Do Political Factors Influence U.S. Crude Oil Imports?

This work uses a gravity model to estimate the determinants of U.S. crude oil imports, introducing two variables tied to the political nature of trade: U.S. arm sales and the level of political freedom in a country. Model results indicate that contiguity is important for U.S. crude oil imports, but trade agreements are not. Being an OPEC member leads to more exports of crude oil to the U.S. These factors combined with the negative relationship with freedom and U.S. crude oil imports, suggests that the U.S. imports a lot of oil from countries that have little cultural and political similarities.


INTRODUCTION
The United States has historically been the world's largest importer of crude oil, importing, on average, 21% of the global value of crude oil imports from 2002 to 2015 (UN Comtrade, 2018). U.S. imports of crude oil have declined from its peak in 2005 largely due to the domestic production growth from the shale oil boom; however, until very recently, the United States was still the largest importer of crude oil. 1 Although crude oil is noted as being the commodity traded the most, its trade is often contentious due to issues such as national security and political factors (Krane and Medlock, 2018;Zhang et al., 2015). The commodity includes, arguably, the most influential cartel in the world, the Organization of Petroleum Exporting Commodities (OPEC), which is responsible for around 42-44% of the world's production of crude oil and 56-60% of oil trade (OPEC, 2017). In addition, crude oil is important enough to influence agricultural commodity prices-through changes in the of short-run changes from political shocks on exports to China; Fuchs and Klann (2013) conclude that a visit from the Dalai Lama to a country reduces their exports to China the next year; and Heilmann (2016) estimates a reduction in trade from political events that call for boycotts. Summary (1989) makes the argument that bilateral cooperation and conflict is positively associated with levels of bilateral trade. U.S. oil trade and politics has been investigated by Kashcheeva and Tsui (2015) who use micro-level data to examine what kinds of firms are more responsive to change in political distance between the U.S. and its trading partners, measured by divergence in their UN General Assembly voting patterns. Mityakov et al. (2013) use a gravity model to examine U.S. imports of crude oil, also using the political distance factor as was done in Kashcheeva and Tsui (2015). Using data from 1962 to 2000, their results indicate that a divergence in UN voting patterns has a small, significant impact on U.S. oil imports.
Our paper is similar to these papers-in that we are interested in determining if certain political factors and energy policy influence U.S. oil imports. In his 2007 State of the Union address, President George W. Bush stated that U.S. reliance on foreign oil has rendered the nation's interests "vulnerable to hostile regimes, and to terrorists who could cause huge disruptions of oil shipments, and raise the price of oil, and do great harm to our economy." To test if political factors influence U.S. oil imports we construct two variables in a gravity model that goes beyond voting patterns: U.S. arm sales to countries, and the amount of political freedom in a country. To examine energy policy factors affecting U.S. oil imports, we test whether OPEC remains an important sources of U.S imports or if the U.S turns to those whom they have free trade agreements with. We are also interested in examining what standard gravity model explanatory variables (common language, distance, and trade openness) impact U.S. oil imports. Crucially, one point we make in this paper is that there is likely a structural break in U.S. oil imports, due to the fall in imports since 2006. We test for the break (and find one), thus we decompose our data into two periods: rising U.S. oil imports (1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006) and falling U.S. oil imports (2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016). This phenomenon would not have been observed in papers such as Mityakov et al. (2013) and Kashcheeva and Tsui (2015) because their data stops at the year 2000 and 2008.

BACKGROUND
Given the importance of crude oil in the economy, the literature has focused on explaining different aspects of the commodity. Among the aspects, examining fluctuations in crude prices and explaining trade patterns have received, perhaps, the most attention (e.g., Prest, 2018;AlKathiri et al., 2017;Zhang et al., 2015). As our paper focuses on trade patterns, we look deeper at the literature on this topic. In addition, background information on U.S. crude oil imports is provided in this section.

Literature Review
The topic of explaining crude oil flows has been investigated in the literature using a variety of methods. AlKathiri et al. (2017) use a simulation model to examine the ability of large crude oil traders to influence inter-regional price differentials. Research has also examined global oil trade using network theory. An et al. (2014) specify a trading-based network model of international crude oil to study the relationship between countries with common trade partners. Ji et al. (2014) construct a global oil trade core network to analyze oil trade using complex network theory. Further work has examined the relationship between oil prices and macro factors. Amin et al. (2021) examine the nexus between oil price and exchange rate for Bangladesh's economy. Alkhateeb et al. (2021) study the role of oil in fiscal policy cyclicality in Saudi Arabia. Endri et al. (2021) study the effect of oil prices on the stock market in Indonesia. Zhang et al. (2015) use spatial econometric modeling of origindestination flows to identify and analyze the driving factors for the formation of current oil trade patterns. They find that most oil exporters have a focus on exports to North America. They note that technological progress and energy efficiency have significantly influenced oil trade, beyond the common supply and demand factors. Other work that has used a gravity-type model on trade in energy commodities include Zhang et al. (2018), who investigate the determinants of liquefied natural gas (LNG) trade. Note that their work includes a political risk index variable, with results of the gravity model indicating that importing countries are more willing to import LNG from regions with stable political environments.

U.S. Crude Oil Imports
Much attention has been made recently about the tremendous increase in U.S. crude oil production, but Figure 1 indicates that the U.S. produced more crude oil than it imported as recently as 1993. U.S. production of crude oil was 2.71 billion barrels in 1991, and decreased to 1.83 billion barrels by 2008 (a decrease in production of 32.4%). This decrease in production was largely a function of low global prices, as the turnaround in U.S. crude oil production after 2008 was fueled by high global prices making fracking and other technologies profitable. As U.S. gross domestic product (GDP) continued to grow from 1991 to 2008, so did its demand for crude oil. This increase in U.S. crude oil demand was filled largely by imports as shown in Figure 1. Indeed, the increase in U.S. crude oil imports and GDP from 1991 to 2008 were 70% and 66%, respectively. U.S. crude oil imports, however, plateaued in the mid-2000s, and declined beginning in 2006. This decline in imports coincided with an increase in U.S. crude oil production.
Despite the reduction in U.S. crude oil imports after 2006, the United States remained the largest importer of crude oil, globally, when China became the largest importer in 2017 (EIA, 2018a). Changes in politics, energy policy and trade agreements have changed the sources of U.S. crude oil imports over time. Figure 2 indicates that Saudi Arabia provided the largest share of crude oil to the U.S. until 1993, when the Canada and Mexico combination became the largest source, providing more than 45% of crude oil imports. However, for individual countries, Saudi Arabia remained the largest source until 2006, since then, Canada has become the largest source of crude oil to the U.S. The share of imports from Saudi Arabia has remained around 15%, but note the large increase in imports from Canada and Mexico. There are two factors for this: the implantation of the North American Free Trade Agreement (NAFTA) in 1994, and the increase in Canada's shale oil production. Crude oil from those two countries have replaced imports from Nigeria and kept the share from other Middle East and Russia small.
Globally, sources of crude oil imports present a different picture than that for the U.S. Figure 3 indicates that the current largest source is the other Middle East composite, which is largely from Iran, Iraq, Kuwait and the United Arab Emirates. The decrease in other Middle East share from 1990 to 1994 was a result of the Gulf War. Saudi Arabia has also experienced a decline in its share of global crude oil imports, but the share has remained above 15% (like that for the U.S.). The largest gain in share has, however, been  for Russia. It went from exporting very little crude oil in 1996 to becoming the second largest exporter after Saudi Arabia. Note, that the U.S. imports very little crude oil from Russia.

GRAVITY MODEL
The work by Tinbergen (1962) first linked the pattern of bilateral trade to national income, population, and distance. Much of the literature using the gravity model has used aggregated data to examine bilateral trade flows between a large collection of countries. There has been less research examining specific commodities; however, the gravity specification can lend insight into the flow of trade of specific commodities. This paper seeks to contribute to the literature examining specific commodities by using the gravity specification to evaluate the determinants of U.S. crude oil trade imports.
In one of the first studies to specify a commodity specific model, Koo et al. (1994) examine the trade flows of meat using a commodity specific gravity model. Dascal et al. (2002) use the commodity specific gravity model in order to analyze the main factors affecting the trade flows of wine in the European Union. Prentice et al. (1998) specify an empirical gravity model for Canadian pork exports to the United States to identify the particular regions in the U.S. where pork exports from each province in Canada could be expanded.

Methods
Here we use the gravity model developed by Anderson and van Wincoop (2003). The basic conceptual idea is presented in Equation (1): At each point in time t, the volume of U.S. crude oil imports from its trading partners i Oil us,t , 2 depends on U.S. GDP (GDP us,t ), the GDP of its trading partners (GDP it ), the average physical distance between the U.S. and the trading partner (Dis us-i,t ), and a number of trade limiting or/and enabling factors, π us−i,t , such as sharing a border (contiguity), having a bilateral free trade agreement (FTA), having a similar official language (common language), stable exchange rate regimes (exchange rate), the level of trade openness (trade openness), membership in similar international trade institutions such as World Trade Organization (WTO) and/or Organization of Petroleum Exporting Countries (OPEC), and the political economy factors (freedom index) and U.S. arm sales (U.S. arm sales).
The estimation is performed using a log-linearize specification, shown in Equation (2) Ln Oil us,t = β o + β 1 ln GDP us,t + β 2 ln GDP it + β 3 ln Dis where the coefficient for β 1 , β 2 , β 3 and some non-dummy variables from β j (such as trade openness, exchange rates) are interpreted as the elasticity of U.S. crude oil imports with respect to changes in the independent variables.
When it comes to estimation techniques 3 , there is no one general best performing estimation, and the selection of the estimation techniques depends on a number of factors (Kareem, 2013). Piermartini and Yotov (2016) noted that gravity estimates in the existing literature suffer biases and inconsistency despite the progress in gravity model applications. They also reviewed some drawbacks and possible solutions. For instance, the presence of zero trade flows in a number of observations would make the Ordinary Least Square (OLS) estimation approach impossible when the volume of trade is transformed into a logarithmic form. This can be solved by use of the Poisson Pseudo Maximum Likelihood (PPML) estimator that was advocated by Santos Silva and Tenreyro (2006). PPML can also provide a consistent estimate in the presence of heteroscedasticity. In addition, Feasible Generalized Least Square (GLS) can also be used with the problem associated with log normal formulation and the presence of zero valued trade flows (Kareem, 2013). As shown next in the data section, we do not anticipate a problem as a result of zero or missing data in most of our data series. Nevertheless, we present the results using GLS, OLS, and PPML estimation techniques for robustness.

Data
The literature notes that trade flows in crude oil are largely a result of advantages in transportation costs and preferences for different qualities (Grant et al., 2006). This makes the gravity model a good candidate to explain trade flows. The gravity model can be used to trace and understand the flow of goods with the understanding that the volume of bilateral trade tends to vary directly with market size and inversely with distance (Prentice et al., 1998).

RESULTS
We begin our discussion of results with the test of a structural break, since Figure 1 indicates that U.S. crude oil imports seemed to face two regimes: increasing imports until the mid-2000s, then decreasing imports since then. We will present the results for the specified gravity model based the resulting timeline indicated from the structural change test.

Structural Break?
We test for a structural break in U.S. crude oil import data, assuming that the break point is not known. This Wald test is done by combining the test statistics computed for each possible break date in the sample. The null hypothesis of the test is that there is no structural break, the results noted in Table 2 Table 3 presents the econometric results for the basic gravity specification. 4 These results indicate that U.S. crude oil imports are inversely related to U.S. GDP; although the results are not statistically significant for the PPML estimation. The literature notes that this coefficient is most often positive in a gravity model; however, recall the decrease in U.S. crude oil imports over the time-frame (and the increase in U.S. GDP). The coefficient on exporters' GDP is positive and significant across all time frames, indicating that the U.S. tends to import crude oil from countries who have an increasing GDP. Distance is inversely related to U.S. crude oil imports, as suggested by the literature. Contiguity is positive, as the U.S. imports more than 45 per cent of its crude oil from Canada and Mexico. The sign on the coefficients and statistical significance are the same across all three econometric methods for contiguity. This basic model gives an adjusted R-squared of around 0.27 for the GLS and OLS estimation, but the PPML specification provides a better fit for the data (i.e., a larger adjusted R-squared). Note that none of these variables are included in Mityakov et al. (2013). Pursuant to our structural break, none of the variables in Table 3 have a change in sign; although the magnitude changes across time frames and U.S. GDP becomes statistically significant in one time-frame, but not the other. The coefficient on WTO membership is positive and statistically significant for many of the specifications in Mityakov et al. (2013), but as mentioned before, their time-frame is until the year 2000. The dependent variable is US crude oil import; all non-dummy variables are in log format. Standard errors in parentheses. *P < 0.10, **P < 0.05, ***P < 0.01

Gravity Model Results with Political Factors
A third model includes all the previous variables, but now includes our two political factors: Freedom Index and U.S. arms sales. Table 5 presents those results. First, focusing on the importance of those two variables in the model, note that the variable for U.S. arms sales is statistically significant in the GLS model, indicating that the U.S. is more likely to import crude oil from countries that it sales arms to. However, this variable is not statistically significant for all time-frames in the OLS and PPML specifications. The Freedom Index variable is statistically significant (at the 1%) level across all methodologies and time-frames (except for the 1993-2006 period in the OLS estimation), and the sign is positive. This indicates that a country with a lower level of Freedom Index is more likely to export crude oil to the United States (recall that the Freedom Index is inversed in terms of freedom). The magnitude for both of these political variables is stronger in the 2007-2016 time-frame compared to the earlier time-frame.
The table also indicates that the same thing occurs for the coefficient on OPEC, WTO, and their interaction term. The coefficients on OPEC and WTO are positive in the 1993-2006 time-frame, but not so in the 2007-2016 time-frame. Again, this occurs because of the interaction term: being a member of both OPEC and the WTO increases the likelihood that the U.S. imports oil from that country. Trade openness is statistically significant and negative across 7 of the 9 time-frame/econometric combinations-indicating that the U.S. imports crude oil from countries that are less open. Contiguity, common language, exchange rate, and the FTA member dummy variable are not always statistically significant, indicating that much of the variance is being accounted for in the political variables. But, overall, this model with political factors has the highest adjusted R-squared compared to the earlier models.

Robustness Results
The final table presents the results for different model specifications when dummy variables are one, as a robustness check of our results. The results are presented for the GLS model, results using the other methods are available upon request. The first three columns of results in Table 6 are for those who are members of the WTO. The coefficients on U.S. and exporters' GDP are the same as in the previous models. Indeed, most of the variables have the same sign and statistical significance as those in the previous model; however, the U.S. arm sales variable is now no longer statistically significant.
The second three column in Table 6 presents the results for FTA partners. This time, the coefficient on U.S. GDP is positive (across the entire time-frame), as can be expected from the increase in U.S. crude oil imports from Canada and Mexico. Some additional variables also have a change in sign when examining the results for the entire time-frame. Distance is now positively related to crude oil imports, mainly because U.S. imports of crude oil are   1993-2006 2007-2016 1993-2016 1993-2006 2007-2016 1993-2016 1993-2006 2007-2016 1993-  almost exclusively from Canada and Mexico for FTA partners. Interestingly, in the 1993-2006 sample, distance is negative (and contiguity is not statistically significant), but this result is reversed in the later time-frame (due to the increase in imports from Canada and Mexico). The exchange rate is statistically significant in this model, and the Freedom Index is statistically significant at the 10% level. This model also features the highest adjusted R-squared at 0.968.
The final specification presents results for those countries who are a member of OPEC. Across the entire sample, U.S. GDP is no longer statistically significant, as is the case for many variables that were statistically significant in the other models (common language, WTO member, trade openness). Indeed, the adjusted R-squared for this specification is the lowest of all models (0.239). U.S. arm sales are statistically significant, as is the Freedom House Index, indicating that the U.S. is more likely to import crude oil from OPEC members that it sales arms to, and those that have less freedom.

CONCLUSIONS
Crude oil is arguable the most important commodity in the world as it provides a large part of energy to the world. The United States is a major player in the market, importing and consuming a large share of crude oil. Therefore, it is important to investigate the variables that impact U.S. crude oil imports.
Because the literature suggests that transportation costs and quality are important to crude oil trade flows, we use a gravity model for this analysis. We include the standard gravity variables, and also consider several factors related to trade. But, the main contribution to the literature is two parts: the specification of two variables related to political relations: U.S. arm sales and the level of freedom in the exporting country. Crude oil is much different than any other commodities, since it is heavily traded and is deemed a national security issue almost worldwide. In addition, we also test for a structural break in our data, with results indicating that the determinants of U.S. crude oil imports differ beginning around 2007.
After conducting several exercises in specification for the gravity model, we can make the following conclusions. First, distance is an important determinant for U.S. crude oil imports, with results indicating that as distance increases, imports to the U.S. decreases. Second, results indicate that U.S. GDP and oil imports are inversely related, which is due to the fall in U.S. crude oil imports over the last eight years. Contiguity is important for U.S. crude oil imports (since it imports so much from Canada and Mexico), but free trade agreements are not.
Being an OPEC member leads to more imports of crude oil to the U.S., while a common language and trade openness are not. These factors combined with the negative relationship with freedom and U.S. crude oil imports, suggests that the U.S. imports a lot of oil from countries that have little or no cultural and political similarities. Finally, the econometric analysis across time-frames indicates that the U.S. is now in a different era then when it was the world's largest importer of crude oil.