Stephen Totten, Director of Quantitative Analysis, oneZero
AI is set to transform many industries and its impact on the FX markets will be particularly dramatic. Staying ahead of the curve for institutional FX traders will require a combination of cutting-edge technology, adaptable industry experts and a deep understanding of market dynamics.
I joined oneZero as Director of Quantitative Analysis at the start of 2023 after years in banking and investing as the shift toward e-FX took place and made currency trading the most electronified of all the major financial markets – as well as the biggest market in the world by trading volume
In my former role as global head of e-FX at UniCredit, and before that as a portfolio manager at Citadel, I witnessed firsthand the transformative power of data.
Having been a consumer of FX data solutions in past roles I have an insight into what is needed – and what is possible – and believe that electronic trading is about to be turbocharged by the impact of AI.
In this first in a series of blogs about AI and ways to harness the power of FX data I will look at how oneZero is using AI to develop classification tools that help market makers to focus on optimising the service they provide to their clients.
Future posts will examine oneZero’s FX simulation tools and other potential benefits from adoption of AI machine learning. I will also explain how improvements to functionality that are due for rollout by oneZero in the second half of 2023 can be used by clients to harness the evolution in AI tools.
Utilising data and analytics efficiently has already become fundamental to boosting revenues, volumes and customer satisfaction.
AI will take this to a new level – and will require adaptation on the part of market participants to ensure that they do not fall behind in a data utilisation race with their competitors.
Volumes vs data
The most comprehensive source of FX trading volume is the triennial survey conducted by central banking group the Bank for International Settlements (BIS).
When its last survey was released in October 2022, the BIS found that global FX trading volumes had hit a new record of $7.5 trillion per day, based on its monitoring of deals conducted in April 2022.
The graph below shows the growth in FX volumes to new highs, as well as the increasing significance of FX swaps, forwards and options relative to spot trades.
Source: BIS Triennial 2022 Central Bank Survey
More recently, in July trading venues including the CBOE and Eurex reported increases in their FX volumes for June on both a monthly and year-on-year basis, and our own transaction statistics back this trend.
But the data generated by FX market activity is increasing at a far faster rate than actual deal volumes.
Macroeconomic conditions and inflation in the past 18 months have led to increased volatility, and new and upgraded participants in the FX space mean that data rates are higher than ever. At oneZero, we see 2.5 times the market data we handled just two years ago, for example. We hear from other organisations that some legacy systems are struggling to keep up with the sheer volume of data, which makes having a reliable technology partner more important than ever.
At oneZero, in the first six months of 2023 alone we successfully captured and analysed over one billion orders, in a testament to our robust system capacity.
Much of this activity is not captured in the BIS triennial survey of trading and oneZero is able to add value with analysis and interpretation of the significant amount of data that is uniquely available from within our own client EcoSystem.
This scale makes oneZero uniquely placed in the market to create and test sophisticated pricing, execution and flow analysis models, some of which I will discuss below.
Capture and Storage
Having a robust data capture pipeline is fundamental in a modern electronic trading business, and requires a significant amount of infrastructure, tooling and monitoring to ensure the system is working reliably, and is able to scale.
It is at times of high market volatility that market makers have the highest volumes and trading revenues, but also the largest risks.
From a data point of view, these are also the periods with the largest bursts of orders and market data, and ensuring your data pipeline can capture and handle the flows during these periods is critical to performing robust post-trade analysis, understanding where the system worked well, and also what can be improved for the future.
It is often said that 80% of time in data science is spent sourcing and cleaning data, and only 20% on modelling, and the engineering effort that oneZero has spent in building a resilient architecture for capturing huge amounts of data is fundamental to our ability to build analytics for our clients.
This is not an effort that we have launched recently in response to the current market fascination with the potential of AI – at oneZero we have a 14-year track record in building and enhancing our data capture pipeline.
This has placed us in a strong position to deploy new AI machine-learning tools to help our clients to improve their own business performance.
From data to information to knowledge
At oneZero, AI is deployed to analyse client flow and markouts. Within larger institutions, reviewing this information is common to building a picture of the value of a client to the franchise. Our Insights packages have a number of reports to present this information to staff sitting on our clients’ trading desks to allow them to decide how to tier or risk manage their customers.
Our goal at oneZero is to use our deep knowledge and experience of the industry, combined with the large data set and information tools to help our clients optimise their franchise.
One example of this is our focus on “negative inceptions”, i.e. trades accepted at a price worse than market mid, and thus immediately loss-making.
Our Insights reports focus on helping our clients discover and eliminate this particular set of problematic trades, offering an actionable way to immediately increase revenues.
AI in action
The advances of AI have ushered in a new era of possibilities in the FX industry. AI algorithms have the ability to analyse vast amounts of data, uncover hidden patterns, and identify valuable relationships in sophisticated ways. AI techniques are evolving fast, with development of clustering models one of the areas where we are seeing the biggest early benefits for our clients.
At oneZero, AI is deployed to analyse the client flow and markout information, categorising each trade in a multi-factor model.
This analysis allows for a significantly more scalable and efficient understanding of clients’ trading styles and preferences, with the ability to segment the information in a number of ways, ensuring that they are priced and risk-managed optimally. Our models allow a desk to scale their clients into the thousands or tens of thousands, automatically reviewing all flow in the same way that a trader would.
In order to effectively train this model with sufficient statistical power, a large set of trades is required. At oneZero, firms of all sizes can benefit from our vast dataset, and by leveraging our proven ability to capture a wide array of data elements in a fast-moving, dynamic EcoSystem.
Firms understandably do not want to give data on their entire trading history to their liquidity provider to run analysis, but as a neutral third party we supply metrics that can be quickly deployed to improve performance – and client/provider relationships.
Classification of client and flow profiles
Once a client’s flow is classified, the oneZero Hub allows market makers to optimise pricing, trade acceptance logic and risk management of each currency pair based on the expected outcome.
All classifications are calculated on a trade-by-trade basis at the currency pair level and then rolled up to the client, streaming, and hedging model.
Our engine also produces a conviction level for the clustering results, so as the engine gathers more and more trade data from a new client relationship, the confidence of the classification will increase. Conversely, if a client adapts its trading style over time, one would also observe a gradual change in conviction as our model begins to categorise the flow differently.
Classifications are calculated in rolling monthly windows. Our Data Source Insights package highlights changes in classification over time, so that market makers can focus their time and effort on optimising service to clients where their behaviour has changed.
Simulation tools and AI
Simulation and backtesting is another area where having large, accurate historical data sets is a significant advantage.
In the next oneZero AI blog, I will examine simulation tools such as oneZero’s Maker Pool Replay, along with other ways to harness the power of historical market data and leverage AI algorithms to optimise trading strategies, foster stronger relationships and unlock new opportunities.
Learn more on this subject by watching an interview between Stephen Totten and Colin Lambert at The Full FX here.