What’s spurring the rapidly evolving electronic trading landscape and what are the implications of this growing demand for automated trading solutions?
Electronic trading is like Amazon, or the iPhone, or the internal combustion engine: it’s the story of technology writ large, applied to trading.
Like all technology, electronic trading represents a phase shift in productivity. Just as the personal computer compressed work timelines from days to minutes, electronic trading vastly improves the efficiency of trading workflows. What previously required a phone call with humans on each side can now be done instantly and without human intervention. This elimination of friction in turn unlocks all sorts of otherwise-uneconomical market activity. And in a competitive market, once this paradigm shift occurs, all firms are compelled to get on board to keep up. The automation train only goes in one direction.
The implications of this transformation are widespread.
For dealers and asset managers, technology becomes a core business requirement, and market share tends to accrue to firms with the most efficient and scalable infrastructure. This tends to lead to consolidation, with less scaled players either acquired by larger firms or competed out of existence. This played out, for example, in equities in the 1990s/2000s, where flow used to diffuse among dozens of smaller dealers, but ended up concentrated primarily among a few large tech-driven liquidity providers.
Greater efficiency also unlocks new business models. For example, the growth of the SMA (separately managed account) industry in the municipal market over the past decade was fuelled in large part by better technology, which allowed scaled providers to efficiently manage tens of thousands of accounts and millions of individual bond holdings. This in turn allowed SMA providers to lower account minimums, expanding accessibility of the asset class and fuelling AUM growth.
At the market level, electronic trading inevitably leads to a reduction in transaction costs, be it in the form of bid/ask spreads or explicit transactional fees. And as trading gets cheaper, you get more of it, leading to an increase in transaction volumes. This increase is often concentrated in smaller increments, leading to a reduction in average transaction size.
Ultimately, the main beneficiaries of electronic trading are investors, who gain better access to a larger selection of investment products at lower cost.
How important is data when it comes to increased automated trading? What is the key thing that needs to be considered?
Data is the lifeblood of any automated trading business. But this should not be surprising, because the same could be said for non-automated trading as well.
When a human is deciding how to make a trading decision, he or she relies on prior experiences and information at hand. Where was the last trade in this instrument? Was there any recent news? Is anything interesting happening in related tickers? How much can I generally charge for a trade in an instrument like this? These questions are answered, in one way or another, by data. The trader probably has a Bloomberg terminal with streaming order book data and a news feed. She also has a database, of sorts, in the form of prior experiences and memories.
Automated trading systems operate similarly. They synthesise various forms of live and historical data into a series of trading decisions. The difference, of course, is that automated systems can utilise much larger quantities of data at much higher precision, and can apply quantitative techniques to such data near-instantaneously. But the general role of data as an input to decision-making is similar.
The biggest mistake firms make during the transition to automated trading is simply throwing data away. People have limited processing capacity, so firms used to human-driven workflows are often sloppy about persisting the large quantities of valuable data flowing through their businesses. This failure mode is particularly pernicious because once data is gone, it is often only recoverable in real-time. If you need a year’s worth of training data for a new model, and you’ve thrown your prior data away, guess what? You now have to wait a year until you’ve rebuilt that dataset. Not fun.
What are the most impactful changes AI is making on electronic trading, and markets in a wider sense?
First, some terminology. When people say “AI” nowadays, they’re generally referring to a specific type of technology known as a large language model, or LLM. What’s an LLM? Basically, think: ChatGPT.
LLMs are an important development, to be sure. However, they are just one part of a much broader ecosystem of techniques collectively known as machine learning. Machine learning contains various forms of statistical techniques to understand data. The boundary between machine learning and 9th grade algebra is somewhat fuzzy – for example, is linear regression machine learning? But generally the term is used to refer to more complicated techniques, such as neural networks and random forests.
I highlight this difference between classical machine learning and LLMs because the impact each has had on electronic trading to this point is quite different.
Classical machine learning techniques have been used in electronic trading for decades. They form the foundational building blocks of many trading algorithms. These techniques appeared first in more liquid asset classes, like equities and futures, but in recent years have proliferated in fixed income as well. For example, since bonds often trade only a few times per day (or less), it can be difficult to estimate the “current” price of a bond. Machine learning techniques such as Kalman filters have been applied to this problem for years.
By contrast, electronic trading use cases for AI/LLMs are in relative infancy. Applying LLMs to trading is less straightforward than for classical machine learning, and the reason is in the name: LLMs are about language, whereas automated trading is about math. ChatGPT can write a pretty convincing rap in the linguistic style of Benjamin Franklin, but it’s not yet great at predicting the price of the next bond trade. Direct applications to trading algorithms are therefore still limited.
The caveat is that all of this is changing rapidly. I anticipate more direct applications to trading strategies over time.
And widening the aperture a bit, AI is beginning to have the same impact in trading as in every other industry: as a major productivity enhancer. For example, copilot-style tooling is increasing the throughput of the software developers who write the code behind trading algorithms. I expect this form of impact to grow significantly over time.
Looking ahead to the rest of 2025, what industry developments/market structure changes are you most conscious of?
The main theme that comes to mind is: we’re so much closer to the beginning of this journey than the end.
We’ve come a long way, to be sure. In investment-grade credit, for example, electronic trading has grown from less than 10% to north of 50% market share over the past decade. High-yield credit is around 25%, and on a similar trend line.
But this is just the start. There are many fixed income markets where electronification is just getting started. Municipal bonds, mortgage specified pools, loans, and others are still voice-dominated markets, with electronic volumes below 20%. Will these markets follow the precise path of credit or equities? No. But the direction of travel is clear.
And electronification of trading volumes is just the first step. When execution is cheap and instant, it unlocks forms of market activity that would be otherwise infeasible. For example, in credit we’ve seen the rise of portfolio trading, where hundreds or thousands of bonds are traded simultaneously as a single package. This kind of workflow could not exist without automation. I expect we will continue to see new forms of market activity like this, built on the back of electronic trading workflows. For example, OpenYield (in which, disclosure: my employer, TD Bank, is an investor) is building innovative trading protocols to create a more equity-like experience for fixed income investors.
I am also watching areas of the market which have not yet garnered as much attention. For example, electronic trading is mostly discussed in the context of secondary markets, but not as much in relation to primary markets. The process of debt issuance hasn’t changed much in 25 years. Timelines are long, processes are manual, and underwriting fees haven’t budged. These conditions are ideal for the emergence of a more automated solution. I would not be surprised to see a push into this space in the coming years.