In 2017, the Republic of Georgia sent $272 million in exports to its neighbor, Azerbaijan. The same year, Azerbaijan reported receiving $74 million—that’s not a typo—in imports from Georgia. Goods worth $198 million seemingly disappeared before they reached Azerbaijani customs. The gap is a big deal. Azerbaijan taxes imports just above 5% on average (weighted for trade), which means its treasury missed out on collecting roughly $10 million in tariffs—0.1% of all government spending in that year—from just a single trading partner.

Many factors could explain the gap (see, for example, here, here, and here). Shippers might have rerouted goods to other destinations, the two countries’ customs offices might value goods differently, or the customs offices could have erred in reporting results or converting them to dollars. But one reason Azerbaijan’s reported imports are so low—not only here, but systemically across trade partners and years—is corruption and associated tariff evasion. Many traders likely undervalue and/or underreport their imports when going through Azerbaijani customs, and the sheer magnitude of the trade gap suggests the complicity or collusion of the authorities. The corruption involved might be petty (e.g., an importer bribing a customs officer to look the other way, or a customs officer pocketing the tax and leaving it off the books) or grand (e.g., a politician with a side business using her influence to shield imports from inspection; see here). A similar dynamic might also be at work in exporting countries: companies may undervalue exports to limit their income tax liability, possibly paying bribes to avoid audits.

Though Azerbaijan may be an extreme case, it is not unique. Economists have examined these export gaps (sometimes called “mirror statistics”) and have found similar discrepancies in, for example, Hong Kong’s exports to China, China’s exports to the United States, and Cambodia’s imports from all trading partners. Most recently, economists Derek Kellenberg and Arik Levinson compared trade data across almost all countries over an eleven-year time period, finding that “corruption plays an important role in the degree of misreports for both importers and exporters.” For lower-income countries, Professors Kellenberg and Levinson showed a positive relationship between a country’s level of perceived corruption, as measured by Transparency International’s Corruption Perceptions Index (CPI), and its underreporting of imports. The authors also showed a strong positive relationship between perceived corruption and the underreporting of exports across all countries.

Mirror statistics are an imperfect measure of customs corruption, to be sure, but they can serve two useful purposes in fighting this sort of corruption, and anticorruption reformers should pay more attention to this type of data.

• First, and most obviously, mirror statistics are a useful diagnostic tool for identifying countries where customs corruption is likely to be a significant problem, as indicated by unusually large discrepancies between reported exports to a given country and reported imports received by that country. Mirror statistics also make it possible for anticorruption advocates to quantify the costs of corruption by presenting in stark numerical terms the value of uncollected tariff revenue. Moreover, with sufficiently granular data, mirror statistics can even pinpoint specific trade relationships, ports of entry, and categories of goods that are at a higher risk. For instance, since we know that Georgia-Azerbaijan trade exhibits an unusually high trade gap, we can drill down to specific products to try to determine which are most prone to underreporting—perhaps the $8 million of live cattle, or the similar value of potatoes, that Azerbaijan reports entering from Georgia? Product-level analysis can also reveal if other explanations are at play. In addition, mirror statistics can help government officials work with their counterparts in partner countries to identify particular trading firms that are underreporting goods and possibly paying bribes.
• Second, mirror statistics can be used to assess the effectiveness of anticorruption measures. When a customs ministry punishes an employee for collecting bribes or implements a reform such as a ban on cash payments, an external audit, or outsourcing customs collection altogether—three measures proposed on this blog for Haitian Customs—the agency can compare trade gaps from before and after implementation to gauge the effect of the measure. Even better, by implementing a pilot reform at some customs points and not others and then tracking changes in trade gaps over time, a customs ministry could use the difference-in-difference method to control for exogenous changes in trade gaps over time, providing a more robust sense of the measure’s effect. In addition to domestic policymakers, foreign governments and international organizations that provide anticorruption assistance could incorporate mirror statistics into their monitoring and evaluation programs.

Mirror statistics are attractive in part because they are an objective measure, making them less vulnerable to the behavioral biases that distort subjective scores. That said, mirror statistics are a proxy, not a direct measure of customs corruption. As with any proxy, there’s a lot of noise in the data. They capture factors other than corruption, exclude certain types of corruption, and may have a limited useful life as corrupt actors learn to circumvent them over time. For these and other reasons, the trade gaps revealed by mirror statistics may not be a suitable alternative measure of national-level corruption as a general matter. Nonetheless, as long as we remain mindful of what kinds of corruption trade gaps can and cannot measure, they can be very useful, providing those doing anticorruption work with a valuable tool for measuring tariff evasion, evaluating policies, assessing risks, and highlighting the costs of customs corruption. They might start with the $198 million in imports from Georgia that Azerbaijan seemingly never taxed.