Why NDF execution algos might still hit the spot

With any new product, there is always a concern that it won’t live up to the hype. Google Glass did not revolutionise computing, the Sinclair C5 did not mark a step change in the field of transport and blockchain has yet to provide any tangible solutions for the world of finance.

Many in the currency markets believe execution algorithms for non-deliverable forwards have similarly failed to meet expectations. During the pandemic, algo execution became the predominant method by which buy-side firms traded G10 spot foreign exchange. This led to significant demand for dealers to develop similar products for restricted currencies, such as the South Korean won and the Indian rupee, and BNP Paribas, Barclays, Goldman Sachs, HSBC and JP Morgan all added NDF execution algos to their arsenals.

Yet despite the initial uptake, banks say their clients have not used the algos as frequently or to trade the amounts they had hoped for. According to the 2022 survey by the Bank for International Settlements (BIS), less than half of the $266 billion-a-day notional NDF market is conducted on electronic platforms. A survey of 100 buy-side heads of FX trading, conducted in the second quarter of 2022 by market research firm WBR, showed that only 8% had adopted NDF algos, while 49% were looking at them but had no immediate plans to use them.

“NDF algos had definitely been overhyped in 2020,” says Vittorio Nuti, head of segregated algo execution for Deutsche Bank’s FX and listed derivatives divisions.

With some restricted currencies, it has been difficult to generate liquidity through electronic trading between banks and their clients. Only a handful of electronic interbank platforms enable NDF transactions. Wider use of FX execution algos has also stagnated, which has had a knock-on effect in terms of NDF algo use. In March, Coalition Greenwich reported that the percentage of asset managers and corporates using FX execution algos did not grow between 2021 and 2022, despite the return of volatility.

“Our client base is finding that risk transfer pricing – trading through voice at a guaranteed rate – is still extremely tight, even in this period of volatility,” says Vivek Sarohia, global head of FX alternative execution services at HSBC. “The appeal of algos and their inherent execution risk has thus been outweighed, and so we have seen reticence for clients to come back to the previous high usage of algos we saw during Covid.”

Others point to roadblocks in the way the market is structured that have prevented clients from accessing banks’ NDF algos. Popular buy-side execution management systems such as FXall and FX Connect have not supported NDF algo submissions, which has made it difficult for clients to select algo execution for their NDF trades.

Our client base is finding that risk transfer pricing – trading through voice at a guaranteed rate – is still extremely tight, even in this period of volatility

Vivek Sarohia, HSBC

“That’s affected the distribution and has stagnated the growth of the NDF algo business,” says Asif Razaq, global head of FX automated client execution at BNP Paribas. “Most of our clients simply could not access the service through the tools that they use in their organisations.”

Yet the NDF market’s microstructure is evolving in a way that could re-energise its electronification. Risk.net understands that, from next month, FXall will enable the real money community to access banks’ NDF execution algos. Refinitiv is also set to launch NDF Matching later this year in Singapore; the new central limit order book will cover the full range of Asia-Pacific and Latin American currencies.

With more electronic NDF markets, the potential for fragmented liquidity would give clients a greater reason to use algos.

“You get an initial uptake, then it plateaus, then more product gets added and you get more liquidity,” says David Wilkins, head of fixed income, currencies and commodities execution services for Europe, the Middle East and Africa and head of global e-FX sales at Goldman Sachs. “You also get more participants coming to get familiarity with the product. We’re certainly not at a level of full maturity and there is still a distance to go, but we’re very much committed to it and seeing good client interest in it.”

Forward planning

NDFs offer a way to hedge and speculate in currencies that are not freely traded, though trading via these instruments also occurs in G10 currency pairs. The cash payout is based on the difference between the agreed forward rate and the spot rate at maturity – typically one- or three-month expiries.

The market made significant strides during the 2010s. According to the BIS, the average daily volume (ADV) for NDFs nearly doubled between 2016 and 2019. During this period, interbank electronic platforms including 24 Exchange, Cboe FX, Euronext FX and EBS also expanded to allow trading in NDFs.

Since then, however, trading has slowed somewhat. The BIS reports that ADV for NDFs only increased by 3% between 2019 and 2022, while the entire FX market grew by 14%.

Although electronic communications networks such as 24 Exchange and Cboe FX are reporting double-digit growth in NDF volumes – with the former setting a new ADV record of $1.9 billion in February – this still represents only a fraction of the overall NDF market. This in turn makes it difficult to build electronic liquidity in NDF pairs outside the Indian rupee, the South Korean won, the Indonesian rupiah, the new Taiwan dollar and the Brazilian real.

“Outside of these pairs, the liquidity is really hard to find in an electronic format, and it’s still very much a voice broker-driven market,” says Razaq.

Liquidity is also extremely patchy in the most traded NDF pairs, where traders can only really get pricing during certain hours. This has resulted in inconsistent pricing and transparency in certain NDF pairs.

“You cannot trade US dollar/Brazilian real NDFs at 8am London time as there just isn’t any liquidity,” says Deutsche Bank’s Nuti. “If you want to trade an Asian currency very late in New York, you’re probably better off to trade by voice. NDFs do not trade 24/7 and there isn’t enough liquidity in the markets to have a two-way price.”

The banks that launched these algos to trade NDFs were also held back by the liquidity pools they could tap into. The predominant way the algos could access liquidity was through a bank’s internal franchise – where the bank is the market-maker of a restricted currency pair and the algo trades exclusively off liquidity supplied by client transactions.

BNP’s Razaq says this approach limits the algo’s capabilities and means the client would typically have to cross the spread. He adds that, as a result, most algos trade small ticket sizes of $30 million on average, with the bulk of activity coming from asset managers.

Brian Andreyko, chief product officer at Edgewater Markets, an FX technology firm that offers white-label trading services, says: “You need a diverse set of trading partners on the other side of the algo that have a genuine interest in the liquidity and a depth of book to really accommodate the size of transactions or frequency of transactions that algorithmic trading can bring. It’s fine that the algorithmic providers produce these NDF algos, but unless you have the trading partners on the other side that have the price streams to support those algos in a consistent and transparent fashion, then those algos can’t work.”

Other banks have started to do things differently. Goldman Sachs was the first to come to market with what it called a “smart algo”, whereby it could pull liquidity from the bank’s internal matching pool and gain access to external venues such as EBS.

BNP then launched a similar product whereby its algo could source liquidity from exchanges and other electronic communications networks in addition to its own. “We saw a significant performance upgrade to the strategies,” says Razaq. “Now [the algo] can post interest into the market, and the client now has the ability to earn spread rather than pay the spread. Because spreads in NDF currency pairs are generally wide, the cost savings were quite significant relative to what the clients were trading.”

However, the only dealer-to-client platform that has enabled asset managers to access bank algos for non-deliverable currencies has been Bloomberg’s FXGO. As a temporary workaround, BNP made its FX algo suite available through the Bloomberg App Portal. This meant clients could download BNP’s Cortex LIVE platform within the Bloomberg ecosystem without having to go through a major integration process.

A client can call the algo desk at BNP and verbally pass the order, which the algo trader will then input on the client’s behalf. When the order has been submitted, it appears on the Cortex app that runs within the Bloomberg terminal. The client can then pilot the algorithm as if it were managing it from its own execution management system. The client can still manage and see the execution live through the app, but it is delegating the order input part to BNP.

Dealers believe that once FXall and, eventually, FX Connect support NDF trading, it should provide a major boost to algo volumes.

Out on a date

Beyond the market structure, there are several improvements algo providers are looking to make. One significant change will involve making algos more accommodating to broken dates.

NDFs with a maturity of one month, and especially those denominated in Asian currencies, are the most liquid and therefore the easiest to trade electronically. However, Ralf Donner, Goldman Sachs’ head of fixed income, currency and commodities execution solutions, points out that not every client is looking for a one-month NDF.

“While some clients have responded to the standardisation of electronic Asia NDF markets around the one-month date by managing one-month risk and rolls to broken dates independently, others prefer a flexible workflow that is convenient for their chosen settlement date,” he says.

Instead, a client may trade on international monetary market dates or other non-standard tenors. Goldman is therefore amending its NDF algos so it can automate the process of rolling positions after they have been split up into smaller ‘child’ orders.

“The roll can be a significant component of both cost as well as the overall outcome of an execution, which means you have to do it effectively,” says Donner. “We’re moving towards a technology where the roll is done at the child order level rather than post-trade. This provides clients with a much better sense of the cost of their roll, so it doesn’t come as a surprise post-trade.”

An increasing number of dealers are also building algos that can execute both non-deliverable and deliverable currencies as a basket. These basket algos allow users to group correlating currencies with the aim of reducing overall execution costs. Some G10 pairs have strong correlations, such as the Japanese yen with the Australian dollar. However, the yen can sometimes be correlated with the South Korean won or the Philippine peso. An algo could automatically identify these correlations and trade both pairs within a single basket.

“A basket algo gives clients the ability to trade the correlation between currency pairs without the need to manage each leg individually,” says HSBC’s Sarohia. “The ability to constantly analyse market conditions across a region and manage the execution speed of each of the constituent pairs dynamically depending on their liquidity and volatility should bring savings in terms of spread capture, reduced volatility risk and greater control. We see that as a very powerful proposition for the emerging market space, and I expect that to be in the NDF space too.”

The hope is that a wider variety of algo users, such as hedge funds and regional banks, will boost electronic liquidity.

Edgewater Markets has seen a sharp increase in local onshore banks trading algorithmically with buy-side firms in global NDF markets. “When you include these local players, where all of a sudden there is organic interest, they have the inventory to warehouse these transactions without market impact,” says Andreyko. “As these markets electronify, it’s not just about a bid-offer anymore. It’s about the access in those markets and getting that true liquidity and transparency to the global markets for participants to execute their trades, and that’s really what’s going to drive algos.”

Unlike smart spectacles and power-assisted mini-tricycles – and, perhaps, blockchain – NDF execution algos may have life in them yet.

Editing by Daniel Blackburn