Intra-national versus International trade in North America: why do national borders matter?

Intra-national versus international trade in North America: why do national borders matter? SAHIL MANEKIA School of International and Public Affairs, Columbia University, 420 West 118th St, New York, NY 10027 Abstract The objective of this paper is to observe border effects among the countries of NAFTA, the United States, Mexico and Canada. In this context, we adopt the specification of the gravity model and distance measures in the established literature. We find effects in the direction we expected based on previous studies of the subject. We find our estimates for the border effect biased as we were not able to account for relative prices, and therefore not wholly reliable. Introduction and Literature Review Internal trade flows (within nations) have long been known to dwarf international trade flows. McCallum (1995) notes that trade flows between two provinces in Canada were about 22 times as large as their trade with US states after controlling for a number of explanatory factors. Chen (2004) studies these effects in the EU and finds that domestic trade volumes are typically 6 times as large as international flows. Anderson and Wincoop (2001) provide an extensive and useful overview of trade costs and borders as barriers to trade. Their findings suggest trade costs matter (they are large), are linked to economic policies, have big welfare implications, and will help resolve major problems in economics. Chen (2004) looks at the specification of the gravity model and different measures of internal distances in her analysis of border effects in the EU. She finds that both significantly alter the coefficient measuring the importance of domestic trade, but are not out of line with the established pattern. She looks at the causes of border effects, and finds a compelling case for the ‘agglomeration thesis’ advanced by (Wolf,1997; Hilberry, 1999; Hilberry and Hummels, 2000), that intermediate and final goods producers agglomerate to avoid trade costs, producing a border effect. This finding suggests that welfare gains from reducing border effects would be quite minimal. Theoretical Model This paper will try to replicate results obtained in Chen for trade between the United States, Canada and Mexico. These countries are less integrated than the countries of Europe, despite the passing of NAFTA. Chen uses data for seven countries and 78 industries in 1996 to look at how bilateral trade is impacted by ‘border effects.’ This paper will follow her methodology, adopting her specification of the gravity model and distance measures. Her paper then runs producer and destination fixed effects to control for relative prices, but that will not be attempted here. ‘Bilateral Exports’ is the dependent variable in our regression. This is defined as the value of exports from one country to another in a common currency. Where trade is internal, Bilateral Exports includes the value of domestic production less the value of all exports. This allows us to compare exports to domestic consumption in a simple OLS model. ‘home’ is a dummy variable equal to 1 if trade is internal (domestic consumption) and equal to 0 if not. The coefficient on home will allow us to predict the border effect and is the main variable of interest. It is expected to be positive, indicating a preference for trading internally. ‘GDP’ is gross domestic product, a measure of economic output. In our model, GDP will always be that of the importing country, unless trade is internal. It is supposed to show a positive relationship with bilateral exports. Distance is a measure of the distance between the economic centres of two countries (in this case their capitals). As countries get further from each other, trade is expected to fall of. Hence the sign here should be negative. For internal trade, there are two different measures of distance that we will employ in this paper. The first, suggested by Wei (1996) is to take a quarter of the distance to the economic centre of the nearest trading partner. This is given by the ‘great circle formula’, and Chen includes helpful guidance in her appendix that allows us to calculate the distance measures for our three countries. The second measure is Leamer (1997) who uses the radius of a circle (the area of which is equal to the area of the country. We will look at a variable for adjacency to see how having a common border might affect trade (adjacent countries are expected to trade more, positive sign), as well a variable for domestic production to see what relationship that might have with trade. We will use the gravity model specified in Chen as follows: lnXij,k = β0 + β1home + β2 lnYi,k + β3ln Yj + β4adjij + β5 ln Dij + εij,k Data and Methodology An explanation of each symbol, its units, the number of observations recorded, their frequency and the year for which they were obtained are recorded in the Table A below. Table A. A Table of the symbols Name lnXij,k Log, Bilateral export flows for exporter (i) to importer (j) in industry k Log, Domestic production of exporter (i) in industry (k) Log, Gross Domestic Product of importer (i) Units Percentages Observations 178 Frequency Annual Year 2004 lnYi,k Percentages 202 Annual 2004 ln Yj Percentages 286 Annual 2004 adjij A dummy variable=1 if two countries share a common border 1 or 0 286 Annual 2004 Ln Dij Log, A measure of internal and Percentages external distances. Will differ under Wei and Leamer. A dummy equal to 1 if trade is internal, 0 if not Is a white noise error term 1 or 0 288 Annual 2004 Home 286 Annual 2004 εij,k Table A.1 Sources for data Source lnXij,k SourceOECD, International Trade by commodity statistics. SITC Rev3 FAOSTAT database Institution SourceOECD, OECD Food and Agriculture Organization Citation http://titania.sourceoecd.org /vl=3102017/cl=28/nw=1/rps v/statistic/s15_about.htm?jnl issn=16081218 Year 2004 2004 http://faostat.fao.org/ lnYi,k FAOSTAT database Food and Agriculture Organization Euromonitor International Foreign Agricultural Service, USDA http://faostat.fao.org/ 2004 EUROMONITOR www.euromonitor.com 2004 FAS Report, Major Oilseeds FAS Report, Table10 http://miniurl.org/lyp http://www.fas.usda.gov/oi lseeds/circular/2004/0412/table10.pdf http://www.fas.usda.gov/g ainfiles/200712/146293251. pdf 2004 GAIN Report CH7623, FAS Market and Trade Data Foreign Agricultural Service, USDA Economic Research Service, USDA OICA, International Organization of Motor Vehicle Manufacturers United Nations Statistics Division. 2007 Cotton and Wool Yearbook 2004 2004 http://miniurl.org/0Mq 2004 http://oica.net/category/pr oduction-statistics/2004statistics/> 2004 World Production Statistics ln Yj GDP at market prices, Key Global Indicators http://data.un.org/Data.asp x?q=GDP&d=CDB&f=srID%3 a29919 Ln Dij Self-generated 2004 Descriptive Statistics Variable Number of Observations 178 202 286 288 288 214 288 288 286 Mean Standard Deviation 3.60 2.65 1.14 0.99 0.64 2.72e+10 815.61 719 6104257 Minimum Maximum Ln Xij Ln Yik LnGDPj LdistWij LdistLij X distWij distLij gdpj 18.66 21.41 14.85 6.61 6.99 5.49e+09 1112 1316 5559307 7.101 15.23 13.89 5.08 6.129 0 161 459 1088128 26.577 26.577 16.47 7.713 7.712 3.49e+11 2239 2237 1.43e+07 Correlation Coefficients Ln Xij 1.0000 0.5172 0.1166 -0.5141 -0.2026 Ln Yik 1.0000 -0.0772 -0.0619 0.0514 LnGDPj LdistWij LdistLij Ln Xij Ln Yik LnGDPj LdistWij LdistLij 1.0000 -0.2693 -0.0752 1.0000 0.4955 1.0000 Scatterplot |Log, Bilateral Export Flow and Log Domestic Production lnx 5 10 15 20 25 16 18 20 lnYik 22 24 26 I used data in the natural log form to normalize my data to make it easier to use in the regression. Log models are invariant to the scale of the variables since they measure percent changes. For large variables such a GDP, or production using logs has much practical value. Furthermore using log of bilateral exports helps determine the elasticity of its response to changes in explanatory variables. The value of domestic production was obtained by simply multiplying the price of the commodity or good on the world markets by the quantity produced domestically, where domestic prices were not available. This may tend to inflate the value of domestic production, and hence the value of goods traded internally [Production-Exports] overestimating the border effect. Data for the value of domestic production in 2004 was sometimes difficult to find or stated in terms of an index of 2000 prices. The OECD Source database was helpful with values for exports, but did not carry numbers for production. The FAO database had values for domestic production, but it was largely limited to agricultural goods. Another database, the United States International Trade Commission database was also helpful, but was restricted primarily to measures of US exports, and did not contain data for Mexico or Canadian exports. Estimation Methods The dataset I have assembled is a cross section for the year 2004. I will run a simple Ordinary Least Squares Regression as performed in Chen. The selected regression will be tested for heteroskedasticity. If prevalent the regression will be re-estimated using robust standard errors. Additional tests to determine multicollinearity and correct for this will also be attempted. Empirical Results The model used in Chen estimates a simple gravity equation as a pooled Ordinary Least Squares Regression of bilateral export flows. This estimation allows us to assess the average border effect value for our three countries. The model is replicated below, and results reported in the adjoining table. lnXij,k = β0 + β1home + β2 lnYi,k + β3ln Yj + β4adjij + β5 ln Dij + εij,k The estimation of this model over the pooled sample allows us to assess the average border effect value for our sample of three countries and 26 industries. Table 1 reports the results using two different measurements of internal distances, Wei (1) and Leamer (2). Standard deviations appear in brackets under each coefficient. The t and p values for each coefficient appear alongside in the adjacent column. Table 1. Average and country specific border effects (1) t-and p values 9.94 (0.00) (2) t-and p values 9.96 (0.00) lnYi,k ln Yj adjij Ln Dij Home 0.718 (0.07) 0.719 (0.07) 0.165 (0.17) 0.97 (0.33) 0.168 (0.16) 1.05 (0.29) 2.963 (0.64) 4.59 (0.00) 2.965 (0.67) 4.45 (0.00) -0.015 (0.43) -0.04 (0.97) -0.0082 (0.32) -0.03 (0.98) 5.884 (1.11) 5.28 (0.00) 5.909 (0.69) 8.45 (0.00) Distance Measures Notes: N =178. Wei Leamer From the basic equation estimated by Wei and Leamer the co-efficient on the variable home appears to highly significant and very strong. For Wei’s distances the co-efficient on home is 5.884 suggesting that the typical country trades over 357 times (=exp(5.88)) as much with itself than with other countries when adjusting for other factors. The coefficient when Leamer distances are employed is even greater. These results are in the same direction as those found in Chen but the effect is far greater here. The coefficient on the home variable is known to be sensitive to different measures of internal distances, which may have biased our coefficient value upwards. Another reason for the high estimates is that relative prices have not been controlled for. As documented in Chen and Anderson and Wincoop (2001) adjusting for relative prices tends to reduce the border effect. For a 1 % increase in domestic production of a commodity (lnYi,k) the model predicts a 0.71% increase in bilateral trade. Adjacent countries trade much more with each other than those further away. This result is as expected, and the model confirms it. The coefficient on adjij tells us that countries bordering each other trade 19 times (=exp(2.96)) more with each other than those which do not. A surprising result is that log GDP (ln Yj) and log distance (Ln Dij) appear to be jointly insignificant at the 5% level, an outcome not observed in the literature. Could the results have been different if the model been adjusted for heteroskedasticity? If it is present, heteroskedasticity would not affect the coefficients, but it would increase the standard deviations and thus the t and p values. The Breusch-Pagan test reveals heteroskedasticity; the F value for the test is very significant at the 5% level. Results from the regressions rerun with robust standard errors are summarized in Table 2. Table 2. Average and country specific border effects, robust (1) t-and p values 9.03 (0.00) (2) t-and p values 9.35 (0.00) lnYi,k ln Yj adjij Ln Dij Home 0.718 (0.079) 0.719 (0.07) 0.165 (0.177) 0.93 (0.35) 0.168 (0.14) 1.16 (0.25) 2.963 (0.84) 3.52 (0.00) 2.965 (0.86) 3.43 (0.00) -0.015 (0.44) -0.04 (0.97) -0.008 (0.32) -0.03 (0.98) 5.884 (1.19) 4.90 (0.00) 5.909 (0.88) 6.65 (0.00) Distance Measures Notes: N =178. Wei Leamer Running the regressions with robust standard errors does not significantly change the outcome. Both log GDP (ln Yj) and log Distance (Ln Dij) are still jointly or individually statistically insignificant at the 5% level. This is observed for distance calculated using Wei or Leamer distances. As discussed earlier there are many possible causes for this outcome. One possibility is that without relative prices, both GDP and distance measures become insignificant. This is a possibility since relative prices reduce the ‘border’ effect- associated with the coefficient on the variable home. The sample of three countries may not be of a sufficient size to capture bilateral trade in more detail. Increasing the number of countries and industries may then increase the significance of some of the explanatory variables in line with has been observed in previous studies. Unfortunately, data for values of domestic production is not that broad; there are several missing values in the data, some of which are simply unavailable in major databases. Conclusion My intention in writing this paper was to see if the results from Chen (2004) could be replicated with data from trade between the US, Canada and Mexico. The final results do fall in line with her observations in important areas. The coefficient on the variable for domestic trade is strong, significant and positive. Adjacent countries also do far more trade with each other, controlling for the size of the economy and other variables. I was not able to observe a strong predictive relationship for some variables, which was unexpected. However if future papers include controls for relative prices, and fully replicate her analysis, there is a possibility of these effects being fully observed. References Papers Anderson, J.E., van Wincoop, E., 2001. Gravity with gravitas: a solution to the border puzzle. National Bureau ofEconomic Research Working Paper 8079. Chen, Natalie, 2004. Intra-national versus international trade in the European Union: why do national borders matter ?. Journal of International Economics 63, 93-118. Hillberry, R., 1999. Explaining the ‘border effect’: what can we learn from disaggregated commodity flow data? Indiana University Graduate Student Economics Working Paper Series 9802. Indiana University. Hillberry, R., Hummels, D., 2000. Explaining home bias in consumption: production location, commodity composition and magnification. Purdue University, mimeo Leamer, E.E., 1997. Access to western markets, and eastern effort levels. In: Zecchini, S. (Ed.), Lessons from the Economic Transition: Central and Eastern Europe in the 1990s. Kluwer Academic Publishers, pp. 503– 526. McCallum, J., 1995. National borders matter: Canada–US regional trade patterns. American Economic Review 85 (3), 615– 623. Wei, S.-J., 1996. Intra-national versus international trade: how stubborn are nations in global integration? National Bureau of Economic Research Working Paper 5531. Wolf, H.C., 1997. Patterns of intra- and inter-state trade. National Bureau of Economic Research Working Paper 5939. Databases SourceOECD, International Trade by commodity statistics. SITC Rev3. OECD. Retrieved May 11 2009. http://titania.sourceoecd.org/vl=3102017/cl=28/nw=1/rpsv/statistic/s15_about.htm?jnlissn=160812 18 ProdSTAT and FAOSTAT Database, Food and Agriculture Organization, United Nations. Retrieved May 11, 2009. http://faostat.fao.org/ (2004). Euromonitor International. Retrieved May 11, 2009, from Euromonitor Web site: http://www.euromonitor.com/ (2004). Table 10. UNITED STATES: OILSEEDS AND PRODUCTS SUPPLY AND DISTRIBUTION. Retrieved May 11, 2009, Web site: http://www.fas.usda.gov/oilseeds/circular/2004/04-12/table10.pdf (2004). Table 4. Major Oilseeds: World Supply and Distribution (Country View), Foreign Agricultural Service, USDA. Retrieved May 11, 2009, Web site: http://www.fas.usda.gov/psdonline/psdreport.aspx?hidReportRetrievalName=BVS&hidReportR etrievalID=703&hidReportRetrievalTemplateID=8 (2007). GAIN Report CH7623, FAS Market and Trade Data, Foreign Agricultural Service, USDA. Retrieved May 11, 2009, Web site: http://www.fas.usda.gov/gainfiles/200712/146293251.pdf (2004). Cotton and Wool Yearbook, Economic Research Service, USDA. Retrieved May 11, 2009, Web Site: http://www.apeda.com/TradeJunction/Report/December_2008/cotton_and_wool_outllok.pdf (2004). World Production Statistics, OICA International Organization of Motor Vehicle Manufacturers. Retrieved May 11, 2009, Web Site: http://oica.net/category/production-statistics/2004-statistics/ (2004). GDP at market prices, Key Global Indicator, United Nations Statistics Division. Retrieved May 11, 2009, Web Site: http://data.un.org/Data.aspx?q=GDP&d=CDB&f=srID%3a29919