Will you take my (s)crap? Waste havens in the global plastic waste trade

Abstract

For two decades until 2017, more than half of the world’s traded plastic waste was exported to China. In that year, however, China announced a complete ban on imports of post-consumer plastic waste to take effect at the start of 2018. The ban forced many countries to seek new ways to deal with plastic waste, including new destinations for exports. These changes may have significant impact on global social welfare and the environment. In this study we first ask whether trade in plastic waste follows a waste haven pattern, shifting environmental burden from richer countries and those with better environmental regulations to poorer countries and those with with weaker regulations. Second, we evaluate how China’s import ban altered the plastic waste trade. Empirical analysis using a gravity model reveals that the plastic waste trade follows a waste haven pattern, and the ban exacerbated this relationship. Differences in per capita GDP drove bilateral trade both before and after the ban, and disparities in stringency of environmental regulations became influential following the ban. Given that post-ban import volumes far exceeded pre-ban volumes in many countries, these results raise two concerns. First, post-ban trade changes were seemingly driven by exporters’ demand for disposal services rather than importers’ demand for plastic waste, thereby increasing environmental burden in importing countries. Second, because countries with weak environmental regulations likely have poor waste management systems, this pattern of plastic waste redistribution may worsen the existing global plastic waste pollution crisis.

1 Introduction

The idea that the wealth of a nation is correlated both with willingness to pay for environmental amenities and with the quality of environmental regulation is widely accepted. For both reasons, less wealthy countries likely have a higher tolerance for waste disposal methods and direct disposal costs that are low in relation to those in wealthy countries, but which may present or augment significant environmental threats. It is also possible that waste may be traded or smuggled among developing countries.¹ In either case, if a larger portion of waste is being disposed of in a weak environmental regulation regime, the probability that such waste ends up as pollution in the global ecosystem is higher. This problem is especially salient for the case of long-lived materials, notably plastics (Kershaw & Rochman, 2015).

The direct motivation for this study comes from a major policy shock affecting international trade in plastic waste. Between 1992 and 2016, China imported over half of the world’s traded waste products (Brooks et al., 2018). Between 2008 and 2016, China and Hong Kong imported over 70% of the global total on average. But in July 2017, China notified the World Trade Organization (WTO) of its intention to ban imports of 24 categories of post-consumer recyclable waste by year’s end, including post-consumer plastic waste (Igini, 2022). Despite some doubts prior to implementation, the ban proved to be effective, and the quantity of plastic waste imported annually by China fell by 99% in 2018 (Staub, 2017b, 2019).

China’s abrupt exit threw the global recycling industry into turmoil.² Countries whose recycling systems depended on China’s willingness to import their scraps³ began to struggle with ever-growing piles of recyclable waste that had nowhere to go.⁴ For lack of alternative outlets, more recyclable waste entered landfills and incineration facilities; recyclable prices dropped as the cost of recycling rose; the recycling industry—especially those companies that sort and sell scrap collected in municipal recycling bins—struggled with reduced profits. In the United States, cities and counties began to scale back the range of items accepted for recycling (Paben, 2019).

Developing countries were severely affected the ban as waste exporters sought alternative destinations. Malaysia, Vietnam, Thailand, and Turkey, among others, experienced a sharp spike in waste imports immediately after the ban was announced. Thailand, for example, was a net exporter of plastic waste before the ban but became a net importer after the ban and saw plastic waste imports from the United States increase by almost 2000% towards the end of 2017, relative to imports in the first half of the same year (Parker, 2018).

This additional trade volume undoubtedly exceeded domestic recycling capacity, so the surge in trade generates both economic and environmental puzzles. It implies that after the ban, not all “recyclable” waste was being traded for the purpose of recycling. Just as the US increased dumping of its own recyclable wastes in the wake of the ban, the new destination countries almost undoubtedly increased their own disposal of recyclable waste due to recycling capacity constraints and lags in the creation of new capacity, despite adopting increasingly restrictive policies (Kojima, 2020). International efforts to reduce and regulate trade in plastic waste were formalized in 2019 as a set of amendments to the Basel Convention⁵; however, it is still too early to know whether these efforts will be effective (Benson & Mortensen, 2021).

Against this background, our study has two empirical goals. The first is to investigate if global plastic waste trade follows a waste haven pattern (Kellenberg, 2012) in which trade shifts the environmental costs of plastic waste from richer to poorer countries. The second is to evaluate whether China’s import ban caused significant change in the country composition of trade in the plastic waste market. The market’s adjustment to this shock likely has significant implications for global social welfare and the environment.

Using a gravity model, we find that post-ban trade diversion in the global market for plastic waste is dominated by increased flows from higher-income to lower-income countries, and from countries with stronger environmental regulations to those where such regulations are weaker. We find further that this trade pattern deepened after the ban, and that bilateral differences in environmental policy regimes were prominent drivers.

Specifically, our estimates show that prior to the China import ban, for every 1% that an importer’s GDP per capita falls below that of an exporting partner, the importer will experience a 0.02% increase in waste imports from the exporter. As for regulatory quality, we find that disparity in the Environmental Performance Index (EPI) scores of exporting and importing countries has a weak negative relationship in the pre-ban period. In the post-ban period, however, our preferred estimates indicate that for every 1% that an importer’s EPI score decreases relative that of an exporting partner, the importer will experience a 0.2–0.3% increase in waste imports from the exporter.

To our knowledge, this is the first empirical analysis to ask specifically whether international plastic waste trade follows a waste haven pattern, and the first to address the effect of the China’s 2018 ban on trade in plastic waste. We base these empirical tests upon a micro-theoretic foundation that rationalizes the actions of entities at both the origin and the end point of global plastic waste trade. The remainder of this paper develops as follows: the second section is a literature review, the third section develops the theoretical framework, the fourth section describes the data, the fifth section reports estimation and results, and the last section concludes.

2 Literature review

Empirically, this paper is closely related to two recent studies. Balkevicius et al. (2020) use trade data from 2010–15 to estimate the effect of China’s 2013 “Operation Green Fence” policy on global trade in non-hazardous waste. Thakur (2022) presents a structural gravity model which evaluates the relationship between country income and the import of high-value versus low-value waste and explores how trade in these responds to trade barriers. To evaluate the impact of China’s 2018 ban, Thakur presents counterfactual outcomes using estimates from a gravity model using cross-sectional data from 2015.

Conceptually, the idea that dirty industries and activities may be moved among countries to minimize cost is widely accepted. But whereas the familiar pollution haven hypothesis refers to the movement of industries to locations where process-related pollution is less costly, the waste haven hypothesis (Kellenberg, 2012) posits that international trade allows for waste to be redistributed from its country of origin to foreign destinations, thereby transferring the costs and negative externalities of waste disposal across international borders. Empirical analyses of trade data consistently find that developing countries serve as waste havens for developed countries. Kellenberg (2015), for example, uses UN Comtrade data to show that, between 1992 and 2011, the volume of global trade in waste commodities increased by roughly 500%, from 45.6 million to 222.6 million tons. Over the same period, the share of waste imported by developing countries grew by more than 40%, while the share imported by developed countries declined.

A growing literature confirms the existence of waste havens (Matsuda et al., 2021; Balkevicius et al., 2020; Kumamaru & Takeuchi, 2021). However, work on identifying the mechanisms that dictate international waste flows remains inconclusive, in part because waste commodities have some special characteristics that set them apart from general traded goods. The literature addresses variables that determine the international trade of commodities in general, notably country and bilateral characteristics that are the default variables in a parsimonious gravity analysis, including economic size (GDP) and trade cost (distance and non-distance) variables. It also addresses variables that are suggested by the pollution haven hypothesis as determinants of international trade and investment flows that enable cross-border movements of goods associated with negative environmental externalities. These are, notably, income (GDP per capita) and environmental regulation quality. Finally, a smaller number of studies consider the specific characteristic of recyclable materials, including plastics. Because they can be either industrial feedstock or simply trash, they have a “double identity” that plays a potentially big role in trade and is thus important to understand.

2.1 Gravity variables

Scrap products are traded at a price just like general market goods. For this reason, factors that determine the trade of general commodities are relevant to the trade of scrap products. The gravity model, the tool of empirical trade studies, has been applied to trade in waste products (Baggs, 2009; Kellenberg, 2012; Kellenberg & Levinson, 2014; Higashida & Managi, 2014; Okubo et al., 2016). While these studies display many variants, they collectively predict that the two major factors that influence the volume of bilateral trade are the economic size of each country and cost of trade between them.

Larger economic size encourages trade due to a greater capacity to consume and produce, while higher costs discourage trade. These relationships have been observed for trade in both hazardous and non-hazardous waste products (Anderson & Van Wincoop, 2003; Baggs, 2009). As with conventional gravity model studies, non-distance determinants of trade costs, such as common language, common colonial ties, and joint membership of international trade agreements are also found to be important. Kellenberg and Levinson (2014), for example find that factors that indicate similarity between trading partners are also associated with higher waste trade volume.

2.2 Waste haven variables

As its name suggests, “waste haven” can be interpreted as a type of pollution haven. The pollution haven hypothesis (PHH) predicts that pollution-intensive industries will relocate from countries with higher income and stricter environmental regulations to countries with lower income and less stringent environmental regulation (Taylor, 2004). Instead of dirty industry relocation, the waste haven hypothesis asserts that differences in income and the quality of domestic environmental regulation drive the transboundary movement of goods associated with negative externalities. In general, empirical studies find that waste trade tends to flow from wealthier countries with stricter environmental regulations to poorer countries with weaker environmental regulations (Balkevicius et al., 2020; Kellenberg & Levinson, 2014).

The two waste haven determinants—wealth of a country and environmental regulation quality—are closely related.

First, institutions and regulatory quality can play a large role in bringing about economic growth (North, 2016; Kaidi et al., 2019). Second, the environmental Kuznets curve (Grossman & Krueger, 1991) suggests that the relationship between GDP per capita and environmental degradation follows an inverse U-shape. A poor country may be willing to accept environmental degradation in exchange for economic growth up to a threshold beyond which willingness to degrade the environment starts to fall. Thus, some studies have used GDP per capita as a proxy for stringency (Baggs, 2009).

Fikru (2012) explores the relationship between regulation and hazardous waste trade within the EU. Using facility-level data, she finds that countries with a greater number of hazardous waste regulations and higher hazardous waste tax rates have a higher propensity to export waste to countries with fewer regulations and lower tax rates.

For global trade in non-hazardous waste, Kellenberg (2012) uses the Global Competitiveness Report (GCR) of 2003–2004 to compute an environmental regulation “gradient” index which measures the difference in the stringency of environmental regulation for each exporter-importer pair. A positive value indicates that the importing country in any pair has relatively weaker environmental regulation. According to the PHH, this should encourage waste flow into the importing country. Results from this cross-section analysis suggest that all else equal, for every 1% that an importer’s environmental regulation quality falls below that of an exporter, the importer will experience 0.22% higher waste imports from that exporter.

A study by Okubo et al. (2016) also uses GCR data to measure the impact of the gap in environmental regulation stringency on the volume of recyclable waste exports from Japan. This study distinguishes three types of GCR scores based on overall regulation, toxic waste regulation, and air pollution regulation. It finds that the bilateral export volume from Japan increases with the gap in all types of GCR scores between Japan and its trade partner.

2.3 Plastic scrap trade variables

As discussed above, trade costs and economic size are considered to be the key determinants of bilateral trade volume in general. Environmental regulation stringency is a determinant that is theoretically relevant to any transboundary activities that enable the redistribution of negative environmental externalities, whether this is the relocation of firms that produce dirty goods or the shipment of waste products.

In addition to these two key determinants, some empirical studies acknowledge factors that are more specific to the waste trade context. Researchers often incorporate these factors, although often without formally establishing a theoretical relationship. In her study of hazardous waste, Baggs (2009) incorporates the capital/labor endowment ratios of exporters and importers into the gravity model as a proxy for technological capabilities in the hazardous waste disposal sector. She finds that countries with higher capital/labor ratios tend to import a higher volume of hazardous waste. However, capital abundance is correlated with GDP per capita, so this effect disappears when the latter variable is included in estimation. Kellenberg (2012) includes recycling industry wage rates to reflect the marginal productivity of workers in the recycling sector, assuming that higher wage rates reflect higher productivity. He computes the recycling wage gradient as the difference in recycling productivity between exporting and importing country pairs. He finds as expected that the more productive the exporting country is at recycling relative to the importing country (positive gradient), the smaller is the volume of waste flow between the country pair.

Our study focuses on trade in recyclable plastic. We know of two studies presenting models designed specifically to explain international trade in recyclable waste. Sugeta and Shinkuma (2012) develop a two-country theoretical model that addresses the role that cross-country heterogeneity in recycling technology plays in determining the pattern of international recyclable waste flow and the corresponding environmental harm. Both countries produce, consume and trade consumption goods and recycled materials but have different recycling technologies, resulting in different recovery rates. The model demonstrates that whether a country gains net benefit or suffers net environmental harm from trade in recyclable waste depends on the recovery rates of the two countries as well as efficiency in the production of consumption goods. The model implies, however, that the incentive to import and export recyclable waste should depend on the production characteristics of the recycling industry as well as the production sector that uses recycled materials as input.

Higashida and Managi (2014) develop a gravity model for recyclable waste trade. They specify the demand and supply equations of recyclable wastes and use them to derive the commodity-specific gravity equation. The model suggests that the trade volume is determined by transportation cost, the scale of the recycling sector, and the ratio of imported waste that enters landfills to total waste imports. It predicts that if final consumption goods that use recyclable waste as production inputs are not freely traded in the global market, then domestic demand for those goods will encourage imports of recyclable waste.

3 Theoretical framework

3.1 The gravity model

In this section we explore ex ante drivers of waste trade patterns and the effects of shocks (specifically, a large negative demand shock) on that trade. The models we develop are intended to inform an empirical exercise using a gravity model of trade, so we begin with a very brief outline of that model.

The basic gravity model embodies two ideas: larger countries trade more, and higher trade costs reduce trade flows. Let $X_{ij}$ indicate the volume of trade flow from country i to country j. Let $G_i$ and $G_j$ be the economic sizes of the two countries. Let $\tau _{ij}$ represent factors hypothesized to affect bilateral trade costs, such as distance, colonial ties, common language, common currency, contiguity, membership in trade agreements, and bilateral tariffs. The most basic form of the gravity equation, known as the “intuitive” gravity model (Capoani, 2023), is:

$$\begin{aligned} X_{ij} = \alpha _0 G_i^{\beta _1} G_j^{\beta _2} \tau _{ij}^{\beta _3} \end{aligned}$$

(1)

The intuitive model contradicts real-world observation in two major areas (Shepherd et al., 2019), however. First, it suggests that a change in trade cost between country pair i and k has no effect on the trade flow between country pair i and j. Second, the intuitive model suggests that an equal decrease (increase) in trade costs across all country pairs, including domestic trade, would result in proportional increases (decreases) in trade volume across all bilateral routes, including the domestic trade. These limitations motivated economists to derive the micro-foundation for the gravity model, producing variants that are collectively known as structural gravity models. Among these, the “Gravity with Gravitas” (GwG) model by Anderson and Van Wincoop (2003) both solves the two major issues faced by the intuitive gravity model and also provides straightforward guidelines for empirical estimation (Shepherd et al., 2019). For this reason, we use the GwG model as a basis for our empirical work.

Shepherd et al. (2019) provide a thorough explanation of the derivation of the GwG model. In summary, the model involves deriving sector-specific demand and price equations for each country. The demand equation is derived from the maximization of aggregate consumer utility (assumed to take CES form) subject to the country’s budget constraint. The model assumes that an economy consists of many sectors, indexed by k. Each sector has a continuum of firms which produce a continuum of varieties of good k that is specific to the sector. The price conditions are derived from the profit maximization problem of the firms. Prices of goods produced in country i and consumed in country j are expressed as a function of country-pair specific trade costs in iceberg form. This structure produces the gravity equation by combining the three elements (demand, price, and trade cost equations) with the general equilibrium identity requiring that in each country, the total value of production must be equal to GDP, that is, that aggregate expenditure is equal to aggregate income.

The final form of this gravity equation explains the export of goods from sector k of country i to country j ($X^k_{ij}$) as a function of country i’s income from selling the domestically produced output of k sector worldwide ($Y^k_i$); country j’s total expenditure on sector k ($E^k_j$); the total world output of sector k ($Y^k$); the elasticity of substitution between varieties of output of sector k from consumer’s utility function ($\sigma _k$); the trade cost ($\tau _{ij}^k$); outward multilateral resistance ($\Pi _i^k$); and inward multilateral resistance ($P_j^k$).⁶ The Anderson and Van Wincoop (2003) gravity equation is:

$$\begin{aligned} X^k_{ij} = \dfrac{Y^k_iE^k_j}{Y^k}\left( \dfrac{\tau ^k_{ij}}{\Pi ^k_iP^k_j}\right) ^{1-\sigma _k}. \end{aligned}$$

(2)

Our estimating models (see Sect. 5) are variants on this fundamental structure, with modifications to account for measures of environmental quality, pre-and post-ban differences, and a variety of fixed effects.

3.2 The dynamics of plastic scrap trade

The fundamental determinants of trade flows for general commodities as suggested by the gravity model are also relevant to trade in plastic scrap. Plastic scrap, however, has an additional characteristic not shared by most commodities. It has a double identity in that it can be a raw material if it gets recycled, or a piece of trash if gets dumped. In fact, since all scrap contains at least some non-recyclable contaminants, it is guaranteed that some additional costs will be incurred—either to remove and dispose of contaminants, or to dispose of the entire shipment if processing is uneconomic. This feature also means that the value of scrap depends not only on the price of virgin plastic, for which it is a substitute, but also on the costs of processing to remove contaminants and disposal of non-recyclable material. Importantly for the trade story, processing and disposal can occur at either (or both) ends of the trade flow; therefore, trade depends on relative processing and disposal costs across trade partners.

This intrinsic feature (the double identity) implies that export of scrap from country i to country j may also function as the purchase of disposal services by country i from country j, where the service fee is embedded in the price of the scrap commodity. In other words, scrap importers may be paying a low price for raw materials (the recyclable portion) in exchange for bearing the cost of disposing of unusable waste (the non-recyclable portion). Thus, trade in plastic scrap is likely driven by two forces: importers’ demand for plastic scrap, and exporters’ demand for disposal services.

In the exporting country, decisions over processing, disposal and/or export originate with recyclers, formally known as material recovery facilities (MRFs). In the importing country, entities that purchase, process and/or dispose of scrap are known as reclaimers. Both are modeled as representative agents in a population of many identical firms with unrestricted entry and exit. We assume these firms to be profit-maximizing, price-taking entities. Trade in plastic waste, when it occurs, originates with recyclers and flows to reclaimers. Our model considers their responses to price changes induced by China’s plastic waste import ban. The following provides a skeletal description of the model and its predictions; both the recycler and reclaimer models are fully and formally set out in section B and C of the appendix.

3.2.1 Supply: recyclers

A recycler with a municipal recycling collection contract takes in an exogenous quantity of domestic recycling with an exogenous initial contamination rate. It processes this waste to produce a mix of recyclable plastics and waste for disposal. The mix produced depends on the initial contamination rate and the effort expended by the recycler to separate contaminants from usable plastic scrap. Additional effort is costly, and so is additional waste disposal. The price at which the recycler can sell recyclable plastic is a diminishing function of the post-processing contamination rate. The higher is the contamination rate of scrap offered for sale by the recycler, the greater is the effort required to be expended by buyers (the reclaimers) in order to generate feedstock for the manufacture of post-consumer resin (PCR) pellets. There is an upper threshold on this contamination rate, based on the reclaimer’s processing costs and the price of PCR pellets; above that threshold the price offered to recyclers is zero. If producing output with a contamination rate below this the threshold is not economically feasible, a recycler may choose to exert no effort and instead dispose of all its collected recyclable waste. A recycler may be willing to produce and sell recyclable plastics at a loss, however, as long as that loss is smaller than the cost of sending what it collects to a landfill.

International trade occurs when recyclers and reclaimers are in different countries. Suppose that the cost of effort is a function of wage rates that may vary across labor markets. Similarly, suppose that waste disposal costs also vary across jurisdictions, according to the stringency of environmental regulations. Then the price and the threshold contamination rate of scrap offered for sale by a recycler may vary among countries with different labor costs and environmental management regimes. This provides the basis for changes in trade volumes and destinations. In addition, the supply of disposal services may be more or less elastic in different markets, so disposal costs may also vary differentially in response to increases in demand.⁷

The withdrawal of China from the plastic waste market affects production and trade decisions world-wide. The China import ban reduces demand for plastic scrap sold by recyclers. This will lower the world price of scrap. In addition, when exports of scrap to Chinese reclaimers cease, then the supply of PCR from Chinese reclaimers also falls, leading to a rise in PCR pellet prices. These predictions conform with reported trends for HDPE plastic scrap prices (Senechal, 2018) and PCR supply and price (Yoshida, 2022).

What are the likely effects on the industry, and on trade? For recyclers, a lower plastic scrap price means reduced profits at all contamination rates. Returns to effort decrease. The model predicts that recyclers that face high disposal costs, such as those in wealthier countries, are more likely to continue producing and selling recyclable plastics, whereas those with lower disposal costs are more likely to reduce effort and increase disposal. Recyclers that continue to sell may increase or decrease effort to remove contamination, depending on how the ban changes the additional value of a lower contamination rate relative to the costs of disposal and contaminant removal. Overall, the model predicts that the global scrap exports will decrease, and total disposal quantity will increase. Sales will increase to markets where disposal services are elastic and/or where the cost of effort to remove contaminants is lower. This corresponds to real-world observations of shipments from wealthy to developing countries that were labeled as recyclable scrap, but which in reality were simply trash, or had contamination rates above the acceptable threshold (Law et al., 2020). In all countries, the activities of recyclers will cause demand for landfill and other disposal services to increase.

3.2.2 Demand: reclaimers

A reclaimer produces post-consumer resin (PCR) pellets using three types of input: plastic scraps, labor, and capital. Pellets are sold into a market in which prices depend on demand and the price of virgin plastic; these prices are exogenous to the reclaimer. The reclaimer’s willingness to pay for plastic scrap depends on its contamination rate as well as unit costs of labor and capital and PCR prices. Production of PCR is a vertically integrated process with two steps. First, plastic waste is sorted to remove contamination. Second, the granulation process turns sorted plastic scraps into PCR pellets.

In the sorting step, a reclaimer employs labor to separate contaminants from clean feedstock. The removed contaminants are disposed of at a non-negative cost. The unit cost of sorting increases with the wage and the contamination rate. Since sorting removes all contaminants, the granulating function (the second step) does not depend on the contamination rate. A reclaimer earns revenue by selling PCR pellets. Taking prices and technology as given, profits are maximized by choosing the contamination rate, the amount of scrap input, labor, and capital to employ in the production process.

The China import ban affects reclaimers in two ways. First, it at least weakly decreases the price of plastic scrap as China exits the market. Second, it at least weakly increases the price of PCR. This is because the ban cuts off Chinese reclaimers’ access to foreign plastic scrap, thereby reducing supply in the PCR market (Yoshida, 2022). The model predicts that these price changes at least weakly encourage reclaimers outside China to increase production. Their demand for capital and labor inputs increases, but changes in the scrap input quantity and contamination rate are ambiguous depending on the sorting function (the rate of change of the sorting output with respect to contamination rate relative to the rate of change with respect to the scrap input volume). This is because an increase in the clean feedstock of plastic scraps can be achieved either by adjusting the input quantity, or the contamination rate of purchased scrap, or both.⁸ In addition, heterogeneity in labor costs contributes to variation in the magnitude of input adjustment across reclaimers in different labor markets.

3.2.3 Determinants of plastic scrap trade

For simplicity, assume that technologies are identical across countries. These include the contamination removal function, scrap sorting function, and granulating function. For ease of interpretation, let us categorize determinants of plastic scrap trade into three groups: (1) general bilateral trade factors, (2) plastic waste supply factors, and (3) plastic waste demand factors.

Two types of bilateral trade factors enter the gravity model. The first is the economic mass of traders. The larger the mass of trading partners, the larger the trade volume. The second is bilateral trade costs such as distance and non-distance trade frictions. The higher the trade costs, the smaller the trade volume. Factors that determine plastic waste supply in the international market are identified by the recycler model. These include waste generation, initial contamination rate, disposal cost, effort cost, and the price of plastic scraps. Factors that determine international demand for plastic waste are identified by the reclaimer model. These include disposal cost, price of plastic scraps, and the price of PCR.

Notice that some factors, such as disposal cost and trade/transaction costs belong in more than one of the three groups. Moreover the quantity of waste generated in the recycler model likely coincides with economic mass in the gravity model.

In addition to these three types of determinants, factors that determine the supply of disposal services may also play a role in shaping the international trade of plastic waste. We have not modeled these, but intuitively, the main factor determining the supply of disposal services is their cost. The higher is this cost, the lower the willingness of any reclaimer to import scrap.

3.3 Predicted response to trade diversion

The recycler model indicates that countries with higher disposal costs are more likely to continue to export plastic scrap after the ban, and countries with lower disposal costs are more likely to switch from selling to disposal. The reclaimer model suggests that countries with higher disposal costs are less likely to expand PCR production as PCR prices rise. Thus, after the ban we expect that other things equal, export of waste from countries with higher disposal costs to those where such costs are lower will increase.

The recycler model shows that, when faced with sufficiently high disposal costs, a recycler may continue to export even at a loss. Wealthier nations tend to have better and more formalized waste management systems, which likely translates to higher domestic disposal costs. In global plastic waste trade, this implies that a recycler may seek to export waste even at a loss if that option, after accounting for additional transaction costs, is cheaper than domestic disposal. This is where waste trade intersects with the pollution haven hypothesis: lower disposal costs reduce the unit costs of processing plastic scrap—whether through recycling or disposal. However, the distinction between these two choices has environmental implications. The larger the portion of imported scrap that enters the PCR production process, the greater are the environmental benefits (other things equal), whereas environmental costs are larger, the greater is the portion that is discarded. In cases where disposal costs are low due to greater tolerance for open dumping and other improper waste management methods, the environmental impacts of disposal will likely be higher.

4 Data

We obtain annual trade data of plastic scrap between 2008 and 2019 from UN Comtrade. The 4-digit HS commodity code that corresponds to plastic scrap is 3915. There are four 6-digit subcategories within HS-3915: 391510 includes ethylene polymers waste, parings, and scrap; 391520 includes styrene polymers waste, parings, and scrap; 391530 includes vinyl chloride polymers waste, parings, and scrap, and 391590 includes other plastics.

The desirability of plastic scraps as raw materials in the production of recycled plastic pellets varies according to the type of plastic. In general, ethylene polymer (391510) is the most desirable type of plastic waste for recycling. The commonly known types for this category of plastics are PET, HDPE, and LDPE (these correspond to recycling numbers 1, 2, and 4 respectively). In this analysis, we use data of the 4-digit code. Accounting for different plastic types is a potentially meaningful extension. However, tariff data are available only at 4-digit level.

We limit the sample to 73 countries that traded more than 20 metric tons.⁹ of plastic scraps every year between 2008 and 2016 or in 2018 or 2019. We exclude China and Hong Kong from our analysis because our objective is to examine how other counties have been affected by China’s import ban. For each exporting country in each year, we define its destination choice set as countries that (1) imported at least some plastic scraps in that year from the exporting country and (2) imported plastic scraps from the exporting country at least once between 2008 and 2019. This makes 2,782 country pairs and 32,705 exporter-importer-year observations.

We obtain data on bilateral characteristics that are commonly used in the gravity model from the CEPII database. In addition to distance, we use the following time-invariant bilateral indicators: contiguous border, common official language, common currency, origin is/was a colonizer of destination, and destination is/was a colonizer of origin. We also use the time-varying indicator of common membership in a regional trade agreement.

Bilateral tariffs on plastic scrap commodities (HS-3915) are computed using most-favored-nation (MFN) tariff rates and preferential tariff rates from WTO. Data on total GDP and GDP per capita data are from the World Bank. Lastly, the Environmental Performance Index (EPI) data are from NASA’s Socioeconomic Data and Application Center (SEDAC).

Our study differs from nearly all previous work on gravity analysis of waste trade in using a time-varying measure of environmental quality. The EPI score is published every even-numbered year since 2006. It is designed to reflect the environmental quality of a country. The score ranges from 0 to 100, with higher scores reflecting better environmental quality. We use the EPI score as a proxy for disposal cost and/or environmental regulation quality. We assume that countries with higher environmental regulation quality also have higher disposal costs.

The overall EPI score is computed from scores assigned to a set of environmental indicators that varies from year to year. The 2020 EPI includes a measure of the share of “solid waste properly managed”,¹⁰ among other indicators, and this provides us an opportunity to check the correspondence of this indicator with the EPI as a whole. Figure 1 shows a scatter plot of these two variables¹¹ and suggests that they are quite strongly correlated. For this reason, we believe that while using the EPI score as a proxy for disposal cost is not perfect, it is a reasonable option given limited data availability.

Fig. 1

Correlation between overall 2020 EPI score and the share of properly managed waste. Note: What A Waste Global Database published by the World Bank

Figure 2 shows the correlation between country GDP per capita and EPI score by income group. EPI score and GPD per capita are positive correlated, however, there is considerable variation in EPI score among countries with similar income levels. Table 1 shows the five countries with the highest and lowest average EPI scores within each income group. Figure 2 and Table 1 show that there are lower-middle income countries with higher EPI score than high income countries. For example El Salvador, a lower-middle income country, has an average EPI score of 60.7 which is higher than the average score of UAE (59.5), a high-income country.

Fig. 2

Correlation between EPI score and GDP per capita by income group (2008–2018 average). Note: GDP per capita data are from the World Bank. Environmental Performance Index (EPI) data are from NASA’s Socioeconomic Data and Application Center (SEDAC)

Table 1 Five countries with highest average EPI scores and five countries with lowest average EPI scores within each income group

Figure 3 compares the distributions of EPI scores by net trade status. In the pre-ban period the EPI distributions of net importers and net exporters are similar. In the post-ban period, however, the net exporter distribution is substantially to the right of the distribution of net importers.

Fig. 3

EPI Score Distributions by Net Trade Status. Note: Trade data are from UN Comtrade

4.1 Summary statistics

Table 2 compares the pre-ban and post-ban averages of bilateral characteristics. The pre-ban period is defined as 2008 to 2016 and the post-ban period includes 2017 through 2019. We include 2017 as part of the post-ban period because even though the ban became officially effective in January of 2018, adjustments to the ban occurred before its official implementation date. The monthly time trend of plastic scraps imported by China shows a sharp drop after July 2017 when the ban was announced (see Fig. 4).

Fig. 4

Change in monthly quantity of plastic scraps imported by the world and China. Note: Trade data are from UN Comtrade

As seen in Table 2, bilateral export and import quantities increased after the ban. As expected, unit value of plastic waste ($ per kilogram) decreased both exports and imports. Origin-minus-destination differences in GDP per capita are positive in all three cases (all-period, pre-ban, and post-ban). This confirms that plastic scraps are traded from richer countries to poorer countries, on average. Origin-minus-destination differences in EPI score are also positive in all three cases, confirming that on average, plastic scrap is traded from countries with stronger environmental regulation to those where it is weaker. The difference is larger in the post-period, but the pre-post ban difference is not statistically significant. The gradient of EPI score, however, increased after the ban and the change is statistically significant. This affirms the hypothesis that plastic waste flows from countries with stronger environmental regulation and/or higher disposal costs to countries with weaker environmental regulation and/or lower disposal costs.

Table 2 Pre-Ban versus post-Ban averages of bilateral characteristics

Table 3 compares averages of characteristics of net importers and net exporters. Panel (A) uses data from the whole period of study (2008–19), while panels (B) and (C) use pre-ban and post-ban data respectively. Overall, these statistics show that while net importers have significantly more import partners (i.e., they import from more countries) than net exporters, they have a similar number of export partners (i.e., number of countries they export to). Net importers have lower GDP per capita than net exporters on average in all three cases. The gap increased by a factor of almost two in the post-period. Net importers have a lower EPI score than net exporters on average in all three cases. The gap increased in the post-period, more than doubling in magnitude. These statistics suggest that poorer countries likely take in more plastic scraps than richer countries and that this pattern became more pronounced after the ban.

Table 3 Average characteristics of Net importers versus Net exporters

Simple mean comparisons of bilateral and country-specific characteristics suggest that the international trade of plastic scraps follow a waste haven pattern where plastic scrap, a high environmental-cost commodity, flows from richer to poorer nations. These simple statistics, however, only serve as a preliminary evaluation of how the waste haven pattern changed after the ban. In the next section, we use the gravity model to examine this question more closely.

4.2 Evidence of trade diversion

Figure 4 shows monthly variation in the quantity of plastic scraps imported by the world (black) and by China (blue) between 2015 and 2019. Each vertical line corresponds to an event that is relevant to China’s ban on plastic scraps import. The first line, in July 2017, marks China’s announcement of intent to ban imports of post-consumer plastic scrap. The second line, in January 2018, is when the ban went into effect. The third line, April 2018, is when China announced its intention to expand the list of banned items; this expansion went into effect in January 2019, which is the rightmost line. Our study concerns the first two of these events. There is a sharp decrease in Chinese imports from announcement date, perhaps because exporters were unable or unwilling to initiate new shipments after that date. For this reason, in our annual data series we treat 2017 as belonging to the post-ban period. As the figure makes clear, world imports of plastic scrap declined along with Chinese imports—but by less than the decline in the latter. This is evidence of trade diversion.

Figure 5 shows the change in net exports of plastic scraps by income group, excluding China and Hong Kong.¹² Light blue columns correspond to pre-ban quantities: the sum of net exports in 2015 and 2016. Dark blue columns correspond to post-ban quantities: the sum of net export in 2018 and 2019. The two left-most columns shows that high-income countries decreased their net export quantity by more than half, from 11.6 to 4.4 million tons, but remained net exporters as a group. The middle two columns show that upper-middle-income countries were collectively net exporters before the ban but became net importers afterward. Strikingly, post-ban net import quantity (2 million tons) is more than double the pre-ban quantity (0.94 million tons). The two rightmost columns show that lower-middle-income countries also changed from net exporters to net importers as a group, with the post-ban net import quantity (0.91 million tons) being about three times the pre-ban quantity (0.32 million tons).

Fig. 5

Change in net-export of plastic scraps by income group: before (2015–16 total) versus after (2018–19 total) the ban. Note: Trade data are from UN Comtrade. Income group data are from UNCTAD

Figure 6 shows yearly variation in import quantity by income group, and Fig. 7 shows variation in export quantity (again, China and Hong Kong are excluded from these data). All country groups experience an increase in import quantity between 2016 and 2018. The increase is the sharpest for upper-middle countries, followed lower-middle income countries, while high-income countries only see a slight increase. Export quantities from all country groups declined around the time of the ban.

Fig. 6

Annual import quantity of plastic scraps by Income Group. Note: Trade data are from UN Comtrade. Income group data are from UNCTAD

Middle income countries where net exporter of plastic waste as a group before the ban. After the ban, however, they become net importer as a group as they increased imports more than they decreased exports. High-income countries, on the other hand, remained net exporters as a group even after the ban because they increased import less than they decreased exports. The distinction is important for analysis of trade diversion and welfare impacts. All countries adjust to the ban by exporting less plastic waste, likely by absorbing more waste domestically. However, middle income countries, particularly upper-middle income countries, were more likely to become destinations for trade diversion after the ban than high-income countries. Given that middle-income countries were net exporters prior to the ban, it is unlikely that they had sufficient capacity to properly process the net import increase.

Fig. 7

Annual export quantity of plastic scraps by income group. Note: Trade data are from UN Comtrade. Income group data are from UNCTAD

Figure 8 shows a scatter plot of the net-import quantity in 2016 (x-axis) and 2018 (y-axis). Countries in the lower left quadrant, such as Japan and the US, were net exporters in 2016 and remained net exporters in 2018. Countries in the upper left quadrant, such as Thailand and Indonesia, were net exporters in 2016 but became net importers in 2018. Out of 74 countries in this study, 14 changed from net exporter to net importer after the China ban.

Fig. 8

Relationship between net import quantities in 2016 and 2018. Note: Trade data are from UN Comtrade

Table 4 lists net import quantities of these countries in 2016 and 2018, ranked in descending order by 2018 net import quantity.¹³ The table also shows 2018 net import quantity as a share of 2016 net export quantity (column 4), the change in net import quantities (column 5), and the change in net import quantity as a share of 2016 net export quantity (column 6). For eight countries, net import quantity more than doubled. For six of these, 2018 net import quantity was more than double its 2016 value. If we assume that the net-import quantity in 2016 reflects the capacity to process plastic waste and that the amount of waste generated by each country remained relatively constant during the ban implementation, it is very unlikely that these countries would have been able to ramp up recycling capacity within less than a year to accommodate the increase in waste inflow from trade diversion.

Table 4 Countries that were net exporters in 2016 but became net importers in 2018

Some countries that were net importers in 2016 and 2018 may also have struggled with capacity constraints. Malaysia, for example, had net imports of 124 million tons in 2016 but by 2018, this increased to 825 million tons, more than a 6-fold increase. Similarly the Czech Republic increased its net import quantity from 5.8 million tons to 45.5 million tons, an increase of more than 600%. This pattern of change suggests that a large share of plastic scrap imports may have entered the disposal system as trash instead of entering the recycling system as raw materials. In this case, countries that started receiving overwhelming quantities of plastic waste in the post-ban period likely faced negative environmental consequences.

5 Empirical analysis

5.1 Estimation models

We augment the standard gravity model with two variables relevant to the waste haven hypothesis (WHH): GDP per capita and EPI score.¹⁴ We use the gravity model to evaluate bilateral trade in plastic scrap (commodity code HS-3915) from origin country o to destination country d in year t, with t between 2008 and 2019. We specify two models: one with country-level WHH variables, and another with variables for each bilateral pair.

5.1.1 Model with country-level WHH variables

Let $Q_{odt}$ be the quantity of HS-3915 exported from country o to country d in year t measured in tons. Let $\varvec{G}$ be a set of additional control variables commonly used in gravity model estimation. These include time-variant country GDP; time-invariant bilateral distance, indicators of contiguity such as common language, colonial relationship, currency; and time-variant bilateral indicator of common regional trade agreement and bilateral tariffs on plastic scrap commodities ($\tau _{odt} = log(1 + \text {tariff}_{odt})$).

The WHH variables in model 1 are country characteristics: origin’s and destination’s GDP per capita ($w_{ot}, w_{dt}$) and EPI score ($e_{ot}, e_{dt}$). Model 1 also includes interactions of the post-ban indicator ($post_{t}$) with each of these WHH variables. This permits a comparison of pre-ban and post-ban relationships between the augmented variables and trade flows. We estimate the model with fixed effects (FE) for year ($\phi _t$), exporter ($\phi _o$), importer ($\phi _d$), and country pairs ($\phi _{od}$). The inclusion of the pair fixed effects causes time-invariant bilateral variables (such as distance) to drop out. Consequently, the set of gravity variables ($\varvec{G}$) in model 1 comprises only of time-variant variables, including country-specific GDP, bilateral tariffs on plastic scrap commodities, and an indicator for common regional trade agreements. Following Yotov et al. (2016), we use the PPML estimator to estimate the following gravity equation.¹⁵

Model 1:

$$\begin{aligned} X_{odt} = exp[&\eta _1 w_{ot} + \eta _2 w_{ot}\times post_t + \eta _3 w_{dt} + \eta _4 w_{dt}\times post_t + \eta _5 e_{ot} + \eta _6 e_{ot}\times post_t + \nonumber \\&\eta _7 e_{dt} + \eta _8 e_{dt}\times post_t + \varvec{G'\xi } + \phi _o + \phi _d + \phi _t + \phi _{od}] \times \epsilon _{odt} \end{aligned}$$

(3)

Model 1 decomposes the impact of GDP per capita and EPI scores into pre-ban and post-ban parts. There are eight coefficients of interest. $\eta _1$ and $\eta _3$ measure the pre-ban average impact of origin’s and destination’s GDP per capita on the export quantity of plastic scraps. $\eta _2$ and $\eta _4$ measure the post-ban change in the average impact of origin’s and destination’s GDP per capita on the quantity of plastic scrap. Similarly, $\eta _5$ and $\eta _7$ measure the pre-ban average impact of origin’s and destination’s EPI score (the proxy for disposal cost) on the export quantity of plastic scraps. $\eta _6$ and $\eta _8$ measure the post-ban change in the average impact of origin’s and destination’s EPI score on the export quantity of plastic scrap.

In comparing estimates in the pre- and post-ban periods, it is necessary to keep in mind that because the data does not include trade with China and Hong Kong, the pre-ban estimates are from data that cover at most one-third of world plastic waste trade, whereas the post-ban data cover virtually all. The pre-ban estimates show patterns for that fraction of trade not going to China, whereas the post-ban estimates reveal the pattern of world trade that emerges when China is absent from the system.

5.1.2 Model with bilateral WHH variables

Model 1 helps answer questions about how country-specific characteristics may influence waste trade. For example, do wealthier countries export more plastic waste? Do countries with poorer environmental regulation quality import more plastic waste? These are meaningful questions but they do not adequately address the waste haven problem because they do not address how bilateral differences may jointly determine the waste trade. For example, does plastic waste tend to flow from richer to poorer countries? Does plastic waste tend to flow from countries with stronger environmental regulations to countries with weaker environmental regulations?

To address these questions, we make use of “gradient” forms of the waste haven variables, following Kellenberg (2012). Let $x_{ot}$ be a time-varying characteristic of origin o and let $x_{dt}$ be the same characteristic of destination d. The gradient of x, which is a time-variant bilateral variable, is defined as:

$$\begin{aligned} g(x)_{odt} = \dfrac{x_{ot} - x_{dt}}{(x_{ot} + x_{dt})/2} \end{aligned}$$

In other words, the gradient of x is the origin–destination difference as a fraction of the pair average.

In his study, Kellenberg uses the gravity framework to analyze the trade flow of scrap commodities (including plastic and other scrap materials). He conducts a cross-section analysis in which the explanatory variable of interest is the environmental regulation gradient, computed using data from the 2003–2004 Global Competitiveness Report. The gradient version of the index is interpreted as a measure of the average percentage difference in environmental regulation between the importer and the exporter, where larger positive values imply that the exporter has more stringent environmental regulation than the importer, and vice versa. Thus a positive coefficient is expected on the environmental regulation gradient variable.

In addition to capturing the effects of pair-specific characteristics, the gradient variant also addresses an identification issue in the country-level model. That model includes origin, destination, and pair fixed effects, thereby controlling for country-specific and pair-specific omitted variables. However, it is still possible that some time-variant country characteristics remain in the error term, causing omitted variable bias. The gradient model addresses this threat by replacing origin fixed effects and destination fixed effects with origin-year fixed effects and destination-year fixed effects. There is a cost, however: the application of country-year fixed effects rules out estimation of the effects of country-specific GDP per capita and EPI scores.

We define the two gradient variables—GDP per capita ($w_{odt}$) and EPI score ($e_{odt}$) as follows:

$$\begin{aligned} g(w)_{odt}&= \dfrac{\text {GDP per Capita}_{ot} - \text {GDP per Capita}_{dt}}{(\text {GDP per Capita}_{ot} + \text {GDP per Capita}_{dt})/2}\\ g(e)_{odt}&= \dfrac{\text {EPI}_{ot} - \text {EPI}_{dt}}{(\text {EPI}_{ot} + \text {EPI}_{dt})/2} \end{aligned}$$

The MRF and reclaimer models outlined in Sect. 3 suggest that higher disposal costs encourage exports of plastic scrap and discourage imports. Therefore, we expect positive coefficients on both the GDP per capita gradient and the EPI score gradient.

Analogous to the model with country-specific WHH variables, the model with bilateral WHH variables includes fixed effects for origin-year ($\phi _{ot}$), destination-year($\phi _{dt}$), and country pairs ($\phi _{od}$). Standard errors are again clustered at the country pair level. The inclusion of these fixed effects implies that the set of gravity variables ($\varvec{G}$) in model 2 only includes time-variant bilateral vairables, namely bilateral tariffs on plastic scrap commodities and the indicator of common regional trade agreements.

Model 2:

$$\begin{aligned} Q_{odt} = exp[&\lambda _1 g(w)_{odt} +\lambda _2 g(w)_{odt} \times post_t +\lambda _3 g(e)_{odt} +\lambda _4 g(e)_{odt} \times post_t \nonumber \\&+\,\varvec{G'\xi } + \phi _{ot} + \phi _{dt} + \phi _{od}] \times \epsilon _{odt} \end{aligned}$$

(4)

Model 2 decomposes impacts of the GDP per capita and EPI score gradients into pre-period and post-period parts. There are four coefficients of interest. $\lambda _1$ and $\lambda _2$ measure the pre-ban and post-ban impacts of the GDP per capita gradient on the export quantity of plastic scrap. $\lambda _3$ and $\lambda _4$ measure the pre-ban and post-ban impact of the EPI score gradient on the export quantity of plastic scrap.

5.2 Estimation results

Results from model 1 are reported in Table 5 and those from model 2 are in Table 6. In each table, columns (1) presents results from estimations that exclude the EPI score, and column (2) presents the full model.

Table 5 PPML estimates of the gravity model with country characteristics

Table 6 PPML estimates of the gravity model with gradient characteristics

5.2.1 Effect of GDP per capita

Origin’s GDP per Capita

We expect origin’s GDP per capita to be positively correlated with the quantity of plastic scrap, for four reasons. Residents of wealthier countries consume more goods per capita, thereby generating more recyclable plastic waste; their countries are more likely to have legal requirements for recycling, allowing for better extraction of recyclable plastics from municipal solid waste; and residents likely have a higher willingness to pay for environmental regulation quality, which may lead to a preference for exporting scraps material to be handled elsewhere. Lastly, higher-income countries likely have higher labor costs. Recycling plastic waste requires some manual sorting, plastic recycling may be relatively costly in higher-income countries.

Model 1 estimation results indicate, as expected that higher GDP per capita in the origin country encourages more trade flow. The estimates show that in the pre-ban years, a 1% increase in the GDP per capita of origin increases quantity exported by 4.32% to 3.96%. The post-ban interaction term is not statistically significant, indicating this relationship was unaffected by the ban.

Destination’s GDP per Capita

The sign on GDP per capita of destination countries cannot be unambiguously predicted. If plastic scrap is imported for use as raw material in the production of recycled primary form plastics to feed demand for raw materials from the domestic manufacturing sector, then industrializing middle-income countries may import more plastic scrap than high-income and low-income countries. In addition, while the sorting process is labor-intensive, the process of turning sorted scrap into primary-form plastic is more capital- and technology-intensive. For these reasons, the relationship between plastic scrap imports and GDP per capita may not be linear. It might follow an inverse U-shape curve, similar to the environmental Kuznets curve.

We find that GDP per capita of the destination has no impact on trade flow in the pre-ban period, but the interaction term indicates that there is a negative relationship in the post-ban period: a 1% increase in the GDP per capita of destination decreases the quantity traded by 0.27% to 0.34%. This finding matches the real-world observation that lower-income countries started importing more plastic waste than did higher-income countries.

Gradient of GDP per Capita

We expect trade in plastic scrap to go from higher income origins to lower income destinations, so we expect the coefficient on the GDP per capita gradient variable to be positive. Table 6 confirms this expectation. The estimates are positive and statistically significant, and indicate that in the pre-ban period, for every 1% that an importer’s GDP per capita decreases relative to an exporting partner, the importer will import 0.02% more waste from that exporter.¹⁶ Post-ban interactions are not statistically significant, indicating that the positive relationship between GDP per capita gradient and plastic scrap trade did not change after the ban.

5.2.2 Effect of EPI score

5.2.2.1 Origin’s EPI Score

We expect the origin’s EPI score to be positively correlated with export quantity, since EPI score is strongly correlated with effective waste management (Fig. 1). We assume that a higher EPI score corresponds to a higher cost of waste disposal, which in turn increases costs for plastic reclaimers and MRFs. Thus, a country with a higher EPI score may have less incentive to import plastic scraps and more incentive to export them. Results in Table 5 show no statistically significant relationship between the EPI score of exporters and trade volume in the pre-ban and post-ban periods.

5.2.2.2 Destination’s EPI Score

We expect the destination’s EPI score to be negatively correlated with traded quantity because our model predicts that plastic recycling is more profitable in countries with lower disposal costs. We find as expected that a lower EPI score of importer is associated with a higher trade volume. A one-unit increase in the EPI score of the destination country is associated with a 1.37% decrease in the quantity of plastic scrap imported.¹⁷ This relationship does not change after the ban.

5.2.2.3 Gradient of EPI Score

We expect plastic scrap trade to flow from an origin with higher disposal costs to a destination with lower disposal costs, so we expect the coefficient on the GDP per capita gradient in Table 6 to be positive. The pre-ban estimate in Table 6 is negative and statistically significant at 10%. The estimate suggests that, in the pre-ban period, bilateral trade in plastic waste other than that with China or Hong Kong was weakly from less to more environmentally stringent regimes after controlling for GDP per capita and other variables. This could be the case, for example, if some fraction of exports to countries other than China or Hong Kong were to countries with relatively advanced recycling and disposal systems. However, the post-ban estimate is positive and statistically significant at 1%—and is more than twice the magnitude of the pre-ban estimate. When sales to China are no longer possible, the pattern of bilateral trade is such that for every 1% that an importer’s EPI score falls below that of an exporting partner, the importer will experience a 0.2–0.3% increase¹⁸ in waste imports from the exporter. Even though the GDP per capita gradient estimate is unchanged from pre to post periods, global trade in plastic waste is clearly and strongly associated with differences in disposal costs, as proxied by the EPI, once exports to China are no longer feasible.

5.2.3 Summary of main findings

GDP per capita of origin is positively correlated with trade quantity in the pre-period. This relationship remains unchanged in the post-period. GDP per capita of destination exhibits no relationship with the trade volume of plastic waste in the pre-ban period but exhibits a negative relationship in the post-ban period. The GDP per capita gradient has a positive impact on trade quantity in the pre-ban period, and this relationship remains unchanged in the post-period. Our analysis indicates that plastic scrap trade follows a waste haven pattern in which international trade allows for the negative externalities of plastic waste to be redistributed from wealthier to poorer countries. China’s ban on imports of plastic scrap does not appear to have changed this pattern.

The EPI score of the origin country exhibits no relationship with the trade volume of plastic waste both in the pre-ban and post-ban period. EPI score of destination is negatively correlated with trade quantity in the pre-period. This relationship remains unchanged in the post-period. The EPI score gradient has a negative relationship with the trade volume of plastic waste in the pre-period, though the estimate is only statistically significant at 10%. In the post period, this relationship became positive, over twice the magnitude of the pre-ban estimate, and statistically significant at 1%. This suggests that the relative disposal cost between trading partners became an important determinant of the international flow of plastic waste after the China import ban was imposed.

Without considering the effect of the ban, we find that trade in plastic scrap follows a waste-haven-type pattern in the following ways. First, plastic waste flows more from a richer origin to a poorer destination, and countries with lower GDP per capita import more waste. This implies exportation of the negative externalities associated with the consumption that generated the waste. In terms of country characteristics, lower disposal costs in an importing country are associated with higher trade volume.

Considering the ban’s impact, we find that the waste haven pattern in terms of GDP per capita observed in the pre-period persists in the post-period. Controlling for the effect of GDP per capita on plastic waste trade, we find that the waste haven pattern in terms of environmental regulation quality becomes pronounced after the ban. This finding aligns with the hypothesis that, in the post-ban period, plastic scrap trade is increasingly driven by exporters’ demand for disposal services. Because the measure of disposal cost (EPI score) is highly correlated with the share of poorly managed municipal solid waste (e.g. dumping, open burning, unsanitary landfills, etc.), it is reasonable to conclude that the environmental problem of plastic waste is likely higher under international trade than under a counterfactual in which exporting countries have to handle their waste locally.

6 Conclusion

We explore the ex ante drivers of global trade in plastic waste and examine trends in global plastic waste trade following a policy ban on imports by the world’s largest importer, China. We find that the bilateral trade in plastic waste exhibits a waste-haven pattern. Among the determinants of bilateral trade in plastic waste, bilateral differences in per capita GDP became more influential after the ban. Bilateral differences in the stringency of environmental regulation emerged as significant drivers of trade after the ban.

While this study cannot address the distribution of the net benefit of the plastic scrap trade, it provides some information about the distribution of the environmental costs of processing plastic waste. The results support concerns that post-ban trade diversion may worsen global plastic waste pollution, because more waste gets diverted to countries with less stringent environmental regulation. If the recycling capacity of these new destinations is insufficient to accommodate all the diverted trade, then it is likely that some of the additional waste imports are simply dumped. This corresponds to documented cases where highly contaminated waste mislabeled as scraps was found in “new destination” countries, especially in Southeast Asia (Staub, 2018; Parker, 2018). The 2019 Basel Convention amendments mandate that plastic waste be recycled or disposed of “as close as possible to source” and aim to regularize waste trade and international disposal through a “prior informed consent” procedure between exporters and importers (Benson & Mortensen, 2021). However, these measures were to be implemented only from 2021, so it is still to early to judge their efficacy.

The supply chain from recycling bins to PCR production is long and complex; no study can rigorously account for all its links. We have modeled the immediate ex ante drivers of bilateral trade and estimated the structure and changes in that trade in light of a major policy shock. Related and ongoing work addresses empirical measures of reclaimers’ demand for plastic scrap as raw material, for example recycling capacity (Kojima, 2020); this will facilitate estimation of the extent to which post-ban trade diversion is determined by demand for plastic scraps in importing countries. Similarly, a longer-run perspective will require examination of factors influencing post-ban investments in recycling capacity, waste disposal, and policy responses in the importing countries.

Appendices

Appendix A: Gravity model

The gravity model is widely used among empirical studies of international trade. The original application of the model is for describing the bilateral trade flow between countries but it has also been applied in other types of economic studies such as migration (Lewer & Van den Berg, 2008) and domestic commuting flows (Thomas & Tutert, 2013). In this section, we first review the “intuitive” gravity model to discuss the principle that underlies all varieties of the gravity model. Then, we discuss in more detail the “Gravity with Gravitas” (GwG) model of Anderson and Van Wincoop (2003) which is the basis for our empirical analysis. Insights below also draw from Shepherd et al. (2019) and Head and Mayer (2014).

The basic idea that motivates the gravity model is simple—countries with larger masses (GDP) trade more, and trade cost reduces trade flow. Let $X_{ij}$ indicate the volume of trade flow from country i to country j. Let $G_i$ and $G_j$ be the economic sizes of the two countries. Let $\tau _{ij}$ represent the bilateral trade cost. Traditionally, distance is considered the most fundamental determinant of trade cost. Non-distance determinants of trade cost or trade friction that are common in the application of the gravity model include colonial ties, common language, common currency, contiguity, join-status in trade agreements, and bilateral tariffs. The most basic form of the gravity equation, known as the “intuitive” gravity model, is:

$$\begin{aligned} X_{ij} = \alpha _0 G_i^{\beta _1} G_j^{\beta _2} \tau _{ij}^{\beta _3} \end{aligned}$$

(5)

The log-linearized form, with the error term is:

$$\begin{aligned} logX_{ij} = \alpha _0 + \beta _1 logG_i + \beta _2 logG_j + \beta _3\tau _{ij} + \epsilon _{ij} \end{aligned}$$

(6)

The model posits that larger countries trade more, and thus $\beta _1$ and $\beta _2$ are expected to have positive signs. High trade cost discourages trade flow, and thus $\beta _3$ is expected to have a negative sign. While this intuitive model is simple and serves the purpose of illustrating the basic idea, it is not endorsed by academic studies. The early gravity model as introduced by Tinbergen (1962) is viewed as lacking in theoretical foundation. Additionally, the intuitive model contradicts real-world observation in two major areas (Shepherd et al., 2019). First, it suggests that a change in trade cost between country pair i and k has no effect on the trade flow between country pair i and j. That is, $\frac{\partial log X_{ij}}{\partial log \tau _{ik}} = 0$. Second, the intuitive model suggests that an equal decrease (increase) in trade costs across all country pairs, including domestic trade, would result in proportional increases (decreases) in trade volume across all bilateral routes, including the domestic trade.

The gravity model started gaining momentum in 1995 with multiple economists officially endorsing its ability to produce robust empirical results (Leamer & Levinsohn, 1995; McCallum, 1995). This led to a series of attempts among economists to derive the micro-foundation for the gravity model and to resolve the intuitive model’s contradiction with real-world observations discussed above. These variants are collectively known as structural gravity models. Among these, the “Gravity with Gravitas” (GwG) model by Anderson and Van Wincoop (2003) became one of the most popular variants. Not only does this model solve the two major issues faced by the intuitive gravity model, but it also provides a straightforward guideline for empirical estimation of the model (Shepherd et al., 2019). For this reason, we use the GwG model as a basis for our empirical framework.

A detailed explanation of the derivation of the GwG model is provided in Shepherd et al. (2019). In summary, the model involves deriving sector-specific demand and price equations for each country. The demand equation is derived from the maximization of aggregate consumer utility subjected to the country’s budget constraint. Consumers have CES utility. The model assumes that an economy consists of many sectors, indexed by k. Each sector has a continuum of firms which produce a continuum of varieties of good k that is specific to the sector. The price conditions are derived from the profit maximization problem of the firms.

The model introduces the “iceberg trade cost”, expressing the price of the goods produced in country i and consumed in another country j as a function country-pair specific trade cost. The model produces the gravity equation by combining these three elements (demand, price, and trade cost equations) with a general equilibrium accounting identity which requires that the sector-specific income of country i must be equal to the income it earns from the total sales of all domestically produced varieties of the good of that sector.

The final gravity equation explains the export of goods from sector k of country i to country j ($X^k_{ij}$) as a function of country i’s income from selling the domestically produced output of k sector worldwide ($Y^k_i$); country j’s total expenditure on sector k ($E^k_j$); the total world output of sector k ($Y^k$); the elasticity of substitution between varieties of output of sector k from consumer’s utility function ($\sigma _k$); the trade cost ($\tau _{ij}^k$); outward multilateral resistance ($\Pi _i^k$); and inward multilateral resistance ($P_j^k$). With C being the number countries in the global economy, the multilateral resistance terms, $\Pi _i^k$ and $P_j^k$, are defined as follow:

$$\begin{aligned} \Pi ^k_i = \sum \limits ^C_{j=1}\left( \dfrac{\tau ^k_{ij}}{P^k_j}\right) ^{1-\sigma _k}\dfrac{Y^k_j}{Y^k}, \end{aligned}$$

$$\begin{aligned} P^k_j = \sum \limits ^C_{i=1}\left( \dfrac{\tau ^k_{ij}}{\Pi ^k_i}\right) ^{1-\sigma _k}\dfrac{Y^k_i}{Y^k}, \end{aligned}$$

The Anderson and Van Wincoop (2003) gravity equation, in multiplicative form, is:

$$\begin{aligned} X^k_{ij} = \dfrac{Y^k_iE^k_j}{Y^k}\left( \dfrac{\tau ^k_{ij}}{\Pi ^k_iP^k_j}\right) ^{1-\sigma _k} \end{aligned}$$

(7)

And the corresponding log-linearized form is:

$$\begin{aligned} logX^k_{ij} = logY^k_i + logE^k_j - logY^k + (1-\sigma _k)\left[ log\tau ^k_{ij} - log\Pi ^k_i - logP^k_j\right] \end{aligned}$$

(8)

The outward and inward multilateral resistance terms of this model ($\Pi _i^k$ and $P_j^k$ respectively) captures the real-world feature of the global market that the quantity of trade flow between country i and country j depends on trade costs across all other trading pairs in the market. In other words, in the GwG model, $\partial X_{ij}/\partial \tau ^k_{ij} \ne 0$. This resolves the intuitive model’s limitation in accounting for the cross-route dependence in trade costs as discussed earlier. Additionally, because the intuitive model does not include these outward and inward multilateral resistance terms, they are theoretically the omitted variables in the intuitive equation, causing omitted variable bias issue in the intuitive model.

1.1 Options for estimating the gravity equation

The outward and inward multilateral resistance terms serve a theoretical model’s justification for the inclusion of exporting and importing country fixed effects in an empirical estimation which mitigates against omitted variable bias. This is not the case for the intuitive model where the only country-specific variables implied by the gravity equation are the economic sizes of the trading partners. Additionally, while the GwG model can be used for describing trade flows of goods from multiple sectors of an economy, it can also be conveniently scaled down to fit an analysis of the trade flow of a single sector.

Baggs (2009), Kellenberg (2012) and Kellenberg and Levinson (2014) apply the gravity model in their analysis of international waste trade. In their studies, the sector-superscript k is omitted and the sector-specific expenditure and income are substituted with the GDP of the two countries. With these adjustments, the multiplicative form of the GwG equation (omitting k) becomes:

$$\begin{aligned} X_{ij} = \dfrac{Y_iY_j}{Y}\left( \dfrac{\tau _{ij}}{\Pi _iP_j}\right) ^{1-\sigma } \end{aligned}$$

(9)

And the log-linearized form is:

$$\begin{aligned} logX_{ij} = logY_i + logY_j - logY + (1-\sigma )\left[ log\tau _{ij} - log\Pi _i - logP_j\right] \end{aligned}$$

(10)

There are two estimation methods that are commonly used for estimating the GwG and other structural gravity equations. First is the log-linearized OLS estimation with fixed effects, and the second is the Poisson pseudo-maximum likelihood (PPML) method.¹⁹

• 1. The fixed effects estimation

  In this case, all the exporter- and importer-specific terms are absorbed by the exporter and importer fixed effects ($F_i$ and $F_j$). In relation to the GwG equations defined above in Eq. (10), these include the income terms ($Y_i$ and $Y_j$) and the outward and inward resistance terms ($\Pi _i$ and $P_j$). Recall that $\tau _{ij}$ comprises of distance and other non-distance bilateral trade cost determinants.

  $$\begin{aligned} logX_{ij} = C + F_i + F_j + (1-\sigma )\left[ log \tau _{ij}\right] \end{aligned}$$

  (11)

  where

  $$\begin{aligned} C&= -logY \\ F_i&= logY_i - log\Pi _i \\ F_j&= logY_i - logP_j \end{aligned}$$

  This estimation is commonly used among empirical applications of the gravity model. Kellenberg and Levinson (2014), for example, apply this method to estimate the impact of the Basel convention on the bilateral flow of waste and scrap products.

• 2. Poisson pseudo-maximum likelihood method (PPML)

  Silva and Tenreyro (2006) argue that the Poisson pseudo-maximum likelihood method (PPML) is generally a superior method for estimating the gravity equation for the following three reasons. First, consider the case where the error term enters the gravity equation in multiplicative form, and thus the error appears in the log-linearized equation as a log term.

  $$\begin{aligned} logX^k_{ij} = logY^k_i + logE^k_j - logY^k + (1-\sigma _k)\left[ log\tau ^k_{ij} - log\Pi ^k_i - logP^k_j\right] + loge^k_{ij}, \end{aligned}$$

  (12)

  In this case, the OLS estimator is biased and inconsistent because the first assumption of OLS is violated. Silva and Tenreyro (2006) suggest that this scenario is highly probable in practice, and the PPML is preferred because it provides consistent estimates of the original non-linear gravity model, under the assumption that the correct set of explanatory variables is included in the model. Second, PPML can handle fixed effects, whereas a number of other non-linear estimation methods, such as the logit regression, are not compatible with fixed effects due to the incidental parameters problem. Third, the PPML can handle the occasion where the bilateral trade data contains zeros for many country pairs while the log-log OLS estimation cannot appropriately handle zeros. Kellenberg (2012) uses the PPML method to estimate the effect of cross-country differences in environmental regulation quality on the volume of international waste trade. Unfortunately, the study does not provide results from a log-linearized OLS estimation for comparison. The study, however, presents results from the “zero-inflated” PPML estimation method (ZIPPML). It states that “the ZIPPML model allows for bilateral trade to be generated by two processes, one that generates zero counts, and another that generates positive counts”. The ZIPPML has the effect of increasing the variance of the model, however. Kellenberg shows that the impact of regulation on hazardous waste flow is robust across the two types of estimation, with the ZIPPML producing larger coefficients in general. The impact is as predicted by the pollution haven hypothesis – a decrease in environmental quality in the importing country j relative to the exporting country i increases waste flow of waste from i to j.

Recycler model

1.1 Supply: recyclers

We develop a model of a profit-maximizing plastic recycler to theoretically explore the determinants of the supply for plastic scraps. A detailed version of this model can be found in Chunsuttiwat (2022).

A profit-maximizing recycler takes in an exogenous quantity of waste (W) that was put in recycling bins as agreed upon in its contracts with municipalities. Let $\alpha$ denote the initial contamination rate of recyclable waste²⁰ which is exogenous to the recycler. A recycler charges a per-unit recycling fee of $p_W$ and produces plastic scraps output by removing contamination. The contamination removal process requires effort (E), which has a unit cost of $p_E$. Let $r(\alpha ; E)$ denote the contamination removal function, with $r(0)=0$, $r'(E)>0$, and $r''(E)<0$. The contamination removal function depends on the initial contamination rate ($\alpha$) which determines the upper bound of the removal function: $r(\alpha ;E) \in [0, \alpha )$ $\forall E$.

The contamination removal process separates total waste input (W) into two piles: the removed contamination, r(E)W, which must be disposed of at price $p_D$, and the plastic scraps, $[1-r(E)]W$, which is the output. The output quantity is decreasing in E. The contamination rate of the output (i.e. its quality) is the difference between the initial contamination rate and the the contamination removal rate: $c(E) = \alpha - r(E)$. If its output cannot be sold, the recycler must dispose of it at price $p_D$.

Let $\phi (c)$ be the market price of plastic scraps with $\phi '(c) < 0$. Let $\bar{l}$ be the maximum “marketable” contamination rate: $\phi (\bar{l}) = 0$ and $\phi (c) > 0 \forall c < \bar{l}$, that is, scraps with $c > \bar{l}$ cannot be sold on the market. Let $\tau > 0$ be an exogenous transaction cost. Let $p(c) = \phi (c) - \tau$ be the unit “net price” of plastic scraps. $p(c) < 0$ if the market price is lower than the transaction cost. This is the gain/loss to a recycler from selling one unit of plastic scraps with a contamination rate c. Let $\tilde{l}$ be the “breakeven” contamination rate: $\phi (\tilde{l}) = \tau$ thus $p(\tilde{l}) = 0$. Any contamination rate that is greater than $\tilde{l}$ results in a negative net price.

A recycler can choose to exert no effort and dispose of all its recyclable waste. In this case, there is no output and the profit function is $\Pi ^D = p_WW - p_DW$, which does not depend on effort or the price of plastic scraps. If a recycler chooses to exert effort and remove contamination, its profit function, $\Pi ^A$, has a point of discontinuity at ${E}$ which satisfies $\alpha - r( {E}) = \bar{l}$.

$$\begin{aligned} \Pi ^A(E) = \left\{ \begin{array}{ll} \Pi ^L(E) = p_WW - p_DW - P_EE, &{} E \in [0, {E}) \\ \Pi ^H(E) = p_WW - [\phi (\alpha - r(E)] - \tau ][1-r(E)]W - p_Dr(E)W - P_EE, &{} E \ge {E} \end{array} \right. \end{aligned}$$

If a recycler exerts effort $E < {E}$, the resulting contamination rate is higher than the marketable threshold: $c(E) = \alpha - r(E) > \bar{l}$. The output cannot be sold in the market and the profit function ($\Pi ^L(E)$) depends on E but not $\phi (c)$. Notice that $\Pi ^L(E)$ is minimized at $E=0$ which renders it identical to $\Pi ^D$.

If a recycler exerts effort $E \ge {E}$, then $c(E) = \alpha - r(E) \le \bar{l}$. The output can be sold at price $\phi (c)$ and the profit function ($\Pi ^H(E)$) depends both on E and $\phi (c)$. Whether $\Pi ^H$ lies below or above $\Pi ^L$ at ${E}$ depends on the relative magnitude of $p_D$ and $\tau$: if $p_D > \tau$ then $\Pi ^H( {E}) > \Pi ^L( {E})$ and if $p_D < \tau$ then $\Pi ^H( {E}) < \Pi ^L( {E})$. To find the optimal E, a recycler first finds $E^H$ that maximizes $\Pi ^H$ then compares $\Pi ^H(E^H)$ to $\Pi ^D$. If $\Pi ^H(E^H) < \Pi ^D$, then it will choose to exert no effort and dispose of all W.

1.1.1 Model prediction

Under the assumption that contamination removal technology and initial contamination rate is constant across all recyclers, the decisions of whether to exert effort at all and how much effort to exert depend on the relative magnitudes of disposal cost ($p_D$), effort cost ($p_E$), and the sale price of the plastic scrap ($\phi (c)$). The decision whether to exert effort at all depends on the relative magnitude of the per-unit disposal cost and per-unit loss from sorting and selling marketable plastic scraps (sale price net transaction cost and per-unit effort cost). A recycler will exert no effort only if the per-unit net loss is greater than per-unit disposal cost. The model thus implies that a recycler facing a higher disposal cost is more likely to be selling scraps in the market, even at a loss.

The decision of how much effort to exert is based on the cost-benefit tradeoff of exerting effort. Exerting additional effort allows for a cleaner scrap output which can be sold at a higher price. The marginal cost of exerting additional effort has three components: (1) the accruing effort cost, (2) the cost of disposing the additional contamination removed, (3) the reduction in the amount of salable scraps (with higher contamination rate) which could have been sold at a lower price. Thus, the faster the value of scrap rises with respect to its quality, the higher the incentive for recyclers to exert effort beyond the minimum amount required to produce salable scraps.

Consider a scenario in which initial contamination rate varies across recyclers. This likely have an important implication for the plastic waste trade. Recyclers in higher-income countries with a well-established municipal waste management system that requires the separation of recyclable waste are more likely to have a lower initial contamination rate. This means a smaller amount of effort may be required to achieve a minimal marketable effort level (${E}$). The model thus predicts that, all else equal, recyclers with a low initial contamination rate are more likely to produce and sell plastic scraps.

The ban affects the plastic scrap price function. More specifically, the market becomes less tolerant to contamination. The maximum marketable contamination rate ($\bar{l}$) at least weakly decreases and the price at least weakly decreases at all contamination levels below the new $\bar{l}$.

The model predicts that recyclers that used to sell before the ban may keep selling or shift to disposal, whereas recyclers that chose disposal before the ban will continue with the same option (they will not switch from disposal to selling). As a result, the global quantity of plastic scrap traded decreases, and the total quantity of disposal increases, at least weakly in both cases.

As a result of the downward price shock, some recyclers may be forced to switch from selling to disposal ($\Pi ^H(E^H) < \Pi ^D$ under the post-ban price function). Recyclers that face a low disposal cost relative to effort cost are more likely to do this external margin adjustment. For recyclers that continue to sell (internal margin adjustment), those with a higher disposal cost are more likely to decrease effort and those with a lower disposal cost are more likely to increase effort after the ban.

If a recycler finds that $E=0$ is the optimal choice, and if the resulting profit is negative ($\Pi ^L(0) = \Pi ^D < 0$), then it may be willing to pay for alternative disposal service with unit cost less than $p_D$. In countries where waste disposal is formalized, the disposal capacity may be more inelastic and the price may be more sensitive than in places where open dumping and other informal methods of disposal are widely practiced. For this reason, it may also be possible that $p_D$ rises in response to the ban in some countries, and $\Pi ^D$ generates a greater loss for recyclers. To minimize loss, recyclers may seek alternative disposal options as described above. This corresponds to real-world observations where developing countries caught many shipments from developed countries labeled as recyclable scraps that were just trash or have a contamination rate that far exceeds the legal threshold (Law et al., 2020).

Reclaimer model

1.1 Demand: the reclaimer

We develop a model of a profit-maximizing plastic reclaimer to theoretically explore the determinants of the demand for plastic scraps. A detailed version of this model can be found in Chunsuttiwat (2022).

A reclaimer uses plastic scraps to produce post-consumer resin (PCR) pellets. The output (Y) is sold at price $p_Y$. We assume that there is no quality variation in the PCR pellets and thus $p_Y$ is an exogenous constant with respect to a reclaimer. A reclaimer uses plastic scraps (S), labor (L), and capital (K) as production inputs. The unit cost of labor is w and the unit cost of capital is r. Let us consider a vertically integrated production function scenario. The production of PCR involves two steps. First, the sorting process serves to remove contamination from the plastic scrap input. Second, the granulation process turns sorted plastic scraps into PCR pellets.

In the sorting process, a reclaimer buys plastic scraps S with contamination rate c at price $\phi (c)$. The first step of the production process is to remove all contamination (cS) and produce clean recyclable plastic scraps, denoted by $X = (1-c)S$. The contamination is disposed of at price $p_D$. Suppose that the sorting process only requires labor input. Let h(c, S) be the amount of labor (e.g. labor hours) required to sort scraps quantity S with contamination rate c. For example, if each unit of contaminant requires one unit of labor, then $h(c,S) = cS$. The second step of the production process – the granulation process – converts the output of the sorting process into PCR granules (Y). Since all contaminants are removed in the first step, the granulating function does not depend on the contamination rate. Let $g(X, L, K) = g((1-c)S, L, K)$ be the granulating function.

The profit function of a reclaimer is:

$$\begin{aligned} \Pi ^R = \underbrace{p_Yg(\overbrace{(1-c)S}^{X}, L, K) - wL - rK}_\text {the granulating process} \underbrace{- wh(c,S) - p_DcS - \phi (c)S}_\text {cost of the sorting step} \end{aligned}$$

(13)

The first term in the profit equation is the sales revenue. The second and the third terms are the labor cost and capital cost of the granulation process. The last three terms are the cost of the sorting process: wh(c, S) is the total cost of labor, $p_DcS$ is the total cost of disposal, and $\phi (c)S$ is the total cost of scraps input. A reclaimer chooses c, S, L, and K to maximize profit.

1.1.1 Model prediction

The ban affects reclaimers in two ways. First, it at least weakly decreases the input price (of plastic scraps). Second, it at least weakly increases the output price (of PCR). These changes encourage reclaimers to at least weakly increasing their output which requires input adjustments. Under some assumptions about the shapes of the sorting and granulation functions,²¹ capital, labor, and scrap input quantity weakly increases and the contamination rate of choice weakly decreases. There may be heterogeneity in the magnitude of adjustments in contamination rate and scrap input quantity due to variation in the wage rate. Whether a higher wage is associated with a larger magnitude of adjustment depends on the rate of change of the sorting function with respect to contamination rate and scrap input volume ($h_{cc}$ and $h_{SS}$). There may be heterogeneity in output adjustment overall (which affects input adjustment) due to variation in transaction cost. Reclaimers who face a larger reduction in transaction cost post-ban have a greater incentive to increase output. Variation in granulation technology may also result in variation in reclaimers’ magnitude of adjustment to the ban. This is true even when wage has no effect on the magnitude of adjustment (when $h_{cc} = h_{SS} = 0$).

Change in gradient variables

Let $X_{it}$ be a time-varying country characteristic. The gradient of X is the origin–destination difference as a fraction of the pair average, which is a bilateral time-variant variable. The gradient of X is defined as:

$$\begin{aligned} Grad(X)_{odt} = \dfrac{X_{ot} - X_{dt}}{(X_{ot} + X_{dt})/2} \end{aligned}$$

The coefficient estimates from the gradient regression analyses can be interpreted as, for example, for every 1% an importer’s EPI score decreases relative to an exporting partner, the importer will experience a x% increase in waste imports from the exporter.

The phrase “1% an importer’s EPI score decreases relative to an exporting partner” refers to $\%\Delta EPI_{orig} - \%\Delta EPI_{dest} = 1\%$. This corresponds to an increase of 0.01 in the EPI gradient. Three tables below show three examples of how this 1% decrease in importer’s EPI score relative to an exporting partner may occur.

Example 1: origin EPI = 80, destination EPI = 80, and gradient = 0.

Starting with origin EPI = 80, destination EPI = 80, and gradient = 0. If origin’s EPI remains unchanged and destination’s EPI decreases by 1% from 80 to 79.2, then the gradient increases from 0 to 0.01. If origin’s EPI increases by 1% from 80 to 80.8 and destination’s EPI remains at 80, then the gradient increases from 0 to 0.01. Similarly, if origin’s EPI score increases by 5% from 80 to 84 and destination’s EPI score increases by 4% from 80 to 83.2, then the gradient also increases from 0 to 0.01. All these cases correspond to a 1% decrease in importer’s EPI score relative to an exporting partner may occur. Example 2 and 3 follow similarly, but with different reference EPI scores and gradient values (Table 7).

Table 7 Example 1 of a change in EPI score gradient

Example 2: origin EPI = 80, destination EPI = 70, and gradient = 0.13 (Table 8).

Table 8 Example 2 of a change in EPI score gradient

Example 3: origin EPI = 70, destination EPI = 80, and gradient = − 0.13 (Table 9).

Table 9 Example 3 of a change in EPI score gradient

IHS estimation results

See Tables 10 and 11.

Table 10 IHS estimates of the gravity model with country characteristics

Table 11 IHS estimates of the gravity model with gradient characteristics