Australian credit can’t count on quant

Quant credit trading is taking off in the US and Europe, with asset managers betting that the asset class is ripe for the same innovation that has upended equity markets over the last 20 years. Australian credit may be less fertile ground for such a rollout, but some market users are exploring the viability of using quant practices and systems to gain an edge.

Dan O’Leary Deputy Editor KANGANEWS

In late 2020, the New York-based asset manager Blackstone, which manages nearly US$1 trillion, bought Diversified Credit Investments – describing it at the time as a pioneer in quantitative credit investing. Other firms, such as Blackrock and Citadel, have since followed with similar forays into quant practices in the credit market.

In today’s market, buy-side firms are implementing quant and AI systems to scour data and identify risks and opportunities that human analysts may not have bandwidth to spot. But the outcome is typically further investigation by human investment managers as opposed to the generation of an automated buy or sell instruction.

The history of quant deployment in the credit sector hints at its use as an aid to analysis rather than for automated trading. Quantitative trading and research techniques are typically associated with the most liquid asset classes, including equity and exchange-traded derivatives. But their involvement in credit stretches back to the market’s inception in the 1980s and 1990s, when bank propriety traders used theoretical models, or screeners, to buy and sell bonds systematically.

These rules-based models were built using a limited range of data, as bonds were voice traded. In listed equity, the growing availability of data as electronic trading was introduced and grew fuelled innovation, and the techniques became known as “quantimental investing”.

Julien Turc, head of the QIS Lab at BNP Paribas in Paris, tells KangaNews early quant credit models derived from equity and corporate data were effective, famously predicting the Xerox bankruptcy. “During my years as a credit quant strategist, my most useful and successful models were actually theoretical in nature,” Turc notes.

He spent years working on a series of capital structure models that made links between bonds, credit default swaps, volatility, dividends, equity prices and company data. These models were very useful until the Lehman Brothers crisis, Turc continues, though they have since gradually been replaced by machine learning as more data has become available.

Again, the key has been intraday trading data as the bond market has gone electronic. Turc says it is now possible in some markets to use AI due to the quantity of data available. Machine learning has been used for various tasks as far back as the 1990s – including producing credit ratings, deriving credit spreads and identifying M&A targets. What is new is that quant traders and researchers have “rediscovered these techniques and adapted them into AI-powered neural networks”, according to Turc.

“It takes an analyst a while following the release of an annual report to digest it and conduct meetings with investor relations – it might take days. Our platform consumes everything instantly.”

ANALYST EVOLUTION

With quant credit trading growing in major global markets, some Australian participants are exploring its local applicability. The reality is that Australian credit, and especially corporate credit, may be too illiquid ever to support the full rollout of quantitative trading (see box). However, KangaNews has learned that a select group of fixed-income asset managers are deploying quant research techniques like those once used by strategists such as BNP’s Turc.

Platforms such as Blue Fire AI’s Emmalyn process publicly available data – including company financials, professional news feeds, credit conditions, share market announcements and asset price market movements – and send analysts early warnings of potential credit events.

Refinitiv’s StarMine, meanwhile, applies credit risk models that track a company’s probability of default. It uses similar techniques to those deployed by the early quant credit traders, which were inspired by a 1973 paper by US economist Robert Merton that assessed the structural credit risk of a company by modelling its equity as a call option on its assets.

Hayden Briscoe, head of fixed income, global emerging markets and Asia Pacific at UBS Asset Management in Hong Kong, says quant research platforms consume thousands of news articles a day across a range of languages. Some also read Mandarin Chinese – a tricky feat for a financial jargon heavy task, he adds.

Briscoe comments: “We use quant research platforms that essentially allow us to consume every broker report and tell us what is relevant. These platforms combine equity analyst reports, which have a good track record of predicting earnings, into their signal. It makes our credit analysts more efficient.”

He notes the AI learns over time and with use. “The AI platforms can learn if one analyst is more accurate, or if a journalist has a good track record of predictions. Maybe a certain Fed [US Federal Reserve] governor speaks with one reporter more than others – if so, the system picks up on it.”

The platforms send a warning to the credit analyst to investigate if they determine the probability of a credit event has increased on a particular name. “We get a number of flags in our group every day that lead our credit analysts to do a deeper dive on a firm,” Briscoe adds. “We can scale up very seasoned credit analysts to become a lot more efficient every day.”

One such platform is Blue Fire AI. The company is headquartered in Singapore and launched in 2016 with support from the local government. It initially focused on AI applications within equity risk, but the firm turned its attention to credit on the back of increased volatility in the asset class – a prerequisite for many quantitative strategies.

David Mcintosh, head of distribution at Blue Fire AI in Brisbane, describes this sort of quant platform as a support rather than a replacement for human investors. “Credit risk has increased significantly and we have noted a dramatic uptick in desire for an extra set of eyes in the credit universe. This does not mean we will replace, for instance, the rating agencies. But we certainly have customers that use our product to augment or challenge their reliance on that model.”

“Credit risk has increased significantly and we have noted a dramatic uptick in desire for an extra set of eyes in the credit universe. This does not mean we will replace, for instance, the rating agencies. But we certainly have customers that use our product to augment or challenge their reliance on that model.”

Mcintosh says the goal is to support human intelligence in part by removing some inherent bias. “We aggregate information and interrogate it, training the algorithms to do one thing – to look for patterns of risk,” he explains. “The product becomes a very good complement to an alpha generating process.”

Specifically, the Blue Fire research process has been trained to look at drawdowns, underperformance, spread blowouts and gap risk, and then to seek patterns in the publicly available data that could indicate future asset price weakness.

Market participants say quant research platforms improve credit analysts, helping them cut through market noise. However, they agree the human element will remain essential for some time.

“Traders still need the full picture so there will always be a human overlay,” one Australian fixed income fund manager tells KangaNews. Traditional fixed-income investment organisations rely too much on rating agencies, the fund manager adds, which might give an edge to investors that supplement this model with quant practices.

HOMEMADE RESEARCH

The next frontier of quant practice may be investors developing their own systems and practices rather than having to rely on third-party products. Quant veterans say the arrival of the Python programming language in particular has democratised AI research. Python’s ease of use, flexibility and suitability for big data, machine learning and cloud computing tasks has led to its adoption by non-computer scientists including in financial markets.

Turc says Python and the explosion of availability of high-quality data gives anyone the ability to test large volumes of package data. “For people like me, maybe this is not such a huge change,” he comments. “We have had these systems in place for some time – maybe it was just a little complicated. But there is now a system that can produce the same result with perhaps two lines of code.”

Market participants say this has the additional advantage of significantly reducing cost for new entrants as they can use a cheaper subscription service rather than employing a team of data scientists. There is little or no expectation among market participants that machine learning will replace human investment management in the foreseeable future, but even in less liquid markets there is room for it to provide an edge for those who adopt it smartly.

Briscoe says integration of third-party quant research platforms has already given UBS AM’s credit analysts a competitive edge – and he believes credit analysts in future will not exist without this technology. “They have to evolve – they cannot be stuck in their old ways,” he cautions. “It takes an analyst a while following the release of an annual report to digest it and conduct meetings with investor relations – it might take days. Our platform consumes everything instantly.”

Quant trading Australian credit

Quantitative trading has revolutionised liquid markets globally and its deployment in the credit space is believed in some quarters to be the next financial revolution. However, data driven techniques first developed in the equity sector will likely not be able to be fully deployed in Australia’s less liquid corporate bond markets without major change.

The issue, market participants say, is that the underlying fuel powering quant credit trading overseas might not be viable in Australia – at least not without serious market development. Indeed, the absolute cutting edge of quant approaches may not be fully deployable even in the world’s most liquid credit markets.

Julien Turc, head of the QIS Lab at BNP Paribas, tells KangaNews quant credit trading using AI in European and US markets is itself problematic on the basis that there may be insufficient data to power the models. The complexity of the credit asset class adds to the difficulty, he says.

“There are a lot of companies and a lot of different instruments out there,” Turc explains. “We are at a stage where statistics and machine learning can be applied successfully thanks to an increased volume of data. We might end up with enough data points to make some use of real AI, but I am not sure we are there yet.”

The Australian credit market is even less well positioned for a quant overlay. The reality is that secondary trading is typically limited in scale and often dries up to a significant extent especially during periods of heightened volatility. For instance, secondary trading all but shut down for at least some time in the middle of 2022 as rising inflation and interest rates shocked the market and dealers hit their risk limits.

JULIEN TURC

We are at a stage where statistics and machine learning can be applied successfully thanks to an increased volume of data. We might end up with enough data points to make some use of real AI, but I am not sure we are there yet.

JULIEN TURC BNP PARIBAS

Australia also lacks a central trade reporting system like those found in other jurisdictions, such as the US’s Trade Reporting and Compliance Engine (TRACE). This acts as a repository of data that traders can use to build and feed models.

Attempts have been made by various entities to develop a third-party equivalent in Australia, using sources of trading information like the ASX’s Austraclear system. But KangaNews is not aware of any platform that replicates what TRACE offers US traders.

Hayden Briscoe, head of fixed income, global emerging markets and Asia Pacific at UBS Asset Management, says Australia would likely need to see significant development in the fixed income exchange-traded fund (ETF) and block trading spaces to build the infrastructure that would make quant techniques viable.

“Quant trading comes about when there is more product in the passive investment space, such as ETFs,” he says. “Quant shops then price ETFs and create liquidity by executing block trades of 50 or 100 names.”

This sort of development is not impossible; Briscoe notes that he is starting to see commercial banks supporting the ETF market in the Asia-Pacific region and says he expects to see the same phenomenon in Australia “at some stage”. Higher rates could attract retail money to the credit market to a greater extent than has historically been the case in Australia, and ETFs are a natural way for this buy-side sector to gain exposure.

While this is not yet the case, asset managers are scaling up and want to trade multiple bonds – or slivers of their entire portfolios – not just single names, Briscoe adds. The development of a more liquid and diverse domestic credit default swap market would also encourage greater liquidity, he notes.

The bottom line, though, is that market participants believe quant trading opportunities will always be limited in Australian credit by the relative lack of diversity and liquidity in the market. A machine can recognise good trade ideas but a trade idea has no value if it cannot be executed in scale.