First mover advantage in a wave of change can be enormous, as LTX has demonstrated with BondGPT
Bond trading platform LTX has launched into an intensely competitive market, but its rapid response to changing technology has allowed it to make massive splash already. It uses a large language model (LLM), BondGPT, to let traders and portfolio managers find which bonds have a set of desirable characteristics, across a massive universe of instruments.
BondGPT’s chat function connects with LTX’s Liquidity Cloud and bond similarity technology so users can query US corporate bonds on the LTX trading platform based upon the user’s criteria. This solves for a major problem in bond investing, which is how to find the right instrument to buy or sell using often non-standard information across a fragmented market to form a coherent picture of market activity, including currently liquidity conditions.
“The response has been really nothing short of extraordinary,” says Jim Kwiatkowski, CEO at LTX. “In any electronic trading business, you get people to the point that they ask for a contract, and a lot of time can go by after that point. Contracts just started coming back. Significant financial institutions that we’d love to do business with, but historically weren’t able to provide quick responses were suddenly calling us.”
The firm’s waiting list quickly built to hundreds of people including many outside of its core markets of US corporate credit, with European traders noting that the tool had a clear application, potentially cutting workload considerably and providing the impetus to win this award in the 2023 EMCAs.
“I think we really hit some key points for people by making LLM technology accessible and relevant to the bond market,” says Kwiatkowski. “No one else had done that. We’re in the market only 60 days, although it seems like longer because we’re implementing enhancements so quickly. We’ve actioned lots of client feedback that has driven more usability.”
Even before that feedback was included, there were issues that LTX needed to address in order to support the use of an LLM in a highly challenging and risk averse capital markets environment.
“The major concerns about other LLM implementations came down to security concerns and hallucination concerns,” he explains. “A GPT is effectively programmed to satisfy the end user. That can be unacceptable behaviour in the financial markets. To address this, firstly, we only included the highest quality sources of data, with calculations that we ourselves were doing. Alongside the answer, the question itself is echoed back to ensure the user can see exactly how BondGPT interpreted their question. Then we took one further step because we’re a broker dealer, so we built an adversarial compliance rules engine, which actively monitors BondGPT responses and evaluates them based on a number of compliance rules, including avoiding responses that contain investment advice.”
At the base of each answer is a check on subtly different questions so the user can check their query has not been misinterpreted based on the highest probability of the natural language search, and links to allow the alternative questions to be asked. “We tried to think a couple of steps ahead in terms of usability,” says Kwiatkowski.
Data protection is also key. “Information security is a paramount concern for LTX – all of its services including BondGPT are provided within the highly secure and audited Broadridge infrastructure.”
Being part of post-trade giant, Broadridge, has provided LTX with enormous resources including decades of fixed income expertise. The LTX platform has used AI since the firm’s inception, allowing LTX to take the lead on its application in capital markets.
“LTX has been using AI in a number of different ways from its foundation,” he notes. “Our Liquidity Cloud, which is really central to our value proposition, is a neural network. The Cloud Match Scores which assess real-time contra-interest for bonds is a product of machine learning. So from the earliest days of LTX , it’s been providing valuable and unique pre-trade signals to users based on AI.”
The roadmap for LTX continues to expand these developments in AI, to develop BondGPT and the firm’s wider offering.
“Our roadmap is based on client feedback,” Kwiatkowski says. “Our clients are fantastic in that they really tell us what they’re not able to get elsewhere in the market. With BondGPT, people embraced the immediate access to high quality data, with answers you can trust and are fully compliant. What we needed to do on top of OpenAI was teach the system a lot of bond market jargon. It’s a huge starting point but it’s not the finished product. What we’ve delivered so far is time saving access to inisghts. Clients are accustomed to a workflow where they gather raw data from platforms that are great at delivering raw data. But that wasn’t what they were looking for. They were looking for the insight.”
Using natural language questions to get that insight has clearly been compelling for end users as it also reduced time to use the new tool and allows a natural process of interrogation.
“What may have taken people five minutes to answer takes 20 seconds, and once we crossed that threshold, people really started to get excited,” he said.
Crucially, that first mover advantage means that LTX is now being built into plans and changes that big buy-side firms are making which is allowing it to become embedded into workflows.
“We had a conversation with one large asset manager, with data scientists and traders on the phone, and one of the data scientists said, ‘Maybe we can build this?’ to which a trader replied, ‘But you haven’t’. We have the ability to work collaboratively with clients and one of the things that we’ve learned building BondGPT is that using this technology, the pace of innovation is just faster.”
This article was first published on Best Execution
©Markets Media Europe 2023
