4.1. Model Specification
To examine empirically the impact of multilateral trade liberalization on resource revenue, we draw on the existing literature on the determinants of public revenue (e.g.,
Ghura 1998;
Khattry and Rao 2002;
Ebrill et al. 1999;
Agbeyegbe et al. 2006;
Brun et al. 2007;
Baunsgaard and Keen 2010;
Thomas and Treviño 2013;
Crivelli and Gupta 2014;
Brun et al. 2015;
Von Haldenwang and Ivanyna 2017). This literature has identified a number of structural factors that influence public revenue. These structural factors (which are control variables used in the current analysis) include the level of development, proxied by countries’ real per capita income; the level of trade policy liberalization, demographic characteristics of countries, measured for example by the population size. We have additionally included the inflation rate, which has appeared virtually in almost all studies cited above, and the oil prices, which acts as a proxy for natural resource prices. We also control for the non-resource revenue variable in the model so as to ensure that the impact of multilateral trade liberalization on resource revenue is a direct one, that is, it does not pass through non-resource revenue. We explain below how each of these factors could influence resource revenue.
Against this background, we postulate the following model:
where
i is the subscript associated with a given country;
t denotes the time-period. The model is estimated using a panel dataset comprising 57 countries, including both developed and developing countries, over seven sub-periods of non-overlapping 3-year average data covering the annual period 1995–2015. These sub-periods include 1995–1997; 1998–2000; 2001–2003; 2004–2006; 2007–2009; 2010–2012 and 2013–2015. The choice of countries and the time period are dictated by data availability. It is worth noting that some variables have been transformed into Logarithm (
Log), while others have not. Indeed, we transform into Log all variables that exhibit a high skewness.
to
are parameters to be estimated.
are country-specific effects. The disturbance term
is assumed to be independently and identically distributed.
Table A1 provides the definition and source of the variables used in model (1), while
Table A2 reports the list of countries contained in the full sample.
Table A3 presents descriptive statistics on these variables.
“RESREV” represents the share (%) of resource revenue in GDP. As this variable contains yet positive values, but also many zeros, we have transformed it into a Logarithm to limit its skewness. Specifically, as using the natural Logarithm would not be appropriate here, we use the Logarithm transformation method suggested by
Yeyati et al. (
2007), which goes as follows:
(2), where “X” denotes the variable that is being transformed and “Y” is the variable outcome of the transformation. This method allows for retaining information related to zero observations (see, for example, also
Dabla-Norris et al. 2015 and
Morrissey et al. 2016 who have used this technique in their respective study).
“NONRESREV” stands for the share (%) of non-resource revenue in GDP. It is the difference between total public revenue and resource revenue, both expressed in % GDP. We expect an increase in non-resource revenue to be negatively associated with the resource revenue share.
“DTP” and “MTP” are respectively the index of domestic trade policy and the index of multilateral trade policy discussed in
Section 3. The literature has demonstrated that domestic trade policy liberalization can exert either a positive or a negative effect on public revenue.
Ebrill et al. (
1999) and
Agbeyegbe et al. (
2006) have reported that the impact of trade openness or trade liberalization on public revenue depends on several factors, including the structure of trade liberalization and the effect of the latter on each component of public revenue. As far as the impact of domestic trade policy on resource revenue is concerned, we argue that trade policy liberalization could attract resource-seeking FDI inflows that intend to exploit natural resources, with a view to exporting them to the home countries (these natural resources may be used as inputs in the production process of final goods in the home countries). In this case, trade policy liberalization would induce higher resource revenue. MNEs could also intend to process natural resources in the host country where they have established their plants, in order to export the final products in the international trade market. This may in turn result in higher resource revenue if the government of the host country does not exempt MNEs from paying taxes on the exploitation of these resources. At the same time, the host-country’s government can decide to exempt resource-seeking MNEs that intend to exploit and process the natural resource in the host country from paying taxes on the exploitation of the natural resources if it intends to collect higher non-resource revenue. This non-resource revenue would flow from the processed natural resources that are either sold in the domestic market and/or exported in foreign markets. The literature on the determinants of public revenue has established that a higher value added in the manufacturing products (in % of total output) generates higher non-resource revenue (e.g.,
Thomas and Treviño 2013;
Brun et al. 2015;
Gnangnon and Brun 2017). In this context, trade policy liberalization would not result in higher resource revenue, but will likely generate higher non-resource revenue. Overall, it would be difficult to anticipate the (average) impact of trade policy liberalization on resource revenue, as this impact can be either positive or negative.
Brief discussion on the expect impact of control variables on resource revenue
“GDPC” is the real per capita income, a proxy for countries’ overall development level. We expect that countries with higher development levels would be better equipped, including in terms of skills and institutional sophistication (for example, strong capacity of the tax administration to collect public revenue) to negotiate better contracts with resource-seeking MNEs. Thus, advanced economies would be able to collect higher resource revenue from the exploitation of natural resources by MNEs than less advanced economies. Therefore, we expect a rise in the real per capita income to be associated with higher resource revenue.
“INF” stands for the inflation rate. We transform this variable into a Logarithm using formula (2) as it contains both negative and positive values. Such transformation also helps reduce the skewness of this variable. According to
Tanzi (
1977), in an inflationary environment, lags in tax payments reduce by the inflation rate, the real amount of tax paid (this effect is further enhanced if the tax system is not protected from inflation). Many studies (among those highlighted above) have reported a negative impact of inflation on public revenue. As far as resource revenue is concerned, we expect that, in an inflationary environment, resource revenue would be adversely (negatively) affected. However, if the government’s contracts with MNEs to exploit natural resources are indexed to inflation, the latter would not affect resource revenue.
“POP” is the size of the total population. The size of the population could influence resource revenue through its possible impact on FDI inflows. MNEs that are interested in exploiting natural resources with a view to adding value to these resources and serving the host-country’s domestic market (and eventually export abroad, including to the regional markets) could be motivated to set up their plants in host countries that have a high market size, i.e., an important size of the population
4. The rise in the size of the population could result in higher FDI inflows and generate higher resource revenue for the host-country’s government, if the latter does not fully exempt MNEs from the payment of taxes on the exploitation of natural resources. However, the host-country’s government could decide to fully exempt MNEs from paying resource revenue (or the government could opt for making MNEs pay miniscule resource revenue) on the exploitation of the natural resources. Such a decision by the government could be based on the expectation that it would collect higher non-resource revenue from the higher value-added products that MNEs would sell in the domestic market, and possibly export. It is important to recall that exports of non-resource products could also be an important source of non-resource revenue as they could generate higher jobs in the tradable sector as well as higher firms’ profits. These jobs and the rise in firms’ profits would lead to higher non-resource revenue through the channels described in
Section 2. In light of all these, it is difficult to anticipate the impact of the size of the population on resource revenue, as this impact could be either positive or negative. The empirical analysis would provide further guidance.
“OILPR” represents the oil prices (deflated by the US consumer price index). It is a proxy for natural resource prices. We expect a rise in natural resource prices to be associated with higher resource revenue.
“INST” represents the indicator (a synthetic measure) of the institutional and governance quality. The importance of institutional and governance quality for public revenue performance has been emphasized in the empirical literature. For example,
Ghura (
1998) and
Bird et al. (
2008) have shown empirical evidence that good institutional and governance quality is associated with higher public revenue performance. Therefore, we expect that countries with strong institutional and governance quality would be able to negotiate better contracts on the exploitation of natural resources, and hence collect higher resource revenue. However, as resource revenue is not a sustainable source of public revenue, such countries may opt for diversifying, over the medium to long term, the sources of their public revenue away from resource revenue toward non-resource revenue. We can expect in this context that the better the institutional and governance quality, the lower is the resource revenue share (probably at a higher non-resource revenue share). This argument is particularly relevant because countries with weak institutional and governance quality usually experience difficulties in collecting non-resource tax revenue, and their alternative source of public revenue would therefore be resource revenue. As it could be observed in
Table A1, the synthetic measure of institutional and governance quality is computed by means of the factor analysis (notably the Principal Component Analysis). In particular, we use the first principal components of six indicators of governance (for details, see
Table A1) (e.g.,
Globerman and Shapiro 2002 and Buchanan et al. 2012). The Principal Component Analysis approach generates linear combinations of object measures (called eigenvectors), which show the greatest statistical variance over all of the objects under consideration (six indicators of institutional and governance quality). Higher values of the “INST” indicator represent better institutional and governance quality.
4.2. Estimation Strategy
To start with, we estimate by means of the within fixed effects estimator a static version of model (1), i.e., model (1) without the one-period lag of the dependent variable as a regressor. In particular, we correct the estimates’ standard errors using the Driscoll and
Kraay (
1998) technique, which helps take into account the presence of cross-sectional dependence and heteroscedasticity in the error term. We denote “FE-DK” the within fixed effects estimator. The estimates arising from the use of the FE-DK estimator could be biased due to several endogeneity issues, including the reverse causality problem associated with many regressors, as well as the absence of the one-period lag of the dependent variable as a regressor (omitted variable problem). Therefore, following the above-mentioned existing studies on the macroeconomic determinants of public revenue, we also estimate different variants of the dynamic model (1) by means of the two-step system GMM estimator proposed by
Arellano and Bond (
1991) and
Blundell and Bond (
1998). The estimation based on the GMM system approach involves a system of equations, i.e., an equation in differences with an equation in levels where lagged first differences are used as instruments for the levels equation, and lagged levels are used as instruments for the first-difference equation. Compared to the first-difference GMM estimator suggested by
Arellano and Bond (
1991), the two-step system GMM estimator performs better when cross-sectional variability dominates time variability and when there is a strong persistence in the time series under investigation (
Blundell and Bond 1998). In addition, when the panel dataset is unbalanced, the difference GMM estimator has a weakness of magnifying gaps (see
Roodman 2009). This estimator has the advantage of addressing the endogeneity problem that could stem from the presence of the one-period lag of the dependent variable as a regressor (this problem is referred to as the Nickell bias, see
Nickell 1981), as well as other endogeneity issues that could arise from the estimation of model (1). Specifically, the variables “NONRESREV”, “GDPC”, and “DTP” could be considered as potentially endogenous, due to the eventual reverse causality from the dependent variable to each of these variables. Therefore, in the regressions, we consider all these three variables as endogenous. Standard errors of the estimates obtained by means of the two-step system GMM estimator are corrected using the
Windmeijer (
2005) finite-sample correction.
We check the validity of the two-step system GMM estimator by relying on three diagnostic tests: the Sargan test of over-identifying restrictions, which helps check the validity of the internal instruments used in the regressions; the Arellano–Bond (AB) tests of first-and second-order serial correlation, respectively denoted AR(1) and AR(2) (the null hypothesis of absence of second-order serial correlation in the disturbances should not be rejected, while the null hypothesis of absence of first-order serial correlation should be rejected). Incidentally, we report the AB test of the third-order serial correlation in the error term, denoted AR(3), to show the lack of autocorrelation at the third order in the error term. We also report the number of instruments used in the regressions, as the above-mentioned diagnostics tests may lose power if the number of instruments is higher than the number of countries (see
Roodman 2009).
In a nutshell, the empirical analysis proceeds in several steps. First, we estimate a static specification of model (1) by means of the FE-DK estimator, and the estimation’s results are presented in column [1] of
Table 1. We then estimate model (1) as it stands, i.e., the dynamic specification of model (1) by means of the two-step system GMM approach. The results of the estimation of this model are reported in column [2] of
Table 1. Second, we estimate another specification of model (1), which allows examining whether the impact of multilateral trade liberalization on resource revenue is the same in the poorest countries and the non-poorest countries in the full sample. We consider the least developed countries (LDCs) as the poorest countries. LDCs
5 constitute a group of countries designated as such by the United Nations, which consider them as the poorest and most vulnerable countries in the world. To perform the analysis, we define a dummy variable “LDC”, which takes the value 1 when a country belongs to the category of LDCs, and 0, otherwise. The list of the LDCs used in the analysis is displayed in
Table A2. This LDC dummy variable is subsequently interacted with the “MTP” variable, and both the dummy variable and its interaction with the “MTP” variable are introduced in model (1). We expect that, as the poorest countries mostly rely on raw materials (including natural resources) for production and exports, and do not have a strong trade capacity
6, they might not be able to take full advantage of the opportunities offered by multilateral trade liberalization by diversifying their production away from natural resources. In this context, multilateral trade liberalization (i.e., greater access to the international trade market) might increase their specialization in the production and export of natural resources, which would generate higher resource revenue to their governments. The results of the estimation of this specification of model (1) are displayed in
Table 2. Before moving to the interpretation of the empirical results, we compare in
Figure 1 the evolution of the variables “RESREV”, “NONRESREV” and “MTP”, using the non-overlapping 3-year average dataset. It is important to note that although the non-resource revenue is not a key variable of interest in the current analysis, its evolution has been presented in
Figure 1 with a view to comparing it with that of resource revenue.
It could be observed from
Figure 1 that multilateral trade liberalization has exhibited an upward trend over the period, thereby suggesting that over time, the world has witnessed greater trade liberalization, albeit steadily. In the meantime, after a decline from 10.3% in 1995–1997 to 8.4% in 1998–2000, resource revenue has steadily increased up to 14% in 2007–2009. It has then shown a declining trend to reach 10.2% in 2013–2015. Non-resource revenue has always moved in the opposite direction to resource revenue. It has remained relatively stable around 14.2% to 14.9%, from 1995–1997 to 2010–2012, and has increased from 14.6% in 2010–2012 to 16.3% in 2013–2015.