For investment firms looking to extract signals from alternative data, finding an interesting dataset is only part of the challenge. Moving that data from research into production brings a different set of requirements around quality, timeliness, consistency and historical accuracy.
Those requirements are particularly important for quantitative investment firms, where apparently small discrepancies in timestamps, identifiers or historical mappings can affect both backtesting and live models.
EventVestor approaches this problem from a somewhat different direction to many alternative data providers. Rather than deriving information from third-party sources such as credit-card transactions, web activity or satellite imagery, the company focuses primarily on information disclosed directly by listed companies.“We consider ourselves first-order alternative data because our data is sourced from primary sources – the companies themselves – rather than third parties,” Anju Marempudi, CEO of EventVestor, tells Market & Alt Data Insight. “Around 95% of our data comes from company disclosures: press releases, IR websites, conference calls and filings.”
EventVestor structures this information across more than 80 corporate-event categories, capturing individual attributes and data points within each category. Crucially, it also records when information became publicly available.
“The timestamp is important because that’s when the market first knows about the information,” says Marempudi. “We maintain the data point-in-time, with a clean structure and clear identifiers.”
Adding context to alternative signals
EventVestor has collected US corporate-event data since 2007, giving quantitative researchers close to two decades of history. More recently it has expanded its coverage into markets including Europe, Japan, India and Korea, largely in response to demand from existing customers.
Its client base is concentrated among quantitative hedge funds, proprietary trading firms and market makers operating in equities and equity options. But Marempudi sees a broader opportunity as more investment managers incorporate quantitative analysis without necessarily becoming fully systematic.
An important part of the proposition is not simply using corporate-event data as a standalone source of signals, but combining it with other datasets to provide context.
A credit-card dataset, for example, might identify a sudden increase in transactions associated with a particular product. But a corporate-events dataset may already show that the company has announced a product launch. Equally, a sharp decline in transactions might be easier to interpret if the company has previously announced a product withdrawal or recall.
“Even if you’re using other alternative data, having corporate-event data as an overlay can provide much better context,” says Marempudi. “Rather than taking third-party alternative data at face value, you can understand what the company itself has already disclosed.”
That also helps address concerns around alternative-data crowding. Multiple firms may consume the same underlying datasets, but the resulting strategies can be very different depending on which events they select, how they combine them with other information and the time horizons over which they trade.
“We can have multiple teams within the same customer trialling the data,” says Marempudi. “One team might find alpha and want to move forward, while another doesn’t find much. Their mandates, risk profiles and approaches are different.”
The point-in-time challenge
These issues become more significant as alternative data moves from research into production.
Marempudi identifies reliability, timeliness and point-in-time integrity as some of the biggest obstacles firms encounter. Data that appears useful in a trial may behave differently when incorporated into a production process, while seemingly mundane issues such as identifier mapping can introduce errors into historical analysis.
Corporate relationships themselves change over time. A brand or business unit associated with one listed company today may previously have belonged to another, for example, following an acquisition, disposal or spin-off. Accurately mapping transactional or other alternative data therefore requires knowing which entity it related to at the time the observation was made.
EventVestor maintains point-in-time ticker, CUSIP and ISIN mappings alongside historical corporate affiliations, allowing customers to relate events and other datasets to the appropriate company at a particular moment.
For Marempudi, another important characteristic of primary-source data is that it can be verified. “If we say a company announced its earnings or guidance at 8:02 this morning, you can go back and validate it,” he says. “You don’t simply have to believe us.”
He cites a recent case where a prospective customer was receiving conflicting information from different vendors about the scheduled time of a company conference call. EventVestor was able to provide the original company disclosure as well as its history of the event over the previous ten quarters. The prospect subsequently began a trial.
From models to markets
The example illustrates a wider challenge for investment firms as alternative data becomes embedded more deeply within investment and trading workflows. An approximate answer may be sufficient for exploratory analysis; production systems have a much lower tolerance for errors.
“With AI and LLMs, being 80%, 90% or 95% accurate can sometimes be considered good,” says Marempudi. “With this kind of data, even 99% isn’t necessarily good enough. You need to be as close to 99.99% as you can get.”
No dataset can be completely error-free – even companies themselves occasionally issue incorrect figures or subsequently amend disclosures. For investment firms, however, the critical requirement is to capture as accurately as possible the information that was available to the market at that point in time.
As alternative data matures, the challenge is therefore shifting beyond simply finding new sources of information. Increasingly, the differentiator is whether those datasets can be reliably combined, mapped, placed in their correct historical context, and ultimately moved from promising research signals into production investment processes.
Anju Marempudi will discuss these issues as part of the “Alternative Data in Production: From Models to Markets” panel at the A-Team Group/Eagle Alpha Alternative Data Conference in New York in September.
Subscribe to our newsletter
