Estimation results

For this paper we estimate three basic models: a trade equation (results summarized in Table 2 [ PDF 72.5KB | 2 page(s) ] and Table 3 [ PDF 69.1KB | 2 page(s) ]), a FDI equation (results summarized in the first two columns of Table 4 [ PDF 75.9KB | 3 page(s) ]), and a cross-border road infrastructure equation (results summarized in the last four columns of Table 4 [ PDF 75.9KB | 3 page(s) ]). The trade equation was estimated using two alternative definitions of trade: one based on major exports transported via land or river, and the other based on total bilateral trade as reported in the IMF Direction of Trade Statistics database. Our preferred estimation procedure is the random effects estimator for panel data. However, we also estimate trade and FDI equations using single years of data on country pairs to gain additional insight as to how the cross-sectional variation in our estimation models evolved over time.

Table 2 [ PDF 72.5KB | 2 page(s) ] presents results of estimates of the value of major exports between GMS countries. Up to 5 commodities (defined at the 4 digit level in the UN Harmonized System of Product Categories) per country pair were selected and summed to generate this measure of trade. The selection of products relied on available (admittedly sketchy) information from customs data for these countries that details or suggests the commodities and goods that are most likely to be transported by road and ferry—where bridges are not available across rivers. Use of disaggregate commodity-specific trade data is preferred to aggregate trade because a larger variety of factors besides cross-border road infrastructure are expected to influence aggregate trade. However, the downside of using the ‘major exports’ is data scarcity and unavoidable subjectivity in the selection of major commodities due to unreliability of customs data at overland points of entry.

Table 2 [ PDF 72.5KB | 2 page(s) ] reports results of five estimation models using the major exports variable. The overall goodness of fit of the models is good, with estimated R2 measures ranging between 35.6 percent (Model 1) and 76.2 percent (Model 5). All five models are highly statistically significant, as indicated by the results of the Wald Chi-square test—which reject the null hypothesis that there is no systematic statistical relationship between the models and major exports at a 99 percent confidence level. However, limits on the estimations that use the major exports variables were rather severe due to the relatively small sample size available across GMS countries. These limits made it difficult for panel data models to be estimated and prevented estimation of models that include some variables of interest, so instead, we reverted to a simpler regular Ordinary Least Squares (OLS) regression in Models 3, 4, and 5 reported in Table 2 [ PDF 72.5KB | 2 page(s) ]. However, use of the OLS estimator is not supported by our results from the Breusch-Pagan Lagrange Multiplier test. In addition to this specification test, the general sensitivity of these models’ coefficient estimates to changes in the number of right hand side variables that are included suggests that the results of models 3 through 5 are non-robust. Along with the results of the Breusch and Pagan test, this further suggests caution is warranted in interpreting the results of Models 3 to 5.

Models 1 and 2 are estimated as random effects panel regressions, and yield coefficient estimates for the basic variables of the gravity model (i.e. GDP, population, and area) that accord with our expectations and with the results generally obtained in gravity model estimates.⁹ A notable exception to the consistency of our results with previous estimates is the non-significant effect that distance is estimated to have on major export flows. This suggests that the distance between capitals may be a poor indicator of the relevant distance in determining overland trade flows between GMS countries, which is understandable since overland trade tends to focus on markets besides the capital city (e.g., regional markets closer to border areas). Unfortunately, limitations of the sample size available prevented estimation of Model 1 in panel form when key variables of interest in addition to the base variables of the gravity equation are added (i.e. the cross-border road measure, an indicator of domestic road infrastructure, and the FDI and tariff measures).

Model 2 includes the cross-border infrastructure variables but only the GDP variable from the base variables of the gravity model. Although not detailed in the table, the variation in trade levels observed for pairs of GMS countries was explained largely by changes in the level of trade between countries over time (as opposed to cross-sectional variation across country-pairs).¹⁰ A key finding from our estimation Model 2 is that intra-GMS trade via land in major commodities has an elasticity of between 0.42 and 0.46 with respect to cross-border road infrastructure on both sides of the border; which implies that a doubling of the density of roads in border provinces or regions would be expected to induce an average increase in trade in major exports of over 40 percent across the GMS countries. However, when we add a variable measuring domestic road infrastructure to our random effects panel estimates, the statistical significance of cross-border road infrastructure no longer holds, although both variables maintain their positive coefficients. The overall conclusion we reach from the two panel estimates reported in Table 2 [ PDF 72.5KB | 2 page(s) ] (Models 1 and 2) is that trade in major commodities within the GMS is positively influenced by the level of cross-border infrastructure, and that such trade flows are largely driven by economic size of the countries involved and to a lesser but still significant extent by cross-border road infrastructure.

To explore the marginal impact of cross-border road infrastructure in addition to the effect of domestic road infrastructure on major exports, we also estimated Models 3 through 5 (also summarized in Table 2 [ PDF 72.5KB | 2 page(s) ]). In these models we find that cross-border road infrastructure has an even larger positive and statistically significant association with trade in major exports than that found in our panel estimate (Model 2). Domestic road infrastructure is found to have a negative and statistically significant effect on trade in major exports. One interpretation of this result is that domestic road infrastructure—when separated from roads in frontier areas—mainly promotes the integration of domestic markets within GMS countries and diverts economic activities away from trade in major commodities across GMS countries. Another interpretation is that domestic road infrastructure in GMS complements other infrastructure necessary for ocean-bound trade but not land-bound trade. However, additional information and study is required to assess the validity of this interpretation with confidence. Another coefficient estimate worth noting is the positive and statistically significant effect that importer tariff rates are found to have on major exports, which runs counter to expectations.

Table 3 [ PDF 69.1KB | 2 page(s) ] presents estimation results on total exports between GMS countries. Because of the greater number of observations of total exports (rather than major exports via land), we are able to estimate all these models using the preferred random effects panel estimator. Use of this estimator is supported by the highly statistically significant results of the Wald Chi-square tests and the results of the Breusch-Pagan Lagrangean Multiplier tests. The models explain between 42.3 and 57.4 percent of the observed variation in aggregate exports between GMS countries. The six variants reported in Table 3 [ PDF 69.1KB | 2 page(s) ] have results that are largely consistent with our expectations and published gravity model results (e.g., negative association between distance and export levels, and the positive association between trading partners’ economic and geographic sizes and their levels of trade). As in earlier studies, the association between trading partner population and total exports is generally negative, although in the majority of cases the association is not statistically significant.

The sound performance of Model 6, which includes only the base variables of our gravity model and, the consistency of base variable coefficient estimates across the 6 models reported in Table 3 [ PDF 69.1KB | 2 page(s) ] suggests that the basic gravity model provides a strong base upon which the effect of other variables of interest for trade levels can be usefully judged. Of our particular interest in Table 3 [ PDF 69.1KB | 2 page(s) ] are the estimated coefficients for cross-border and domestic roads, indicators of trade policy and trade environment, and FDI inflows. Models 8 and 9 include an indicator of the trading partners’ cross-border road infrastructure. Such roads have a positive but not statistically significant effect on total trade in Model 8, and have a positive and statistically significant effect on exporter’s total trade in Model 9—which also includes a measure of domestic road infrastructure that also has a positive and statistically significant association with total exports of the exporting economy. This provides limited evidence that cross-border roads favorably influence total exports, although the relationship is clearly weaker than was the case for selected major exports via land. Model 9 also indicates that cross-border and domestic road infrastructure play a complementary role to each other with respect to enhancing aggregate exports among GMS countries, which is contrary to the result we reported in Table 2 [ PDF 72.5KB | 2 page(s) ] in terms of selected major exports via land.

Models 7 and 10 in Table 3 [ PDF 69.1KB | 2 page(s) ] show that the average tariff rate has a negative association with total exports, although the association is statistically significant only for the exporting country (while one would typically expect the importing economy’s tariffs to have a greater effect on bilateral trade). This result may be obtained either because tariff barriers are the lesser obstacles to trade than quantitative restrictions and other non-tariff barriers, or because the weighted average tariff rates automatically include all kinds of exemptions as well as “missed” collections by customs authorities, and therefore, understate official tariff rates. In Model 10, our export and import environment dummy variables had signs contrary to our expectations, but neither were statistically significant. This could be because these variables are represented by the extent of administrative time taken by exporters and importers and may have left out other important informal barriers to trade. FDI inflows has no statistically significant association with trade flows, indicating FDI flows may be independent of aggregate trade flows. Lastly, relative real prices across the trading economies—measured by the ratio of the purchasing power parity conversion factor and the official exchange rate—has strong effects on trade with the expected signs.

The first two columns of Table 4 [ PDF 75.9KB | 3 page(s) ] present estimation results for FDI inflows. Most of the coefficients show expected signs with statistical significance: e.g., positive association with economic size of the receiving (importer) country; negative association with population size of the exporter country (the larger the economy, the less impetus to invest abroad); and positive association with FDI environment. Both models are statistically significant overall and explained 60 and 81 percent of the variation in FDI, respectively. The finding that the larger the GDP (and land area) of FDI importer, the higher the level of FDI likely reflects a PRC effect. In model 2, FDI is associated positively with the cross-border infrastructure of the receiving (importer) country but negatively with that of the sending (exporter) country. This may suggest that countries develop cross-border infrastructure in order to entice FDI or that such infrastructure is developed as a condition of FDI. FDI flows are positively and significantly associated with the domestic road infrastructure of the sending country but negatively with that of the receiving country, which is consistent with expectations that capital tends to flow from richer to poorer countries within the GMS and that richer countries tend to have more developed road infrastructures.

The last four columns in Table 4 [ PDF 75.9KB | 3 page(s) ] present estimation results on cross-border road infrastructure. By these, we intended to test whether the development of road infrastructure is itself an outcome of the level of trade, FDI and the other standard variables included in a gravity model. Cross-border road infrastructure appears influenced positively by the country’s economic size, both on exporter’s and importer’s sides, presumably due to greater fiscal capacity of larger economies in investing in roads. Similarly cross-border road infrastructure is largely influenced positively by population size. On the other hand, it is largely influenced negatively by the land area, presumably due to the greater difficulty of spreading road network in geographically larger areas. Combined with our results in Table 2 [ PDF 72.5KB | 2 page(s) ], economic and population sizes seem to be the dominant drivers of both trade levels and investment levels in road infrastructure, while cross-border road infrastructure has some identifiable influence on trade levels. It is not clear whether and in which directions the cross-border road infrastructure is associated with FDI inflows.

Table 5 [ PDF 72.4KB | 2 page(s) ] summarizes estimation results on total exports in individual years. Our main motive was to investigate stability and trend over time of the relationship between trade level and standard explanatory variables in a gravity model. The associations with distance (negative), economic size (positive), and land area (positive) are fairly stable and consistent with expectations. However, the association with population size is unstable over time, which has been found in previous gravity model studies and in this instance may particularly reflect the massive changes in the People’s Republic of China’s economic relationship with the other GMS countries over time.

Table 6 [ PDF 77.8KB | 2 page(s) ] summarizes estimation results on FDI inflows in individual years. One interesting result in the table is the positive and fairly stable association between distance between trading partners and FDI flows. This may suggest that greater distance spurs businesses to move closer to the markets assuming that home-market-oriented FDI is dominant between GMS members – contrary to the production-integration-oriented FDIs that are increasing among the firms in advanced economies. This interpretation is consistent with FDI’s positive and stable association with economic size of the receiving (importer) economy.

Download this Discussion Paper [ PDF 309.9KB| 35 pages ].

Comment(s)

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