Geopolitical distance and international trade (OeNB Bulletin Q2/25)

Ana Abeliansky, Julian Mayrhuber ¹

JEL classification: F1, F5, P0

Keywords: international trade, geopolitical distance, globalization

1 Introduction

In an increasingly polarized global landscape, geopolitical fragmentation is reshaping economic exchange and, in particular, international trade dynamics. Rising geopolitical tensions, economic sanctions and shifts in trade alliances have created new uncertainties for global markets, challenging long-established patterns of trade and financial flows (e.g., Aiyar et al., 2023a and Gopinath et al., 2025). Understanding these shifts is crucial, as disruptions in trade can impact economic growth, inflation, exchange rates and financial stability.

Recent research highlights the growing economic consequences of geopolitical fragmentation (e.g., Alvarez et al., 2023; Attinasi et al., 2023; Góes and Bekkers, 2022 and Javorcik et al., 2024). Campos et al. (2023) examine how trade flows are affected by shifting geopolitical alliances, showing that geopolitical fragmentation can lead to significant trade diversion and welfare losses. Fernández-Villaverde et al. (2024) provide empirical evidence on the macroeconomic consequences of fragmentation, particularly its impact on global trade networks and financial stability. Gopinath et al. (2025) further emphasize the role of geopolitical shocks in altering trade patterns, affecting supply chains and influencing global monetary policy responses.

Given the central role of trade (policy) in economic growth, inflationary pressures and exchange rate movements, central banks should take into account these emerging risks to effectively navigate an evolving economic landscape (Frankel and Romer, 1999; Ambrosino et al., 2024; Broda and Romalis, 2011). Along these lines, Attinasi et al. (2024) discuss the challenges central banks face in a fragmenting global trading system, noting that geoeconomic fragmentation increases the complexity and unpredictability of the operating environment. Similarly, Aiyar et al. (2023a) highlight that trade and investment flows are being redirected along geopolitical lines, indicating that there are signs of fragmentation beneath the surface of global trade. Additionally, Qiu et al. (2025) underscore the heightened policy uncertainty and unpredictability stemming from geopolitical tensions, which complicate the economic outlook for central banks. Their study also finds supportive evidence for the negative effect of geopolitical fragmentation on trade values – stemming from effects on prices and quantities.

Building on these findings, this paper investigates the link between geopolitical fragmentation and international trade, exploring how shifting geopolitical relations influence the trade of goods between country pairs. To do so, we use the Base pour l’Analyse du Commerce International (BACI) bilateral trade dataset, which spans from 1995 until 2023. We merge the dataset with standard control variables from the literature (e.g., free trade agreements and common currency for the fixed effects model, or additionally geographical distance and common language for the pooled ordinary least squares regression). Afterward, we combine it with the ideal point distance (IPD) from Bailey et al. (2017) to investigate whether geopolitical differences influence bilateral trade. We then proceed with a heterogeneity analysis across time, type of importer/exporter and geopolitical blocs. Our study improves on past literature by using a wide array of countries, a longer time frame and a broad heterogeneity analysis that has not been done before. In other words, we investigate the relationship between geopolitical differences and international trade across time and also within country groups and blocs. Overall, our findings aim to provide insights into how adverse the effects of geopolitical fragmentation are, and how important it is to build economic resilience in an era of uncertainty. The paper is structured as follows: Section 2 presents the empirical strategy. Section 3 discusses the main findings and assesses their robustness. Finally, section 4 concludes.

2 Empirical strategy

In this section, we introduce the empirical model and describe the data used.

2.1 Empirical model

Most of the large literature investigating the impact of trade barriers on international trade has been based on the theoretically micro-founded model of Anderson and van Wincoop (2003) ² . The authors propose that bilateral trade can be explained by the economic size of the exporter and importer countries (divided by world GDP), by trade costs needed to move goods across countries (that act as barriers to trade) and by multilateral resistance terms, i.e. the ideal price indices of the exporter and importer countries. The latter essentially allow to control for third-country effects, that is to consider in the analysis the relative importance of a trade barrier compared to trading with alternative trading partners.

Our empirical model for estimation is as follows:

\(y_{ijt} = e^{(\alpha*x_{ijt - 1}\ + {\ \ \beta*IPD}_{ijt - 1}\ + \ \gamma*z_{ij} + \ \delta_{it}\ + k_{jt})\ }\eta_{ijt}\) (1)

where \(y_{ijt}\) is the bilateral import trade share calculated as the value of exports from country i to country j in year t standardized by the value of total imports from country j in year t . Unless otherwise stated, trade shares are import trade shares. We follow other studies such as Eaton et al. (2012), Head and Mayer (2014), Novy (2013) and Carrère et al. (2020) and use trade shares as the dependent variable rather than the value of bilateral trade. ³ Sotelo (2019) and Head and Mayer (2014) propose that the shares diminish the relative importance of large values of trade, since large trade flows are typically directed toward countries with high overall import volumes. Using bilateral trade shares rather than values thus ensures that all countries are weighted equally in the estimation procedure.

Following the standard in the literature, we use a Poisson Pseudo Maximum Likelihood (PPML) regression model to include zeroes in the dependent variable. As Santos Silva and Tenreyro (2006) highlight, this regression model should prevent the heteroskedasticity from a log-normalization of the error when we linearize a multiplicative model like that of Anderson and van Wincoop (2003). Sotelo (2019) also shows that using the trade shares in the PPML regression combined with destination country fixed effects is equivalent to a Multinomial Pseudo Maximum Likelihood (MPML) regression described by Gourieroux et al. (1984).

On the right-hand side of equation (1), we include time-variant variables (\(x_{ijt - 1}\)) that proxy for time-variant bilateral transport costs such as having a deep trade agreement ⁴ in place and sharing a common currency. We expect these two variables to enable trade by diminishing the economic costs of trading resulting, e.g., from tariffs or exchange rate uncertainty. Then, we have our main variable of interest, the geopolitical distance \({\ IPD}_{ijt - 1}\). We expect that geopolitical de-alignment would – if anything – deter trade between countries. Countries usually do not trade as much with “enemies,” as they do with “friends.” The variables \(x_{ijt - 1}\) and \({\ IPD}_{ijt - 1}\) are lagged by one period to ameliorate endogeneity concerns, mainly stemming from reverse causality.

Furthermore, we include country-pair fixed effects (\(z_{ij}\)) to absorb any variable measured at the bilateral level which is constant over time and might be correlated with the other covariates. In the specifications without country-pair fixed effects, we include variables that proxy for trade barriers such as the natural logarithm of geographical distance between capitals, having a common border, a common colonial past, a common language and a common religion. While longer distances are expected to diminish trade, since they increase transportation costs, all other variables should increase trade. Our preferred specification is the one that includes country-pair fixed effects, as it also captures all other possible time-invariant factors (at the bilateral level) and we are not specifically interested in one particular variable. Finally, to control for multilateral resistance (and anything that is country- and time-specific), we include the country of origin (i) and time (t) as well as the country of destination (j) and time fixed effects (\(\delta_{it}\) and \(k_{jt}\), respectively). \(\eta_{ijt}\) is the error term, which we assume to be well-behaved. To allow for correlation across time within country pairs, we cluster the standard errors at the country-pair level.

2.2 Data sources and summary statistics

To capture geopolitical (de-)alignment between countries, we rely on the ideal point distance (IPD) from Bailey et al. (2017) that builds on the voting behavior in the United Nations General Assembly (UNGA). The IPD reflects how far apart countries are in terms of their foreign policy positions and it is a widely used measure to study the effects of geopolitical fragmentation (e.g., Abeliansky et al., 2024; Aiyar et al., 2024 and Gopinath et al., 2025). This measure is based on deriving each country’s individual foreign policy position (ideal points) with respect to the US-led liberal order from its UNGA voting behavior. Chart A1 in the annex presents the unilateral ideal points for all the countries of our final sample. Higher ideal points indicate closer alignment with the US-led liberal order, whereas lower values suggest the opposite. Then, the bilateral IPD is calculated as the absolute difference between two countries’ ideal points. Another widely used measure for geopolitical de-alignment is the S-score proposed by Signorino and Ritter (1999), which is simpler to construct than the IPD and reflects how similarly two countries vote in the UNGA. ⁵ Bailey et al. (2017) argue that the IPD is more suitable than the S-score for time series analysis, as it accounts for shifts in the UNGA agenda. For this reason, we use the IPD as a measure for geopolitical (de-)alignment in our baseline specification. ⁶

For the bilateral trade shares, we use the BACI dataset from Gaulier and Zignago (2010), which is based on UN COMTRADE data and provides consistent bilateral trade data across countries. It adjusts for discrepancies between reported data from importers and exporters to create a balanced bilateral trade dataset. Furthermore, we also use standard gravity controls (i.e., geographical distance, contiguity, common colonial history, common language and religious proximity) from the CEPII Gravity database (see Conte et al. (2022) for a description of the dataset). Moreover, for information on deep trade agreements, we use the World Bank Deep Trade Agreements database by Hofmann et al. (2017). The variable taken from this dataset covers agreements which support economic integration with respect to trade in goods and services, capital flows and movement of people and ideas (Mattoo et al., 2020). Finally, for data on common currency, we rely on the dataset available from the website of José de Sousa ⁷ .

Overall, our sample covers annual data ranging from 1995 to 2023 ⁸ . The analysis includes 180 countries, with 34 classified as advanced economies (AEs) and 146 as emerging market and developing economies (EMDEs), following the definition used by the IMF ⁹ . Table 1 shows the summary statistics for the sample used throughout the analysis. On average, trade shares are around 1% (and this average remains essentially unchanged when zero values are excluded). Moreover, while the between standard deviation of IPD is larger, there is also sufficient within variation, which is important for the analysis at the country-pair level. We also find that in about 18% of the observations, some form of deep trade agreement is in force, while a common currency is shared in around 1% of the sample observations. While trade shares, IPD, common religion and geographical distance are continuous variables, the rest are dichotomous.

Table 1

Since the focus of our study is the relationship between trade and geopolitical distance (IPD), we decided to first provide a graphical illustration of this relationship. Using a binned scatter plot, chart 1 shows that higher geopolitical distance is associated with lower trade shares when controlling for origin and destination country fixed effects. Furthermore, it shows that the link is more negative in 2023 compared to 1995 ¹⁰ , suggesting that geopolitical alignment has become increasingly important for bilateral trade relations. To analyze the link between trade and geopolitical distance more formally, the next section presents the results of a gravity model analysis.

Chart 1

3 Empirical results

In column (1) of table 2, we try to explain bilateral trade shares by our main variable of interest, i.e. geopolitical distance (IPD). In this specification, we find that IPD is statistically significant at the 1% level. In column (2), we add several control variables which are time-invariant, weakening the statistical significance. The coefficient decreases in size, as would be expected given the correlation of the variables with IPD and with international trade (e.g., two countries that share a common language are also more likely to be geopolitically aligned and trade more). All of the control variables have the expected size, and their sizes are comparable to those in the gravity literature. In column (3), we add two time-variant variables: economic integration agreements and common currency. While we find that the first variable helps explain trade (as expected), the latter does not. Surprisingly, IPD loses its statistical significance, maybe driven by omitted variables. In column (4), we add country-pair fixed effects to control for all time-invariant country-pair heterogeneity, implying that all other bilateral time-invariant variables have to be dropped. In this case, we find that IPD regains its statistical significance at the 1% level. Here, a one-point increase in IPD is associated with a reduction of the share of bilateral trade over the country’s total imports by approximately 7% ¹¹ . ¹² Now, the importance of a common currency in promoting trade also gains importance, as economic integration agreements (deep trade agreements – DTAs) also remain as an enabling factor for trade. The estimated coefficient of DTA is now smaller in size compared to column (3), but this is in line with previous results in the literature and expected, since we only consider time variation within country pairs. We conduct the Ramsey RESET test to evaluate the appropriateness of the specifications of the models used for columns (3) and (4). As expected, the test statistics suggest that the model used in column (4) is more adequate.

In table A1 in the annex, we assess the robustness of our main specification, i.e. that of column (4) from table 2. Column (1) is for reference only. In column (2), we use the S-score as an alternative index of geopolitical distance. While the coefficient is not comparable in size, the sign and the statistical significance remain the same. In column (3), we calculate the trade share based on the total exports of the exporting country for each year, rather than based on the imports of each importing country for each year. The results remain qualitatively the same. Moreover, in columns (4) and (5), we exclude both China and the United States from the sample. ¹³ Here, too, we find that the coefficients remain barely unchanged compared to those of column (1). Finally, column (6) shows that excluding zero values for the trade share variable has little effect on the results and only slightly reduces the coefficients. ¹⁴

In terms of sign and size, our results are broadly aligned with the literature. Bosone and Stamato (2024) also find that larger IPD relates to lower trade (measured as total values in their analysis). Gopinath et al. (2025) use the Russian invasion of Ukraine as an event affecting geopolitics and find lower trade after the invasion between different blocs of countries. Our study improves on the past literature by using a wide array of countries, a longer time period and a rich heterogeneity analysis which follows.

Table 2

3.1 Heterogeneity analysis

Globalization, here measured as the movement of goods across borders, has been described as having different phases, or as evolving with time (see Aiyar et al. 2023a). We want to investigate this further and analyze the relationship between IPD and different time periods. Column (5) of table 2 shows that the importance of IPD as a deterring factor for trade increases over time measured in five-year periods ¹⁵ . An alternative approach to investigating the relationship between IPD and time is to interact IPD with year dummies, in a similar approach as before when we interacted IPD with the five-year periods. To make the results more illustrative, we plot the point estimates of the interaction of IPD with each year dummy, as well as with their respective confidence intervals (chart 2). Here, too, we observe overall a rather stable negative and small relationship for the initial years of the sample, but then a descending trend surfaces: The more time elapses, the more geopolitical distance diminishes trade between country pairs. In the annex, chart A2 adds to the results of chart 2 by including the trade shares with respect to the exporter and by replicating the results with trade in levels. As can be seen, both estimates overlap throughout the time frame, supporting the robustness of our results. We posit that the results show that trade could be increasingly used as a punishing mechanism for political differences (see, e.g., Fuchs and Klann (2013) who documented how China reduced imports from countries who received the Dalai Lama as a visitor).

Chart 2

An alternative approach to investigating the heterogeneity analysis is to look at the country groups by origin/destination, by types of country pairs and by country blocs. Chart 3 shows the time-IPD interaction plotted for two different types of economies as exporters – EMDEs and AEs. While we see that for EMDEs the estimated coefficient of IPD is always negative, there is an unexpected positive pattern at the beginning for AEs (which is also found for FDI flows in the complementary analysis by Aiyar et al. (2024)). Subsequently, a non-significant period follows, and finally, a negative period starting from 2015 onward. In these last years, there is an overlap with the estimate for EMDEs. The corollaries for world trade of changing geopolitical distance have become even more negative with the passing of time. We argue that geopolitical distance plays a critical role in shaping trade relationships due to the inherent limitations of complete contracting (Nunn, 2007) and the difficulties associated with contract enforcement. In such contexts, trust becomes a key factor (Guiso et al., 2009), often rooted in perceived friendship or political alignment between countries. If we think of the EMDE countries, these are normally characterized by a weaker institutional environment, and therefore the negative correlation between IPD and trade is stronger (in absolute value) for these countries throughout the sample period. We see the same pattern for destination country groups in chart 4. Why would these geopolitical differences matter more broadly for the economy? Fernández-Villaverde et al. (2024) argue that these are of ultimate importance, since trade disruptions bring inefficiencies in the form of reduced specialization, decreased competition and resource misallocation. The authors reinforce their statement with the argument from Aiyar et al. (2023a) about how difficult it is to replace (imported) goods in the short run (low elasticity of substitution). Our results support those obtained by Fernández-Villaverde et al. (2024) and Hakobyan et al. (2023) who also find that higher fragmentation is especially harmful for emerging economies in terms of their international trade.

Chart 3

Chart 4

For chart 5, we split the origin and destination countries into four groups based on the AE and EMDE classification. We observe that a higher IPD always deters trade between EMDE countries, which is as expected and aligned with what we also observed in charts 3 and 4. On the contrary, for trade between advanced economies IPD does not seem to really matter. It might therefore prove more interesting to see what happens when we consider trade between different country groups. For trade from AEs to EMDEs, we observe that IPD starts having a negative and statistically significant effect (increasing in absolute value with the passing of time) only after the year 2016 (see top right panel of chart 5). For trade from EMDEs to AEs, this is the case from 2019 onward, with the negative association being stronger (see bottom left panel of chart 5). This period coincided with stock market volatility, which has put higher pressure on emerging economies, with the US-China trade war in 2018, which has increased uncertainty, and with trade tensions and deglobalization trends like Brexit.

Chart 5

Finally, we observe the dynamics between IPD and trade across country blocs, based on their geopolitical distance to the US and China, respectively. More precisely, we define the blocs as countries that are positioned within the closest quartile in terms of IPD either to the US or China for every year. The remaining countries are defined as non-aligned countries. ¹⁶ First, we look at trade between countries belonging to the same blocs. IPD mostly does not seem to matter for trade within blocs – in fact, belonging to the same bloc seems to act as a “trust mechanism” where small differences do not seem to play a role. Between non-aligned countries, changes in IPD do not seem to change bilateral trade during the beginning and end of our sample period. However, in the middle period, they seem to do – one should also not forget that these were years of increased economic uncertainty due to the financial crisis. For trade within the US bloc, we see that after the financial crisis increases in IPD do deter trade. We postulate that, unlike in the case of the China bloc, belonging to the same bloc does not necessarily provide a “trust mechanism.” While China has a tight relationship with partners in terms of trade, FDI (which are interrelated, see Abeliansky and Martínez-Zarzoso (2019)) and migration, this is not the case for the US. Also, the US is a more closed economy, while China relies more on its internationalization as a growth strategy. Non-aligned countries as importers mostly do not seem to be affected by IPD (see middle column of chart 6). Non-aligned exporters only seem to be influenced by larger IPD when trading with the US bloc (see middle row in chart 6). Interestingly, the negative relationship between trade and IPD between the US and China blocs (China–US; US–China) seems to increase with time, being insignificant at the beginning of the period. Thus, the US bloc seems to be most sensitive to IPD as an importer, while this is the case for the Chinese bloc only when trading with the US plus allies as a counterpart.

Chart 6