“The trading desk is the largest source of untapped alpha,” said Seth Merrin, New York-based CEO of Liquidnet, a peer-to-peer trading platform for institutional buy sides.
Today, traders measure performance in terms of “implementation shortfall,” or I.S., the difference between the execution price and the next available price in market at the time of the order. It is always an attempt to minimize market impact, to make execution as cheap as possible.
“But if dealers become part of the investment process, they can add trading alpha,” Merrin said. “This isn’t about saving a few bps [basis points] on I.S.”
This is a new effort by Liquidnet to marry machine learning to execution, so that traders can become advisors to portfolio managers, rather than order-takers.
Buy-side traders say they need to see results, particularly in Asia, where execution costs are relatively high, but that this is the direction the industry should head.
George Molina, head of trading for Asia at Franklin Templeton in Hong Kong, said, the next evolution in trading having the right tools to allow alpha retention on the trading side, which is then passed on to the funds.
“Traders and P.M.s are working closer than ever in protecting and contributing to performance through spread savings, block trading and data analytics,” he said.
From OMS to OTAS
Merrin created the first order management system (OMS) in 1982 while a junior at Oppenheimer Funds. It literally involved putting 50-odd paper tickets per trade into alphabetized binders, so that orders could be cleared faster.
Wall Street was disdainful of technology; everything was about relationships: “The traders considered a keyboard something just for the back office guys; they wouldn’t touch one,” he recalled.
From there he built his own business refining the OMS into something more scalable, using spreadsheets and adapting it to Microsoft Windows. A later stint in Silicon Valley introduced him to peer-to-peer networking, which eventually led to the idea of an anonymous pool to handle block trades among institutions, cutting out brokers. He launched Liquidnet in 2000.
The trading desk is the largest source of untapped alpha
Today the platform has 135 brokers as well, drawn to the liquidity it has amassed, and it also trades $20 billion a day in corporate bonds, expanding from its equities roots. Merrin says stock exchanges have been great at helping price discovery, but Liquidnet is better for finding liquidity, because it is electronic, global, and anonymous.
Last year the company acquired OTAS Technologies, a London-based firm that’s developed machine learning for alternative data sets (anything that’s not an exchange’s data feed), and is now embedding it into Liquidnet’s bigger infrastructure; the company has pipes into over 1,000 firms.
With OTAS, Merrin believes the firm can lead another wave of technological efficiencies and value creation in capital markets. Despite the ubiquity of low-touch electronic trading in equities, many sell sides still rely on human sales traders and relationships.
“Contacts and relationships are fine, except [sell-side] firms don’t make money in equities, and so banks are downsizing,” he said. Both buy and sell sides will still need traders, but he says the industry remains analog and is going to experience far more digitalization.
Already the platform uses digital tools to route orders to likely sources of liquidity. Next will be giving traders alternative data sets and A.I.-derived insights, to let them partner with their portfolio managers.
For example, in Merrin’s view, a trader might use A.I. to collate many factors (CDS spreads, vol, ETF movements) to determine why, for example, a stock is trading 15% higher than peers at 2x volume. The trader can recommend her P.M. sell or short an instrument.
“Right now traders are spending time on compliance and implementing orders, at a time when alpha generation is sorely needed,” Merrin said. “Trading has been a seat-of-the-pants thing.” Post-trade measurements of performance, called transaction-cost analysis (TCA), has become an important metric, but Merrin says the practice involves “fuzzy math”. Liquidnet/OTAS is trialing a product called Discovery with a dozen or so buy sides, including a handful in Asia. Merrin intends to integrate P.M.s next year.
A.I. to the fore
Buy-side dealers are likely to welcome tools that make them more important in their organization. But managements will probably listen too.
Data from ITG, an independent electronic broker that runs the most widely followed TCA rankings, shows that Asia and emerging markets remain expensive to trade.
According to ITG’s lastest cost report, from August, the average regional commission by market capitalization is 7.2 bps for Asia ex-Japan, versus 3.5bps in the U.S. It’s 10.1bps for global emerging markets. ITG data shows that although total I.S. costs in Asia ex-Japan have declined, from 60.0bps in 2009 to 47.3bps in 2018, that’s still a substantial expense in a highly competitive business. (Broker costs and commissions raise the total cost of execution in Asia to 54.5bps.)
Moreover, if buy sides can realize Merrin’s vision of adding alpha to fund performance, instead of just being a cost of doing business, the impact on firm’s market standing could be palpable.
Research by Morningstar states that buy sides are in such close competition now that an increase of just 10bps in performance could move a fund ahead of 13 competitor funds in Morningstar’s rankings.
So machine learning in execution could make a difference to bottom lines. But buy-side traders to whom DigFin spoke say they haven’t yet seen Liquidnet Discovery’s results. “Scrubbing information and data for emerging markets is a challenge,” said one – which is why large buy sides are starting to hire data analysts to do this, as a means of deciding how to route orders to which broker’s algo.
“Execution without analysis is now dated,” he said. “Seth just jumped over his competitors...[Liquidnet] could become a new ITG for pre-trade.”
He says the next innovation he wants to see is for Liquidnet to rate broker algos. “That would let us see beyond the high-level TCA of an ITG, and go deep into brokers’ coding,” the dealer said. “It would let our programmers work with the sell side, to customize algos.” This is already starting to happen in the U.S., but not yet in Asia.