A transfer with no payment in it
Almost everything in anti-money laundering assumes that moving money leaves a payment behind. Monitoring reads payments. Screening screens payments. Correspondent banking is about whose payments travel through whom.
Mirror trading moves money between countries without a cross-border payment at all. Nothing is remitted. No message crosses a border. Instead, two trades are arranged: one customer buys a security in one country and pays in local currency, and at the same moment a connected customer sells the identical security in another country and is paid in hard currency. The securities offset. The money does not: it has changed country, currency and owner.
Why this is worth a course of its own. It defeats the normal apparatus at every point:
- There is no payment to monitor, so transaction monitoring built on payment patterns sees nothing to pattern.
- Each leg, examined alone, is an ordinary trade at an honest price, executed in a liquid market.
- The two legs usually sit in different legal entities, different systems, different countries and different teams, so the only place the pair is visible is a view that nobody has.
- The customers are different people, so a control that asks "is this customer behaving oddly?" answers no, twice.
And it is not a historical curiosity. The arrangement needs only a liquid instrument, two accounts in two places and someone willing to arrange both sides. That describes a large part of the world's capital markets infrastructure, and the same logic works with foreign exchange, derivatives, repurchase agreements and commodities, which is lesson four.
The uncomfortable part. Matched, offsetting trades between related parties are also completely normal. Market makers hedge. Groups move positions between entities. Funds transfer holdings between share classes. A control that flags every offsetting pair will drown, and a control that flags none is blind. Separating the two is the skill this course is about, and it is why mirror trading is rarely taught: the answer is not a red flag but a piece of analysis.
Where the detail comes from. The case this course is built on is the FCA's Final Notice to Deutsche Bank AG of 30 January 2017. It is a public document of forty-four pages and it is unusually precise about the mechanics, which is why it is used here rather than any secondary account.
The case, exactly as the regulator found it
On 30 January 2017 the Financial Conduct Authority fined Deutsche Bank AG £163,076,224, made up of disgorgement of £9,076,224 and a penal element of £154,000,000, for breaches of Principle 3, which concerns management and control, and of the senior management arrangements, systems and controls rules SYSC 6.1.1R and 6.3.1R, over the period 1 January 2012 to 31 December 2015. The bank settled early and received the corresponding discount.
The arrangement. There were more than 2,400 mirror trades between April 2012 and October 2014, each of which involved two legs:
- the Moscow Side: a Russian customer of DB Moscow bought highly liquid Russian securities from DB Moscow, paying in Roubles; and
- the London Side: at the same time, a non-Russian customer of Deutsche Bank sold the same number of the same securities to Deutsche Bank in exchange for US Dollars.
The customers on the two sides were connected to each other. The orders for both legs were received by DB Moscow's front office, and that office executed both sides at the same time.
The result. The trades were used to transfer more than USD 6 billion out of Russia, through Deutsche Bank in the United Kingdom, to overseas bank accounts including in Cyprus, Estonia and Latvia.
Two mechanics worth naming. The onboarding of the non-Russian customers was initiated and facilitated by the Moscow front office — the same desk that took both sides of the trades. And the London Side was executed by that Moscow desk on Deutsche Bank's behalf through remote booking, a process by which Deutsche Bank offices in certain locations, including Moscow, could book trades directly into Deutsche Bank's trading books in the United Kingdom.
Why it ran for two and a half years. The Notice attributes it to widespread deficiencies in the bank's anti-money-laundering control framework, particularly in establishing new customer relationships and in the ongoing monitoring of transactions.
And the finding most people get wrong. The Authority found that the failings were not committed deliberately or recklessly, and that there was no evidence that senior management at Deutsche Bank, or any Deutsche Bank employee in the United Kingdom, was aware of or involved in the suspicious trading. This is a control failure story, not a conspiracy story, and that is what makes it instructive: it can happen in a bank where nobody in the relevant office knows.
The economics, and why it is cheap
To see why anyone would do this, look at the money rather than the compliance.
The pair is close to risk-free. Buying a quantity of a security in one place and selling the identical quantity in another, at the same moment, leaves no net market position. Whatever the price does next is irrelevant, because the two legs cancel. That is the opposite of speculation: the arranger is not taking a view, and does not need to be right about anything.
So the only cost is friction. Commissions, the bid-offer spread on both legs, the currency conversion and the custody and settlement charges. Add them up and you have the price of moving the money, expressed as a percentage. For the customer the question is simply whether that percentage is lower than the alternatives, and compared with the cost of physically moving cash, or of layering through a chain of shell companies that each needs maintaining, it generally is.
Why liquid instruments are chosen. A liquid security can absorb large quantities without moving the price, can be executed immediately, and has a price everyone can verify. That last point matters more than it first appears: because the price is honest, nothing in the trade looks like market abuse. A surveillance system hunting for mispricing, front-running or manipulation finds nothing, because none of those things is happening.
The trade is real. This is not a sham or a fictitious transaction. The securities exist, the trades settle, the title passes, and an auditor examining either leg finds a properly documented transaction at a fair market price. There is nothing to unpick in the paperwork, because the paperwork is correct.
Which is why the signal is never in one trade. It is in the relationship between two trades: identical instrument, identical quantity, opposite direction, near-simultaneous, different currencies, different jurisdictions, and counterparties connected to each other. Remove any one of those and the pair stops working as a transfer; keep them all and the pair is the transfer.
A note on who profits. The intermediary earns the spread and the commission on both legs, honestly, and the volume is enormous. That is the quiet reason this kind of business survives internal challenge: it is not a loss-making favour to a difficult client but a profitable flow for the desk that arranges it, which is exactly the circumstance in which a control needs to come from outside the desk.
The same trick in other instruments
Mirror trading is a name for one instance of a general method, and recognising the general method matters more than remembering the Russian example.
The invariant. Two legs, in two jurisdictions or two entities, whose market risk offsets, between parties connected to each other, with the currencies or the places deliberately different. Wherever those conditions can be arranged, value can be moved without a cross-border payment.
Matched foreign exchange. One party buys a currency pair and a connected party sells the same amount of the same pair at the same rate elsewhere. The net position is nil and the cash has changed country.
Back-to-back lending. A customer lends in one country while a connected customer borrows the same amount in another, each against collateral. No money crosses the border, yet each party's balance sheet has what it wanted. The loans may be entirely genuine, which is what makes them hard to argue with.
Total return swaps and other derivatives. Offsetting derivative positions achieve the same transfer with smaller initial cash flows and considerably more complexity, which is to say considerably less chance of anyone reconstructing the pair.
Repurchase agreements. Securities sold with an agreement to buy them back, mirrored elsewhere, which adds the appearance of secured financing to the arrangement.
Commodities and physical trade. Matched purchases and sales of a liquid commodity, sometimes with the goods never moving. Here the method meets trade-based laundering, which this catalogue covers separately; the difference is that mis-invoicing lies about value, whereas a mirror pair lies about nothing at all.
Insurance and reinsurance round-trips. Premiums paid in one jurisdiction and recovered through connected reinsurance arrangements in another.
What this means for a control. Do not build a mirror-trade detector. Build the capability the general method requires you to have: the ability to see both legs of related activity across entities and jurisdictions, and to ask whether a customer's activity has any purpose other than the transfer it accomplishes. An institution that can do that for equities and not for foreign exchange has solved one instance of the problem.
And keep the legitimate cases in view. Every structure above has a real use: hedging, financing, tax-neutral group reorganisation, genuine risk transfer. The question is never whether the instrument can be misused, since all of them can, but whether this customer's use of it has an explanation that survives contact with the facts.
Why ordinary controls do not see it
It is worth being precise about which control failed, because the answer is not that monitoring was badly tuned. Most of the standard apparatus is aimed somewhere else entirely.
Transaction monitoring is built for payments. Scenarios are written about credits, debits, velocity, round amounts, new beneficiaries and corridors. A securities purchase settled in the ordinary way produces none of those patterns. There is no beneficiary, no corridor and no remittance.
Trade surveillance is built for market abuse. It looks for manipulation, insider dealing, spoofing and mispricing. A mirror pair executes at fair prices in a liquid instrument, so the surveillance that exists over the trading floor is looking for a different crime and correctly reports nothing.
Each leg is lawful. Nothing in either trade is irregular. The Final Notice is explicit that the bank failed to obtain sufficient information about its customers' businesses, with the consequence that their trading could not be effectively monitored even if the bank had had the facilities to do so. That sentence is the heart of this lesson: without knowing what a customer is for, no amount of transaction data tells you that a trade is inexplicable.
The pair is split across the organisation. One leg sits with a Moscow entity and its customer; the other with a London book and a different customer. Different systems, different jurisdictions, different supervisors, different teams, different legal entities. The pair exists only in a view that spans them, and in most banks nobody owns that view.
"Not my customer" is the structural excuse. The London desk's customer behaved unremarkably. The Moscow customer was somebody else's relationship. Each side could be reviewed in good faith and found unobjectionable.
And the desk that arranged both sides also brought in the customers. The Notice records that the onboarding of the non-Russian customers was initiated and facilitated by the same Moscow front office that received and executed both legs, and that the London leg reached the UK books by remote booking. When the same team sources the customer, takes both orders and books the result into another country's accounts, the separation that controls rely on has gone, and no individual control has to fail for the whole to fail.
The lesson to take. This is not a tuning problem or a model problem. It is an architecture problem: the information needed to see the activity was never assembled anywhere.
What detection would actually have to look like
If the pair is the signal, the surveillance has to be built on pairs. That is a different query from anything a payment monitoring system runs, and it is specific enough to write down.
The matching conditions. Look across entities and books for two trades where:
- the instrument is identical, matched on an unambiguous identifier rather than on a name;
- the quantity is identical, or close enough that rounding explains the difference;
- the direction is opposite, one buy and one sell;
- the timing is close, from simultaneous to within a day or two, since a delay only adds market risk the arranger does not want;
- the currencies differ, or the settlement places differ, or both;
- and the counterparties are connected, by ownership, by address, by signatory, by introducer, or by the desk that onboarded them.
Then rank, rather than alert. Matched pairs are common and mostly innocent, so the output of the query is a population to be scored, not a queue of suspicions. The factors that move a pair up the ranking are the ones that have no innocent reading: connected counterparties with no group relationship to explain it, a customer whose entire activity is one side of such pairs, repeated pairs with the same partner, a corridor between a currency under pressure and a hard currency, and settlement into a jurisdiction unrelated to either party's business.
The negative test is the strongest one. Ask what else the customer does. A market maker's offsetting trades sit inside a book of thousands of other positions. A group's internal transfers sit inside a visible corporate structure. A customer whose account shows the mirror legs and nothing else has no second explanation available.
Three practical obstacles, to be honest about. Instrument identifiers are inconsistent across systems, so the match often fails on data quality before it fails on logic. Quantities are reported in different units and lots. And the two legs may be days apart in the records of systems whose clocks and booking conventions differ, which turns "simultaneous" into a judgement about timestamps.
Who should run it. Not the desk, and not a payments monitoring team that has no securities data. This sits with a function that can see trade data across legal entities, which in most institutions means it has to be built deliberately rather than found. That is the subject of the last lesson.
The due diligence that makes any of it possible
The Final Notice makes a point that is easy to read past and is the most portable lesson in the case: the bank failed to obtain sufficient information about its customers' businesses, so their trading could not be effectively monitored even if it had had the facilities to do so.
Read that as a statement about order of operations. Surveillance is a comparison between what a customer does and what a customer is expected to do. Without the second half, the comparison cannot be made, and the sophistication of the first half is irrelevant.
So what should be on file for a trading customer?
- What the customer is, in the ordinary sense: the business, its revenue, its markets, and for a fund or a vehicle, who is behind it and whose money it invests.
- Why it is trading with you, which for an institutional customer should be a sentence a person could repeat: hedging a known exposure, allocating a mandate, financing inventory.
- Expected activity, stated in terms the surveillance can use: instruments, approximate volumes, currencies, settlement locations, and expected counterparties.
- Where the securities or the money comes from, which for a securities customer is the analogue of source of funds, and is almost never asked.
- Who introduced the relationship, and whether that person has an interest in the trading.
Then the part that is usually missing: review against it. Expected activity recorded at onboarding and never compared with actual activity is a filing exercise. The comparison need not be clever. A customer onboarded as a corporate treasury hedging foreign exchange, whose entire activity is buying one equity in one currency and selling it in another, has departed from its own description, and that departure is visible to anybody who looks.
A warning about introduced business. Where a relationship arrives through a desk, a broker or an affiliate, the file frequently contains the introducer's assurance in place of the customer's own explanation. In this case the onboarding of the non-Russian customers was initiated and facilitated by the very office that then traded both sides. An introducer's assurance about a customer it profits from is not due diligence; it is a referral.
And the uncomfortable implication. Most institutions' securities customers have thinner files than their retail customers, because institutional clients are presumed sophisticated and are commercially harder to question. That presumption is what the arrangement in this case relied on.
Governance: desks, introductions and remote booking
The structural features that let this run are governance features, and each has a name and a fix.
A front office doing its own onboarding. When the desk that will profit from the flow also sources and sponsors the customer, the first control in the chain is held by the party with the clearest interest in the answer. The fix is not more paperwork but a different owner: onboarding approval outside the revenue line, with the authority to refuse.
Both sides of a trade with one desk. A desk that receives both legs from connected customers and executes them together has, in effect, arranged a transfer and recorded two trades. Where that is a legitimate business, it should be an identified business with its own controls and its own reporting, not an emergent property of how orders happen to arrive.
Remote booking. The Notice describes the London leg being executed by the Moscow office through a process that let offices in certain locations book trades directly into the UK trading books. Remote booking is ordinary and useful; it also separates who executes from where the risk and the record sit, and therefore separates the activity from the people supervising it. Any institution that permits it needs to be able to say, for each remote booking arrangement, which entity's controls apply, who reviews the activity, and which supervisor is responsible.
Supervision across borders. A desk in one country executing into another country's books is supervised by two regimes and belongs comfortably to neither. The gap this creates is not theoretical: it is the specific gap in this case, and it persisted for two and a half years.
And the finding that should discipline how this is discussed. The Authority found the failings were not deliberate or reckless, and found no evidence that senior management or any UK employee knew about the trading. There was no villain in London to find. The arrangement survived because responsibility was distributed across entities thinly enough that no one person held enough of the picture to be alarmed by it.
The governance question to put to any institution. For every material flow, who is the single person who can see the whole of it? If the answer is nobody, the institution's controls depend on an assumption that is not true, and the size of the resulting loss is set by how long it takes someone outside to notice.
Building the capability, and its limits
The honest summary of this topic is that detecting it requires a capability most institutions have not built, and that the capability is specific enough to plan.
What to build, in order of value.
- One view of trade activity across legal entities, with instrument identifiers reconciled. Nothing else on this list works without it, and it is the expensive item.
- Pair-matching surveillance on the conditions in lesson six, producing a ranked population rather than alerts.
- Expected activity on file for every trading customer, and a periodic comparison against it.
- Onboarding approval outside the revenue line, especially for business introduced by a desk.
- A register of remote booking arrangements, each with a named owner, the applicable control framework and a review.
- Connected-party identification that works across entities: common ownership, addresses, signatories, introducers and settlement instructions.
How to measure whether it works. Not by counting mirror trades found, which may well be none. Measure instead: how many trading customers have expected activity recorded; what proportion of trade records can be matched on a clean instrument identifier; how long it takes to answer the question "show me all activity by parties connected to this customer, across the group"; and how many introduced relationships were approved outside the introducing desk.
The limits, stated plainly. Offsetting trades are a normal part of markets, so this will never be a control that blocks. It produces questions, and questions need someone able to ask a sophisticated customer an awkward one, which is a cultural capability rather than a technical one. Data quality will defeat the matching more often than cleverness will defeat the controls. And an arranger who is prepared to accept a day's market risk, or to use two unrelated institutions, degrades the pattern considerably.
Why it is still worth doing. The method moves very large sums at low cost, and the sums in the case were measured in billions over two and a half years at one bank. A control that merely shortens how long such an arrangement can run, from years to months, is worth more than a great deal of alert tuning elsewhere.
And the sentence to carry away. The failure was not that a system missed a pattern. It was that nobody had assembled the information in which the pattern would have been visible, and the regulator's finding that nobody intended any of it is precisely what makes that worth fixing.
Thirty questions
Answer all thirty questions, then check your answers. You need 27 out of 30 to pass and receive a certificate. If you score less, you will see which answers were right and wrong, and then go through the course again before you retake the check. Your answers, progress and times are kept only in this browser.
Your answers
Your certificate of completion
Enter your name as you want it to appear, then save the certificate as a PDF. In the print window, choose Save as PDF. A certificate is issued once per completion of the course.
Rate this course
How would you rate this course? Your rating is anonymous — just a number, with no name and no account.
Bring this to your team
Your colleagues can take this course too — it is free, needs no sign-up, and nothing they enter leaves their browser. If your institution wants the same ground covered as structured training for a team, with an instructor and your own cases, we run courses in a classroom or online for groups of ten or more.
Talk to us about training your team → See the paid courses →
Saolix does not record who takes this course, so it cannot verify these certificates. The certificate confirms completion of a free self-paced course and is not an accredited qualification.
Sources
The official documents this course relies on. Laws and guidance change, so check the current version.
- Final Notice to Deutsche Bank AG, 30 January 2017 (mirror trading; Principle 3 and SYSC 6.1.1R and 6.3.1R) · Financial Conduct Authority, United Kingdom
- Directive 2014/65/EU on markets in financial instruments (MiFID II) · EUR-Lex
- Regulation (EU) No 909/2014 on securities settlement and central securities depositories · EUR-Lex
- Directive (EU) 2015/849, Articles 13 and 14 (customer due diligence and ongoing monitoring) · EUR-Lex



