A partnership with academia

Building knowledge for trade and development

Vi Digital Library - Text Preview

Misreported Trade

Working paper by Mohammad Farhad, Michael Jetter, Abu Siddique, Andrew Williams, 2018

Download original document (English)

This paper introduces a methodology to measure misreported trade in a consistent way across countries and over time. Our methodology does not require any assumptions about which countries may be more or less likely to misreport – rather, all indices are derived endogenously with available trade data. We derive seven specific indices related to overall misreporting, as well as over- and under-reporting of exports and imports. Applying this method to existing bilateral trade data on the HS 4-digit level from 1996-2015, we present several rankings and describe a few prominent cases, such as China. Overall, our indices can explain intuitive developments well and should help researchers to study countries’ trade misreporting in a global dimension that is comparable across countries and over time. We conclude the paper with an application, focusing on the role of tariff and VAT rates as predictors of import under-reporting. As predicted by economic theory, case studies, and economic intuition, we find positive correlations for both tariff and VAT rates with import under-reporting. These results are robust to the inclusion of potentially confounding factors, as well as country- and time-fixed effects.


7150
2018


July 2018


Misreported Trade
Mohammad Farhad, Michael Jetter, Abu Siddique, Andrew Williams





Impressum: 
 
CESifo Working Papers 
ISSN 2364‐1428 (electronic version) 
Publisher and distributor: Munich Society for the Promotion of Economic Research ‐ CESifo 
GmbH 
The international platform of Ludwigs‐Maximilians University’s Center for Economic Studies 
and the ifo Institute 
Poschingerstr. 5, 81679 Munich, Germany 
Telephone +49 (0)89 2180‐2740, Telefax +49 (0)89 2180‐17845, email office@cesifo.de 
Editors: Clemens Fuest, Oliver Falck, Jasmin Gröschl 
www.cesifo‐group.org/wp 
  
An electronic version of the paper may be downloaded  
∙ from the SSRN website:           www.SSRN.com 
∙ from the RePEc website:          www.RePEc.org 
∙ from the CESifo website:         www.CESifo‐group.org/wp 
 
 
 
 









CESifo Working Paper No. 7150
Category 8: Trade Policy






Misreported Trade



Abstract

This paper introduces a methodology to measure misreported trade in a consistent way across
countries and over time. Our methodology does not require any assumptions about which
countries may be more or less likely to misreport – rather, all indices are derived endogenously
with available trade data. We derive seven specific indices related to overall misreporting, as
well as over- and under-reporting of exports and imports. Applying this method to existing
bilateral trade data on the HS 4-digit level from 1996-2015, we present several rankings and
describe a few prominent cases, such as China. Overall, our indices can explain intuitive
developments well and should help researchers to study countries’ trade misreporting in a global
dimension that is comparable across countries and over time. We conclude the paper with an
application, focusing on the role of tariff and VAT rates as predictors of import under-reporting.
As predicted by economic theory, case studies, and economic intuition, we find positive
correlations for both tariff and VAT rates with import under-reporting. These results are robust
to the inclusion of potentially confounding factors, as well as country- and time-fixed effects.


JEL-Codes: F130, F140, H260.


Keywords: international trade, trade misreporting, tariffs rates, VAT rates.




Mohammad Farhad*
University of Western Australia


35 Stirling Highway
Australia – Crawley 6009, WA


mohammad.farhad@uwa.edu.au


Michael Jetter
University of Western Australia


35 Stirling Highway
Australia – Crawley 6009, WA


mjetter7@gmail.com


Abu Siddique
University of Western Australia


35 Stirling Highway
Australia – Crawley 6009, WA


abu.siddique@uwa.edu.au


Andrew Williams
University of Western Australia


35 Stirling Highway
Australia – Crawley 6009, WA
andrew.williams@uwa.edu.au




*corresponding author



July 5, 2018




1 Introduction


In 1996, the US recorded a $39.5 billion trade deficit with China (Feenstra et al., 1999). However,


China reported that value to be $10.5 billion. These official trade figures, reported by the


world’s two largest economies, differ by $29 billion – a number equivalent to the collective GDP


of Uruguay and Zimbabwe at that time. But which number is correct or, more realistically, to


what degree are both incorrect? The literature has produced evidence suggesting (i) an under-


reporting of Chinese exports to avoid the value-added tax (VAT), as well as (ii) tariff evasion at


the US border through under-reporting of imports (for example, see Ferrantino et al., 2012). In


fact, if the latter were true, this $29 billion gap may be even higher. This simple and prominent


US-China example illustrates that discrepancies in reported trade statistics are not explainable


by the development status of reporting countries alone. For example, similar gaps in reported


trade numbers have been identified between Canada and the US, two of the richest OECD


countries (Feenstra et al., 1999). Thus, it is not sufficient to simply assume the US numbers to


be correct and the Chinese numbers to be inaccurate.


But why would such discrepancies in reported trade data matter? In reality, fabricated


trade statistics can put policymakers in difficult situations, since trade data play a central role


in macroeconomic policymaking, as well as in trade and foreign policy considerations. Exam-


ples include public policies related to protectionist tariff measures, trade negotiations, capital


controls, or export support programs.1 Trade data might also substantially influence coun-


tries’ internal democratic decision making processes. For instance, the magnitude of the US


trade deficit with China played a substantial role in the 2016 presidential elections (Schneider-


Petsinger, 2017). Similarly, trade relationships with China played a crucial role in the UK voters’


decision in the Brexit referendum (Colantone and Stanig, 2018). Perhaps most importantly from


a fiscal perspective, misreporting trade data can directly decrease public resources, for example


1For example, Feenstra et al. (1999) describe how bilateral trade deficit acts as one of the principle drivers in the
US trade disputes with East Asia; UNCTAD (2016) finds the extensive use of export under-reporting as a main tool
of capital flight from four resource-rich developing countries (Côte d’Ivoire, Nigeria, South Africa, and Zambia).
Kar and Spanjers (2015) claim there was around $1 trillion in illicit capital outflows from emerging countries in
2013, and over 83 percent of that number are suggested to be transported through trade misinvoicing. Finally,
Jara and Escaith (2012) gives a detailed account of how important international trade statistics are for national and
international economic policy making.


1




via lost revenue from tariff evasion or the misuse of export support programs. Further, any


evidence-based policy making or empirical analysis using misreported trade data might indi-


cate misleading outcomes of targeted policy interventions.2 Similarly, measuring international


trade costs or the costs of trade (for example, the costs of cheap Chinese imports on employment)


might be erroneously estimated if trade data are systematically misreported.3


Overall, we can summarize this discussion with three key points: (i) trade data are im-


portant for policymaking, (ii) misreporting trade data exists and is unlikely exclusive of rich


countries, and (as a consequence of the previous point) (iii) it is insufficient to use one country’s


data as the automatic benchmark for correct reporting of any bilateral trade estimate. To date,


several studies exist that analyze and quantify underlying incentives for misreporting. For ex-


ample, Fisman and Wei (2004), Javorcik and Narciso (2008), Mishra et al. (2008), and Ferrantino


et al. (2012) estimate the impact of tariffs on under-reporting imports; Ferrantino et al. (2012)


investigate under-reporting of exports to avoid tax payments. However, these studies usually


have to rely on the assumption that countries commonly labeled as developed report their bilat-


eral trade data correctly, whereas developing countries do not. In addition, the vast majority of


the associated studies focus on individual country pairs or a small group of selected trading


partners to investigate trade misreporting, whereas a misreporting index that is comparable


across countries and over time has remained elusive.


In the following pages, we aim to provide just that: To objectively derive a trade misre-


porting index that is (i) non-discriminatory (i.e., without an a priori definition of one country’s


reports as more credible than another’s), (ii) scale-independent (i.e., independent of country,


economy, and population size), and (iii) comparable across countries and over years. We want


to briefly sketch our methodology that constitutes the main contribution of this paper. First, we


identify a country’s numerical reporting distance of each reported trade flow to its respective


2Egger and Larch (2012) find that disregarding tariff evasion suggests unrealistically higher welfare effects of a
full liberalization of import tariffs.


3For example, The World Bank and the United Nations Economic and Social Commission for Asia and the Pa-
cific (UNESCAP) jointly publish a global data set of bilateral trade costs (available at https://data.worldbank.org/
data-catalog/trade-costs-dataset). To measure bilateral trade costs, they simply ignore the issue of misreport-
ing, which might significantly alter the estimated trade costs. Autor et al. (2013) find that a cheap Chinese import
surge resulted in higher unemployment and low wage rates in the US manufacturing sectors, which is estimated
using the UN Comtrade database that disregards any possibility of misreporting.


2




counterpart’s reported value. Second, we aggregate that country’s reporting discrepancies with


(i) all trading partners (ii) for all goods and services (iii) in a given year to derive a country-


and year-specific weighting factor. Intuitively, that weighting factor proxies “how much we can


believe that country’s trade numbers in that year, according to all their trade partners’ reports”.


Third, these weighting factors allow us to calculate a weighted trade value for each individual


trade entry. Thus, the resulting estimates of each individual bilateral trade flow are solely de-


termined by available data and remain free from any a priori assumptions about who may or


may not be reporting accurately. Fourth and final, we put the estimated trade flows in relation


to the actual trade flows to derive a general trade misreporting index ranging from zero to one.4


We then repeat these steps to derive six specific over- and under-reporting indices for exports


and imports – each of which is designed to analyze particular types of misreporting.


Applying this methodology, we then access bilateral trade data for 160 World Trade Orga-


nization (WTO) member countries from 1996-2015, incorporating over 58 million pairs of trade


observations at the HS 4-digit level. For 2015, we find Togo to be the largest overall trade misre-


porting country, followed by Antigua and Barbuda, Panama, and Afghanistan, whereas Canada


emerges as the least misreporting country. In general, high-income OECD countries misreport


the least, whereas low-income countries misreport relatively more over the entire sample pe-


riod. However, and perhaps somewhat surprisingly, high-income non-OECD countries are the


second highest export-misreporting country group, including Kuwait, Saudi Arabia, and the


United Arab Emirates. These nations rely heavily on exporting oil and other natural resources,


which could explain their large degree of misreporting as a tool of illicit cross-border capi-


tal movement. These findings are also commensurate with the regional average, placing the


Middle East and North Africa as the top export misreporting region. Finally, North America


remains the least misreporting region, both in terms of exports and imports, while Sub-Saharan


Africa emerges as the largest import misreporting region.


As one particular case study of our indices, we then turn to the example of China, conclud-


ing the country’s average export misreporting index to be around 40 percent higher than that


4Specifically, we employ a variation of a Contest Success Function (CSF, e.g., see Buchanan et al., 1980) to measure
Index = estimatedestimated+actual , where 0 ≤ Index ≤ 1.


3




of OECD countries. However, that is not consistent across all types of misreporting: China’s


average import misreporting is comparable to the OECD average throughout the 1996-2015 pe-


riod. Further, China’s imports are dominated by over-reporting, while our results suggest that


exports are largely under-reported. Interestingly, however, these trends are reversed in recent


years. Chinese overall trade misreporting started to decline significantly right before 2001 – the


year when the country joined the WTO. Quite possibly, this could reflect the transparency and


gradual liberalization requirements the country had to comply with for its accession into the


multilateral trading system. In addition, our indices suggest that possible illicit capital flight


outflows through import over-reporting and export under-reporting declined over the years,


corresponding to China’s gradual relaxation of its capital outflow control regimes. However,


possible inflows of ‘hot money’ through export over-reporting may have increased during this


period.


To conclude the paper, we provide one empirical application of one of our derived indices


to provide an example of a practical application. Specifically, we further explore import under-


reporting – the type of trade misreporting that has received the most attention in the literature


to date. Intuitively, as indicated by various country-specific studies, importers may intentionally


under-report to evade tariffs (e.g., see Fisman and Wei, 2004, Mishra et al., 2008, Ferrantino et al.,


2012). Indeed, we find evidence consistent with this hypothesis as applied tariff rates remain


a positive and statistically powerful predictor of our import under-reporting index throughout


a series of regressions, using panel data for our sample countries from 1996-2015. This result


prevails even when we control for country- and year-fixed effects, in addition to potentially


interfering variables such as trade openness, democracy, or corruption levels. Finally, we also


find value-added tax (VAT) rates to be positively associated with import under-reporting. In


addition to the intrinsic implications of these results, we hope this application provides an


example for the usefulness of our indices in analyzing a range of research questions related to


misreported trade data in a panel dimension across many countries and years.


Overall, we aim to contribute to the research community in two ways. First, to the best


of our knowledge, we present the first method to measure country- and time-specific misre-


4




porting of trade data which is free from a priori ad-hoc assumptions about who does and does


not report correctly. In practice, this method could be applied to any level of disaggregated


trade data. Second, we provide a ready-to-use set of trade misreporting indices, which are


comparable across countries and over time. Specifically, we derive seven distinct indices that


explore (i) overall trade misreporting, (ii) export misreporting, (iii) import misreporting, (iv)


export over-reporting, (v) export under-reporting, (vi) import over-reporting, and (vii) import


under-reporting. Depending on the research questions, we hope that these indices can help


us to better understand both the determinants and the consequences of various types of trade


misreporting on a global level.


The paper proceeds with a short background discussion of existing types of trade misreport-


ing measurements. Section 3 introduces our theoretical framework, whereas Section 4 takes the


developed indices to the data and presents initial findings, including a case study on China.


Section 5 presents one empirical application of one of our indices. Finally, Section 6 offers


concluding remarks.


2 Background


In theory, international mirror trade data should be comparable, since each transaction is re-


ported twice by the trading partners to the corresponding public authorities of their coun-


tries. However, similar to other publicly recorded economic activities where deviations from


actual figures generate rents, discrepancies in reported trade data have become a historical phe-


nomenon, and their existence widely recognized in the economics literature.5 These discrep-


ancies in reported bilateral trade statistics, which Ferrantino et al. (2012) describe as “endemic


globally”, continue to stifle economic research and policymaking. Exporting and importing


parties may have several incentives for misreporting trade data. For example, tariffs or other


protectionist trade policies can encourage importers to under-report; capital controls may lead


5For example, 19th century Italian economist Galileo Ferraris (1885) measured the movement of gold from France
to Great Britain from 1876-1880 and 1881-1884, finding that only a varying part of the total exports and imports of
any country was recorded in the official published statistics. Morgenstern et al. (1963) and Bhagwati (1964) provide
an early account of trade misreporting.


5




to misreporting in order to channel capital into or out of the country; export support programs


might inspire exporters to inflate export earnings.6


While these motivations of misreporting trade are much better understood, measurement


methods used to assess misreporting have received relatively little attention. The few existing


studies concerned with measuring discrepancies in trade data can broadly be divided in two


groups. Early works simply measure differences of reported mirror trade flows by bilateral


trading partners as misreporting (for example, see Morgenstern et al., 1963, Bhagwati, 1964,


Sheikh, 1974, among others), while Fisman and Wei (2004) and studies thereafter focus on the


difference in logarithms of bilateral mirror trade flows (also see Javorcik and Narciso, 2008, 2017,


Mishra et al., 2008, and Fisman and Wei, 2009). Initially, they calculate reporting discrepancies


as gap value = log(export value) − log(import value). However, because of its logarithmic
definition, this specification ignores transactions where one partner recorded some trade but


the corresponding partner recorded nothing. To take into consideration these extreme cases of


so-called “complete smuggling”, Mishra et al. (2008) and Fisman and Wei (2009) use a second


measure, where the reporting gap is measured as evasion = log(1+ imports)− log(1+ exports).
These methods merely capture the trade reporting gap. This gap can be attributed to misre-


porting by a specific country only when one assumes that the partner country’s reported trade


data is correctly recorded. For example, Javorcik and Narciso (2008) consider Germany’s re-


ported trade data as accurate when exploring the misreporting of its ten Eastern European


trading partners. Similarly, estimating import under-reporting by India, Mishra et al. (2008)


regard the trade data reported by its top 40 trading partners as correct. Following a similar


assumption, Ferrantino et al. (2012) analyze US imports from China and explore the possibility


of exports being under-reported at the Chinese border, while considering the US data as accu-


rate. In turn, Ferrantino et al. (2012) propose the possibility of import under-reporting by the


US only when the Chinese data are assumed fixed. In sum, all these studies have to make an


ad-hoc assumption that one side of each trade relationship is correctly reported, whereas the


other is not.


6See Bhagwati (1964, 1967, 1981) for details on different types of trade misreporting, their underlying motivations
and economic implications, as well as possible ways of faking trade invoices in practice.


6




Perhaps as a consequence of this lack of a comparable and consistent trade misreporting


index, the literature usually focuses on one trading partner (e.g., Fisman and Wei, 2004, and


Ferrantino et al., 2012) or the few major trade partners of one country (e.g., see Mishra et al.,


2008, or Javorcik and Narciso, 2008). A few studies consider a select group of countries, such as


Javorcik and Narciso (2017) who analyze bilateral exports from Germany, the US, and France,


as well as imports by 15 countries that joined the WTO between 1996 and 2008.7 Moreover, the


prevailing literature rarely attempts to capture the extent of all four types of trade misreporting


by a particular reporting country. For example, Fisman and Wei (2004), Javorcik and Narciso


(2008); Mishra et al. (2008); Ferrantino et al. (2012), and Javorcik and Narciso (2017) try to


capture and explain import under-reporting; Ferrantino et al. (2012) also explore export under-


reporting (also see Arslan and van Wijnbergen, 1993). As one of the few exceptions, Buehn and


Eichler (2011) aim to capture all types of trade misreporting, but employ aggregate trade data


between the US and 86 countries. Again, Buehn and Eichler (2011) start from the premise that


one country (in this case the US) reports trade data accurately.


Overall, we lack a consistent empirical method that is comparable across countries and over


time to estimate trade misreporting without making ad-hoc a priori assumptions about who


does and does not report correctly.


3 Theoretical Framework


The dual nature of reported trade data provides us with a straightforward way to identify


the existence of misreporting. Nevertheless, assigning any discrepancies to one of the trading


partners is challenging since differences may be induced by either or both parties involved.


As an example, consider the export of coffee (HS 4-digit code 0901) from Brazil to Tunisia.


Let us assume that, in a given year, Brazil reports exporting $100,000 worth of coffee to Tunisia;


however, Tunisia reports only $60,000 worth of coffee imports from Brazil. Who is misreporting?


We will use this example throughout this section to illustrate the derivation of our index. To


7Kellenberg and Levinson (2016) make an attempt to examine misreporting using a larger panel including trade
data between 126 countries over 11 years. However, they use aggregate trade data which may not be able to capture
the extent of trade misreporting correctly – an aspect we consider in our data section.


7




keep it simple, we assume that both values are in so-called free-on-board (FOB) values.8 To


facilitate readability, we omit time subscripts t throughout this section as all calculations are of


a static nature, i.e., take place in the same year.


3.1 Step 1: Deriving Weighted Trade Values


Our first step to derive a comparable index of trade misreporting consists in identifying the


degree to which a given country misreports its exports and imports in a given year. Then, we


use these numbers to calculate the weighted value for each bilateral trade transaction. Thus,


we begin by considering the ‘reporting distance’ of all bilateral trade relationships reported by


a country and all of its trading partners.


3.1.1 Export Weighting Factors


Beginning with exports, consider the top panel of Table 1, displaying the hypothetical relation-


ships between exporting Brazil and importing Tunisia. We can observe three types of trade


links: exports that are reported by Brazil but unreported (as imports) by Tunisia; exports re-


ported by both countries, indicated by the shaded grey areas; and imports reported by Tunisia


that are not reported as exports from Brazil. We can then extend this picture to all countries


that Brazil is linked to in terms of exports. To keep things simple in this example, Table 1


assumes Brazil’s exports are linked to no more than three countries overall in a given year:


Tunisia, Bangladesh, and Australia. (Note that this includes countries that report having im-


ported something from Brazil but Brazil does not record any of those exports.)


Our first step consists in using the absolute reporting distance of Brazil’s reported export


values with the respective importer-reported import values. We consider the unreported trade


values as zero trade where one party reports non-zero trade, whereas the corresponding partner


reports nothing. In the example of Table 1, Brazil’s reported exports total $1,060,000, whereas


its partners report importing a total of $1,180,000 from Brazil in aggregate. However, the total


8Following the IMTS (2010) recommendation, countries use the FOB valuation for exports (at the border of
the exporting county) and the cost, insurance, and freight (CIF-type) valuation for imports (at the border of the
importing country) while reporting their trade values. We will return to this difference in our empirical section.


8




Table 1: Mirror trade flow reported by exporter Brazil (s1) and all destination countries: Tunisia
(d1), Bangladesh (d2), and Australia (d3).


HS-4 code Source Destination Export value Import value Absolute Reporting
($000) ($000) distance ($000)


(s1 d1)





0110 Brazil (s1) Tunisia (d1) 15 15
0806 Brazil (s1) Tunisia (d1) 20 20
0901 Brazil (s1) Tunisia (d1) 100 60 40
4040 Brazil (s1) Tunisia (d1) 40 50 10
5050 Brazil (s1) Tunisia (d1) 50 40 10
6060 Brazil (s1) Tunisia (d1) 25 25
7009 Brazil (s1) Tunisia (d1) 10 10
8080 Brazil (s1) Tunisia (d1) 5 5
8(3) 5(3) 6(3) 225(190) 190(150) 135


(s1 d2)





1010 Brazil (s1) Bangladesh (d2) 30 30
2020 Brazil (s1) Bangladesh (d2) 85 70 15
3030 Brazil (s1) Bangladesh (d2) 60 50 10
4040 Brazil (s1) Bangladesh (d2) 80 100 20
5050 Brazil (s1) Bangladesh (d2) 100 150 50
6060 Brazil (s1) Bangladesh (d2) 80 80
7009 Brazil (s1) Bangladesh (d2) 40 40
8080 Brazil (s1) Bangladesh (d2) 20 20
8(4) 5(4) 7(4) 355(325) 510(370) 265


(s1 d3)





1010 Brazil (s1) Australia (d3) 20 20
2020 Brazil (s1) Australia (d3) 100 125 25
3030 Brazil (s1) Australia (d3) 120 140 20
4040 Brazil (s1) Australia (d3) 240 200 40
5050 Brazil (s1) Australia (d3) 15 125
5(2) 4(2) 3(2) 490(460) 480(465) 120


(s1 dn, n = 3) 21(10) 14(10) 16(10) 1,060(975) 1,180(985) 520


Notes: The above table shows hypothetical trade reportings between exporting Brazil and importing Tunisia in a
given year. Both exports and imports are considered here in comparable FOB values to eliminate discrepancies
resulted from FOB and CIF price reportings by the exporter and the importers, respectively.


9




absolute reporting distance of Brazil reports exports from its counterparts’ reported imports


is $520,000. Thus, we derive the export weighting factor (EWF) for Brazil in this example as


one minus the ratio of the total absolute reporting distance divided by the sum of Brazil’s


reported exports and its partners’ reported imports. In this case, we derive a value of 1 −
520,000


1,060,000+1,180,000 = 0.755. Intuitively, the closer the EWF comes to zero, the less misreporting we


detect; as the EWF approaches one, more and more misreporting is detected.


From this example, we can now formalize the derivation of the EWF. Considering total


reported exports x of all products K (with k ∈ [1, ..., K]) from all source countries S (with
s ∈ [1, ..., S]) to all destination countries D (with d ∈ [1, ..., D]), we can write


XKsD =
K



k=1


D



d=1


xksd. (1)


In our simple example from Table 1, this corresponds to the reported exports of $1,060,000.


Further, the total reported imports (m) of all K products by all importing (destination) countries


D from each source country S are calculated as


MKDs =
K



k=1


D



d=1


mkds, (2)


which corresponds to the reported imports of $1,180,000 in Table 1. From here, we calculate


the reporting distance (δksd) of each product as the difference of each reported export value (x
k
sd)


from its mirror import value (mkds) reported by the corresponding import partner as


δksd = m
k
ds − xksd. (3)


Now we calculate the total absolute reporting distance (δKsD) of all Brazil’s reported export values


from its counterparts reported import values as


δKsD =
K



k=1


D



d=1
|δksd|. (4)


In Table 1, this corresponds to $520,000. Finally, the EWF (wxs ) for Brazil is then derived as one


10




minus the ratio between the total absolute reporting distance and the sum of Brazil’s reported


total exports and all importing countries’ reported total imports from Brazil as:


wxs = 1−
δKsD


XKsD + M
K
Ds


. (5)


Intuitively, if a country reports export values that are close to the reported import values by the


respective importer, the country will score a high wxs ; on the other hand, countries having higher


discrepancies with their counterparts’ reported imports will score a lower value. Naturally, the


wxs ranges between zero and one.


3.1.2 Import Weighting Factors


If we consider trade from the importing country’s perspective, we can derive an analogous


weighting factor for imports. An example is provided in Table A1 in the appendix, where we


consider Tunisia’s imports, assuming three respective source countries: Brazil, Bangladesh, and


Australia. We now consider the total value of Tunisia’s reported imports from all its import


partners and all its import sources’ reported export values to Tunisia. We then calculate the


total absolute reporting distance of Tunisia’s reported imports from its counterparts’ reported


exports. Finally, we derive the import weighting factor (IWF, wmd ). Formally, the IWF is derived


analogously to equation 5 with


wmd = 1−
δKdS


MKdS + X
K
Sd


, (6)


where δKdS = ∑
K
k=1∑


S
s=1 |δkds|, MKdS = ∑Kk=1∑Ss=1 mkds, and XKSd = ∑Kk=1∑Ss=1 xksd. These three terms


constitute the counterparts of equations 4, 1, and 2 from the export perspective.


3.1.3 Calculating Weighted Trade Values


The EWF and IWF values provide proxies for the reliability levels with which each country


reports its exports and imports, based entirely on reported data, as opposed to ad-hoc assump-


tions about the reliability of one country’s data over another. With this information, we can


now revisit each trade entry – for instance, our example coffee exports from Brazil to Tunisia. If


11




Brazil reports an exported value of $100,000, but Tunisia reports importing $60,000, then which


entry is more reliable and by how much? We can now use the EWF and the IWF values to


weigh these values according to how reliable the respective country’s reporting is. Formally,


we can calculate the weighted export value of product k (e.g., coffee) from source country s


(e.g., Brazil) to destination country d (e.g., Tunisia) as


x̂ksd =
wxs


wxs + wmd
× xksd +


wmd
wxs + wmd


×mkds. (7)


Intuitively, if Brazil had a strong EWF and Tunisia had a weak IWF, then the first fraction


( w
x
s


wxs+wmd
) would be closer to one. Consequently, the value reported by Brazil would carry more


weight, i.e., the predicted actual export value (x̂ksd) would be closer to $100,000. Alternatively,


if Tunisia’s IWF was more credible, x̂ksd would converge closer to $60,000. In our example, the


predicted export of coffee from Brazil to Tunisia is ( 0.7550.755+0.665 × 100, 000+ 0.6650.755+0.665 × 60, 000) =
$81, 268 (see Table A1 for Tunisia’s IWF in this example). This method allows us to derive a


weighted value for every reported trade entry, including situations where one country reports no


export but its corresponding import partner does report a non-zero value.9


Likewise, we can derive predicted import values (m̂kds) for each reported import product,


using the importing country’s IWF (wmd ) and the corresponding export country’s EWF (w
x
s ).


Formally, this translates to


m̂kds =
wmd


wmd + w
x
s
×mkds +


wxs
wmd + w


x
s
× xksd. (8)


In sum, equations 7 and 8 provide us with a weighted value for every import and export


entry in the product-country-year dimension.


9To illustrate this, consider our example from Table 1, where Tunisia reports $10,000 worth of glass mirror
imports (HS 4-digit code 7009) from Brazil. Brazil, on the other hand, reports no export of this item to Tunisia.
Using equation 7, we can estimate a weighted export value of glass mirrors from Brazil to Tunisia, which in this case
is ( 0.7550.755+0.665 × 0+ 0.6650.755+0.665 × 10, 000) = $4, 683.


12




3.2 Step 2: Constructing Trade Misreporting Indices


With these derivations, we are now ready to construct misreporting indices. Specifically, for


every country and year, we can derive (i) an overall misreporting index, (ii) an under-reporting


index, and (iii) an over-reporting index for exports and imports. We begin with considering


exports and then move to imports in Section 3.2.2 before considering overall misreporting in


Section 3.2.3.


3.2.1 Export Misreporting Indices


First, we find the misreported export value (x̃ksd) for each product as the difference between the


reported value (xksd) and the weighted value (x̂
k
sd):


x̃ksd = x
k
sd − x̂ksd. (9)


To gain an overall picture of a country’s export misreporting, we need to sum up their product-


wise misreported export values. However, exporters of a country might under-report some


values but over-report others, based on the individual incentives. Therefore, a simple summa-


tion of product-wise misreported values would cancel out some of the negative and positive


misreported values and, hence, we would fail to capture the actual magnitude of trade misre-


porting in that country and year.


To circumvent this issue, we sum the absolute values. Formally, we calculate the total


absolute misreported export value X̃Ks for each source country s and all its export products k to


all its export destinations d as


X̃Ks =
K



k=1


D



d=1
|x̃ksd|. (10)


X̃Ks gives us a dollar estimate of the total absolute export misreporting of any given country


in any given year. However, this would still make a comparison across countries and time


difficult, since clearly countries that trade more and in larger volumes would report higher


values of X̃Ks . To derive a comparable index that is naturally bounded between zero and one,


our final step consists in putting X̃Ks in perspective to the sum of the country’s total reported


13




export values (XKs ) and the total absolute export misreporting value (X̃Ks ). This step is perhaps


best comparable to a so-called Contest Success Function (CSF, e.g., see Buchanan et al., 1980).


Formally, we label the overall export misreporting index for source country s as MRIxs with


MRIxs =
X̃Ks


X̃Ks + XKs
. (11)


Further, if we are specifically interested in export under-reporting, we can sum up the under-


reported export values only. Thus, we consider only those x̃ksd values from equation 9 that are


negative. Denoting these with xksd, we arrive at the total under-reported export value of


XKs =
K



k=1


D



d=1
|xksd| (12)


and the export under-reporting index becomes


URIxs =
XKs


XKs + XKs
. (13)


Similarly, assume we are interested in over-reported exports only, labeling these xksd. In this case,


we only consider those values from equation 9 that return positive values, i.e., the reported


export value is higher than the weighted value. Consequently, we derive the total over-reported


export value via


XKs =
K



k=1


D



d=1
|xksd| (14)


and the export over-reporting index becomes


ORIxs =
XKs


XKs + XKs
. (15)


In sum, we can derive three distinct export misreporting indices: (i) the overall export


misreporting index (MRIxs ), (ii) the export under-reporting index (URIxs ), and (iii) the export


over-reporting index (ORIxs ).


14




3.2.2 Import Misreporting Indices


The corresponding indices for import misreporting follow analogously and we only sketch


them briefly here. Specifically, if we are interested in the overall degree of import misreporting,


we first calculate misreported import values (m̃kds) for each product as the difference between


the reported value (mkds) and the weighted value (m̂
k
ds) as


m̃kds = m
k
ds − m̂kds. (16)


From here, we get the total overall misreported import value for each importer by taking abso-


lute values of equation 16, leading to


M̃Kd =
K



k=1


S



s=1
|m̃kds|. (17)


Next, to derive an overall import misreporting index (MRImd ), we calculate


MRImd =
M̃Kd


M̃Kd + M
K
D


. (18)


Finally, we can construct import under- and import over-reporting indices via


URImd =
MKd


MKd + M
K
d


(19)


and


ORImd =
MKd


MKd + MKd
. (20)


Overall, this gives us three distinct import misreporting indices: (i) the overall import mis-


reporting index (MRImd ), (ii) the import under-reporting index (URI
m
d ), and (iii) the import


over-reporting index (ORImd ).


15




3.2.3 Overall Misreporting Index


Depending on the underlying research question, one may sometimes be more interested in


misreporting exports or imports and over- or under-reporting in either domain. For example, if


one was interested in questions related to tariff evasion, the import under-reporting index may


be of particular interest. In turn, if we were studying the potential abuse of export subsidies,


the export over-reporting index may be most appropriate to consider.


However, in its most general context researchers may be interested in an overall index that


describes the degree of trade misreporting by a country in a given year. Following our method-


ology laid out in the previous pages, we can derive a trade misreporting index of country i in


year t (TMRIit) via


TMRIit =
X̃Ks + M̃Kd


(XKs + MKd ) + (X̃
K
s + M̃Kd )


. (21)


This provides our seventh and final index to measure trade misreporting. With these concepts


in mind, we now turn to the data to illustrate the respective indices, followed by a country case


study and an application of one of the developed indices.


4 The Index in Practice


4.1 Trade Data


We retrieve trade data using the World Trade Solution database (WITS), which is derived from


the United Nations International Trade Statistics Database (UN Comtrade).10 UN Comtrade


contains bilateral import and export statistics on an annual basis from over 200 countries. The


International Monetary Fund (IMF), the World Bank, the Food and Agriculture Organisation


(FAO), and the International Trade Center (ITC) also publish and disseminate trade data on an


10The World Bank, in collaboration with the United Nations Conference on Trade and Development (UNCTAD)
and in consultation with organizations such as the International Trade Center, United Nations Statistical Division
(UNSD), and the World Trade Organization (WTO), developed the World Integrated Trade Solution (WITS). The
database is available under http://wits.worldbank.org/about_wits.html. For more detailed information about
the UN Comtrade data collection, coding, valuation, and processing system, we refer to the United Nations Interna-
tional Trade Statistics Knowledgebase, available under https://unstats.un.org/unsd/tradekb/Knowledgebase/
50075/What-is-UN-Comtrade.


16




annual basis. We use UN Comtrade as our single source of trade data since it is considered


as the most comprehensive and primary source of international trade statistics. We refer to


ChathamHouse (2018) for a detailed discussion about different available sources of merchandise


trade statistics and the comprehensiveness of the UN Comtrade.


The existing literature analyzing trade misreporting uses both aggregated and disaggre-


gated trade data to identify and measure misreporting.11 However, a country could misreport


export and import products, which may cancel out in aggregate. Therefore, aggregated trade


data would not allow us to isolate the actual extent of misreporting, and neither could we


distinguish between export- and import-specific over- and under-reporting. Consequently, we


employ disaggregated trade data and select the HS 4-digit level for our analysis. Although one


could well disaggregate trade down to the 6-digit level, a simple example may illustrate why


the 4-digit level may be most appropriate when exploring misreported trade by minimizing


unintentional misclassifications. To see this, consider our coffee example. The HS 2-digit level


identifies Coffee, Tea, Maté, and Spices; the 4-digit level considers Coffee, whether or not roasted or


decaffeinated; the 6-digit level identifies Coffee, not roasted and not decaffeinated. It is quite conceiv-


able that one party could easily mistake roasted for decaffeinated coffee (or vice versa), whereas


it is more difficult to mistake coffee for tea. Of course, one could easily exploit more (or less)


disaggregated levels of classifications in deriving our indices and we refer to Section A.2 for a


more detailed explanation of why we choose the 4-digit level.


Another challenge in identifying misreporting from bilateral mirror trade flows comes from


separating insurance and freight costs from reported import values.12 In fact, the majority


of the associated literature does not specifically consider this issue (e.g., see Fisman and Wei,


2004, Fisman and Wei, 2009, and Mishra et al., 2008), whereas some studies employ an average


11For example, Kellenberg and Levinson (2016) and Egger and Larch (2012) use aggregated trade data from UN
Comtrade and Buehn and Eichler (2011) use aggregated trade figures from IMF’s Directions of Trade statistics
(DOTS). Ferrantino et al. (2012), Fisman and Wei (2004), and Mishra et al. (2008) use HS-6 digit data from UN
Comtrade, whereas Ferrantino and Wang (2008) use 8-digit trade data for China and Hong Kong from the Customs
General Administration of China and the Census and Statistical Department of Hong Kong, respectively. Further,
Ferrantino and Wang (2008) employ 6-digit data from USITC’s Oracle database to analyze discrepancies in reported
trade data. Javorcik and Narciso (2017) also use HS 6-digit trade data from UN Comtrade.


12Most countries report import data on the cost, insurance, and freight (CIF) basis, while exports are reported
based on a free on board (FOB) value.


17




adjustment factor of 1.1, as suggested by the IMF, 1993 (e.g., see Buehn and Eichler, 2011, and


UNCTAD, 2016). However, the economics and transport literature describes a declining trend


in transport cost over the decades (see Hummels, 2007, and Timmer et al., 2012, among others).


In March 2017, the IMF introduced a new CIF/FOB factor of six percent to convert imports


CIF into exports FOB (and vice versa; Marini et al., 2018; and Miao and Fortanier, 2017). We


readily use this conversion factor in our analysis. In reality, this definition does not produce


substantial changes in our results and we derive virtually identical indices when employing


the traditional conversion factor of 1.1 (see Section A.3). Similarly, the role of entrepôt trade


has been investigated with respect to discrepancies in reported trade data (e.g., see Feenstra


et al., 1999). Nevertheless, our indices only change marginally if we address those issues;


for example, once we consider Hong Kong (the largest entrepôt worldwide) and China as one


trading country, the correlation coefficient with our baseline overall misreporting index becomes


0.99. Thus, although the role of entrepôt trade may affect the ranking of individual countries


in our indices, it does not affect the overall rankings and indices in general. (Nevertheless, it


would of course be straightforward to follow our methodology and adjust accordingly.)


Overall, we incorporate bilateral trade data reported by 160 WTO members at the HS 4-digit


product level from 1996-2015, using the HS1996 version (also known as HS1).13 After excluding


products under Chapter 99 (representing Commodities not specified), this produces 58,515,054


pairs of trade data.


4.2 Most Recent Country Rankings


Following our theoretical framework outlined in Section 3, we derive seven trade misreporting


indices for each reporting country per year for the period of 1996-2015. By construction, all


indices range between zero and one, where values approaching zero represent less misreporting


and higher values indicate more misreporting. As an example, Table 2 lists the top and bottom


ten countries for the total misreporting index (TMRI) among the 127 countries for which data are


13Information on the Harmonized Commodity Description and Coding Systems (HS) can be found at the World
Customs Organization website under http://www.wcoomd.org/en/topics/nomenclature/overview.aspx and the
World Integrated Trade Solution (WITS) website under https://wits.worldbank.org/wits/wits/witshelp/
content/Annex/Annex1.About_WITS_HS_Combined.htm.


18




available in 2015, the most recent year in our database.14 Our index suggests that, in 2015, the


countries that misreport most in their trade statistics are Togo, Antigua and Barbuda, Panama,


Afghanistan, and Malta, whereas Canada, Peru, Chile, Mexico, and the US are the countries


with the lowest misreporting index. (In Table A4, we display correlation coefficients among the


respective trade misreporting indices, whereas Table A5 presents correlations coefficients with


popular country-level variables. Finally, Table A6 reports summary statistics of the indices.)


Table 2: Empirical results for the overall trade misreporting index (TMRI) for top and bottom
ten countries in 2015.


Top 10 Misreporting Country Bottom 10 Misreporting Country


Rank Country Overall Trade Rank Country Overall Trade
misreporting index misreporting index


1 Togo 0.784 118 Brazil 0.154
2 Antigua and Barbuda 0.713 119 Japan 0.148
3 Panama 0.712 120 Germany 0.144
4 Afghanistan 0.636 121 Italy 0.140
5 Malta 0.614 122 Argentina 0.137
6 Benin 0.613 123 United States 0.133
7 Kuwait 0.592 124 Mexico 0.133
8 Sierra Leone 0.561 125 Chile 0.124
9 Solomon Islands 0.494 126 Peru 0.123
10 Niger 0.481 127 Canada 0.098


To provide a quantitative example as to what the index means in practice, consider the case


of Togo. A score of 0.784 in the TMRI indicates that for every US$100 of reported trade, Togo


misreported its trade value by approximately US$363. This follows directly from our index


calculation in equation 11 since for reporting US$100, we get 0.784 = mm+100 , which, after some


simple algebra, produces m = 363. Since the TMRI incorporates all possible types of trade


misreporting, it may be worth to distinguish further between imports and exports, as well


as under- and over-reporting. Tables 3 and 4 provide the respective lists. These distinctions


provide us with more detail about how a particular country received a high or low score on the


overall TMRI. For example, Togo’s misreporting in 2015 is primarily driven by under-reporting


14A full list of all trade misreporting indices for 160 WTO members for the 1996-2015 period can be accessed
under https://farhadm.weebly.com/trade-misreporting-index.html


19




imports, and the country remains absent from all three the top ten lists for export misreporting.


Although Table 3 suggests some notorious misreporters that may have been expected, they


also produce results that are perhaps surprising at first sight. For example, export over-


reporting may be much less of an issue among top offenders than export under-reporting,


as indicated by the top values in either index (0.991 and 0.433; see Panels B and C of Table 3).


Consequently, the ten countries that are suggested to misreport exports most are also those who


under-report exports most. In turn, the top five countries in the export over-reporting category


are African. Further, although five out of the ten countries that are under-reporting exports


the least are within the European Union (EU), no EU country makes that list when it comes to


export over-reporting.


Table 4 turns to our three import misreporting indices. As with exports, the values of the top


ten suggest that under-reporting imports is more of an issue that over-reporting imports. Seven


OECD nations are among the bottom ten when it comes to misreporting imports in general,


whereas the EU nations Croatia, Spain, Denmark, Portugal, the United Kingdom, Italy, and


Romania are suggested to be least prone to over-reporting imports.


It may also make sense to consider the corresponding indices within averages of country


groups over the entire timeframe, as displayed in Table 5. On average, high-income OECD


countries misreport the least, whereas low-income countries misreport their trade values most


during the 1996-2015 period. This may be reflective of a weak state of governance, more restric-


tive policies, and capacity constraints to record and report trade statistics accurately. However,


and perhaps surprisingly, high-income non-OECD countries are the second-highest trade mis-


reporting country group, and the highest export under-reporting country group. Interestingly,


this high-income non-OECD country group includes Kuwait, Saudi Arabia, and the United


Arab Emirates – all of which heavily reliant on exporting oil and other natural resources. Al-


though speculative at this point, this could indicate possible illicit outflows of capital through


export under-reporting. These findings are also commensurate with the regional average, plac-


ing the Middle East and North Africa as the top export misreporting region. North America


20




Table 3: Empirical results for the export misreporting index for top and bottom ten countries
in 2015.


Top 10 Misreporting Country Bottom 10 Misreporting Country


Rank Country Index Rank Country Index


Panel A: Overall Export Misreporting


1 Antigua and Barbuda 0.991 118 El Salvador 0.121
2 Macao 0.983 119 Bolivia 0.121
3 Kuwait 0.889 120 Germany 0.116
4 Sierra Leone 0.884 121 Mexico 0.113
5 Panama 0.841 122 Angola 0.107
6 Yemen 0.823 123 Argentina 0.104
7 Hong Kong 0.806 124 Chile 0.101
8 Saudi Arabia 0.737 125 Peru 0.100
9 Cyprus 0.707 126 Brunei 0.074
10 United Arab Emirates 0.695 127 Canada 0.068


Panel B: Export Under-Reporting


1 Antigua and Barbuda 0.991 118 Mongolia 0.054
2 Macao 0.983 119 Slovak Republic 0.053
3 Kuwait 0.883 120 Poland 0.050
4 Sierra Leone 0.872 121 Czech Republic 0.049
5 Panama 0.836 122 Belgium 0.049
6 Yemen 0.810 123 Bolivia 0.047
7 Hong Kong 0.791 124 Germany 0.046
8 Saudi Arabia 0.725 125 Paraguay 0.036
9 Cyprus 0.693 126 Canada 0.035
10 United Arab Emirates 0.680 127 Brunei 0.028


Panel C: Export Over-Reporting


1 Sierra Leone 0.433 118 New Zealand 0.052
2 Niger 0.420 119 Japan 0.051
3 Central African Republic 0.398 120 Chile 0.050
4 Zimbabwe 0.393 121 Macedonia 0.050
5 Zambia 0.380 122 Brunei 0.048
6 Kuwait 0.303 123 Peru 0.048
7 Afghanistan 0.301 124 Angola 0.048
8 Mozambique 0.292 125 St.Vincent and Grenadines 0.048
9 Yemen 0.288 126 Argentina 0.041
10 Hong Kong 0.276 127 Canada 0.036


21




Table 4: Empirical results for the import misreporting index for top and bottom ten countries
in 2015.


Top 10 Misreporting Country Bottom 10 Misreporting Country


Rank Country Index Rank Country Index


Panel A: Overall Import Misreporting


1 Togo 0.810 118 India 0.150
2 Panama 0.694 119 Chile 0.147
3 Antigua and Barbuda 0.660 120 United Kingdom 0.145
4 Afghanistan 0.644 121 Japan 0.144
5 Malta 0.641 122 Italy 0.143
6 Benin 0.638 123 Peru 0.142
7 Sierra Leone 0.488 124 Romania 0.139
8 Kyrgyz Republic 0.461 125 Canada 0.125
9 Central African Republic 0.454 126 Botswana 0.116
10 Brunei 0.453 127 United States 0.109


Panel B: Import Under-Reporting


1 Togo 0.801 118 India 0.061
2 Panama 0.681 119 Costa Rica 0.058
3 Malta 0.618 120 Peru 0.058
4 Antigua and Barbuda 0.618 121 El Salvador 0.058
5 Benin 0.610 122 Japan 0.054
6 Afghanistan 0.565 123 China 0.053
7 Kyrgyz Republic 0.415 124 United States 0.052
8 Guinea 0.383 125 Botswana 0.050
9 Cambodia 0.374 126 Mexico 0.048
10 Brunei 0.369 127 Canada 0.046


Panel C: Import Over-Reporting


1 Sierra Leone 0.362 118 Croatia 0.084
2 Central African Republic 0.349 119 Spain 0.083
3 Afghanistan 0.336 120 Denmark 0.083
4 Niger 0.318 121 Portugal 0.081
5 Burkina Faso 0.280 122 Hong Kong 0.078
6 Burundi 0.274 123 United Kingdom 0.077
7 Macao 0.265 124 Botswana 0.073
8 Guyana 0.264 125 Italy 0.069
9 Solomon Islands 0.254 126 Romania 0.068
10 Uganda 0.254 127 United States 0.064


22




remains the least misreporting region, both in terms of exports and imports, while Sub-Saharan


Africa is the top import misreporting region. With these descriptive data in mind, we now turn


to exploring misreported trade data for China as a prominent example.


4.3 Trade Misreporting: The Case of China


The case of China has generated particular interest in the trade misreporting literature (for in-


stance, see Feenstra et al., 1999, Fisman and Wei, 2004, and Ferrantino et al., 2012). China’s mas-


sive economic growth over the past three decades has seen the country rise to the world’s largest


merchandise trader in 2015. Figure 1 visualizes China’s TMRI over our sample period from 1996


to 2015. Interestingly, the index starts to decline sharply right before 2000 and through to 2011,


indicating a constant improvement in trade reporting relative to its trading partners. In terms


of magnitude, this sizeable drop from 1998 to 2011 is equivalent to more than two-thirds of a


standard deviation of the TMRI across all countries and years. Interestingly, China formally


joined the WTO in 2001, after the respective negotiations negotiations lasted a couple of years.


In theory, joining the WTO required China to liberalize much of its trading sectors, along with


streamlining its trade reporting system and providing more transparency. Although a range of


motivations and policy responses may influence trade misreporting, China’s accession to the


WTO, that required steep reduction of its tariff and other non-tariff barriers, drastic overhaul-


ing of its state owned enterprises (SOEs) and gradual opening of the financial system (see, for


instance Lu and Yu, 2015, Khandelwal et al., 2013, Bajona and Chu, 2010, Prasad et al., 2005


and He et al., 2014), as well as substantial reduction in trade policy uncertainty with respect to


its trading partners (see Feng et al., 2017 and Brandt et al., 2017), is likely to be reflected in this


declining trend of the TMRI.


Further, Figure 2 illustrates China’s development when it comes to under-reporting and


over-reporting. Intuitively, import-overinvoicing or export under-invoicing can be used to cir-


cumvent outward capital controls and therefore transfer money abroad via official channels


(e.g., see Bhagwati, 1964, 1967). In turn, foreign capital can be channelled into the country


through over-invoicing of exports or under-invoicing of imports. One hypothesis that could


23




Ta
bl


e
5:


A
ve


ra
ge


s
ov


er
ti


m
e


an
d


by
in


co
m


e
an


d
re


gi
on


al
gr


ou
ps


fo
r


di
ff


er
en


t
tr


ad
e


m
is


re
po


rt
in


g
in


di
ce


s,
us


in
g


al
l


da
ta


fr
om


19
96


-2
01


5.


O
ve


ra
ll


tr
ad


e
Im


po
rt


Im
po


rt
Im


po
rt


Ex
po


rt
Ex


po
rt


Ex
po


rt
m


is
re


po
rt


in
g


m
is


re
po


rt
in


g
un


de
r-


re
po


rt
in


g
ov


er
-r


ep
or


ti
ng


m
is


re
po


rt
in


g
un


de
r-


re
po


rt
in


g
ov


er
-r


ep
or


ti
ng


Pa
ne


l
A


:A
ve


ra
ge


s
by


in
co


m
e


gr
ou


ps


H
ig


h-
in


co
m


e,
O


EC
D


0.
18


4
0.


18
8


0.
10


6
0.


10
2


0.
17


7
0.


08
9


0.
10


5


H
ig


h-
in


co
m


e,
no


n-
O


EC
D


0.
36


4
0.


31
8


0.
21


2
0.


16
2


0.
44


4
0.


37
8


0.
15


5


U
pp


er
m


id
dl


e
in


co
m


e
0.


28
6


0.
28


0
0.


16
4


0.
16


2
0.


29
4


0.
21


5
0.


12
3


Lo
w


er
m


id
dl


e
in


co
m


e
0.


31
5


0.
31


2
0.


19
6


0.
17


4
0.


31
7


0.
23


1
0.


14
4


Lo
w


in
co


m
e


0.
43


2
0.


40
5


0.
27


3
0.


22
5


0.
46


3
0.


34
3


0.
24


8


Pa
ne


l
B


:A
ve


ra
ge


s
by


re
gi


on
al


gr
ou


ps


Ea
st


A
si


a
&


Pa
ci


fic
0.


26
7


0.
25


5
0.


14
4


0.
15


2
0.


29
9


0.
22


4
0.


12
6


Eu
ro


pe
&


C
en


tr
al


A
si


a
0.


25
8


0.
26


2
0.


16
9


0.
13


1
0.


25
8


0.
16


1
0.


13
3


La
ti


n
A


m
er


ic
a


&
C


ar
ib


be
an


0.
31


5
0.


30
2


0.
19


7
0.


16
0


0.
34


1
0.


28
0


0.
11


7


M
id


dl
e


Ea
st


&
N


or
th


A
fr


ic
a


0.
35


2
0.


27
6


0.
16


3
0.


16
0


0.
41


8
0.


35
4


0.
14


8


N
or


th
A


m
er


ic
a


0.
11


2
0.


11
1


0.
04


9
0.


06
9


0.
11


9
0.


07
2


0.
05


5


So
ut


h
A


si
a


0.
26


0
0.


26
9


0.
14


2
0.


17
0


0.
25


2
0.


17
2


0.
11


1


Su
b-


Sa
ha


ra
n


A
fr


ic
a


0.
38


4
0.


37
3


0.
23


3
0.


21
9


0.
39


8
0.


28
5


0.
21


1


W
es


te
rn


Eu
ro


pe
0.


20
6


0.
20


9
0.


13
2


0.
10


1
0.


20
0


0.
10


4
0.


12
2


W
or


ld
av


er
ag


e
0.


29
6


0.
28


5
0.


17
7


0.
15


7
0.


31
2


0.
22


6
0.


14
3


24




Figure 1: Trend of overall trade misreporting by China, 1996-2015


(at least in part) explain China’s changes in these indices over time is related to possibly illicit


flows of capital. Recently, Chen and Qian (2016) developed extensive measures to capture the


ongoing changes in China’s capital control regime, using detailed information from the IMF’s


Annual Report on Exchange Arrangements and Exchange Restrictions (AREAER) for the 1999 to 2012


period. Their de jure and hybrid indices reflect a persistent process of liberalizing China’s capital


account since 2000. Specifically, Chen and Qian (2016) report that China liberalized its capi-


tal outflow controls faster than its control on capital inflows, which may encourage outward


FDI to support China’s ‘going global’ policy initiative of 2002. Interestingly, both our import


over-reporting and export under-reporting indices for China in Figure 2 exhibit a consistent


downward trend since 2001. In fact, our import over-reporting index correlates positively with


Chen and Qian’s (2016) de jure and hybrid capital outflow control indices, with correlations


of 0.90 and 0.75, respectively. Further, our export under-reporting index also correlates posi-


tively with Chen and Qian’s (2016) de jure and hybrid capital outflow control indices, with even


stronger correlation coefficients of 0.98 and 0.93. In sum, our derived indices are consistent


with the specific explanation put forth by Chen and Qian (2016).


Finally, Chen and Qian’s (2016) hybrid index shows higher magnitudes of inflow controls


25




0.
10


0.
15


0.
20


0.
25


0.
30


1995 2000 2005 2010 2015
Year


Import over−reporting Export under−reporting


Import over− and export−underreporting
by China, 1996−2015


0.
04


0.
06


0.
08


0.
10


0.
12


1995 2000 2005 2010 2015
Year


Import under−reporting Export over−reporting


Import under− and export−overreporting
by China, 1996−2015


Figure 2: Trade misreporting in China, 1996-2015.


than their de jure index. Chen and Qian (2016) report that China has experienced an episode


of ‘hot money’ inflows since 2003 and the Chinese government’s constant initiatives to restrain


such capital inflows. Similarly, Ferrantino et al. (2012) suggest the possibility of ‘hot money’


inflows from the US into China during the 2003-2008 period. Interestingly, our export over-


reporting index for China reveals a constant upward trend beginning in 2003, which is consis-


tent with that hypothesis. However, it has reversed and started to decline from 2013 onwards,


which may reflect China’s sharp relaxation of its capital inflow regime during that period. We


particularly notice China’s new rules on FDI in late 2011, which officially allows foreigners to


invest on the Chinese mainland with offshore funds.15 China’s import under-reporting index


remains almost stable since 2000.


Overall, these descriptions are, of course, purely suggestive at this point. Nevertheless, it


is interesting to see that our indices show developments that are consistent with hypotheses


about China’s development and closely correlated with other China-specific indices. With this


in mind, we now turn to the final contribution of this paper with a specific empirical application


of our import under-reporting index to the role of tariffs and value-added taxes.


15For details on these measures, we refer to the Global Legal Monitor of the Library
of Congress of the US (available under http://www.loc.gov/law/foreign-news/article/
china-new-rules-on-foreign-direct-investment-with-renminbi/) and the IMF’s AREAER dataset (avail-
able under http://www.elibrary-areaer.imf.org/Pages/ChapterQuery.aspx).


26




5 Empirical Application: Tariff and VAT Rates


5.1 Setting


In this section, we provide one application of our misreporting indices, predicting the import


under-reporting index with tariff and VAT rates in our panel dataset. We choose to examine the


under-reporting of imports because it remains the main focus of the existing literature on trade


misreporting (for example, see Javorcik and Narciso, 2008, Mishra et al., 2008, or Ferrantino


et al. (2012)). Intuitively, an economic agent may try to curb their import costs by avoiding (or


at least minimizing) tariff payments – a value that is usually based on the import value. In


other words, everything else equal, we would expect import values to be more under-reported


when tariff rates are high. Of course, a range of other factors may play an independent role


and we will shortly discuss the list of control variables we consider.


Indeed, the existing literature finds empirical evidence of systematic under-reporting of im-


ports motivated by burdensome tariffs. Bhagwati (1964) reports strong evidence of understated


imports at the Turkish end, which is systematically correlated with tariffs and import controls;


Fisman and Wei (2004) find Chinese imports from Hong Kong to be to be under-reported;


Mishra et al. (2008) identify similar dynamics for Indian imports from its major trading part-


ners; Ferrantino et al. (2012) suggest the same when it comes to trade between the US and


China, as US importers are likely trying to avoid paying import tariffs. (As discussed before,


these studies focus on either reported trade between a pair of bilateral trading partners or


reported trade between a particular country of interest and its major trading partners.)


Our objective here is twofold. First, we want to examine whether our import under-


reporting index is also supported by the economic intuition of import under-reporting, as ev-


idenced by previous literature. Second, we want to evaluate whether import under-reporting


motivated by tariff evasion could be a global phenomenon or whether this phenomenon re-


mains unique to some selected bilateral trade relationships. In addition to the potential role of


tariff rates, we also investigate whether VAT rates, which are calculated and payable according


to the reported import value, are positively correlated with the import under-reporting index.


27




5.2 Econometric Specification


We estimate a simple linear regression model, predicting the import under-reporting index with


tariff and VAT rates in country i and year t. To properly isolate potential relationships, we also


control for several other variables that may independently affect the reporting of imports. Fur-


ther, we account for country- and year-fixed effects to control for any country- and time-specific


phenomena that could drive under-reported imports. For instance, a country’s geography or


regular trading partners (perhaps stemming from historical connections, such as colonialism)


may systematically influence the reporting of trade data. Similarly, persistent cultural and insti-


tutional characteristics could affect misreporting. With respect to time-specific unobservables,


global recessions or booms could systematically drive global misreporting rates. Two-way fixed


effects are able to isolate our analysis from any such dynamics. Formally, we estimate


URImi,t = β0 + β1Tari f f i,t + β2VATi,t + Xi,tγ+ αi +ωt + ε i,t, (22)


where URImi,t refers to the import under-reporting index for country i in year t. Tari f fi,t mea-


sures the trade-weighted applied tariff rates for all products from all source countries to each


importing country i at time t, whereas VATi,t represents the value added tax rates applicable


to all imports by the importing country. Xi,t constitutes a vector of other observable country


characteristics that may carry an independent effect on reporting behavior. Specifically, we in-


clude measures for (i) capital account openness, (ii) trade openness, (iii) democracy, (iv) and


corruption. Bhagwati (1964) and Ferrantino et al. (2012) discuss the possibility of misreporting


of trade data as one of several methods to avoid capital controls, while Fisman and Wei (2009)


reports a positive correlation between corruption and trade data discrepancies. Kellenberg and


Levinson (2016) also employ capital controls and corruption while explaining misreported trade


and tariff evasion. Further, we control for trade openness and democracy since higher levels


of integration with the global trade network and a more democratic system, associated with


more inclusive political institutions and the prevalence of the rule of law, may well form in-


dependent drivers of misreporting trade numbers. In addition, country- and time-fixed effects


28




are captured by αi and ωt, whereas ε i,t represents the usual error term. Throughout our esti-


mations we report both robust standard errors and clustered at the country level. Finally, we


multiply our import under-reporting index by 100 to facilitate the quantitative interpretation of


coefficients.


5.3 Data Sources


We access data on corruption levels from the Corruption Perceptions Index (CPI, provided by


Transparencey International, 2017).16 From 1995 to 2011, the CPI ranged from zero to ten, but


since 2012 the index ranges from zero to 100, following an update in methodology. We rescale


earlier data to match the post-2011 range from zero to 100. Note that the CPI codebook specifi-


cally mentions this switch in measurement comes because researchers should not compare data


before 2012 with those since then. In our case, however, accounting for time-fixed effects should


account for such measurement issues. (Nevertheless, all our findings are consistent when ex-


cluding the CPI.)


GDP per capita (constant 2010 US$), trade weighted applied tariff rates, value added tax


(VAT) rates, and population data are collected from the World Bank’s “World Development In-


dicators” (Group, 2016). For capital account openness, we use the Chinn-Ito index (KAOPEN),


which measures a country’s degree of capital account openness.17 The scale of the KAOPEN in-


dex ranges from the “most financially open” valued of 2.37 to the “least financial open”, scored


at -1.90. In addition, we use the polity2 variable from the Polity IV dataset to measure the coun-


try’s degree of democracy in the respective year (Marshall and Jaggers, 2017). This variable


captures the regime authority spectrum on a 21-point scale ranging from -10 (complete autoc-


racy) to +10 (consolidated democracy). Table A7 presents summary statistics of all variables


used in this econometric analysis.


16The CPI has been developed by Transparency International since 1995, providing “country level annual corrup-
tion scores” based on the perceived levels of corruption, as determined by expert assessments and opinion surveys.


17The KAOPEN index was initially introduced by Chinn and Ito (2006) and the latest update covers the time
period of 1970-2015 for 182 countries.


29




5.4 Empirical Results


The results from our econometric specifications are reported in Table 6. Note that we display


robust standard errors in parentheses under the respective coefficients and standard errors clus-


tered at the country level in brackets. We begin by examining the univariate relationships be-


tween the import under-import reporting index and our two variables of interest: tariff and VAT


rates. The corresponding coefficients are displayed in columns (1) and (2). Regression (3) then


considers tariff and VAT rates as simultaneous predictors of the under-reporting of imports. In


column (4), we introduce our set of control variables, while columns (5) and (6) incorporate


country- and year-fixed effects. To facilitate the comparison of results across regressions, we


only employ observations in which information for all variables is available. Nevertheless, all


results are robust when using all available observations for the respective specifications.


The results concerning tariff and VAT rates provide strong support for the hypothesis that an


increase in either rate is associated with a significant increase in the under-reporting of imports.


These results emerge for all six specifications and are consistent with the discussed country-


specific studies. It may also be useful to consider the derived magnitudes of the effects. In


the most complete specification (column 6), the implied magnitudes for tariff and VAT rates are


quite comparable. A one standard deviation increase in tariff rates (equivalent to approximately


4.6 points) would be associated with a 0.9 point rise in the import under-reporting index, on


average. When it comes to VAT rates, a one standard deviation increase (equivalent to 5.3


points) corresponds to a 1.1 point increase in the import under-reporting index.


Finally, we can put these magnitudes in context with a simple back-of-the-envelope calcula-


tion. For example, what would a 2% change in the import under-reporting index really mean?


Let’s take the example of India. In 2015, the value of the import under-reporting index for India


was 0.060, meaning India under-reported its imports by around US$6.4 for US$100 reported. In


2015, total reported imports of India was US$390,745 million and the country’s trade-weighted


average tariff rate was 6.35 percent. Therefore, an estimated $1.6 billion of Indian tariff rev-


enue is suggested to be lost due to under-reporting of imports. Thus, a hypothetical 2 percent


decrease in the value of the import under-reporting index of India would correspond to an


30




Table 6: Predicting the import under-reporting index with tariff and VAT rates in an unbalanced
panel of 107 countries with annual data from 1996-2015.


(1) (2) (3) (4) (5) (6)


Dependent variable: Import under-reporting index (mean=14.90)


Tariff 0.654 0.733 0.548 0.222 0.195
(0.097)∗∗∗ (0.100)∗∗∗ (0.104)∗∗∗ (0.084)∗∗∗ (0.087)∗∗
[0.261)∗∗ [0.263] ∗∗∗ [0.257]∗∗ [0.084]∗∗∗ [0.087]∗∗


VAT 0.317 0.411 0.423 0.192 0.202
(0.057)∗∗∗ (0.064)∗∗∗ (0.066)∗∗∗ (0.065)∗∗∗ (0.064)∗∗∗
[0.153]∗∗ [0.162]∗∗ [0.165]∗∗ [0.065]∗∗∗ [0.064]∗∗∗


Capital account openness -0.064 -0.263 -0.274
(0.254) (0.631) (0.634)
[0.589] [0.631] [0.634]


Trade openness 0.038 0.014 0.016
(0.005)∗∗∗ (0.020) (0.025)
[0.016]∗∗ [0.020] [0.025]


Democracy (polity2) -0.122 0.085 0.064
(0.089) (0.143) (0.137)
[0.232] [0.143] [0.137]


Corruption (CPI) -0.093 -0.023 -0.008
(0.014) ∗∗∗ (0.049) (0.049)
[0.037]∗∗ [0.049] [0.049]


Country-fixed effects Yes Yes
Year-fixed effects Yes
Observations 1,344 1,344 1,344 1,344 1,344 1,344
R-squared 0.069 0.024 0.108 0.173 0.145 0.111


Notes: The dependent variable is import under-reporting index as defined in equation 22 in the text. Robust standard errors are


displayed in parentheses and robust standard errors, clustered by reporting country, are listed in brackets. ∗ p < 0.10, ∗∗ p < 0.05,
∗∗∗ p < 0.01.


31




increase of around US$550 million of tariff revenue in the year 2015.


6 Conclusion


This paper proposes a novel methodology to estimate a country’s degree of trade misreport-


ing. Our methodology is not based on ad-hoc assumptions about who may or may not report


accurately; rather, it incorporates the full range of available data to compute the trade report-


ing patterns of a country with all of its trading partners in a given time period. We use this


information to weigh each reported trade entry and eventually we derive seven specific trade


misreporting indices, capturing overall trade misreporting, as well as under- and over-reporting


of exports and imports. Another unique aspect of the indices developed here is that they are


scale independent, making them comparable across countries with different trade values and


over different time periods.


After introducing the theoretical derivation, we apply our measurement technique to bilat-


eral annual trade data from 1996-2015, covering over 58 million trade entries at the HS 4-digit


level reported by 160 WTO members, accounting for approximately 98 percent of world mer-


chandise trade. To our knowledge, this constitutes the first trade misreporting indices that are


comparable across countries and over time, as well as independent of a priori definitions about


countries’ reporting accuracies. In a descriptive analysis of the associated country rankings,


we find low income countries to misreport relatively more, possibly reflecting their capacity


constraints and overall restrictive policy regimes, as well as weak governance and institutional


quality. Emerging economies, including primary resource exporting countries, are more likely


to over-report exports – an indication for illicit capital flight. We then specifically analyze the


prominent case of China’s trade data and our indices suggest the country’s overall trade report-


ing started to improve substantially when negotiations over joining the WTO began in the late


1990s. Further, China’s relaxation of its restrictive capital control policies coincides with a fall


in the country’s export under-reporting.


Finally, to we present an empirical analysis of import under-reporting, using our full (un-


balanced) panel data set of 107 countries from 1996-2015. Specifically, economic intuition, as


32




well as several country-specific studies, suggest that as tariff or VAT rates rise, so should the


incentive of importers to under-report, thereby avoiding additional taxation. Indeed, our results


provide evidence consistent with that hypothesis on a global level, even after accounting for a


list of potentially confounding factors, as well as country- and year-fixed effects.


Beyond these specific results, we hope that our derived indices can be of value for re-


searchers interested in a batter understanding of the determinants and consequences of misre-


ported trade data on the global level. For example, the indices may be used to study a range of


trade policy analyses, such as estimating the welfare effects of trade facilitation programs (e.g.,


tariff liberalization or preferential trading arrangements); devising effective export support and


capital control programs; or supplementing bilateral and multilateral trade negotiations and


foreign policy making, to name a few. Naturally, we do not claim these indices to be perfect.


However, we hope they provide a starting point to global studies on trade misreporting.


33




References


Arslan, I. and van Wijnbergen, S. (1993). Export incentives, exchange rate policy and export
growth in Turkey. The Review of Economics and Statistics, 75(1):128–133.


Autor, D. H., Dorn, D., and Hanson, G. H. (2013). The China syndrome: Local labor market
effects of import competition in the United States. American Economic Review, 103(6):2121–68.


Bajona, C. and Chu, T. (2010). Reforming state owned enterprises in China: Effects of WTO
accession. Review of Economic Dynamics, 13(4):800–823.


Bhagwati, J. (1964). On the underinvoicing of imports. Bulletin of the Oxford University Institute
of Economics & Statistics, 27(4):389–397.


Bhagwati, J. (1967). Fiscal policies, the faking of foreign trade declarations, and the balance of
payments. Bulletin of the Oxford University Institute of Economics & Statistics, 29(1):61–77.


Bhagwati, J. (1981). Alternative theories of illegal trade: Economic consequences and statistical
detection. Weltwirtschaftliches Archiv, Bd. 117(H. 3):409–427.


Brandt, L., Van Biesebroeck, J., Wang, L., and Zhang, Y. (2017). WTO accession and performance
of Chinese manufacturing firms. American Economic Review, 107(9):2784–2820.


Buchanan, J. M., Tollison, R. D., and Tullock, G. (1980). Toward a theory of the rent-seeking society.
Number 4. Texas A & M Univ Pr.


Buehn, A. and Eichler, S. (2011). Trade misinvoicing: The dark side of world trade. The World
Economy, 34(8):1263–1287.


ChathamHouse (2018). resourcetrade.earth.


Chen, J. and Qian, X. (2016). Measuring on-going changes in China’s capital controls: A de jure
and a hybrid index data set. China Economic Review, 38:167–182.


Chinn, M. D. and Ito, H. (2006). What matters for financial development? Capital controls,
institutions, and interactions. Journal of Development Economics, 81(1):163 – 192.


Colantone, I. and Stanig, P. (2018). Global competition and Brexit. American Political Science
Review.


Egger, P. H. and Larch, M. (2012). Tariff evasion effects in quantitative general equilibrium.
Economics Letters, 116(2):262 – 264.


Feenstra, R. C., Hai, W., Woo, W. T., and Yao, S. (1999). Discrepancies in international data: An
application to China-Hong Kong entrepôt trade. The American Economic Review, 89(2):338–343.


Feng, L., Li, Z., and Swenson, D. L. (2017). Trade policy uncertainty and exports: Evidence
from China’s WTO accession. Journal of International Economics, 106:20–36.


Ferrantino, M. J., Liu, X., and Wang, Z. (2012). Evasion behaviors of exporters and importers:
Evidence from the U.S.-China trade data discrepancy. Journal of International Economics,
86(1):141 – 157.


34




Ferrantino, M. J. and Wang, Z. (2008). Accounting for discrepancies in bilateral trade: The case
of China, Hong Kong, and the United States. China Economic Review, 19(3):502 – 520.


Fisman, R. and Wei, S. (2004). Tax rates and tax evasion: Evidence from ‘missing imports’ in
China. Journal of Political Economy, 112(2):471–496.


Fisman, R. and Wei, S.-J. (2009). The smuggling of art, and the art of smuggling: Uncovering the
illicit trade in cultural property and antiques. American Economic Journal: Applied Economics,
1(3):82–96.


Group, W. B. (2016). World Development Indicators 2016. World Bank Publications.


He, H., Chen, S., Yao, S., and Ou, J. (2014). Financial liberalisation and international market
interdependence: Evidence from China’s stock market in the post-WTO accession period.
Journal of International Financial Markets, Institutions and Money, 33:434 – 444.


Hummels, D. (2007). Transportation costs and international trade in the second era of global-
ization. Journal of Economic Perspectives, 21(3):131–154.


IMF (1993). A Guide to Direction of Trade Statistics. International Monetary Fund (IMF).


IMTS (2010). International merchandise trade statistics: Concepts and definitions 2010.


Jara, A. and Escaith, H. (2012). Global value chains, international trade statistics and policy-
making in a flattening world. World Economics, 13(4):5–18.


Javorcik, B. S. and Narciso, G. (2008). Differentiated products and evasion of import tariffs.
Journal of International Economics, 76(2):208 – 222.


Javorcik, B. S. and Narciso, G. (2017). WTO accession and tariff evasion. Journal of Development
Economics, 125(Supplement C):59 – 71.


Kar, D. and Spanjers, J. (2015). Illicit Financial Flows from Developing Countries: 2004-2013. Global
Financial Integrity.


Kellenberg, D. and Levinson, A. (2016). Misreporting trade: Tariff evasion, corruption, and
auditing standards. Technical report, National Bureau of Economic Research.


Khandelwal, A. K., Schott, P. K., and Wei, S.-J. (2013). Trade liberalization and embedded insti-
tutional reform: Evidence from Chinese exporters. American Economic Review, 103(6):2169–95.


Lu, Y. and Yu, L. (2015). Trade liberalization and markup dispersion: evidence from China’s
wto accession. American Economic Journal: Applied Economics, 7(4):221–53.


Marini, M., Dippelsman, R., and Stanger, M. (2018). New estimates for direction of trade
statistics. IMF Working Papers, (WP/18/16).


Marshall, M. G. and Jaggers, K. (2017). Polity IV project: Political regime characteristics and
transitions, 1800-2017.


Miao, G. and Fortanier, F. (2017). Estimating transport and insurance costs of international
trade. OECD Statistics Working Papers, (2017/04).


35




Mishra, P., Subramanian, A., and Topalova, P. (2008). Tariffs, enforcement, and customs evasion:
Evidence from India. Journal of Public Economics, 92(10):1907 – 1925.


Morgenstern, O. et al. (1963). On the accuracy of economic observations. Princeton University Press.


Prasad, M. E., Wang, M. Q., and Rumbaugh, M. T. (2005). Putting the cart before the horse? Capital
account liberalization and exchange rate flexibility in China. International Monetary Fund.


Schneider-Petsinger, M. (2017). Trade policy under President Trump: Implications for the US
and the world. Reserach paper, The Royal Institute of International Affairs, Chatham House.


Sheikh, M. A. (1974). Underinvoicing of imports in Pakistan. Oxford Bulletin of Economics and
Statistics, 36(4):287–296.


Timmer, M., Erumban, A. A., Gouma, R., Los, B., Temurshoev, U., de Vries, G. J., Arto, I.-a.,
Genty, V. A. A., Neuwahl, F., Francois, J., et al. (2012). The world input-output database
(WIOD): contents, sources and methods.


Transparencey International (2017). Transparency International: Corruption perceptions index.


UNCTAD (2016). Trade Misinvoicing in Primary Commodities in Developing Countries: The cases of
Chile, Côte d’Ivoire, Nigeria, South Africa and Zambia. Number UNCTAD/SUC/2016/2. United
Nations Conference on Trade and Development (UNCTAD).


36




A Appendix


A.1 Example of mirror import data


Table A1: Mirror trade flow reported by importer Tunisia (d1) and all source countries: Brazil
(s1), Bangladesh (s2), and Australia (s3).


HS-4 code Destination Source Import value Export value Absolute Reporting
($000) ($000) distance ($000)


(d1 s1)





0110 Tunisia (d1) Brazil (s1) 15 15
0806 Tunisia (d1) Brazil (s1) 20 20
0901 Tunisia (d1) Brazil (s1) 60 100 40
4040 Tunisia (d1) Brazil (s1) 50 40 10
5050 Tunisia (d1) Brazil (s1) 40 50 10
6060 Tunisia (d1) Brazil (s1) 25 25
7009 Tunisia (d1) Brazil (s1) 10 10
8080 Tunisia (d1) Brazil (s1) 5 5
8(3) 6(3) 5(3) 190(150) 225(190) 135


(d1 s2)





1010 Tunisia (d1) Bangladesh (s2) 20
2020 Tunisia (d1) Bangladesh (s2) 40 60 20
3030 Tunisia (d1) Bangladesh (s2) 60 80 20
4040 Tunisia (d1) Bangladesh (s2) 80 100 20
5050 Tunisia (d1) Bangladesh (s2) 100 90 10
6060 Tunisia (d1) Bangladesh (s2) 100
7070 Tunisia (d1) Bangladesh (s2) 75
8080 Tunisia (d1) Bangladesh (s2) 20
8(4) 5(4) 7(4) 300(280) 525(330) 285


(d1 s3)





2020 Tunisia (d1) Australia (s3) 10 10
3030 Tunisia (d1) Australia (s3) 150 300 150
4040 Tunisia (d1) Australia (s3) 120 100 20
5050 Tunisia (d1) Australia (s3) 110 100 10
7009 Tunisia (d1) Australia (s3) 80 80
8080 Tunisia (d1) Australia (s3) 30 30
6(3) 4(3) 5(3) 390(380) 610(500) 300


(d1 sn, n = 3) 22(10) 14(10) 16(10) 880(810) 1,360(1,020) 720


Notes: Both imports and exports are considered here in comparable FOB values to eliminate discrepancies resulted
from CIF and FOB price reportings by the importer and the exporters, respectively.


A.2 Mis-Reporting or Mis-Classification? Using HS 4-Digit Product Level Trade


Data


The Harmonized Commodity Description and Coding System (or simply HS), developed, main-


tained, and monitored by the World Customs Organization (WCO) was introduced in 1988 and


37




has since been adopted by most countries worldwide as a basis for collecting international


trade statistics. It currently covers more than 98 percent of merchandise international trade


globally and national customs authorities of more than 200 WCO member countries.18 The HS


comprises approximately 5,300 product descriptions that appear as headings and subheadings,


arranged in 99 chapters, grouped in 21 sections.


The uniform product classification across countries only goes down to HS 6-digt level of


disaggregation, while national product classifications often extended up to 8 to 10 digit level


(e.g., India and Singapore use 8-digit product classification, while China, UK and USA use


10-digit national product classification.) Thus, internationally available trade data comparable


across countries allow us to use at best HS 6-digit disaggregated data for measuring trade


misreporting. One might tends to attribute a portion of discrepancies in reported bilateral


international trade data to different product classifications used by different countries and the


possibility of unintentional misclassification of products by national customs authorities. This


demands a brief discussion of HS Nomenclature and Classification of Goods.


Table A2 shows an example of the HS nomenclature. The six digits HS product code can


be broken down into three parts. The first two digits (HS 2-digit) identify the chapter the


goods are classified in, e.g., 09 corresponds to ‘Coffee, Tea, Maté, and Spices’. The chapter is


further divided by adding two digits (HS 4-digit) to identify groupings within that chapter,


e.g., 09.01 is associated with ‘Coffee, whether or not roasted or decaffeinated’. Finally, the next


two digits (HS 6-digit) are even more specific, e.g., 09.01.11 identifies ‘Coffee, not roasted and


not decaffeinated’. Up to the HS 6-digit level, all countries classify products in the same way.


Thus, while the probability of unintentional misclassification is not completely ruled out (mix-


up between coffee, not roasted and roasted, or not decaffeinated and decaffeinated) at the HS


6-digit level, there should not be any such unintentional misclassification at the HS 4-digit level


(since coffee and tea are completely different products). Therefore, to avoid potential issues of


‘unintentional misclassification’ of products by some countries, our analysis focuses on the HS


4-digit product level of disaggregation.


18As per the WCO website, accessed on 3 November 2017; available under http://www.wcoomd.org/en/topics/
nomenclature/overview/what-is-the-harmonized-system.aspx


38




Table A2: An example of HS product classification by the WCO: First two headings of Chapter
9.


Chapter Heading Sub heading
(HS Code)


Product description


09 Coffee, tea, maté and spices
09.01 Coffee, whether or not roasted or decaffeinated; coffee husks and skins; coffee sub-


stitutes containing coffee in any proportion.
- Coffee, not roasted:


0901.11 - - Not decaffeinated
0901.12 - - Decaffeinated


- Coffee roasted:
0901.21 - - Not decaffeinated
0901.22 - - Decaffeinated
0901.90 - Other


09.02 Tea, whether or not flavoured.
0902.10 - Green tea (not fermented) in immediate packings of a content not exceeding 3 kg
0902.20 - Other green tea (not fermented)
0902.30 - Black tea (fermented) and partly fermented tea, in immediate packings of a content


not exceeding 3 kg
0902.40 - Other black tea (fermented) and other partly fermented tea


Notes: The 2012 edition of the WCO HS Nomenclature is available at
http://www.wcoomd.org/en/topics/nomenclature/instrument-and-tools/hs_nomenclature_previous_


editions/hs_nomenclature_table_2012.aspx.


Further, while the WCO reviews and amends the HS every five years, these revisions mainly


targeted the fine-tuning and ensure better coverage of trade statistics at the HS-6 level.19 There-


fore, by focusing on the HS 4-digit product level we also alleviate concerns about all countries


potentially not reporting their trade data using the same version of the HS nomenclature.


A.3 Using different CIF/FOB conversion factor


Since the use of IMF recommended 6 percent CIF/FOB conversion may still leave some doubts,


as this estimate is also based on flawed (misreported) data, and one would argue it is useless to


impose such an average number since transport and insurance widely varies across product cat-


egories, trading partners including its distance from the counterparts and mode of transports.


To check the sensitivity of our estimated indices to the use of CIF/FOB conversion factor, we


19For example, the HS Nomenclature 2017 Edition includes 233 sets of amendments, mostly featuring the
environmental and social issues of global concern. For a detailed discussion on the changes introduced
in the 2017 edition, we refer to http://www.wcoomd.org/en/topics/nomenclature/instrument-and-tools/
hs-nomenclature-2017-edition/amendments-effective-from-1-january-2017.aspx.


39




test our index estimation with the traditional factor of 1.1. However, this exercise does not have


any significant effect on our original indices apart from some trivial changes in the index values


(see for example Table A3). This is also reflected in the correlation coefficients with our original


indices, which are around 0.99 for overall misreporting index as well as other sub-indices.


Table A3: Comparison of overall trade misreporting index (TMRI) estimated using different
CIF/FOB conversion factor for top and bottom ten countries in 2015.


Top 10 Misreporting country Bottom 10 Misreporting country


Rank Country TMRI using TMRI using Rank Country TMRI using TMRI using
CIF/FOB 1.06 CIF/FOB 1.1 CIF/FOB 1.06 CIF/FOB 1.1


1 Togo 0.784 0.788 118 Brazil 0.154 0.153
2 Antigua and Barbuda 0.713 0.717 119 Japan 0.148 0.146
3 Panama 0.712 0.719 120 Germany 0.144 0.148
4 Afghanistan 0.636 0.640 121 Italy 0.140 0.144
5 Malta 0.614 0.620 122 Argentina 0.137 0.137
6 Benin 0.613 0.620 123 United States 0.133 0.135
7 Kuwait 0.592 0.591 124 Mexico 0.133 0.135
8 Sierra Leone 0.561 0.563 125 Chile 0.124 0.124
9 Solomon Islands 0.494 0.490 126 Peru 0.123 0.124
10 Niger 0.481 0.481 127 Canada 0.098 0.106


A.4 Correlations with common macroeconomic indicators and correlations between


the indices


Table A4 provides simple correlations among the misreporting indices, and Table A5 displays


correlation coefficients between all seven misreporting indices and most common macroeco-


nomic indicators including population size, GDP per capita, a democracy score (using the


polity2 variable from the Polity IV indicators), corruption levels, capital account openness, and


trade openness.


40




Table A4: Correlation coefficients among different trade misreporting indices.


Index: Overall trade Import Import Import Export Export Export
misreporting misreporting under-reporting over-reporting misreporting under-reporting over-reporting


Overall trade misreporting 1.00


Import misreporting 0.88∗∗∗ 1.00
(0.00)


Import under-reporting 0.81∗∗∗ 0.92∗∗∗ 1.00
(0.00) (0.00)


Import over-reporting 0.60∗∗∗ 0.67∗∗∗ 0.35∗∗∗ 1.00
(0.00) (0.00) (0.00)


Export misreporting 0.89∗∗∗ 0.64∗∗∗ 0.57∗∗∗ 0.47∗∗∗ 1.00
(0.00) (0.00) (0.00) (0.00)


Export under-reporting 0.84∗∗∗ 0.60∗∗∗ 0.58∗∗∗ 0.38∗∗∗ 0.97∗∗∗ 1.00
(0.00) (0.00) (0.00) (0.00) (0.00)


Export over-reporting 0.57∗∗∗ 0.44∗∗∗ 0.30∗∗∗ 0.50∗∗∗ 0.60∗∗∗ 0.39∗∗∗ 1.00
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00)


Note: P-values are displayed in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.


Table A5: Correlation coefficients between trade misreporting indices and common macroeco-
nomic indicators on the country level.


Index: Overall trade Import Import Import Export Export Export
misreporting misreporting under-reporting over-reporting misreporting under-reporting over-reporting


Population size (log) -0.40∗∗∗ -0.42∗∗∗ -0.32∗∗∗ -0.40∗∗∗ -0.37∗∗∗ -0.36∗∗∗ -0.17∗∗∗
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)


GDP per capita (log) -0.39∗∗∗ -0.44∗∗∗ -0.31∗∗∗ -0.51∗∗∗ -0.27∗∗∗ -0.20∗∗∗ -0.37∗∗∗
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)


Democracy (polity2) -0.39∗∗∗ -0.31∗∗∗ -0.18∗∗∗ -0.42∗∗∗ -0.39∗∗∗ -0.37∗∗∗ -0.28∗∗∗
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)


Corruption (CPI) -0.32∗∗∗ -0.35∗∗∗ -0.24∗∗∗ -0.39∗∗∗ -0.22∗∗∗ -0.20∗∗∗ -0.20∗∗∗
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)


Capital account openness -0.26∗∗∗ -0.28∗∗∗ -0.19∗∗∗ -0.33∗∗∗ -0.20∗∗∗ -0.17∗∗∗ -0.16∗∗∗
(0.00) (0.00) (0.00) (0.00) (0.00) (0.00) (0.00)


Trade openness -0.10∗∗∗ -0.10∗∗∗ -0.11∗∗∗ -0.00 0.15∗∗∗ 0.13∗∗∗ 0.09∗∗∗
(0.00) (0.00) (0.00) (0.95) (0.00) (0.00) (0.00)


Note: P-values are displayed in parentheses. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.


41




A.5 Summary Statistics of all Trade Misreporting Indices for all countries and data


employed in Econometric Application


Table A6: Summary statistics: All trade misreporting indices for all countries.


Variables Obs Mean Std. Dev. Min Max


Overall trade misreporting index 2,472 0.30 0.14 0.08 0.96


Export misreporting index 2,461 0.31 0.19 0.06 1.00


Export under-reporting index 2,461 0.23 0.21 0.03 1.00


Export over-reporting index 2,461 0.14 0.09 0.01 0.50


Import misreporting index 2,464 0.28 0.13 0.08 0.90


Import under-reporting index 2,464 0.18 0.13 0.03 0.89


Import over-reporting index 2,464 0.16 0.07 0.04 0.50


Table A7: Summary statistics of data used in our application.


Variables Obs Mean Std. Dev. Min Max


Import under-reporting index [0 to100] 1,344 14.90 10.79 3.32 84.10


Tariff rate (applied, trade weighted mean, all products) (%) 1,344 4.64 4.33 0.00 28.55


Value added tax (VAT) rate (%) 1,344 10.65 5.30 0.05 67.74


Capital account openness [-1.90 to 2.37] 1,344 1.04 1.49 -1.90 2.37


Trade openness (trade % of GDP) 1,344 86.59 49.99 16.44 441.60


Democracy (polity2) [-10 to +10] 1,344 6.92 4.58 -9.00 10.00


Corruption (CPI) [0-100] 1,344 50.03 22.52 12.00 100.00


Note: This table is based on the sample used in the regression presented in Table 6


42




Login