Artificial intelligence is being applied to bridge and unite information in bond trading. Liquidity is notoriously gappy in credit markets, and market data naturally reflects this. Bursts of trading activity are not the only points of isolation. Trading happens across many points, some bilateral, some multilateral, and information is therefore distributed between both points in time and points of execution.
There are tools that capture a majority of credit market activity, and the US has full post-trade transparency, via TRACE, albeit structured to mitigate information leakage that would prevent certain trade sizes from becoming unviable.
Building a picture from this very disparate data universe is a constant but rewarding challenge. If patterns can be found in pricing and trading activity, they can inform pre-trade decision making, and support the knowledge that each trader is able to bring to the table.
In every market, with or without a post-trade tape, the technology needed to aggregate data is crucial to gaining an advantage in trading.
Artificial intelligence is delivering value in several key areas, according to bond traders. First is its ability to find patterns in unstructured data. Due to the negotiated nature of bond trades, much of the information which exists around a trade is unstructured, either recorded as voice or written information.
That unstructured information is exploding in volume as a greater number of communications are recorded, and potentially contains information on trades and those negotiations that did not result in trades. Parsing data on this scale would be impossible manually, and rules-based systems would struggle to handle unstructured data.
Analysis of this data set can support buy- and sell-side desks to assess potential trades, alongside compliance and reporting models, but only with systematic processes in place in order to clean it.
Secondly, structured data can also be incredibly challenging to process due to the aforementioned market fragmentation. Machine learning tools can be trained on data sets in order to find patterns that support a more systematic approach to trading, including model development and signal generation. Where there are gaps in pricing, models can be used to assess where pricing and liquidity is likely to be.
The third use of AI is as an interface, bringing the results of analytics to the fore for users without requiring specialist query knowledge. Tools are already being used to assess relative value and alternative bonds to use for an investment idea.
Clearly, by providing a better picture of a fragmented market, via an interface which can easily be understood by traders, an investment manager can become more efficient at assessing a situation. It can scale up activity by reducing the level of work required to grow.
That said, there is a more noise around AI than actual activity. Risk exists around the explicability of any system, with regulators often demanding that decisions can be transparent.
For the moment, Incorporating new tools that use AI into the trading workflow, as opposed to changing systems, can potentially improve the investment process more easily. Supporting traders with more, better information helps the desk at a point where margins are tightening.
The final gap AI might be able bridge is between the ‘haves’ and ‘have nots’. If it democratises access to data and analytics, reducing the heavy lift historically associated with more systematic processes, greater competition might be the ultimate result.
