Type to search

Cover Stories Feature Q3 2020 Journal Archives Buy-side

The Evolution of Automated Trading in Fixed Income

With Dwayne Middleton, Head of Fixed Income Trading, and Amit Deshpande, Head of Quantitative Fixed Income Investments & Research, T. Rowe Price

What has been the evolution of automation in fixed income trading and what is the current state of play?

Dwayne Middleton, T. Rowe Price

Dwayne Middleton: The evolution of automated trading in fixed income, from a buy-side perspective, emanated from efficiency and scalability efforts. Data aggregation initiatives, the acceptance/growth of third party algo driven pricing and the increasing role of the buy-side trader as a price decision maker have all contributed to provide flexible, rules based, data driven opportunities to trade. Whether it is setting an execution level vs. one of the composite level pricing sources, a range of liquidity scores, and/or a required number of quotes received, the ability to generate automated orders for execution that fit these rules based constraints has been beneficial to increasing the efficiency of the trading desk. 

Amit Deshpande: While the buy side has made significant progress in automating fixed income trading, we are still very early in the game, especially when compared to some other asset classes. The market is increasingly seeing use of electronic trading platforms that has led to more efficiency. However, much of this gain has been in the segments of the market that have been traditionally more liquid, and for orders that are homogenous. Large and complex executions such as portfolio trades still take manual intervention and can take hours to complete. The inherent complexity of fixed income markets has slowed the evolution somewhat, especially compared to equities and currencies. 

What are the benefits of automation in fixed income trading?

DM: A key benefit is capturing the pricing information, dealer quotes, and Trace data at that point of time and being able to prove out the quality of our executions. Capturing the full liquidity opportunity set for a security is critical today given how fast the velocity of information and data flows across the desk. Feeding this data back into our pre/post trade analysis tools is a benefit to our investment process across portfolio management, research and trading. Time management and multitasking under pressure situations are key characteristics of a desk that operates at a high level, so moving repeatable trades of securities to a more automated workflow and gaining access to that data in near-real time frees up the desk to participate in more value-added discussions in the investment process. 

Amit Deshpande, T. Rowe Price

AD: In addition to the important points above, elimination of friction caused by manual trading is an added advantage of such automation. In most cases, better price discovery is a natural consequence of increased reliance on systematic trading algos. In the ideal state, fuller automation would lead to more orderly and predictable markets. This means lower inter-temporal variation of the liquidity premium and better value attribution to the various market risk premia. Over time this could lead to a dampening of the volatility-of-volatility from today’s levels and improve risk adjusted returns for the intelligent investor. This would also lower trading costs resulting in more seamless and efficient market making. 

What are the challenges/limitations of automation in fixed income trading?

DM: Fixed income has many facets of high touch trading in sectors/markets that benefit structurally from human intervention to facilitate best client outcome. Extreme volatility as we saw in March 2020 posed some challenges for some of the low touch automated workflows but as that volatility subsided, we did see the percentages of automated volumes return. To benefit from the efficiency and data aggregation benefits from automation, the workflow is limited to securities where there is robust price visibility and quality composite pricing to make rules-based execution consistent with our efforts for best client outcome. Our technology platform must incorporate broker restrictions, guideline compliance and the ability to capture all of the prices/quotes at the point of execution. 

AD: It is almost a fallacy to lump all bonds into one “fixed income” category. Emerging Markets Local, for instance, has as much to do with Treasuries as Oil has with to do with Chinese Real Estate Trust. Most fixed income asset classes have neither the transparent price discovery of equities nor the advantage of liquidity enjoyed by FX. Trade execution is improved greatly by automation for certain assets but not significantly for others. One challenge is optimization of electronic order flow in a market traditionally dominated by manual processes, including both buy-side orders and inter-dealer platforms. We are at a stage where dealers have efficient execution platforms for the more liquid instruments but still use non-automated environments for others. The paucity of historical data combined with lack of transparency has hampered development of trading algorithms which could facilitate a greater electronic presence in execution. These issues are exacerbated by market volatility as Dwayne pointed out.

What areas within fixed income trading are suitable to automate and which may not be? 

DM: We have seen an uptake for trade automation in more liquid sectors such as on-the-run Treasuries. Credit securities with a well-defined market context and visible pricing also have seen growth in this area. Liquid portfolio products trades such as generic trades in rates derivatives and CDX/iTraxx indices are areas where automation works. There have been innovations in mortgage backed securities (MBS) and even parts of emerging market debt. Less liquid segments of fixed income or where the human touch is required do not now fit this workflow. Security selection driven by the unique insight of our research teams combined with the risk parameters set by portfolio managers also lean to more high touch trading. Even within high touch trading, we are looking for ways to reduce clicks and streamline the execution. 

AD: I agree with Dwayne. Liquidity and trade automation seem to be inexorably tied in a self-referential loop. As long as price is a function of volume and trader interest, we will continue to see this dichotomy. There is a reason why fixed income traders – and not just the Physics majors — are familiar with Heisenberg’s Uncertainty Principle. 

For a buy-side firm, is automation in fixed income trading about automating in-house, or is it more of a sell-side story?

DM: Fixed income trade automation is unique to each firm, in my opinion. We need to partner with dealers and platforms regarding our efforts around automation. We are focused on getting smarter and more resourceful in our search for liquidity. This is the intersection within fixed income between defining efficient methods to deploy our trading, data science and quantitative efforts to deliver investment results for clients. Automation fits within our strategic efforts on maximizing our trading platform and efforts to streamline repeatable processes. 

AD: Buy-side firms can gain a lot of leverage by harnessing the power of data and trade automation. Almost every firm of size has access to price, volume, and position information but it is often highly fragmented. Centralization of this data on an intelligent digital platform can help a firm identify dislocations of price and demand. Taking advantage of the changing cost of liquidity, for instance, can add an uncorrelated source of alpha to client portfolios. Integration of research and trading platforms can lead to better relative value calls. It is well-recognized that the shelf life of research has shrunk as the asset management industry has grown. The upshot is that arbitrage strategies are becoming more dependent on technicals than on fundamentals. Automation can help the buy-side reduce costs, gain flexibility, and ultimately be more nimble with opportunistic trades. 

What is T. Rowe Price doing to advance automation in fixed income trading, whether that be in-house or in terms of leaning on sell-side partners to automate? 

DM: We are in various stages of planning and development to build our trading ecosystem to be as future-proof and scalable as we can. Data and connectivity go hand-in-hand for an effective automated workflow. We are learning from our colleagues on the equity side of the house who have lived through this electronification and are using automation for certain workflows. As traders, we partner with internal colleagues including portfolio managers, analysts, technology, compliance, and legal as well as external partners for order workflow and data aggregation initiatives. Conversations with our dealer partners start first with our focus on clients. We are working on several initiatives to bring a higher-quality level of engagement with our dealer partners. Direct connectivity is one avenue that will create value-added engagement on behalf of our clients. Dealers play a leading role in risk transfer and their efforts to bring innovative tools in partnership with the buy side are welcome. 

What role does artificial intelligence play in the automation of fixed income trading?

AD: The use of AI in traditional fixed income trading platforms is so rudimentary that nearly every major area could potentially benefit from such techniques. More accurate pricing is one such opportunity. Most pricing services use recent trades to update their marks. However, it is not a trivial exercise to update prices for bonds that have not traded for days, if not weeks. If, for example, we see a large trade in VZ 2.5 of 5/16/30, it will likely have a price impact on other bonds across the issuer’s curve, on the Wirelines sector, and on the BBB curve. This propagation has traditionally been a high touch activity with significant differences between firms especially in High Yield, EM, and Munis. AI techniques like cluster analysis and backpropagation algorithms can help with faster price discovery. Another potential area is in identifying transient and long-lasting sources of value. Traditional investing has favored the latter because of its better signal-to-noise ratio. But with the help of deep learning algos, one could identify short lived pockets of local dislocation that mean revert faster than traditional value metrics. There are several other use cases where supervised and unsupervised learning can complement manual trading processes. Over time, we will see the emergence of higher-frequency and high mean reversion trade ideas add significant alpha to real money portfolios, as they have done to alternative investment strategies.  

What role do algorithms play in the automation of fixed income trading? 

DM: Within equity and FX markets, algos are more prevalent. The growth of automated market making across credit markets has dramatically improved the response rates and hit ratios across many of the electronic platforms. This in turn led to the development of rules-based constraints for automated workflows via the order management system on electronic platforms. The composite pricing created by the platforms also plays a key role. This afforded the buy side the ability to implement automated executions by having readily accepted reference pricing used to evaluate the dealer-sourced bids and offers in combination with the other constraints such as liquidity scoring. 

AD: Algos have the potential to add significant value by making data-driven and rules-based trades in real time with minimal human intervention. Minimizing market impact of large order executions, generating alpha through arbitrage opportunities and designing efficient hedges are some areas well suited for quantitative and algorithmic trading. As Dwyane said, we have seen growth in algo-driven trades in the more liquid asset classes like FX and Rates. The most direct way to use algos in fixed income is for non-directional, risk-neutral relative value pairs. An example would be curve trades that take advantage of temporary misalignments amongst different maturities of the same issuer. The advantage of algos is not as much as in generating trade ideas as in the speed of identifying such dislocations. In March 2020, we saw multiple sigma moves in the NAV basis of many large ETFs as markets struggled to find fair value for the underlying securities. At the same time, the cash-synthetic basis in credit derivatives showed a similar pattern. It would have been impossible to manually flag and monetize each opportunity, but an algorithm could easily create a basket trade in near-real time. 

What is the role of trading platform providers (e.g. MarketAxess, Tradeweb) in advancing automation in fixed income trading?

DM: The electronic platforms provide a lot of value outside of their original execution models. The composite pricing is one. Another is the enhanced integration with the key OMS providers and EMS providers. They have been essential to the fixed income trading infrastructure developments that have occurred over the last few years. The other role is around data capture and delivery back into our internal systems. Straight through processing is key for efficiency efforts. The platform plumbing also provides a path for all-to-all trading where the desk may passively aggress around liquidity that best meets client outcomes. 

What is the future of automation in fixed income trading? Where will it be in five or 10 years? 

DM: The role of pure execution fixed income trading is largely over. Traders must add value in the investment process and that includes market intelligence, flexible trading protocols and intellectual curiosity. The complexity of client portfolios, the pace of information flow and operating in a global multi-asset fixed income world requires a high degree of intensity and focus from today’s fixed income trader. Our efforts around efficiency will only increase going forward. 

AD: The focus of the modern buy-side trading desk is to seek alpha through idea generation and superior execution. Large platforms such as ours sit at the crossroads of global order flow, real time price discovery, and proprietary research. A big change we have seen in the last decade has been the explosion in data available to make decisions. FINRA alone reports processing more than 75 billion records every day! The nature of data has changed as well. Historical tick data used to dominate earlier, but now trading histories, reference data, and even alternative data like Natural Language Processing (NLP) are gaining importance. However, analytic tools available to transform data into decisions have lagged quite a bit. While human traders are still likely to be calling the shots in a 5-10-year timeframe, they will most probably be assisted by smart decision-making algorithms. We see the mainstreaming of integrated trading/portfolio management tools that will use ML/AI techniques to combine external feeds like news events and price innovations with prop information like quant signals and fundamental research. Automation will help accelerate the value added by trading in the investment process.