COPING WITH EXTREME MARKET STRESS.
Market infrastructure and systems passed the volume test in March, thanks to flexibility, capacity, and cloud technology. Dan Barnes reports.
In 2020 trading firms are increasingly using cloud-based and machine-learning tools to overcome inefficiencies in trading workflows. Market infrastructure has been sorely tested since the 2008 financial crisis. Many of the world’s largest exchanges suffered outages that hurt traders and investors over the past decade, as trading in and out of positions was paralysed.
The last two years have seen trading delays or outages on the Chicago Mercantile Exchange, Deutsche Boerse, Hong Kong Exchange (HKEx), the London Stock Exchange, the Moscow Exchange, NASDAQ’s Baltic and Nordic markets and the New York Stock Exchange.
In March 2020, when a combination of an oil trade war and Covid-19 responses triggered a sell-off across the globe, the markets largely held up – although Malaysia deliberately closed its stock exchange. However, this was not business as usual.
Trade processing was challenged. The Australian Securities and Investments Commission (ASIC) requested high-frequency trading (HFT) firms reduce the volume of orders they were running, warning: “If the number of trades executed continues to increase, it will put strain on the processing and risk management capabilities of market infrastructure and market participants.”
The Australian Stock Exchange (ASX) itself reported challenges when the previous daily record volume more than doubled to 7.1 million trades on 13 March 2020 and ASX systems experienced some processing delays via its CHESS settlement platform.
There were also clear weaknesses in trading workflows away from market infrastructure, particularly in more complex instruments.
Analyst firm Acuiti, in its April 2020 ‘Derivatives Insight Report’, found that 21% of firms on the buyside which included proprietary trading firms and asset managers, suffered a decline in the instruments they could trade at the height of market volatility. Some 23% of sellside firms noted communication problems with clients and 16% highlighted issues communicating with colleagues, while 17% of buyside firms had challenges speaking with counterparties.
More significantly it found that 58% of dealers had major back office problems, with buyside clients reporting surprise at the issues that were uncovered. In the Acuiti report, buyside firms also saw significant concerns around margin calls, both in their frequency and capacity to meet them.
In part, legacy systems and workflows were found to be at fault. One custody banking executive noted that peers struggled to reconcile errors that crept in, and that had a knock-on effect for the timely processing of margin calculations and instructions.
Hard to trade
Challenges were also severe in markets which trade manually or by voice. The inefficiency of exchanging information by Bloomberg or Symphony messages and by voice when compared to systems with straight-through-processing is significant.
This is particularly tough in fixed income primary markets, as bond issuance spiked because of central bank policy in lowering rates and buying corporate bonds for the first time.
This is one of the more manual areas of trading, with significant back-and-forth of information taking place between the syndicate banks managing the issuance process and the buyside traders, who act as intermediaries to the portfolio managers.
Record levels of bond issuance, including US$1.92 billion of bonds issued in the US up to September 2020, created enormous pressure on the desks of investment managers and tied up sellside traders in high volume, low-value communication creating inefficiency on both sides of the street.
Technology adoption on the sellside has been slow in this area, which creates risk for investors as portfolio managers may not be provided with timely or correct information, but that is likely to change says Nick Hall, managing director and head of fixed income at IHS Markit.
“We are going to see a lot more of these manual processes being automated in order to mitigate that operational risk which will increase as volumes rise,” he observes.
The appetite for improved primary market activity is being driven by several initiatives including a sellside consortium, DirectBooks, but regardless of the model adopted, buyside firms want to see an electronic system, that creates auditability, real-time transparency and more automated workflows.
Keenan Choy, managing director for Fixed Income Syndicate & Capital Markets at Wellington Management says, “I think speed and transparency go hand-in-hand. While we want to increase speed through efficiency, we also want safety enhancements to protect investors by going faster. We want to have insight into how the deals themselves are running in real-time, since a lot of this data is useful for our PMs and traders to evaluate.”
Clouds for safety
The value of using cloud-based systems with low-level machine learning to help automate processes has been supported by the stresses in legacy technology. Any successful automation hinges upon data upon which low-level decisions can be made without human intervention. It is those markets with most limited access to data who suffer the most in a volatile period. Even where systems exist to enable trading, the absence of data can push traders to a more manual process.
Cloud-based tools can allow trading system providers to process large amounts of transaction data without risk of falling over, and in addition providing considerable processing firepower to support the calculations needed.
Vuk Magdalenic, CEO of Overbond, which applies multiple data sources to a machine-learning model to pinpoint prices, says, “Sellside trading desks can boost their request-for-quote (RFQ) hit ratio with a pricing feed featuring a margin optimisation model add-on, and grow their RFQ response volumes 80-120%.”
This not only offers greater efficiency, but its cloud-based model allows it to scale up at times of peak volume while maintaining accuracy in pricing.
“We performed a test with a large European credit trading desk partner and between January 2019 to today, for all their executed trades the algorithm delivered seven cents price precision for Tier 1 category that can be executed with no-touch, and 8.5 cents precision – on an average and on a price basis – in Tier 2 trade category that is a candidate for one-touch execution. It can manage data sourcing and aggregation in the cloud and model and visualise all analytics in the cloud portal.”
Cloud based-tools are not unlimited in capacity, said Steve Toland, CEO of hosting and trading system provider TransFICC, speaking at the Fixed Income Leaders Summit on 14 October, but they do give a system the potential to expand as needed.
“One of the things that we do at TransFICC is host customers both in co-location data centres and cloud data centres,” he explained. “Trading venues like Bloomberg, Tradeweb, and MTS have a throughput limit on their APIs and if you send too many messages in a certain period of time, or too many messages at the same time the API will cut you off. We use computer memory to keep track of how many quotes we are sending at any one time. But our customers just send us more and more quotes because they don’t have to bother about that throttling, it’s our problem.”
He said that a few months ago in 2020, one of his partners had alerted him of the need to upgrade the memory capacity of all of the firms’ computers.
“About 10 minutes later he came back me and said, ‘All the cloud customers are already done. For each customer who is in the cloud we logged onto our AWS cloud environment, changed the virtual computers that we use, picked out some computers that are already in the data centre, and we picked ones that have more memory capability and deployed them immediately for the test environment and at the weekend for the customer environments.”
By comparison for customers co-located in the data centre, TransFICC had to buy physical servers, which was a more expensive and less tailored process.
“So changing your requirements in the cloud is really easy, it’s cost effective,” he added. “It is much more difficult in a co-location data centre.”