Improving investment returns via new insights from new data sources

It is universally acknowledged that information is power and since time immemorial there has been a race for the power that an information advantage can confer. Venetian merchants used Galileo’s telescopes not to study the heavens, but rather to study the cargoes of approaching ships a few hours before their competitors. Recently, a range of alternative big data sources have emerged that offer this promise of an information advantage to the buy side. Right now, 31% of hedge funds are using alternative data and analytics in their investment process, while another 21% expect to adopt it in the next two to three years. Leading asset managers are following suit and have begun the journey of integrating new data sources into their investment process, to augment their existing fundamental analysis. In this article, we’ll consider examples of the big data sources that are now available, illustrations of their potential benefits, the barriers to realizing these benefits, and the consequent opportunities and threats for the buy side.

Traditional and alternative data

Traditionally, investment analysts have used a range of long-established data sources, such as company filings, earnings announcements, investor presentations, sell-side research, market data feeds and financial news. Now, in addition, there are many new alternative big data sources that can augment these traditional sources to provide new insights for the investment process. In the same way, there are many different types of alternative data: web, app and social media; satellite; location; email; local price and economic data; and credit card transactions, to name but a few. Potentially, each can provide new or earlier insights compared with traditional sources. Leading asset managers have recognized this opportunity and are acting.

Examples of the potential benefit of using alternative data

In order to demonstrate the potential advantages of applying alternative data in the investment process, we will present four examples and illustrate their potential benefits.

… there are many new alternative big data sources that can augment these traditional sources to provide new insights for the investment process.- Phil Tattersall, Director, EY

Example 1 illustrates how social media sentiment can provide new insights. ‘Samsung exploding’ was being reported via social media many days ahead of a 10.6% fall in the Samsung stock price over two trading days. At the time, Samsung was the largest constituent of the MSCI Global Emerging Markets Equity Index.

Example 1

Example 2 illustrates how satellite imagery can also provide new insights. On 22 April RS Metrics used satellite imagery to forecast Tesla sales volumes accurately 10 days before the official company announcement, which precipitated a rapid price fall of 12.5% in three days.

Example 2

Finally, Example 3 illustrates how location and app data can provide new insights. On 12 April, Foursquare used foot traffic stats from its app data to forecast Chipotle sales more accurately than the earnings consensus. This was 14 days ahead of the official company announcement and a significant price fall (over 6% in a day).

Example 3

These new alternative data sources can augment traditional sources in various ways:

• A much richer picture than before can be built from orchestrating information across different alternative data sets, e.g., credit card transaction data plus web search plus app downloads plus foot traffic and so on. In addition, “Big data can enable asset managers to see hidden connections and relationships between companies, including across industries … leading to a potential advantage in selecting investments,”¹ it is stated in a recent report by Goldman Sachs.
• Moreover, in all the above cases, the insight from the alternative data sources was available earlier than from traditional sources.

The challenges of using alternative data in the investment process

Generating insights from alternative data sources is hard: firms will face many and varied challenges. A Citi report states that “Most investment firms are not yet focused on big data because they lack the institutional momentum, the skill set, and the business case to build out these capabilities in the short term.”²

The challenges of using alternative data can be summed up as follows:

1. There is not yet a universal recognition of the potential benefits of using alternative data sources.
2. Identifying the new alternative data sources that will be useful and additive to a firm’s investment process is difficult. For example, there would be no point in investing in sourcing alternative data that was highly correlated with traditional sources, such as earnings revisions. Fundamentally, firms will need to experiment to learn what is most useful for them.
3. Engineering the platform to access the alternative big data sources is also difficult. Many asset managers have already been on a long journey establishing the data platforms for their investment processes. Alternative data is very high volume, more varied and less structured, needs more cleansing and is often less accessible than traditional sources. New data architectures are required.
4. Generating insights from all this alternative data is tough and needs new techniques. No one would have been monitoring social media for ‘Samsung exploding,’ but the negative sentiment and connotations from this term could have been identified.

Applying machine learning to interpret the mass of alternative data

Machine learning aims to construct solutions that automatically improve with experience, without being explicitly programmed and without human intervention or assistance. There are different techniques but, for supervised learning, the algorithm is trained on a data set (the training data) that is sufficiently large and diverse to enable the algorithm to perform effectively in the real world.

Two recent impressive achievements are in the fields of Go and poker:

• AlphaGo, the board game-playing artificial intelligence (AI) program from Google’s DeepMind subsidiary, beat the world champion Lee Sedol early in 2016.
• In January 2017, an AI poker bot called Libratus played a 20-day tournament of no-limit Texas hold’em against four top professional players making a profit of US$1.8 m.
• In each case, the algorithm played itself repeatedly (billions of Go games for AlphaGo, trillions of poker hands for Libratus) and learned to improve over time. Given the volume and data quality issues, these sophisticated machine learning approaches are a necessity in analyzing the huge, complex alternative data sources and identifying relevant insights for the investment process.

The future: more data and more powerful processing to generate better insights

In the future, the torrent of alternative data will continue to increase exponentially; there will be many, many more sources of potential insight. A primary reason is that more and more firms will attempt to monetize the data that is produced as part of their everyday operations. For example, Twitter’s primary purpose is as a communications platform, but the data gathered, the ‘data exhaust,’ has value to others, and aggregated data from the Twitter platform is now being sold to hedge funds and others. The Internet of Things will accelerate this trend, as it generates an unimaginable amount of real-time data on everything from wearables to autonomous vehicles and sensors.

Firms will become increasingly innovative. For example, in March 2017, Acadian Asset Management announced a partnership with Microsoft’s Bing Predicts (a machine learning project that mines internet search and social media data for factors to try to forecast events, such as a corporation’s quarterly results).

The machine learning techniques to generate insights will also continue to be enhanced:

• Many of the machine learning techniques originated in the 1980s, but it’s only with the increase in computing power that they’ve become viable. As computing power continues to double every 18 months, so will the power of the algorithms.
• A long-standing tenet of machine learning is that more training data results in improved performance³the exponential increase in alternative data allows ever more powerful training of the machine learning algorithms.
• New, more sophisticated machine learning algorithms will be devised as industry use of machine learning techniques matures.

An increasing number of active asset managers will be informed by additional insights from more alternative data sources that have been analyzed by machine learning techniques.

Conclusion: there are significant opportunities and threats for asset managers

All investment performance is necessarily relative: league tables and beating the benchmark are more important than raw investment returns. The opportunity for asset managers is to identify new insights that others don’t have, and augment their investment process to improve their relative performance.

The leaders are investing in refining their capabilities in this space. ‘We believe that in order to generate sustained alpha, investors should embrace acquiring, analysing, and understanding the fast-growing universe of data. Those who are unable to do so run the risk of falling behind in a rapidly changing investment landscape.’[4]

Conversely, the threat is that relative returns will be eroded if your investment process does not use these new insights when other firms do. Many asset managers do not yet have a mature capability in this area, and developing it will be difficult; particularly given the range and diversity of the potential sources of insight.

Investors primarily buy the ability to generate investment outcomes from asset managers. If this ability is eroded even marginally, because others in the industry have successfully innovated with new ways of improving their investment process and returns, this is a significant risk for an asset manager. We recommend all active managers begin their journey to augment their investment process with these new insights from alternative data sources.

Disclaimer:
This article contains information in summary form and is therefore intended for general guidance only. It is not intended to be a substitute for detailed research or the exercise of professional judgement. Member firms of the global EY organization cannot accept responsibility for loss to any person relying on this article.

About the author

Phil Tattersall, Director, EY

Phil is a Director in EY’s UK Wealth & Asset Management Data and Analytics advisory practice. With 25 years in the industry, he has extensive experience helping firms enhance their data and technology platforms, including developing firms’ data strategies, operating models, architectures & governance frameworks, as well as selecting and implementing new solutions and platforms. Most recently, Phil has been focusing on how asset managers and security servicers can apply advanced analytics and robotics throughout the value chain.

1. The Data Revolution, Goldman Sachs Asset Management, 2016.

2. Big Data & Investment Management: The Potential to Quantify Traditionally Qualitative Factors, Citi, 2015.

3. M. Banko and E. Brill, Scaling to very large corpora for natural language disambiguation, Annual Meeting of the Association for Computational Linguistics, 2001.

4. Raffaele Savi, Jeff Shen, Brad Betts, Bill McCartney, The evolution of active investing: finding big alpha in big data, BlackRock, 2015.