2.1. Theoretical Basis of Economic Integration

According to Molle’s (Reference Molle1997, 281–295) definition, economic integration refers to a static situation or a dynamic process whereby member countries relinquish their own cross-border barriers mutually and integrate into a greater union. The purpose of the union could be, in general, economic and political. It is expected to enhance economic welfare by advancing specialisation and the division of works according to the principles of comparative advantage and reciprocal cooperation while eliminating trade impediments under market segmentations among members. Economic integration is also intended to lower the risk of military conflict by enhancing the degree of economic interdependence. In some cases, it helps to promote democracy and improve human rights in newly transitioned economies.

Economic theories that address the benefit of integration mainly take two routes. One involves ‘neoclassical economics’. Based on assumptions about perfect competition, a constant return to scale in production and a diminishing return of production factors, economic growth could be generated in the short and medium term but cannot be sustainable in the long term. Global free trade serves as the best channel for promoting economic welfare for individuals. All the protection measures against free trade, such as non-tariff barriers, would cause immobile factors of labour, an incontestable market and an unelaborated comparative advantage and eventually move resources away from efficient allocation (Gilpin Reference Gilpin2001, 126–127).

From a static viewpoint, neoclassical economics suggests clarifying the benefit of integration with the ‘trade creation effect’ and the ‘trade diversion effect’, as proposed by Viner (Reference Viner1950). The former suggests that better trading terms among members within an integrated area could result in more effective resource allocation and contribute positively to welfare. The latter suggests that asymmetry of trading terms between the members and the non-members of the integrated area would alter the original resource allocation and affect welfare negatively. If the former is larger than the latter, the net effect would be positive and vice versa (Ginsberg Reference Ginsberg2007, 99–103; Zervoyianni and Argiros Reference Zervoyianni, Argiros, Zervoyianni, Argiros and Agiomirgianakis2006, 93–134; Robson Reference Robson1998, 18–26).

Contrary to the static effect of neoclassical economics, the new economic geography theory, originating from the seminal work of Krugman (Reference Krugman1991, 483–499), places more focus on the contribution of dynamic effects. Krugman emphasised that the real economic world is indeed characterised by an imperfect competitive market, an increasing return to scale in production and product differentiation. These market characteristics, including an increasing marginal return, could make economic growth permanently sustainable (Cuaresma et al. Reference Cuaresma, Ritzberger-Grunwald and Silgoner2008, 650–643; Gilpin Reference Gilpin2001, 126–127). Regional integration would thus eventually create at least the following effects on markets or industries and, in turn, promote social welfare (Robson Reference Robson1998, 268–277): (1) tariff elimination could amplify the market contestability and lower the market power of individual firms internally; (2) the enhancement of market competition would increase the advantages of operation and products for firms with greater efficiency and accommodate flexibility; (3) product diversification would be inspired by the greater product competition; (4) production efficiency and scale economies would be enhanced by stronger specialisation and division of works; and (5) external bargaining power could be increased. In sum, regional integration would strengthen market contestability and eventually raise the productivity of members to achieve growth through the competitive effect, the scale effect and the investment effect.

Politically, ‘intergovernmentalism’ and ‘new functionalism’ are the two main ideologies of integration. Intergovernmentalism refers to a theory explaining why integration originates and develops in the realist context of international relations, despite the sovereign rights of governments being restrained. New functionalism is a critical and introspective theory opposing realism. Both these political theories interpret regional integration; however, they expound poorly whether integration is beneficial for economic welfare. ^{Footnote 1}

2.2. Studies on the Growth Effect of Integration

Past studies on the productivity change of the eurozone have mostly glorified the effect of integration and can be distinguished roughly into two strands. One strand handled the topic at the macro level. Cuaresma et al. (Reference Cuaresma, Ritzberger-Grunwald and Silgoner2008, 650–653) empirically examined the effect of European integration on the long-term growth of the current EU member states. They concluded that EU membership had a significant positive effect on economic growth and that the effect was relatively stronger for poorer countries. Kutan and Yigit (Reference Kutan and Yigit2007, 1370–1395) adopted a stochastic endogenous growth model to examine the impact of EU integration on productivity change and convergence. The results showed that integration is beneficial to members, especially from a long-term perspective, while cohesion and structural funds assist newcomers in catching up with the core members’ standard of living.

Badinger (Reference Badinger2005, 50–78) compiled an index of economic integration accounting for the global and regional integration of the EU member states to test for growth effects within a growth accounting framework, using a panel of 15 EU member states over the period 1950–2000. The results suggested that the growth effect is inclined to be temporary and that integration has contributed 20% to 44% of income growth to the EU areas since 1950. Johansson (Reference Johansson2001, 1–20) investigated how regional integration and the different treatment of members and non-members affect the relationship between imports and productivity growth. The study focused on the EU and found that intra-union imports have a greater effect on total factor productivity growth than imports from outside the EU.

The other strand dealt with the subject at the industrial or firm level. Bottasso and Sembenelli (Reference Bottasso and Sembenelli2001, 167–186) examined the impact of the EU Single Market Programme (SMP) on the market power and total factor productivity of Italian firms. It was revealed empirically that, for sensitive firms, the market power decreased by 50% in the SMP implementation period and that there was a positive transitory shock to the productivity growth rates after the announcement of the reform project. Cincera and Galgau (Reference Cincera and Galgau2005) explored the impact of the product market reforms of the EU on firm entry and exit rates. The results showed that an increase in the level of deregulation caused by the SMP led to an increase in both the entry and the exit rate, a decrease in price controls and hidden import barriers and an increase in regulatory quality. These characteristics in turn contributed to the macroeconomic performance.

Not all the literature has praised the contribution of integration, however; some studies have taken a neutral or opposite viewpoint, and some have concerned regional imbalance and inequality. Wooster et al. (Reference Wooster, Banda and Dube2008, 161–182) suggested that, under the prerequisite of regional (but not global) integration, a growth promotion effect actually exists in the integrated area; nonetheless, there are also greater scale economies and competition effects outside the area. The conclusion thus depends on the circumstances. Martin and Sanz (Reference Martin and Sanz2003, 205–236) emphasised that economic gain will be not generated automatically from integration but depends on whether the individual members have valid corresponding strategies and conditions to create an attractive business environment for the technological spill-over effect from foreign direct investment. Foreign direct investment was identified as a spur for European economic integration by Gorg and Ruane (Reference Gorg and Ruane1999, 346) and Robson (Reference Robson1998, 119).

Venables (Reference Venables2003, 747–761) pointed out that not all members can be winners and benefit from integration and that the outcomes depend on their comparative advantage relative to other members and the rest of the world. It was concluded that developing countries are likely to be served better by ‘north–south’ than by ‘south–south’ agreements. Aristotelous (Reference Aristotelous2006, 17–21) also supported the coexistence of positive and negative effects on trade for members after reviewing the experience of the EU. Heidenreich and Wunder (Reference Heidenreich and Wunder2008, 19–20) and Pardo (Reference Pardo2005, 459–467) noted that regional integration could shrink the income gap among members within the integrated area but widen the economic inequality across countries from different regions.

2.3. Risks Involved in Integration

Amidst a large body of literature that has continuously focused on the effect of integration, the mainstream of the world economy continuing to push forwards to a higher degree and a wider scope of integration, an interesting question is whether there is a trade-off in integration. Theoretically, integration should enhance the extent of risk taking. No one can guarantee that the appropriateness and harmony of those corresponding policies and the constitution will sustain an economy. Breuss (Reference Breuss2002, 245–274) anticipated this worry to a certain extent. He adopted a world macroeconomic model to check the effect of EU enlargement and empirically asserted that EU enlargement seemed to be a win–win situation, although the effect was quite different across the union members. Enlargement acts as an exogenous structural shock leading to asymmetric disturbances due to the different regional impacts. The study thus concluded that enlargement may not be solely beneficial but might be a risk for the EU.

When an economy starts to open from autarky to free trade, or even more to integration with others, there is an immediate risk from two aspects. The first is the quantity of allied members. Each member increase represents the dismantling of a firewall and the opening of a risk propagation channel to the member that was originally unconnected. The more members there are, the higher the risk of being involved is. The second is the degree of alliance. Other things being equal, the propagation capacity of risk depends on the extent of blockage power of the firewall. A null degree of alliance implies full blockage, and a full degree of alliance naturally means complete risk. The frightening point, however, is that integration seems to be an irreversible one-way street. Stiglitz (Reference Stiglitz2010, 388–389) pointed out that, in theory, the conventional idea of the integration of global financial markets leading to greater financial stability is doubtful. Owing to the greater risk sources and linkages across economies that would accompany the integration, a local failure could lead to a system-wide failure without firewalls. He thus asserted that full financial integration may not be desirable.

Forbes and Rigobon (Reference Forbes and Rigobon2002, 2223–2261) categorised the risks that are transmitted or disseminated among member states into three formal definitions: ‘the monsoonal effect’, ‘the spill-over effect’ and ‘pure contagion’. The monsoonal effect implies that stresses appear to be contagious because the underlying macroeconomic variables are correlated. A spill-over effect signifies that hazards in one economy might affect the other economies through external connections such as trade. Pure contagion denotes a significant increase in cross-market linkages after a shock to one country or a group of countries, which could make the market jump from a ‘good’ to a ‘bad’ equilibrium. The former two effects were categorised as ‘interdependence’, and the latter was defined as ‘contagion’.

This section reviewed the theoretical interpretations of the effect of integration, empirical examinations of integration experience and the potential risk accompanying integration. It showed that using integration as a way to pursue income growth or promote welfare may produce uncertain results due to the corresponding risk that always exists. Logically, the derived risk should be taken into account endogenously prior to the decision about joining the union and cannot be removed exogenously after initiating the integration process. Either way, risk taking should at least be considered in the framework of productivity measurement. More specifically, the additional effect of the risk premium should be culled from performance, because a biased estimation from ignoring risk is preferable to overly exaggerating the benefit. Without having a clear understanding about this, while the eurozone or other regions continue to integrate further, the behaviour might just be a risky trial or a great gamble.

4.1. Capital Market Return and GDP Growth

This subsection initially provides a review of the fluctuation in the capital market return and GDP growth. Figure 1 provides the first glance at the EZ, EU, G8, G20 and EM economies during the sample period, with histograms. The figure shows that the EZ economy lacks a prominent performance in its economic growth and stock market return compared with other economies. Meanwhile, we can see that the directions of the GDP growth and stock market return for the EZ are similar to those for other areas, although the margins of variation are inconsistent. Among the five classification economies, the G20 shows the largest increase in its GDP. During the global financial tsunami in 2008–2009, the EU and EZ suffered deeper negative GDP growth. The subsequent rebound of their capital market return was not stronger than that of other regions.

Figure 1.

Histograms of GDP growth and stock price return for the EZ, EU, G8, G20, and emerging economies (EM) during the sample period.

Note: y-axis: change %; x-axis: annual period.

Figure 1 has some implications. First, GDP growth generally reflects the change in the total output and consists of productivity and non-productivity factors. It is difficult to identify the latent productivity merely from the surface numbers of the GDP. The stock market returns are reflected in the price changes, which could be considered not only as a response to macro-and micro-economic situations but also as a reaction to the degree of risk bearing of an economy. It is, of course, impossible to obtain the degree of risk bearing merely from the price change in the stock market. Accordingly, it becomes apparent that the trends and margins of the GDP growth and of the capital market return are not necessarily consistent. In the meantime, these figures cannot indicate whether the EZ gained superior productivity growth or bore a higher economic risk.

Table 1 displays additional comparisons of economic performance between the EZ and the other economies, using the regressions specified in equation (8). In Table 1, the regressions show that the time trends of the GDP growth and stock market return are insignificant. Relatively, the period after the year 1999 had significantly higher GDP growth than the period before it. There is no similar effect for the stock market return. Neither the EU nor the EZ show significantly higher GDP growth or stock market return relative to the other economies.

Table 1.

GDP growth and stock price return regression for the EU, EZ, G8, G20 and emerging economies.

Notes: ***, ** and * denote coefficient significance at 1%, 5% and 10%, respectively. The figures in the parentheses are standard errors. All the figures in this table are provided with 2 digit significance.

^a: Dummy variable equals 1 if the sample country belongs to the European Union, otherwise 0. ^b: Dummy variable equals 1 if the sample period is after January 1999, otherwise 0. ^c: Dummy variable equals 1 if the sample country is a member of the Eurozone.

With these ambiguous results from Figure 1 and Table 1, it is conceivable that, to reveal the real economic performance of an economy, there is a need to unearth the potential productivity from the GDP on the surface and to extract the degree of potential risk bearing from the capital market price fluctuation. These are the tasks of the subsequent subsections.

4.2. Risk Bearing and Productivity Estimation

In accordance with equation (1) in Section 3, and to gauge the degree of risk bearing for each country in the group, we start with the stock market return and confirm the white noise lags for the mean and variation equations. The appendix includes all the tested results of ARMA(p, q)–GARCH(m, n). The standard mGARCH system regressions, shown as equations (1) and (2) in Section (3), can be estimated with the above information and other controlled variables. The conditional variance and conditional covariance of the capital market return for each economy can then be estimated. We then take the square root of the conditional variance for each economy as the degree of risk bearing, S, and the conditional covariance between each economy and the EZ, cov(S), as the degree of risk linkage.

Table 2 shows the regressions of S and cov(S) on the four dummies $D_1^{{\rm{EU}}}$ , $D_2^{1999}$ , $D_3^{{\rm{EZ}}}$ and T, as in Table 1. Despite the similarity of the explanatory variables, the empirical results become particularly interesting. For regression S, the negative sign of T shows that the degree of risk bearing for the sample period demonstrates a roughly downward trend over time. The negative sign of $D_1^{{\rm{EU}}}$ reveals that the degree of risk bearing for the EU economies is higher than that for the non-EU economies. The negative sign of $D_2^{1999}$ shows that the degree of risk bearing for the period after 1999 is lower than that for the years before 1999. A more interesting point is that $D_3^{{\rm{EZ}}}$ shows a significant positive result. This indicates that the EZ economies bear a higher degree of risk than other economies, which is also higher than the risk before joining the eurozone.

Table 2.

Conditional variance and covariance for the sample counties.

Notes: ***, ** and * denote coefficient significance at 1%, 5% and 10%, respectively. The figures in the parentheses are standard errors. All the figures in this table are provided with two-digit significance. a: Dummy variable equals 1 if the sample country belongs to European Union, otherwise 0. b: Dummy variable equals 1 if the sample period is after January, 1999, otherwise 0. c: Dummy variable equals 1 if the sample country is a member of the Eurozone. d: Square root of the estimated conditional variance for the sample countries are used as the dependent variable. e: estimated conditional covariance between the sample countries and the Eurozone are used as the dependent variable.

In the cov(S) regression, the negative sign of T suggests that the degrees of risk linkage for the sample during the period also show a downward trend. The positive sign of $D_1^{{\rm{EU}}}$ reveals that the degrees of risk linkage between the EU and the EZ economies are higher than those between the non-EU and the EZ economies. The negative sign of $D_2^{1999}$ shows that the degree of risk linkage between the sample economies and the EZ is generally lower after the year 1999. The positive sign of $D_3^{{\rm{EZ}}}$ reveals that the degree of risk linkage between the EZ and the overall EZ economies is significantly higher than that between the non-EZ and the EZ economies.

Generally speaking, according to Table 2, consistent with intuition, integration induces a degree of risk linkage among economies because of the increased number of members and the deepened connection among members. Meanwhile, as the sources and channels of risk rise, the degree of risk bearing is correspondently enhanced. With the estimated S, the production function can further be estimated, and, by considering the risk input and the total factor productivity index, it can thus be calculated with the frontier parameters as well.

Following equation (3) in Section 3, Tables 3 and 4 demonstrate the maximum likelihood estimation (MLE) of production functions in translog form. It is important to notice the difference between the two tables. Specifically, Table 3 refers to the results under ordinary settings, which do not include the degree of risk-bearing variable S in the model and are used as the control for Table 4. In comparison, Table 4 refers to the results using the parameters of this study; it incorporates the degree of risk-bearing variable R into the model and is used as an experiment relative to the results in Table 3.

Table 3.

MLE estimation under pooling, separating and metafrontier specifications without consideration of risk-bearing.

Notes: ***, ** and * denote coefficient significance at 1%, 5% and 10%, respectively. The figures in the parentheses are standard errors. All the figures in this table are provided with two-digit significance. a: All the countries are used to estimate the stochastic frontier regardless of the segmentation of the technological difference. b: EU countries are used to estimate the stochastic frontier. c: Non-EU countries are used to estimate the stochastic frontier. d: log-likelihood of the regression estimation. e: Likelihood ratio test; H₀: the frontiers of the EU and non-EU countries are identical; H₁: the frontiers of the EU and non-EU countries are distinct. The LR statistic is defined by λ=–2{ln[L(H₀)]–ln[L(H₁)]}, where the ln[L(H₀)] is the value of the log-likelihood function for the frontier estimated by pooling all the banks, while the ln[L(H₁)] is the sum of the values of the log-likelihood functions for the two group frontiers. f: Likelihood ratio test; H₀: the frontiers estimation with and without all the inputs are identical; H1: the frontiers estimation with and without all the inputs distinct. The LR statistic is defined by λ=-2{ln[L(H₀)]-ln[L(H₁)]}, where the ln[L(H₀)] is the value of the log-likelihood function for the frontier estimated without all the inputs, while the ln[L(H₁)] is the value of the log-likelihood functions for frontier estimated with all the inputs.

Table 4.

MLE estimation under pooling, separating and metafrontier specifications with consideration of risk-bearing.

Notes: ***, ** and * denote coefficient significance at 1%, 5% and 10%, respectively. The figures in the parentheses are standard errors. All the figures in this table are provided with two-digit significance. a: All the countries are used to estimate the stochastic frontier regardless of the segmentation of the technological difference. b: EU countries are used to estimate the stochastic frontier. c: Non-EU countries are used to estimate the stochastic frontier. d: log-likelihood of the regression estimation. e: Likelihood ratio test; H₀: the frontiers of the EU and non-EU countries are identical; H₁: the frontiers of the EU and non-EU countries are distinct. The LR statistic is defined by λ=–2{ln[L(H₀)]–ln[L(H₁)]}, where the ln[L(H₀)] is the value of the log-likelihood function for the frontier estimated by pooling all the banks, while the ln[L(H₁)] is the sum of the values of the log-likelihood functions for the two group frontiers. f: Likelihood ratio test; H₀: the frontiers estimation with and without all the inputs are identical; H₁: the frontiers estimation with and without all the inputs distinct. The LR statistic is defined by λ=–2{ln[L(H₀)]-ln[L(H₁)]}, where the ln[L(H₀)] is the value of the log-likelihood function for the frontier estimated without all the inputs, while the ln[L(H₁)] is the values of the log-likelihood functions for frontier estimated with all the inputs. g: Likelihood ratio test; H₀: the frontiers estimation with and without considering risk-bearing (lnR) inputs are identical; H₁: the frontiers estimation with and without considering risk-bearing (lnR) inputs are distinct. The LR statistic is defined by λ=–2{ln[L(H₀)]–ln[L(H₁)]}, where the ln[L(H₀)] is the value of the log-likelihood function for the frontier estimated without lnR, while the ln[L(H₁)] is the values of the log-likelihood functions for frontier estimated with lnR.

There are at least five points to check from Table 3. First, the ‘pooling’ column indicates that the regression is estimated without segmenting the observations into distinct technological groups. The estimation implicitly assumes that all the economies belong to the same technological group and face an identical technological frontier. The log likelihood ratio of this MLE is shown to be 651.1490. The ‘EU’ and ‘non-EU’ columns denoting the regressions are estimated separately for the EU group and the non-EU group, respectively. The log likelihood ratios of the MLE are 327.0024 and 331.3901, respectively. To find out whether separate estimates are better, we adopt a likelihood ratio (LR) test, which regards the pooled estimate as a restricted model and the separate estimates as an unrestricted model. The null hypothesis here means that the log likelihood ratios of the two regressions are not different. The results show that the λ value of the LR statistic is 74.4870, which is significantly higher than the critical value of the chi-square distribution of 34.8053 under the conditions of the 1% significance level and 18 degrees of freedom. The result notably supports the separate estimates empirically. The ‘meta’ column denotes the metafrontier regression, which is a set of estimations based on the group frontiers, which implies that economies can overtake the existing technological upper limit in the long term and reach a potentially feasible common technology level. We also estimate that, with the overall output elasticity of factors, the scale elasticity is 1.0340 for the EU economies on average. This suggests an increasing return to scale. However, the scale elasticity is 0.9599 for the non-EU economies, which is a decreasing return to scale. One point worth noting here is that all the results shown in Table 3 and discussed above are estimates that ignore the degree of risk taking in the models.

Similar to Table 3, Table 4 provides another set of MLEs for the frontier regressions. The difference between these two tables is that Table 4 now refers to the estimation when considering the degree of risk bearing in the models as the experiment of this study, according to the specification in Section 3 with the translog form. Several interesting points are revealed by this table. First, to determine whether the group frontiers are more suitable than the pooled frontier, the likelihood ratio test shows that the separate regressions have a significantly better fit at the 1% level. Second, to investigate whether the model that considers the degree of risk bearing is better, we conduct another log-likelihood ratio test. We use the results of Table 3 as the restricted regression, which ignores the degree of risk bearing, and the results of Table 4 as the unrestricted regression, which considers the degree of risk bearing. The null hypothesis is that the estimates of the two regressions are not significantly different. The chi-square statistic shows that, at the 1% level and with a degree of freedom of 5, all four regressions consistently and significantly reject the null hypothesis. This implies that the regression considering the degree of risk bearing has a better fit. It is reasonable to incorporate the risk variable into the models.

Third, concerning the overall output elasticity of factors, we also find that, after considering the risk bearing, the EU’s scale economy is 1.0465; that is an increasing return to scale. Meanwhile, the non-EU scale economy is shown to be 0.9680; that is a decreasing return to scale. Both these values are a little higher than those revealed in Table 3. Fourth, regarding risk bearing as an intangible factor input, the output elasticity of risk input for the EZ and the non-EZ can be revealed. It is clear from the table that, while the degree of risk bearing increases by 1%, the contribution to the overall output is 0.0342% for the EU and –0.0162% for the non-EU. It is also worth remembering again that the ‘output’ in the overall output elasticity of factors is the total output, the GDP. In this study, however, we continually emphasise the total factor productivity and the contribution of risk bearing to the total factor productivity, that is, the risk premium. We thus utilise the parameter estimates obtained from Tables 3 and 4 to conduct the calculation.

Regarding the gMMPI and its decompositions from Section 3, Table 5 lists the calculated results. All the figures are shown in the table for the five economies classified in the sample and distinguished as being obtained with and without considering the degree of risk bearing. The t-statistic is also used to test the mean difference. We begin the check with the technical efficiency. First, for the EZ economies, the overall metafrontier technical efficiency (MTE) is shown to be 0.9956, if the risk bearing is not controlled, but it becomes 0.7796 when considering risk bearing. This implies that the efficiency score of the MTE would be significantly overestimated if the risk bearing is ignored. Second, the MTE is made up of the group technical efficiency (GTE) and the technology gap ratio (TGR). The GTE is estimated to be 0.9971, very close to 1, if the risk bearing is poorly controlled, and becomes 0.8719 if the risk bearing is controlled. The value will again be overestimated if the risk bearing is ignored. The TGR is estimated to be 0.9634 if the risk bearing is ill-controlled and becomes 0.8952 when controlling the risk bearing. The significant gap again reveals evidence of overestimation if risk bearing is ignored in the model.

Table 5.

Measures of efficiency and productivity scores and their mean difference tests estimated under with and without risk-bearing considerations.

Notes: ***, ** and * in the column ‘Diff. test’ denote coefficient significance at 1%, 5% and 10%, respectively. The figures in the parentheses are standard deviations.

We then check the productivity index. First, the gMMPI shows that the estimate is 1.0240 annually on average, if ignoring the risk bearing, and 1.0177 when considering risk bearing. It again shows that ignoring risk leads to an overestimation. Second, the gMMPI is made up of the technical efficiency change (TEC), technical change (TC) and scale efficiency change (SEC) and demonstrates that the TEC is about 1.0002 on average if ignoring the risk bearing and becomes 0.9873 after controlling the risk. It is implied that controlling the risk results in a reversal from efficiency improvement to degeneracy. The TC is estimated to be 1.0154 if ignoring risk bearing and 1.0340 if controlling the risk. This implies that ignoring the risk results in underestimation of technical change here. The SEC is shown to be 1.0083 if ignoring risk bearing and 0.9966 if controlling the risk. It again can be interpreted as an overestimated result. The average scores are also similar for the other economies. It can be seen from the difference test that ignoring the potential risk bearing could result in bias in an empirical study.

4.3. Productivity Comparison and Analysis

After reviewing the productivity scores, we now start to check the key empirical outcomes of the entire process for this study. Table 6 lists the regressions of estimated productivity and their components on the independent variables, as in Table 2. The table partitions the regressions into two parts: Panel A and Panel B. Panel A shows the results of ignoring risk bearing, while Panel B shows the results of controlling risk bearing.

Table 6.

Efficiency and productivity regression with the dummies of EU, EZ countries and time after 1999.

Notes: ***, ** and * denote coefficient significance at 1%, 5% and 10%, respectively. The figures in the parentheses are standard deviations. The number of observations for all the regressions in the table are 32 countries, 17 annual periods and thus 576 in total. All the figures are provided with two-digit significance.

^a: Dummy variable equals 1 if the sample country belongs to the European Union, otherwise 0. ^b: Dummy variable equals 1 if the sample period is after January, 1999, otherwise 0. ^c: Dummy variable equals 1 if the sample country is a member of the Eurozone.

We check the regressions of the gMMPI first. In Panel A, T shows a significant negative outcome, which implies that the gMMPI shows a downward time trend and the productivity for all the sample economies shrank with time. The $D_1^{{\rm{EU}}}$ manifests a significant positive coefficient. It suggests that the sample economies of the EU seem to have higher productivity performance than the non-EU economies. Further, $D_2^{1999}$ shows a significant negative result. This suggests that the productivity change after the year 1999 was lower than that before the year 1999 for the sample economies. With control of the above, we then check $D_s^{{\rm{EZ}}}$ . $D_s^{{\rm{EZ}}}$ has a significant positive result. The coefficient demonstrates that the EZ economies have superior productivity performance compared with the non-EZ economies and before they were affiliated with the EZ. Despite the interesting outcome, we need to keep in mind that this might be a spurious result, as the degree of risk bearing was ignored to control for the process of estimating the productivity scores.

The situation becomes much more interesting when we check Panel B and further compare it with Panel A as follows. In Panel B, T shows a significant positive outcome. It implies that the gMMPI has an upward time trend and that the productivity for all the sample economies was roughly expanding over time. This result is a complete reversal of that shown in Panel A. $D_1^{{\rm{EU}}}$ now also manifests a significant positive coefficient. This suggests that the sample economies of the EU seem to have higher productivity performance than the non-EU economies. This result is consistent with that observed in Panel A. Further, $D_2^{1999}$ shows an insignificant result. It suggests that the productivity change after the year 1999 is not significantly higher or lower than that before the year 1999 for the sample economies. This result is quite different from that shown in Panel A. With control of the above, we then check $D_s^{{\rm{EZ}}}$ . $D_s^{{\rm{EZ}}}$ does not reveal a significant result. The coefficient shows that the EZ economies did not have superior productivity performance to the non-EZ economies and before they were affiliated with the EZ. This result is different from that obtained in Panel A. It is worth noting that the gMMPI in Panel B refers to the productivity when considering the risk-bearing input; that is, the risk premium is excluded from the productivity scores. The outcomes in Panel B are therefore relatively trustworthy when applied to Panel A to inspect the real productivity performance.

In the other regressions for the components of productivity, the coefficients of $D_s^{{\rm{EZ}}}$ of Panel A and Panel B only differ in values, and there is no notable reversal or contradiction in the positive and negative signs. With control of the time trend, the EZ shows significant superior technical change (TC) and scale efficiency change (SEC), but the technical efficiency change (TEC) is shown to be relatively lower.

To demonstrate further the relationship between the degree of risk bearing and the productivity scores, Table 7 provides the regressions for the gMMPI^R, gMMPI^UR and risk premium on the degree of risk-bearing S. The empirical work then reveals some very interesting results. First, if regarding the risk premium from bearing more risk as part of the productivity gain, that is, measuring productivity without considering risk bearing, S then has a significantly positive relationship with the gMMPI^R. It can be tested whether one unit of risk bearing can create 0.201 units of the gMMPI^R. Second, if considering the risk premium as not belonging to the productivity gain, the degree of risk bearing should be controlled in the process of measuring productivity. S then has a significant negative relationship with the gMMPI^UR. Whether one unit of risk bearing would result in a decline in the gMMPI by 0.77 can be checked. The reason for this could be that higher risk would induce conservatism in business and restrain economic activity.

Table 7.

Regressions of productivity gain and risk premium on risk-bearing.

Notes: ***, ** and * denote coefficient significance at 1%, 5% and 10%, respectively. The figures in the parentheses are standard deviations. All the figures in this table are provided with two- digit significance.

Finally, we test for the relationship between risk bearing and the risk premium. Table 7 clearly shows a figure of 0.97. This implies that the ratio of risk bearing to the risk premium should be 1:0.97, almost equal to unity rather than larger than unity. It could be concluded that it is not surprising that the EZ made no extra productivity gain relative to the other economies in Table 6.