Determinants of Regional Infrastructure

In the previous section, I demonstrated that governance and institutions are positive ly correlated with infrastructure development. However, thus far I have not examined the functional relationship between governance and infrastructure, while controlling for exogenous factors. I therefore try to verify whether governance influences regional infrastructure development in interactions with other exogenous variables.

The bivariate associations discussed in the previous sections indicate that governance (institution quality) has a positive and significant effect on income and infrastructure. As Figure 1 indicates, trade (integration) can also have a positive impact on governance, infrastructure, and output, suggesting that trade (integration) can have an indirect effect on incomes by improving institutional and infrastructure qualities. Figure 11 [ PDF 17.3KB | 1 page ] better captures this complex interplay among these variables.

In order to find the probable determinants of regional infrastructure, I defined regional infrastructure as a product of (i) the scale and structure of country's economic size; (ii) domestic and international demand, through production and international trade; and (iii) governance in institutions, among others. I then estimated the following baseline equation:

where i represents a country, t time and ei is the error term. The dependent variable Infra, is the PII representing regional infrastructure. Gov is a composite measure of governance (represented by the governance index), X is a vector of additional regressors, and Region is a dummy variable rep resenting geographical regions with regional infrastructure : (i) Asia (=1 for Asian countries, 0 otherwise); (ii) Europe (=1 for European countries, 0 otherwise); and (iii) Latin America (= 1 for Latin American countries, 0 otherwise).²⁴ Additional regressors (X) include some control variables to represent internal and external demand for infrastructure, such as per capita income, population, industry and trade. All regressions include country fixed effects (ai).

I introduced an interactive term between Gov and Region to examine the impact of regional governance on regional infrastructure development. Equation (1) then becomes:

The base year for all the variables is 2006, except otherwise mentioned. I included all countries for which data is available for the dependent and independent variables. That left us with a sample of 98 countries, a relatively large dataset for this type of exercise. Moreover, I included all 16 EAS countries and 35 Asia -Pacific members of ADB, to represent the region Asia in this analysis.

Since there are significant and systematic variations in infrastructure development across countries, a satisfactory model should explain substantial heterogeneity at the country level. Using cross-section pooled data can better explain the relevant relationships between regional infrastructure and governance over time when I have both time-variant and timeinvariant regressors; because of its structural nature, it will take a long time before the potential impact of infrastructure development on the economy is realized. The use of crosssection pooled data also has the advantage of better capturing the dynamic relationship between endogenous and exogenous variables, by introducing more variability, less collinearity, more degrees of freedom, and more efficiency. I therefore tested the baseline equation (2) using both cross-section (2006) and cross-section and pooled (1996 and 2006) frameworks.

Given the bivariate associations discussed in the previous section, I have yet to ascertain the functional relationship between endogenous and exogenous variables. To do this, I used both the linear (OLS) and non-linear (ordered probit) models. To check the relative robustness of the model, I replaced the PII with the infrastructure index of the WEF, for the cross-section analysis. I selected Generalized Least Squares (GLS) in Model 2 for two technical reasons: (i) the Hausman test (1978) rejected fixed effect (OLS) and select random effect (GLS), and (ii) GLS provided a higher R-squared, compared to OLS. Estimation results are presented in Table 8 [ PDF 13.7KB | 1 page ] and Table 9 [ PDF 14.5KB | 1 page ].

The following results bear highlighting. First, the linear models with PII as dependent variable are a better fit; both models explain 81-90% of the variation in observations. The basic results of the estimation were as expected—most of the estimated coefficients are statistically significant and robust, with the correct signs and magnitudes. The good fit in both models tell us that good governance positively influences the development of regional infrastructure: every one point improvement in governance lea ds to a two point rise in regional infrastructure in Model 1 (Table 8), and a 0.85 increase in Model 2 (Table 9). At the sample average of the index of governance -0.04, and the value of the coefficient 2.010 (in Model 1, Table 8) and 0.851 (in Model 2 in Table 9) in the baseline regression, the size of the effect with respect to the index of governance would vary between 1 to 2 points.

Second, the significant and positive interaction term (Governance*EU) in both models suggest that, other things being eq ual, membership in the EU is not critical for the development of regional infrastructure; what is important is the introduction of good governance in the region. The results for Latin America are completely opposite, while the results for Asia fall somewhere in between. The results imply that appropriate institutions and policies are required for effective governance and regional infrastructure development. In the case of the EU, they also suggest a degree of regional diffusion—taken together, regional institutions and governance have a direct and positive effect on the local governance of each country in the region, which in turn facilitates regional infrastructure development. However, the negative coefficient of the Latin America dummy (highly significant in both models) implies that membership in regional institutions alone has not helped the development of regional infrastructure in Latin America. It also suggests that the region has yet to significantly improve regional governance.

Third, in both models and for both periods, the results remain largely unchanged even after I replaced the aggregate governance index with its individual components as regressors (Table 10 [ PDF 20.7KB | 2 page ] and Table 11 [ PDF 21.5KB | 2 page ]). The one exception was political stability, which had a negative but statistically insignificant coefficient. Interestingly, it is government effectiveness which had the strongest influence on regional infrastructure: a one point improvement in government effectiveness may lead to a rise of 1.28 to 1.55 points in regional infrastructure, ceteris paribus. The most striking result is the significance of the EU dummy. The Asia dummy appeared with a negative sign, but was not statistically significant. These results suggest that government effectiveness, rule of law, regulatory quality, control of corruption, and voice and accountability are all important determinants of regional infrastructure development.

Tables 9 and 10 show that income levels are highly significant for regional infrastructure development. Population is likewise revealed to be significant. Trade and manufacturing did not come out as significant in the models, although they had the correct signs. This may be the result of income and population (and also their variations) neutralizing the significance of trade and manufacturing in the models. The results also suggest that a country's growth is just as important as governance (both at the national and regional level) for regional infrastructure development.

To conclude, countries (and regions) with higher income, stronger institutions, better governance, and more open economies are likely to have higher levels of regional infrastructure. Indirectly, the estimated results of the baseline models also suggest that that our efforts to promote regional infrastructure must not be limited to traditional policy measures aimed at attracting investment in infrastructure, but must also address policy reform across a number of areas. Thus, institutions and governance must play an important complementary role in strengthening Asia's regional infrastructure.

Robustness checks

The relationships described above cannot be interpreted as causal until I rule out the possibility of endogeneity in equation (2). To address this problem, I used a dynamic GMM estimator (system-GMM) to analyze changes across countries and over time. The estimator also effectively deals with reverse causality by using a set of instruments for the endogenous variables, and includes the lagged dependent variable to account for the persistence of the infrastructure indicator.

Figure 12: Estimation Strategy [ PDF 20.4KB | 1 page ]

One of the main advantages of the system-GMM estimator is that it does not require any external instruments other than the variables already included in our dataset. It uses lagged levels and differences between two periods as instruments for current values of the endogenous variable, along with external instruments. For the infrastructure index and the period 1996 and 2007, for example, the system-GMM method uses the following as instruments: (i) levels of infrastructure for the periods 1995 and 2006 and earlier; and (iii) differences in infrastructure, namely, differences between the periods 1995 and 1996, and 2006 and 2007. More importantly, the estimator does not use lagged levels or differences by itself for the estimation, but rather employs them as instruments to explain variation in infrastructure development. This approach ensures that all information will be used efficiently, and that focus is given to the impact of regressors (such as governance) on infrastructure, and not vice versa.

For this analysis, I started with a relatively simple specification:

where the variables are same as in the previous models, except for two additions in equation (3): (i) Infra_it-1 and Infra_it-2, which represent the lagged dependent variable in the previous period, and (ii) Z_it is a set of instruments for Gov_it and X_it. Here, Gov_it is the variable of interest. X_it denotes the set of control variables, and e_it stands for the error term. Estimating equation (3) by OLS for the typical pooled cross-country time series analysis with “small T and large N” is likely to produce biased coefficients, if the independent variable is endogenous. As a remedy, I followed the procedure suggested by Arellano and Bond (1991) and, as a first step, eliminated the country-specific effects using first differences:

where ?Infra_it = Infra_it – Infra_it-1. As a second step, I estimated equation (4) using system- GMM.²⁵ The system-GMM approach estimates equations (3) and (4) simultaneously, by using lagged levels and lagged differences as instruments.²⁶ I favored the system-GMM estimator, as Infra is very likely to be persistent. Because I used lagged levels and lagged differences, the number of instruments can be quite large in a system-GMM estimator. I used 15 instruments in the analysis. I also report the results of IV regressions (2SLS). To test the appropriateness of the instruments used, I used the Sargan test of over-identifying restrictions in the case of 2SLS, and Hansen J statistics in case of system-GMM in Table 12. The Sargan and J- statistics show that the applied instruments are valid. The estimation results are reported in Table 12 [ PDF 13.9KB | 1 page ].

The following observations are worth mentioning. First, the signs and statistical significance of the coefficients confirm the results obtained for regional infrastructure development in Tables 8 and 9. Moreover, the results obtained in the 2SLS and system-GMM are very similar to each other, except for the size of the coefficients. In the 2SLS, governance (and institutional quality) remains a strong predictor of regional infrastructure development, despite the fact that the magnitu de of its coefficient is lower compared to the estimates obtained in equation (2). On the other hand, the coefficient for governance is much bigger in the system-GMM in equation (4). The results have therefore improved compared to those reported in Table 9, an indication of the general robustness of the relationship between regional infrastructure and governance.

Second, I find substantial improvements in the results for the dummy variable Asia and its corresponding interaction term in equation (4), compared to equation (2). The interaction term in equation (4) yielded the best results in terms of significance and the overall explanatory power of the regressions. The estimated coefficients for Asia and the interaction term are significant at the 5% level in the 2SLS and at the 10% level in the system-GMM, thereby suggesting that (i) national and regional governance must move in parallel in order to have optimal regional infrastructure development in Asia; and (ii) regional governance is perhaps more importan t than national governance, for regional infrastructure to develop in Asia. A one point improvement in regional governance would lead to roughly a two point increase in regional infrastructure in Asia, other things being equal. With the average of the index of governance (-0.04), the size of the effect with respect to the index of governance would vary between 1 to 1.5.²⁷

Third, when the governance index takes on a negative value (in the range -2.5 to +2.5), the interaction term (Gov*Asia) actually becomes negative. This suggests that, in the case of corrupt countries with inefficient governments and weak institutions or governance, regional infrastructure may not facilitate integration with the international market.

Table 14: Marginal Effects of Governance on Regional Infrastructure [ PDF 14.3KB | 1 page ]

Fourth, the size of country-level governance effects on regional infrastructure varies between 0.02 and 6.92 (Table 13 [ PDF 14.3KB | 1 page ]).²⁸ Improvements in regional governance have a greater effect on regional infrastructure development (between 3.64 and 0.19), compared to improvements in national governance (between -0.17 and 3.28) in Asia. In some cases, deficiencies in national governance may be overcome if these are complemented by improved regional governance, ceteris paribus.

Finally, the total average effect of governance, which depends on these complex interactions, can be estimated by calculating the marginal effects. To facilitate the interpretation of the results in Table 12, I computed the marginal effects for the variables of interest.²⁹ Given the underlying equation (3), these marginal effects can be interpreted as variations relative to the mean value at a given income level. In other words, they quantify the observed improvement in regional infrastructure when a country has improved governance, relative to other countries at the same income level. The estimated marginal effects further strengthen our arguments: both national and regional governance facilitates regional infrastructure development.

Download this Paper [ PDF 569KB| 48 pages ].

Comment(s)

There are [0] comment(s) for this entry. Post a comment.