There are dozens of official sources of data on international trade. We write this post because, if you compare these different sources, you will find that they do not agree with one another. Even if you focus on what seems to be the same indicator for the same year in the same country, discrepancies are large.
For example, for China in 2010, the estimated total value of goods exports was $1.48 trillion according to World Bank Data, but it was $1.58 trillion according to WTO Data. That’s a difference of about 7%, or a hundred billion US dollars.
Such differences between sources can also be found for rich countries where statistical agencies tend to follow international reporting guidelines more closely. In Italy, for example, Eurostat figures of the value of exported goods in 2015 are 10% higher than the merchandise trade figures published by the OECD.
And there are also large bilateral discrepancies within sources. According to IMF data, for example, the value of goods that Canada reports exporting to the US is almost $20 billion more that the value of goods that the US reports importing from Canada.
In this blog post we explain how international trade data is collected and processed, and why there are such large discrepancies.
What data is available?
The data hubs from several large international organizations publish and maintain extensive cross-country datasets on international trade. Here’s a list of the most important ones:
- World Bank Open Data
- IMF Data
- WTO Statistics
- UN Comtrade
- UNCTAD World Integrated Trade Solutions
- Eurostat
- OECD.Stat
In addition to these sources, there are also many other academic projects that publish data on international trade. These projects tend to rely on data from one or more of the sources above; and they typically process and merge series in order to improve coverage and consistency. Three important sources are:
How large are discrepancies between sources?
In the visualization below we provide a comparison of the data published by several of the sources listed above, country by country, since 1955 up until today.
For each country, we exclude trade in services, and we focus only on estimates of the total value of exported goods, expressed as shares of GDP.4
As we can clearly see in this chart, different data sources tell often very different stories. And this is true, to varying degrees, across all countries and years. You can use the option labeled ‘change country’, at the bottom of the chart, to focus on any country.
Constructing this chart was demanding. It required downloading trade data from many different sources, collecting the relevant series, and then standardising them so that the units of measure and the geographical territories were consistent.
All series, except the two long-run series from CEPII and NBER-UN, were produced from data published by the sources in current US dollars, and then converted to GDP shares using a unique source (World Bank).5
So, if all series are in the same units (share of national GDP), and they all measure the same thing (value of goods exported from one country to the rest of the world), what explains the differences?
Let’s dig deeper to understand what’s going on.
Why doesn’t the data add up?
Broadly speaking, there are two main approaches used to estimate international merchandise trade:
- The first approach relies on estimating trade from customs records, often complementing or correcting figures with data from enterprise surveys and administrative records associated with taxation. The main manual providing guidelines for this approach is the International Merchandise Trade Statistics Manual (IMTS).
- The second approach relies on estimating trade from macroeconomic data, typically National Accounts. The main manual providing guidelines for this approach is the Balance of Payments and International Investment Position Manual (BPM6), which was drafted in parallel with the 2008 System of National Accounts of the United Nations (SNA 2008). The idea behind this approach is recording changes in economic ownership.6
Under these two approaches, it is common to distinguish between ‘traded merchandise’ and ‘traded goods’. The distinction is often made because goods simply being transported through a country (i.e. goods in transit) are not considered to change the stock of material resources of a country, and are hence often excluded from the more narrow concept of ‘merchandise trade’.
Also, adding to the complexity, countries often rely on measurement protocols that are developed alongside these approaches and concepts that are not perfectly compatible to begin with. In Europe, for example, countries use the ‘Compilers guide on European statistics on international trade in goods’.
Even when two sources rely on the same broad accounting approach, discrepancies arise because countries fail to adhere perfectly to the protocols.
In theory, for example, the exports of country A to country B should mirror the imports of country B from country A. But in practice this is rarely the case because of differences in valuation. According to the BPM6, imports and exports should be recorded in the balance of payments accounts on a ‘free on board (FOB) basis’, which means using prices that include all charges up to placing the goods on board a ship at the port of departure. Yet many countries stick to FOB values only for exports, and use CIF values for imports (CIF stands for ‘Cost, Insurance and Freight’, and includes the costs of transportation).7
The chart below gives you an idea of how large import-export asymmetries are. Shown are the differences between the value of goods that each country reports exporting to the US, and the value of goods that the US reports importing from the same countries. For example, for China, the figure in the chart corresponds to the “Value of merchandise imports in the US from China” minus “Value of merchandise exports from China to the US”.
The differences in the chart below, which are both positive and negative, suggest that there is more going on than differences in FOB vs CIF values. If all asymmetries were coming from CIF-FOB differences, then we should only see positive values in the chart (recall that, unlike FOB values, CIF values include the cost of transportation, so CIF values are larger).
What else is going on here?
Another common source of measurement error relates to the inconsistent attribution of trade partners. An example is failure to follow the guidelines on how to treat goods passing through intermediary countries for processing or merchanting purposes. As global production chains become more complex, countries find it increasingly difficult to unambiguously establish the origin and final destination of merchandise, even when rules are established in the manuals.8
And there are still more potential sources of discrepancies. For example differences in customs and tax regimes, and differences between “general” and “special” trade systems (i.e. differences between statistical territories and actual country borders, which do not often coincide because of things like ‘custom free zones’).9
Even when two sources have identical trade estimates, inconsistencies in published data can arise from differences in exchange rates. If a dataset reports cross-country trade data in US dollars, estimates will vary depending on the exchange rates used. Different exchange rates will lead to conflicting estimates, even if figures in local currency units are consistent.
Wrapping up
Asymmetries in international trade statistics are large and they arise for a variety of reasons. These include conceptual inconsistencies across measurement standards, as well as inconsistencies in the way countries apply agreed protocols. Here’s a checklist of issues to keep in mind when comparing sources.
- Differences in underlying records: is trade measured from National Accounts data rather than directly from custom or tax records?
- Differences in import and export valuations: are transactions valued at FOB or CIF prices?
- Inconsistent attribution of trade partners: how is the origin and final destination of merchandise established?
- Difference between ‘goods’ and ‘merchandise’: how are re-importing, re-exporting, and intermediary merchanting transactions recorded?
- Exchange rates: how are values converted from local currency units to the currency that allows international comparisons (most often the US-$)?
- Differences between ‘general’ and ‘special’ trade system: how is trade recorded for custom-free zones?
- Other issues: Time of recording, confidentiality policies, product classification, deliberate misinvoicing for illicit purposes.
These factors have long been recognized by many organizations producing trade data. Indeed, international organizations often incorporate corrections, in an attempt to improve data quality along these lines.
The OECD’s Balanced International Merchandise Trade Statistics, for example, uses its own approach to correct and reconcile international merchandise trade statistics.10
The corrections applied in the OECD’s ‘balanced’ series make this the best source for cross-country comparisons. However, this dataset has low coverage across countries, and it only goes back to 2011. This is an important obstacle, since the complex adjustments introduced by the OECD imply we can’t easily improve coverage by appending data from other sources. At Our World in Data we have chosen to rely on CEPII as the main source for exploring long-run changes in international trade; but we also rely on World Bank and OECD data for up-to-date cross-country comparisons.
There are two key lessons from all of this. The first lesson is that, for most users of trade data out there, there is no obvious way of choosing between sources. And the second lesson is that, because of statistical glitches, researchers and policymakers should always take analysis of trade data with a pinch of salt. For example, in a recent high-profile report, researchers attributed mismatches in bilateral trade data to illicit financial flows through trade misinvoicing (or trade-based money laundering). As we show here, this interpretation of the data is not appropriate, since mismatches in the data can, and often do arise from measurement inconsistencies rather than malfeasance.11
Hopefully the discussion and checklist above can help researchers better interpret and choose between conflicting data sources.