Data vendor Six is working on a prototype to mine data on US dividends. The prototype reads and filters out press releases on dividends from news feeds Six consumes, and transforms the text messages into a structured record that can be processed electronically.

Dominique Tanner, head of content management at Six Financial Information, says if the prototype goes live—which Tanner estimates could happen in about a year—customers will save time by not having to wait until the corresponding exchanges release dividend announcements. 

Currently, Six sends out the structured corporate actions information once the relevant exchanges release it. “What we’re seeing from our customers in terms of the demand is that we have to be faster with supplying the corporate actions information,” he tells WatersTechnology.

Clients are looking for more timely delivery of corporate actions data to have a better handle on growing volumes and complexities of corporate actions events. The prototype won’t necessarily mean real-time delivery of dividend information, but it could be near real-time.

“When a company is declaring a dividend—so for example, IBM says, ‘I paid $0.25 on this date,’ and they announced it at 10 a.m. US time—people would expect that maybe an hour or two later, they would see it on our systems as well, or get the notification from us,” he says.

Usually, Six would get the corporate actions information from the corresponding exchanges and process it into structured records, which would only then be fed into its datafeeds to clients.  

Exchanges act as a collector and aggregator of corporate actions information for their listed stocks. Typically, that is put into a file sent out overnight or at the end of the day.

“You’ll get a file from the exchange that says, ‘Here are the 25 companies that have announced dividends today.’ That has a time lag because the company might have already issued the press release in the morning or the day before,” he says.

This is problematic for customers who want faster access to the information due to growing volumes and compressed processing deadlines.

Eiichiro Yanagawa, senior analyst at research and advisory firm Celent, says growing volumes and complexities driven by increasing digitalization require better straight-through processing (STP). The evolution of regulations and market practices will require greater process automation and automated workflows, he adds. Technologies such as robotic process automation and machine learning-enabled solutions can help in this area. 

“As always, the mother of innovation is its necessity, and the father is competition. Since this space is a globally common competitive area, many vendors have already shifted from the race for its development to the race for its delivery [business] model. A key success factor is to provide a low-cost and robust utility service around the exchanges,” he says.

If Six’s prototype is put into production, it would be imperceptible to customers—they would simply get the same dividend information they were getting previously, only much faster.

The prototype delivers dividend information within 15 to 30 minutes of the company announcement, which would represent a significant latency improvement.

“We would have received the information from the exchange and then would have processed it overnight. Usually, it would be sent out overnight as well. But, overnight, no one is really looking at it. So customers would have seen it the next morning. So we could really reduce this delay, which is probably up to a 24-hour delay, to maybe 15 to 30 minutes,” he says.

This applies particularly for US markets, while other rules could apply in other markets.

“The US is already very efficient, but other markets are less efficient and there might be a time delay of maybe two to three days. Less developed markets could also benefit from gaining time efficiencies,” he says.

The prototype is currently experimental, and Tanner says there is no set date for when Six plans to put it into production. “[When we began], we said we want to increase our know-how in applying these technologies and gain experience,” he says.

‘Pretty simple’ machine learning

Tanner says Six deployed “pretty simple” machine learning to the prototype, as dividend press releases are generally released in the same format in the US, compared with other markets. “They follow a very similar format, and it doesn’t matter whether Microsoft or Google is announcing a dividend; they’re pretty standard text messages,” which is more conducive to machine learning than if they were completely different, he says.

Using IBM as an example, Tanner says a release would read something like, ‘IBM; NYSE is announcing a quarterly dividend for the first quarter of 2021 of $0.25, payable on this date.’

The prototype currently has recognition rates of close to 80% for simple dividends.

“That actually means we can do more automation or STP on our end using AI to transform simple press releases into structured records that our customers would want to process,” he says.

For the prototype, Six used examples from past press releases and tagged the relevant information. “The tagging is kind of a manual task,” he adds. “You take the press release, then you tag, let’s say, a date as the record date, then you mark another as the currency, then the stock symbol—so we actually know which stock the dividend belongs to—then the exchange, and so on.”

Tanner says it is rare for an issuer to communicate the ex-date, adding that the corresponding exchange usually sets it. But because Six knows the rules for various exchanges, it can derive the ex-date based on its records.

“Now we have those 15 data fields, which the machine can extract from the message, but there is one field missing, which is the ex-date. Then we have a rule that says if it’s NYSE-listed, we take an offset of plus one day from the record date adjusted for any non-business days, and that’s the ex-date,” he says.

A Six data expert can see the structured data on one side of the screen and the original message on another screen. Once the exceptions are managed, amended, and completed, the record is fed into the database, which would then be delivered to Six’s customers.  

Six aims to get the prototype to 90% recognition for simple dividends, and to 75% to 80% for more complex dividends, before it is put into production.

Virginie O’Shea, CEO and founder of Firebrand Research, says one of the greatest challenges in the corporate actions space is the number of manual processes involved in rekeying and confirming data related to various events.

Research conducted by Firebrand indicates that an average of 64.2% of events are automated. The rest are plagued by costly manual intervention and higher operational risk of errors made while rekeying data.

While simpler event types such as dividends tend to be automated at the event capture stage, she says the later stages of the lifecycle can be trickier due to the nature of the data flows across multiple intermediaries. 

“The best way to contextualize the challenge is that issuers are working with lawyers that generally don’t think about the downstream processing of the data—their priority is meeting legal obligations and therefore the documentation is often not easily translated into standardized information,” she says.

AI is good at recognizing patterns but requires a lot of model training, O’Shea adds. It could be trickier for other corporate actions data, which is generally not standardized. 

“Exchange data tends to be a lot more standardized than the original documentation received from issuers, so hopefully this will save time for firms, and dividends are on the simpler end of the 70-or-so corporate actions types. Of course, firms may still choose to double-check data because of the cost of errors,” she says.