(1)
Master Thesis
The Escalating U.S/China Trade War and its Indirect Impact on Trade:
Evidence for Trade Diversion
By Michael Barry McShane (S3736547) Email: m.mcshane@student.rug.nl Supervised by Dr. Tarek M. Harchaoui
Co-assessor: Dr. Giampaolo Lecce Abstract
In this paper we investigate the immediate trade diversion effect that has resulted from the escalating trade war between the US and China. The diversion effect is an important consideration in trade policy as it deducts from the trade wars intended goal of decreasing Chinese imports to the U.S. by simultaneously increasing imports from countries outside the bilateral trade war. We take advantage of the unprecedented and unconventional use of tariffs on Chinese products to formulate the treatment needed within a Generalized difference-in-difference estimation framework. Based on a subsample that uses monthly product-level import data from the top 35 importers to the U.S we discover (1) Chinese product imports affected by the tariffs have on average decreased by 26 percentage points relative to products that remained unaffected (2) All products are not created equal; the type of product targeted by the tariff played a key role on the effectiveness of reducing Chinese imports, with Capital, Consumption and non-differentiated products showing the largest declines (3) Trade diversion effects has on average increased imports from 34 of the top US importers by 3 percentage points (4) Trade diversion effects range considerably amongst regions (5) Trade Diversion effects are strongest for Capital, and Final goods. Our paper has strategic implications on the conduct of trade wars for policy makers, whilst adding considerable evidence on the product-level behaviour of international imports in response to a tariff, to the trade literature.
(2)
1. Introduction
Those who cannot remember the past are condemned to repeat it1 - is perhaps a platitude too commonly
cited to excite, however through its commonality the quote suggests an ingrained relevance in reality, for if
history does not repeat itself, it at the very least shares a rhythm across time. On the 22nd of January 2018,
the Trump administration approved tariff safeguard measures aimed at reversing the damage incurred on the solar panel and washing machine industries, from import competition (Bown, 2018). The safeguard measure mostly in the form of tariffs came by recommendation from the United States International Trade Commission (USITC) under Section 201 of the Trade Act of 1974. An act that enabled U.S industry an avenue to address unfair competition practices from oversea competitors. This approval was a historically significant shift in trade policy, as it was first time an industry had utilised section 201 since 2001, let alone
for the president to approve the resulting recommendation2 (Bown 2017).
The rapid successive waves of tariffs that came to follow unravelled more than half a century of cross-border U.S led efforts to reduce international trade barriers, and mostly hit by this shift in trade policy was China. Utilising the Trade Act of 1974, the Trump administration self-initiated an investigation under Section 301 to determine whether China had been engaging in unfair trade practices. This was an unprecedented move as section 301 was not an import restricting legislative tool comparable to section 201, rather it was intended as a vehicle for U.S industry to gain exporting footholds into disputed foreign markets by threat of unilateral U.S tariffs. Section 301 previously fell out of favour as the international optics did not favour the U.S. undertaking the roles of investigator, prosecutor, jury, and judge on what is a bilateral dispute (Bown 2017). More importantly by 1994 the establishment of the rules-based World Trade Organisation (WTO) and the associated implementation of an effective international dispute settlement system resulted in the overtly aggressive Section 301 mechanism being rendered obsolete. The Trump administrations purposeful bypassing of the international trade co-operative that is the WTO in favour of US centric Section 301 legislation, led to a tit for tat escalation, and accusations of political motivation between the two largest economies in human history. The resulting trade war led to the average volume weighted tariff applied on all Chinese exports to the U.S. to increase from 3.1% in January 2018 to 18.4% in September 2019. Chinese retaliatory measures increased tariffs on U.S exports to China from 8% in January 2018 to 21.8% in September 2019.
So why China? The U.S Trade Deficit finished on $853 Billion or 3.98% of total GDP for the year 2019. This is not a new development; the U.S. has run a trade deficit since 1975 with a notable expansion from the early 2000’s onwards as noted in Figure 1. The rapid expansion in the U.S trade deficit also
1 A quote most likely attributed to writer and philosopher George Santayana
(3)
happened to coincide with Chinas admission into the WTO on the 11th of December 2001, with the trade
balance deficit between the two nations growing from an already significant $83 billion in 2001 to a substantial $375 billion in 2017 (Figure 2). The ongoing and growing trade deficit has garnered attention and calls of resolution from politicians, commentators, and economists alike, and when we apply the growing trade deficit with China as a proportion to the total trade deficit of the U.S. as we did in Figure 3, it is easy to understand why. Donald Trump’s administration has actively voiced blame for the decline in American manufacturing and loss of middle-income occupations on the deteriorating trade position of the United States, to which China often accounts for almost half of. It is worth noting with an empirical glance that the trade war which began in July 2018 has resulted in noticeable improvements for the US/China trade balance as noted in both Figure 2 and 3, although this improved bilateral trade position doesn’t appear to transition to the U.S. overall trade position as seen in Figure 1, thus suggesting a neutralising counterforce, which this paper will investigate as trade diversion.
(4)
The US/China Trade War is a unique development by the standards of post-war global economics, although the political and economic rationale tying this confrontation together are undoubtably an established feature of economic past. Our aim in this paper is to quantify the degree of success this unprecedented return to protectionism has had on improving the US trade position. We achieve this by investigating any changes in import behaviour the top 35 U.S. importers underwent during the US/China trade war, to determine whether the tariff induced reduction in Chinese imports has led in part, to an increase of imports elsewhere. This movement of imports from one country to another is known as trade diversion, and it undermines the pursuit of an improved trade balance and the economic autarky at the heart of Donald Trump’s 2016 campaign slogan “Make America Great Again”.
Figure 3
Using a multi period, multi treatment Generalized Difference-in-Difference (GDD) two-way fixed effect model, the paper aims to measure the success of the trade war by quantifying the direct impact Trumps tariffs has had on Chinese import volumes, and the indirect effects it has had on global import volumes. We have selected this framework due to the low data requirement relative to the more commonly used models
within the trade literature3, therefore allowing us to undertake a concurrent analysis so early into the trade
war. By using a Difference-in-Difference (DD) type model, we can compare changes in import volumes of HTS8-level products receiving tariff treatment relative to those that have not, thus creating the treatment and control set up required for the model. A fundamental condition of this type of model is the exogeneity of the treatment. A case of heightened concern as issues of endogeneity are commonplace whenever trade policy 3 One of the mostly commonly used models for estimating the real impact of trade policy is the gravity equation model. Due to the model
(5)
is used to regress trade data (Urata and Okabe, 2014). Fortunately the requirement of exogeneity is easily achieved for importing countries other than China, as U.S. administered tariffs on specific products within Chinese imports will be identified as a positive exogenous demand shock for those same specific products when imported from a third party country that did not receive the tariff, an example being Vietnam. Regarding China the condition of exogeneity is more difficult to prove as the trade policies are specific to the country. Nonetheless the concern is somewhat neutralised by the escalating tit for tat nature of a
politically driven trade war, an ad hoc selection process for the composition and size of tariff groups4, and
the overall specificity of the unilateral tariff policy to the Trump administration, all of which in tandem suggest a suite of tariff policy unparalleled by existing trade trends (Meinen, Schulte, Steinhoff and Cigna 2019).
By utilising country-product data on China and 34 of the other top importers to the United States which we will henceforth name “third-party” countries, we can investigate whether the reduction in Chinese
imports has taken the form of a) trade destruction5 and thus fulfilling the intended goal of protecting domestic
U.S industry and improving the US trade balance, or b) trade diversion, whereas trade has simply redirected from efficient Chinese sources to less efficient sources elsewhere to the benefit of other countries and to the detriment of the U.S.. Deriving the contribution of both trade destruction and trade diversion from the overall decline in trade from China allows the paper to provide insight as to whether the unquantifiable social, political, and economic costs of the trade war can be at all justified by the quantifiable protectionist outcome of an improved trade balance. We also further our analysis by classifying the HTS8 product level data into product type and regional subsamples, doing so enables us to investigate whether trade destruction and trade diversion effects are influenced by the product type of the tariffed good or the region from which it originates. The data required to achieve our aim has been obtained from the USITC in the form of monthly 8-digit Harmonised Tariff System (HTS8) product-country codes, covering imports into the U.S. from January 2015 to December 2019, with 35 of its top trading partners, overall representing 92% of the U.S total trade. HTS8 codes have the benefit of being both the official measurable identifier of traded goods whilst being the level of aggregation tariffs are applied.
In this paper, we estimate a 26 percentage points (pp) decline on the Chinese import of products affected by tariffs relative to those that are not. We also establish a correlation between the strength of the tariff and the extent of import decline the product experiences. Furthermore, there is significant statistical evidence that over a third of the decline in Chinese imports resulting from the 25% tariff is offset by trade 4 We suggest randomness in the composition of tariffs for the following reasons: a) the tariffs do not favour any product type b) The import
volumes of each tariff appeared to focus more on size than composition e.g. $50, $150, and $200 Billion, and c) The Office of the US Trade representative has retracted thousands of HTS10 product-level tariffs, this high level suggest a lack of foresight in their development. https://ustr.gov/issue-areas/enforcement/section-301-investigations/ search
(6)
diversion effects from third-party countries, with the remaining two thirds being explained by bilateral trade destruction. We also find evidence the product type of the tariffed good plays a strong role in the level of trade destruction that occurs, whilst a lesser but still significant role in the indirect trade diversion effect. Finally, we find that the level of trade diversion differs greatly across geographical regions, with low income Asian countries and Taiwan experiencing the largest gains. Of all continents, Asia comes out the biggest beneficiary of the trade war, although the EU and North America do see trade diversion effects. We have also robustly checked our model against the influence of price on trade volumes, and found our initial statistically significant findings of trade diversion effects on the 10% tariff was indeed explained by an abnormal increase in price of the tariffed selected goods, whilst price played a statistically insignificant role for the 15% and 25% tariffs, nonetheless the overall trade diversion effect across all tariff groups remain the same due to the small role the 10% tariff played in the overall trade war.
The remaining structure of the paper is as followed: Section 2 is the literature review that covers the diverse nature of trade literature and the position of our research within it. Section 3 is the quantitative framework, where we establish the empirical foundation of our research question, and map out the reasoning, methodology and execution of our chosen difference-in-difference model. Section 4 is our implication and caveats section; here we will demonstrate the relevance of our research in the trade literature and its implications for policy whilst bearing a conscience effort to acknowledge potential pitfalls. Finally, Section 5 is our conclusion.
2. Literature Review
The effectiveness of tariffs is an ongoing discussion within the trade literature. This is perhaps a result of the diversity in their implementation, their permanence throughout history, and the fact they are highly consequential by design. Although on a more philosophical note, the ongoing prevalence of trade policy in the literature may very well stem from the cause and effects of trade policy being intertwined with the unpredictability of politics and the subjectiveness of economics, a marriage of unbounded complexity.
(7)
war in many ways is the antithesis to a preferential trade agreement.6 One such study is by Sologoa and
Winters (2001) who extend the basic gravity model7 to include dummies that identify the annual trade flows
between trade block members and those from outside the block, and applied it to the swathe of PTA’s
emerging globally throughout the 90s8. In the context of a PTA, a trade diversion effect takes the form of
increased intrabloc trading coming at the expense of lower trade from outside the block. Overall Sologoa and Winters finds mild evidence for trade diversion amongst two of their nine regressed PTA’s, and evidence of trade creation in none. Another common method used to answer the same research question is demonstrated by Fukao, Okubo and Stern (2003). They develop a conceptual framework based on a partial-equilibrium model of differentiated products under monopolistic competition, and then implements the resulting equation empirically through a fixed effect panel model using US import data at the HS2 level. Focusing on the North American Free Trade Agreement (NAFTA), the authors find statistically significant evidence of trade diversion occurring in 15 of the 70 regressed HS2 product classes, with the strongest evidence in textiles and apparel. The results of Fukao et al (2003) greatly contrasts the work of Sologoa et al (2001) and other renowned economists such as Clausing (2001) and Krueger (1999), who find no evidence of trade diversion resulting from the NAFTA agreement.
A paper by Romalis (2004) identifies two key reasons as to why the trade literature has failed to cohesively identify trade diversion effects on a PTA as relevant and widely studied as NAFTA. First, high levels of aggregation masks the regression from the multitude of product level shocks occurring at the disaggregated level (Clausing, 2001), the second reason is the inability of the empirical models that dominate the literature to accurately distinguish a trade shock incurred by NAFTA from concurrent shocks within the
same period.9 To overcome these issues the author disaggregates the data typical of the literature10 to the
more detailed HS6 product-level, and calculates the demand and supply elasticities of each individual US import product using a difference-in-difference framework. The author then replicates this framework for EU imports of the same products, to enable a reference database of product elasticities that can account for changes in US import volumes that have resulted from a change in production costs rather than the PTA. The resulting framework enables a product level analysis that can factor in the unique product-specific application of the PTA and the concurrent international trends influencing trade volumes, to determine the PTA’s actual effect on the import of a product. An example would be the NAFTA three levels of reduced
tariffs11 applied to Mexican imports to the USA, the DD approach at the HS6 level enables all three tariff
6 A PTA will lower tariffs and trade barriers for its members, and usually stems from a pre-existing trend of growing economic/political
integration. A trade war aims to increase tariffs and barriers to trade, and stem from a pre-existing trend of political/economic grievance.
7 In the basic gravity model, trade flows between countries are estimated on factors such as their economic weight (GDP, population),
geographic size (landmass, access to ports) and trade barriers (cultural similularities, distance, ease of travel). The empirical robustness of this model has made it the workhorse in the trade.
8 The Common Market of the South (MERCOSUR) in 1991, ASEAN Free Trade Area (AFTA) in 1992, North American Free Trade
Association (NAFTA) in 1994, are just a few examples.
(8)
levels (which includes the no change control group essential to identify concurrent trends in growth) to be estimated separately whilst the inclusion of comparable EU import elasticities accounts for concurrent production trends shared between the two trading blocks. Romalis finds evidence of trade diversion, and attributes this contrary result to the inability of other models to factor in the rapid growth in third-party imports that would have occurred in the absence of NAFTA.
Another article that utilises the difference-in-difference approach whilst embarking on a more similar research question to our own, is written by Meinen et al. (2019). Upon utilising a basic DD design, the authors determine that US imports of Chinese products affected by Trumps tariffs have grown on average 30 percentage points less than those unaffected. By May 2019 they determine no significant evidence of trade diversion had materialised in third party countries, thus concluding trade destruction as the only explanation behind the drop in Chinese trade. This papers results differ from our own and can be partly explained by the following: 1) the level aggregation that occurred on the product-country data. The model within the paper regresses import data at the HS6 level, this is problematic as the tariffs themselves are applied at the HS8 level, this discrepancy as discussed in the data section of our paper does not come without compromise. 2) the data only covers until May 2019, ultimately resulting in less than one years’ worth of tariff data, which is especially problematic as it places far too much explanatory power on the largest tariff which was designated to be upgraded from 10% to 25% in May 2019, the month their analysis ends. We will discuss later in the paper the import volatility surrounding this period, as firms pre-emptively adjust their trading in anticipation. 3) The dependant variable is the unweighted growth rate of the HS6-level product, this will result in the underrepresentation of larger HS6 products (by import volume) that drive the change in import volume, whilst overrepresenting smaller more volatile HS6 products. A few other papers of the trade literature that utilise the difference-in-difference approach are; Lloyd and Solomos (2019) estimation of the British 1932 General Tariffs effect on domestic industry, they found the tariffs led to substantial short-run and long-short-run positive effects for both productivity and net output; Fotopoulos and Psallidas (2009) estimates the effect the adoption of the Euro had on trade, they find significant trade creation within the European Monetary Union although no evidence of trade diversion from outside the block.
(9)
Estimating an overall 25 pp decline in Chinese imports to the U.S. being compensated by trade diversion
effects in the magnitude of 48c on the dollar for the first phase12 of tariffs and 33c on the dollar for goods in
phase two, by the 2nd quarter of 2019.
Another trade phenomenon in the literature building on the concepts of trade destruction and trade diversion, is trade rerouting. Trade rerouting occurs when the tariff affected country uses a third-party country as a vehicle to avoid tariffs. This is achieved by exporting a good to a third-party country, so it can be re-exported to the original country of destination. Trade rerouting in relation to the US/China trade war is worse for the US than trade diversion, at the resulting increase in third-party imports is not offsetting the decline in Chinese imports, it is Chinese the imports. Using the US/China trade war and Vietnam as the acting third-party country, an example would be as follows (assuming ceteris paribus and full trade rerouting): China stops exporting a good to the U.S, China exports the good to Vietnam instead (trade deflection), while Vietnam exports that good to the U.S. (trade diversion). The example results in a correlated increase in trade diversion and trade deflection, whilst U.S. imports and Chinese exports of the same good remain unchanged (trade rerouting). A paper by Liu and Shi (2018) provides evidence on trade rerouting occurring upon U.S. implementation of anti-dumping duties on selected Chinese products between 2002 and 2006. Using product‐country‐month‐level data within a difference in difference framework, the authors discover the correlation between trade diversion and trade deflection is strongest upon third-party countries which are geographically closer to China, and especially those with large ethnically Chinese population. They argue the stronger correlation stems from the need to minimise the risks involved in rerouting trade as it involves falsifying country of origin certificates (CO’s) and breaking the Rules of Origin (RoO) criteria established in the Free Trade Agreement (FTA)/Preferential tariff agreement (PTA), the third-party country is part of, and thus jeopardizes their privileged status. Secondly their work finds less-differentiated products are more likely to be rerouted as the CO’s can be more easily manipulated due to the difficulty of tracing the origin of a non-unique or unrefined product.
(10)
first would appear and progresses the case for trade diversion. This reasoning can extend to China although the import/export divergence is less significant due to its commonality and therefore can only put forth a mild case for trade deflection, and therefore a mild case for trade rerouting.
Figure 4: Import/Export data with Third-Party Countries for both China and U.S.
Although Liu and Shi (2018) found strong evidence of trade rerouting, it is in fact an uncommon occurrence as discussed in Felbermayr, Teti and Yalcin’s (2019) paper. Liu and Shi’s paper covered a period of an underdeveloped China which only recently entered the WTO, a period of adjustment and unsettled markets rifled with opportunism. Whether this level of trade agreement defiance could exist now is debateable. Felbermayr et al (2019) extends this mode of thinking by researching whether the burdensome and costly Rules of Origin criteria placed upon firms are even necessary to prevent trade rerouting. They answer this question by estimating the profitability of trade rerouting for every country pair within their dataset. They find that once one factors in the cost of external tariffs and transport, 86% of all bilateral product level comparisons within an FTA is unprofitable, and therefore cannot incentivise trade rerouting. They also find this portion increases to 98% for the unilateral trade agreements typically drafted between poor and rich nations.
(11)
duty price of tariffed Chinese products did not fall, suggesting a full pass-through effect of the tariff onto the duty-inclusive price. This suggests that it is American consumer and firms who paid for the raised tariffs on imports to the tune of $51 billion or 0.27% of GDP. Although once gains for both domestic producers and tariff revenue are accounted for, the net loss in real income decreased to $7.2 billion, with the majority of the impact effecting those targeted directly by the Chinese retaliatory tariffs, which are mostly republican counties. Another general-equilibrium paper that uses varying assumptions on competition to determine the welfare cost is by Balistreri, Bohringer, and Rutherford (2018). They find when the model is applied under the assumption of monopolistic competition and frictionless entry/exit of firms, that the welfare cost to the US increases substantially to $124 Billion USD or 1% of total private consumption, the authors also find long-term trade diversion effects across multiple regions, especially in Europe, at the expense of both China and US. Amiti, Redding and Weinstein (2019) through the use of more standard economic methods, reaffirms this previous analysis’s by confirming the full pass through of the tariff onto consumers, and an aggregate real income loss of $6.9 billion in the first 11 months, a number they acknowledged was increasing by the month as the trade war continued.
(12)
By bringing together the rich tapestry of research we can begin to understand the complexity and far reaching consequences of tariffs, and furthermore mend the inconclusiveness of the topic. The US/China trade war is a highly consequential event and justifiably demands attention from the trade literature, even so the published literature remains scarce. Popular models such as the gravity equation are unable to estimate the trade flows (and therefore trade diversion) because the trade war is to recent for the heavy data demands of the model, whilst the general equilibrium models of Fajgelbaum et al. (2019) and Balistreri et al. (2018) differ in results due to the market assumptions essential to their method. We propose reinforcing the existing literature by introducing a model that is free of market assumptions and is already established in fields outside of trade. The DD model and our GDD variant has already been tried and tested by Romalis (2004), and with the lessons learnt from that paper we can build on more recent DD approaches by Meinen et al. (2019) and Liu et al. (2018) to uncover trade diversion effects hidden within the aggregated data. Furthermore, by uncovering a trade diversion effect we can discount the perceived success of the trade war (lowered Chinese imports), and add onto the tapestry of welfare costs, and retaliatory measures, already established in the literature
3. Quantitative Framework
3.1. The model and the identification strategy
(13)
𝐥𝐧 𝑰𝑴𝑷𝒄,𝒊,𝒕𝑼𝑺 = 𝜶 + 𝛃 𝝉𝒊,𝒕𝑪𝑵+ 𝜸𝒄𝒊+ 𝜸𝒕+ ∈𝒄,𝒊,𝒕 (1)
whereas subscript c denotes the country of origin; subscript i denotes the product at the HS8 level; and
subscript t denotes time (month and year). Therefore 𝐥𝐧 𝑰𝑴𝑷𝒄,𝒊,𝒕𝑼𝑺 represents the natural log of the monthly
import of product i to the U.S from country c at time t. 𝛃 𝝉𝒊,𝒕𝑪𝑵 is a multi-treatment, multi-time ordinal dummy
variable that is applicable to the product of all countries, and switches to 1 from the month the US applies the 10% tariff on product i imported from China (CH), 2 from the month the U.S. applied the 15% tariff, and
3 from the month the US applied the 25% tariff. 𝜸𝒄𝒊 is a country-product fixed effect that can control for
time-invariant characteristics of a product within a country, such as trade policy or the capacity of the industry. The Fixed effect works by filtering out average country-product specific growth rates, an important
addition given the potential for differing trends given the number of the ID’s in our panel. 𝜸𝒕 is a monthly
fixed effect accounting for time specific shocks to our analysis. To address the issue of inconsistent standard errors, we apply the Arbitrary Variance-Covariance Matrix method outlined by Bertrand et al. (2004), this essentially involves clustering the standard errors at the HTS8 8-digit product level.
Our model is a newer variant of the difference-in-difference employed by Meinen et al. (2019) and Liu et al. (2018). We have opted for the more flexible GDD two-way fixed effects model as specified by section 6.5.2 of Imbens and Wooldridge (2009). The traditional DD approach is undoubtedly intuitive although its inability to accommodate any complexity beyond the two-group two-period model, and the
requirement of a very strict parallel trend assumption13 makes it unsuitable for most real-world
applications14. Due to the escalating nature of the U.S/China trade war, there are multiple treatment groups
varying in strength over multiple periods in time, and although the DD approach can address any single one of these groups effectively, it is unable to approach the problem holistically. A key feature of the GDD model is the inclusion of a two-way fixed effect that can trace out group effects (HTS8-Country) and time trends (monthly) which are caused by underlying differences in unknown covariates across groups and time periods (Wing, 2018), and controls for them even though they are not explicitly accounted for, therefore lowering the data requirements of our model (Magee 2008). It is implied the outcome of each group, once two-way fixed effects are accounted for, should differ by a predictable and consistent amount in the absence of treatment for every period. The GDD also relaxes the parallel trend assumption to a group specific common trend assumption, enabling the treatment groups and control group to maintain different baseline trends
13 The Parallel Trend Assumption: Is the assumption that both the treatment and control group have the same trends prior to treatment and
would have maintained the same trends if the treatment were never administered.
(14)
between units as long as they maintain consistency in the absence of treatment (Kim, 2016), this is an important development as the parallel trend assumption rarely upholds in reality. This relaxation to the common trend assumption allows for far more variation in research design compared to the traditional DD as illustrated by Lee’s (2016) multi treatment example that has shifted the outcome on multiple occasions, an illustration most fitting to our tariff varying trade war. Aside from the common trend theory, is the requirement of exogeneity upon the treatment itself. Amiti et al (2019) argue that Trump’s election came as a shock to most analysts and was not anticipated by nearly all sectors within the economy. This is furthered by our discussion in the descriptive evidence, whereas the short duration between tariff announcement and implementation, and the lack of certainty surrounding any development, led to firms having little or no time to prepare for the new tariffs thus maintaining exogeneity.
3.2. The source data and the clean dataset
Our main data source is the United States International Trade Commission (USITC), an independent bi-partisan federal agency of the U.S. with the role of providing expertise trade advice to both the Executive and legislative branch. Through their Dataweb portal we can obtain U.S tariff and trade data on a per country basis at the required HTS8 level. In total a selection of 35 countries have been selected for the regression
based on being the top importers to the United states in 201715. Combined, these 35 countries represent
93.3% of 2019 imports into the U.S
The Harmonized Tariff Schedule 8 (HTS8) level data has been selected as the level of detail for our model, as it is the level of disaggregation that can link both tariff product codes and trade data codes too one unique identifier. Tariffs are applied at the HTS8 level whilst trade data can aggregate anywhere from HS2 to HTS10. In the example presented in Table 1, we demonstrate the importance of using HTS8 data instead
of a more easily manageable aggregate level such as HS6, HS4 or SITC16 codes, which are commonly used
in the literature17. In the Table 1 example there is two HTS8 product codes for two types of Frozen haddock.
One for the fillets (0304.72.10) and another for the other parts of the fish (0304.72.50). Combined they are simply identified as frozen haddock meat at the HS6 level (0304.72) or General fish meat at HS4 (0304). The issue arises when a tariff is applied to one HTS8 product code and not the other, as has been done to frozen haddock other and not fillets. At the HS8 level we can run a regression and treat the product codes 15 Australia, Austria, Belgium, Brazil, Canada, Chile, China, Colombia, Denmark, France, Germany, India, Indonesia, Ireland, Israel, Italy,
Japan, Malaysia, Mexico, Netherlands, Philippines, Poland, Russia, Saudi Arabia, Singapore, South Africa, South Korea, Spain, Sweden, Switzerland, Taiwan, Thailand. Turkey, United Kingdom, Vietnam
16 Standard International Trade Classification (SITC) codes are commonly used to compare industries across countries and are constructed by
grouping HS6 product level data.
(15)
individually, although at the HS6 level a decision needs to be made as whether or not the aggregated product code is identified with a tariff or not, or excluded altogether, regardless a compromise had to be made.
Table 1. Difference between the HTS8 and those Utilized in the Literature and Implications
Our database is represented in Table 2 and takes the form of Panel Data. With a monthly time-series spanning January 2015 to December 2019 and a Country-Product (HS8) unique ID. An example of a panel ID in our dataset would be Australia – Frozen Haddock fillets. Our dataset contains a total of 2,602,644 unique Country-Product panel IDs across 35 countries and is represented by (2) Reduced Dataset in Table 2. To control for simultaneous direct and trade diversion effects occurring outside the U.S/China trade war, we have created a reduced dataset by dropping all products associated to tariffs targeted against multiple
countries, this includes Solar Panels, Washing Machines and Aluminium18- see (3) Reduced dataset. Due to
the disaggregate nature of HTS8 product level data, there is a high proportion of zeroes in the database, for example the (3) Reduced Dataset contains 12.55 million observations in total, 62.2% of which are observations of zero. This therefore runs a risk of very low trade volumes being overrepresented within the dataset. As a result, we created a subsample of consistent Country-Product code time-trends with no instances of zero in their monthly import data throughout the duration of 2015 – 2019. This subsample contains a more manageable 1.97 Million observations between 355,800 country-product IDs, at the cost of losing an additional 21% of US imports – see (4) Subsample at the HTS8 level. We acknowledge the creation of the subsample will lead to a selection bias as it removes Country-Product IDs that have started/ceased trading during the regression period, traded in very low volumes, or sporadically traded in large purchase goods such as military equipment and infrastructure.
18 Dropped Product codes: Energy products: HS4 (2709 – 2716), Iron/Steel products HS4(7201 – 7326), Aluminium products HS4(7601 –
7616), Washing Machines: HS4(8450), Solar Panels: HS6(854140)
HS4 HS6 HTS8 Chinese Imports to USA Tariff
0304 0304.72 0304.72.10 $6,900,000,000 None
0304.72.50 $56,000,000,000 10% in Sep18 and 25% in May19
$62,900,000,000
(16)
Table 2. U.S. Imports: Data Cleaning Process and the Resulting Reduced Datasets/Subsamples
We nonetheless apply (4) Subsample at the HTS8 level as the main dataset for our model as it more accurately reflects the Country-Product pairs of interest, those of established industries more sensitive to price. The rationale behind the removal of zero trade observations is two-fold: first, it is in the interest of our research question to capture the change in imports in response to a tariff, for this to be achieved it is essential the common trend assumption is not violated by omitted variables uncaptured by the two-way fixed effects, the inclusion of import data with instances of no monthly trade data may very well violate this assumption by introducing omitted variables (Wing, 2018) in the form of: administrative errors in paperwork, changes
in policy for a regional specific good, and large purchases in goods that are insensitive to price19. Secondly
and more importantly, a high proportion of zeroes in a dataset can lead to large bias due to the
19 Military equipment, large building equipment and telecommunications infrastructure are some examples.
Year Country-Product Imports(Billions) IMP/Total IMP Country-Product Imports(Billions) IMP/Total IMP (1) Complete Dataset: All imports
2015 - 483 21% - 1,766 79%
2016 - 462 21% - 1,724 79%
2017 - 505 22% - 1,835 78%
2018 - 540 21% - 2,001 79%
2019 - 452 18% - 2,046 82%
(2) Reduced Dataset: Top 35 countries
2015 121,968 483 21% 2,480,676 1,601 71% 2016 121,968 462 21% 2,480,676 1,563 71% 2017 121,968 505 22% 2,480,676 1,656 71% 2018 121,968 540 21% 2,480,676 1,807 71% 2019 121,968 452 18% 2,480,676 1,874 75% (3) Reduced Dataset: Top 35 countries - Products Removed
2015 114,588 464 21% 2,323,236 1,391 62% 2016 114,588 445 20% 2,323,236 1,392 64% 2017 114,588 488 21% 2,323,236 1,445 62% 2018 114,588 521 21% 2,323,236 1,562 61% 2019 114,588 437 18% 2,323,236 1,641 66% (4) Subsample at the HTS8 level: Top 35 countries - Products Removed - Country-Products with zero removed
2015 46,008 428 19% 309,792 997 44% 2016 46,008 408 19% 309,792 992 45% 2017 46,008 447 19% 309,792 1,033 44% 2018 46,008 477 19% 309,792 1,118 44% 2019 46,008 401 16% 309,792 1,174 47% (5) Subsample at the HTS10 level: Top 35 countries - Products Removed - Country-Products with zero removed
2015 46,008 301 13% 309,792 646 29% 2016 46,008 282 13% 309,792 640 29% 2017 46,008 307 13% 309,792 669 29% 2018 46,008 325 13% 309,792 729 29% 2019 46,008 273 11% 309,792 764 31% Source: Authors calculations based on data sourced from the USITC Dataweb portal, Feb 2020. Country-Product column represents the number of unique Country-HTS8 identifiers within a reduced dataset for the given year. Column Imports(Billion) represents total imports in Current USD Billions for a reduced dataset for a given year. Column IMP/Total IMP represents the Imports(billion) column of a reduced
dataset, as a proportion of total(Complete dataset) US imports for a given year.
(17)
overrepresentation of zeroes in the resulting regression, whilst adding a volatility typically associated to low
level trading.20 An alternative is to undertake a strong transformation of the model that can compensate for
the high proportion of zero, although this within itself can add a bias. We address these issues within the robustness section of the paper. Our last subsample disaggregates the (3) Reduced Dataset into the HTS10 product-level. This subsample is used to robust test our model against fluctuations in price being the driving force behind Country-Product import changes. HTS10 is used as it is the only level of aggregation that contains a complete price dataset.
Tariffs data is obtained from section 99 of the USITC January 2020 publication: Harmonized Tariff Schedule of the United States (2020) Revision 1. This publication is the reference guide for US importers and exporters as it contains all US product codes as well as accompanying duty/tariff information. In a normal year, the schedule would be published and then revised once or twice depending on the circumstances. In 2019 it was revised 20 times to accommodate the escalating trade war. I have selected and maintained the January 2020 publication as it contains the up-to-date trade information relevant to my time-series and nothing more. Anything beyond January 2020 adds no benefit whilst running the added risk of niche tariff exemptions or inclusions, which are typically registered at the HTS10 level, going under the radar in my analysis. The trade war applied tariffs in waves. Each successive wave containing a group of products defined at the HTS8 level, all linked together with a tariff wave HTS8 identifier beginning with 9903.88. Below in Table 3 is a detailed representation of the successive waves.
Table 3. List of U.S. Tariffs applied to Chinese Imports
20 Growth figures around the 0 mark can dramatically bias the dataset. If the import of a product jumps from $10 one month, to $10 million the
next, to $500,000 the month thereafter, than it would seriously bias the results by portraying excessively large growth rates on what is a relatively low trade volume compared to the rest of the dataset. This type of volatility is common at the HTS8 and HTS10 product level, and an example of its occurrence would be one-off (or seasonal) large stock purchase of a very niche product.
Tariff Wave ID Tariff Strength Date Applied Products Covered USD (Millions) % of total Imports
9903.88.01 25% 6/07/2018 818 $32,185 6.37% 9903.88.02 25% 23/08/2018 279 $13,789 2.73% 9903.88.03 10% 24/09/2018 Increased 25% 10/05/2019 9903.88.04 10% 24/09/2018 Increased 25% 10/05/2019 9903.88.15 15% 1/09/2019 3,229 $101,045 20.00% 9903.88.16 15% Withdrawn 542 $151,087 29.91% 10,638 $486,778 96.35%
Source: Authors’ Calculations based on 2017 Chinese import data which is sourced from the USITC DataWeb Portal, all in Current USD. Tariff data retrieved from section 99 of the USITC January 2020 publication: Harmonized Tariff Schedule of the United States (2020) Revision 1
(18)
A major takeaway from analysing the successive tariffs is they differ in size, timing, and scope quite dramatically, whilst covering a substantial 96.35% of all Chinese imports. The remaining 3.65% of imports left unaffected by tariffs would have made too small a control group for a difference in difference regression,
thankfully tariff wave 9903.88.16 which was planned for the 15th of December 2019 was withdrawn three
days before implementation, thus enabling us to add it to the control group. To overcome the issue of multiple treatment periods (tariff implementation dates) and varying degrees of treatment strength (10%, 15%, 25%), whilst accounting for changes of tariff strength within treatment periods, we utilised a GDD model. From this point onwards we will refer to the tariff waves by their last two digits, as an example tariff wave 9903.88.01 will henceforth be named wave 01.
Although the U.S. has applied tariffs to a total of 10,638 different products at the HTS8 level, only 9,266 have traded for a minimal of one month between 2015 and 2019. The remaining 1,372 represent products that have not been imported from China once during the five years. For this reason, we kept the product codes within the database but have disassociated them with their corresponding tariff wave and reregistered the products as: no tariff status. We reason that if we are measuring tariffs as exogenous shocks to product imports from both China and third-party countries, then the intended effect on Chinese imports cannot transpire as it never changes from zero. In contrast, third-party country importers of the same product type cannot experience the exogenous shock of tariffed Chinese competition, because the Chinese competition simply didn’t exist to begin with. This is effectively the equivalent of no tariff being applied, and we have, therefore, treated it accordingly within our analysis.
3.3. Documentation of Facts
Our research question covers the US/China trade war from the first Section 301 induced tariff wave against China (tariff wave 01) in July 2018, to the seven subsequent tariffs aimed at China until December 2019 (Table 3). The US Trade war began even earlier on February 7, 2018 when the US imposed broad safeguard tariff measures against all imports of solar panels and washing machines, this was followed by the Section
23221 tariff against Steel and Aluminium products from selected countries22, including China. We have
restricted our research question to China and the U.S. and therefore have dropped Steel, Aluminium, Solar Panel and Washing Machine related product codes as the trade destruction and diversion effects will be distorted by the multilateral effects of the tariffs.
(19)
Figure 5 provides a timeline of the escalating trade war specific to China and the U.S. as outlined in Table 2 and described earlier. Using a weighted average tariff against total imports we can evaluate the escalation of a tariff wave relative to total imports. The unprecedentedness scale of the trade war by modern economic standards is truly reflected in Figure 5, as from February 2018 the 3.1% average tariffs quickly escalated to 18.4% in September 2019, a period of only 19 months. With the strongest escalation of 5.6% occurring on May 10, 2019 when Tariff waves 03 and 04 increased from 10% to 25%. The red line represents Chinese retaliation tariffs assembled by the Peterson Institute for International Economics (PIIE).
Figure 5. A Timeline of the Escalating US/China Trade War
The year 2018 contains 3 tariff waves. Although they occurred in very quick succession and can easily be summed up as a 25% tariff on $46 billion worth of imports in July/August and a 10% tariff on $190 billion in September. We refrain from using 2018 numbers in the proceeding descriptive evidence as the year contains very volatile Chinese imports at the time of tariff implementation, suggesting importers adapted to changing import costs by bulk buying goods before the cut-off date. The year 2018 is further complicated by the fact the trade war did not truly escalate till 7 months into the year. By focusing on the results of 2017 and 2019 we can more accurately compare a pre trade war trade environment to a 2019 environment deep into the trade war.
(20)
war is how the improved bilateral trade position the U.S achieved with its largest trading partner China, has not translated into the U.S. global trade position. With total U.S. imports increasing from a pre trade war $2,340 Billion in 2017 to $2,498 Billion in 2019, a substantial increase of 6.8%. Whilst the U.S. trade deficit of which half is often attributed to China (see Figure 3) has worsened, from $794 Billion in 2017 to $853 Billion in 2019, a worsening of 7.5%. This drastically improved trade balance between the U.S. and its largest trading partner China, coinciding with an overall deterioration of the total U.S trade balance delivers the foundation to our research question.
The occurrence of a rapidly improving trade balance with China coinciding with an overall deteriorating trade deficit is an interesting development, especially when China accounts for more than 40% of the deficit. Although this does not necessarily suggest it cannot be explained by pre-existing trends. Perhaps in the absence of a trade war, the trade balance would have fallen even further into deficit, or the rapid drop in Chinese imports coincides with rapid existing increases from EU, Canada, or Mexico, or it just may be a commodity driven change. To understand what is occurring we first need to understand where the change is occurring. In Figure 6 we observe a rapidly deteriorating trade position with large trading partners in North America and South East Asia, whilst the EU varies country by country.
(21)
Plotting some of the worse affected yearly trade deficits onto Figure 7, we notice the trade balance positions worsen in unison whilst the Chinese deficit improves for the year 2019, with mixed results in 2018.
The worsening of the trade balance for the year 2019 and a lesser extent in 201823 aligns itself with our
theory of trade diversion, whereas goods imported from China in the past are now imported from third-party countries to avoid the tariff. We have selected trade balances and not imports, as changes of imports could simply reflect a stronger trade relationship (an accompanying increase in exports). To prove occurrences of trade diversion we will need to prove the deteriorating trade balance is being disproportionately driven by increases in imported goods of the tariffed type with consideration to the timing and strength of the tariff.
Figure 7. The Deteriorating U.S Trade Balance with Canada, Vietnam, Mexico, and Taiwan
To investigate the occurrence of Trade Diversion we first need to determine the effectiveness of the tariffs on their targeted Chinese products. Figure 8, which displays the (seasonally adjusted) allocation of Chinese imported products to their specified tariff wave, highlights the rapid growth of Chinese imports during the pre trade war period of January 2015 until June 2018. This growth is shared amongst all tariff groups. The introduction of the first Tariff in July 2018, as specified by the 01 in the bottom right corner of the graph, shows a substantial initial drop in imports followed by a sustained suppression in trade for the $32 Billion worth of tariff affected products within the tariff group. What is worth noting is the other tariff groups appear entirely unaffected by the implementation of tariff wave 01, until their designated tariff wave occurs. This pattern of indifference followed by a steep import volume decline at time of tariff implementation,
23 We focus our empirical analysis on 2019 Vs 2017 instead of 2019 Vs 2018, because the year 2018 contains both a 6-month
(22)
followed by sustained suppressed trade volume thereafter is repeated for tariff wave 02 and 15. The notable exception to the pattern is the jointly applied tariff wave 03 & 04 which contradictorily experiences strong growth shortly after the application of the tariff in September 24 ,2018.
The counter intuitive increase of imports can be partly explained by the unpredictability and swiftness of Trump’s tariff announcements. For instance, on July 10, 2018 a draft copy of tariff wave 03 & 04 was
announced. On August 1st, it was suggested he wanted a 25% tariff rather than a 10%, and on September 17,
one week before the September 24th tariff wave implementation, the final revision of the draft was confirmed
with a 25% increase clause designated for the 1st of January 2019. The fast-paced nature and unpredictability
of Trump’s announcements made it hard for importers to appropriately adapt, whilst the relatively low 10% tariff lessened the urgency to do so swiftly, especially with the increase to 25% now looming over wave 03
& 04 for the 1st of January, thus explaining the surge in December 2018 imports.
Figure 8. Chinese Imports to the U.S. divided into their designated tariff wave
Ultimately, the increase to a 25% tariff on wave 03 and 04 was delayed on numerous occasions until
the 10th of May 2019, although by that stage the series of quasi credible announcements leading to the
eventual confirmation on May 5, 2019 led to the dramatic rise and fall in imports, well before the May 10 roll out. The sharp surge in December imports suggest a pre-emptive firm response to the initial January deadline for the 25% increase in tariffs, thus suggesting an attempt by importers to avoid the increased tariff. This revised deadline resulted from promising Chinese/US trade talks during the G-20 meeting in Buenos
(23)
and conflicting press releases, the details of the agreement and the deadline itself remained masked in uncertainty.
Our research question will depend heavily on the unaffected non tariffed $170 Billion black line which under empirical inspection maintained its pre-existing trend line, and therefore prepares a strong argument for candidacy as the control group. An added benefit of the GDD framework is the introduction of pre-treatment tariff observations are included in the control group thus bolstering the control group numbers. It is the aim of our research question to quantify the behaviour displayed in Figure 8, so we can estimate how Chinese imports respond to a rapid succession of exogenous tariff shocks, and then reapply this same methodology to third-party countries to see if these same shocks had led to the increase in imports suggested by figure 7.
3.4. Empirical Results
We use regression analysis to identify trade diversion behaviours based on an increase of imports from third-party countries occurring simultaneously upon treatment with a decrease in imports of the same product from China. Our main findings are presented in Table 4. We find significant evidence of U.S tariffs having a direct negative effect on Chinese imports on all three levels of tariff strength (column 2), with the relative negative effect on imports correlating with the strength of the tariff applied. Using our baseline GDD regression with two-way fixed effect, we find Chinese Imports ended 15.9, 16.5 and 36.4 percentage points lower than that of unaffected products, upon receiving the 10%, 15% and 25% tariff treatment, respectively (column 2) or a
25.9 pp decrease overall (column 1) 24. The resulting decrease in Chinese imports is lower than the general
findings of Meinen, et al. (2019) DD approach, to which they discovered a 30pp decrease amongst all tariffed Chinese imported products. Our work also differs considerably in its ability to separate the resulting direct effect by its associated tariff strength, to discover the 25% tariff in isolation has decreased Chinese imports relative to their unaffected equivalents by a considerable 36.4%. Our model’s ability to extend beyond the
restrictive two-period two-group requirement of the DD25 has enabled us to factor in the layering of
treatments over time, thus creating a more dynamic regression reflective of the escalating trade war. To determine how much of this decline in Chinese imports is attributed to trade destruction, we will need to calculate the trade diversion effect.
24 Percentage change calculated by applying the coefficient x to: (𝑒 − 1) ∗ 100 = Percentage change of outcome. For the 25% tariff this
would be (𝑒 . − 1) ∗ 100 = 36.04%
25 The analysis took advantage of the quick succession of the 2018 July/August/September tariffs to create the one treatment effect required of
(24)
Table 4. Results On – Direct Effects and Trade Diversion Effects
Contrary to Meinen et al(2019) we find significant evidence of trade diversion. Our results show third-party imports of products affected by the tariffs have on average ended 3.05 (column 3) percentage points higher than those of unaffected products, or 2.2, 2.7 and 3.9 percentage points higher upon implementation of the 10%, 15% and 25% tariff treatment, respectively (column 4). Our finding breaks away from the current DD literature on the US/China trade war and instead reflects those of Nicita (2019) whom upon utilising a cross-section regression with a dichotomous variable that indicated the presence of tariffs, found for every dollar lost in Chinese imports from tariff wave 01 and 02, there was a 48c increase in imports elsewhere (trade diversion). The trade diversion result in our regression is the average increase in imports across 34 countries, as a result it is currently not comparable to the findings of Niciti (2019). To change this, we will need to account for the relative size difference in import volume between China and third-party countries, and then adjust our trade diversion effect accordingly.
This can be achieved by first dividing U.S import volumes into origin, and then sub dividing further by tariff strength as done in Table 5. Using the 25% tariff for our example, we will start by using 2017 as
our reference year due to endogeneity issues26 with 2018 and 2019, we can see in 2017 that the $201 Billion
worth of Chinese imports of the 25% tariffed type, paled in comparison to the $668 Billion in equivalent
26 Both 2018 and 2019 have already undergone the direct and indirect effects of the trade war, and as a result the import volumes of those
years will be endogenous to the treatment and therefore bias, we will therefore use the next most recent pre tariff-war year of 2017 as a proxy for the relative size difference.
Direct Effect Diversion Effect
China Third-Party (1) (2) (3) (4) Tariff Dummy 10% -0.173*** 0.022** (0.018) (0.009) 15% -0.180*** 0.027** (0.023) (0.011) 25% -0.447*** 0.038*** (0.020) (0.009) Tariff Dummy
General tariff effect -0.300*** 0.030***
(0.015) (0.007) Constant 13.855*** 13.855*** 12.766*** 12.766*** (0.011) (0.011) (0.006) (0.006) Observations 230,040 230,040 1,548,960 1,548,960 R-squared 0.052 0.057 0.015 0.015 Number of Country_Product 3,834 3,834 25,816 25,816
Product-Country FE Yes Yes Yes Yes
Monthly FE Yes Yes Yes Yes
(25)
products from third-party countries, by a ratio of 3.33. By multiplying this volume ratio of 3.33 by our calculated 25% trade diversion effect of 3.9, we can adjust our trade diversion effect to a more comparable 12.99% (3.33 * 3.9). Next, to obtain the per dollar trade diversion effect, we need to divide the adjusted trade diversion effect by the equivalent direct treatment effect of 36.4% to obtain a proportional trade diversion effect of: 35 cents (12.99/36.4). We therefore find for every dollar lost in Chinese imports to the 25% tariff, there is a 35-cent increase in third-party imports. For the 15% tariff we find a 35-cent increase, and for the 10% tariff a 33-cent increase, with an average 35 cent increase in third-party imports for every dollar lost in Chinese imports across all treatments. The uncovering of an overall trade diversion effect of 35 cents on every dollar lost in Chinese imports is a significant finding and suggests at first glance that trade diversion scales proportionally to the decrease in imports from China, which in turn scales increasingly to the strength of a tariff. Furthermore, by calculating the per dollar trade diversion effect we can now deduce a trade destruction effect of 65 cents of every dollar lost in Chinese import.
Table 5. Import Volume split into Tariff Strength: China vs. Third-Party Countries
Another area of interest within the literature and one we will investigate within our model is the effect the HTS8 product class has on trade diversion and trade destruction effects. For instance, differentiated goods as discussed by Rauch (1996) have greater search costs and frictions between international buyers and sellers, compared to their homogeneous good counterpart which are typically traded on exchanges or have an easily obtainable fair market price. In relation to our research question, we would reason homogenous goods would be greater affected by the direct and indirect effects of tariff as they tend to be more price sensitive commodities that a more easily substituted by US importers Rauch (1996). By creating subsamples
USA Import Volume by Tariff Strength 2015 2016 2017 2018 2019 China No Tariff 151 141 157 163 161 10% - Tariff 147 146 163 183 125 15% - Tariff 94 86 89 94 88 25% - Tariff 183 180 201 220 151 Third-Party Countries No Tariff 173 174 180 193 218 10% - Tariff 371 376 391 424 441 15% - Tariff 189 187 191 200 202 25% - Tariff 640 636 668 729 758 Ratio (Third-party / China)
No Tariff 1.15 1.23 1.14 1.19 1.35 10% - Tariff 2.52 2.57 2.39 2.32 3.52 15% - Tariff 2.02 2.17 2.15 2.12 2.28 25% - Tariff 3.50 3.53 3.33 3.32 5.01
(26)
in Table 6 that utilises the differentiated/non-differentiated product type database27 created by Rauch’s
(1996) we can see product differentiation does indeed play a big role in the direct effect on Chinese imports, with non-differentiated (column 1) product coefficients far exceeding those of differentiated products (column 2) at all tariff strength at the 99% confidence level. Surprisingly, these results do not extend to trade diversion (columns 3 and 4), with neither non-differentiated nor differentiated products displaying a significant change in imports resulting from the implementation of tariffs.
Table 6. Subsample Regression: Differentiated Vs Non-Differentiated Products
A few possible reasons as to why trade diversion retrieved insignificant results for both Non-differentiated and differentiated goods, could stem from the wide array of variation in the effect across geographical regions, furthered by an omittance of products unable to be classified by the Rauch database. To overcome the first issue, we have further regressed the model by region, to see if non-differentiated products can demonstrate any significant role in the trade diversion effect. In table 6’ we see a continuation of insignificant results for NAFTA, The European Union (EU) and even Asia, although when we further restrict the results to regions within Asia more strongly associated to low tech manufacturing and commodities such as Low-Income Asia and South-East Asia, we find a substantial trade diversion effect. With Low-Low-Income Asia demonstrating a strong trade diversion effect at the 25% tariff strength of 17.3% at the 90% significant level,
27 Rauch uses the Standard International Trade Classification Rev 2 database to divide products into three categories; those sold on an
organized exchange, those with a market reference price, and differentiated. With the first two being considered homogenous due to their possession of a reference price.
Direct Effect Trade Diversion
Non-Differentiated Differentiated Non-Differentiated Differentiated
(1) (2) (3) (4) Tariff Dummy 10% -0.355*** -0.095*** -0.037 0.001 (0.057) (0.032) (0.027) (0.017) 15% -0.478*** -0.177*** 0.071 0.026 (0.179) (0.040) (0.048) (0.021) 25% -0.527*** -0.441*** 0.029 0.005 (0.069) (0.044) (0.029) (0.019) Constant 13.348*** 13.537*** 13.028*** 12.292*** (0.032) (0.022) (0.016) (0.012) Observations 32,520 63,900 169,140 381,540 R-squared 0.059 0.065 0.017 0.011 Number of Country_Products 542 1,065 2,819 6,359
Product-Country FE Yes Yes Yes Yes
Monthly FE Yes Yes Yes Yes
(27)
and South-East Asia displaying an even more substantial 30.7% trade diversion effect at the 95% significant level. A further interesting development is the occurrence of the inverse effect for Taiwan, displaying a significant 16% trade diversion effect for differentiated goods at the 95% confidence level for the 25% tariff.
Table 6’. Subsample Regression by Region: Differentiated Vs Non-Differentiated Products
(28)
Another common means of distinguishing product types in the trade literature is their designated use. In Table 7 we create subsamples based on whether the product is classified as a capital good, final good, or intended as an intermediate good. These classifications are derived from the Broad Economic Categories Rev 5 (BEC5), a widely used and persistently updated economic tool provided by the statistics branch of the United Nations (UNSTATS). Table 7 shows particularly strong declines in capital orientated Chinese imports (column 3) relative to intermediate and final goods, with 25% tariffed Chinese imports declining by
a substantial 46%28 compared to their unaffected equivalents, this is nearly 10 pp stronger than the average
36.4 pp decline on all 25% tariffed goods. This strong decline in capital goods at the 25% tariff treatment is accompanied by a strong trade diversion effect of 5.1 pp. This strong trade diversion effect extends to final goods at the 25% tariff level, although most prevalently at the 10% tariff level. This abnormal final goods trade diversion effect at the 10% tariff must contain products most heavily hit by the trade diversion effect, a product described as a good that is final and therefore ready to be consumed. Overall, there is significant evidence that final goods are the most affected by tariffs in both direct effects and trade diversion.
Table 7. Subsample Regression: – Final Vs Intermediate Vs Capital Goods
Aside from product types there is evidence trade diversion effects vary across geographical regions. By applying our baseline regression onto a series of subsamples defined by regions, we find trade diversion does indeed vary greatly. South America for instance has no statistically significant coefficients whilst High 28 Percentage change calculated by applying the coefficient x to: (𝑒 − 1) ∗ 100 = Percentage change of outcome. For the 25% tariff this
would be (𝑒 . − 1) ∗ 100 = 46%
Direct Effect Trade Diversion
Final Intermediate Capital Final Intermediate Capital
(1) (2) (3) (4) (5) (6) Tariff Dummy 10% -0.159*** -0.268*** -0.367*** 0.066*** -0.011 0.032 (0.027) (0.028) (0.063) (0.014) (0.014) (0.021) 15% -0.310*** -0.150*** -0.121* -0.015 0.035* 0.004 (0.031) (0.043) (0.071) (0.019) (0.019) (0.032) 25% -0.545*** -0.487*** -0.610*** 0.047** 0.019 0.050** (0.040) (0.030) (0.053) (0.022) (0.013) (0.021) Constant 14.067*** 13.618*** 14.438*** 12.473*** 12.751*** 13.219*** (0.018) (0.015) (0.025) (0.012) (0.007) (0.014) Observations 85,080 123,000 34,380 471,060 890,640 279,960 R-squared 0.083 0.060 0.076 0.014 0.017 0.018 Number of Country_Products 1,418 2,050 573 7,851 14,844 4,666
Product-Country FE Yes Yes Yes Yes Yes Yes
Monthly FE Yes Yes Yes Yes Yes Yes
(29)
Income (HI) Asian nations Japan, Taiwan, South Korea and Singapore, as defined by having a per capita income in excess of $25,000 USD per capita, have experienced increased imports of 5.3%, 6.7% and 5.8% percentage points higher than those of the unaffected products, upon implementation of the 10%, 15% and 25% tariff treatment on Chinese products, respectively (column 7). Taiwan showed the strongest gains from trade diversion with a 5.4, 9.4 and 17.7 pp increase at the 10%, 15% and 25% tariff treatment, respectively. With hindsight, the result of Taiwan being the greatest beneficiary to the trade war does make sense, as their cross-strait economic co-dependence has only intensified since the early 1990s when Taiwanese entrepreneurs invested immense sums in mainland Chinese manufacturing (Chiang, 2013). Whether it be trade diversion resulting from the two countries sharing similar economic composition, or an instance of trade deflection as firms take advantage of the cultural and economic similarities to reroute trade as demonstrated by Liu et al (2018), is unknown and outside the scope of this paper. Asia in general appears to have benefited greatly from the trade war as expected by Liu et al. (2018) due to the regional proximity to China, with Low Income (LI) Asia (column 6) being the subgroup containing the biggest increase. An interesting development is the large declines across LI and SE Asia for products associated to the 15% tariff group. This is an interesting development as the 15% tariff has been in effect for only four months before our dataset ends in December 2019. Overall, we see region as a large factor in the occurrence of trade diversion, with the strongest trade diversion effects occurring in the LI Asian subgroup and Taiwan.
Table 8. Subsample Regression: Trade Diversion Effects of Different Regions
NAFTA EU S.America Asia SE Asia LI Asia HI Asia Taiwan
(1) (2) (3) (4) (5) (6) (7) (8) Tariff Dummy 10% 0.012 0.016 -0.072 0.059*** 0.081*** 0.085*** 0.052*** 0.053* (0.019) (0.012) (0.046) (0.014) (0.025) (0.020) (0.018) (0.032) 15% 0.016 0.051*** 0.084* -0.002 -0.085*** -0.068*** 0.065*** 0.090** (0.024) (0.015) (0.051) (0.018) (0.030) (0.025) (0.023) (0.038) 25% 0.041** 0.040*** -0.008 0.049*** 0.071** 0.076*** 0.056*** 0.163*** (0.019) (0.012) (0.044) (0.015) (0.028) (0.022) (0.018) (0.032) Constant 13.429*** 12.426*** 12.552*** 12.902*** 12.924*** 12.806*** 12.990*** 12.765*** (0.011) (0.009) (0.037) (0.009) (0.018) (0.014) (0.012) (0.020) Observations 271,920 610,980 40,320 510,540 162,660 244,080 266,460 80,280 R-squared 0.013 0.019 0.021 0.019 0.022 0.027 0.016 0.023 Number of Country_Product 4,532 10,183 672 8,509 2,711 4,068 4,441 1,338
Product-Country FE Yes Yes Yes Yes Yes Yes Yes Yes
Monthly FE Yes Yes Yes Yes Yes Yes Yes Yes
Trade Diversion by Region
