The Does Commercial Diplomacy Overcome Impediments to International Economic Flows? The Case of Australia

The Supplementary Material Appendices SA-SF for this article can be found here.

Total foreign investment in Australia is the sum for each country of: i) direct investment in Australia, equity capital and reinvested earnings; ii) direct investment in Australia, other capital; iii) portfolio investment liabilities, equity securities; iv) portfolio investment liabilities, debt securities; v) financial derivative liabilities; and vi) other investment liabilities. The data are deflated by the GDP deflator and expressed in 2010 prices. Annual inbound real foreign direct investment As above Foreign direct investment in Australia is the sum of i) and ii) above.

Annual inbound real portfolio investment
As above Foreign portfolio liabilities in Australia are the sum of iii) and iv) above.

Appendix SA: Data Sources
Appendix SA documents the data sources for the models of exports and investment. Data for the models were collected for the period from 2010-2015, or for T=6 years. There are 181 countries in the sample. Annual real effective exchange rates

Appendix SB: Characteristics of the Dataset
Appendix SB contains a range of indicators of the characteristics of the dataset. Table SB1 contains descriptive statistics, while Table SB2 presents the correlations of the real exports' variable, the foreign investment variables, commercial diplomacy and the trading/investment partner characteristic variables. Table SB3 summarises the number of countries that register positive foreign investment stocks by the number of years that positive numbers are recorded. Note that the model for foreign investments treats zero investment as being economically determined.

Appendix SC: Model of Exports
Drawing on the work of Rose (2007),1 the gravity model for exports used in this article is expressed in equation (1) as follows: ln(x jt )=β0 + γdip jt + β1 ln (dist j ) + β2 ln (pop jt ) + β3 ln(gdp jt ) + β4 land j + β5 island j + β6 common j + β7 noEnglish j + β8 fta jt + β9 ln (reer jt ) + β10 ln (ecfree jt ) + υ jt . (1) The dependent variable (x jt ) is annual real exports from Australia to country j expressed in terms of natural logarithms as indicated by the operator ln(.). The explanatory variables are as follows: -dip jt is a count variable taking a value equal to the total count of embassies, consulates and trade offices that Australia has in country j. This value is 0 if there are no diplomatic entities in country j; -dist j is the distance between Australia and country j; -pop jt is the population of country j; -gdp jt is the annual real GDP of country j; -land j is a binary variable taking a value of 1 if country j is landlocked and 0 otherwise; -island j is a binary variable taking a value of 1 if country j is an island and 0 otherwise; -common j is a binary variable taking a value of 1 if country j is or has been a Commonwealth member and 0 otherwise; -noEnglish j is a binary variable taking a value of 1 if the official language of country j is not English and 0 otherwise; -fta jt is a binary variable taking a value of 1 if a regional or free trade agreement exists between Australia and country j; -reer jt is the world real effective exchange rate of country j; -ecfree jt is an index of the economic freedom of country j. The m parameters of the model not pertaining to economic diplomacy are denoted by β m . The key coefficient for the variables capturing the incremental effects of economic diplomacy for Australia's real exports is γ.

Diagnostic Tests
The model in equation (1) is specified as a random effects model where the error term of the equation is υ jt = u j + e jt , where the unobserved effect (u j ) is uncorrelated with each of the explanatory variables in all time periods. To test formally that validity of this specification, the Hausman test statistic for the benchmark model that is equation (1) without the diplomatic variable (dip jt ) is 14.54 with a p-value of 0.042, confirming the choice of random effects model over the fixed effects model at the 0.01 level of significance. Similarly, the Hausman test statistic for the model with the diplomatic variable is 17.03 with a p-value of 0.030, again confirming the choice of random effects model over the fixed effects model at 0.01 level of significance. A fixed effects model is not our preferred model, because the commercial diplomacy variables are encompassed in the fixed effects terms, meaning that the effects of the non-time varying variables cannot be separated. The Breusch and Pagan Lagrange multiplier test statistics for a random effects model for the exports model is 1397.56 with a p-value of 0.000, confirming the choice of the random effects model over a pooled, ordinary least squares model.

Appendix SD: Model of Foreign Investment
Estimation of the model of inbound foreign investment is more complicated than the model for merchandise exports because of a large mass of zero observations in the data on foreign investment between Australia and many countries in the sample. The percentage of observations that take zero values is 66 per cent for total foreign investment, 87 per cent for direct investment and 84 per cent for portfolio investment. Further, some countries record foreign investment in some years of the sample but not others. The data on direct investment are available for 35 out of 181 sample countries, with only fifteen countries recording investment stocks every year of the sample period spanning 2010-2015. The situation is similar for portfolio investment where a complete dataset in each year of the sample is only found in thirteen countries. Appendix SB1 contains descriptive statistics on the inbound foreign investment data, and Table SB3 summarises the number of countries with positive foreign investment data.
The mass of zero observations makes the log-based model specification that works for the exports model inaccurate, as we cannot take the log of zero. However, an observation of zero is an economically determined number, so it is not appropriate to deal with the issue by merely removing the zero observations from the sample. To address this issue, we adopt the Silva and Tenreyro Poisson pseudo-maximum-likelihood (PPML) estimator of the gravity model to quantitatively assess the effects of commercial diplomacy and gravity-specific variables on foreign investment flows to and from Australia, as the Poisson estimator is able to account for the zero investment flows in the data.2 The specification of the gravity model for inbound foreign investment is shown in equation (2) as follows: (2) The dependent variable (tfi jt ) is annual real inbound foreign investment from country j to Australia, measured in levels and expressed in 2010 dollars by deflating by the GDP price deflator. The pertinent parameter estimating the effects of diplomacy for foreign investment is δ, with the diplomacy variable (dip jt ) and other explanatory variables defined in Appendix SC.

Diagnostic Tests
The effect of diplomacy on the components of inbound foreign investment, foreign direct investment and portfolio investment liabilities is examined by replacing the (tfi jt ) variable with the foreign direct investment variable (fdi jt ) and portfolio investment liabilities variable (pi jt ) respectively.
We use the Ramsey's regression specification error test (RESET) to test whether our model is right for the investment data. The Poisson regressions all pass the RESET test with a chi-squared statistic of 0.08 and a p-value of 0.782 for the model for inbound foreign investment. The RESET test results confirm that there is no evidence of misspecification of the gravity equations estimated using the PPML estimator, confirming that our model fits the data well.

Appendix SF: Complete Set of Results
The first set of results for the gravity model of Australian exports is contained in Table SF1. Table SF1 reports several specifications of the model, including a model containing no diplomacy variable in column 1, and a model containing the count variable representing commercial diplomacy in column 2. The remaining columns report the robustness of the results to three alternative estimation methods: pooled ordinary least squares (OLS); allowance for a first-order autoregressive error structure (AR error); and estimation using instrumental variables (IV), where all covariates are treated as exogenous. The standard errors are in brackets, and the significance of the variables are indicated by the stars. The coefficients on the commercial diplomacy variable from the alternative estimation methods lie in a narrow range above the estimate of the random effects model. The coefficients of diplomacy for these alternatives are 0.153, 0.142 and 0.123 respectively, and are significant at the 1 per cent level except for the AR error term model, which is significant at the 5 per cent level. Adjusting these coefficients shows that Australian exports are 16.5, 15.0 and 13.0 per cent higher respectively than those in countries without diplomatic representation, placing the contribution of diplomacy using our alternative estimation methods quite in line with our preferred model reported in the above article and in column 2 of Table SF1 below. Table SF2 contains the parameter estimates for the model of commercial diplomacy for the log of real foreign investment for the three categories of total, direct and portfolio investment, while Tables F3 and F4 contain the parameter estimates for the model of commercial diplomacy for the log of real exports and for real foreign investment for alternative country types. These tables are discussed in the text.
Not all countries have data for all six years. For example, data for the real effective exchange rate have 1,020 observations and data for the variable of importer GDP have 1,066 observations. By including these two variables in the model, the number of observations (N) of the model is N=1000, with the number of clusters 169. If we had data on all variables in all years, the number of observations would be N=1086 and the number of clusters would be 181.   Table SF1 Parameter estimates for the model of commercial diplomacy (cont.) Note: Standard errors are in parentheses; the asterisks represent the 1% (*** p<0.01), 5% (** p<0.05) and 10% (* p<0.1) level of significance. Note: Standard errors are in parentheses; the asterisks represent the 1% (*** p<0.01), 5% (** p<0.05) and 10% (* p<0.1) level of significance.    Table SF4 Parameter estimates for the model of commercial diplomacy (cont.) Note: Standard errors are in parentheses; the asterisks represent the 1% (*** p<0.01), 5% (** p<0.05) and 10% (* p<0.1) level of significance.