Implications of COMTRADE Compilation Practices for Trade Barrier Analyses and Negotiations

To facilitate meaningful cross-country comparisons, the United Nations Statistical Office (UNSO) justifiably requests that members report import statistics to the Commodity Trade (COMTRADE) Statistics Division on a common cost-insurance-freight (c.i.f.) basis, even in the case of countries like the United States, Canada, Australia, and New Zealand, where free-on-board (f.o.b.) valuation tariffs are used. As a result, COMTRADE overstates the actual dutiable value of these countries' imports. A relevant question, which appears to have been given little consideration, is whether this valuation procedure invalidates the use of COMTRADE for analysis of these countries' tariffs and similar trade restrictions.

A second important point relates to the “reporting system” used for the compilation of COMTRADE statistics. Most countries employ the “general” system, which includes imports for direct consumption as well as imports under customs bond or into officially designated foreign trade zones (FTZs). The latter are not subject to national tariffs unless they eventually clear customs controls. Statistics compiled by the United States International Trade Commission (USITC) can be used to determine if inclusion of FTZ transactions in COMTRADE causes the U.N. data to significantly further misstate the dutiable value of imports. These questions are of importance because recent efforts have attempted to utilize COMTRADE for tariff change simulations and related issues in multilateral trade negotiations.

Before proceeding, one important qualification should be noted. In this study, the term “COMTRADE bias” is sometimes used. This is in no way intended as a criticism of the methodology or procedures used in constructing the database which are fully consistent with the appropriate and intended applications of the U.N. statistics. Rather, the term is directed at users who appear to be unaware of important basic characteristics of COMTRADE and have attempted to employ the data in ways that are incorrect and inappropriate.

I. Characteristics of Trade Projection Models

During the Uruguay Round, the World Bank and UNCTAD (1987) developed a partial equilibrium projection model (named SMART, for Software for Market Analysis and Restrictions to Trade) to simulate the effects of proposed tariff cuts in the negotiations. The projections employed tariff line level import statistics contained in the WTO Integrated Data Base since it was acknowledged that the COMTRADE data then available were too aggregate, were sometimes tabulated in inappropriate values, and were compiled using a methodology not appropriate for tariff simulations. The fact that far more detailed six-digit Harmonized System (HS) statistics have now become generally available has generated renewed interest in the possible use of COMTRADE for tariff analyses and simulations (see http://go.worldbank.org/IJIR5D0T80).

Although different procedures have been used for estimating trade diversion, valuation biases in COMTRADE will generate similar errors for these projections. For example, Baldwin and Murray (1977) simulate trade diversion using a methodology which incorporates the trade creation projection estimate as an explanatory factor. In this and other empirical approaches (see World Bank and UNCTAD 1987), the higher the COMTRADE valuation bias, the larger will be the resulting error in trade diversion projections.

II. The Magnitude of Cross-Product Valuation Biases

The United States International Trade Commission (USITC) provides online public access to national import statistics down to the level of individual ten-digit HTS products (see http://USITC.gov). Aside from the f.o.b. value of imports for direct consumption, this system (named “trade data web”) also provides information on international transport and insurance costs incurred in bringing the product to the first U.S. port of entry. As requested by the United Nations Statistical Office, these c.i.f. import values are reported for inclusion in U.N. COMTRADE records. As a result, COMTRADE overstates the dutiable value of imports for countries like Australia, Canada, New Zealand, and the United States that utilize f.o.b. import tariffs. A key question relating to tariff analyses and projections concerns the magnitude of this bias.

Since the USITC “trade data web” provides both free-on-board import values and international transport costs for all imported products it can be employed to quantify the tariff valuation bias for countries trading with the United States. Specifically, the nominal bias for imports from a given country can be derived by taking the ratio of all transport and insurance costs for a given product to its landed f.o.b. value.

For an empirical assessment of the bias, the following procedure was employed. First, a selection of 45 “test” countries was chosen with an effort made to achieve as much geographic and economic diversity as possible. The selection included countries in Asia, Africa, Europe, and South America, island countries like the Maldives, Fiji, and Sri Lanka, as well as several that were landlocked (Nepal and Paraguay). Next, available statistics were drawn from the USITC trade data web to compute nominal transportation costs for all six-digit HS level products imported from these countries by the United States. These freight factors were then ranked in ascending order, that is, from the lowest nominal transport cost for each product to the highest. The percentile values for each country's distribution of biases are given in Table 1.

As an illustration, the table shows that the United States imported 132 individual 6-digit HS products from the Cote d'Ivoire with a total 2007 value of $639 million. (All dollar amounts are current U.S. dollars.) The median nominal freight factor (i.e., the valuation bias) was 12.7 percent, while three-tenths of all shipments had a freight factor of 25 percent or more. A similar share of imports from Ghana, Nepal, Togo, and the Maldives incorporated biases exceeding 20 percent, while the biases for one-tenth of all imports from Belize, Senegal, Honduras, Nepal and other countries exceeded 35 percent. Ten percent of the combined U.S. imports from all countries listed in the table incurred nominal freight costs exceeding 26 percent.

Overall, for the 45 countries, only five (Uruguay, Bangladesh, Sri Lanka, Vietnam, and Pakistan) faced median Unites States tariffs that exceeded their median ad valorem transport costs. As a result, the percentage error in trade creation estimates for most countries will exceed the actual percentage change in this parameter, often by very high margins. The implications are that COMTRADE statistics can generate highly inaccurate estimates of the level and composition of the trade response to tariff changes.

While Table 1 examined the distribution of individual countries' valuation biases across all exports, an important related question is whether significant differences exist in the biases between product groups. If so, one would have an interest in identifying those sectors where COMTRADE most seriously overstates the dutiable value of imports. Second, if the valuation biases within specific product groups fall in a relatively narrow range this might suggest the possibility of employing a standard f.o.b.–c.i.f. correction factor to offset, or reduce, the bias. This approach would be similar to the ten percent factor employed by the IMF to account for differences in export and import statistics.

For relevant information, U.S. trade statistics for the countries in Table 1 were combined into six different regional groups (i.e., Southern Cone South America, Other South America and the Caribbean, Oceania, sub-Saharan Africa, Southeast Asia, and South Asia. Next, the COMTRADE bias was computed for U.S. imports of several broad categories of goods from these country groups. For four of the six country groups the valuation bias for textiles and footwear products exceeds the median bias for all goods, while meat, fish and vegetable imports frequently incur higher than average nominal transportation costs. Almost one-third of all meat exports from Oceania had nominal transport costs exceeding 21 percent, while the corresponding freight rate for sub-Saharan African vegetable products was about 31 percent. One possible explanation for these results is that due to their perishable nature, food products rely more heavily on relatively expensive air transport to access U.S. markets.

Other products sectors with relatively high valuation biases include hides and leather goods, as well as articles of plastic and glassware products. Conversely, nominal freight costs for vehicles and machinery, optical goods, and scientific instruments are generally among the lowest in the table. The overall diversity of the biases across groups provided little evidence that the COMTRADE valuation problem could be corrected by a standard f.o.b.–c.i.f. adjustment factor.¹

III. Effects of General Trade Reporting Practices

While utilization of an inappropriate cost-insurance-freight valuation base can significantly overstate the dutiable value of U.S. imports, there is an additional problem that may produce even greater biases. Specifically, two methodologies are used for compiling import statistics; namely, the general and special recording systems. Special trade statistics tabulate the value of goods imported directly for final consumption. This exchange encounters any existing tariffs and related trade control measures so special trade data are submitted by the U.S. to the WTO Integrated Data Base. In contrast, general trade statistics record the value of merchandise imports, either for direct immediate consumption or into bonded warehouses and foreign trade zones (FTZs) under customs custody. Imports under the general trade regime destined for FTZs are exempt from tariffs unless they are redirected toward domestic markets for consumption. Due to these special import provisions, general trade statistics, which are employed in the COMTRADE database, have major shortcomings for analyses of trade restrictions.²

As a result of their compilation procedures, general statistics may seriously misstate dutiable import values, and may also fail to correctly identify the goods facing trade restrictions. A hypothetical example can illustrate this point. Assume the U.S. imports $10 billion of crude petroleum (HTS 270900) into a foreign trade zone. In the zone the shipment is further processed (refined) into distillate fuel oils (HTS 271019) and this product is then transferred to the domestic market for consumption—at which point applicable U.S. tariffs are assessed. Under normal zone procedures, importers generally have the option of paying duties on the original materials imported into the zone or on the finished fabricated product. Since the nominal equivalent of specific tariffs on distillate fuels is relatively lower than those on crude oil, the former would be reported on customs forms. The statistical records of these transactions would be as follows: Table 2 provides examples of the magnitude of the general system bias using statistics on selected six-digit HS U.S. imports from Korea. The table shows the dutiable f.o.b. value for each item (column 3), which is reported to the WTO Integrated Data Base. These numbers represent the value of goods imported for direct consumption and, as such, are subject to existing tariffs. In addition, column 5 shows the corresponding c.i.f. general import value reported to COMTRADE. Differences between these values indicate the magnitude of bias associated with the use of the U.N. data for tariff analyses and/or projections.

• COMTRADE would record statistics on the actual value ($10 billion) of crude oil imports. In contrast, the WTO Integrated Database would not report any imports of crude oil, because this specific product did not cross the customs frontier for domestic consumption.

• The IDB would report statistics on the distillate fuel oil imports, since these are the shipments upon which relevant United States import duties are assessed. In contrast, U.N. COMTRADE would not report import information for these goods due to their significant physical transformation within U.S. geographic territory, that is, the foreign trade zone.

Column (4) has been added to help assess the relative size of the f.o.b.–c.i.f. valuation bias in COMTRADE data as opposed to biases originating from tabulating data using the general trade system. As an example, the c.i.f. value of dutiable imports of gear boxes (HTS 870840) is $1.5 million higher than the dutiable customs value, while the difference between the latter and the c.i.f. general import value is $319 million. These comparisons indicate that the general trade compilation practices account for almost all of the differences in import values reported in the IDB and U.N. COMTRADE.

The clear impression from Table 2 is that COMTRADE biases are of a magnitude that invalidates use of these statistics for tariff simulations or negotiations. Biases incorporated in the U.N. data may seriously misdirect national priorities for a liberalization across products, and may also significantly overstate overall potential trade gains. Specifically, Overall, for the products listed in Table 2 the combined import values reported to COMTRADE are approximately $2.1 billion higher than the true dutiable value of these goods. These comparisons indicate the U.N. statistics overstate the dutiable value of imports (and potential trade creation gains from a tariff liberalization) by about 80 percent—see the column totals. In general, major statistical discrepancies occur between COMTRADE and IDB data for raw materials, semi-finished goods, and components imported into FTZs for further processing.³

• COMTRADE overstates the dutiable value of the first item (HTS 252329 — Portland Cement) by about $57 million, or 56 percent. Since the customs c.i.f. and general import values are equal ($158.9 million), the bias is entirely attributable to the inappropriate (for tariff analysis) valuation base employed for the U.N. statistics.

• Differences of just under 1 billion dollars (approximately 67 percent) occur in the import value for petroleum oils (HTS 271019) reported to the IDB for domestic consumption and the general import total ($2.4 billion) recorded in COMTRADE. Only about $63 million, or 7 percent of the difference, is attributable to the alternative valuation bases (see columns 3 and 4).

• COMTRADE-IDB differences exceeding several thousand percent occur for rare earth imports (HTS 284610). The table indicates that about 95 percent of all imports of this product were destined for the processing zones. As such, the IDB correctly does not record most shipments of this item. Similarly, the import value reported in COMTRADE for organic solvents (HTS 381400) is more than ten times greater than the dutiable value of imports for domestic consumption.

• COMTRADE overstates the dutiable value of electrical or nonelectrical tapes (HTS 391990) by about $44 million, or approximately 300 percent. Compilation practices for general imports were a major cause of the overall difference between these statistics and the data reported to the IDB.

• If COMTRADE were used as a trade base for simulations of the effects of import tariff changes for woven synthetic fabrics (HTS 540753) the upward bias in the projection error would be 246 percent. Similar problems occur for dyed woven fabrics (HTS 551612), where COMTRADE overstates the dutiable value of imports by approximately 190 percent. General imports account for most of the difference between COMTRADE data and the relevant numbers in the IDB.

• Similarly, COMTRADE would generate an upward bias in projections of the effects of tariff changes for lock keys (HTS 830170) exceeding 600 percent. The corresponding bias for transmission bearings (HTS 848330) would be over $11.6 million, or almost 270 percent. The general reporting system accounts for almost all of the differences between the WTO and U.N. statistics.

• Differences of approximately $700 million occur between the true dutiable values for the combined imports of radio remote control apparatus (HTS 852692) plus gear boxes and parts (HTS 870840) and the general import totals ($1.0 billion) recorded for these items in COMTRADE. The U.N. statistics overstate the dutiable value of gear boxes by over 500 percent.

• The value reported in COMTRADE for imports of photographic lenses and filters (HTS 900220) is almost 50 times greater than the dutiable value of imports for consumption reported in the Integrated Data Base. Other examples in the table also reflect similar, very high, COMTRADE biases.

Even stronger negative conclusions concerning the magnitude of COMTRADE biases emerge from statistics on U.S. imports from Austria. For example, Table 3 shows general imports of cars with cylinders under 3,000cc (HTS 870323) from Austria are reported as $3.0 billion in COMTRADE, while the dutiable value of these items imports reported to the IDB is roughly $2.1 billion (or 218 percent) lower. General imports of $333 million are reported for spark ignition engines (HTS 840732) which is almost fifty times the dutiable value of these shipments. Overall, the reported general imports of the 18 items listed in Table 4 are approximately $3 billion higher than the dutiable customs value of imports. Differences for several products, like articles of magnesium, actually exceed one thousand percent.⁴

Two key points emerge from these statistics. A further major defect of COMTRADE is that it may indicate no, or relatively limited, domestic consumption of a good occurred when there was in fact significant dutiable trade. This situation is the converse of that reflected in Tables 2 and 3 where COMTRADE overstated the customs value of imports. The underreporting problem occurs when imported raw materials, semi-finished goods, or components experience significant transformation in a foreign trade zone before shipment to domestic markets. As a result of the processing activity the originally imported good may be classified under a different COMTRADE HTS heading than the final product. In these cases, statistics on the end products for domestic consumption are reported to the IDB, but not the components that originally entered the FTZ. COMTRADE would not record statistics on the final product actually consumed, but would report imports of the raw materials.

• – First, COMTRADE data have the capacity to significantly misdirect national priorities for a tariff liberalization across products. This point is reflected in the fact that rankings of general import product values may differ substantially from those based on actual dutiable values of imports. As an example, the value of U.S. general imports of spark ignition engines ($333 million) is the third highest in the table even though the actual dutiable import value ($7 million) was exceeded by 8 other items—often by very high margins.

• – Second, the COMTRADE general statistics may also provide very inaccurate and unreliable information concerning the magnitude of potential overall gains resulting from a trade liberalization. This point is reflected in the fact that the actual dutiable value of U.S. imports from Austria is approximately 40 percent lower than totals reflected in the general trade statistics (see the memo item in the table). This figure represents the potential overall error in trade creation projections that utilize COMTRADE statistics.

While it can be difficult to precisely identify the goods fabricated in FTZs, the situation is less problematic for energy imports since refinery operations generally create a range of products that are clearly petroleum or natural gas derivatives. For example, Table 4 shows the IDB is reporting total U.S. crude petroleum (HTS 270900) imports whose value is about $88 billion less than COMTRADE. The overstated COMTRADE statistics could cause exporters to place a higher than warranted priority on liberalization of existing U.S. specific tariffs on crude oil imports. However, this strategy would be misdirected since much of the discrepancy can be accounted for by differences in reported IDB-COMTRADE imports of refined energy products. This transformation is the reason why reported IDB imports of petroleum coke and bitumen are approximately $18 billion higher than the corresponding figures in U.N. COMTRADE.⁵

Aside from energy products, COMTRADE may significantly underreport dutiable imports of a broad range of products that experience transformation in FTZs. Statistics on foodstuffs, beverages and tobacco, machinery, electronics, and transport equipment are among the sectors often affected by this problem. In extreme cases, COMTRADE fails to report any imports of items for which the IDB shows dutiable trade exceeding millions of dollars occurred. Examples include U.S. imports of unwrought zinc (HTS 790111) and marine propulsion engines (HTS 840721) from Korea. In other cases, COMTRADE significantly under-reported dutiable imports by over $400 million in the case of motorized transport equipment (HTS 870323) from Korea, and by over $2 billion for cellular telephones (HTS 851712) from China.

IV. Conclusions

This study examined characteristics of COMTRADE statistics to assess their utility for tariff analysis and related applications in multilateral trade negotiations. This issue is of major importance since recent attempts have been made to use COMTRADE for tabulating the value of imports subject to tariff and nontariff restrictions and simulating the trade response to negotiated tariff changes. Accurate and reliable information on these points are key requirements for the formulation of national trade strategies, or to support multilateral trade negotiations. For several reasons, negative conclusions were reached regarding the utility of unadjusted U.N. statistics for such efforts.

First, a serious problem exists concerning the valuation base employed for COMTRADE. These statistics overstate the dutiable value of all United States imports, often very significantly, since they are expressed in cost-insurance-freight values although the U.S. employs free-on-board import tariffs.⁶ As a result, the error in COMTRADE-based trade creation projections could seriously misdirect national priorities in multilateral negotiations. This projection error extends across all regional groups of countries, as well as major product categories. Evidence was cited that indicates these problems also occur for other countries like Australia, Canada, and New Zealand that employ free-on-board import tariffs.

Second, the general trade compilation procedure used for COMTRADE may greatly amplify the detrimental effect of the valuation bias. In some cases, general import statistics overstated the dutiable value of individual six-digit HS products by several hundred percent, or by billions of dollars. Automotive equipment, machinery, electronics and energy products were often prone to this statistical bias. Third, COMTRADE may incorrectly identify specific items which are, or are not, subject to tariffs and other trade restrictions. This is due to the fact that the U.N. records tabulate information on products entering a country's geographic territory, but may fail to record relevant information on the nature and value of the goods actually clearing customs. This problem occurs when imports experience significant transformation in foreign trade zones and then clear customs under a different HTS code than that recorded in COMTRADE.⁷ Another possible cause is that the processed products were forwarded to final destinations in third countries and did not clear U.S. customs in any form. As a result, COMTRADE may both seriously overstate and understate dutiable import values.

Each of these factors is of major importance by itself. However, taken together, the combined biases can reach magnitudes that clearly preclude the legitimate use of unadjusted COMTRADE data for trade projections and negotiations. For example, recent USITC statistics report FTZ imports of $285 billion that were exempt from tariffs. To put this value in perspective, these imports were only slightly lower than the total combined customs value of U.S. imports from Japan, Germany, and the United Kingdom. In addition, transport and insurance charges on all U.S. imports were $65.8 billion—a figure that reflects the further overall COMTRADE bias associated with tabulation of cost-insurance-freight trade values.

Third, statutory regulations exempt certain imports from tariffs. These include all U.S. government imports, imports for the treatment of specific medical problems, and imports by overseas territories—all of which were about $24 billion. Altogether, these combined biases totaled $366 billion, or almost 20 percent of the customs value of all United States imports. However, as previously noted, the relative importance of the general trade bias may be larger in other countries (particularly those in East Asia) where international production sharing is practiced more extensively than in the United States. These biases would normally be incorporated in COMTRADE data given the U.N. recommendation that trade statistics be tabulated using general reporting practices:

“The general trade system provides a more comprehensive recording of the external trade flows than does the special system. It is recommended, therefore, that countries use the general system for compilation of their international merchandise trade statistics” (U.N., Department of Economic and Social Affairs (ESA/STAT/AC.137.5, p. 30).

The key point that follows is that analyses of tariffs and other trade barriers should ideally utilize tariff line level import statistics compiled on the same valuation base employed for assessing these duties. Furthermore, the data must accurately account for specific exemptions like those normally afforded foreign trade zones or government entities, as well as for country specific exemptions associated with preferences or the withholding of “most-favored-nation” trade status. Brenton and others (2009) presents a useful illustration of the nature of these required adjustments within the context of formulating national structural adjustment policies. The United Nations (2007) provides comprehensive information on the trade compilation practices of about 40 countries, which should also be useful for identifying required adjustments. As this study shows, a failure to properly account for these factors may adversely influence a country's strategies in multilateral negotiations or the formulation of national trade policies.

References

Notes