Thursday was Thanksgiving in the United States—maybe elsewhere, too, but American hegemony insists that I only care about Pilgrims and Indians. I’m thankful for many things in my life, but when it comes to work, beyond the amazing reporters I get to work with on a daily basis, I’m very thankful for our subscribers.

This might be a little bit “inside baseball,” but the majority of our revenue here at WatersTechnology—and, for that matter, at our sibling publications Risk.net, Central Banking, and FX Markets—comes from subscriptions, and not events, marketing services, or advertisements. It’s largely because of this business model that we’ve been able to successfully navigate the pandemic (thus far).

When the Financial Crisis in 2008 hit, I saw lots of excellent trade magazines go under, and the pandemic hasn’t spared them either, like we saw with the shuttering of Profit & Loss, which is truly a loss for the industry. One thing most in the trade space learned in 2008, and again in 2020, is that if you don’t have people paying for your content, you don’t have much of a business model. 

Which brings me to this column. I started putting together the Waters Weekly Wrap back in mid-July and had it in front of our paywall so as to give people a taste of what this post would be about. This edition, and all subsequent editions, will be behind our paywall. I hope that the information provided here—and on the website, as a whole—is valuable. Obviously, you’ll let me know whether it is by continuing to subscribe, but do not hesitate to let me know how we can improve: anthony.malakian@infopro-digital.com.

Thank you for your support and your readership.

ML, NLP & CoV-2

I’ve written it before, but the pandemic is a proving ground for all the product or services that promise to perform under stress to…well, prove it. Lately, in the world of buy-side trading, it’s machine-learning algorithms coming under the microscope.

To highlight the environment facing asset managers today, Faye Kilburn spoke with portfolio managers at eight buy-side firms to see where their ML models struggled during the pandemic, and how they’re tweaking them to better perform under unexpected stresses in the future. Michael Heldmann, head of multi-factor equity investing for North America at Allianz Global Investors, told Faye that these models failed “pretty spectacularly.”

“They have been hammered during Covid, especially in the beginning, when they were betting on things that have been very successful in the long term, like betting on a big drawdown not being followed by another big drawdown. That has led to massive underperformance in those models,” he said.

I’ll let the article speak for itself, but the broad theme is that quants train ML models “on lots of long-term data,” but it is difficult to build enough adaptability into an ML model so that when situations change on a short-term horizon, the model can adapt without overfitting decades of data.

“The recent struggles of some models underscore the static nature of the way that many quants view the world: what worked decades ago should work decades from now. But the reality of a changing world seems to repeatedly interfere with the theory,” said Andrew Beer of Dynamic Beta Investments.

One way to balance these models is to train them using shorter time windows, and here is where natural language processing (NLP) comes into play. From the story:

One of Allianz’s best-performing signals during the last six months, for example, has been to use NLP analysis of earnings call transcripts and news flow to train neural networks to predict future returns. Heldmann attributes its success to the speed at which the strategy picks up on change.

Because NLP algos often are trained on more abundant data from outside finance, quants can afford to train them on a much shorter time period, Heldmann says. “For these NLP models, you need just a couple of years and not 10 years. And you can train them in a pretty reasonable way.”

As Joanna Wright wrote earlier this year, advancements made in the field of NLP—and the use of open source in this field—have helped to democratize these models, allowing cap markets firms of all stripes to experiment by combining NLP with ML and other forms of AI to build unique alpha-driving strategies and tools. Some examples: Lazard, JPMorgan AM, BNP Paribas AM, Ping An Insurance Company of China, Bloomberg, Manulife, Jefferies, MSCI, Brown Brothers Harriman, Morgan Stanley…I could go on.

I wrote about this previously, but just three years ago, most of the talk in the capital markets was about using machine-learning models and algorithms, but now I get the feeling that it’s NLP getting all the attention. Obviously, there’s overlap between ML and NLP, but it does seem that more advancements are being made in the field of open-sourced NLP tools.

I guess it’s more of a balancing act than one side winning over the other, but to me it feels like ML teams are on an island, whereas NLP is a world of collaboration and experimentation, and that’s where cap markets tech development is zeroing in.

Think I got it wrong or you have a different idea, let me know: anthony.malakian@infopro-digital.com.

(If you’re interested, the companies Faye spoke to for her feature were Allianz Global Investors, BNP Paribas Asset Management, AllianceBernstein, XAI Asset Management, Abu Dhabi Investment Authority, Dynamic Beta Investments, Numerai, and PineBridge…seriously, in my opinion, when it comes to depth of buy-side sources that incorporate machine learning, there is not a better reporter in the world than Faye—that’s not hyperbole. Some other examples: ML & ESG, ML & alt data, ML & interpretability, ML & correlations, ML & risk modeling.)

Touching the Cloud

Earlier in November, Hamad Ali caught up with Adrian Ip, director of product management and technology sales at Aquis Exchange, after the exchange operator’s technology division had announced a successful proof-of-concept with Amazon Web Services (AWS) and Singapore Exchange (SGX). The project is aiming to show that complex exchange trading platforms can work as efficiently in the cloud as in physical datacenters when it comes to the low-latency transfer or data and latency variation, or jitter.

What was interesting is that Hamad learned that Aquis was able to cut jitter down from 200 microseconds—a “quite troubling” number, as Ip put it—at the start of the PoC in late spring, down to 4 microseconds in the cloud. For context, the Aquis Trading Protocol—the exchange’s own binary trading protocol—has a jitter range of just 1 microsecond.

This news story builds off of a feature from earlier this summer, where Hamad spoke with execs from Nasdaq, London Stock Exchange, AWS, Google Cloud, Microsoft Azure, and others to find out if exchange operators can move all of their exchange operations to the cloud.

The problem with the cloud when it comes to core trading platforms at exchanges is that it doesn’t support multicast data delivery, which is the accepted method for distributing data from one source to multiple recipients at the same time. One of the biggest problems with “streaming” exchange market data in the cloud is the latency, or more specifically, that latency is not deterministic—i.e. the speed isn’t low-latency, nor constant, so messages can theoretically arrive out of sequence.

For the Aquis PoC, the exchange is working with AWS. Nasdaq is also using AWS as it looks to move to the cloud, and CME Group has a market data project underway with Google Cloud. I recently wondered if these partnerships might prove risky for exchanges—what happens if the cloud providers decide that they’re better at providing market data and trading tools than the exchanges—but as I have also been hammering home in this column, cloud is not the future—it’s the here and now—and every institution on Wall Street needs to get up to speed, and that obviously includes exchanges.

Will we see a full shift to the cloud in the future? Well, Nikolai Larbalestier, head of cloud strategy at Nasdaq, recently told us that it’s not a matter of “if” but “when”, and he thinks that the day we’ll see exchanges running on the cloud will come sooner than many might think. Larbalestier is far more intelligent than yours truly, so I’m not going to counter that prediction—and it would appear that Aquis’ progress with AWS and SGX is inching toward Larbalestier’s prediction.

The image at the top of the page is Frederic Edwin Church’s “Niagara” courtesy of the National Gallery of Art.