For quantitative strategies, pools of capital that aim to deliver alpha by analyzing vast amounts of data, the advent of Artificial Intelligence (AI) technology such as Natural Language Processing (NLP) powers them towards a new frontier by multiplying in orders of magnitude the amount of information they can harness to deliver investment success.

One area that has proved especially fruitful when it comes to leveraging  this explosion in data availability and computing power has been the ability to construct groups, or networks, of companies and use these to reveal connections and insights that inform investment decisions and strategies. Uncovering the hidden connections among and between companies stands out as a particularly rich and underexploited source of alpha.

Specifically, instead of basing an investment decision solely on a company and its fundamentals, we can now also assess information about the other businesses that the company in question is connected to and use the learnings from those relationships to build more resilient portfolios.

These networks can help us better understand a company’s place within its ecosystem, how it is perceived by competitors, how it interacts within its supply chain and how information about these connections seeps into the market. Capturing this data and using it to fuel investment decisions blends economic insight with data science and can improve returns, for example, by refining the implementation of proven quant strategies.

Today, the most compelling opportunities lie in the nuanced, the uncorrelated and the hidden. A quant strategy that has the ability to identify and capture these previously unavailable data sets, and applies it with economic rigor, gives itself a significant edge.

How we capture information

Over the last decade, the ability to accurately extract information from text has evolved significantly. At the same time, the cost of the computer power required to achieve this has shrunk equally dramatically.

The advent of AI, NLP and large language models (LLMs) has enabled the translation, interpretation and automated processing of company filings and reports, without the need for human (read: fundamental) analysts.

It is estimated that around 50-60% of global data is text-based, rising to 70-80% when assessing data relevant to businesses — annual reports, earnings calls, news articles, and so on.

Even if we narrow the field to a more manageable size, the amount of text is still overpowering — or would have been in the pre-AI age. For example, there are almost 8,300 public companies in the MSCI ACWI IMI index that, in aggregate, have issued ca. 230,000 past and current filings in 21 languages. The computing power available today makes short work of digesting and interpreting this vast wealth of data, irrespective of the language it is in. From an investment standpoint, this makes for an invaluable treasure trove of actionable intelligence.

Any kind of investing relies on access to information, but this is especially the case with quant strategies. We use AI tools to analyze and collate this enormous cache of intel and we use it to build and assess networks; collections of  ‘nodes’ and ‘edges’ representing companies and their linkages. The nodes are the individual entities within the network and the edges are the connections between them.

For us to apply our network theory in practical, investment terms, we need first to vectorize this textual information — that is, we must convert the raw text into numerical data that can be understood and used by algorithms that drive quantitative investment processes. Once we have the data in the form we need it, we can then start to use it to build networks.

Take a supply chain network as an example. Each node is a company within that network while the edges are where they are related, such as supplying parts or distributing products. We might build out this network by assigning weightings to the edges according to the transaction value between two companies in the network, or as a percentage expressing the relative importance of the relationship.

Another example might be a network of companies that are linked by technology. We can build this by using patent-ownership data, such as scraping information from global patent databases. In this scenario, we can assign numerical ratings according to each company’s share of each technology class.

These are but two ways we could map the relationships between sets of businesses. In general, any type of connection between companies can be used to construct a network, applying similar techniques to that described above.

Implications

Having demonstrated ‘the how’, the obvious follow-up question is ‘the why’. What advantage does creating these networks provide to our quantitative strategies in practice?

One of the main benefits is in the way we deploy proven style factors, such as momentum. In momentum investing, an investment is typically predicated on investors under-appreciating good news. One buys into a stock on the expectation that it has further upside. There is a good deal of academic and practitioner insights that suggests that momentum effects are pervasive and persistent. However, these positive trends can reverse, which is clearly something we would want to avoid.

By complementing long-term proven momentum signals, for example based on price information, with network-based momentum, that crash risk can often be reduced substantially. We are looking at companies on a more granular level and gaining a better understanding of the drivers of momentum, based on a stock within its cohort and its network.

Using this network analysis as part of a momentum approach might mean you are not as prone to buying a stock simply because it sits in a certain sector or country that has done well recently. Rather, you rather buy it because of stock specific momentum drivers. These tend to be a much more stable drivers of excess returns and less prone to reversals.

These network insights help us to understand the more persistent drivers of momentum. This allows us to be more efficient and deliberate in allocating our clients’ risk budgets for improved portfolio outcomes.

Looking ahead

As quantitative investors, we are always seeking the next edge. In a world awash with data, the challenge is not scarcity, but discernment. Network-based signals offer a powerful new lens — one that is grounded in economic theory, enabled by modern technology, and validated by empirical results.

The future of quant investing is not just about better models or faster computers. It’s about better context. And in that future, networks allow us to make even better informed and risk-efficient investment decisions.

We want to be able to add consistent incremental excess returns across the portfolio for our clients. And having these risk-managed signals and additional insights from network effects helps to achieve that.

In a world that can feel overwhelming given the sheer volume of information flying at us at all times, having a structured approach to separating fact from fiction and capitalizing on the genuinely useful is another step forward for quant investing.