The Geographic Structure of Cuban Imports and the Incidence of Over-Invoicing.

Ernesto Hernandez-Cata

This note examines the relation between Cuban imports of goods from major trading partners as reported by Cuba’s statistical agency, and exports of goods to Cuba as reported by her partners. At first sight, these numbers should be equal—two faces of the same coin. In fact, however, these numbers differ for various reasons, including transportation and insurance costs and in some cases over invoicing.

The examination is based on the following simple equation.

(1) mk = αk + (ßk + δk) xk

where:

mk is the value of Cuban imports from country k, as reported by Cuba

xk is the value of export to Cuba reported by country k,

ßk is the cost and insurance factor, expressed as a proportion of exports.

δk is the proportion of over invoicing,

αk is a constant term, and the sub-index k indicates the partner country

By international convention, imports are measured on a cost insurance freight (cif) basis and exports on a free on board (fob) or free alongside ship (fas) basis. The parameter ß is generally estimated at 0.1 (10% of the value of trade). Therefore the parameter (ß+δ) is expected to be equal to 1.1, or larger if over-invoicing is present (δ>0). There is no strong reason to believe there should be a constant difference between mk and xk, and therefore we should expect the constant term α to be zero.

The analysis was based on a sample of Cuba’s 14 major trading partners selected as those whose exports to Cuba reached $100 million or more in 2014 or 2013 (see Fig. 1). Data for Cuban imports came from the Oficina Nacional de Estadisticas e Informacion (ONEI), and data for partner-reported exports from the United Nations’ Com Trade database.

Regressions based on equation (1) were estimated for each of Cuba’s 14 top trading partners using data from 2005 to 2014 and the main results are shown in Table 1. Estimates in the first two columns are based strictly on equation (1) in that all parameters, including the constant terms αk’s, are unconstrained and estimated by ordinary east squares. As expected, the estimated slope coefficient (ß + δ) is equal or larger than 1.1 in most equations. Thus, assuming a cif/fob factor ß of 10%, imports from these counties appear to be significantly affected by over-invoicing. However, (ß+δ) is significantly smaller than 1.0 for Algeria, Brazil Canada, the Netherlands and Russia, which is contrary to our expectations.

The estimated constant term α is insignificantly different from zero for all countries. As noted earlier, there is no reason to believe that there is a constant difference between the m’s ad the x’s. Accordingly, the equations were re-estimated by constraining the constant term to be zero. The new estimates of the slope coefficients are equal or larger than 1.1 for all counties except Algeria and the Netherlands, were they are significantly lower than 1, which is theoretically unacceptable and suggests the presence of more fundamental measurement problems. In general the slope coefficients in the constrained equations are equal or larger than those in the unconstrained equations, and their standard errors are much lower.

The estimates for the constrained equations in Table 1 indicate the presence of over-invoicing for most of Cuba’s major trading partners. The degree of over-invoicing ranges from 0-8% of the value of import for China, Brazil, and Germany; and 10-17% for Mexico, Italy, the United States, Argentina, France and Ukraine. All the δ coefficients for those countries are significantly larger than zero based on a one-tailed There is no evidence of significant over-invoicing for Canada, and Spain. In value terms over-invoicing is estimated to have ranged between 7% (in Germany) and 17% (in Argentina). The estimated total cost to the nation in 2014 would have been be around $330 million.

Over-invoicing also affect the interpretation of the balance of payments estimates. The presence of recorded but inexistent imports implies that the true current account surplus is underestimated, which in turn implies positive net errors and omissions.