This article is written by Robert Simek, Head of Sell-side Analytics at Bloomberg.
Traders face the challenge of responding to price requests for securities such as bonds – Request-for-quote or RFQs as quickly as possible and with the best price, sometimes dealing with up to 10,000 orders or RFQs each day.
With ever-increasing volume and under challenging time constraints, many are finding it difficult to keep up — missing opportunities and not working as efficiently as they could be. Hence many are looking to technology to aid them and help optimize this process.
Incorporating AI and machine learning is an approach currently in the limelight, but it’s not as surefire a solution as it may appear. Challenges facing those trying to implement machine learning models include the vast amounts of data available, as well as the ability to specify the problem in such a way as to be able to apply AI successfully. Then, with data, it’s imperative to normalize it, enrich it and line it up so the machine can recognize patterns. Finally, organizations must have the talent to be able to understand the domain and apply AI to the problem.
AI offers a more efficient process for RFQ response
Currently, hard-coded rules are used to help automate the process of routing trades and responding to RFQs, however, the “if this, then that” approach is difficult to scale. A trading desk can quickly run into the complexity of managing these rule sets and determining which rules offset or override other rules.
For example, a sell-side trader can set a single rule to auto-respond to trades of less than five bonds in size, but the lack of detail could lead to complications. So instead, she codes the workflow to auto-respond to trades less than five bonds:
- if it’s in a certain sector and duration
- if it’s above a certain price
- if the trader has a position.
This rule set can grow very quickly, and while the deterministic auto-respond solution is helpful to traders, it requires constant maintenance and monitoring.
In contrast to the above, a different technology approach can help traders solve the problem in a more efficient way:
1. Signals: Add signals that are dynamic to the ruleset. This helps adjust rules based on changes in the market pricing, liquidity, client demands, risk and balance sheet. For example, if negative news is published about an issuer, traders should not auto quote that bond.
2. Machine learning: Add a machine learning model to suggest which orders or RFQs the trader should work on first. This is similar to what Netflix does when it suggests movies a viewer might like. Machine learning technology can do the same thing by learning from the trades users priced in the past. A feedback loop can pick up changes in patterns and adjust.
The biggest benefit of a dynamic automation for traders; it allows them more time to focus on high-value decisions.
From deterministic to dynamic
A deterministic system is a rule-based system, or what people refer to as an “expert system” that is used to automate workflows. Essentially, a subject matter expert creates rules on how the system should work. This automation is typical of what you see in most systems on Wall Street today.
The problem with the deterministic system is that you have to constantly maintain the system, and it’s difficult to manage during ever changing market conditions.
A dynamic system, on the other hand, is a machine learning approach used to automate workflows. The approach uses historical data to understand patterns and make suggestions, like ranking a list of orders. There is also a feedback loop, which can inform the machine learning model of changes. The model can then adjust by understanding what is important and what is not important. The power in a dynamic system lies in the fact that it is data-driven and can adjust over time.
Bloomberg is building solutions like this in the TOMS system to enable dealers to grow their business. This allows sell-side dealers to scale their business and optimize resources given known constraints and to do so efficiently.
For every RFQ that comes into a trader’s view, Bloomberg’s technology analyses the bond’s characteristics, the amount, the side, the counterparty, the time of day and other factors.
From there, the RFQ receives a score, which is applied in the blotter and then sorted. This gives the trader a view of prioritized items that they should respond to first, rather than having to fish through or keep up with orders as they come in.
While it’s true that many sell-side participants and vendors are already trying to solve problems via a machine learning framework, it’s challenging to do it at scale and at a cost that has a positive ROI. Many are trying, but few — beyond big tech firms — have produced results and moved to production.
Bloomberg has seen success using machine learning to solve complex problems. On the sell side, advantage is everything. What makes our solutions distinctive is that we’re developing a framework that delivers unique analytics to each dealer, rather than a generic one-size-fits-all solution.
Learn more about Bloomberg Solutions for the Sell Side here.