Fixed Income Trading: Big Data Boost

By Carl James, Global Head of Fixed Income Trading, Pictet Asset Management

carl-jamesAccess to more data is improving the efficiency and raising the sophistication of the buyside fixed income trade execution process.

Fixed income buy-side firms increasingly have the capability to analyse data to achieve a better outcome for clients and to satisfy regulatory requirements. To some degree the reason is quite simple: until recently data was sparse, now it is becoming more and more abundant.

Banks make money when there is opacity, but regulators are forcing them to be more transparent about their activities, justify their fees and validate transaction pricing across asset classes. Moreover, sophisticated technology is getting cheaper which encourages disintermediation and disruption to traditional business models by new entrants. Banks and brokerages are compelled to respond, or else suffer shrinking market share.

The buy-side is in a similar position. The ineluctable rise of passive investing and the intrusion of robo-wealth advisers are piling pressure on established asset managers. Truly, the tectonic plates are shifting.

The technology needed to gather and make sense of the information and then derive recommendations is rapidly improving. Indeed, we have just launched a five-person trading technology team to take advantage of these new opportunities.

The information is generated internally and accumulated from external sources. The internal data accrues from messaging from counterparties, orders, trades and their hit ratios and a myriad other derivations and interpretations. External data, for instance, declared orders and reported trades, was previously proprietorial and therefore scarce or expensive, but is now widely available.

Although there isn’t yet a consolidated tape, measures such as the Trade Reporting and Compliance Engine (TRACE) for over-the-counter bond transactions in the US is a step in that direction. In the UK, the London Stock Exchange’s approved publication arrangement (APA) has raised the level of trade reporting in the fixed income markets, with some third party vendors now aggregating APAs. The product captures a significant part of the total market, and usually large enough to draw valid conclusions especially when used in conjunction with internal data.

Post-trade analysis is critical. As more data is available, a trader becomes better able to gauge whether they handled an order correctly (for example, a request for quote), whether they approached the best counterparty for a particular bond (through examination of hit-ratios) and whether they transacted at the optimum time in the day.

It is also worth emphasising that an important effect of the Markets in Financial Instruments Directive (MiFID) II is to ensure that a fixed income trader’s experience and intuition is underpinned by evidence. A decision to execute a trade and the process towards that decision must be seen to be rational.

Electronic and automated trading of liquid, benchmark bonds is evolving, with increasingly reliable and systematic recalibrations of constituent bond issues taking place. One obvious benefit for traders is that more bandwidth is created for them to deploy their skills and exploit their networks to concentrate on illiquid or esoteric bond issues.

Limits to machine learning effectiveness
Of course, the potential of artificial intelligence (AI) and machine learning (ML) for the trading process has attracted a lot of attention. However, despite the hype, their deployment in fund management in general is at a very early stage: there simply isn’t yet the commercial imperative for their application.

Portfolio managers might use ML to track a benchmark (either explicitly or covertly); traders might eventually find a use for ML if orders are delivered in a less sequential fashion than now. However, the fragmentary nature of the fixed income markets and, despite the proliferation of data, the still incomplete knowledge of all liquidity sources, different trading protocols, diverse instruments and partial price information inherent in an over-the-counter market limits the efficacy of ML in the trading process.

Furthermore, many bond issues trade infrequently. Although most European equities trade around 500 times a day, many bond issues rarely trade for days or even weeks. This, perhaps, is the most significant reason why ML has restricted relevance to the fixed income markets – at least, beyond the regularly traded, liquid benchmark issues which are more amenable to systematic trading.

The extensive use of algorithms caused a behavioural shift in the equities markets, by reducing the size of individual trades. Greater electronification of fixed income trading is also likely to lead to behavioural changes, as orders get focussed on more tightly defined data points and as the underlying process becomes more methodical.

The difficulty for buy-side firms is how to embrace new technologies successfully. There isn’t a standard model with an unambiguous record of success – partly because it’s too early to make an accurate assessment, and partly because many buy-side firms have historically relied (and in many cases still do) on the sell-side for technologies and systems, for instance their trading algorithms. In any case, the buy-side firms need to decide quickly – and must expect to fail as they experiment.

Ultimately, it’s only possible to solve problems and adapt to external innovations to the extent of your capabilities. As in the solution to the riddle, how do you eat an elephant? Answer: one bite at a time.

We’d love to hear your feedback on this article. Please click here

recommend to friends
  • gplus
  • pinterest