Source: Markets Media
Over the past few years, regulatory and competitive pressures have driven a steady rise in the adoption of transaction cost analysis (TCA) in the fixed income space. Market participants are under increasing scrutiny to demonstrate best execution, manage outliers and review their own performance.
Primarily, participants have turned to post-trade analytics to accomplish these goals: capturing and analyzing data following a trade’s execution. Can this same data, however, be employed to provide insights pre-trade? The migration of TCA to earlier in the trade cycle is still in its infancy, but as it matures, it could provide tremendous value for trading desks in all areas of fixed income. What pre-trade analytics are possible? One of the most encouraging—and self-evident—use cases for pre-trade TCA is giving traders the tools to help them predict what their expected execution costs will be. Predictive modelling has been prevalent in equities for many years, but its implementation in fixed income is challenged by the fundamentally less-predictable nature of the asset class. Liquidity shifts alter perceived order difficulty based on size and price availability. It is much harder to model exact predictions for thinly traded instruments.
However, predictivity does not need to be exact to be useful. Having a relative target implementation cost is attractive to traders, as it allows them to manage expectations right from the inception of the order by putting the achieved costs in context.
Fixed income traders are beginning to build predictive models based on a growing array of post- trade observations but out of a sea of data, which should they choose? The micro-structure of fixed income makes this a tricky question. Models predicated on specific instruments where precise volume data is available might work for equities, but in fixed income—where maturity and seasoning change the liquidity profile of a bond throughout its life—the opportunity is emerging to implement pre-trade models based on a new, fixed income specific approach.
The best models will blend client-specific observations with more general factors such as maturity, rating, age and duration. Although there is no single instrument context, thorough back testing indicates it is possible to achieve a strong signal, and this is key. Applying transparency through use of signal strength scoring will allow users to evaluate reliability of predictions against achieved results. This will in turn provide increased confidence for those users to select them and defend their use in post-trade review.
In addition to making predictions, pre-trade TCA can also feed off post-trade insights to create a virtuous cycle. Every day, traders are confronted with basic questions such as: Who do I route RFQs to? How many should I send? When should I send them? Many traders are used to relying on instinct and experience for these decisions, but by analyzing how these decisions turned out for previous trades, they can help traders prepare for the next order.
While previous performance does not guarantee future results, traders need to be mindful not to get too granular or vague when analyzing their prior trades. Aggregating data is a balancing act between not being so general that the signal is weak, and having so many characteristics that the signal is strong but is based on too small a sample of data to be significant. Challenges to implementation
Even armed with these use cases, implementing pre-trade TCA is much easier said than done. Traders will need to overcome one (or more) of three key challenges to successfully implement a pre-trade TCA program:
1. Simplicity: It is important to have a pre-trade TCA model that is understandable and whose value proposition is easily communicated. When it comes to pre-trade, as the complexity of a model grows, its utility decreases. Decision-makers on the trading desk must be able to understand what factors go into a model, as well as the circumstances under which it is strongest and weakest. To grow this confidence the key is to conduct thorough back testing. Being able to illustrate quantitatively these results will deliver transparency which will help speed pre-trade TCA’s adoption among even the most skeptical traders.
2. Discoverability: Even the most insightful data is useless if it is buried in a table or report and impossible to find. The key is to integrate the pre-trade analytics directly to the trading workflow, so as new orders arrive, the OMS and EMS can automatically update the pre-trade analytics.
The trader’s dashboard should be able to see an order and give traders answers based on current market conditions. How long will this trade take to complete? Which dealers have given us the best results in the past? What is my expected leakage from RFQ to execution? Integrated pre-trade analytics will allow traders to make these determinations quicker than they ever have. This can be taken to another level if pre-trade analytics are used to route an RFQ directly or flag it for high-touch treatment.
3. Actionability: As with all analytics, there is a symbiosis between pre-trade and post-trade. You trade, you analyze the result, you implement improvements, and then you trade again. In addition to building the workflow to ensure pre-trade support is presented at the right time, traders must make sure a true feedback loop is in place to maximize the value of TCA.
Feedback can be implemented manually, but your system can also recalibrate models automatically if observations are “tagged” to link results directly to the pre-trade tool. The clearer the identification, the easier it is to isolate areas for improvement in models and decision support algorithms. That said, traders need to strike a balance, as recalibrating too quickly can create volatility in their results. The adjustment process should be gradual and thoughtful.