Can AI price bond new issue indications?
A real pleasure to share the stage with many of our fellow capital markets fintechs, banks and investors on yesterday’s EventsRadar panel.
For me, the most interesting part was a question about whether AI can take over the pricing of bond new issue indications.
First off, the goal and value proposition of our platform was never about automating the entire new issue pricing process, or replacing the talented individuals in DCM and Syndicate who are currently doing this the traditional way. It was, and still is, about streamlining the process, organising and serving the right data at the right time in the right format and automating the mundane, processing aspects of the job.
We are acutely aware of the nuances that go into bond new issue pricing and have no intention of overstepping into areas that fundamentally shouldn’t be automated.
However, it depends on the use case.
In the secondary market, algo-supported electronic trading accounts for an ever increasing share of volumes. But algo pricing remains mostly constrained to smaller-ticket transactions and portfolio trades. For block trades, or in particularly volatile markets, traders still pick up the phone or chat with their counterparties bilaterally (shout out to the FT for their ongoing good coverage on this topic).
In the same way, a differentiated conversation between a Syndicate banker and a borrower in the run up to a benchmark transaction is unlikely to benefit much from an algo or AI pricing.
However, for the hundreds of thousands of pricings that each bank blasts out to their issuers every year as a service, to keep them appraised of broad market conditions, relative value and issuance opportunities, there is significant scope for automation.
We currently support this automation via a proprietary algo that achieves good results and acceptance by users. Crucially, the algo feeds off user inputs, which ensures that pricings remain unique for each bank.
In many cases it’s still used as a reference rather than being accepted blindly - and that’s fine. Some users prefer not to be influenced by an algo at all and opt to price the traditional way. That’s fine too and those users still benefit from all the data and process streamlining our platform provides.
The trend is clearly towards more automation of these low-value, low-effort pricings.
For example, one focus area for us is to provide users with more fine-grained controls over algo pricing inputs and improve the explainability of results, which we believe will increase the share of no-touch pricings.
With the acceleration of AI capabilities over the last few years we’re of course also exploring whether certain elements of the new issue pricing process can be better modelled via a big data approach vs an algo approach.
The jury is still out on that, but one important point to keep in mind is that one size does not fit all in this particular use case. Unexplainable, black-box AI pricing is of limited use in primary market pre-trade workflows.
AI may still have a role to play, but in my view as one component of a bigger process, not as a wholesale replacement of what Syndicate bankers do today.
Co-founder & CEO
Bond Origination Technologies