The central hypothesis which underlies this article is the concept of equilibrium in personal income distribution. The famous economist Vilfredo Pareto (Pareto, 1895) is already known to have believed that the social distribution of personal incomes moves towards a stable equilibrium over time. He based his statement on the observation that personal income dispersion fluctuates neither internationally nor inter-temporally. Much later, Hans Jürgen Ramser (Ramser, 1987) identified stationarity in the secondary distribution of personal incomes (i.e. the income net after government intervention with taxes and transfers) and not in the primary distribution of incomes (out of the market process). Recent studies support this empirical finding (Genc, Miller & Rupasingha, 2011).
The existing skewness of (personal) income distribution may be interpreted as a display of social preferences, thereby implying that preserving a specific degree of income inequality is intentional (Blümle, 1992, p. 224) and not arbitrary. Although distributional justice continues to be a fundamental goal of economic policies, it does not focus strictly on achieving a perfect equitable income distribution. Both short- and long-term scenarios accept the unwarranted existence of a certain degree of inequitable income distribution in society (Blümle, 1992, p. 225). In reality, such equilibrium would seldom be achieved to a full extent, although the policymakers have good reasons to push towards the ‘steady state’ and thereby help reach convergence.
When it comes to the question of breaking down this idea to a possible empirical analysis, the world seems to be more complex: Will countries be moving always towards a similar equilibrium level of inequality? This question is all but trivial. Works by Esping-Andersen (1990, 1994, 1998), by Esping-Andersen and Myles (2009) and by Hall and Soskice (2001) suggest something different and highlight the fact that among developed countries there exist long-run differences in the institutional structure and in so-called ‘ institutional complementarities in the macroeconomy’ (Hall & Gingerich, 2004), which can also be conceived as different equilibria in income distribution. These different equilibria may be attributed not only to the differences in the institutional set-up, but also to the observed variety of historical experiences in the respective countries. Taking into account the role of business in national economies and the fact that there is more than one path to economic success (‘liberal market economies’ vs. ‘coordinated market economies’) explain the further differences. However, different historical experiences and/or unalike institutional structures become much less important, once countries share a longer period of a common economic, social and political history, as is the case for the member countries in the EU. This insight holds even if countries realise to a different and necessarily often changing extent what Esping-Andersen calls the ‘three worlds of welfare capitalism’ (liberal, conservative, social-democratic systems; see Tiemann, 2006 for a critical empirical evaluation of this non-undisputed concept). Furthermore, it is understood that the institutional complementarity and implicit coordination between a ‘supra-national’ European Central Bank (ECB) on the one hand, and national institutions, such as national unions, pursuing a ‘national’ wage policy on the other hand, is a common challenge for members of the Eurozone.
The article is organised as follows: after a brief review of relevant literature, we present our equilibrium model in personal income distribution. An exhaustive empirical part—considering a period of extreme economic turbulence—follows, whose findings support the concept of steady state and convergence in personal income distribution. The article concludes with a discussion of the theoretical and empirical results of the present study, the limitations of the analysis and the possibilities for future research endeavours.
In principle, there exist two strands of the literature which are relevant to the subject under consideration in the present study. While one of the two perspectives focuses on the relationship between an
In a pre-crisis study (Van Kerm & Alperin, 2013), the authors reported that the arrangement of the countries of the world in descending order of annual income inequality for the period of 2003–2007 puts “Portugal and Baltic states (such as Estonia, as reported by the authors) at the top, and most Scandinavian countries (such as Finland and Denmark, as reported by the authors) at the bottom” (p. 937). This result was overly unspecific. The research papers published by Dolls, Fuest and Peichl (2011), De Beer (2012) and Kaitila (2013) concerned the issue of economic crisis and income distribution in Europe. Dolls et al. (2011) conducted two controlled experiments (simulations) of macro shocks to income and employment and observed that “both shocks lead to higher differences between the Gini coefficients based on equivalent disposable and market income” (ibid., p. 240). This effect applied to all the 19 European countries considered by the authors in their study. De Beer (2012), in our view, utilised too short a period (2008–2009) to conclude that “the economic crisis has not so far led to a general widening of income disparities and a rise in poverty” (ibid, p. 23).
The second group of contributions has a considerable tradition and follows the seminal paper of Ravallion (2003): as a long-time member of World Bank research groups, his variety of natural interest concentrated on developing countries’ fate. However, his 2003 paper also delivered important methodological aspects for measuring (conditional and unconditional) convergence in personal income distribution. The critical hypothesis that Ravallion tested is whether the trend in inequality depends on its initial level (ibid., p. 352). He finds inequality convergence, “with a tendency for within-country inequality to fall (rise) in countries with initially high (low) inequality” (ibid., p. 355).
A specific example of follow-up investigations is the contribution of Alfani and Ryckbosch (2015). In their long-run historical perspective (1500–1800), they compared changes in inequality between central and northern Italy on the one hand and the southern and northern Low Countries on the other hand. Similarly, Martinez-Carrion and Maria-Dolores (2017) explored inequality and regional convergence in Spain and Italy for the long period of 1850–2000. Lessmann and Seidel (2015) investigated regional inequality (based on Gini indexes and other measuring instruments) and convergence based on much more recent satellite night-time light data for a vast sample of developed and developing countries. Sell (2015, p. 15–20) also analysed both developed and developing countries and obtained as a result that “globalisation and possibly other forces linked to the revolution in communication and information technologies have contributed to an almost worldwide convergence in the distribution of personal incomes. More precisely, one can say that developing (developed) countries’ income distribution has become more equal (unequal)” (ibid, p. 16).
A smaller section of this body of literature deals directly with the convergence issue applied to overall and/or parts of Europe (for example, the EU and/or the Eurozone) and the recent past. For example, Paas and Schlitte (2007) analysed a cross-section of 861 EU-25 regions from 1995 to 2003 to detect between- and within-country disparities in income distribution—measured by the Theil index. Melchior (2008) studied regional inequality and convergence in Europe (1995–2005), that is, for a period which preceded exactly the financial and world economic crisis of 2008/2009. She found that for the EU-27 as a whole, there was a modest increase in within-country regional inequality, but convergence across countries (ibid, p. 31).
In 2013, Kaitila, in turn, reported a result that was considered close to the findings of the present study, although a slightly different approach (involving the calculation of the sigma convergence) was followed in that study. According to Kaitila’s findings, “For the EU–15 (a little less for the EU–27, the authors), we found that the national Gini coefficients have converged considerably during these (1995–2011, the authors) years” (ibid., p. 14). A most recent contribution to the EU inequality subject stems from Savoia (2019): his sample covers the years 1989–2013 (i.e. he necessarily misses part of the Eurozone crisis episode) and countries from the so-called ‘NUTS 2’ regions. He finds a clear tendency that supports the results of Kaitila: “inequality is converging, but to a higher level” (ibid., p. 29).
A minor methodological remark: It is surprising to notice that regressions between GDP per capita and Gini coefficients of disposable income are seldom run for the enormous projects concerning European inequalities, and the focus of these projects remains restricted to the role of redistributive policies (Medgyesi & Toth, 2009, p. 135; Paulus, Figari & Sutherland, 2009, p. 154). An obvious step towards convergence remained undetected between 2000 and 2005, a period when the EU-27 was without Cyprus, Malta, Slovakia and Luxembourg. As stated by a previous study, “the level of inequality at the beginning of the period does not seem to influence the direction and the magnitude of the change in inequality” (Medgyesi & Toth, 2009, p. 140). This finding, however, appears flawed. The discrepancy is attributed to the addition of countries to the EU, which turned EU-15 into EU-25 in 2004 and into EU-27 later.
Therefore, the question of how economic crisis, equilibrium and convergence in personal income distribution could be addressed scientifically in a comprehensive approach remains demonstrated so far. We describe this task in the following sections.
It is surprising to see that irrespective of the definition of income, the economy in question or the time period under consideration, the distribution of incomes is skewed positively (i.e. skewed towards the right-hand or steep on the left-hand).
This fact has significant consequences for the parameters of the density function, for which the maximum value, which is referred to as the modus (
The consequences of this are extensive. According to Blümle (2005), most economic agents would receive an above-modus income. Based on this observation, the agents would have the impression of being well-paid, and therefore their attitude towards a redistribution (the existing distribution) of incomes should be quite critical (benevolent). The density function depicted in Figure 1 may be approximated, rather accurately, using a log-normal distribution of incomes, which is represented as follows:
The expected or similar average wage rate is then obtained using the following expression (Beichelt & Montgomery, 2003, pp. 46–8):
Taking the full differential of the above-mentioned expression from left to right yields the following expression:
Furthermore, we consider the following expression:
Taking the full differential of the above-mentioned expression yields the following expression:
Finally, we obtain the following expression:
In Figure 2, an increase in the standard deviation, σ, has the effect derived in Propositions 1–3: the modus is shifted to the left and the arithmetic mean is shifted to the right while the position of the median remains unchanged.
It is now assumed that an increase in inequality or a higher concentration of incomes is perceived by an individual
Assuming the law of diminishing increases of damage, one obtains:
Therefore, the corresponding iso-damage curves are concave. It is essential to apply a kind of budget constraint to determine an optimal solution. Such a budget constraint may be found in the properties of the log-normal distribution. Its properties reveal that an increasing dispersion of incomes does not alter the median of the distribution (Proposition 3), reducing and shifting the modus to the left (Proposition 2). These findings also indicate a likely increase in households’ share with an income above the (new) modus whenever the concentration of incomes, as measured by the standard deviation σ, increases.
The following expression was used to determine the mathematical solution for the Cobb–Douglas utility function which is presented in Figure 3:
In the diagram depicted in Figure 3, one may identify and locate the equilibrium in the personal income distribution. The modus (
Point P signals towards a situation where a preferably low iso-damage curve is tangential to VV. In a sense, P represents an equilibrium in the income distribution. Note that point P stands for what Chiang (1984, p. 231–232) labels a ‘goal equilibrium’. The “equilibrium state is defined as the optimum position for a given economic unit (a household, a business firm, or even an entire economy) and in which the said economic unit will be deliberately striving for the attainment of that equilibrium” (ibid, p. 232). A ‘nongoal equilibrium’, on the contrary, “dictates an equilibrium state ... in which ... opposing forces (demand and supply, for example, the authors) are just balanced against each other, thus obviating any further tendency to change”. (ibid, p. 231).
In comparison, the points Q and R represent the suboptimal solutions. Although Q and R fulfil the ‘budget constraint’ of the log-normal distribution, they are located on the less favourable iso-damage curve I2. The equilibrium level of inequality for the individual and the society, economy and political system as a whole is now also being derived formally. As after learning the optimal solution’s intuition through the graphical analysis, the study has to proceed to a straightforward mathematical solution of what has been described ahead.
The individual
First, the personal desired level of income dispersion, σ*
Inserting the expression for
The Lagrangian for this maximisation problem is:
Taking the first-order condition (FOC) yields the following expression:
Inserting Eq. (2) in Eq. (1) yields the following expression:
Substituting
Multiplying both the sides with
Therefore, the desired level of income dispersion by the individual
In the aggregated utility function,
When the utility function is represented by a first-order Taylor series approximation around y
The Lagrangian for the maximisation problem now becomes the following:
Taking the FOC yields the following:
Substituting
Substituting
Dividing both sides by
Dividing both the sides by y
One should notice that both the constraints imply
Of course, external shocks will destroy any earlier equilibrium and lead to deviations from the former equilibrium level of inequality. The question as to which income groups will favour a subsequent redistribution policy depends on the external shock’s nature. If the shock tends to an increase (decrease) in the concentration of personal incomes, modern theory of inequity (equity) aversion (Bolton & Ockenfels, 2000; Sell, 2015) would suggest that the group of inequity (equity) -averse agents will push government to progressive (regressive) redistribution policies. It is important to note that our model, now as it stands, captures only the preference of inequity aversion.
The theoretical section of this article describes the development of a model which demonstrated that the (collective) choice of the preferred variance of income distribution results from a trade-off between the preference for one’s above-modus income, which is expressed as the first root term in the utility function, and the preference for a low concentration of the income distribution in the society, which is expressed as the second root term in the utility function. A testable hypothesis derived from the model was that there exists a long-term equilibrium value of the income distribution variance, towards which the society converges.
However, to conduct the empirical test, two issues had to be resolved first. The first issue was that the empirical data often contain Gini coefficients instead of the variance of the income distribution. The second issue was that it sounds counter-intuitive that the societies at different time points in history and those at different developmental stages should all converge to ‘the’ optimal variance of the income distribution, which would be clear later, is also not supported by the data.
The first issue could be conveniently resolved because of the assumed log-normal distribution of the incomes, which encompasses several real-world features (including a left-steep/right-skewed income distribution with a modus lower than the median and a median lower than the mean), thereby serving as a good approximation of the real-world data. Since the log-normal distribution serves as a fair approximation of the empirical distribution of incomes, it is a well-known fact that the Gini coefficient is a monotonically increasing function of the variance or its root, the standard deviation σ. More precisely,
The second issue could also be resolved conveniently by introducing a preference parameter into the developed basis model. The utility function would then become:
The introduced parameter γ stands for the weightage which the individuals of a society assign to the preference for an equal-income distribution (or against a higher σ), relative to the preference for a larger above-modus own income. The larger the value of γ, the lower would be the equilibrium standard deviation σ* (see Appendix A1). Although it is possible that in addition to varying across regions, σ* also changes over time due to shifts in preferences, it should remain relatively stable, such that an equilibrium state of σ* would nonetheless be observed in the absence of the underlying force driving the change.
To test our model, we need more restrictive assumptions about γ. Given that it represents the relative preference of more equal income over a higher own income, or solidarity over competition, we postulate that γ does not only change slowly over time, but that it also converges to similar values for countries which are closely linked each other. Thus, the Gini steady-state value in an integration process would not diverge but rather converge to similar values. This fact makes our model prediction similar to the neoclassical growth model (NGM), which predicts a convergence of Gini coefficients of various countries linked via free trade. However, the difference between our model and the NGM is that we are explicitly dealing with after-tax income distribution. Thus, the parameter of preference γ plays an essential role through the collective decision process. Furthermore, our model is about steady-state values of Gini, not Gini per se. Thus, we do not strictly assume that Gini’s change would be (negatively) proportional to its initial value. The following empirical test shows that our model better fits the data. When modelling Gini’s change, we also assume that a negative relationship exists between the initial value and the change of
European data on income distribution stem from “EU Statistics on Income and Living Conditions (EU-SILC) (2015).
The
Figure 4 plots the Gini coefficients against the calendar years; each of the 28 countries is on a separate line. Missing data are plotted as a dotted line in between using linear interpolation. The year 2008 is marked with a grey vertical line. Figure 5 rearranges the data by plotting the Gini coefficients against membership years in the EU (negative numbers correspond to years before joining the EU) instead of plotting against the calendar years. The year of joining the EU is plotted as a grey vertical line through the zero point. The year 2008 now appears as a short grey vertical line through the corresponding data points. The Gini coefficients of the countries, as one can see in Figure 5, converge (are getting closer) to each other, the longer the respective EU membership.
This convergence looks further clearer when we plot the Gini coefficients minus the group mean against membership years in the EU in Figure 6. The group mean is computed as the mean value of all EU members in each year, namely, Belgium, Denmark, Germany, Finland, France, Greece, Ireland, Italy, Luxembourg, the Netherlands, Austria, Portugal, Spain, Sweden and the United Kingdom from the beginning on, then plus Estonia, Latvia, Lithuania, Malta, Poland, Slovakia, Slovenia, Czechia, Hungary and Cyprus from 2004 onwards. Bulgaria and Romania joined in 2007 and Croatia in 2013. Because the dataset ended in 2019, the UK has always been counted as an EU member.
Whether this convergence is due to the EU membership, as a first glance on the graphs would suggest, or rather due to the worldwide observed convergence in Gini coefficients as documented in Ravallion 2001, Ravallion 2003 and Savoia 2019, is unclear. To distinguish the two types of convergence, we also plotted Gini coefficients minus the overall mean, namely, the mean value of all 28 countries each year, in Figure 7. Here, the same degree of convergence can be detected. Thus, the observed convergence may also be part of a worldwide inequality convergence process, derived from the NGM, as pointed out in Ravallion 2003. Figure 7 is almost identical to Figure 6. The convergence process is not very well visible in Figure 4 because the old member countries in our data are from the beginning (in 1995) near the overall mean; thus, they remain hidden among the data points and do not stand out. When plotted against years of membership in the EU instead of calendar years in Figures 5–7, the old member countries are put on the right side while the new member countries are on the left side; then the convergence becomes more visible. That the old member countries are from the beginning onwards closer to the overall mean may be because they have been more extended members in the EU, but it could also be because they have been longer integrated into the world market economy.
To investigate why Figure 7 is almost identical to Figure 6, which share the same
In Figures 5–7, we can also see that the Gini coefficients of the old member countries are closer to the overall mean, without having the same value, and that they show less convergence towards the overall mean. This may be because they converge more slowly, and they do this because they are closer to the steady state as predicted by the NGM (which predicts the same long-run steady state for all countries). It may also well be that the countries implied have slightly different steady states and are pending around them as pointed out by CCT. CCT explains that some long-lasting differences among countries are due to some hard-to-change underlying country-specific institutional factors. Hence, countries that are similar concerning important institutional factors converge better. Since the steady states are quite close, and the change is small, one cannot determine which theory is approximately more correct. What we can observe, however, is that if there were some hard to change differences among the EU countries, then they are relatively small, and being in the EU did not force the long-term members to all converge to the same
The NGM predicts a negative relationship between the change in and Gini coefficients’ initial value as documented in Ravallion (2001), Ravallion (2003) and Savoia (2019). Our data mostly confirm the prediction of the convergence theory, as shown in Figure 9. Since our data set is unbalanced, i.e. not every country has the same
It would be interesting to see if this negative relationship remains if we break the data into as many subsamples as possible, namely, when each subsample only encompasses two years. Then, we are able to plot the annual change of the Gini coefficient, Δ
The convergence hypothesis, widely represented in the literature, is usually expressed as a negative relationship between the change in the respective Gini coefficient and its initial value. The higher the initial value, the lower the expected change (the Gini coefficient increases slower or it decreases faster). Alternatively, one could formulate this negative relationship as follows: the higher the initial value above the overall mean, the lower the change. Graphically spoken, one would plot
Convergence in the Gini coefficients can also be explained by the fixed effects (FE) approach or by country dummies, an idea similar to CCT. When Δ
To see how similar two parameters are to each other, one could directly look at the parameters and ignore their standard errors, as we did in the bottom part of Figure 12 for the country dummies. Alternatively, one can look at their pairwise
It is interesting to see what happens when regressing Δ
Our parameter
The respective standard errors are given in parentheses.
Finally, we present the ADF test results for the Gini coefficients in Table 2. Since the ADF test we use cannot deal with missing values, we only use the most extended recent time series without any missing value for each country. According to Lin and Huang (2012), the NGM also implies stationarity in time series of inequality measures such as the Gini coefficient. Findings based on our data set weakly support this view. While 13 out of 28 countries are stationary according to at least one of the three alternative measures, namely, AIC, BIC and t-stat, for 9 countries the unit root hypothesis cannot be ruled out at the 10% confidence level, with any measure. Similar results (with US data) can be found in Lin and Huang (2012); they showed that the unit root hypothesis cannot be ruled out for several states if omitting structural breaks and then they also showed that by taking into account up to two structural breaks, almost all time series are stationary. Although we have a structural break candidate with the year 2008, not do we did the structural break test because the start year of our time series is in the range of 2003–2010, which makes the test less meaningful, and because our own model does not require empirical stationarity; although our model does not suggest unit root, it does not rule out that
According to Lessmann and Seidel (2015), regional inequality depends on the respective per capita GDP,
Dependent variable
−0.053*** (0.013) | −0.249*** (0.030) | −0.265*** (0.030) | ||
0.321*** (0.110) | ||||
−0.210 (0.255) | 6.417*** (0.837) | 6.676*** (0.835) | ||
−0.171 (0.255) | 6.588*** (0.852) | 6.855*** (0.850) | ||
0.707** (0.312) | 9.187*** (1.067) | 9.481*** (1.063) | ||
−0.267 (0.389) | 7.329*** (0.989) | 7.497*** (0.982) | ||
0.246 (0.324) | 7.888*** (0.973) | 8.109*** (0.968) | ||
−0.143 (0.312) | 6.075*** (0.807) | 6.201*** (0.802) | ||
0.294 (0.283) | 6.528*** (0.800) | 6.741*** (0.796) | ||
−0.663** (0.292) | 7.619*** (1.039) | 7.932*** (1.036) | ||
0.085 (0.261) | 6.342*** (0.796) | 6.553*** (0.793) | ||
−0.205 (0.268) | 7.022*** (0.910) | 7.318*** (0.908) | ||
−0.020 (0.261) | 7.017*** (0.886) | 7.278*** (0.883) | ||
−0.259 (0.249) | 8.180*** (1.048) | 8.548*** (1.047) | ||
−0.100 (0.292) | 6.679*** (0.865) | 6.875*** (0.860) | ||
−0.271 (0.255) | 7.475*** (0.968) | 7.806*** (0.966) | ||
−0.175 (0.261) | 7.775*** (0.993) | 8.110*** (0.991) | ||
−0.071 (0.312) | 8.871*** (1.121) | 9.172*** (1.117) | ||
−0.060 (0.302) | 8.694*** (1.097) | 9.002*** (1.093) | ||
0.076 (0.255) | 7.023*** (0.874) | 7.302*** (0.872) | ||
0.071 (0.312) | 6.972*** (0.885) | 7.142*** (0.879) | ||
−0.115 (0.261) | 6.558*** (0.844) | 6.811*** (0.841) | ||
−0.473 (0.302) | 7.269*** (0.979) | 7.511*** (0.974) | ||
−0.281 (0.255) | 8.541*** (1.094) | 8.926*** (1.093) | ||
−0.179 (0.312) | 8.331*** (1.071) | 8.605*** (1.066) | ||
−0.408 (0.324) | 5.829*** (0.813) | 5.960*** (0.808) | ||
0.006 (0.292) | 5.880*** (0.761) | 6.017*** (0.757) | ||
0.025 (0.261) | 8.351*** (1.037) | 8.710*** (1.036) | ||
0.369 (0.292) | 6.662*** (0.809) | 6.847*** (0.805) | ||
0.118 (0.283) | 8.179*** (1.011) | 8.528*** (1.010) | ||
1.482*** (0.389) | ||||
ADF test result
Austria | ** | ** | ** | ** | ** | ** | 17 |
Belgium | ** | 17 | |||||
Bulgaria | 14 | ||||||
Croatia | *** | 10 | |||||
Cyprus | * | 15 | |||||
Czechia | 15 | ||||||
Denmark | *** | *** | *** | 17 | |||
Estonia | *** | *** | * | 16 | |||
Finland | 16 | ||||||
France | ** | ** | ** | 15 | |||
Germany | ** | ** | *** | *** | *** | 15 | |
Greece | 17 | ||||||
Hungary | ** | * | ** | ** | ** | 15 | |
Ireland | *** | *** | *** | 16 | |||
Italy | *** | *** | *** | 15 | |||
Latvia | ** | ** | ** | *** | *** | *** | 15 |
Lithuania | 15 | ||||||
Luxembourg | * | * | 17 | ||||
Malta | * | * | * | * | * | * | 15 |
Netherlands | *** | *** | *** | *** | *** | *** | 15 |
Poland | ** | ** | ** | 15 | |||
Portugal | ** | ** | ** | 16 | |||
Romania | ** | ** | 13 | ||||
Slovakia | 14 | ||||||
Slovenia | 15 | ||||||
Spain | 16 | ||||||
Sweden | *** | *** | *** | ** | ** | ** | 16 |
United Kingdom | 14 |
Dependent variable
210.401** (101.18) | 463.717*** (146.76) | |||
−22.158** (10.25) | −46.512*** (15.17) | |||
0.765** (0.35) | 1.541*** (0.52) | |||
previous |
0.950*** (0.01) | 0.351*** (0.05) | ||
Austria | −253.595** (104.02) | −1753.78*** (482.91) | ||
Belgium | 272.008*** (104.02) | −1437.44*** (479.71) | ||
Bulgaria | −1451.324***(108.02) | −1902.24*** (385.60) | ||
Croatia | 500.411* (283.47) | −1291.01** (557.56) | ||
Cyprus | −377.443*** (145.40) | −1706.16*** (489.32) | ||
Czechia | 171.240 (145.40) | −1537.05*** (512.88) | ||
Denmark | −535.120*** (118.87) | −1892.70*** (480.87) | ||
Estonia | 323.865*** (100.53) | −1448.45*** (508.94) | ||
Finland | 123.248 (100.53) | −1460.13*** (480.53) | ||
France | −265.087** (109.12) | −1737.01*** (486.63) | ||
Germany | −387.554*** (117.02) | −1825.68*** (491.62) | ||
Greece | 87.160 (104.02) | −1331.06*** (460.23) | ||
Hungary | −332.505*** (104.51) | −1832.71*** (500.85) | ||
Ireland | 155.019 (113.32) | −1520.09*** (458.52) | ||
Italy | −291.261** (120.55) | −1686.74*** (480.19) | ||
Latvia | 459.139*** (145.40) | −1280.34** (507.37) | ||
Lithuania | −487.852*** (117.02) | −1868.91*** (510.14) | ||
Luxembourg | −488.867*** (104.02) | −1838.25*** (454.26) | ||
Malta | −181.419 (145.40) | −1882.67*** (541.84) | ||
Netherlands | 58.677 (104.51) | −1571.00*** (476.53) | ||
Poland | 421.547*** (117.02) | −1298.42*** (501.97) | ||
Portugal | 686.918*** (110.50) | −1085.10** (483.61) | ||
Romania | −15.311 (183.48) | −1518.38*** (488.56) | ||
Slovakia | 596.187*** (162.52) | −1212.39** (526.13) | ||
Slovenia | −199.579* (104.51) | −1764.69*** (504.61) | ||
Spain | −240.605** (100.53) | −1695.74*** (486.51) | ||
Sweden | −578.102*** (113.37) | −1972.66*** (481.94) | ||
United Kingdom | 324.630*** (113.35) | −1358.44*** (484.78) | ||
Austria*TIME | 0.140*** (0.05) | 0.122** (0.05) | ||
Belgium*TIME | −0.122** (0.05) | −0.035 (0.05) | ||
Bulgaria*TIME | 0.739*** (0.05) | 0.198* (0.11) | ||
Croatia*TIME | −0.233* (0.14) | −0.109 (0.13) | ||
Cyprus*TIME | 0.203*** (0.07) | 0.099 (0.07) | ||
Czechia*TIME | −0.073 (0.07) | 0.012 (0.07) | ||
Denmark*TIME | 0.279*** (0.06) | 0.192 (0.06) | ||
Estonia*TIME | −0.145*** (0.05) | −0.030 (0.06) | ||
Finland*TIME | −0.048 (0.05) | −0.024 (0.05) | ||
France*TIME | 0.146*** (0.05) | 0.115** (0.05) | ||
Germany*TIME | 0.207*** (0.06) | 0.159*** (0.06) | ||
Greece*TIME | −0.027 (0.05) | −0.087* (0.05) | ||
Hungary*TIME | 0.179*** (0.05) | 0.160*** (0.05) | ||
Ireland*TIME | −0.062 (0.06) | 0.008 (0.07) | ||
Italy*TIME | 0.161*** (0.06) | 0.090* (0.06) | ||
Latvia*TIME | −0.210*** (0.07) | −0.112 (0.07) | ||
Lithuania*TIME | 0.260*** (0.06) | 0.180** (0.08) | ||
Luxembourg*TIME | 0.257*** (0.05) | 0.165*** (0.05) | ||
Malta*TIME | 0.104 (0.07) | 0.185** (0.08) | ||
Netherlands*TIME | −0.016 (0.05) | 0.032 (0.05) | ||
Poland*TIME | −0.194*** (0.06) | −0.105** (0.07) | ||
Portugal*TIME | −0.324*** (0.06) | −0.209*** (0.05) | ||
Romania*TIME | 0.025 (0.09) | 0.006 (0.10) | ||
Slovakia*TIME | −0.284*** (0.08) | −0.149* (0.08) | ||
Slovenia*TIME | 0.111** (0.05) | 0.125** (0.05) | ||
Spain*TIME | 0.136*** (0.05) | 0.095** (0.05) | ||
Sweden*TIME | 0.300*** (0.06) | 0.231*** (0.06) | ||
UK*TIME | −0.145*** (0.06) | −0.073 (0.05) | ||
constant | −624.115* (331.96) | 1.450*** (0.43) | ||
Observations | ||||
In the previous subsection, we found that
The respective standard errors are given in parentheses.
In estimation model B(3), we regress
The country dummies’ estimates are not very relevant (currently, they stand for the predicted
Combining the estimates of the dummies and their interaction terms, we could calculate a predicted value of
The
Taking all together, the data support our model quite well. Our regression results are also supportive for the convergence theorem derived from the NGM, though our less restrictive assumptions in line with the CCT can increase the model fit significantly. Our findings also suggest that the CCT can fit our data only if the predicted differences between countries are not too high (the right-hand part in Figure 5 suggests a roughly ± 5 band somewhere around somewhat less than 30 for the EU countries’
The empirical model of an N-shaped income-determined Gini coefficient is interesting for our model because income was included in our model but then cancelled out after the optimisation and aggregation. This cancelling-out effect is intended because, as mentioned earlier, the per capita GDP, or its log, is not bounded, while
Convergence in inequality is a big issue in empirically oriented research on personal income distribution. The majority of contributions in our empirical part test the NGM/convergence theory and CCT, as well as our approach. While the first suggests that market forces (international and intranational trade, factor movements, etc.) push the tendency towards convergence, the latter (and theoretical strands close to it; see the introduction) explains why primarily institutional factors—similar within clubs, but not similar outside—tend to put forward some sort of ‘conditional convergence’ in income inequality. Our approach is close to CCT but stresses much more the political decision process as a determinant for the steady state in income distribution.
Our empirical research tends to support our hypothesis, but the findings can be differentiated as follows:
That convergence in the Gini coefficients can also be explained by the FE approach or country dummies, which are similar to CCT. Our empirical findings reveal that combining the convergence prediction of NGM with FE/CCT is a viable strategy.
We ran additional regressions that detected the impact of time trend, previous Gini coefficients, country dummies and their interaction terms and of real per capita GDP on the actual inequality of personal incomes. Furthermore, we could confirm the Lessmann and Seidel result (2015), according to which regional inequality depends on the respective per capita GDP,
The present study’s limitations are related to the availability of consistent and comparable data on the personal income distribution’s statistical moments, such as the modus of incomes. Consequently, it was not possible to test, directly, all the implications of the theoretical model. Future research on personal income distribution in Europe could intend to deepen the knowledge regarding the convergence-stimulating effects of the economic crises or the change in the institutional setting and could also interconnect the analysis of personal income distribution with the development of macroeconomic shares of total income (profits and wages).