Revisiting national border effects in foreign trade in goods of Canadian provinces

ABSTRACT

A significant body of empirical studies demonstrates sizable national border effects in foreign trade of Canadian provinces throughout the 1980s and 1990s. This paper revisits and expands the scope of the border effects analysis by estimating the border effect in trade with US states as well as countries in the European Union (EU) and the G-20 using more recent data from 2001–10. Furthermore, we perform the Blinder–Oaxaca nonlinear decomposition to decompose the border effects into various components, including the transaction costs, the tariff and non-tariff measures, and the unexplained component. Results from the Poisson pseudo-maximum likelihood model show that, compared to existing estimates from the 1980s and 1990s, the size of the border effect in trade between Canadian provinces and US states has declined. The border effects for Canada–EU and Canada–G-20 bilateral trade flows sit at somewhat elevated levels. About a third of the border effects in overall trade with EU and G-20 countries can be attributed to the variables related to transaction costs in foreign trade. While the significance of tariffs has declined, the prevalence of non-tariff measures seems to be on a rise. That said, we find that the welfare-changing measures combined – tariff and non-tariff measures – play a limited role in explaining the border effects in comparison with the role of transaction costs and the unexplained component.

Keywords:

• gravity model
• border effects
• Canada-USA trade

JEL CLASSIFICATIONS:

• F14
• F1

Acknowledgement

I would like to thank Alexander Fritsche, Gregory Bauer, Laurent Martin at the Bank of Canada and conference participants at the 49th Annual Conference of the Canadian Economics Association.

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1. Based on 1988 data.

2. Obstfeld and Rogoff (2000) viewed McCallum's findings as one of the major puzzles in international macroeconomics.

3. Group of countries in G-20 in this paper excludes France, Germany, Italy and the United Kingdom because they are included in the group of EU countries. The United States is also excluded from the G-20 group. The list of EU and G-20 member countries included is provided in Table A1. In contrast with the most of the existing literature that focus on the 30 largest (and/or adjacent to Canada) US states, our study extends the analysis to all US states.

4. The states are Alabama, Arizona, California, Florida, Georgia, Idaho, Illinois, Indiana, Kentucky, Louisiana, Maine, Maryland, Massachusetts, Michigan, Minnesota, Missouri, Montana, New Hampshire, New Jersey, New York, North Carolina, North Dakota, Ohio, Pennsylvania, Tennessee, Texas, Vermont, Virginia, Washington and Wisconsin.

5. Theoretical derivation of trade costs and border effects in bilateral trade does not exclusively hinge on Andersen and van Wincoop's (2003) model. Eaton and Kortum (2002) developed a Ricardian-type model emphasising the supply side whereas Melitz (2003), Chaney (2008), and Melitz and Ottaviano (2008) established a stream of trade literature that accommodated the role of border effects through the model of heterogeneity of firms in productivity and trade. Both streams of literature conclude that, all things held constant, a rise in border related costs in international trade reduces bilateral trade flows.

6. The list of most populous is in Table A1.

7. Andersen and van Wincoop (2003) incorporate the remoteness terms both for exporter and importer. However, due to a strong correlation between the two terms in our data we dropped the exporter's remoteness term to avoid multicollinearity. The choice of the remoteness term turns out to be irrelevant in practice in our data. Experiments with the exporter's remoteness term give, by and large, the same results for the border effects.

8. Zero trade flows are quite common in international trade. This creates a problem when estimating log-linear gravity equations. Ignoring zero bilateral trade flows or replacing zeros with a small positive number creates a bias in the parameter estimates as noted earlier. The models proposed by Tobin (1958), Heckman (1979) and Helpman et al. (2008) were used to tackle the problem. The Tobit model approach was scrutinized, most prominently by Linders and de Groot (2006) who argued that hypothetical trade cannot be negative. As a result, trade flows cannot be censored from below zero. Sample selection models by Heckman (1979) and Helpman et al. (2008) have been criticized on two grounds – complexity of satisfying the exclusion restriction imposed by two-stage models and lack of controls against heteroscedasticity, which tends to be present in international trade data (Santos and Tenreyro 2009; and Flam and Nordstrom 2011).

9. Santos and Tenreyro (2006) argue that, in estimating equation (3)(3) , the issue of endogeneity may also occur if the errors in the original model are heteroscedastic. This is due to the fact that the expected value of the logarithm of a random variable depends on high-order moments of its distribution, which makes OLS or two-stage (including Heckman sample selection) models inefficient.

10. The problem of endogeneity may arise in augmented gravity models when estimating the impact of variables related to trade policies (Baier and Bergstand 2007). Countries are likely to introduce favourable trade policies for partners with which they already trade frequently. If this is the case, Tar_ijt, NTB_ijt, WTO_it, and WTO_jt in equation (7)(7) may be correlated with the error term  ϵ_ijt. To explore whether endogeneity of trade policy variables is a problem in our model, we re-estimate equation (6)(6) and equation (7)(7) using one-period lagged levels of potentially endogenous variables (i.e., Tar_{ijt − 1}, NTM_{ijt − 1}, WTO_{it − 1}, and WTO_{jt − 1}) as instruments in the Poisson IV model. While the endogeneity may be also sourced in other explanatory variables, such as GDP, its effect on a bias of parameter estimates in the gravity model has been estimated to be limited (Cyrus 2002). Next, we use the Wald test to compare estimated coefficients from the Poisson IV regressions with the original estimates from equation (6)(6) and equation (7)(7) . Under the hypothesis of no endogeneity, the estimates are not expected to differ significantly. The technique has been frequently applied in the literature on endogeneity in panel data. The detailed treatment of this technique may be found in Holtz-Eakin, Newey, and Rosen (1988), Angrist and Krueger (2001), Kristensen and Wawro (2003), and Cameron and Trivedi (2008).

11. http://data.worldbank.org/indicator/NY.GDP.MKTP.CD

12. In this paper, we do not attempt to explain why certain provinces' home biases are more or less pronounced. That said, we are planning to discuss the issue of heterogeneity of the provincial border effects in a follow-up study.

13. We have also estimated equation (6)(6) and equation (7)(7) using one-period lagged values of potentially endogenous variables Tar_ijt, NTM_ijt, WTO_it, and WTO_jt as instruments in the Poisson IV model. Results from the Wald test indicate that the differences in estimated coefficients for trade policy variables in the Poisson IV model and the original equation (6)(6) and equation (7)(7) are not statistically significant at 5 per cent significance level. The results are available upon request.

14. Detailed description of the strategy can be found in Seizing Global Advantage: A Global Commerce Strategy for Securing Canada's Growth & Prosperity. Ottawa. Public Works and Government Services Canada. Government of Canada. 2009, available at http://www.international.gc.ca/commerce/assets/pdfs/gcs-en.pdf.

15. According to Grundke and Moser (2014), some of the NTMs imposed by the United States recently are consistent with the existence of counter-cyclical protectionism. Some of the recent examples of protectionist regulations in the United States include imposing mandatory labelling of the country of origin on beef; inspecting imported catfish; expanding the Lacey Act (2008) to include lumber and other forestry products; banning clove cigarettes; and restricting the shrimp-tuna trade.