Polarizing and equalizing trends in international trade and Sustainable Development Goals

Indicator framework for SDGs

The UN SDGs are nationally focussed, thus countries determine how SDGs are incorporated into their planning processes and development strategies⁵⁶. The 193 signatories to the UN 2030 Agenda provide data on their SDG progress to the UN annually⁵⁷. Guidelines developed by the UN provide detail on what structure the SDG reports could take and what data are required⁵⁸. The regional UN offices provide region-specific information and resources regarding SDGs and their indicators. Despite efforts to improve and support national data collection and SDG reporting, many Global South countries still have limited resources for data collection and monitoring processes. The UN global indicator framework for the SDGs and targets of the 2030 Agenda for Sustainable Development provide a measurable way to progress towards each target of the SDGs⁵⁹.

Given the assumption that national efforts collectively add up to global progress on the SDGs, spillover effects (that is, impacts taking place outside countries’ borders) are not explicitly covered in the 2030 Agenda. Thus, nations are not required to report how their consumption patterns influence other countries’ SDG progress. However, recognizing the growing literature on burden shifting via global trade⁶, the Sustainable Development Solutions Network has developed a Spillover Index to shed light on how SDG progress in one region might hinder it in another⁶⁰. These spillover effects were not considered in the conceptualization of the SDGs and the 2030 Agenda⁶¹, thus a perfect alignment does not exist between the indicators often used for assessing spillover effects and those proposed to measure progress towards the SDGs. In this vein, this study develops consumption-based proxy quantities (‘proxies’) linked to specific SDG targets or different elements of the SDGs for undertaking a quantitative assessment to analyse the influence of international trade on polarizing and equalizing trends in the Global North and Global South.

Connecting SDGs to quantitative proxies

We express SDG performance using quantitative proxies that cover the environmental and social dimensions of sustainability (Supplementary Information 2.3). These proxies are related to several SDG targets; however, the majority of these (except for ‘material footprint’) are not part of the official global indicator framework for the SDGs⁵⁹ for reasons explained above. Hence, the selected environmental and social proxies are linked to 12 SDGs as representative consumption-based proxy quantities to capture consumption-based effects that are not otherwise considered in the SDG framework. Understandably, the proxies used in this study do not capture all the different dimensions encompassed by each of the SDGs (described by all their targets), however by being linked to specific single targets of the selected SDGs, our proxies show the extent to which international trade has contributed or not to some aspects of the SDGs in question.

In addition, an assessment of temporal change in outsourcing requires data on the selected proxies at a detailed country and sector level, hence their selection is determined by the choice of the economic global trade database used in the study. Thus, the selected proxies are linked to the GLORIA multi-region input–output (MRIO) framework¹⁷, with data on environmental proxies taken from the Sustainable Consumption and Production Hotspot Analysis Tool (SCP-HAT)¹¹ and on social proxies from a range of sources, as described below. When linked to the MRIO framework, the environmental and social proxies are called satellite accounts.

Outsourcing trends as assessed in this study for environmental and social proxy quantities can be understood as being associated with international trade⁶. For example, petroleum refining for meeting foreign demand of oil directly results in environmental repercussions, such as greenhouse gas (GHG) emissions—this is an example of the environmental effects of trade-related connections between producing and consuming countries. The connection between trade and social effects (for example, poverty, occupational accidents) is less direct due to inherent human dimensions. For example, poverty is a multi-faceted issue that depends on income, education, health, threat of violence⁶² and much more. MRIO analysis enables the tracking of production and consumption of commodities linked with workers below the poverty threshold; thus, connecting the demand for products and services with production in countries with a poorly paid population. We use environmental and social proxies to analyse SDG performance in terms of polarizing or equalizing outsourcing trends. Uncovering link of consumption with GHG emissions, water stress, land-use and biodiversity threats (which take place during the production of consumed products); and the link of consumption with the prevalence of poverty and occupational accidents contributes to the understanding of underlying relationships between production and consumption at a global level and also in understanding (using a time-series assessment) polarizing and equalizing effects of international trade on the SDGs.

MRIO analysis

MRIO analysis is based on statistical data that capture connections between sectors and regions in a global economy. Originally conceived by the Nobel Prize Laureate Wassily Leontief⁶³, the technique has been widely used for assessing environmental and social repercussions of international trade⁶. To date, multiple MRIO databases have been developed for assessments at local, national and global scales^8,9,10; for analysis of disasters⁶⁴; and for underscoring the use of the MRIO technique in implementing an SDG reporting and monitoring system¹⁷. Here we demonstrate the power of MRIO analysis in appraising temporal performance of countries over time in relation to the SDGs.

A MRIO system is represented by a set of matrices \(T^\,(P I)\times (R J)\times T\), \(y^(R J)\times S\times T,\,Q^Q\times (P I)\times T\) with data for \(q=1,\ldots ,Q\) primary factors (such as labour, resources or pollution), \(p=1,\ldots ,P\) producing (that is, extracting, employing, emitting) regions, \(i=1,\ldots ,I\) producing (extracting, emitting) industries, \(r=1,\ldots ,R\) selling (processing, manufacturing, trading) regions, \(j=1,\ldots ,J\) sold (processed, manufactured) products and \(s=1,\ldots ,S\) buying (consuming) regions, all for \(t=1,\ldots ,T\) years. T is an intermediate transactions matrix, y is final demand and Q is a so-called satellite account showing the regions’ inventories in terms of the primary factors relating to the SDG proxies. There are two important derived quantities: the matrix \(q^Q\times (P I)\times T\) holds proxy intensities with \(q=Q\widehat\left(T\,1+y1\right)^l-1\) and \(L^(P I)\times (R J)\times T\) is the famous Leontief inverse with \(L={\left[I-T{\widehat{\left(T\,1+y1\right)}}^-1\right]}^-1\). 1 is a suitable row summation operator, and the hat (^) accent denotes vector diagonalization. The footprint of proxy quantities contained in satellites Q are defined by their most general tensor form as \(F_ij,t^\;q,prs=\left\qLy\right\_ij,t^q,prs=q_i,t^q,pL_ij,t^pry_j,t^rs\). Given Leontief’s national accounting identity \(\sum _ij^rsF_ij,t^\;q,prs=\sum _i^pQ_i,t^q,p\forall q,t\), the quantities Q and F constitute dual perspectives on resource use (environmental and human) and pollution, one called the territorial or producers’ view, and the other the supply-chain or consumers’ view.⁶⁵

In line with the underlying MRIO database—GLORIA—the satellite accounts Q were established for multiple environmental and social proxies at a detail of 97 sectors in 164 countries.

Definitions and data sources for proxies

In the following we describe the 12 selected environmental and social proxies linked to the SDGs. For ease of interpretation, we define the proxies such that they are considered bad/detrimental for the accomplishment of the SDGs. In other words, if a downward trend is observed over the assessed time period for the proxies, then that can be interpreted as a desirable outcome. The selected proxies do not serve to comprehensively cover all targets and indicators in the 12 selected SDGs. Instead, the proxies are designed to link to relevant aspects of selected SDGs for assessing polarizing and equalizing international trade trends over time. Briefly, the proxies are (Supplementary Information 2.3):

• GHG emissions: we source data from the EDGAR v.5.0 database^66,67 to characterize region- and sector-specific anthropogenic GHG emissions. Construction of this proxy was done by implementing a ‘satellite account’ linked with the MRIO database, which required total sectoral output as additional data for allocating emissions to the MRIO database (supplementary information of Lenzen et al.⁶⁸ for details). This proxy is presented in the units of tonnes and has been linked to SDG 13 (Climate Action), target 13.2, indicator 13.2.2.

• Air pollutants: we cover three main air pollutants: sulfur dioxide (SO₂), particulate matter (PM_2.5) and nitrogen oxides (NO_x). This proxy is developed by applying characterization factors to emissions data from EDGAR v.5.0 database^66,67, expressed in disability-adjusted life years. This proxy is presented in the units of tonnes and has been linked to SDG 11 (Sustainable Cities and Communities), target 11.6, indicator 11.6.2

• Land: we cover six land-use classes: annual crops, permanent crops, pasture, extensive forestry, intensive forestry and urban, based on data from the Food and Agriculture Organization Corporate Statistical (FAOSTAT) Land Use domain. We map these land-use classes to the GLORIA MRIO table following the approach outlined in Annex VIII of the Sustainable Consumption and Production Hotspots Analysis Tool (SCP-HAT)–technical documentation⁶⁹. This proxy is presented in the units of hectares. This proxy has been linked to SDG 2 (Zero Hunger), considering that its target 2.4 argues for the need to ensure sustainable food production systems that increase productivity.

• Biodiversity: this proxy builds on the land-use proxy by applying characterization factors for global species loss on land-use categories to yield a measure of potentially disappeared fraction of species (that is, temporary loss of biodiversity from land occupation, refer to land-use-specific UN Environment Programme global guidance for Life Cycle Impact Assessment indicators⁷⁰ for characterization of biodiversity impacts from land use). This proxy is presented in the units of potentially disappeared fraction of species. This proxy has been linked to SDG 15 (Life on Land), target 15.5.

• Energy: we source data on primary energy production from the International Energy Agency⁷¹ to develop the energy proxy, which includes primary energy production from 21 energy products that are grouped into six broad groups: coal and peat; oil and natural gas; nuclear; solid biofuels; captured energy and heat. Annex XI in the SCP-HAT–technical documentation⁶⁹ describes the allocation of these sources to MRIO table. This proxy is presented in the units of joules. This proxy has been linked to SDG 7 (Affordable and Clean Energy), target 7.2.

• Materials: we source data on materials from the UN International Resource Panel Global Material Flows Database⁷², which presents direct material flows for 124 different material categories that can be categorized into four broad materials groups: biomass, metal ores, non-metallic minerals and fossil fuels¹⁷. This proxy is presented in the units of tonnes. This proxy has been linked to SDG 12 (Responsible Consumption and Production) in view of the material footprint being a UN proposed indicator for target 12.2 (that is, indicator 12.2.1) of SDG 12.

• Water stress: we compile this proxy based on data on water use⁷³ combined with the AWARE characterization factor⁷⁴. Development of this proxy is explained in detail in the SCP-HAT’s technical documentation⁶⁹. This proxy is presented in the units of m3 H₂O-equivalent. This proxy has been linked to SDG 6 (Clean Water and Sanitation) under the consideration that water stress is proposed by the UN as an indicator (indicator 6.4.2) for this goal’s target 6.4.

• Nitrogen emissions: we cover three categories of nitrogen emissions: nitrogen oxides (NO_x) and ammonia (NH₃) to air and nitrogen (N) leached to freshwater. We characterize nitrogen flows to water based on the FAOSTAT’s Climate Change Emissions domain⁷⁵. We then capture airborne nitrogen emissions based on the GHG emission and air pollution dataset as described above. We obtain nitrogen leaching information from the FAOSTAT Climate Change Emissions domain⁷⁵, encompassing data on nitrogen utilization and leaching to water. This proxy is presented in the units of tonnes. This proxy has been linked to SDG 14, target 14.1, indicator 14.1.1 (Life Below Water), given that nitrogen is a nutrient associated with eutrophication.

• Women’s participation: we distinguish employment by gender (male and female employment) based on data from the International Labour Organisation (ILO)⁷⁶, expressed in units of number of people. The ILO data feature more aggregated sectors than the GLORIA MRIO database, which are disaggregated using data on compensation to employees. The raw data for this proxy are then converted into a ratio by applying a weight of 0 to female employment and 1 to male employment, followed by normalization as described in Supplementary Information 2.3. Thus, this resultant proxy corresponds to women’s participation in the workforce, and it is expressed in the units of percentage of males employed. A negative trend where the percentage of males decreases over time is considered a positive outcome for this indicator. This proxy has been linked to SDG 5 (Gender Equality), as target 5.1 calls for the end of discrimination against women.

• Labour skills: we distinguish different levels of skills (high-, medium- and low-skilled) based on data from the ILO⁷⁶. Allocation of these data to the MRIO database follow the same approach as the women’s participation proxy described above, using data on compensation of employees (Supplementary Information 2.3 for details). This resultant proxy ‘labour skills’ is expressed in the units of percentage of low-skilled workforce and has been linked to SDG 10 (Reduced Inequalities), given that SDG 10’s target 10.1 argues for the income growth of the lower population percentiles. Because this proxy is based on employees’ compensation, a negative trend where the percentage of low-skilled workforce decreases over time is considered a positive outcome for this indicator.

• Poverty: we construct this proxy by considering individuals living below the poverty line (US$1.90 per person per day), with the daily income per person estimated as the ratio of total income to employment⁷⁶. The income data are obtained from the value-added block of the GLORIA database. This proxy has been linked to SDG 1 (No Poverty) in units of number of people, considering that its target 1.1 seeks to eradicate extreme poverty (people living below the international poverty line).

• Occupational accidents: this proxy covers fatal and non-fatal accidents⁷⁷ in connection with work. This proxy is presented in the units of cases and has been linked to SDG 8 (Decent Work and Economic Growth), taking into account that fatal and non-fatal injuries is the indicator (8.8.1) proposed by the UN to measure progress towards SDG 8’s target 8.8.

Outsourcing

Whereas the producers’ perspective reports primary factors of labour, resource extraction and emissions as attributes of the producing region, the consumers’ perspective re-allocates these to whoever ultimately consumes the product for which these factors have initially been expended. If a regions’ footprint (F, consumer perspective) is larger than its factor inventory (Q, producer perspective), the region is said to be a net importer of this factor and vice versa. A net-importing region is said to be outsourcing factor use to net-exporting regions, and the commodities it imports are said to embody the factor use while being traded.

The phenomenon of outsourcing is well known and has assumed various other connotations such as quantities that leak (for example, for carbon⁷⁸) or that are virtual (for example, for water⁷⁹). This work is concerned with whether and how outsourcing has changed over time, both in its direction and magnitude.

On the basis of the quantifications Q and F of the producer and consumer perspectives, we define outsourcing as a matrix \(\barS^Q\times R\times T\), with \(\barS_t^q,p=\sum _iQ_i,t^q,p-\sum _ij^srF_ij,t^\;q,srp=\barQ_t^q,p-\barF_t^\;q,p\), or in matrix notation \(\barS=\barQ-\barF\), with the bar signifying summation over products. Note the reversal of the regional indices for the footprint tensor; in the summation above, p is taken as the producing and consuming region. \(\barS_t^q,p > 0\) indicates that in year t, region p is a net exporter of primary factor q and vice versa a net importer for \(\barS_t^q,p < 0\). We then normalize outsourcing so that we can compare across indicators and countries: we define the outsourced fraction \(\bars=\left(\barQ-\barF\right)\oslash \barF\), where \(\oslash \) is element-wise division. This quantity describes what fraction of the regional footprint is outsourced (as opposed to of domestic origin). Similarly, \(\bars_t^q,p > 0\) indicates that in year t, region p is a net exporter of primary factor q and vice versa a net importer for \(\bars_t^q,p < 0\).

In our results, we use outsourcing averaged over time (\(\overline\overlineS\), denoted by a double bar), defined as \(\overline\overlineS=\overline\overlineQ-\overline\overlineF\), with \(\overline\overlineQ^\;q,p=\sum _t\barQ_t^q,p=\sum _itQ_i,t^q,p\) and with \(\overline\overlineF^\,q,p=\sum _t\barF_t^\;q,p=\sum _itF_i,t^\;q,p\). Similarly, we define \(\overline\overlines=(\overline{\overline{Q}}-\overline\overlineF)\oslash \overline\overlineF\) as the (time-)average outsourced fraction. As with the outsourced fraction, this quantity describes what fraction of the regional footprint is outsourced (as opposed to of domestic origin), but appraised over the entire study period.

Underlying data for constructing MRIO tables, such as GLORIA, vary in quality and resolution by country. Standard deviation estimates are published for GLORIA MRIO tables, which we use for performing an uncertainty assessment, as described in Supplementary Information 4.

Temporal change of outsourcing

As explained above, changes of outsourcing patterns over time are at the heart of this work. We plot outsourcing trends against net trade status and define four archetypes:

1. i.

   Quadrant III: \(\overline\overlines^\,q,p < 0\) and \(\bars^\dotq,p < 0\); regions in this quadrant are (on average) net importers that have been boosting their net-importer status over time.

2. ii.

   Quadrant II: \(\overline\overlines^\,q,p > 0\) and \(\bars^\dotq,p > 0\); regions in this quadrant are (on average) net exporters that have been boosting their net-exporter status over time.

3. iii.

   Quadrant I: \(\overline\overlines^\,q,p < 0\) and \(\bars^\dotq,p > 0\); regions in this quadrant are (on average) net importers that have been diminishing their net-importer status over time (or have even become net exporters).

4. iv.

   Quadrant IV: \(\overline\overlines^\,q,p > 0\) and \(\bars^\dotq,p < 0\); regions in this quadrant are (on average) net exporters that have been diminishing their net-exporter status over time (or have even become net importers).

Arranging these archetypes in a four-quadrant system allows us to identify situations where trends work to either further aggravate historical outsourcing (polarizing trends; quadrants II and III) or further alleviate historical outsourcing (equalizing trends; quadrants I and IV). Polarizing trends deteriorate existing disparities; equalizing trends mitigate them.

To this end, we regress \(\barS_t^q,p=\barS^q,p\left(t\right)=m^q,pt+\barS_t=0^q,p\forall p\) by following the weighted least squares approach by taking country footprints as weights. The regression slope m_r is the average temporal trend \(m^q,p=\frac\partial \barS^q,p(t)\partial t\) of the outsourcing of region p in terms of primary factor q. We use \(m^q,p\) to determine the temporal change of the outsourced fraction \(\bars\), calculated as \(\bars^\dotq,p=\frac\partial \bars^q,p(t)\partial t=\frac{\partial [\barS^q,p(t)/\barF_t^\;q,p]}\partial t\approx \frac{\partial \barS^q,p(t)/\partial t}{\sum _t\barF_t^\;q,p/T}=\fracm^q,p{{\overline\overlineF}^\;p/T}\), with the dot (·) accent denoting the temporal derivative. Regions with \(m^q,p < 0\) have been increasingly outsourcing and vice versa.

Finally, regressing a regional four-quadrant cloud of outsourced fractions \(\overline\overlines\) and their temporal changes \(\dot\bars\) yields four-quadrant slopes \(b^q=\partial \dot{\bars}/\partial /\overline{\overline{{{s}}}}\); Fig. 1). Regions in quadrants II and III are part of a polarizing trend (\(b^q > 0\)) and regions in quadrants I and IV are part of an equalizing trend (\(b^q < 0\)).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.