Welfare regimes: A brief literature overview

The most widespread classification of socio-economic models is that proposed by Reference Esping-AndersenEsping-Andersen (1990), according to which welfare models can be divided into three groups, namely: the Liberal, Continental and Scandinavian models. Although methodologically still very relevant, this taxonomy was based on evidence before the 1990s. Reference Hay and WincottHay and Wincott (2012) propose a dated version, extending the Esping-Andersen classification to five models by adding the Mediterranean group and the CEEC groupings. The peculiarities of a sort of ‘Southern’ model had already emerged in Reference FerreraFerrera (1996). Moreover, welfare patterns have further diverged since 1990: the Scandinavian model has followed the so-called ‘compensation thesis’ by increasing welfare in order to cope with the challenges of globalisation; the Continental model has maintained or slightly increased the level of welfare spending; and three other groups, namely, the Liberal, Mediterranean and CEEC, have converged in reducing public social expenditure during the past two decades, clearly following the so-called ‘efficiency thesis’ in confronting globalisation.

More broadly, countries can also be classified according to their type of economic system, characterised by particular institutional forms and macroeconomic factors (domestic competition, role of the state and international openness among others). Following this approach, both Reference AmorosoAmoroso (2003) and Reference JessopJessop (2002) identify four types of economic systems: the Anglo-Saxon model (or competitive capitalism); the Corporative model (or corporative capitalism), the Dirigiste model and the Social-Democratic model. To these, Reference ChoiChoi (2006) and other scholars (Reference Qian and RodrikQian, 2003; Reference YeagerYeager, 1999) added the current model of Socialist Markets, represented in particular by China and Vietnam. A similar story has been narrated by Reference AmableAmable (2004) who proposes five different ideal types of capitalism, taking into consideration several institutional features (product market competition, wage–labour nexus, financial sector, social protection and education). More specifically, Amable combines the Dirigiste and the Corporative models (shaping a Continental European model) and adds two new models (namely, Asian and South European). Thus, the Reference AmableAmable (2004) classification is as follows: (1) the Market-Based economy (USA and UK are the closest to this cluster), (2) Continental European capitalism (lead by Germany and France), (3) the Social-Democratic economies (represented by Scandinavian economies), (4) South European capitalism and (5) Asian capitalism. In parallel to the classification of welfare models, a theory of varieties of capitalism has been proposed, in particular by Reference Soskice and HallSoskice and Hall (2001), which considers welfare state and capitalism as complements, not as adversaries (see Reference RuberyRubery (2009) for a gender-based critique of the latter).

Inequality and welfare models: An empirical investigation

While broadly sympathetic to the argument of Reference Soskice and HallSoskice and Hall (2001), according to which the welfare state and capitalism should not be regarded as incompatible, we share the classification of Reference Hay and WincottHay and Wincott (2012) and add some elements to the picture by taking into consideration the evolution of different economic features in the past two decades among advanced economies. In detail, we combine traditional traits such as welfare spending (Reference BonoliBonoli, 1997) along with some features related to the emerging challenges of globalisation (trade openness and financial flows), as well as inequality and labour market flexibility. Our investigation refers to 34 OECD countries.^{Footnote 2} By merging some elements of Reference Hay and WincottHay and Wincott (2012) and Reference AmableAmable (2004), we cluster 30 of these countries in six groups – namely, Anglo-Saxon, Continental, Mediterranean, Scandinavian, CEEC and Asian, as in Table 1 – while Chile, Israel, Mexico and Turkey have not been grouped owing to their specific features.

Table 1. Country classification by welfare model (34 OECD countries).

Table 1 reports our classification of countries based on different welfare regimes. Country designations based on ISO 2-digit code (ISO, 2013).

In addition, we emphasise that a broad process of financialisation of the economy emerged in parallel to the more known trade-globalisation phenomena (Reference Engelen, Konings and FernandezEngelen et al., 2010; Reference EpsteinEpstein, 2005; Reference KrippnerKrippner, 2005). Capital mobility increased substantially and foreign direct investment (FDI) pursued mainly cheaper labour costs and higher returns. Moreover, at the international level, a sort of tax competition has taken place among countries in the past two to three decades to attract more capital, and this has helped shift power relations between labour and capital, advantaging capital, combined with a strong decline in trade union participation and a progressive weakening in labour market institutions. This has led to negative consequences for income distribution and has increased inequality (Reference Borjas and RameyBorjas and Ramey, 1995; Reference Chusseau and DumontChusseau and Dumont, 2012; Reference GordonGordon, 2012). Wage shares declined dramatically in particular during 1980–1990, and stagnated later, with a negative impact on aggregate demand, as argued by some post-Keynesian authors (Reference LavoieLavoie, 2014; Reference StockhammerStockhammer, 2015).^{Footnote 3} Reference Cynamon and FazzariCynamon and Fazzari (2013) found that inequality affects demand growth and creates a drag on the economy because high-income groups spend a smaller share of income;^{Footnote 4} while Reference Goda and LysandrouGoda and Lysandrou (2014) argued that economic inequality is boosted by credit consumption, and in turn this affects negatively stable economic growth. In this respect, Figure 1 reports the dynamics of average labour income shares (i.e. wage share adjusted by self-employment) by country groups.

Figure 1. Labour income shares by welfare model (1970/2013).

Source: Organisation for Economic Co-operation and Development (OECD), 2017.

Shows dynamics of labour income share ratios (adjusted wage shares) clustered by welfare model. Based on simple average value of labour income share for each group, calculated annually. Because of lack of historical data, the CEEC group starts from 1993.

Empirical evidence shows Asian countries as having the highest labour income shares, despite huge decreases in the past decades. Furthermore, wage shares remain, on average, higher in Scandinavian and in Continental European countries than in Anglo-Saxon, Mediterranean and CEEC countries. The Anglo-Saxon and the Mediterranean economies have suffered the most from the restructuring process (particularly financial globalisation)^{Footnote 5} that has progressively intensified since the 1980s. In this respect, globalisation has posed several challenges to national economies and governments. One of the most important factors has been the pressure on labour relations and its impact on income inequality, both within and between countries, as well as its consequences for welfare state sustainability (Reference Hay and WincottHay and Wincott, 2012). The debate is very lively and has produced two main interpretations of the problem.

The first interpretation states that globalisation depresses the size of welfare states because social provision constitutes a cost for firms: since welfare state expansion leads to higher income taxes, social costs and contributions, it tends to reduce prospective profits and increase companies’ costs. Firms will hence be pushed to transfer capital abroad unless government retrenches welfare spending and reduces taxes; then, in order to maintain higher levels of investment and employment, globalisation processes require the welfare state to be confined, with adverse consequences for income inequality. This ‘efficiency thesis’, developed within the so-called ‘neoliberal’ paradigm, argues that globalisation has forced states to retrench social spending in order to achieve a market-friendly environment, to increasingly attract international capital and to foster external competitiveness (Reference Allan and ScruggsAllan and Scruggs, 2004; Reference BlackmonBlackmon, 2006; Reference Castells, McGrew and HeldCastells, 2004). This view sees welfare states as a cost for firms: owing to capital mobility, companies will shift to the lowest cost location for production, putting pressure on governments to lower their welfare provision.

Against this argument, a second approach emerged: the ‘compensation thesis’ maintains that since globalisation increases income inequality, welfare states need to be expanded to mitigate vulnerability created by globalisation. In other words, increasing trade openness and capital mobility actually pressure governments to expand welfare support in order to compensate those who are damaged by the globalisation process (Reference Brady, Beckfield and Seeleib-KaiserBrady et al., 2005; Reference RodrikRodrik, 1998; Reference SwankSwank, 2002). Globalisation can produce net gains at the national level but where, within nations, there are winners and losers, then losers should be compensated by (partial) redistribution from the winners. Extending the ‘compensation’ argument, welfare expansion would allow countries to further pursue globalisation. An extensive interpretation would then see welfare expansion not as a result but as a condition of globalisation: briefly, in order to continue (or to start) with the process of globalisation, policy-makers must expand social safety nets.

Nonetheless, with respect to income inequality, through outsourcing practices and FDI outflows, globalisation has improved the bargaining position of capital relative to labour in higher income countries (Reference FeenstraFeenstra, 1998). Declining trade union power, particularly within export-oriented industries, may well account for a portion of increasing wage inequality in the United States and in other countries (Reference Borjas and RameyBorjas and Ramey, 1995; Reference Lavoie and StockhammerLavoie and Stockhammer, 2013). Moreover, the actual (or even the potential) decision by firms to relocate capital and production across countries has distributional effects since it can worsen the position of low-skilled workers in industrial countries by a combination of (1) growing globalisation and (2) availability of new technologies. With regard to the first pressure, as well as exacerbating competition among workers, global capital mobility may increase the bargaining power of capital in pressuring the state for tax reductions and welfare retrenchment: governments willingly embark on tax competition with each other in order to keep investment and production at home. The second effect, according to the so-called skill-biased technical change argument,^{Footnote 6} is a direct and negative impact on unskilled workers’ employment and their earnings – and consequently on real output in wage-led economies – in a context of lowering welfare support and social institutions.

As a consequence of these processes, intensified globalisation (particularly during the 1990s and the 2000s) was accompanied by increased income inequality, not only in emerging economies but also in advanced countries.^{Footnote 7} Figure 2 reports the Gini coefficient across OECD countries, indicating greater levels of inequality in Mediterranean and Anglo-Saxon countries than in Scandinavian countries (despite an overall upward trend).

Figure 2. Inequality by welfare model (1990–2013).

Source: OECD, 2017.

Displays the dynamic of Gini coefficients clustered by welfare model, based on simple annual average value of Gini among countries in each cluster. Uses current Organisation for Economic Co-operation and Development (OECD) definition: disposable income, post taxes and transfers where higher values mean higher concentrations of income.

Most recent literature describing the new socio-economic model built around the world in the past two decades sees it as a consequence of the new macroeconomic consensus linked to the process of globalisation (Reference Arestis, Charles and FontanaArestis et al., 2013; Reference GalbraithGalbraith, 2012; Reference PalleyPalley, 2012; Reference StockhammerStockhammer, 2015). This ‘Washington consensus’ has called for the implementation of institutional forms that better suit the globalisation process, such as the financialisation of the economy and the introduction of labour flexibility (Reference TridicoTridico, 2012). Over the past 20 years, to varying degrees in different advanced economies, labour market institutions have increased the freedom of entrepreneurs to hire and fire workers and to cut wages (Reference Storm and NaastepadStorm and Naastepad, 2012). However, flexibility has taken different forms. Some countries, such as Germany, Austria, Belgium and France, still have quite rigid labour markets; some, such as Denmark, Sweden, Finland and the Netherlands, have introduced higher levels of both flexibility and security; others, such as the USA, the UK and Ireland, have intensified job insecurity or maintained a very flexible labour market. Finally, Mediterranean countries and most of the former communist economies in Europe have built a hybrid combination of liberal and corporative elements with an increasing level of labour flexibility. It is useful to assess labour flexibility using Employment Protection Index (Employment Protection Legislation (EPL)), an OECD indicator measuring the level of worker protection in the labour market. This metric, which ranges from 0 (very low protection) to 6 (very high protection), is an indicator of the extent to which national legislation controls regular employment, temporary employment and collective dismissal – in other words, it reflects the degree to which employers are free to fire and hire workers at will (OECD, 2004). In general, labour flexibility has increased consistently in recent times as policy-makers have facilitated firms’ rapid adjustment of their productive systems to new processes in response to globalisation, regardless of negative effects on wages and labour productivity.

Inequality and labour flexibility have largely (although not everywhere) gone hand in hand during the past two decades, particularly in advanced economies whose governments have accepted the doctrine that globalisation will boost income growth. It is therefore interesting to notice an inverse relationship between inequality levels and the EPL index, as captured by Figure 3: the higher the labour flexibility, the higher the inequality. More specifically, Continental and Scandinavian European countries exhibit a higher EPL (i.e. lower labour market flexibility) and lower inequality, while Anglo-Saxon and Mediterranean countries generally show higher inequality and a lower EPL (Reference TridicoTridico, 2013).

Figure 3. Inequality and employment protection.

Source: OECD, 2017.

Correlation between the index of Employment Protection Legislation (EPL) and the Gini coefficient in 2013. Correlation is −0.403.

Based on Organisation for Economic Co-operation and Development (OECD) data on employment protection (2013) (regular contracts, version 1, 1985–2013) and inequality (Gini coefficient 2013: disposable income, post taxes and transfers). Closest year data used for countries where 2013 Gini coefficient not available. Excludes countries not clustered in the six identified models.

As labour flexibility allows for the reduction of labour costs and thus wage saving at the expense of wage earners (i.e. people with high propensity to consume), income inequality can be expected to increase and, consequently, aggregate demand is likely to be restricted owing to decreasing consumption (Reference KaleckiKalecki, 1954).^{Footnote 8} Nevertheless, we argue that what determines whether inequality increases or decreases under the condition of globalisation seems to be the socio-economic model that each country has been willing to build. More specifically, what is most relevant is the set of policies that each country is actually able to implement in order to cope with the challenges of globalisation, in terms of both income distribution and competitiveness (Reference RodrikRodrik, 1999). Of particular importance have been the institutions and conflict management policies that countries have adopted during the past two decades – social protection against unemployment and lower wages, social expenditure against poverty, public programmes on health and disease, national policy for housing and so forth. A proxy for these patterns can be offered by the relationships in our set of OECD countries – which can be considered the most advanced market economies – between inequality and welfare expenditure. As Figure 4 shows, countries that spend more on welfare appear to have, on average, a lower level of inequality.

Figure 4. Inequality and public social expenditure.

Source: OECD, 2017.

Correlation between public social spending as share of GDP and Gini coefficient in 2013 Correlation is −0.206.

Nine areas of social policy spending included, as per Organisation for Economic Co-operation and Development (OECD): old age, survivors, incapacity-related benefits, health, family, active labour market, unemployment, housing and others. Data on inequality refer to the Gini coefficient for 2013 or closest year, as per Figure 3. Excludes outliers, as per Figure 3.

However, supporters of the ‘efficiency thesis’ claim that expanding welfare provision may hamper economic growth, considering it an obstacle to efficiency and competitiveness. Conversely, ‘compensation thesis’ advocates see growing public social support as contributing to growth and facilitating globalisation. Within this debate, our contribution provides relevant empirical evidence. We start with a simple correlation between the growth rate in public social spending^{Footnote 9} (in real terms) and the growth rate of per capita real GDP (constant prices, constant purchasing power parities (PPPs)).

On the basis of Figure 5, the positive relationship between per capita income growth and public social spending growth offers some challenge to the ‘efficiency thesis’: welfare expansion cannot be regarded as inconsistent with real growth. However, this evidence is not enough in itself to provide empirical validation for the ‘compensation thesis’. In this regard, evidence concerning the relation between welfare expenditure (expansion or retrenchment) and growth in the context of globalisation is often found to be inconsistent and mixed. On one hand, Reference BlackmonBlackmon (2006) argues that globalisation has forced welfare state retrenchment in order for markets to achieve more competitiveness, and other economists have claimed that increasing welfare spending is not conducive to better economic performance (Reference Alesina and PerottiAlesina and Perotti, 1994) and that reducing the size of government will lead to higher incomes and improved competitiveness (Reference MitchellMitchell, 2005). On the other hand, other scholars argue that social spending has contributed to, rather than inhibited, economic growth (Reference Furceri and ZdzienickaFurceri and Zdzienicka, 2012; Reference LindertLindert, 2004; Reference RodrikRodrik, 1998). We investigate these divergent claims through a multivariate regression model which will allow us to analyse the relationships between welfare expansion and per capita real growth, taking into account both a set of control variables (able to play a relevant role in determining income growth) and causality issues.

Figure 5. Income growth and public social expenditure.

Source: OECD, 2017.

Correlation between public social spending growth rate and per capita real GDP growth rate in 2013. Correlation is +0.441.

Public social spending growth rate (2013) computed on the basis of public social expenditure at constant prices. Per capita real GDP growth rate computed on the basis of 2013 GDP per head (constant prices, constant purchasing price parities (PPPs)). For the sake of data comparability, here we considered the overall set of 34 Organisation for Economic Co-operation and Development (OECD) countries used in the econometric surveys.

The model

In order to test whether the ‘compensation thesis’ empirically holds, in this section we deploy an econometric analysis based on a panel dataset using annual data for 34 OECD economies from 1990 to 2013.^{Footnote 10} The dynamic growth equation to be estimated is the following

$\begin{array}{c}{g}_{it}=c_i+{\beta }_y.y_{it-1}+\sum _{s=0}^1{\beta }_s.{\stackrel{.}{PSS}}_{it-s}+\sum _{k=0}^1{\beta }_k.{\stackrel{.}{INV}}_{it-k}+{\beta }_o.OPE{N}_{it}+\\ +{\beta }_q.INE{Q}_{it}+{\beta }_f.\Delta FD{I}_{it}+{\beta }_e.E{D}_{it}+{\beta }_c.UL{C}_{it}+{\delta }_t+{ϵ}_{it}\end{array}$

where

• • g_it is per capita real income growth rate (USD, constant PPP);

• • y_it is the log of lagged per capita real income, used in order to control for the expected reduction in real growth as per capita income rises (Reference Baldacci, Clements and GuptaBaldacci et al., 2004) and in order to better cope with feasible autocorrelation issues;^{Footnote 11}

• • $P\stackrel{.}{S}{S}_{it}$ is the growth rate of public social spending in real terms; as we expect that countries have higher income growth if they invest more in welfare, we test both current and lagged growth of public social spending in order to control for reasonable anti-cyclical effects (i.e. automatic stabilisers); according to the ‘compensation thesis’, we expect β_s coefficients to be positive;

• • $I\stackrel{.}{N}{V}_{it}$ is the growth rate of investment, measured as gross fixed capital formation (both public and private) in real terms; we expect β_k coefficients to be positive since investment growth should reflect increasing capital stock, which in turn would positively affect real growth;

• • $OPE{N}_{it}$ is the degree of trade openness, measured as export plus import (goods and services) divided by GDP; since this indicator is a proxy for globalisation, the coefficient β_o indicates whether high-income growth is associated with a greater degree of openness in terms of trade;

• • $INE{Q}_{it}$ is the level of inequality, measured by the Gini coefficient (current definition, disposable income post taxes and transfers); as we argue that an inequitable income distribution can hinder economic growth, we expect the coefficient β_q to be negative (a high Gini coefficient means greater concentration of income, hence higher inequality);

• • The set of three control variables, considered for an ‘augmented’ model specification, as well as for robustness, is composed of: (1) ΔFDI_it, a metric of financial globalisation, where FDI represents the share of FDI inflows on GDP; (2) ED_it is a metric of education (human capital), which indicates the share of the population aged 25–35 with tertiary qualifications; (3) ULC_it is the unit labour cost in the total economy, that is, wages adjusted for labour productivity (a widely used indicator of external competitiveness);

• • $c_i$ country-specific fixed effects identified by the type of welfare model, while δ_t denotes year dummies; both random and fixed effects are used with respect to the panel variable (country or welfare model, alternatively), while the use of year dummies allows us to consider time fixed effects (to deal with common time-related shocks and thus to remove correlations in errors across countries);

• • ${\epsilon }_{it}$ denotes the error term.

In Table 2, we report the results, which refer to four different model specifications estimated by means of feasible generalised least squares (FGLS)-based regressions. First, we estimated the baseline model (column I) without considering the set of control variables by means of a generalised least squares (GLS) regression and without controlling for country-specific fixed effects (thus assuming c_i = c for all i). Then, in the second column, we extended the analysis (column II) to our vector of controls, namely, FDI-inflows dynamics, tertiary education and competitiveness), using the same estimation methodology as for the baseline model. Finally, using FGLS regressions, we estimated the baseline model by considering panel heterogeneity, thus alternatively using country-specific fixed effects (column III) and country fixed effects identified by the type of welfare model (column IV).

Table 2. Regression models on panel data (34 countries, 1990–2013), different specifications.

Data sources: Organisation for Economic Co-operation and Development (OECD) 2017. In detail, data on GDP, per capita GDP_PPP, Export and Import drawn from dataset 1. GDP; public social spending drawn from OECD Social Expenditure Database (SOCX); Tertiary Education drawn from Educational attainment and labour-force status; Gross fixed capital formation and ULC drawn from Economic Outlook No 101, June 2017; FDI drawn from FDI flows by industry; Gini coefficients drawn from dataset Income Distribution and Poverty; Real effective exchange rate (REER) drawn from data set Factbook Country Statistical Profiles, 2015/2016 edition.

GDP: gross domestic product; FDI: foreign direct investment; PPP: purchasing price parity; FE: fixed effect; ULC: unit labour cost.

Robust standard errors clustered by countries in parentheses.

*** p < 0.01;

** p < 0.05;

* p < 0.1.

As can be seen by comparing the different columns in Table 2, the model is remarkably robust to different specifications in terms of both additional predictors and fixed effects (both country and welfare models). With respect to the ‘baseline’ version of the model (Table 1, Column I), both current and lagged growth rates of public social spending are positively related to per capita income growth, and particularly the current growth rate is highly statistically significant. Moreover, both investment growth rates (current and lagged) and trade openness are positively related to per capita income growth, while income inequality is negatively related (β_q < 0 at 1% significance level). Regrettably, the inclusion of control variables reduces substantially the number of observations. Notwithstanding this, our main findings are robust to the inclusion of additional variables. In this respect, some elements can be added to the picture: specifically, the dynamics of FDI inflows is proved not to be statistically functional to per capita real income growth, while increasing unit labour cost (ULC) – that is, decreasing external cost-competitiveness – is negatively related to growth (β_c < 0 at 1% significance level). Moreover, the level of tertiary education is positively related to per capita income growth (β_e > 0 at 5% significance level).

Since the extended specification might suffer from the poor availability of data, we opted for testing the baseline model’s robustness by considering panel heterogeneity. To this end, we estimated a two-way fixed effects model by means of FGLS: when we allow for country or welfare model fixed effects, respectively, regression coefficients are almost completely in line with the baseline estimation.

Of course, a key challenge associated with our econometrics is that public social spending is likely to be partly endogenous: we have to deal with the potential issue of reverse causality. This criticism is to some extent mitigated by the inclusion of per capita GDP level within the model, whose estimated coefficient (β_y) is negative in all model specifications (i.e. the expected reduction in real growth rates when per capita GDP is higher actually occurs). However, we additionally performed a bi-variate causality test (Reference GrangerGranger, 1969) based on a panel-vector autoregression (VAR) methodology (Reference Abrigo and LoveAbrigo and Love, 2016) between public social spending growth rate and per capita GDP growth rate.^{Footnote 12} As shown in Table 3, bidirectional Granger-causality arises between public social spending and real income growth rates. Nevertheless, a more accurate assessment emerges from Table 4, where Granger-causality is tested in a tri-variate specification: when investment growth rate is included (as in the empirical model), growing public social spending Granger-causes income growth, but not vice versa.^{Footnote 13}

Table 3. Causality between public social spending and GDP: Bi-variate Granger-causality test.

GDP: gross domestic product; VAR: vector autoregression; GMM: generalised method of moments; PSS: public social spending.

Robust standard errors in parentheses; final GMM Criterion Q(b) = 1.36e–34; No. of obs. = 677; No. of panels = 34; Ave. no. of T = 19.912.

H_o: the ‘row variable’ does not Granger-cause the ‘column variable’.

*** p < 0.01;

** p < 0.05;

* p < 0.1.

Table 4. Causality between public social spending and GDP: Tri-variate Granger-causality test.

GDP: gross domestic product; VAR: vector autoregression; GMM: generalised method of moments; PSS: public social spending.

Robust standard errors in parentheses; final GMM Criterion Q(b) = 2.80e–33; No. of obs. = 677; No. of panels = 34; Ave. no. of T = 19.912.

H_o: the ‘row variable’ does not Granger-cause the ‘column variable’.

Δ% PSS = public social spending growth (constant prices).

Δ% GDP = per capita GDP growth (GDP per head, constant prices, constant purchasing price parities (PPPs)).

Δ% INV = gross fixed capital formation growth (constant prices).

*** p < 0.01;

** p < 0.05;

* p < 0.1.

Furthermore, in order to detect whether the ‘efficiency thesis’ may hold with regard to competitiveness – that is, to verify whether increasing welfare provision actually reduces external competitiveness due to increasing costs – we performed a bi-variate Granger test between the growth rate of public social spending and the dynamics of ULC. As shown in Table 5, increasing public social spending has no statistically significant impact on the forward dynamics of ULC (in other words, Granger-causality does not emerge in this direction), while the ULC growth rate has significant effects on the forward dynamics of public social spending (decreasing external competitiveness is followed by increasing welfare provision, consistently with automatic stabilisation policies). The same conclusions can be traced with respect to the real effective exchange rate (REER),^{Footnote 14} an additional metric for price-competitiveness: according to Table 6, the public social spending growth rate does not Granger-cause REER dynamics. Ultimately, the typical narrative of the ‘efficiency thesis’, according to which increasing welfare spending would hinder economic growth by decreasing external competitiveness (with respect to both ULC and REER), is not supported by our results.

Table 5. Causality between public social spending and external competitiveness: Bi-variate Granger-causality test.

ULC: unit labour cost; VAR: vector autoregression; GMM: generalised method of moments; PSS: public social spending.

Robust standard errors in parentheses; final GMM Criterion Q(b) = 6.87e–35; No. of obs. = 562; No. of panels = 29; Ave. no. of T = 19.379.

H_o: the ‘row variable’ does not Granger-cause the ‘column variable’.

*** p < 0.01;

** p < 0.05;

* p < 0.1.

Table 6. Causality between public social spending and external competitiveness: Bi-variate Granger-causality test.

VAR: vector autoregression; GMM: generalised method of moments; PSS: public social spending; REER: real effective exchange rate.

Robust standard errors in parentheses; final GMM Criterion Q(b) = 4.46e–35; No. of obs. = 660; No. of panels = 34; Ave. no. of T = 19.412.

H_o: the ‘row variable’ does not Granger-cause the ‘column variable’.

Δ% PSS = public social spending growth (constant prices).

Δ% ULC = unit labour cost growth (2010 = 100).

Δ% REER = real effective exchange rate dynamics (2010 = 100).

*** p < 0.01;

** p < 0.05;

* p < 0.1.

Finally, in addition to the usual regression statistics, some diagnostic issues were also explored. Since we have to deal with a relatively small (imperfect) collinearity between some predictors, and some endogeneity concerns can be also be raised, we carried out a variance inflation factor (VIF) test, which aims to exclude systematic multicollinearity among explanatory variables. When VIF is high, there is high multicollinearity, and consequently, regression coefficients would be unstable: despite the fact that there are no formal thresholds for determining the presence of multicollinearity, values of VIF that exceed 10 are often regarded as indicating multicollinearity. As can be seen in Table 7, the highest VIF value in our econometrics is 1.75 (while the VIF mean is 1.41); hence, collinearity is unlikely to be biasing the estimated coefficients.^{Footnote 15}

Table 7. Multicollinearity variance inflation factor (VIF) test on extended GLS model.

GDP: gross domestic product; FDI: foreign direct investment; GLS: generalised least squares; ULC: unit labour cost.