The mechanism takes one line to state. Over-invoice an export and value moves to the exporter; under-invoice an import and value stays with the importer. The goods exist, the quantity is honest, and the declared price carries the whole transfer.
What is not simple is proving it. “The price looks high” is not a finding, and a reviewer who escalates on that alone loses the argument the first time someone asks what it was compared against. Nearly every price-variance finding that collapses under challenge does so for one of five reasons, and none of them are about the price.
The incoterm trap
This is the most common way to manufacture a variance out of nothing.
An incoterm allocates cost and risk between buyer and seller, and in doing so it determines what the invoiced figure includes. Under CIF the seller’s price covers cost, insurance and freight to the named destination port. Under FOB it covers the goods placed on board at origin and nothing after that.
Compare a CIF invoice against an FOB benchmark and the invoice appears inflated by roughly the freight and insurance content of that lane. On a long container route carrying something bulky and low in value density, that gap alone reaches double digits as a percentage. The trade is honest. The comparison was not.
The check is mechanical: read the incoterm off the invoice, establish the basis of the benchmark, and normalise one to the other before computing anything at all. Where the benchmark’s basis cannot be established, the variance is unknown, and reporting it as unknown is more useful than reporting a number.
Aggregation level
The Harmonized System is agreed internationally to six digits, with further digits added at national level. Benchmarks often get built at four, because that is where public data is plentiful.
Four digits is frequently too coarse to carry meaning. A single heading can span goods differing in unit value by an order of magnitude according to grade, specification or generation. Averaging across them yields a median that no individual trade should be expected to sit near, and a variance measured against it is arithmetic without content.
The corollary is unwelcome for anyone who wants broad coverage: for a good many headings there is no benchmark worth quoting, and saying so plainly beats quoting one anyway.
Quantity and contract effects
Unit prices fall with volume, and they fall on a different curve for spot purchases than for draw-downs against a supply agreement. A 200-unit order priced above a benchmark assembled from 20,000-unit shipments is not evidence of anything.
The counterparty relationship works the same way. A first transaction between unfamiliar parties, a distressed sale, an urgent replacement shipment and a long-standing annual contract price differently for reasons unconnected to laundering. A benchmark blind to those distinctions generates variances at a rate that teaches reviewers to dismiss them, which leaves the institution worse off than having no benchmark at all.
Period
Component and commodity prices move, sometimes violently, sometimes on a cycle measured in weeks. A benchmark drawn across twelve months and applied to a shipment at the end of a run-up will misprice it badly.
The window has to be short enough to track the market and long enough to hold comparable trades. Where those two requirements genuinely conflict, the honest output is a range rather than a point.
Mirror statistics measure something else
Aggregate trade-gap analysis, the method behind Global Financial Integrity’s work on trade misinvoicing, compares what two partner countries report about the same flow and treats the disagreement as a value gap.
It is a legitimate macro measurement and a poor case-level one. Reported gaps arise from misinvoicing, and also from valuation convention, since import figures are commonly recorded CIF and export figures FOB. They arise from timing across period boundaries, from transshipment attributing goods to the wrong partner, from re-exports, and from ordinary statistical error. A corridor-level gap tells you where to spend attention. It says nothing about whether one particular shipment inside that corridor was mispriced, and cited as though it did, it will not survive scrutiny.
What a defensible finding looks like
None of the above makes price analysis impossible. The point is to produce a variance the reviewer can stand behind, which means recording, alongside the number, what produced it:
- the HS heading, at the finest level available on both sides of the comparison;
- the incoterm on the invoice and the basis of the benchmark, normalised to each other;
- the quantity band and the contract type;
- the period, and how many comparable trades fell inside it;
- the direction, because over- and under-invoicing implicate different parties.
A variance reported with its basis attached invites a specific argument, and the argument resolves either way: the basis is wrong, in which case the benchmark improves, or it holds, in which case the finding hardens. A variance reported as a score invites dismissal, and usually gets it.
Price alone is a weak signal
A benchmark, however carefully assembled, establishes a question rather than an answer. FinCEN’s 2010 advisory is explicit that no single activity by itself clearly indicates trade-based money laundering, and that indicators have to be read against the activity expected for that customer.
What turns a price question into a finding is the company it keeps. The advisory’s own list is the place to look: a letter of credit amended without reasonable justification, a customer who cannot produce the invoices supporting the transaction, significant discrepancies between the goods descriptions on the bill of lading, invoice, certificate of origin and packing list, payment made by an intermediary apparently unrelated to the buyer or seller, and no apparent business relationship between the originator and the beneficiary.
A price 38% above a defensible benchmark is, by itself, a conversation with the front office. The same price on a shipment whose transport document records fewer units than the invoice, paid by a remitter with no role in the contract, is a case. The difference lies not in the price analysis but in whether anyone opened the other records.