30.05.2023 Views

FX_Algo_News_May_2023-HIGH-RES-PAGES

  • No tags were found...

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

ISSUE 30 | MAY <strong>2023</strong> WWW.<strong>FX</strong>ALGONEWS.COM FOLLOW US AT:<br />

TOP STORIES<br />

Goldman Sachs launches Franchise<br />

Matching algo<br />

Goldman Sachs has rolled out a<br />

Franchise Matching algo to clients,<br />

which offers a new way of using bank<br />

internalisation to offset risk. The algo<br />

has been developed to send part of the<br />

parent order to a new venue, which<br />

triggers an e-book skew to make it more<br />

attractive to clients on the opposing side<br />

in order to fill the algo. Ralf Donner,<br />

Head of Marquee Execution Solutions at<br />

Goldman Sachs, says that the difference<br />

between the new franchise matching<br />

and the existing internal matching is<br />

that it does not require there to already<br />

be an opposing interest from a suitable<br />

set of clients. “The Franchise Matching<br />

algo can be a really interesting new<br />

way of internalising as well as being an<br />

effective way to fill larger order sizes. It<br />

was initially developed for our internal<br />

traders, who use it as a risk transfer<br />

tool, and so we have already seen that<br />

it works very well at sourcing liquidity.<br />

In particular, it works best in G4 pairs<br />

but it also works well in the rest of G10.<br />

Franchise matching is a client opt-in<br />

available as part of our flagship dynamic<br />

hybrid and pegged algos,” Donner adds.<br />

Ralf Donner<br />

LSEG’s <strong>FX</strong>all enhances algo offering<br />

to include NDFs<br />

London Stock Exchange Group’s <strong>FX</strong>all<br />

will extend its existing algo offering to<br />

include NDF currency pairs later this<br />

month. The move will allow customers<br />

to access liquidity providers’ existing<br />

algos for NDF trading providing them<br />

Audra Scharf<br />

with a new execution method for<br />

trading NDFs. Audra Scharf, Head of<br />

<strong>FX</strong>all at LSEG, says that the venue<br />

has been working closely with its<br />

partner banks to bring the additional<br />

functionality to its customers.<br />

“Buyside demand for bank LP algo<br />

execution continues to increase.<br />

Our buyside customers want to<br />

leverage the algo functionality they<br />

are already using on <strong>FX</strong>all for new<br />

products. We see this as a natural<br />

progression for <strong>FX</strong>all. We are building<br />

the capability and partnering with<br />

our LPs based on customer demand,”<br />

Scharf says. She adds that LSEG <strong>FX</strong>all<br />

expects the move will further support<br />

the evolution of NDF algos in the<br />

market and the platform also looks<br />

forward to making further product<br />

announcements later in the year.<br />

IN THIS ISSUE<br />

p1: TOP STORIES<br />

The latest industry stories<br />

p3: NEWS FEATU<strong>RES</strong><br />

More in-depth news<br />

p4: INDUSTRY REPORT<br />

<strong>FX</strong> algo adoption is looking up.<br />

p6: PRODUCT PROFILE<br />

State Street’s new Portfolio <strong>Algo</strong><br />

p8: INDUSTRY VIEWS<br />

Aligning outcomes with client expectations<br />

p14: ALGO OF MONTH<br />

TWAP from BNY Mellon<br />

p16: CASE STUDY<br />

Tradefeedr’s <strong>FX</strong> algo forecasting tools<br />

p18: TRADERS WORKSHOP<br />

Do Limits improve algo performance?


CitiVELOCITY<br />

LIVE NOW<br />

REAL TIME<br />

INTERACTIVE TCA<br />

• Drive in-flight decision making based on real time insight<br />

• Full execution breakdown<br />

• Performance against key benchmarks<br />

• Fully customisable view<br />

• Access full post-trade TCA report<br />

STILL INNOVATING<br />

Contact your <strong>FX</strong> esalesperson to learn more<br />

100+<br />

PLATFORM<br />

AWARDS<br />

12<br />

PATENTS<br />

e for education<br />

$66.9M<br />

RAISED<br />

© <strong>2023</strong> Citigroup Global Markets Inc. Member SIPC. All rights reserved. Citi Velocity, Citi Velocity & Arrow Design, Citi, Citi with Arc Design, Citigroup and Citi<strong>FX</strong> are<br />

service marks of Citigroup Inc. or its subsidiaries and are used and/or registered throughout the world. This product is offered through Citibank, N.A. which is authorised<br />

and regulated by the Financial Conduct Authority. Registered Office: Canada Square, Canary Wharf, London E14 5LB. FCA Registration number 124704. VAT Identification<br />

Number GB 429 625 629. Citi Velocity is protected by design and utility patents in the United States (9778821, 9477385, 8984439, D780,194, D780,194, D806,739) and<br />

Singapore (30201501598T, 11201505904S), and design registrations in the EU (0027845156-0001/0002, 002759266-0001).<br />

2 <strong>May</strong> <strong>2023</strong>


Bloomberg’s <strong>FX</strong>GO expands<br />

<strong>FX</strong> algo and analytics coverage<br />

In addition to the growth in <strong>FX</strong> algo use on Bloomberg’s <strong>FX</strong>GO, clients are also<br />

becoming increasingly sophisticated and interested in more advanced functionality,<br />

analytics and using algos to execute instruments outside of spot. Oleg Shevelenko,<br />

<strong>FX</strong>GO Product Manager at Bloomberg, explains how the platform works closely with<br />

algo providers to ensure that its algo offering remains cutting-edge.<br />

TOP STORIES<br />

NEWS FEATU<strong>RES</strong><br />

Oleg Shevelenko<br />

How has the algo offering and<br />

related toolsets developed over the<br />

past year?<br />

Driven by a need to reduce the cost and<br />

increase the efficiency of trading, algos<br />

emerged as one of the mainstream<br />

ways for a broad spectrum of clients<br />

to access the market. <strong>Algo</strong> offerings<br />

are no longer only the prerogative of<br />

major banks, as regional and non-bank<br />

providers are stepping in to offer their<br />

expertise to a growing population of<br />

clients. More than 40 providers are<br />

now offering algos on <strong>FX</strong>GO across<br />

close to 200 strategies. Initially, clients<br />

were relying on their algo providers<br />

to manage orders from start to finish.<br />

However, today many are taking a more<br />

active role in their order management as<br />

more data about algo behavior becomes<br />

available. Often clients would start<br />

their order execution with a passive<br />

strategy to capture the spread, and then<br />

gradually re-adjust to more aggressive<br />

ones to meet their desired time horizon.<br />

Therefore, advanced execution options<br />

such as in-flight strategy amendments,<br />

pause and resume features, and the<br />

ability to immediately fill the order<br />

became essential and are actively used<br />

by <strong>FX</strong>GO clients.<br />

How do you help support algo<br />

clients to meet their liquidity access<br />

needs?<br />

The fragmented nature of <strong>FX</strong> liquidity<br />

is one of the primary reasons for the<br />

rise in algos as they allow clients to<br />

access multiple liquidity destinations<br />

via a single strategy. However,<br />

additional liquidity is not an immediate<br />

guarantee of a better performance.<br />

As fill level venue analytics become<br />

more mainstream - showcasing various<br />

nuances of liquidity microstructure -<br />

clients are looking to incorporate this<br />

knowledge into their execution by<br />

controlling and customizing the pools of<br />

liquidity available to algos. We continue<br />

to work with liquidity providers to offer<br />

those customizable liquidity options<br />

and collect execution statistics related<br />

to each venue to feed into the TCA<br />

process.<br />

What support do you offer for algo<br />

execution in instruments beyond<br />

spot?<br />

As <strong>FX</strong> instruments beyond spot,<br />

including Forwards, NDFs and Precious<br />

Metals, are becoming increasingly<br />

more electronic from both pricing and<br />

hedging perspectives, liquidity providers<br />

are able to expand their algorithmic<br />

offering to cover those instruments as<br />

well. At this juncture, close to 20% of<br />

order execution on <strong>FX</strong>GO is performed<br />

in instruments other than spot. It is<br />

worth noting that as a large percentage<br />

of algo orders are passive they also<br />

enhance liquidity, which is especially<br />

beneficial for less liquid instruments.<br />

How have you enhanced your <strong>FX</strong> algo<br />

analytics products to support the<br />

increasingly sophisticated demands<br />

of users?<br />

As each algo provider offers a whole<br />

range of strategies with proprietary<br />

parameters, clients are looking for tools<br />

to help them navigate and rationalize<br />

this complex environment to allow for<br />

more intelligent decision making prior<br />

to and during the order execution.<br />

Bloomberg’s <strong>Algo</strong> Analytics hosting service<br />

helps address this demand by allowing<br />

providers to host their pre-trade and<br />

running order analytics within the order<br />

execution workflow of their <strong>FX</strong>GO clients.<br />

At present, there are 5 algo providers<br />

offering their integrated analytics<br />

services for their clients to leverage. In<br />

addition, <strong>FX</strong>GO is now fully integrated<br />

with Bloomberg cross asset TCA offering<br />

(BTCA) to allow clients to analyze their<br />

algo execution against various Bloomberg<br />

market data sources and benchmarks<br />

to further enhance their pre-trade algo<br />

provider and strategy selection.<br />

Do you have any future<br />

developments or areas of focus<br />

looking ahead that you would be<br />

happy to share?<br />

Partnering with our existing algo<br />

providers and onboarding new ones<br />

to expand our instrument coverage<br />

into emerging markets, NDFs, precious<br />

metals and derivatives is going to be<br />

key. We are also excited to work on the<br />

next generation of pre-trade decision<br />

support tools to integrate our composite<br />

pricing, news, analytics and cost models<br />

into all relevant trading workflows to<br />

optimize and automate the trading<br />

process for our clients.<br />

<strong>May</strong> <strong>2023</strong><br />

3


INDUSTRY REPORT<br />

The present<br />

and future of <strong>FX</strong><br />

execution algorithms<br />

According to the findings of a recent report published by Coalition Greenwich, after<br />

years of mediocre growth, <strong>FX</strong> execution algorithm adoption is looking up.<br />

The report was based on the<br />

responses of corporates, banks, fund<br />

managers and hedge funds and<br />

compared their reported use of <strong>FX</strong><br />

execution algos each from 2019 to<br />

2022. According to the findings, most<br />

buyside segments have witnessed a<br />

healthy rise in the use of algos since<br />

2020, especially on the real money<br />

side. Both fund managers and hedge<br />

funds have seen an uptick of algo<br />

use, with adoption now at 35% and<br />

46%, respectively. Growth among<br />

corporates and banks has been slower,<br />

however, only rising slightly over the<br />

four years to 13% and 20%.<br />

Report co-author Stephen Bruel, Head<br />

of Derivatives and <strong>FX</strong>, Market Structure<br />

and Technology at Coalition Greenwich,<br />

says that the trend is a general broader<br />

adoption of <strong>FX</strong> algos, particularly for<br />

<strong>FX</strong> spot. “There are a lot of reasons<br />

for this, including improved algo<br />

quality and better technology behind<br />

the algos,” he adds. “We are also<br />

Stephen Bruel<br />

seeing more investment in workflow<br />

management and firm being able to<br />

now integrate post-trade settlement<br />

data into their algo execution<br />

strategies. This all helps firms to feel<br />

more comfortable about using algos.”<br />

The increased use of TCA has also<br />

contributed to the uptake of <strong>FX</strong> algos,<br />

Bruel explains. Firms are now able to<br />

get a better sense of the quality of<br />

the algo under specific circumstances,<br />

whether that is by size, time of day,<br />

he adds. “Improvements in analytics<br />

reporting will also drive further algo<br />

use,” says Bruel. “That might be<br />

particularly specific to spot now, but<br />

these lessons will certainly be applied<br />

to some of the less automated asset<br />

instrument types going forward. Once<br />

<strong>FX</strong> spot algos are established even<br />

further, it makes sense that it will<br />

just cascade into some of the other<br />

instrument types.”<br />

BENEFITS OF ALGO USE<br />

According to the report, growth<br />

in algo adoption is a solution to a<br />

long-lived problem. The buyside is<br />

increasingly dealing with fragmented<br />

liquidity and a need to execute<br />

transactions more efficiently. The<br />

report contends that by using <strong>FX</strong><br />

algos, the buyside benefits from<br />

improved technology designed to<br />

address gaps in liquidity, data and<br />

other inefficiencies tied to over-thecounter<br />

markets.<br />

In addition, the report highlights the<br />

emphasise on transparency embedded<br />

in the <strong>FX</strong> Global Code (<strong>FX</strong>GC). “It<br />

turns out, improved technology and<br />

greater efficiencies on the desk are<br />

not the only drivers of algo adaption.<br />

Principle 18 of the <strong>FX</strong>GC includes<br />

guidance for algo providers, shining<br />

a light on how algos are marketed<br />

and managed,” the report notes. The<br />

<strong>FX</strong>GC also emphasises transparency<br />

and disclosure requirements that are<br />

intended to help consumers of algos<br />

understand more thoroughly the<br />

nature of the tools they may (or may<br />

not) use. “It appears the more the<br />

industry knows about a product, the<br />

more likely they are to use and trust<br />

it,” the report authors suggest.<br />

The report also questions whether<br />

the adoption of <strong>FX</strong> algos has now<br />

plateaued. “Because <strong>FX</strong> algo use has<br />

not achieved the same market share<br />

as in other asset classes, it is fair to<br />

ask if adoption is levelling off. Our<br />

data shows the answer is a resounding<br />

‘no’. Trading in size in difficult markets<br />

require finesse and creativity, which<br />

algos can provide,” the report adds.<br />

EXPANSION BEYOND <strong>FX</strong><br />

SPOT<br />

Furthermore, the report observes that<br />

creativity also comes into play when<br />

deciding which algo is most suited<br />

for the market conditions in which<br />

participants are trying to execute. This<br />

often requires sell-side guidance to<br />

help buyside traders use the optimal<br />

tool for the market conditions, the<br />

report notes, and warns that despite<br />

the popularity of algos, relationships<br />

still matter. “In equity markets, these<br />

challenges spawned ‘algo wheels’<br />

4 <strong>May</strong> <strong>2023</strong>


- something likely coming to <strong>FX</strong><br />

markets,” the report adds.<br />

The opportunity for additional algo<br />

growth sits squarely in other <strong>FX</strong><br />

products, according to the report.<br />

The value of algorithmic execution<br />

in <strong>FX</strong> spot is mostly understood,<br />

and adoption will continue to rise,<br />

according to the report. “The ability of<br />

algo providers to apply this technology<br />

to other instruments such as NDFs<br />

represents an area of potential usage.<br />

To date, uptake has been limited,” the<br />

report adds.<br />

Talk of the use of algos in certain <strong>FX</strong><br />

swaps and options - particularly nonstandard<br />

dates - is probably sometime<br />

away. However, if the trajectory<br />

of electronic trading and algo use<br />

gleaned from other asset classes is<br />

any indicator of the future of <strong>FX</strong> algo<br />

adoption, the prospects of broad<br />

adoption are very good, the report<br />

concludes. Bruel adds that while algo<br />

use in spot is certainly a well told<br />

story, the focus now is on some of<br />

the less automated instrument types,<br />

such as NDFs and <strong>FX</strong> options. “Dealers<br />

recognise that there probably will<br />

be demand going forward and we’re<br />

starting to see a lot of questions being<br />

asked around the ability to more<br />

effectively execute your NDF and your<br />

options trades, but we have yet to see<br />

a wholesale introduction of algos into<br />

those markets.”<br />

Read the full report here: https://www.<br />

greenwich.com/blog/present-andfuture-fx-execution-algorithms<br />

Coming shortly to deliver even more insight and<br />

commentary about algorithmic <strong>FX</strong> trading:<br />

MARKETWATCH<br />

• The launch of the enhanced<br />

<strong>FX</strong><strong>Algo</strong><strong>News</strong> website. The<br />

new site has been designed<br />

for easier navigation and<br />

will include our updated <strong>FX</strong><br />

algo provider database and<br />

other new resources.<br />

• Our new regular Anatomy of<br />

an <strong>Algo</strong> and Dequantification<br />

features. These will drill down<br />

on the technical structure<br />

and inner workings of these<br />

increasingly powerful trade<br />

execution toolsets.<br />

• Our Summer <strong>2023</strong><br />

Supplement. This will be<br />

exploring outsourced <strong>FX</strong><br />

algorithmic trading services<br />

and the benefits these can<br />

deliver for both buyside and<br />

sellside market participants.<br />

For more information please visit: www.fxalgonews.com<br />

<strong>May</strong> <strong>2023</strong><br />

5


PRODUCT PROFILE<br />

State Street unveils<br />

new Portfolio <strong>Algo</strong><br />

The complexities involved in managing multi-leg, multi-currency trades has proved<br />

to be a significant pain point for portfolio trading, particularly among Real Money<br />

clients. Mary Leung, Global Head of Client <strong>Algo</strong>s at State Street, explains how the newly<br />

launched Portfolio <strong>Algo</strong> helps to automate this process and shares the many innovative<br />

features that the new strategy has to offer.<br />

Mary Leung<br />

Please share more details about<br />

what the new algo is?<br />

State Street is very excited to announce<br />

the launch our new strategy – the<br />

Portfolio <strong>Algo</strong>. This strategy is designed<br />

to execute a portfolio or basket<br />

of orders across various currency<br />

pairs. This innovative solution will help<br />

navigate the complexities of executing<br />

multiple currency pairs simultaneously,<br />

and reduce overall execution costs<br />

when compared to executing each<br />

currency pair individually.<br />

Another goal of the strategy is to<br />

enhance efficiency of the buy and<br />

sell netting opportunities in the<br />

basket orders by breaking the legs<br />

into tradeable, netted orders. We<br />

then optimize the trade-off between<br />

cost and risk of the aggregated<br />

execution. We do this by analysing<br />

the characteristics and correlations of<br />

the netted currency pairs to generate<br />

optimal execution paths for each pair.<br />

These cost and risk trade-offs can be<br />

reviewed on our pre-trade analytics<br />

tool. During execution, clients can<br />

monitor the execution of each leg<br />

with our real time TCA. Post execution,<br />

clients will receive additional TCA<br />

analysing the performance of each<br />

of the netted legs, as well as the<br />

aggregated performance against the<br />

initial portfolio orders.<br />

What do you think are the<br />

differentiating features of your<br />

portfolio algo?<br />

State Street’s Portfolio <strong>Algo</strong> is more<br />

than just a workflow automation<br />

solution. The strategy computes<br />

optimal execution trajectories by<br />

modelling temporary and permanent<br />

market impact cost, volatility cost<br />

and the cross-correlations of different<br />

currency pairs. These trajectories aim<br />

to minimize the mean-variance of<br />

the Implementation Shortfall cost of<br />

the basket (i.e. reduce slippage from<br />

the arrival price of the portfolio). We<br />

construct an efficient frontier based<br />

on different levels of risk aversion.<br />

Each point on the frontier represents<br />

distinct execution pathways designed<br />

to optimally liquidate the basket on an<br />

aggregate level.<br />

An important concept leveraged in<br />

the Portfolio <strong>Algo</strong> is the efficient<br />

frontier. The structure of the efficient<br />

frontier is expressed on two ends of<br />

the execution spectrum: executing<br />

everything now, at a known but high<br />

cost (e.g. aggressive sweep or risk<br />

transfer); versus executing a simple,<br />

evenly distributed strategy slowly<br />

over a time horizon (i.e. TWAP) at an<br />

unknown, but low expected cost. Our<br />

Portfolio <strong>Algo</strong> pre-trade analytics allow<br />

clients to observe the transition from<br />

one extreme efficient frontier point<br />

to another and shows how we can<br />

reduce the expected costs with only<br />

small increases in standard deviation.<br />

This allows our clients to express<br />

their varying risk profiles by choosing<br />

different urgencies. Once an urgency<br />

is selected they will be able to review<br />

the execution paths before executing<br />

the portfolio.<br />

Another differentiating feature is<br />

the flexibility to constrain the basket<br />

USD exposure to its initial levels, or<br />

to be completely USD neutral. It is<br />

inherently challenging for a human<br />

trader to manage multiple legs of<br />

currency pairs with different risk and<br />

cost characteristics while maintaining,<br />

balancing or neutralizing the USD<br />

exposure of the original order. State<br />

Street’s Portfolio <strong>Algo</strong> is able to derive<br />

6 <strong>May</strong> <strong>2023</strong>


Why did you develop the new<br />

algo?<br />

We maintain a consistent dialog with<br />

our clients, listen to their execution<br />

needs and partner with them to find<br />

solutions. The Portfolio <strong>Algo</strong> was built<br />

based on increased demand, interest<br />

and feedback from our clients who<br />

wished to simplify their workflow and<br />

optimize the cost of trading a basket of<br />

currency pairs.<br />

Figure 1: Efficient frontier - Each point translates into specific risk aversion/urgency and the<br />

associated costs/standard deviation.<br />

Client experience is a key theme to<br />

State Street’s <strong>FX</strong> algo offering. We<br />

actively engaged with our clients<br />

to develop this new strategy. From<br />

gathering essential requirements, to<br />

reviewing the concept and design of<br />

the model and workflow, we built the<br />

Portfolio <strong>Algo</strong> to ensure the solution<br />

can provide additional value.<br />

What has been the response/<br />

feedback? Has it been well<br />

received?<br />

The offering has been very well<br />

received. During recent walkthrough<br />

presentations we received tremendous<br />

interest from clients eager to try the<br />

new strategy.<br />

During a recent global algo tour<br />

several of our clients, who have yet to<br />

adopt <strong>FX</strong> algos, commented that the<br />

Portfolio <strong>Algo</strong> is a “game changer” for<br />

them.<br />

Figure 2: Optimal execution paths based on a chosen urgency<br />

the optimal execution pathways of the<br />

different currency pairs and ensure<br />

desired constraints are observed<br />

throughout execution.<br />

Can the algo be customised to<br />

allow clients greater control of<br />

their executions?<br />

Absolutely. The Portfolio algo, while<br />

sophisticated, is designed to easily<br />

deploy. Clients can let the algo decide<br />

the duration to run the order, or<br />

for clients who desire more control,<br />

they will be able to select their own<br />

duration and urgency. Once these<br />

customized selections are made the<br />

strategy will optimize the basket<br />

orders on an aggregate level based<br />

on client inputs. We understand that<br />

clients also like the flexibility to stay<br />

engaged at critical moments similar<br />

to our flexible design of our single<br />

order algo strategies. The portfolio<br />

algo offers the ability to amend<br />

basket orders details, speed up/slow<br />

down execution, add/remove limit<br />

prices and pause/resume individual<br />

legs or the entire basket. The client<br />

can also manage particular legs of<br />

the basket individually by using any<br />

existing strategy in our algo suite or<br />

through a risk transfer price. If any<br />

of these flexible tools are utilized the<br />

portfolio algo will be able to adjust<br />

and recalculate the optimal execution<br />

curves in real time to accommodate<br />

changing parameters.<br />

In what ways will the new<br />

strategy benefit clients? Are<br />

there particular groups of clients<br />

who will find the algo especially<br />

effective?<br />

A consistent pain point for our core<br />

Real Money client base occurs during<br />

monthly and quarterly portfolio<br />

rebalance and asset allocation trades.<br />

Managing multi leg, multi-currency<br />

trades is time consuming, gives rise<br />

to in-efficient execution, increased<br />

implementation shortfall, and<br />

operational headaches.<br />

The Portfolio <strong>Algo</strong> is specifically<br />

designed with these challenges in mind<br />

and solves them with essentially one<br />

click that can save our Real Money and<br />

asset manager community countless<br />

hours and reduce execution costs.<br />

<strong>May</strong> <strong>2023</strong><br />

7


INDUSTRY VIEWS<br />

Aligning <strong>FX</strong> algo<br />

trading outcomes<br />

with customer<br />

expectations<br />

Image by shutterstock<br />

8 <strong>May</strong> <strong>2023</strong>


Nicola Tavendale<br />

The Bank for International Settlements recently released a working paper on the<br />

foreign exchange market, part of which explores the expansion of algorithmic trading.<br />

In particular, the paper charts the rise in algo use on EBS to 2022, by which time both<br />

bank and non-bank <strong>FX</strong> algo trading dominated with each accounting for just over 40%<br />

of trading volume on the platform. As algo use continues to grow, so too have client<br />

expectations around algo performance and the level of service they expect to receive<br />

from algo providers. Yet where can improvements still be made and what will be next<br />

in the evolution of the algo trading experience? Nicola Tavendale writes.<br />

In addition to the growth of <strong>FX</strong> algos<br />

deployed on primary CLOBS, the BIS<br />

Working Paper Banks notes that, due<br />

to the role that such platforms play<br />

in the market, algorithmic trading<br />

has also had an important impact on<br />

price discovery in <strong>FX</strong>. Furthermore, the<br />

paper adds that execution algorithms<br />

are now also being used directly<br />

by some of the more sophisticated<br />

customers in the <strong>FX</strong> market and that<br />

these users “increasingly rely on<br />

smart order routing and execution<br />

algorithms to spread large orders over<br />

time and across multiple electronic<br />

venues.”<br />

Joel Marsden, e<strong>FX</strong> Senior Currency<br />

Trader at ANZ, says that algo trading<br />

literacy is also increasing considerably<br />

among the wider institutional and<br />

multi-national corporate client<br />

base, which is a broadening from<br />

earlier investor and asset manager<br />

adoption. He explains that with<br />

<strong>May</strong> <strong>2023</strong><br />

9


INDUSTRY VIEWS<br />

Joel Marsden<br />

“Clients better<br />

understand the<br />

trade-off between<br />

capturing spread on<br />

their execution against<br />

the market risk they<br />

assume”<br />

this knowledge comes an expectation<br />

around aligning algo performance<br />

to execution goals, which at its most<br />

fundamental comes down to long<br />

term outperformance against risk<br />

transfer pricing.<br />

“Clients better understand the tradeoff<br />

between capturing spread on their<br />

execution against the market risk they<br />

assume, when for instance, making<br />

the decision to be passive, neutral,<br />

or aggressive in their execution. Pretrade<br />

analytics are key to helping<br />

clients with these decisions,” Marsden<br />

says. He continues: “One of our key<br />

algo product differentiators is how<br />

we manage and curate our liquidity<br />

sources. Clients broadly favour spread<br />

capture along with minimal market<br />

impact, so the choice of liquidity<br />

sources is critical to achieve both<br />

those objectives.”<br />

EVOLVING EXPECTATIONS<br />

At ANZ, Marsden explains that the<br />

same liquidity sources and algorithms<br />

used by ANZ for internal e-commerce<br />

hedging also form the building<br />

blocks for its client algo offering,<br />

meaning that both are objectively<br />

aligned. Using TCA by liquidity pool,<br />

he explains that clients can visualise<br />

and review granular, as well as overall,<br />

execution performance. “We believe<br />

TCA should always be exclusive and<br />

transparent of any brokerage/mark<br />

up, so customers understand the raw<br />

wholesale market pricing and impact<br />

of every individual order,” Marsden<br />

says. “Clients expect that we are<br />

always sourcing and routing the most<br />

efficient, low impact, liquidity on their<br />

behalf.”<br />

In addition, the general requirements<br />

of nearly every client now includes<br />

pre-trade analytical tools with insights<br />

into the aggregated electronic market,<br />

insights into spreads, insights into the<br />

potential cost of execution as well as<br />

comparative studies of risk transfer<br />

versus algo execution, says Ralf<br />

Donner, Head of Marquee Execution<br />

Solutions at Goldman Sachs. He adds<br />

that most clients tend to also expect<br />

some kind of real time order monitor<br />

or even a real time order manager,<br />

depending on the platform, while<br />

post trade they would certainly expect<br />

both a post trade report from the<br />

bank as well as a menu of third party<br />

transaction cost analysis, in addition<br />

to some bank aggregated reporting<br />

as well.<br />

“Typically, the onboarding process for<br />

a new account algo account will also<br />

require some kind of evidence of past<br />

performance from the algo provider.<br />

Any new algo client that we are<br />

onboarding now asks us for a study of<br />

the most recent set of algo executions,<br />

across all currency pairs and what the<br />

distribution is, how did they perform?<br />

Clients these days will often ask us<br />

some really insightful questions,”<br />

Donner says. Some of the some of<br />

the larger algo providers such as<br />

Goldman Sachs now have a dedicated<br />

algo team, which Donner notes is<br />

not that common. “Particularly in the<br />

Asia session, some clients really like<br />

the fact that they can speak to an<br />

algo expert anytime of the day. That’s<br />

part of the personalised service we<br />

provide,” he adds.<br />

DEVELOPING INTELLIGENT<br />

SOLUTIONS<br />

“However, what I hope we do not end<br />

up doing in foreign exchange, and<br />

there are some worrying signs from<br />

certain hedge fund clients that we<br />

are going in this direction, is having<br />

to actually create too many bespoke<br />

products for clients. A bespoke<br />

algo product is likely to be a short<br />

term gain and long term pain as it<br />

exponentiates testing and thus slows<br />

future development.” says Donner.<br />

Instead, he says that his team does<br />

sometimes need to help clients who<br />

are using an API connection via a<br />

multi-dealer platform to complete<br />

bespoke work on their Fix connection,<br />

such as a translation of their settings<br />

on their end to the bank’s Fix<br />

messaging, for example. “Many of<br />

the multi-dealer platforms we deal<br />

with are insufficiently on top of this,”<br />

Donner adds. “It should be their<br />

job to ensure this translation layer is<br />

done. It is also in their best interest<br />

to ensure that their clients using Fix<br />

connections are provided with good<br />

service, but some of the smaller multidealer<br />

platforms still seem unable or<br />

unwilling to do this.”<br />

Manual vs. algorithmic execution on EBS Market<br />

The increased availability of algo data<br />

has also raised client expectations<br />

10 <strong>May</strong> <strong>2023</strong>


Evolving with the Market<br />

Added Functionality to Support <strong>Algo</strong>s & Allocations Now Live<br />

®<br />

GUI<br />

Live<br />

RFS<br />

Functionality<br />

Added<br />

London<br />

& Tokyo<br />

Offices Open<br />

<strong>FX</strong>|Insights<br />

Analytics Tool<br />

Launched<br />

USD11T<br />

Supported<br />

for 2020<br />

<strong>FX</strong> Spot<br />

Streaming<br />

Only<br />

<strong>FX</strong> Fwds,<br />

Swaps Added<br />

Streaming<br />

Precious<br />

Metals Added<br />

NDF/NDS,<br />

PM Swaps<br />

Added<br />

Total Reaches<br />

15 LPs<br />

<strong>Algo</strong>s &<br />

Allocations<br />

Functionality<br />

Added<br />

®<br />

<strong>FX</strong>SpotStream is a bank owned consortium operating as a market utility, providing the infrastructure<br />

that facilitates a multibank API and GUI to route trades from clients to LPs. <strong>FX</strong>SpotStream provides a<br />

multibank <strong>FX</strong> streaming Service supporting trading in <strong>FX</strong> Spot, Forwards, Swaps, NDF/NDS and<br />

Precious Metals Spot and Swaps. Clients access a GUI or single API from co-location sites in New York,<br />

London and Tokyo and can communicate with all LPs connected to the FSS Service. Clients can also<br />

access the entire <strong>Algo</strong> Suite of the FSS LPs, and assign pre- and/or post-trade allocations to their<br />

orders. <strong>FX</strong>SpotStream does not charge brokerage fees to its clients or LPs for its streaming offering.<br />

<strong>Algo</strong> fees from an LP are solely determined by the LP.<br />

<strong>May</strong> <strong>2023</strong><br />

11


INDUSTRY VIEWS<br />

Ralf Donner<br />

“A further area to<br />

explore is the potential<br />

for taking a more multiasset<br />

approach to algo<br />

development”<br />

around the outcomes they can expect<br />

to achieve using algo execution,<br />

says Asif Razaq, Global Head of <strong>FX</strong><br />

Automated Client Execution at BNP<br />

Paribas. Client expectations in the<br />

past were more exploratory, but<br />

the readiness of analytics and TCA<br />

has resulted in those expectations<br />

becoming much more focused, he<br />

adds. “That is a big change for the<br />

market,” Razaq says. “We continue<br />

to meet their expectations through<br />

the ongoing development of our<br />

algo solutions, which have evolved<br />

considerably since the launch of our<br />

first algorithm almost 10 years ago.”<br />

Today, BNP Paribas offers fourth<br />

generation algorithms, or interactive<br />

algos, which can provide real-time<br />

feedback and, using the digital trading<br />

system ALiX, can provide commentary<br />

to the client mid-execution so the<br />

client can modify the algo’s flight path<br />

if needed and achieve a much better<br />

outcome on the execution. “We have<br />

been constantly building more and<br />

more tools, which give clients much<br />

better control over the execution and<br />

a much better overall understanding<br />

of how the algos work,” adds Razaq.<br />

Another area of significant<br />

development is the demand from<br />

clients to build customised solutions.<br />

According to Razaq, while the BNP<br />

Paribas algo suite is relatively small<br />

by design, it should cater for 90%<br />

of client execution needs. “However,<br />

if a client does have a very bespoke<br />

requirement to adapt an existing algo<br />

to make it a perfect match to their<br />

needs, then we are able to tailor<br />

the algo accordingly,” he says. “This<br />

framework, Flex, allows us to evolve<br />

our algo family without confusing<br />

clients with too many different core<br />

algos, but instead tailoring the existing<br />

algos to specific client needs. We now<br />

have around 20 different variants<br />

of Chameleon and 16 different<br />

variants of Iguana, all designed to<br />

do something very specific for those<br />

clients’ needs.”<br />

IMPACT OF ANALYTICS DATA<br />

In addition, BNP Paribas has launched<br />

fifth generation algorithms, complex<br />

strategies designed to perform<br />

more than just execution, such as<br />

automating the client’s pre-execution<br />

workflow. One example of this is<br />

Rex, a portfolio hedging or basket<br />

algorithm designed for clients who<br />

typically do not have just one currency<br />

pair to execute, but five or six. “This<br />

takes optimising client executions one<br />

stage further. Rex can automate the<br />

entire workflow, construct a project<br />

plan for the client and show them how<br />

it can use algorithms in a collective,<br />

daisy chain, fashion,” Razaq says. He<br />

notes that much of the success of the<br />

fifth generation algos has been the<br />

Increased availability of algo data has also raised expectations around the outcomes clients can expect to achieve using algo execution<br />

12 <strong>May</strong> <strong>2023</strong>


esult of the time and effort taken<br />

by his team to integrate the algos<br />

with the multi-dealer platforms that<br />

the clients use. “This makes it much<br />

easier for clients to use the algos<br />

because they are already integrated<br />

to their tool that they use on a day<br />

to day,” Razaq adds. “The uptake<br />

from clients has been really positive<br />

as a result. We’re now brainstorming<br />

with clients on other ideas and other<br />

complex workflows that they have as<br />

an organization that we can look to<br />

automate.”<br />

Marsden believes that pre-trade<br />

analytics are also becoming<br />

increasingly important in educating<br />

and guiding clients with their specific<br />

execution objectives. As clients assume<br />

the market risk on the time taken<br />

for the execution, it is imperative<br />

that they are equipped with an<br />

understanding, on average, how long<br />

an algorithm takes for any given time<br />

of day or season given prevailing<br />

market conditions, he explains. “To<br />

that end, the pre-trade analytics<br />

which ANZ provide overlay liquidity<br />

heatmaps with the recent historical<br />

expected trading volume and velocity,<br />

which is particularly important for<br />

clients that execute more passive<br />

or implementation shortfall style<br />

algorithms where the time taken to<br />

complete is variable,” says Marsden.<br />

“Our pre-trade analytics also visually<br />

guide clients to be cognisant of known<br />

event risk that have the potential to<br />

materially move the market, such<br />

as news releases, major economic<br />

indicators along with daily benchmark<br />

fixes, which many institutional clients<br />

may otherwise not be aware of.”<br />

According to Donner, there is also<br />

much more that can still be done to<br />

increase transparency around how the<br />

algos with different liquidity sources.<br />

“<strong>Algo</strong> analytics feels like it now pops<br />

up everywhere,” he explains. “Yes,<br />

there is a huge amount of analytics<br />

out there, but it’s daunting. It is like<br />

stepping into an aircraft cockpit and<br />

there are dials, wheels, bells and<br />

whistles all over the place. It’s very<br />

difficult to make sense of the entire<br />

the entire thing. Clients are looking<br />

for something that is a little bit more<br />

streamlined, something that shows<br />

you everything at a at a glance.”<br />

NEW INNOVATIONS AND<br />

DEVELOPMENTS<br />

Achieving this the main focus which<br />

will help clients in terms of the<br />

usability of the algo analytics, says<br />

Donner. He adds that analytics tools<br />

also tend to be very disparate. “There<br />

will be one pre-trade tool, another<br />

separate intra-trade tool and then<br />

yet another for post-trade,” he says.<br />

“Harmonizing those tools is an aim<br />

worth pursuing.” A further area to<br />

explore is the potential for taking a<br />

more multi-asset approach to algo<br />

development, Donner says. From<br />

a client perspective, he explains<br />

that <strong>FX</strong> does not sit on its own but,<br />

particularly from a client perspective,<br />

it sits together with other tradables.<br />

Another recent release is a new<br />

Franchise Matching algo, which was<br />

developed as a new way of leveraging<br />

internalisation to fill the algo order.<br />

Donner explains: “This is all about<br />

liquidity. Liquidity used to be a bit of<br />

a challenge with algos. There were<br />

only a handful of venues and there<br />

was a great deal that had to be done<br />

externally.” He adds that algos in<br />

general have improved over the years<br />

and have become much smarter about<br />

how they source internal liquidity.<br />

“Our new Franchise Matching algo<br />

is yet another example of being up<br />

being smart about sourcing internal<br />

liquidity,” Donner says. “It encourages<br />

the opposing interest, as distinct from<br />

previous versions of internalisation<br />

which sought existing opposing<br />

interest. Basically, it’s a new form of<br />

internalisation.”<br />

Razaq adds that at BNP Paribas, they<br />

are already ahead of the curve and are<br />

recognised as innovators in the algo<br />

development. “We’re still enhancing<br />

our strategies on a day to day basis<br />

and we’re looking at more fifth<br />

generation algo opportunities to build<br />

upon,” he says. “But we now feel<br />

we’ve conquered <strong>FX</strong> in terms of our<br />

capability. The bigger opportunity for<br />

us is to now look at how we can apply<br />

the <strong>FX</strong> algo technologies to different<br />

asset classes.” This involves building<br />

the <strong>FX</strong> family of algos, Chameleon,<br />

Viper and Iguana, but making them<br />

available to trade different asset<br />

classes, such as commodities, precious<br />

metals, and Futures with underlying<br />

Asif Razaq<br />

“Client expectations<br />

in the past were more<br />

exploratory, but the<br />

readiness of analytics<br />

and TCA has resulted<br />

in those expectations<br />

becoming much more<br />

focused”<br />

<strong>FX</strong>, bonds and equities, adds Razaq.<br />

“We are planning to launch these new<br />

algos later this year. Our focus now is<br />

how can we apply the same principles<br />

and the same technology in different<br />

asset classes, giving the end client a<br />

unified view of algos. Because if they<br />

know what Chameleon does in <strong>FX</strong>,<br />

then they’ll know what Chameleon<br />

will do when it comes to trading gold<br />

or when it comes to trading futures,”<br />

he says.<br />

According to Razaq, that common<br />

understanding of how the algo works,<br />

using the same interface to keep the<br />

look and feel of the <strong>FX</strong> algos, as well<br />

as having Alex to provide algo users<br />

with commentary on their futures<br />

execution and <strong>FX</strong> execution, will<br />

bring valuable unity within the algo<br />

offering. “The cost pressures we are<br />

seeing now in the world, which our<br />

clients are feeling as well, mean that<br />

everyone’s being asked to do a lot<br />

more with a lot less. Clients are no<br />

longer just an <strong>FX</strong> execution desk, but<br />

they’re increasingly becoming multiasset<br />

execution desk. The more we can<br />

do to help them to align our products<br />

to those different asset classes,<br />

the easier we can make life for our<br />

clients,” Razaq concludes.<br />

<strong>May</strong> <strong>2023</strong><br />

13


A<br />

ALGO OF THE MONTH<br />

TWAP from BNY Mellon<br />

With Joseph Café, Global Head of e<strong>FX</strong> Sales<br />

NAME OF THE STRATEGY:<br />

TWAP<br />

DESCRIPTION &<br />

CAPABILITIES<br />

• A time-based algorithm that<br />

works the order during a<br />

user-specified horizon by<br />

spreading trades along a linear<br />

distribution.<br />

Joseph Café<br />

In what ways does the strength and experience that BNY Mellon has built<br />

up in <strong>FX</strong> over many years put it in a good position to meet the growing<br />

demand and interest in <strong>FX</strong> algorithmic trading strategies amongst<br />

buyside firms?<br />

BNY Mellon has been at the forefront of change in the financial world for nearly 240<br />

years. We have weathered times of calm and crisis, and we have done it by adapting,<br />

refining and consistently improving our services. That spirit of discovery has enabled<br />

us to stay ahead of the curve and deliver consistent solutions for our clients. We are<br />

the world’s largest custodian and the singular clearer of U.S. Treasuries. This unique<br />

position in financial markets has enabled deep relationships and trust with our clients,<br />

resulting in new and improved solutions built on client feedback. Many of our clients<br />

have voiced the importance of a differentiated liquidity provision when it comes to<br />

execution algos. Additionally, many of our clients’ value BNY Mellon’s unique principal<br />

liquidity which is driven by our uncorrelated custody flows.<br />

What specific benefits do your <strong>FX</strong> algo solutions provide and why are<br />

they proving increasingly popular?<br />

Many of the benefits of our <strong>FX</strong> <strong>Algo</strong> solutions reside in our liquidity provision.<br />

Given our unique market position our liquidity pools offer flexibility by combining<br />

our custody, principal, and external market sources. This liquidity provision, and<br />

value added features, like the ability to auto-roll algo orders, provides clients<br />

improved execution, and reduces manual steps.<br />

In what ways do your algorithmic and TCA toolsets help clients to gain a<br />

more holistic and detailed view of their <strong>FX</strong> trading activities?<br />

Buy-side clients continue to demand more transparency in OTC <strong>FX</strong>. Our TCA toolsets<br />

are valuable for buyside clients in providing more transparency by benchmarking<br />

execution across several metrics. Examples of these metrics include performance<br />

against risk transfer, arrival-mid, and market impact. In our efforts to help clients with<br />

transparency we have partnered with an independent third-party TCA provider. These<br />

conversations lead to more informed conversations between the sell-side and buy-side.<br />

EXECUTION OBJECTIVES<br />

• The strategic goal is to<br />

minimize slippage against<br />

the interval TWAP, and to<br />

achieve this by spreading the<br />

order evenly over the trading<br />

horizon.<br />

• Additionally, the strategy has<br />

passive order placement logic<br />

which strives to reduce spread<br />

costs.<br />

WHEN TO USE IT<br />

• It’s useful when a client wants<br />

to execute larger orders over<br />

a specific time period while<br />

minimizing market impact and<br />

reducing spread costs.<br />

KEY PARAMETERS &<br />

FEATU<strong>RES</strong><br />

• This strategy will access BNY<br />

Mellon’s unique internal<br />

liquidity as well as external<br />

venues vetted for performance<br />

and market impact.<br />

14 <strong>May</strong> <strong>2023</strong>


ODSC Europe <strong>2023</strong>:<br />

Machine Learning For<br />

Finance Track<br />

At ODSC, learn essential quantitative modelling frameworks as<br />

well as how to use machine learning and reinforcement techniques<br />

to invest, trade, and manage risk in finance. Topics covered at this<br />

event include : Recommendation System for Trading, Machine<br />

Learning for <strong>Algo</strong>rithmic Trading, Deep Reinforcement Learning for<br />

Quant Trading, Streaming Analytics for Quant Trading<br />

Alternative Data For Quant Trading, Sentiment Analysis for Quant<br />

Trading, Artificial Intelligence in Finance and more…<br />

EDUCATION & TRAINING<br />

https://odsc.com/europe/ml-for-finance/<br />

Qube<strong>Algo</strong><br />

Qube<strong>Algo</strong> offers a<br />

unified ecosystem for rapid<br />

and robust development of<br />

sophisticated automated<br />

trading models, algorithms, and<br />

strategies. It has been live with<br />

a select group of clients and<br />

is now being rolled out to the<br />

broader market, providing highly<br />

customizable solutions that<br />

empower quants and developers<br />

to build bespoke, multi-asset,<br />

electronic trading applications,<br />

often with minimal code.<br />

https://www.qubealgo.com/<br />

WEBSITE OF THE MONTH<br />

Finance Hive FICC<br />

Mexico Members<br />

Meeting<br />

13th June, <strong>2023</strong><br />

Discussion topics will focus around and analysis<br />

of recent EM vs Mexico drivers and current market<br />

conditions, how to get real time pre-trading transaction<br />

cost analysis (TCA) to change and improve the<br />

outcomes of your trading strategy, determining the<br />

benefits of aligning with the Global Code of Conduct,<br />

how the <strong>FX</strong> algo space is developing elsewhere in the<br />

world and assessing the current level of algos usage in<br />

Mexico across the desks and much more.<br />

FOR THE DIARY<br />

https://www.thehive-network.com/finance-events/the-finance-hive-live-meeting-mexico-members-meeting-13-june-<strong>2023</strong>/<br />

<strong>May</strong> <strong>2023</strong><br />

15


CASE STUDY<br />

Deutsche Bank uses<br />

Tradefeedr’s <strong>FX</strong> algo<br />

forecasting tools<br />

to enhance client<br />

engagement<br />

Earlier this year Tradefeedr unveiled its <strong>FX</strong> algo forecasting service, a collection of<br />

analytics tools developed to help clients make a more informed choice when selecting<br />

the most appropriate algo for their execution needs. Since then, leading algo providers<br />

have identified its potential in enhancing client discussions around the performance of<br />

their own algo suites. Tim Cartledge, Chief Data Officer at Tradefeedr, and Vittorio Nuti,<br />

Global Head <strong>FX</strong> <strong>Algo</strong>s at Deutsche Bank, explain the benefits.<br />

TIM CARTLEDGE, CHIEF DATA OFFICER AT<br />

TRADEFEEDR:<br />

Following extensive consultation with the industry, our unique<br />

algo forecasting suite is now live and in use by both algo clients<br />

and providers alike. We have developed two products which use<br />

our forecasting tool, one for pre-trade forecasting and the other<br />

for post-trade. Both differ considerably from existing pre- and<br />

post-trade analysis tools in their ability to compare for the first<br />

time individual algo performance against the market average.<br />

Our product suite is available as an API and compares actual<br />

trades and shows a forecast of what the performance would<br />

have been for the currency pair at that time and in those same<br />

market conditions. This enables algo users to compare their<br />

execution performance to see if it was in line with the rest of<br />

the industry, or if their performance was better or worse, faster<br />

or slower, more risk or less.<br />

On the pre-trade side, algo users are also able to forecast<br />

what the expected performance would be for their execution.<br />

It can look at a particular currency pair in a particular size<br />

and compare the expected performance for different levels of<br />

aggression for the algo. That’s available now as an API and we<br />

will also be releasing an interactive front end in the coming<br />

weeks.<br />

The tool which has already been live for some time - and which<br />

Deutsche Bank has been working with - is the post-trade tool.<br />

This allows you to look at your actual trades and compare<br />

execution performance with the average expectation for<br />

those same conditions. We produce three headline numbers:<br />

Tradefeedr Global Forecast, which is based on all of the algo<br />

trades in our database; Tradefeedr Fast, which is based on the<br />

fastest third of algos in the market, and Tradefeedr slow, which<br />

is based on the slowest third of algos in the market.<br />

Tim Cartledge<br />

What is proving valuable to algo providers, such as Deutsche<br />

Bank, is that the tool can also forecast the performance of<br />

individual bank algos as well. Using the same methodology,<br />

the forecast can look at the expected performance of individual<br />

algos and compare those to the market. What we are doing<br />

is complimentary to the work of the banks, who can now<br />

combine the insights from our forecasting tools and provide<br />

16 <strong>May</strong> <strong>2023</strong>


additional background around the algos, flows, market<br />

conditions etc to help clients achieve their execution goals.<br />

The really novel element to our service is the standardisation<br />

of the algo executions. Not all algo executions are the same<br />

depending on currency pair, time of execution, market volatility<br />

etc. This is really the first time clients are able to compare algos<br />

using a level playing field to show what the algo execution<br />

would look like indifferent conditions. It takes the noise out of<br />

the process and allows for really fair comparisons.<br />

The modelling we do is based on the liquidity seeking algos<br />

in the market, which are allowed to run at the natural speed<br />

of the market. Even if you’re manually trading, you now have<br />

this database which shows what the market response was<br />

when the liquidity seeking algos tried to do a certain amount<br />

in a currency pair and you now also know how long it took<br />

them. That’s really useful information about market impact and<br />

fundamental market behaviour, which feeds into the <strong>FX</strong> sales<br />

teams and the advisory work they do with their clients. We can<br />

also see that most clients who use a certain algo are having<br />

good experience with that algo, but they will still need to be in<br />

communication with the algo provider to get the best out of<br />

their performance.<br />

VITTORIO NUTI, GLOBAL HEAD <strong>FX</strong> ALGOS AT<br />

DEUTSCHE BANK:<br />

We are very excited by the Tradefeedr algo forecasting tools<br />

as they offer a more pragmatic and quantitative approach to<br />

analysing algo performance. The tools provide clients with a<br />

holistic view of the algo market, they allow clients to view<br />

different algos and compare their strengths and weaknesses.<br />

Here on the <strong>FX</strong> team, we very much appreciate what the<br />

tools have to offer as they have made it much easier to talk<br />

about the algo performance data with our clients. We are<br />

able to use the data to help our clients to try and improve<br />

their execution performance based on real results. Having it<br />

based on independent data allows the client to compare the<br />

performance of an algo, instead of having to judge based only<br />

on their experience of what worked well or not. At Deutsche<br />

Bank, we are able to demonstrate to clients using this service<br />

that our algos perform well above the global average, both in<br />

terms of speed and lower cost of execution. The tools are based<br />

on independent analysis, so this provides us with an objective<br />

way to discuss the performance of an algo with clients as they<br />

are able to compare the execution to the rest of the market.<br />

The algo data in Tradefeedr represents our standardised settings.<br />

Once clients can see that performance, we can also discuss<br />

how these can be customised to their execution targets, such as<br />

more passive or slower execution versus our standard set.<br />

Tradefeedr have collected the data and aggregated it into two<br />

matrices, which simplifies this very complex data in a way that<br />

the user can readily understand. This quantitative approach has<br />

resulted in a very robust, powerful tool for algo users which will<br />

add to their understanding of algos and further enhance their<br />

execution performance.<br />

Vittorio Nuti<br />

<strong>May</strong> <strong>2023</strong><br />

17


TRADERS WORKSHOP<br />

To limit or<br />

not to limit?<br />

Using price limits with execution algorithms has been a somewhat contentious<br />

subject. While limits can be used for risk management purposes or substantiated<br />

by market insight on the execution horizon, they may also add noise or a suboptimal<br />

variable to an optimized execution logic. Sait Ozturk and Yangling Li of<br />

BestX have recently shed more light on this topic by examining the practical impact<br />

of using limits in conjunction with a range of different execution algorithms.<br />

• Opportunistic: algos that do not<br />

have a strict benchmark-dictated<br />

schedule, so have the flexibility to<br />

execute aggressively or passively<br />

according to market conditions.<br />

Sait Ozturk<br />

In our paper “Do Limits Improve <strong>Algo</strong><br />

Performance?”, we explored how setting<br />

a limit affects execution performance and<br />

whether it was possible to optimize limit<br />

placement. The focus of the paper was<br />

on algo trades executed in less than 24<br />

hours and for G10 currency pairs, which<br />

avoided data issues without sacrificing<br />

significant statistical power. The bulk of<br />

the research concentrated on algos with<br />

only one limit, where limit placement<br />

is a more straightforward matter than<br />

dynamic limit optimisation.<br />

Yangling Li<br />

amount of risk executed as quickly<br />

and efficiently as possible and less on<br />

minimising market impact.<br />

• Pegged: algos that aim to execute<br />

orders at levels within the prevailing<br />

best bid/offer. Passive by nature, since<br />

they follow the market and usually<br />

trade at no worse than mid-price.<br />

• Interval: interval-based algos, such as<br />

TWAP, which aim to minimise slippage<br />

to a benchmark where the algo slices<br />

the parent notional according to an<br />

interval- or time-based schedule.<br />

BestX® Expected Risk Transfer Cost 1<br />

was used as the primary benchmark<br />

to evaluate algo performance with and<br />

without limits. This takes into account<br />

market conditions at the time of<br />

execution to arrive at a fair price estimate<br />

of the trade when algo execution begins<br />

and can be compared with the achieved<br />

average price across algo fills.<br />

Our research focused on five algo styles 2 :<br />

• Get Done: aggressive algos, where<br />

the priority is on getting a specific<br />

18 <strong>May</strong> <strong>2023</strong>


• Volume: volume-based algos, such<br />

VWAP, which aim to execute in line with<br />

a specified percentage of volume traded<br />

within the market.<br />

PRELIMINARY ANALYSIS<br />

As a robustness check, we used the<br />

performance of a TWAP algo without a limit<br />

during the execution period as a secondary<br />

benchmark, which more strongly favours<br />

having a limit than the primary analysis.<br />

However, this out-performance of limits may<br />

be quasi-mechanical, because a TWAP with a<br />

limit will on average outperform by avoiding<br />

prices when the limit is engaged and the algo<br />

is stopped. On the downside, the principal<br />

ways a TWAP with a limit can underperform<br />

are by being unable to execute the intended<br />

trade amount fully or by completing it too<br />

late.<br />

This TWAP under/over performance is<br />

important in the context of finding an<br />

optimal limit value, because having a very<br />

tight limit may produce the illusion of outperformance<br />

purely by executing only part<br />

of the intended amount and only under very<br />

favourable market conditions. To control for<br />

this, we created a smaller data set composed<br />

of mostly clean intended execution amount<br />

data to double-check the results inferred<br />

from our larger data set, while only including<br />

fully executed trades.<br />

Figure 1 (opposite) compares the Risk Transfer<br />

Performance distributions of trades with and<br />

without a limit. The limit is clearly already<br />

improving the median performance 3 in both<br />

data sets.<br />

Price limit placement is just a binary decision,<br />

so we also needed to evaluate where the<br />

limit should be set to assess its utility. We<br />

concentrated on trades with only one limit in<br />

their lifetimes, partly for simplicity and also<br />

because these constituted the vast majority of<br />

trades in our data set.<br />

Figure 2 shows how different limit distances<br />

affect algo performance. We can see a major<br />

difference between the datasets: the large<br />

dataset has numerous outperforming trades,<br />

while there are far fewer trades with negative<br />

distance in the smaller dataset.<br />

Figure 3 shows trade performance by limit<br />

engagement, with negative limit distance<br />

cases combined under Limit Started Engaged,<br />

while the positive limit distance cannot<br />

engage during the algo execution period<br />

(Limit Not Engaged) or engage after the<br />

<strong>May</strong> <strong>2023</strong><br />

19


TRADERS WORKSHOP<br />

start (Limit Engaged). Much of the<br />

limit performance derives from trades<br />

with an engaged limit at start, while<br />

trades with limit engaged later mostly<br />

underperform those where the limit<br />

never engages or limitless trades.<br />

Figures 4 and 5 show the other side of<br />

the trade-off: order completion, over<br />

limit distance from market arrival midprice.<br />

STATISTICAL ANALYSIS<br />

Our starting point for statistical analysis<br />

of limits was a regression model for<br />

the effectiveness of using only single<br />

limits with a positive distance. We then<br />

extended this model to accommodate<br />

dynamic limits by adding variables for<br />

trades with multiple limits during their<br />

lifespan.<br />

We found that for both the large and<br />

small data sets a single limit:<br />

• Degraded the performance of<br />

Opportunistic algos for almost any<br />

positive limit distance.<br />

• Improved Interval and Pegged algos<br />

with statistical significance<br />

• Had a directly diverging effect for<br />

each data set for GetDone algos<br />

• Improved Volume algo performance<br />

for the larger data set, but had<br />

no statistically significance for the<br />

smaller data set.<br />

The results for one limit do not change<br />

qualitatively when indicator variables<br />

for multiple limits are added. Overall,<br />

having multiple limits over the lifetime<br />

of a trade is associated with statistically<br />

significant performance deterioration<br />

across all algos, except for Interval algos.<br />

Figures 6 and 7 display the estimation<br />

results for the large and small data<br />

sets together for limit distance. The<br />

dashed 95% confidence line indicates<br />

whether the solid limit performance<br />

line is statistically significantly different<br />

from the zero line representing the<br />

performance without an algo limit.<br />

Examination of Figures 6 and 7 reveals that:<br />

• The conflicting results for GetDone<br />

algos between the two data sets<br />

20 <strong>May</strong> <strong>2023</strong>


also applied when performance was broken down<br />

by limit distances. While the limit performance<br />

line is below the zero-line for the large data set,<br />

in the small data set tighter limits are statistically<br />

significant above the line when up to 5% of the<br />

daily volatility away from the limit. Any limit farther<br />

away than this fails to improve the performance.<br />

• Also in line with the previous results, the limit<br />

performance line is below the zero-line for almost<br />

any positive limit distance for Opportunistic algos.<br />

The one exception, proving the rule, is the very<br />

tight 1% of daily volatility away from the arrival<br />

market mid, which comes with order completion<br />

issues.<br />

• A tight limit can boost the performance of Pegged<br />

algos with a limit placed up to 15% of daily<br />

volatility away from the market mid-price still<br />

improving the results.<br />

• A number of limit distances enabled performance<br />

improvements for Interval algos. The limit<br />

performance estimates are almost always above the<br />

zero-line, but mostly not statistically significantly<br />

above. A relatively tight distance of 2.5%-5% as<br />

well as having a limit >50% of daily volatility away<br />

(up to 4 times the daily volatility) can improve the<br />

limit.<br />

• Volume algos exhibit a similar, but weaker structure<br />

to Interval algos, with out-performance from a limit<br />

10%-15% and 1.5-2 times daily volatility away<br />

from market mid. Both these algos aim to follow<br />

the market movements without too many attached<br />

smart features, so these relatively distant limits may<br />

be exploiting mean reversion in the market to avoid<br />

trading in transitorily bad market conditions.<br />

CONCLUSION<br />

Our original paper analysed whether setting a limit<br />

aids algo performance and where to set a limit to<br />

optimize any beneficial effect. We explored the<br />

performance/completion trade-off, where a tighter<br />

limit causes better performance at the expense of<br />

only partial execution of the intended trade amount.<br />

We found that setting limits improved Interval<br />

and Volume algo performance, particularly when<br />

distant from the market mid, thereby avoiding<br />

extreme unfavourable market moves. The evidence<br />

for GetDone algos was mixed, but tight limits were<br />

favoured, while taking the partial completion risks<br />

into consideration. It was challenging to improve<br />

Opportunistic algos, which lie on the smarter end of<br />

the spectrum investigated, by using limits.<br />

1 . https://www.bestx.co.uk/glossary<br />

2 . Fixing algos, where the aim is to match the relevant fix price around the official fixing<br />

window and outperforming a fair value price is less relevant, were excluded.<br />

3 . Risk Transfer Performance is defined as the performance of the algo relative to the BestX®<br />

Expected Risk Transfer Cost.<br />

<strong>May</strong> <strong>2023</strong><br />

21


Subscribers receive<br />

12<br />

4 printed copies of the publication each year together with unlimited access<br />

to the registered section of the <strong>FX</strong><strong>Algo</strong><strong>News</strong> website (www.fxalgonews.com)<br />

1 Year Year (4(12 editions) editions) at at £150 £200 or £175 or £275 Rest Rest of World of World<br />

2 Years (8(24 editions) at at £210 £375<br />

or £240 or £500 Rest Rest of World of World<br />

SJB Media Ltd<br />

SJB Media Ltd., Suite 153, 3 Edgar Buildings, George St., Bath, BA1 2FJ, UK<br />

References<br />

Richard S. Sutton and Andrew G. Barto, 2018, Reinforcement Learning: An Introduction, Second<br />

Edition<br />

Cameron Davidson-Pilon, “Bayesian Methods for Hackers: Probabilistic Programming and Bayesian<br />

Inference “, 2016, Addison-Wesley Data & Analytics<br />

Merton, Robert, 1980, On estimating the expected return on the market: An exploratory investigation.<br />

Journal of Financial Economics, Volume 8, Issue 4, December 1980, Pages 323-361<br />

Junya Honda and Akimichi Takemura, Optimality of Thompson Sampling for Gaussian Bandits<br />

Depends on Priors, Proceedings of Machine Learning Research, Volume 33: Artificial Intelligence<br />

and Statistics, 22-25 April 2014<br />

Tze Leung Lai and Herbert Robbins. Asymptotically efficient adaptive allocation rules. Advances in<br />

applied mathematics, 6(1):4–22, 1985<br />

Daniel Kahneman, 2011, Thinking, Fast and Slow, Penguin<br />

Please e-mail this form back to:<br />

Charles.Jago@fxalgonews.com<br />

*On receipt of payment you will be sent a Username and Password for the registered section of the <strong>FX</strong><strong>Algo</strong><strong>News</strong> website<br />

148 |<br />

22<br />

november 2018 <strong>May</strong> <strong>2023</strong> e-FOREX


BOOK OF THE MONTH<br />

Advanced <strong>Algo</strong>rithmic Trading<br />

<strong>Algo</strong> Trading with ChatGPT<br />

BLOG OF THE MONTH<br />

Advanced <strong>Algo</strong>rithmic Trading provides real world application<br />

of time series analysis, statistical machine learning and<br />

Bayesian statistics, to directly produce profitable trading<br />

strategies with freely available open source software.<br />

quantstart.com/advanced-algorithmic-trading-ebook/<br />

This blog explores using the power of OpenAI’s ChatGPT<br />

model in algorithmic trading and discusses the various<br />

steps that are involved.<br />

blog.quantinsti.com/algorithmic-trading-chatgpt/<br />

J.P. MORGAN E-TRADING SURVEY <strong>2023</strong><br />

tradertv.net/j-p-morgan-e-trading-survey-<strong>2023</strong>-how-will-trading-desks-beimpacted<br />

ANZ FOREIGN EXCHANGE EXECUTION ALGORITHMS<br />

youtube.com/watch?v=25WN6q2wIo8&t=1s<br />

VIDEO VAULT<br />

Charles Jago<br />

Editor<br />

charles.Jago@fxalgonews.com<br />

+44 1736 74 11 44<br />

Nicola Tavendale<br />

<strong>News</strong> editor<br />

nicola@ntavendale.com<br />

+44 1736 74 11 44<br />

Susan Rennie<br />

Managing Editor<br />

susie.rennie@fxalgonews.com<br />

+44 1208 821 802<br />

Charles Harris<br />

Advertising sales<br />

charles.harris@fxalgonews.com<br />

+44 1736 74 11 44<br />

David Fielder<br />

Subscriptions manager<br />

david.fielder@fxalgonews.com<br />

+44 1736 74 11 44<br />

Tim Hendy<br />

Digital & Web services<br />

tim@thstudio.co.uk<br />

+ 44 1209 217168<br />

Matt Sanwell<br />

Design & Origination<br />

matt@designunltd.co.uk<br />

+44 7515 355960<br />

Larry Levy<br />

Photographry<br />

larrydlevy@gmail.com<br />

Michael Best<br />

Events manager<br />

michael.best@fxalgonews.com<br />

+44 1736 74 11 44<br />

SJB Media Ltd<br />

Suite 153, 3 Edgar Buildings<br />

George Street, Bath, BA1 2FJ<br />

United Kingdom<br />

Tel: + 44 (0)1208 82 18 02 (switchboard)<br />

Tel: + 44 (0)1736 74 11 44 (Sales & editorial)<br />

Fax: + 44 (0)1208 82 18 03<br />

Printed by Headland Printers<br />

Published quarterly. ISSN 2056-9750<br />

Although every effort has been made to ensure the accuracy of the information contained in this publication the publishers can accept no liabilities for inaccuracies that may<br />

appear. The views expressed in this publication are not necessarily those of the publisher. Please note, the publishers do not endorse or recommend any specific website featured<br />

in this newsletter. Readers are advised to check carefully that any website offering a specific <strong>FX</strong> trading product and service complies with all required regulatory conditions and<br />

obligations. The entire contents of <strong>FX</strong>ALGONEWS are protected by copyright and all rights are reserved.<br />

<strong>May</strong> <strong>2023</strong><br />

23


Reimagining the power<br />

of <strong>FX</strong> <strong>Algo</strong>s<br />

UBS <strong>FX</strong> <strong>Algo</strong>rithms help our clients reduce market impact, improve<br />

performance and add resilience to their trading workflow through:<br />

• Sophisticated Smart Order Router<br />

• Comprehensive liquidity access including UBS internalization<br />

• Advanced machine-learning framework<br />

• Robust strategies from liquidity seeking to passive execution<br />

Find out more, search UBS <strong>FX</strong> <strong>Algo</strong><br />

For Professional and Eligible Counterparties / Institutional / Accredited Investors only.<br />

The value of investments may fall as well as rise and you may not get back the amount originally invested. As a firm providing wealth management services to clients, UBS<br />

Financial Services Inc. offers investment advisory services in its capacity as an SEC-registered investment adviser and brokerage services in its capacity as an SEC-registered<br />

broker-dealer. Investment advisory services and brokerage services are separate and distinct, differ in material ways and are governed by different laws and separate<br />

arrangements. It is important that clients understand the ways in which we conduct business, that they carefully read the agreements and disclosures that we provide to them<br />

about the products or services we offer. For more information, please review the PDF document at ubs.com/relationshipsummary. © UBS 2022. These materials are provided<br />

solely for informational purposes. For further important country specific information visit: ubs.com/disclaimer. All rights reserved. UBS Financial Services Inc. is a subsidiary of<br />

UBS AG. Member FINRA/SIPC.<br />

24 <strong>May</strong> <strong>2023</strong>

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!