Trader TV: Generative AI in Fixed Income Trading
LTX CEO Jim Kwiatkowski sat down with Josephine Gallagher of Trader TV to discuss the application of AI in fixed income trading and how we’ve evolved BondGPT with our enterprise BondGPT+ offering.
Jo Gallagher Welcome to Trader TV, your insight into institutional trading. I'm Jo Gallagher. Today I'm joined by Jim Kwiatkowski at LTX to discuss where AI can be applied to trading functions and fixed income and the challenges in building these solutions.
Jim, welcome to the show.
Jim Kwiatkowski Thank you for having me here.
Jo Gallagher How can AI and GPT powered solutions be used to support trading functions and portfolio construction in fixed income?
Jim Kwiatkowski Many of the fixed income market participants will say the market's opaque, and people talk about the lack of transparency. But other people talk about the abundance of data, the fact that there's much more data than they can lay hands on quickly and easily when they need to.
So we wanted to build a dashboard on our trading platform, the LTX trading platform. And what we learned was that it was difficult to satisfy all the customers. They were all looking for different types of data.
And in building a dashboard like that, our product would've looked like so many other products where there's a complicated menu system and difficulty training the customers on how to use it.
We started to experiment with the new GPT technologies late last year when ChatGPT first came on the scene. With BondGPT, we're able to just take a simple English language question and then access multiple different data sources, do the bond calculations, and give people the simple answers to the questions that we're looking for, which might be a number, it might be a table of bonds, it might be a chart.
And then it takes up whatever screen real estate is necessary to deliver the answer.
Jo Gallagher And what kind of appetite and response have you seen for these types of AI powered solutions?
Jim Kwiatkowski The response to BondGPT's release only a couple of months ago has been tremendous. We already have over 700 users from 200 client firms.
And the clients saw immediate value in BondGPT in its ability to deliver them lots of data from disparate data sources and do it incredibly quickly.
The client started to ask us what more we can do for them. And one of the requirements that we heard most frequently would be the ability to access their own data and their own financial models.
And so we created BondGPT +. And with BondGPT +, we've given clients a range of integration features, some governance and entitlement control capabilities, some additional user customization features.
And that has us speaking with many of the largest asset managers in the world about how we can integrate all of this information for their unique user requirements.
Jo Gallagher And what are some of the barriers to building AI applications and driving client adoption?
Jim Kwiatkowski I think a lot of the concerns from users come down to fear of the unknown. Will the answers that they get be accurate and what's being done with the questions that they ask and how might those be used as a source of information in a way that they're uncomfortable with?
So we worked hard to ensure that accuracy is built into the product, but we worked equally hard to demonstrate that accuracy. Because we want the customers to feel confident with the answers so that they can use them in their jobs.
We also worked really hard to educate our customers about how data is being used because they're now asking us to integrate their own data into the product. So a lot of our focus has been on the education around security and compliance.
Jo Gallagher How have you baked data security and accuracy into your product design?
Jim Kwiatkowski First and foremost, we use only curated sources of data so that we know that the data being used to answer any question is a very high accuracy. Next, the way we use the GPT, we ensure that we're only gathering the information, we're not generating new information for the customers.
Third, we wanted the accuracy to be apparent to the customers. So we built in a simple feature called Show Your Work. And it's not too different from someone asking an analyst to do some work for them and then saying," Can you please show me how that answer was arrived at?" So the Show Your Work feature does that for the user right within the AI.
With regard to security, we made sure the data they provide us is entitlable and only available to the users it's intended for. And we made sure that the way their questions are used, that they're always anonymized and that information is never used for the training of any AI.
Jo Gallagher So what are the next steps for developing BondGPT and how can vendors like yourself, future- proof their technologies to deal with evolving policy changes?
Jim Kwiatkowski Well, it's amazing. We only launched BondGPT in June, and here we are in October launching BondGPT +. With the GPT technology, the ability to innovate quite a bit more quickly than people are accustomed to is what we're experiencing.
And with that, we're seeing lots of imagination coming from our customers. Many of our customers have large data science organizations and they're trying to deliver capabilities for these customers, and they're seeing that we're offering a foundation for them to build on top of.
So I think for us to stay in front will be just continuing to listen to our customers and building for them the things that they're asking us to build next.
Jo Gallagher I'd like to thank Jim for his insight, and of course you, for watching. For our other episodes and to catch Trader TV this week on Monday at London time at 6: 45 AM, go to tradertv. net.