Analysis of the international trade networks of COVID-19 medical products

This research aimed to gain a deeper understanding of how and for what reasons the world trade networks of medical products were reorganized during the novel coronavirus (COVID-19) pandemic. To do this, first, the trade data of eight COVID-19-related product categories (such as medical test kits and protective garments) were collected for the years 2019 and 2020. Then it was examined which countries’ exports and imports changed the most between the studied time period in each product category. In addition, gravity models containing additional economic, geographic, and COVID-19-related variables were used to analyze the impact of the pandemic on the investigated trade networks. Based on the results, China achieved the highest cumulative export growth, surpassing the second-highest value by approximately 14.66. Hungary, with a population of only 9.7 million, stood out as a major importer of ventilators. Additionally, a higher incidence of COVID-19 among importers typically led to reduced traded values, while European Union membership and innovation capacity had the opposite effect.

on world trade.Hayakawa and Imai (2022) examined the trade of medical products, including medicines, personal protective equipment, and health and medical equipment, and Grassia et al. (2022) tried to estimate the effects of restrictions on the export of these products.As Grassia et al. (2022) emphasized, this area of research not only helps to understand the causes and effects of restrictions on free trade but also sheds light on how individual countries react to a global crisis caused by the pandemic.
Research in this field often relies on the gravity equation of trade (Kabir et al. 2017), an econometric model that aims to explain bilateral trade flows between countries by taking into account factors such as economic size, distance, and other relevant variables.Some of these empirical works focus on a specific country, such as China (Fuchs et al. 2020;Liu et al. 2022;Pu et al. 2023) or Malaysia (Zainuddin et al. 2021).Others explore multiple economies (Jindřichovská and Uğurlu 2021) or, similar to our paper, the global economy (Hayakawa and Mukunoki 2021).
The aim of this research was to gain a deeper understanding of how and for what reasons the world trade networks of medical products were reorganized during the COVID-19 pandemic.To do this, first, the trade data of eight COVID-19-related product categories (such as medical test kits and protective garments) were collected for the years 2019 and 2020.Then it was examined which countries' exports and imports changed the most between the studied time period in each product category.Finally, gravity models containing additional economic, geographic, and COVID-19-related variables were used to analyze the impact of the pandemic on the investigated trade networks.
Based on the results, China exhibited the highest cumulative export growth, mainly due to a significant increase in protective garment exports, surpassing the secondhighest value by a factor of approximately 14.66.Despite its small population of 9.7 million, Hungary stood out as one of the main importers of ventilators.Additionally, gravity models showed that a higher incidence of COVID-19 among importers typically led to reduced traded values, while European Union (EU) membership and innovation capacity had the opposite effect.The main implications of the results can be summarized as follows: • The results emphasize China's pivotal role in global supply chains for medical products, specifically regarding protective garments, which have emerged as one of the most crucial product categories in the fight against the pandemic.• The import patterns of medical products were less concentrated, showing no clear asymmetry among the top importing countries.However, Hungary, despite its small population, stood out as a major importer by purchasing a substantial number of ventilators from China.• Regional trade agreements (RTAs) and EU membership of exporter countries positively influenced the traded value, while they had a negative effect on importers.This suggests that trade integration within regional blocs played a significant role in shaping trade dynamics during the pandemic.• The number of COVID-19 cases in exporting countries had a significant impact on the trade of protective garments, indicating that countries heavily affected by the pandemic experienced significant trade restrictions for this specific product category.• Future research should consider focusing on the ego networks of key exporters and importers could provide valuable insights into the broader effects of the pandemic on various economic sectors.In addition, long-term investigation and comparison of trade networks could help elucidate the lasting effects of the COVID-19 pandemic and reveal potential shifts and consolidations in global trade relations in the postpandemic era.
The rest of this paper is organized as follows.The "Data and methodology" section introduces the data employed in this study, as well as the applied methodology and software.
The "Results and discussion" section presents and discusses the results of the social network-and gravity model-based analysis.Finally, the "Conclusions and future work" section provides a summary and conclusions.

Data sources
In this paper, three databases are employed.The first is the source of international trade data related to medical products, while the economic, and geographic variables, as well as the COVID-19-related health data, are obtained from another two data sources.
Exported values from the above-mentioned medical products are aggregated by product category for 2019 and 2020.In the following, for a given product category, the total exported values from country i to the country j in 2019 and 2020 are denoted by EXP 2019 i,j and EXP 2020 i,j , respectively.

Extended Gravity database
The majority of the independent variables used in the analysis are obtained from the CEPII Gravity (Conte et al. 2022) database, which contains a wide range of uni-and bilateral variables related to international trade between 1948 and 2019. 3From this database, economic and geographic indicators are collected for the year 2019.According to Kurbucz (2023), the traded values of the investigated medical products are highly correlated with the innovation capacity of the countries; thus, we complement these data with data drawn from the Global Innovation Index (GII) (Soumitra et al. 2020), which aims to capture the multidimensional aspects of innovation.4Note that indicators with a higher absolute Pearson correlation coefficient than 0.8 are not included together in the final model.

COVID-19 database
Data on COVID-19 cases and deaths are collected from the WHO's COVID-19 Dashboard (COVID 2020). 5For each country, we use the latest data available from 2020.Since data on cases and deaths are highly correlated, only the former is applied during the analysis.

Applied dataset
The dataset compiled using the three data sources (see the Data sources section) is presented in Table 1.More information about the variables of the Gravity database can be found at http:// www.cepii.fr/ DATA_ DOWNL OAD/ gravi ty/ doc/ Gravi ty_ docum entat ion.pdf (accessed: 14 July 2023).

Multilevel network representation
The trading data is represented as a multilevel network (see,e.g., (Hammoud and Kramer 2020)).Multilevel networks include multiple layers that can contain a subset of all available nodes and edges.In our case, the eight product categories form eight layers, nodes are the countries, and directed edges represent their trading activities from the given product category.The weight of the edges within a product category is determined based on the difference between the values exported in 2020 and 2019 as follows: where EXP 2020 i,j is the aggregated exported value from country i to j in 2020, measured in thousand United States dollars (USD), i, j ∈ {1, 2, . . ., L} and L ∈ N.
Formally, the trading data is represented by a graph which is a tuple defined by the sets of nodes ( N ), edges ( E ), and eight layers ( S ) as follows: where K = 8 .The applied multilevel data structure is illustrated in Fig. 1. (1)

Table 1 Applied dataset
Remarks: † : Separate variable for each product category.‡ : Measured in logarithmic scale.Total export from country i to country j in 2020 (thousand USD) † (a) Total export from country i to country j in 2019 (thousand USD) † (a)

Centrality measures and other descriptive statistics
To measure the extent to which the products' total imported and exported values changed in each country between 2019 and 2020, the directed version of strength (i.e., weighted degree) centrality (see, e.g., (Yook et al. 2001;Barrat et al. 2004)) is applied.For each product category, the in-and out-strength centrality-measured in thousand USD-can be defined as follows: where L j=1 e j,i and L j=1 e i,j alone measure the in-and out-degree centrality of the node i, respectively.
In addition, to provide further details about the examined networks, we have also identified the strong and weak components of each layer.Strong components of a network are subsets of nodes where there is a directed path from any node to any other node within the subset, while weak components are subsets of nodes where there is a path (not necessarily directed) between any two nodes within the subset.

Regression model
The gravity equation of trade defines trade as a positive function of the attractive "mass" of two economies and a negative function of the distance between them (Lewer and Van den Berg 2008).Similarly to other researchers (see,e.g., (Johnston et al. 2015;Hussain et al. 2021;Fontagné et al. 2022)), we use additional variables to control for demographic, geographic, ethnic, linguistic, and economic conditions.For each product category, the applied regression model is as follows: where P = 8 and R = 9 are the numbers of uni-and bilateral independent variables (apart from log(EXP 2019 i,j ) and dist i,j ), respectively (see Table 1).The parameters, denoted by α , β , γ , and δ , are estimated by ordinary least squares (OLS) regression in which the ǫ i,j is the random error term.

Applied software
The statistical programming language R is applied to compile and analyze the dataset presented in Table 1.To generate the figures and calculate the strength centrality measures, MuxViz (version: 3.1) R package (De Domenico et al. 2015;De Domenico 2022) is applied.This package enables the visualization and analysis of interconnected multilayer networks.More information can be found at https:// github.com/ manli us/ muxViz (accessed: 14 July 2023). (3) e j,i w j,i ,

Changes in the trade of medical products
Import and export growth between 2019 and 2020 is examined using in-and outstrength centrality measures, supplemented with other descriptive statistics that provide further insights into the investigated networks.First, these descriptive statistics are presented in Table 2, and then Fig. 2 illustrates the changes of the trade networks on a map.  Figure 3 shows the twenty-five countries with the largest total import and export growth between 2019 and 2020 (i.e., the countries with the highest in-and out-strength centralities).Finally, the top five countries that have the largest import and export growth in each of the examined product categories are presented in Table 3 individually.According to Fig. 3 and Tables 2 and 3, the traded value of protective garments (category G), medical test kits (category A), and disinfectants and sterilization products (category B) increased the most.China (CHN) alone increased its exports of protective garments (category G) by 71, 362, 502 thousand USD, which is approximately 20.38 times higher than the second-highest export growth in this product category, which was recorded in Malaysia (MYS).For these products, imports increased the most in the United States (USA), United Kingdom (GBR), and Germany (DEU)-22, 224, 368, 9, 287, 062, and 7, 321, 810 thousand USD, respectively; however, as Fig. 2 shows, most countries in the world had significantly increased imports of protective garments by 2020.In categories A and B, China (CHN) and Germany (DEU) achieved the highest   export growth (6, 155, 374 and 3, 507, 406 thousand USD, respectively), while the highest import growths were recorded in the United States (USA) and Switzerland (CHE) (7,060,897 and 5,100,978 thousand USD,respectively).
The international trade of vehicles (category H) typically decreased during the investigated time period, and the related network was much more fragmented than that of other types of products (see Table 2).In this category, the highest export growth was only 67.694 thousand USD, which was reached by the United Arab Emirates (ARE).In the case of categories C to F, China (CHN) achieved the largest increase in exports by 3, 329, 381, 3, 708, 636, 1, 346, 074, and 3, 522, 334 thousand USD, respectively.While the highest import growth was recorded in the United States (USA) from categories C to E (1,157,565,794,627,1,089,646 thousand USD, respectively), despite its population of only 9.7 million, Hungary (HUN) achieved the largest increase in imports of oxygen therapy equipment and pulse oximeters (category F) by 670, 942 thousand USD.This import growth is related to the Hungarian government's procurement of more than 16, 000 ventilators from China (CHN). 6fter aggregating each product category (see Fig. 3), it can be seen that China (CHN) achieved the largest export growth by 90, 833, 770 thousand USD, which is approximately 14.66 times higher than the second-highest cumulative export growth (6, 194, 989 thousand USD), which was recorded in Belgium (BEL).In the case of aggregate import growth, there was no such degree of asymmetry between the top two countries, which were the United States (USA) and Germany (DEU) by 31, 853, 163 and 15, 268, 697 thousand USD, respectively.

Regression results
The results of the regression analysis are presented in Table 4.
Based on the results, it can be seen that the global F-test was highly significant for all models, while the adjusted R 2 values varied between 0.199 and 0.805.This difference in adjusted R 2 mainly reflects the stability of trade relations (network structure) between 2019 and 2020.Thus, e.g., the highest adjusted R 2 was observed for medical test kits (category A), in which the logarithm of the 2019 and 2020 trading values had the highest Pearson correlation coefficient ( ρ = 0.873 ).Regardless of product category, the sig- nificant values of RTA and the EU membership of the exporter countries increased the traded value, while the effect of the EU membership was shown to be the opposite on the importer's side.The coefficients of WTO membership were only slightly significant.While they were typically positive on the importer's side, the sign of the effect varied in the case of exporting countries.The GDP of both the exporter and importer countries increased the traded value, while the coefficient of the exporter countries was approximately 1.5-2.0times greater than the coefficient of the importer countries.
Cumulative COVID-19 cases were significant in only half of the categories and typically reduced trade value, regardless of whether they were measured for exporting or importing countries.An exception to this was the disinfectants and sterilization products (category C), in which the higher COVID-19 cases in exporter countries typically increased the traded value.Based on these coefficients, the COVID-19 cases recorded in the exporting countries reduced the traded value of protective garments (category G) the most.In addition, the GII score of the exporting countries significantly reduced the trading value only in this product category.This result reflects that the production of protective garments (category G) had significantly lower technological demands than other investigated product categories.Finally, the contiguity of countries increased the traded value the most in the case of disinfectants and sterilization products (category B).After we controlled for this factor, the trade of medical test kits (category A) was the least distance-dependent category, while the distance between trading countries reduced the trade value of other medical consumables (category C) and protective garments (category G) the most.

Conclusions and future work
In this work, we investigated how and for what reasons the world trade networks of medical products were reorganized during the COVID-19 pandemic.To do so, we first collected the trade data of eight COVID-19-related product categories for the years 2019 and 2020.Afterward, with the help of the strength centrality measure, we examined which countries' exports and imports by product category changed the most between the examined periods.Finally, we used gravity models containing additional economic, geographic, and COVID-19-related variables to analyze the impact of the pandemic on the investigated trade networks.
In line with the descriptive statistics, China achieved the largest cumulative export growth of 90, 833, 770 thousand USD, which is approximately 14.66 times higher than the second-highest cumulative export growth (6, 194, 989 thousand USD), which was recorded in Belgium.The dramatic increase in Chinese exports comes primarily from the sale of protective garments (71, 362, 502 thousand USD).In the case of aggregate import growth, there was no such degree of asymmetry between the top importer countries.In addition, with a population of only 9.7 million, Hungary stands out among the main importers of medical products by purchasing more than 16, 000 ventilators from China.
According to the results of the regression analysis, the structure of the trade network changed the least in the case of medical test kits.Regardless of product category, the significant values of RTA and the EU membership of the exporter countries increased the traded value, while this effect was shown to be the opposite on the importer's side.The trade of protective garments was the most sensitive to the number of COVID-19 cases in the exporting countries.Furthermore, the GII score of the exporting countries significantly reduced the trading value only in this product category.
Based on our results, further examination of the trade networks-e.g., by focusing on the ego networks of key exporters and importers-could reveal important details related to the effect of COVID-19 on the global economy in more areas than the case of medical products investigated herein.In addition, the longer-term investigation and comparison of trade networks could show not only the effect of the COVID-19 pandemic but also the consolidation of trade relations in the post-COVID-19 era.

Fig. 1
Fig.1Applied multilevel data structure.(Remarks: v i : node i, e i,j : edge from node i to j.Each e i,j edge has a w i,j weight.)

Fig. 2
Fig. 2 Import and export growth of different product categories between 2019 and 2020.(Remarks: node color: log(s in i ) , from white to red, node size: log(s out i ) , from small to big.)

Fig. 3
Fig. 3 In-and out-strength centralities of different product categories.(Remarks: Top 25 countries based on aggregated measurements.Weights are export differences between 2020 and 2019.Country codes can be found in the Appendix in TableA1.)

Table 2
Descriptive statistics

Table 3
Countries with the largest total import and export growth in each product category Remarks: Total import and export changes are defined by s in i and s out i , respectively.Both are measured in thousand USD.Country codes can be found in the Appendix in TableA1

Table 5
Country codes GEO Georgia