International money laundering and international illicit financial flows (IFFs) are major problems for poor and rich countries alike, but especially the poor. Both are strongly rooted in fraudulent treatment of imports and exports as they cross international borders. This includes smuggling, misclassification of products to change their valuation, and straight mis-invoicing, i.e. deliberately declaring erroneous product values to make it possible to shift money from one country to another secretly and illegally. Spanjers and Salomon (2017) estimate that 87% of the IFFs in the developing countries over the period 2005-2014 resulted from trade mis-invoicing. Detecting undervaluation and misclassification is critical to the fight against tax evasion and IFFs.

How can mirror statistics help detect fraudulent activity?

It is possible to detect these illicit activities by comparing figures on trade flows between pairs of countries. In a perfect world, the details and values of all recorded exports from Albania to Zambia should match the details and values of all recorded imports into Zambia from Albania. These are what are termed mirror statistics. In a perfect world, the reflections would match perfectly. In reality, they never do. Investigating how and why they fail to match is in principle a very good way of helping customs administrations to find out what is going on, and to improve their risk analysis and fraud detection.

Our recent paper in the World Customs Journal explores how this tool has been used, and could be improved. Trade economists first used mirror statistics in the 1960s (see Bhagwati 1964). It is usually assumed that import figures would be more reliable. These are the basis on which import duties and taxes are calculated, so generally receive more official scrutiny. But some countries do not communicate their international trade data at all. It has been possible to build up estimates of their trade patterns using mirror statistics for other countries (Anderson & Van Wincoop, 2003; Carrère, 2006). There is also the problem that import statistics are generally reported on a CIF basis (i.e. inclusive of insurance and freight costs) and export statistics as FOB (i.e. Free On Board, excluding insurance and freight costs). Trade economists have long been using mirror analysis to improve estimates of international transport costs.

What have we learned from the research?

Discrepancies are beyond pure transportation and logistic costs

Statistics discrepancies – the failure of the mirror images to reflect each other perfectly – do not in themselves indicate mis-invloicing or other forms of fraud. There are routine reasons for the failure to match perfectly. An export from Albania for the 2019 calendar or financial year might arrive in Zambia only in the 2020 financial year. Or a consignment of copper originally sent from Zambia to Albania might be legitimately re-sold en route and diverted to Belgium. However, the discrepancies are often so big that we can be fairly sure that something else is going on – and that it is probably illicit.

Hummels and Lugovskyy (2006) examined the trade statistics for 17,790 pairs of countries. They found that in almost 50% of cases the discrepancies were clearly out of any reasonable bounds: i.e. where the declared figures implied that the international transport costs were either negative or more than 100% of the product value at export. Similar results were reported by Raballand et al. (2012). Ayadi et al. (2014) demonstrated that the values of Tunisian imports from Libya and Algeria were on average underreported by a factor of around 3, and information they collected at borders confirmed their findings. Studies have also been conducted by Bensassi et al (2017) in Mali and Chalendard et al. (2017) in Madagascar. Two findings emerge from virtually all studies:

  1. The higher the tariff levels and more complex the tariff structure, the more likely mirror discrepancies are to emerge.
  2. Across countries, large statistical gaps are generally limited to less than 10 products (out of hundreds/thousands of traded goods) but for the same kind of products (usually due to the high taxes at stake), such as food products (vegetable oil, sugar, rice), clothes and footwear, manufactured products (motorcycles, phones), construction materials (cement), fuel, low-selling price per unit products (like fertilisers, due to risk of mis-classification), or products exempted from duties/VAT.

Discrepancies are correlated with governance issues

Research conducted over the past two decades has demonstrated that large valuation discrepancies identified through mirror analysis tend to correlate with various indicators of poor governance, including tax evasion and collusion. Using a gravity equation model and a worldwide panel of very detailed trade data, with 3.5 million observations, Carrere and Grigoriou (2014) found that discrepancies were consistently higher when tariffs – and thus the incentive to evade them – were higher; when there was more direct foreign investment (and thus more incentives to shift profits illicitly); and when there was more reported corruption. Rijkers et al. (2017) demonstrate from Tunisian data that the size of discrepancies is larger when the importer is close to political power.

Therefore, in countries with weak governance, relatively high tariffs, and complex tariff structures, it is highly likely to find statistical discrepancies of the main items imported using mirror analysis.

Leveraging mirror statistics for more effective customs administration

Mirror analysis is relatively easy. It does not require expansive investments or equipment, and customs officers are already used to working with statistical or database management software (see Cantens (2015) for a step-by-step guide). Mirror analysis has been used for customs reform, including as a part of technical assistance to customs administrations, for example in Cameroon and Madagascar. While mirror analysis does not confirm misinvoicing and fraud, it provides a basis for suspecting it. As such, mirror analysis should be supplemented with other types of information, and supported by the routine work of customs data, intelligence, and investigation services.

It is also important to note that mirror analysis cannot be performed on a real-time basis because there are long lags in the reporting of trade data. Automatic exchange of information between national customs organisations, with support from the World Customs Organisation, would reduce these lags and make routine mirror analysis more effective as a means of controlling fraud. Similarly, it is more powerful in concert with other customs reforms, such as increased recording and monitoring of the activities of individual customs inspectors. Therefore, while mirror statistics and analysis is a very useful tool, it must be complemented by other measures for more effective customs and tax administration.

Gaël Raballand

Gaël Raballand is a lead public sector specialist in the World Bank. He holds a PhD in economics and a degree in political science and international public law. He co-authored four World Bank books on transport and trade in Africa and has also been involved in several customs reforms in Sub-Saharan Africa.

Christopher Grigoriou

Christopher Grigoriou joined the CERDI, University of Auvergne, in 2007 as an Associate Professor in Economics and Econometrics. Currently on leave, he is working in Geneva as consultant, for international institutions including the IMF, World Bank and UNCTAD, on international trade and customs and tax administration matters. He is also an invited professor at the Swiss Graduate School of Public Administration (IDHEAP, Lausanne University), where he teaches econometric modelling in the field of public finance. He holds a PhD in economics from CERDI (2006). He is currently conducting research on international trade (including electronic trade), illicit cross-border trade measurements, and risk management for government services.