Everything You Need to Know about Dynamic M-ELO, the First Exchange AI-powered Order Type

Last month, the SEC approved Nasdaq’s filing to introduce the first exchange Artificial Intelligence (AI) powered order type, a watershed moment for how AI is leveraged in the markets.

This new order type allows like-minded buyers and sellers of equities to minimize the extent to which their transactions impact prices. That may sound like a straightforward goal, but it poses a significant technological challenge and solutions have historically meant a range of drawbacks. “I suspect we will reflect on this release as the most significant step since digitalization for improving how markets function,” said Doug Hamilton, Nasdaq’s Head of Artificial Intelligence Research and Engineering. “Markets are more reliable and faster today than they have ever been, and Dynamic M-ELO is an important step in our efforts to modernize the markets.” 

“What we’ve built is the first-ever capability to make markets operate in more adaptive and more proactive ways that make interacting with them easier and that make them more efficient in the long run,” he added.

Let us take you under the hood of this technology to help explain where it is coming from, and where it may be going.

Simply Put, What Is Dynamic M-ELO?

Dynamic M-ELO is an updated version of the Midpoint Extended Life Order, which was launched in 2018 to attract and unite counterparties with longer-term investment horizons. Mechanically, it is a midpoint order that waits a short period of time before being eligible to execute against other M-ELO orders that have also waited the same period. This holding period – which was originally 500 milliseconds (half of a second), was then reduced in 2020 to 10 milliseconds – making it more likely that both parties were transacting in a price-stable environment which drives efficiency for market participants. The holding period time of the legacy M-ELO order type has been fixed for all market participants, but the new Dynamic M-ELO will control the duration by leveraging AI. Specifically, it uses a learning algorithm that seeks to maximize liquidity while minimizing the impact that trades could have on price.

What Is M-ELO Solving?

The idea for M-ELO came from demand from Nasdaq’s customers. “There has been a continued desire by institutional investors – asset managers such as mutual funds– to find new ways to trade with similarly situated counterparties,” said Chuck Mack, Nasdaq’s Head of Strategy for North American trading services. “They generally have a long-time horizon as far as their investment, and they're not aiming to be lightning-fast traders. What they’re really thinking about is: How do I safely express my view on various investment opportunities in the market?”

Historically, these investors have looked to dark pools, or Alternative Trading Systems (ATSs), where they can segment the types of participants they match with, and thereby exert some control in finding like-minded counterparties. But if these institutional investors were looking for options on more closely regulated exchanges, there were few viable options for them. In response, Nasdaq developed M-ELO with the idea that a forced holding period before execution would both discourage any traders looking to take advantage of momentary price shifts and it would naturally make transactions more likely to be stable.

As Hamilton explained: “The reason why these sorts of markets are attractive is that volatility tends to cluster at every level of analysis: So, a volatile day today means it’s more likely to be volatile tomorrow, and if it’s not volatile today, it’s less likely to be volatile tomorrow. This also extends to intra-day events, so if you look at the stability of a price, when the price starts to become unstable, it tends to stay unstable. And when a price is stable, it tends to stay stable.”

“So, if you are placing an order into an unstable environment, having that order hold for a certain period of time increases the likelihood that you'll execute in a stable environment,” Hamilton added. “Everybody on the planet would rather wait a few milliseconds to get that better price.”

If The Product Worked Well, Why Did It Need to Become “Dynamic”?

“Ultimately, the goal is to make M-ELO a more attractive order type through that dynamism to bring in additional volume and give our clients a better execution experience,” Hamilton added. In other words, by determining which situations were which, and then dynamically adjusting the timing accordingly, an improved system could theoretically achieve the same levels – or better – in terms of price stability while maximizing the liquidity for M-ELO users. When M-ELO was first introduced, the team chose a holding period of 500 milliseconds as a starting point. While the product was generally well received, the feedback from Nasdaq’s customers was that their order placement strategies were sophisticated enough that they didn’t typically require such long waits and that they were concerned that such long waits were making it tougher to fill orders.

Consequently, the move to a 10-millisecond hold led to higher fill rates – and more investors began using M-ELO. In fact, in just two weeks of making that change, daily M-ELO volume increased from approximately 8 million shares to more than 15 million shares. Today, that volume is approximately 25 to 30 million shares per day. Despite the strong growth in the demand for M-ELO orders, the total addressable market looks like hundreds of millions more shares per day. That’s where Dynamic M-ELO comes in. “We asked ourselves: Why 10 milliseconds? The initial research indicated that was a reasonably successful holding period – but is it always the best number?” Hamilton explained. “There are certain conditions where you could imagine if prices don't move at all, if they're incredibly stable, you could have a very low timer and still achieve the same exact level of market stability while achieving higher fill rates – and there are very unstable conditions where you might need a longer delay.”

How is Nasdaq using AI to make M-ELO dynamic?

Nasdaq’s Artificial Intelligence lab approached the M-ELO team with an idea: Let’s build an algorithm that can dynamically adjust the holding time to maximize the fill rate and minimize the impact that trades have on the market price.

The specifics of the Dynamic M-ELO model were published in a white paper, but the highlights are that it takes into account 140 factors, the specific delays for each symbol are updated every 30 seconds, and the timing decisions are made by a 35,000-parameter AI that is constantly learning and improving. The only hard-coded element of the program is a stability-control mechanism that rejects any recommended delays that exceed fixed parameters.

While the exact approach to determining the delay length for each symbol at any given moment will improve over time, the basics are simple enough for even non-AIs to understand: Two important factors that the program will be looking at are how stable the symbol has recently been and how much activity the stock is currently seeing. Less volatility and more volume will suggest shorter waiting periods and more volatility, and less volume will suggest longer holds.

Will Dynamic M-ELO Lead to Better results?

After refining the model and testing it against the current M-ELO results, the findings surpassed all expectations. “As they started building models, we posed something we thought was quite challenging: A 20% increase in fill rate without impacting the post-trade performance of that execution,” Mack recalled. “Based on modeling in the lab, the AI program ended up delivering that 20.3% increase in fill rate we set as a high bar – plus an 11.4% improvement in performance, which is just kind of amazing,” Mack said. “We thought we'd be giving up a little performance, which still would have been a good tradeoff to increase the liquidity available.” To put that 11.4% improvement into context: Nasdaq measures post-trade performance by looking at the magnitude of price changes after the trade, the so-called “markout.” Regular M-ELO already succeeded in price stability for 87% of trades, so the AI improvement brings that number approximately 10 percentage points higher. And those results are just based on the initial testing. Dynamic M-ELO is programmed to learn – readjusting its model weekly to continuously improve outcomes.

So - Why Is This New Order Type So Significant?

Dynamic M-ELO’s results speak for themselves: stronger order fill rates while minimizing the risk of adverse selection. That is a formula that institutional investors have been looking for, and now they can get it in a “lit” and regulated environment. “At the end of the day, this really is for the end investor,” Mack explained. “While AI may be a headline-grabbing term, what we're doing is taking the best developments in technology and finding ways to put them to work for our clients.”

“The U.S. market ecosystem is exceedingly competitive and it's not every day that you have the opportunity to reduce your instability exposure by 10-plus percent while also getting filled much more often,” Hamilton said. On a broader level, the deployment of Dynamic M-ELO will also mark the first time that an exchange has used AI to improve how markets themselves function. Artificial intelligence already influences strategy and decision-making for many market participants, and related technologies are applicable for trade execution, but Nasdaq will be the first example of an exchange using an order type that has a dynamic learning model helping to improve the infrastructure of our financial system. And that’s where the comparison to market digitalization comes in. “The difference between an open pit and a digitalized market in terms of market efficiency is huge: The information gets much more symmetric across different players, the rules are clearer, and the prices are much more transparent,” Hamilton said. “By the same token, the ability to have markets tune themselves and tune their parameterization to improve efficiency is the next frontier in this march toward more efficient and more transparent exchanges.”

When and how is Dynamic M-ELO Being Released?

The timing for Dynamic M-ELO’s release is still pending – but, with SEC approval secured, the new technology has cleared the major hurdles on the way to a public debut. In its initial rollout, Dynamic M-ELO will be available for just a few symbols, rigorously tested, and then slowly expanded to more and more of the market. As tests continue, Nasdaq will continue collecting side-by-side data to determine exactly how much better Dynamic M-ELO performs live compared to the current version. Once testing has been completed, every symbol traded on Nasdaq will have the option for a Dynamic M-ELO order. And, from there, the age of intelligent markets will have officially begun. 

Who will be able to use it?

Anyone transacting on Nasdaq will have access to the M-ELO order type when it launches.