Exploring the role of internalisation in FX algo execution performance

Client’s feedback suggests that internalisation is not actually defined in any one common way across the FX market, which itself can lead to misunderstanding or lack of clarity around exactly what it is and the benefit it can bring, says James McGuigan, FX Algo Product Manager at Citi. He explains that, in its most basic form, internalisation can simply mean all or a part of an order is filled by the provider bank rather an on an external venue. However, McGuigan notes this is more a description of the mechanics of internalisation and is not a direct indication as to whether it is beneficial to the client or not. “Internalisation is a very important topic to us and many of our clients,” he adds. “Looking at internalisation from a risk standpoint provides a better view on the value it can bring and some insight into the relative value each of the various mechanisms available might offer.” 

So when an algo provider fills part of an order internally, it will either be risk reducing, risk neutral or risk increasing to their franchise, McGuigan says. “The risk reducing and risk neutral cases mean the bank has either pre-existing inventory or opposing algo order flows that can be matched against, the likelihood of which is directly correlated to the size of a banks FX franchise, an important part of which can be the banks own internal users of its FX algos. Citi has a large internal user base which presents greater opportunities for algo to algo matching. It is the risk increasing scenario that presents the more nuanced risk management requirement in order to truly internalise,” McGuigan adds.

On the other hand, where the bank actively and aggressively clears the risk position, it has clearly been externalised rather than internalised, says McGuigan. “Yet where the bank shows a price so skewed to its client base as to instantly induce offsetting flow, the definition of internalisation can be met without honouring the spirit of it,” he adds. “We might more accurately, but less tangibly, define internalisation as both refraining from hedging in the market, and also extending the holding time of the position out to longer than the natural rate of trading interest, so as to not cause additional market impact. For this reason, only market makers with both adequate client franchise, and also risk-taking appetite, can truly internalise.”

Improved execution performance

Asif Razaq, Global Head of FX Algo Execution, BNP Paribas, agrees, adding that while internalisation set-ups will differ from bank to bank, from an algo desk perspective there are three core ‘flavours’ of internalisation. He explains that the first form of internalisation from an algo execution perspective is also akin to what the market making desk classes as internalisation. “This is probably the most sought after flavour of internalisation,” Razaq says. “From an algo perspective, this particular form of internalisation can be done on very large sizes, with the two algo orders on opposite sides not needing to be of equal sizes to match against each other. This results in a trade that’s impossible to find anywhere else in the marketplace. Furthermore, because that trade is executed within BNP Paribas, there is no information leakage and no market impact on the back of that trade. That is key measure of what algo clients are looking for with internalisation.” 

However, Razaq notes that how well a bank is able to perform this depends on how busy its algo business is and the size of their client franchise. If a provider’s algo business is not very busy, the bank will not be able to find that offsetting interest to trade the two sides of the order at the same time, he explains. “But if the bank has a busy book and has clients leaving algo orders throughout the day, it increases the matching of the two orders and allows both clients to get the benefit of internalisation,” he says. “The caveat behind this form of ‘block matching’ internalisation is that it relies on the bank having an active algo business.”

The second form of internalisation according to Razaq is getting the algo to trade against a bank stream i.e. a BNP algo trading against BNP market making liquidity. “The algo desk is not directly internalising, it is indirectly internalising that order flow through our market making unit. The market making desk are not trying to externalise that flow, their default is to sit on that risk and find another client to trade the other side of that position, which we call ‘indirect’ internalisation – essentially minimising noise and impact,” Razaq says.

Reduced market impact 

He adds that the third flavour is a relatively new one, but again working with BNP’s market making desk. This third type is the ability to place orders with the market making desk, allowing the client to become BNP’s price to our client franchise, Razaq explains. “Clients now have the option to become BNP’s price on one side, improving the pricing in turn,” he adds. “If someone trades against that more attractive price, we are able to give that trade back to the algo client. So, this is a way that the client is utilising BNP’s market making desk to get further internalisation against our client franchise by way of the market making desk.” This third type is delivered via the bank’s BIX (BNP Internal Xchange) service, says Razaq.

At NatWest Markets, internalisation is also viewed as the offsetting of a position, or of a client order, with opposing flow from either internal trading desks or clients, says Dr John Quayle, Head of Client Algo Execution at NatWest Markets. He explains that if, for example, the child orders of a client algo are 100% filled with either opposing desk or client flow, then you can say that the algo offers 100% internalisation. “This is something that we can offer with the NatWest Peg Clipper algo,” adds Quayle. “The nuances then come in when thinking about which activity is ‘client activity’ and this definition may vary between liquidity providers (LPs).” He notes that the key benefit is reduced market impact, which is achieved because the child orders do not need to be placed on external venues. This in turn provides the LP with much more control over which counterparties can ‘see’, and therefore trade with, the algo order, such as limiting then to skew-safe counterparties, Quayle explains. 

“Additional benefits are reduced brokerage charges and simpler algos,” he continues. “Both of these will benefit the client outcome as there will be less slippage vs inception mid or the risk transfer price – the main measure of transaction cost for an algo – assuming that the internalisation framework is well implemented. The NatWest approach is very skew aware and focused on curated liquidity and internalisation first, which attempts to capture these benefits as much as possible, the results of this speak for themselves.”

Increased transparency 

In addition, there is now a greater understanding of the market impact reduction that can be achieved by internalising compared to that which can be achieved when filling on a more agency/external manner, according to Quayle. “The principal way that internalisation improves performance is in significantly reducing market impact,” he adds. “An algo that internalises using well curated low-market-impact liquidity simply performs better, a fact that is supported by our own data at NatWest, as well as ‘peer universe’ data from providers such as Tradefeedr.” In turn, Quayle notes that when an algo is interacting only with a given LP’s clients or trading desks, then that LP has total control over which clients the algo can match with and can therefore ensure the algo only interacts with those clients who are skew-safe.

According to Quayle, the challenge then becomes how to identify those skew-safe clients, and internalisation of algos is inextricably linked to that task. “Internalisation is only as good as the ability to identify those skew-safe clients, and internalisation itself does not necessarily guarantee a good outcome for the algo user – this cannot be stressed enough. It is only when used in combination with low-market impact liquidity that an algo will perform well. Hence, we at NatWest put as much effort into detecting skew, analysing and curating the liquidity as we do in designing the algo, and this clearly feeds through into our performance data,” he says.

Internalisation is also a key differentiator because it is an area where the algo provider’s franchise can come into play, argues Dr Ralf Donner, Head of Marquee Execution Solutions at Goldman Sachs. While a bigger franchise is generally better, it is still imperative to offer clients an intelligent, nuanced way to internalise, resulting in a better experience for the client. “From a client perspective, the number one expectation would be that there is a better mark-out profile from internalisation, which in turn improves the overall performance of the algo as well,” says Donner. “Some clients are also not so keen on last-look found in many of the secondary markets, but that can be avoided by internalising as the internalised prices are always firm. From a dealer perspective, we obviously do not have to pay brokerage for internalising, so we may hope to reduce the all-in cost of execution while offering the client a better experience.”

Point of differentiation

To this end, Goldman Sachs offers four flavours of internalisation: required, selective, matching, and skew, Donner explains. The first type, required internalisation, is used by algo clients if they have, for example, a twap or a vwap algo that has to finish at a certain scheduled time, he explains. In that case, it may be required to send a piece of the algo to a bank e-book in order to ensure that the scheduling requirements are still achieved. The second flavour, selective internalisation, is used when there is an external price and the algo essentially queries internally to see if it is possible to fill the algo at the external price, Donner says. He adds: “The reason why that could be an advantage is if the external price could be missed for some reason, such as via latency or because the external price is not a firm price but has last look conditions attached to it. In that case it might be better to just try and fill that internally and not have any kind of mark out that you might experience with the external price.”

The third internalisation offering is matching, Donner says, in which the bank operates a matching engine, an internal order book with various levels. The algo places limit orders and if a match on price occurs with another internal or external participant, then that results in an internal fill for both parties. Internal participants include FX and non-FX desks at GS. By the way, GS only permits external participation via our algos, which helps ensure the quality of the pool, he explains. Then the fourth and final type is skew internalisation, for when there is not an offsetting trade as there would be in the matching case, Donner explains. Instead, an offsetting trade is created by triggering a skew in streaming bid offer. “The e-book goes and it improves its bid, or improves its offer, and attracts liquidity to fill the algo,” Donner says.

McGuigan also notes the trend seen among certain clients who are now wanting to utilise bank algo offerings to effectively market make into the banks clients base, much like a venue. “As such, the ability of a bank to absorb this flow internally can be seen as being very important,” he adds. “In addition, the maturation of TCA in the FX algo space has also led to a better understanding around what good algo execution looks like and part of this has been the value true internalisation can provide.”

Moving into the mainstream

For BNP Paribas, the algo suite being able to access the three different flavours of internalisation is also seen as a core value proposition and one that clients quickly recognise the benefit of, says Razaq. In addition, all three internalisation types are available to the client by default but still have the option to opt out easily if they do not require a particular feature. “All three types are enabled on all our algo strategies,” Razaq adds. “We articulate to clients that BNP is a top flight player in the algo market and we have a very active algo execution business, which means that their chances of finding that block match in BIX, for example, is much higher at BNP than at any of our peers.”

BNP’s BIX also started life in the algo sector, providing opportunity between client algos trading against another client algo. Razaq shares that because the algo execution business is now very mature, there are a number of various different sectors of the bank that use BNP algos. “Our algos are deployed not only with our clients, but everywhere within the BNP group. Anyone within the group that has FX exposure to execute alsouse BNP algorithms to execute their trades,” he says. “In turn, this increases the matching probability in BIX because now we are not only matching against external clients, we are also matching against internal BNP departments as well. And as all of these trades are executed within  BNP, we can guarantee that the trades will have no market impact when they are matched in BIX. This is a very attractive proposition for our clients.” 

There is also no disadvantage to internalisation, according to Razaq and is on the whole highly sought after by clients. In some cases, clients may want to opt out of internalisation, if for example they need to respect the time horizon for the algo trade or they have some fiduciary duty to trade in the external market, he adds, but notes that they accept that risk because that is their execution objective. In addition, BNP provides full transparency by itemising the different internalisation types in their TCA reports. “Many banks actually mix and match and do not provide that level of transparency,” he says. “The TCA may show one category of internalisation but it is not clear what kind, so clients will often challenge what that actually means for them.”

Future of algos and internalisation 

Looking ahead, BNP has also developed another flavour of internalisation for clients in the systematic hedge fund space, Razaq reveals. “There is a large community of systematic hedge funds who build their own algos who are not able to benefit from the internalisation capability that using a BNP algo has to offer and have to instead trade against bank streams or go to the external market to source that liquidity. Increasingly this client base is realising they are actually missing out on BIX liquidity and they want a piece of the pie,” he says. “In response, BNP is creating an API product, called BIX FX, to offer BIX and BIX Peg services directly to clients so their internal algos can also leverage and start posting interest into BIXand BIX peg to leverage that flavour of internal liquidity.” In general, Razaq believes that clients will increasingly look at internalisation rates when they evaluate algo providers. “Internalisation is a key value proposition in BNP’s offering,” he adds. “At BNP, we are constantly looking for more ways to increase internalisation. It is definitely here to stay.” 

Historically, however, when there was just one form of internalisation, it  could be a double-edged sword and was not always the optimal choice for clients, explains Donner. Yet now clients tend to themselves seek greater internalisation because the offering is much more intelligent than before, he adds.

“It is universally viewed as being more useful and that is also borne out by third-party TCA. Now we have TCA providers also telling clients that if they increase internalisation that will be better for their algo execution outcomes,” Donner says. “Clients are far more sophisticated these days in their analysis of trades. We have migrated from a situation where we did have some clients who prefer not to internalise and even requested us to switch off internalisation for them, to now where nearly every client is not only okay with internalisation but is in fact very keen on it.” 

As a result, Donner also believes that internalisation is here to stay. “If you look at the FX liquidity landscape, primary markets continue their steady decline in volume. So if primary markets are not the answer to sourcing liquidity, what is? It probably is not the various secondary markets, given this patchwork of liquidity, some of it recycled, and much of it with limited transparency on transacted volumes,” he adds. “The solution is likely to be much more bespoke liquidity, curated pools and KYC liquidity. Internalisation is a great way to do that and as an algo provider, it is a great way for us to offer that sort of liquidity where we understand who is on the other side.”