Jonas Thulin broadcasts weekly video presentations to showcase his predictions for investors and the public. In the early days of the Covid-19 pandemic, he predicted the March 2020 trough in the US equities market, and later, the ensuing V-shaped recovery (economists visualize recoveries and recessions as V’s, L’s, U’s or W’s—the shapes they make when charted).

“We were fortunate to publicly call out the equity markets’ trough on March 24 last year, following the pandemic sell-off; we were lucky enough to hit the exact day,” says Thulin, head of the asset management business at Swedish private bank Erik Penser Bank. “And we argued back in April that based on the data already coming out then that in the third quarter, the US would be above 30%.”

Erik Penser is a very small bank, Thulin says, but that prediction helped it punch above its weight and win second place on Bloomberg’s ranking of forecasts for pinpointing the US gross domestic product (GDP) growth number.

For asset managers that, like Thulin, use macroeconomic data to understand the economy, high-frequency and alternative datasets to bolster macroeconomic analysis have become crucial in understanding Covid-19’s impact on the markets, and now that vaccines are becoming widespread, and some countries have reopened after lockdowns, they’re becoming crucial for understanding that also. For these asset managers, traditional measures like GDP just don’t get published often enough. Most countries calculate GDP on a quarterly basis; what if you need to understand what is happening in the equities markets this month, or even this week?

Thulin joined Erik Penser four years ago from Nordea, and immediately threw open its strategy and architecture to all asset classes. Rather than running an internal fund, his team’s strategy is to allocate assets to external managers in the EU, US, and Sweden, selecting these investments by crunching data—a lot of data.

Thulin says the bank has some 1,600 models and 28,000 time series that it reviews each week. This is, as he puts it, his team’s “core view of the world” on which they model all their decisions. “The strength in [using] data over, for example, allocating to an internal fund company is enormous,” he tells WatersTechnology. “Simply put, letting data, products and strategies compete in an objective methodology is key to the future of asset management from our point of view.”

This core, however, is only the short-term, high-frequency data, he adds. “When I do the [fundamental] analysis like the backdrop, asking where the business is going, then we are looking at tens of thousands more factors,” Thulin says.

Erik Penser Bank has partnered with Sweden-based macro data vendor Macrobond since the latter was founded in 2008, and draws all of its macro data from the vendor. During the pandemic, Macrobond added alternative datasets, such as plastic goods shipments, that Thulin’s team have used to make predictions.

Thulin says the thinking goes like this: analysts can’t wait for GDP figures, so they turn to indexes of industrial production to understand economic growth. These, however, are still insufficiently high-frequency, so the next step in understanding how much a country is producing in orders—economists use the US Census Bureau’s durable goods orders, for example, to measure industrial activity.

“But then all of a sudden, orders aren’t giving you the actual amount of order intake, so you go into shipments. And if you go into shipments, you have to find out what goods are ahead of the curve,” Thulin says.

Plastic goods are a good indicator, as orders for these are made early in the production cycles of goods, such as cars, that incorporate them.

“And if you want to understand the core of plastics, you must do the data analysis to see who has the fastest plastics data, and that turns out to be air shipments out of Frankfurt. So that is a great dataset to look at,” Thulin says.

Macrobond takes in data from some 2,000 sources, ranging from the traditional (World Bank statistics) to more alternative data from about 350 sources. Many of the newer datasets it has added in post-lockdown days—US airport passenger security checks, the movement of people in Japan, London pedestrian traffic—are data on human mobility.

Macrobond chief commercial officer Howard Rees says an analyst would have to wait a long time to see the impact of the relaxation of Covid-19 restrictions on GDP. “But you can start to look at some high-frequency datasets: how many restaurant bookings are there on [online restaurant reservation service] OpenTable, how many people are visiting different types of shops, and how many people are passing through airports? “We’re seeing more and more commingling of the traditional backward-looking, survey-type data and those higher frequency, snapshot-type datasets that tell you what’s happening right now,” Rees says.

Using Macrobond mobility data, Erik Penser’s asset managers calculated the relative strength of the reopening economies of the US and Europe, to arrive at an indicator that explained what was happening in the equities market, Thulin says.

The bank almost exclusively uses Macrobond data, but for very technical instruments, it combines it with Bloomberg data. “If you want to take positions on the US rate curve, for example, there is some pricing data that is quite nitty-gritty,” Thulin says. “Or when I have an inflation swap and I want to decide exactly where we want that inflation swap to start and stop, then we use Bloomberg.”

The firm also runs a sustainability fund for which it uses MSCI data. “We use about 20 million data points to study the sustainability effects of a company, and each company has 100 pages of numerical analysis behind it. When we enter that realm of super detailed, raw quant stuff, we are using complements to Macrobond,” Thulin says.

Data differentiator

Any analyst can build models and do math, Thulin says: it’s finding unusual data that is the differentiator of asset managers in a competitive market. Clients of providers like Macrobond are thirsty for data, and they are increasingly sophisticated, able to consume huge amounts of it systematically into models and algorithms. While a quant might use a GUI for a sanity check or some quick insight, their main business will be finding alpha in any given content set.

At the same time, clients’ migration to cloud providers like Amazon Web Services, and increasingly, Snowflake, has pushed Macrobond to become agnostic in how it sends data to customers, says Greg Haftman, head of data sales at Macrobond. Some clients might want to load the data on-premise, some in a cloud-hosted environment, but however they want it, it has to arrive in the same way and be accessible via a language like Python, he says.

Haftman joined Macrobond in November 2020 from FactSet. In his previous job, Haftman says, he saw how clients were beginning to want to integrate new types of data (audio for example, so they could analyze earnings calls) into relational databases, which struggle with data that is not uniform. Increasingly, clients want to diversify, and store unstructured data and semi-structured data. Macrobond doesn’t solely rely on web scraping to ingest data from its sources, but also captures data in semi- and unstructured forms.

Haftman says Macrobond’s founders made the decision early in the company’s history to store its time series data as Blobs—or “binary large objects,” data such as audio or PDF files or images that is stored as a single entity—in a relational database, rather than in tabular form, as is more common. This decision meant that Macrobond could scale the data it stored and sent to clients. In those days, it had about six million time series. When Haftman joined, it had 175 million; it now has 245 million, he says.

“Whatever growth we have, when we onboard new time series, when a user requests data from our server, they request entire objects, which are fully indexed, and that is available via the web API,” he says.

Haftman was brought into Macrobond to help build its new datafeed business, which it announced with the launch of a web API in May. The datafeed is a recognition that consumers need a high volume of data, delivered directly into their operating environments and accessible in statistical applications and programming languages like Python.

“Quants need to have access to data without limitation on what they can consume; in our case, in the number of time series they can consume. How much time does it take to pull out statistics on 2,000 time series, or a million time series? So there are two main ways to deliver data: either you push it or you pull it out: either Macrobond pushes it directly to the client’s environment, or the user requests the data on our server. So for us, our API was the strategic decisions we made for a pull mechanism,” Haftman says.

He says Macrobond is seeing a growing need for equities and fixed income data in almost real time for multi-asset strategies, and for trying to predict the present, or nowcasting. “So we are trying to get more and more data that has a higher frequency, and also diversify the contributors with alternative data, or data that we haven’t really looked at before in terms of investment strategy. That could be data that is not necessarily tied to a particular country, but could be tied to a city or state, for instance,” Haftman says.

Discoverability is another difficulty in providing clients with a large volume of data that is constantly updated: how can clients know what’s available and then find what they need? If Macrobond had, say, 1,000 time series, that could be listed in a PDF or online, or made available via an interactive portal. But it has 245 million. The company deals with this by categorizing the data in terms of themes, Haftman says. A user could search for a particular sector, for example, and the data is always linked to a country or region.

“The challenge is to make our users and prospects understand that 245 million time series might not mean anything. It’s when you start digging into how the data has been categorized into themes and regions, this is how it starts to make sense,” Haftman says.

Thulin says for the mobility use case, Erik Penser’s analysts could search for the term “mobility”, or read an article for clues on other kinds of terms that might yield useful data.

“Macrobond also highlights when they add data to their catalogs. The mobility data was one of those points. Our job is to make the math work and say, OK is this actually explaining something in the market? Is this what we are trading on till the paradigm shifts and we move on to something else?” Thulin says.

Erik Penser Bank has been a client of Macrobond since 2008, and was a client of its predecessor company Ecowin, which was sold to Thomson Reuters in 2005.

While large data vendors like Bloomberg provide macro data, there are some other specialist vendors. Macrobond’s main competitors are Refinitiv’s Datastream, CEIC, and Haver Analytics.