New technologies changing asset management

Technology is transforming the way we work. Asset managers are no exception as they seek better investment outcomes for clients.

Each second, oceans of new data are being generated by the Internet, smartphones, satellites and other innovations. This data is commonly referred to as “Big Data” or “alternative data.” Many asset managers are seeking to harness the power of Big Data by using technologies like natural language processing, image recognition and machine learning to analyse it and uncover new investment insights. BlackRock’s Systematic Active Equity (SAE) investment team has been using these technologies to analyse alternative data for more than a decade. Over this time, the team has learned four key lessons about how to make the most effective use of these technologies within an investment process.

Capital at risk. The value of investments and the income from them can fall as well as rise and are not guaranteed. Investors may not get back the amount originally invested.

Lesson 1: Target the “right” data and multiple data sources

Investment teams looking to harness the power of Big Data must cast a wide net when seeking to find the right data to enhance investment outcomes. Why is that? The global economy and financial markets are highly complex, so the data they generate are often unstructured and noisy. It is important to evaluate as much data as possible to determine which may be most useful within an investment process. SAE researchers trialled over 70 new datasets in 2017 alone.

Yet data alone does not translate to alpha. It requires proper processing and analysis to arrive at investable insights. Asset managers have tripled their spending on data, as shown in the chart below. Yet more important than the amount of data obtained is the distillation and utility of that information in investment decision-making.

More dollars directed to data

Total buy-side spend on alternative datasets ($mm)

More dollars directed to data


Source:, as of June 30, 2019. The above estimates are those of There is no guarantee that forecasts made will come to pass.

The data itself needs to be tested for its quality and additivity to a forecasting model. When SAE onboards a new dataset, it is cleaned and run through a battery of statistical tests to determine the value of the information it provides. In technologically advanced regions such as the U.S. where company information is abundant, the bar to finding new data that is useful is high. After years of experience, SAE has learned that new data is not always additive to existing models. However, you can’t know that until you try. And if not successful immediately, these are often ideas that can be used for future projects.

Once the “right” data is found, it is highly unlikely that a single dataset will produce accurate forecasts. Take analysing data on consumer behaviour when seeking to forecast future sales growth of a retailer. Geolocation data sourced from mobile phone beacons can be useful in recording consumer foot traffic in physical stores. But foot traffic alone does not mean any sales occurred.

Fortunately, there are many data sources that can shed light on consumer intentions and help inform a view of a retailer’s sales growth. At one end of the spectrum, analysing aggregated Internet search activity can capture changing consumer sentiment toward a company’s brand or products. Still, data on Internet search activity has a long forecasting horizon and less forecasting accuracy, as consumers must take several steps after an Internet search before they purchase. At the other end of the spectrum, aggregated consumer transaction data recorded by banks and credit card providers can help track actual consumer spending (and ultimately, reported revenues).

We find that using multiple data sources to corroborate one another and answer the same investment question can significantly improve the quality of forecasts. When using aggregated consumer data, there needs to be a strict procurement and compliance framework to protect individual privacy. As a part of BlackRock, SAE has a variety of compliance rules it needs to adhere to when using data sources. SAE only uses aggregated and anonymised data to deliberately avoid accessing personally identifiable information.

The more data sources the better

Alternate data sources used to project sales growth

Alternate data sources used to project sales growth


Source: BlackRock, as of August 2019. Provided for illustrative purposes only, not meant to depict actual data

Lesson 2: Technology is only as good as the hands it’s in

The ability of technology to analyse a massive amount of data within seconds is impressive, but it is human expertise that brings technology’s power to life. Take machine learning. Not only can machine learning algorithms analyse reams of data in a flash and generate insights, they can determine relationships between a vast number of data inputs in a way that adapts to changing data patterns. In doing so, machine learning can learn faster than humans while also quickly analysing large amounts of data.

Though the computational and adaptive ability of machine learning exceeds human ability, the ultimate success of the application of machine learning (or any technology) relies on human expertise to develop and refine. SAE has built up its data and computer science capabilities over the past decade through hiring and developing the talent to both create and improve the technologies used in our process.
The chart at below shows the sharp increase of alternative data employees in the asset management industry.
Risk. There is no guarantee that research capabilities will contribute to a positive investment outcome.

In Demand

Full-time alternative data employees in asset management, 2000-2018

Full-time alternative data employees in asset management, 2000-2018


Source:, as of June 30, 2019

What does that research expertise look like in practice? Take signal combination. Signal combination is a machine-learning model SAE researcher developed to understand the relationship between stock returns and a vast array of quantitative data, including accounting information and analyst forecasts. Our research found the model was treating periods of stable performance and infrequent episodes of drawdown with equal importance when analysing market data. (Source BlackRock, March 2020) This observation led to an important change where the algorithm was refined to penalise drivers of extreme negative short-term performance. Put another way: SAE researchers had to calibrate and train the model to think and learn like an investor.

When designing a machine learning technique, it is also important to consider what investors have already learned about markets. Making sure new technology is adding value to existing models is a key tenet of any piece of SAE research. Thus, the expertise that is needed encompasses not just the ability to create and refine technology but also knowing to ask the right questions to vet ideas. For example, a highly sophisticated model using cutting-edge techniques, which simply rediscovers well-documented stock market behaviours would represent an inefficient use of research resources. In practice, we encourage the technology we use to reveal insights about what we as investors know less about.

Lesson 3: Technologies can― and do―lose their potency

Markets are continuously changing, and that means investment opportunities have a lifecycle. To keep up with new opportunities as others fade, an investment team needs to be relentless about 1) finding new data sources and 2) innovating the technologies used to analyse that data.

Case in point: SAE has evolved its natural language processing techniques for analysing corporate language to help forecast company prospects. The initial work focused on measuring sentiment in broker reports or company filings through counting the use of positive words that had been predefined by SAE researchers (e.g., “upgrade,” “growth,” “opportunity”) versus negative (e.g., “downgrade,” “threat,” “competition”). As time went on, SAE began to focus on the source and target audience of the text. One finding: remarks by CEOs tended to be more positive and scripted, so focusing on the CFO’s remarks or the Q&A sections of the call proved consistently more useful. Comparing sentiment from one source (such as information given to the market) with another source (such as information given to a regulator) also helped to identify instances of company spin versus hard facts (see the example below).

press release


Source: SEC EDGAR. Paul Ma, Information or Spin? Evidence from Language Differences Between 8-Ks and Press Releases, 2012.

Finally, a material innovation came through combining natural language processing with advanced machine learning techniques. SAE innovation no longer relied on researchers having to program a set of predefined words. By applying learning algorithms, the technology learns in an adaptive way what words are most important for forecasting future returns and fundamentals for a given stock.
SAE’s continuous innovation in machine learning has made corresponding insights very different from the ones used eight years ago, thereby creating an ability to find new investment opportunities.

Lesson 4: Collaboration breeds innovation

The reams of data generated by technologies like the Internet and smartphones can be structured to provide investment insights. While dedicated data and computer science expertise is required within an investment team to analyse the data, our experience in integrating teams with diverse skill sets has shown us that having specific expertise on a team is not enough.

Successful innovation requires a culture of collaboration and constructive debate across diversifying disciplines. Team members—from senior to the most junior—must be encouraged to work together regularly to improve current data techniques and to test ideas in a beta environment, without fear of failure.

SAE teams regularly engage in “hackathons” where team members work together and compete to create technologies and new uses for current technologies that can aid the investment process. These cutting-edge tools may find new sources of alpha for all SAE strategies.

Collaboration requires knowledge sharing, which can be facilitated with the right tools. For example, SAE’s more than 30 years* of research is accessible to all its team members. This open archive supports efficient idea generation through documenting both successful and unsuccessful projects.

A culture that enables innovation to thrive can require many years to create as it depends on the right leadership and team chemistry. Yet, without it, effective innovation may not be sustainable.

* Includes time at predecessor firms.

Jeff Shen
PhD, Managing Director, Co-Head of Systematic Active Equity (SAE)
Raffaele Savi
Managing Director, Co-Head of Systematic Active Equity (SAE)
Richard Mathieson
CA, Managing Director, BlackRock's Systematic Active Equity Investment Group.