Why do IMF Direction of Trade (DOT) figures differ from national statistics? Because they are continuously supplemented with estimates from IMF staff.
As explained in detail by the IMF in its own documentation:
Estimation procedures
The estimation procedures are described in A Guide to Direction of Trade Statistics (1993), available on the CD-ROM and from the Fund’s Publications Section (http://www.imf.org/external/pubs/cat/longres.cfm?sk=154.0).
Reported data, including total imports and exports reported for publication in the Fund’s International Financial Statistics (IFS), are the basis of all estimates in DOTS. The entire DOTS database is continuously supplemented with estimates.
Briefly, the estimation procedures are as follows:
Estimation occurs if a reporting country does not report trade with its partners for a specific period. Data are estimated for all partners and not for some of the partners. In the absence of some or all of the monthly DOTS, quarterly or annual reported DOTS are used. If quarterly data are available, then they are distributed over the relevant months using (1) the available monthly DOTS reported for other partner countries, (2) IFS monthly totals, or (3) an even distribution.
If only annual data are available, then estimation procedures include, in addition to the techniques described above, DOTS reported from the most recent annual report or extrapolations of the most recent data or estimates. Where possible, estimates are computed first for months, and then annual totals are obtained by summation.
If data on total trade are reported for publication in IFS but DOTS data are not reported, then the sum of the direction of trade estimates is constrained to coincide with the IFS totals. In such cases, the data for total exports and total imports shown in the lines IFS World Total and DOTS World Total will be identical. If IFS data are not available, then extrapolation will determine the estimated value of trade, and IFS data will not appear on the country page.
When information is not reported and is inadequate to support the estimation techniques, the data are extrapolated using a matrix of trade among broad country groups. Projected trade growth by these country groups, consistent with trade growth estimates used in the Fund’s World Economic Outlook, is combined with the available DOTS to derive extrapolation factors.
Estimates are not provided for trade flows between countries where data are unavailable for both trading partners during the latest ten years or more.
In summation, only a small portion of world trade is omitted from the DOTS; this portion comprises (1) a small amount of unreported trade among the developing countries; (2) a small amount of unreported trade between developing countries and the countries comprising the group “other countries n.i.e.”; and (3) the trade among “other countries n.i.e.”.
No estimates, however, are compiled for periods prior to 1981 or based on benchmark data referring to 1980 or earlier.
Moreover, data reported from countries’ own records, even with longer delays, continually broaden the base of reported data, thereby replacing previously estimated figures.
Area and world totals are compiled from reported data and these estimates. For a given year the percentage of world trade that is estimated declines over time, as data reported by countries replace the estimates (see Table 1, from DOTSY).
When partner data are used directly or indirectly to derive estimates, the data first are adjusted by a c.i.f./f.o.b. factor of 1.10 to allow for the cost of freight and insurance. Reported imports c.i.f. are divided by the c.i.f./f.o.b. factor to give partner country estimates of exports f.o.b. Similarly, reported exports f.o.b. are multiplied by the c.i.f./f.o.b. factor to give partner country imports c.i.f. For example, if country B has not reported data from its own records but country A has done so, then A’s data for imports from B (reported c.i.f.) are divided by 1.10 to give the f.o.b. value of B’s exports to A. Conversely, A’s data for exports to B (reported f.o.b.) are multiplied by 1.10 to give the c.i.f. value of B’s imports from A. Given the absence of timely data on cost, insurance, and freight, the 10 percent c.i.f./f.o.b. factor represents a simplified estimate of these costs, which vary widely across countries and transactions.