use centralized bargaining, while English-speaking countries, except Ireland, have prefer more fragmented approaches. (Cahuc et al., 2014, pp. 408-410.) If the wage-bargaining structure affects cost-competitiveness of an economy (see Calmfors and Driffill, 1988;Carlin and Soskice, 2006, p. 114), it is plausible to assume that the wage-bargaining structure also affects the level of the current account balance.
For open economies, Traxler and Brandl (2012) propose that the macro effects of bargaining on price competitiveness depend on how well the bargain takes into account inter-sectoral productivity differentials. Industry-level bargaining is superior where the exposed sector dominates wage coordination. They find that the wage-bargaining structure has a statistically significant effect on the growth rate of nominal labor cost and current account balance. With respect to current account surplus, they specifically argue that exposed-sector pattern bargaining outperforms other wage bargaining structures. Du and Liu (2015) assert that labor market flexibility affects the real Mika Nieminen, Kari Heimonen and Timo Tohmo Current accounts and coordination of wage bargaining 6 exchange rate. Both papers suggest that labor market institutions such as the degree of coordination of wage bargaining may affect the current account balance. The distinction of Traxler and Brandl (2012) of productivity differences between exposed and sheltered sectors, however, does not address the real-world issue of large productivity differentials between individual firms within the tradable sector (e.g., Syversen, 2011;Bernard et al., 2012).
An economy can absorb shocks by means of current account adjustment. If aggregate-level shocks are the sole drivers of productivity growth, adjustment should be faster in centralized-bargaining regimes than in countries that use industry-level bargaining (see Aidt and Tzannatos, 2008, pp. 263-264;Carlin and Soskice, 2006, pp. 748-749). Canals et al. (2007), Del Rosal (2013) and Freund and Pierola (2015) all find, however, that a large fraction of aggregate volatility in exports or net exports results from firm-specific shocks.
Typically, highly centralized bargaining systems do a poorer job at accounting for conditions of individual firms. The groundbreaking work of Ju et al. (2014) provides the first microfoundations for understanding cross-country heterogeneity in the current account adjustment rate. They show that an economy's response to a shock involves a combination of intertemporal trade (current account adjustment) and intra-temporal trade (goods trade). Their theoretical model and empirical results indicate that labor market rigidities make the adjustment of current account toward its longrun equilibrium level slower. Cuñat and Melitz (2012) build a theoretical model that highlights the importance of labor market flexibility as volatility (variance of firm-specific shocks in a sector) increases. Labor market flexibility is a source of comparative advantage in high-volatility sectors. They also provide empirical evidence consistent with their model. The exports of countries with relatively flexible labor markets are biased toward high-volatility sectors.
If a government or central organizations have a strong preference for wage moderation, centralized wage bargaining might have a positive effect on the current account balance by enhancing cost-competitiveness. Correspondingly, if a large fraction of aggregate volatility in exports results from firm-specific shocks, centralized wage bargaining might have a negative effect on the speed of current account adjustment. Firm-specific competitiveness is most easily sustained by firmlevel wage bargaining. The remainder of the paper proceeds as follows. Section 2 provides descriptions of the data and empirical methodology. Section 3 is a presentation of our results on determinants of the current account balance and rate of current account reversion. Section 4 includes conclusions and discussion. Since the degree of coordination of wage bargaining is not measured on the metric scale, we mostly use the mode of the sample period for the wage-bargaining coordination variable and model different degrees of coordination of wage bargaining with a set of binary dummy variables. If we include the mode of coordination of wage bargaining in lieu of a set of binary dummy variables, we impose a monotonic relationship between the degree of coordination of wage bargaining and the dependent variable. 3 Models (4)-(5) in the panel regressions of current account balance and models (14)- (23) in the one-step procedure to obtain the rate of current account reversion use annual observations of the degree of coordination of wage bargaining rather than mode of coordination.
The set of control variables for the current account regressions is derived from the current account literature (e.g., Chinn and Prasad, 2003

Current account regressions
To answer our first research question: "Does the degree of coordination of wage bargaining have an effect on the current account balance in the long run?" we run both cross-sectional regressions with multi-year averages and panel data regression with annual observations. These are standard methodologies in the current account literature ( e.g., Chinn and Prasad, 2003).
For models (1)-(3), we estimate the following cross-sectional regression model by the OLS estimator: where the dependent variable is the long-run current account balance (ratio to GDP), α is an intercept, Coordji is a binary variable for coordination of wage bargaining in country i in regime j, xi is a column vector including all control variables for country i, and εi is a residual.
For models (4)-(5), we estimate the following panel data regression model: where the dependent variable is current account balance (ratio to GDP), α is an intercept, μt are time fixed effects, Coordjit measures the degree of coordination of wage bargaining, xit is a column vector including all explanatory variables for country i in period t, and εit is a residual.

Current account adjustment regressions
To answer our second research question: "Does the degree of coordination of wage bargaining have an effect on the speed of current account adjustment?" we apply two approaches. Our two-step procedure is adopted from Ju et al. (2014), while the one-step procedure follows, among others, Chinn and Wei (2013). Ju et al. (2014) note that, despite higher efficiency than the two-step procedure, the one-step procedure includes possible bias due to potential heterogeneity in steady-state current accounts across countries.
In the two-step procedure, we measure the country-specific speed of current account adjustment by estimating the following equation using the OLS estimator for each country: where ∆CAit is the first difference of the current account balance (ratio to GDP) of country i in period t, β0,i and β1,i are country-specific coefficients, CAit-1 is the current account balance (ratio to GDP) of country i in period t-1, and εit is a residual. Values of β1,i close to minus one imply fast adjustment of the current account toward its long-run equilibrium, whereas values close to zero imply slow adjustment of the current account toward its long-run equilibrium. 6 Potential serial correlation in the residual is eliminated by including higher orders of the lags of the dependent variable.
In the second stage of the two-step procedure for models (10)-(13), we estimate the following cross-sectional regression model by the OLS estimator: where β1,i is the speed of adjustment of the current account toward its long-run equilibrium (i.e. β1,i in equation (3)) in country i, α is an intercept, Coordji is a binary variable for coordination of wage bargaining in country i in regime j, xi is a vector of control variables of country i, and εi is a residual.
Within the one-step procedure, we can measure the speed of current account adjustment by using two different approaches. The first approach with models (14)-(17) relies on estimating the following equation using the OLS estimator for each category of wage-bargaining coordination separately: where CAit is the current account balance (ratio to GDP) of country i in period t, ρ0 is an intercept, and εit is a residual. 7 In the second approach for models (18)-(19), we estimate the following equation using the OLS estimator: where CAit is the current account balance (ratio to GDP) of country i in period t, ρ0 is an intercept, Coordjit measures the degree of coordination of wage bargaining, and εit is a residual. 8 We augment the two-step procedure with the one-step procedure for three reasons. First, it allows us to check if we produce the same results with respect to degree of coordination of wage bargaining as the two-step procedure. Second, it enables us to calculate a measure for the half-life of current account balance deviations for different degrees of wage-bargaining coordination. Third, it provides us a simple way to account for asymmetric effects. This implies that the current account adjustment may depend on the sign of the current account balance.

Empirical results and discussion
In section 3.1, we empirically test whether the degree of coordination of wage bargaining is related to the current account balance. In section 3.2, we examine the association between the degree of coordination of wage bargaining and the speed of current account adjustment.  (1) and (2). Compared to our reference category for no coordination in wage bargaining (fragmented firm-level wage bargaining), a high degree of coordination in wage bargaining (centralized wage bargaining) has a positive effect on current account balances.

Coordination of wage bargaining and current account balance
With regard to control variables, our models produce standard results. 9 The result on the coordination of wage bargaining is statistically significant in both the cross-sectional and panel data regressions. In models (2) and (4), we make a distinction between the two subcategories of centralized wage bargaining. 10 In models (3) and (5), we include coordination of wage bargaining as such.
This identification assumes that the relationship between the current account and the degree of coordination of wage bargaining is monotonic. According to our estimations, the current account surplus (deficit) increases (decreases) monotonically with the degree of coordination of wage bargaining. The result is not driven by an outlier (see Figure A1 in the Appendix). Furthermore, it is robust to including a set of control variables that has become standard in the current account literature. If we split the sample according to the IMF country classification and compare the two subsamples, we find that the positive relationship between coordination of wage bargaining and current account balance is stronger for advanced economies than emerging market and developing economies (see Figures A2-A3 and Table A4 in the Appendix). 11 Table 2 Coordination of wage bargaining and current account balances 12  Notes: The dependent variable in models (1)-(3) is the long-run current account balance (ratio to GDP). Heteroscedasticity robust standard errors are in parentheses. The dependent variable in models (4)-(5) is the current account balance (ratio to GDP). Panel robust standard errors are in parentheses (clustering on the panel variable). *, ** and *** denote statistical significance at the 10%, 5% and 1% levels. Fragmented firm-level wage coordination is the reference category for the coordination of wage bargaining in models (1), (2) and (4).

BOFIT-Institute for Economies in Transition Bank of Finland
BOFIT Discussion Papers 20/ 2017 13 Our models in Table 2 predict that the current account surplus (deficit) of a country with a high degree of coordination of wage bargaining is 2.5% of GDP larger (smaller) than for a country with no coordination of wage bargaining. This is a rather substantial effect, and comparable with the effects of other institutional factors reported in previous studies (e.g., Gruber and Kamin, 2007, Figure 2). The apparent channel here is via wage moderation. Thus, in Table 3 we estimate the following equation using the OLS estimator: where ∆NULCi is change in nominal unit labor costs in country i, GDPi is GDP per capita in country i and εi is a residual. 13 Table 3 presents the results from estimating equation (7). The relation between the coordination of wage bargaining and change in nominal unit labor costs seems to be sensitive to whether or not Bulgaria and Romania are included (see also Figures A4-A5 in the Appendix) and whether or not we control GDP per capita. If we include the change in nominal unit labor costs as an additional control to models (1)-(5), the results for the relation between the degree of coordination of wage bargaining and current account balances do not change (see Table A6 in the Appendix). This implies that our result is not entirely driven by wage moderation. Notes: The dependent variable is change in nominal unit labor costs. Heteroscedasticity robust standard errors are in parentheses. *, ** and *** denote statistical significance at the 10%, 5% and 1% levels. In models (8)-(9), the two countries with the largest increase in nominal unit labor costs and the lowest GDP per capita (Bulgaria and Romania) were excluded.

3.2
Coordination of wage bargaining and speed of current account adjustment 3.2.1 Two-step procedure Table 4 presents the results from estimating equation (4). Compared to our reference category no coordination in wage bargaining (fragmented firm-level wage bargaining), a high degree of coordination in wage bargaining (centralized wage bargaining) decreases the speed of adjustment of the current account toward its long-run equilibrium. Due to the central role of the US dollar in the world economy, the US has enjoyed an exorbitant privilege that relaxes its external constraint Rey, 2007, 2014;Prasad, 2014). When we include a dummy variable for the US in model (11), the result becomes even stronger. 14 In model (12), we make a distinction between the two subcategories of centralized wage bargaining. In model (13), we include coordination of wage bargaining as such. This identification assumes that the relationship between the speed of current account adjustment and the degree of coordination of wage bargaining is monotonic.
According to the two-step procedure, the speed of adjustment of the current account toward its long-run equilibrium decreases monotonically with the degree of coordination of wage bargaining. The result is not driven by an outlier (see Figure A6 in the Appendix). 15 This finding seems theoretically plausible as the aggregate shocks in the exporting sector are largely driven by idiosyncratic shocks and the presence of idiosyncratic shocks increases the importance of labor market flexibility. It is also consistent with the theory on collective bargaining and wage dispersion and with microlevel evidence on wages as firm-level wage bargaining increases the responsiveness of wages to firm-specific profitability (e.g., Barth and Zweimüller 1995;Guertzgen 2009;Garloff and Guertzgen 2012). If we split the sample according to the IMF's country classification and compare the two subsamples, we find that the negative relationship between the co-14 The financial openness index and GDP per capita were excluded from the main tables because they were statistically insignificant in all models (see Tables A7, A10-A12 in the Appendix). This finding comports with previous studies (Ju et al. 2014, Tables 3-4;Chinn and Wei 2013, Tables 5-11;Ghosh et al. 2013, Table 3). Table A8 in the Appendix presents results when a dummy variable for the EMU-12 countries is included as a control variable. EMU membership decreases the speed of current account adjustment, but the result for the relationship between the speed of current account adjustment and degree of coordination of wage bargaining remains unchanged. 15 In an earlier version of this paper, we tested whether the length of collective wage agreements affects the speed of current account adjustment. It initially seemed that there would be a negative relationship between the two (longer collective wage agreements would make the current account adjustment slower), but it turned out that this result was driven by only one country -India. If India is excluded, the coefficient of the length of collective wage agreements is not statistically significant even at the 30% level.
ordination of wage bargaining and speed of current account adjustment is equally strong for advanced economies and emerging market and developing economies (see Figures A7-A8 and Table   A9 in the Appendix). 16 Notes: The dependent variable is a country-specific regression coefficient for an AR process with lags that characterizes the speed of adjustment of the current account toward its long-run equilibrium, i.e. β1,i in equation (3). Heteroscedasticity robust standard errors are in parentheses. *, ** and *** denote statistical significance at the 10%, 5% and 1% levels. Fragmented firm-level wage coordination is the reference category for the coordination of wage bargaining in models (10)-(12).  16 We tested a dummy variable for the advanced economies for the regression models presented in Table 4 and found it to be statistically insignificant. 17 See the country coverage and sample period in Table A3 in the Appendix. under firm-level wage bargaining. The difference between the extreme opposite categories is statistically significant at the 5% level (see Figure A9 in the Appendix). 18 In addition all regressions include a constant.

One-step procedure
Notes: The dependent variable is current account balance (ratio to GDP). CAt-1 is the lagged term of current account balance. Panel robust standard errors are in parentheses (clustering on the panel variable). *, ** and *** denote statistical significance at the 10%, 5% and 1% levels. Table 6 presents the results from estimating equation (6). Both approaches of the one-step procedure give similar results. In model (19), the relationship between the degree of coordination of wage bargaining and rate of current account reversion is monotonic and countries with a high degree of coordination of wage bargaining experience slower current account reversion. 19 Thus, the one-step procedure confirms the finding of the two-step procedure.
18 Table A10 in the Appendix presents results with financial openness index included as a control variable. 19 Table A11 in the Appendix presents results when the financial openness index is included as a control variable. We tested a dummy variable for the advanced economies for the regression models presented in Table 6 and found it to be statistically insignificant.
CAt Notes: The dependent variable is current account balance (ratio to GDP). Panel robust standard errors are in parentheses (clustering on the panel variable). *, ** and *** denote statistical significance at the 10%, 5% and 1% levels.
Fragmented firm-level wage coordination is the reference category for the coordination of wage bargaining in model (18).

Asymmetric effects
As the degree of coordination of wage bargaining affects the current account balance in the long run and the speed of adjustment of the current account toward its long-run equilibrium, it is important to consider whether current account reversion rates depend the running of a surplus or deficit. Following Chinn and Wei (2013), we test the asymmetric effects by estimating the following equation using the OLS estimator: where CAit is the current account balance (ratio to GDP) of country i in period t, ρ0 is an intercept, posCAit-1 is a dummy variable that equals one if CAit-1 is positive and εit is a residual. In this identification, the coefficient ρ1 represents the rate of reversion when the current account is negative, and the sum of two coefficients ρ1+ ρ2 is the rate of reversion when the current account is positive. Table 7 presents the results from estimating equation (8). In most categories, current account deficits are associated with higher speeds of current account adjustment. However, the difference is not statistically significant and the positive association between the degree of coordination of wage bargaining and the current account balance (section 3.1) does not explain why the speed of adjustment of the current account toward its long-run equilibrium decreases monotonically with the degree of coordination of wage bargaining. Irrespective of whether current account is positive or negative, centralized wage bargaining is associated with slower current account reversion than fragmented firm-level wage bargaining.  In addition all regressions include a constant.
Notes: The dependent variable is current account balance (ratio to GDP). CAt-1 is the lagged term of current account balance. posCAt-1 is a dummy variable which equals one, if CAit-1 is positive. Panel robust standard errors are in parentheses (clustering on the panel variable). *, ** and *** denote statistical significance at the 10%, 5% and 1% levels.

Wage bargaining, exchange rate stability and current account
As the literature on the rate of reversion of the current account concentrates on examining the role of exchange rate regimes, we analyze the interaction effect of wage-bargaining coordination and exchange rate stability on the speed of current account adjustment. We measure exchange rate stability with the continuous exchange rate stability index proposed by Aizenman, Chinn and Ito (2010) rather than usual dichotomous de facto exchange rate regime classifications. We estimate the following equation by the OLS estimator:

BOFIT-Institute for Economies in Transition
where β1,i is the speed of adjustment of the current account toward its long-run equilibrium (i.e. β1,i in equation (3)) in country i, α is an intercept, Coordji is a binary dummy variable for the coordination of wage bargaining of country i in regime j, ERSi measures exchange rate stability in country i and εi is a residual. 20 Notes: The dependent variable is a country-specific regression coefficient for an AR process with lags that characterizes the speed of adjustment of the current account toward its long-run equilibrium, i.e. β1,i in equation (3). Heteroscedasticity robust standard errors are in parentheses. *, ** and *** denote statistical significance at the 10%, 5% and 1% levels. Fragmented firm-level wage coordination is the reference category for the coordination of wage bargaining in model (24).
Current accounts and coordination of wage bargaining 20 Table 8 presents the results from estimating equation (9). Model (24) shows that, while exchange rate stability decreases the speed of current account adjustment, there is a strong negative interaction between the effects of coordination of wage bargaining and exchange rate stability on the speed of current account adjustment. In model (25), we see that imposing the assumption of monotonicity on the coordination of wage bargaining does not change the result. 21  of wage bargaining increases, the effect of exchange rate stability on current account adjustment diminishes.
A negative interaction term between coordination of wage bargaining and exchange rate stability suggests 1) that the level of wage bargaining should be adjusted for the prevailing exchange rate regime to obtain the desired speed of current account adjustment, and 2) that firm-level wage flexibility and economy-wide exchange rate flexibility are not substitutes for shock absorption.
We are unaware of any theoretical model that explicitly analyzes the interaction effect of these two adjustment channels on the speed of current account adjustment. Intuitively, it appears    Regression type Cross-sectional Cross-sectional Cross-sectional Notes: The dependent variable is the long-run current account balance (ratio to GDP). Heteroscedasticity robust standard errors are in parentheses. The sample consists of advanced economies in model (3ADV). The sample consists of emerging market and developing economies in models (3E&D1)-(3E&D2). Romania was excluded in model (3E&D2). *, ** and *** denote statistical significance at the 10%, 5% and 1% levels.

Table A5
Coordination of wage bargaining and current account balances  Notes: In models (1C)-(3C) the dependent variable is the long-run current account balance (ratio to GDP). Heteroscedasticity robust standard errors are in parentheses. The dependent variable in models (4C)-(5C) is current account balance (ratio to GDP). Panel robust standard errors are in parentheses (clustering on the panel variable). Advanced economy dummy equals one if the country is classified as advanced economy in the IMF's country classification, and zero otherwise. *, ** and *** denote statistical significance at the 10%, 5% and 1% levels.  Notes: In models (3A)-(3B) the dependent variable is the long-run current account balance (ratio to GDP). Heteroscedasticity robust standard errors are in parentheses. The dependent variable in models (5A)-(5B) is current account balance (ratio to GDP). Panel robust standard errors are in parentheses (clustering on the panel variable). In models (3B) and (5B), the two countries with the largest increase in nominal unit labor costs and the lowest GDP per capita (Bulgaria and Romania) were excluded. *, ** and *** denote statistical significance at the 10%, 5% and 1% levels. Notes: The dependent variable is a country-specific regression coefficient for an AR process with lags that characterizes the speed of adjustment of the current account toward its long-run equilibrium, i.e. β1,i in equation (3). Heteroscedasticity robust standard errors are in parentheses. *, ** and *** denote statistical significance at the 10%, 5% and 1% levels. Fragmented firm-level wage coordination is the reference category for the coordination of wage bargaining in models (10A)-(12A). Notes: The dependent variable is a country-specific regression coefficient for an AR process with lags that characterizes the speed of adjustment of the current account toward its long-run equilibrium, i.e. β1,i in equation (3). Heteroscedasticity robust standard errors are in parentheses. EMU dummy equals one, if a country adopted the euro by 2001 and zero otherwise. *, ** and *** denote statistical significance at the 10%, 5% and 1% levels. Fragmented firm-level wage coordination is the reference category for the coordination of wage bargaining in models (10B)-(12B). Notes: The dependent variable is a country-specific regression coefficient for an AR process with lags that characterizes the speed of adjustment of the current account toward its long-run equilibrium, i.e. β1,i in equation (3). Heteroscedasticity robust standard errors are in parentheses. In model (13ADV), the sample consists of advanced economies. In model (13E&D), the sample consists of emerging market and developing economies.*, ** and *** denote statistical significance at the 10%, 5% and 1% levels. All regressions include a constant.
Notes: The dependent variable is current account balance (ratio to GDP). CAt-1 is the lagged term of current account balance. Panel robust standard errors are in parentheses (clustering on the panel variable). *, ** and *** denote statistical significance at the 10%, 5% and 1% levels.  Notes: The dependent variable is a country-specific regression coefficient for an AR process with lags that characterizes the speed of adjustment of the current account toward its long-run equilibrium, i.e. β1,i in equation (3). Heteroscedasticity robust standard errors are in parentheses. *, ** and *** denote statistical significance at the 10%, 5% and 1% levels. Fragmented firm-level wage coordination is the reference category for the coordination of wage bargaining in model (24A). Notes: The dependent variable is a country-specific regression coefficient for an AR process with lags that characterizes the speed of adjustment of the current account toward its long-run equilibrium, i.e. β1,i in equation (3). Heteroscedasticity robust standard errors are in parentheses. *, ** and *** denote statistical significance at the 10%, 5% and 1% levels. Fragmented firm-level wage coordination is the reference category for the coordination of wage bargaining in model (24B). i.e. 1 in equation (3) 1 2 3 4 5 Coordination of wage bargaining (1=Firm-level,..., 5=Centralized)