The company we keep can reveal quite a lot about us. Social networks are often shaped by invisible patterns of similarity, influencing what shows we watch, where we study, or even where we choose to invest our money. Sociologists call this phenomenon homophily:[1] the tendency of like to attract like. It’s why people form tight-knit social groups, and why investors, often unknowingly, cluster around similar strategies and holdings.

A study by Mislove et. al[2] demonstrated how these invisible bonds can be amplified to reveal remarkably accurate personal details—without ever looking directly at the person in question. What if the same principle could uncover networks of capital allocation?

Amplifying Connections

Mislove et. al explored how homophily—the tendency for individuals with similar traits to form connections—can be amplified to reveal personal characteristics within a social network. By crawling social media platforms, their team demonstrated that the bonds created through shared traits like education, location, or interests can be leveraged to make surprisingly accurate inferences about individuals, even without direct access to their profiles.

Using a metric called normalised conductance — which measures how tightly connected a group is within a broader network — the researchers were able to infer personal characteristics of individuals connected to a seed profile. For instance, if one user attended a specific university, there was a high probability that their less-visible connections had a similar background, such as attending the same university, graduating in the same year, or living in the same residence hall. Crucially, this method proved effective across networks of varying sizes and complexities, making it useful for drawing inferences about characteristics while only needing a very small amount of qualitative information.

Social Networks to Capital Markets

The ability to estimate characteristics of institutional investors’ portfolios is, indeed, reminiscent of what Mislove et. al explore in their work. We use the same concept to amplify homophilic networks amongst funds. Through algorithms inspired by normalised conductance but tweaked for Irithmics and capital networks (instead of social ones). This allows us to accurately estimate characteristics of investors for our forecasts based on the regulatory disclosures of their peers.

It's not so important what the fund discloses, but rather the company it keeps. Since people with like characteristics attract and bond with others with similar characteristics, it is therefore possible to infer characteristics of an individual based on their social network, with whom they interact and connect. In turn, by knowing when and how certain investors allocate capital, we can make statistically significant estimations about other, similar funds and their behaviour.

Estimating Market Behaviour

Markets are social. Not in the sense of making friends, but in the deeper, structural ways that institutions mirror one another. Just as people gravitate toward others like themselves, investors too form tight, homogenous networks of behaviour. Thanks to the homophilic amplification Mislove et. al engineered, this is the reason we can estimate such information through a small pool of data.

While the social profiles in Mislove et. al form a web that’s interconnected by education, location, and profession, large institutional portfolios create their networks differently - funds of broadly similar size or portfolio focus often hold similar types, sectors and quantities of shares.

Institutional investors do not form the same social networks as people per se. yet, they are still interconnected in webs of like strategies and holdings. This is observed by Paul DiMaggio and Walter Powell through a phenomenon called isomorphism.[3] Whether being coerced by their peers, attempting to mirror their success, or simply following prominent norms, firms lean towards acting in similar ways.

And, thanks to metrics like normalised conductance, Irithmics can estimate quite a lot from a very small amount of data. Mislove et. al used a seed profile as the starting piece of data upon which to base their connection inferences. Irithmics uses regulatory and voluntary disclosures as our seed data. These filings give us a starting point to estimate similar characteristics of like funds on which to base our forecasts.

From Disclosures to Insights

We rarely notice the invisible threads that link us — until someone shows us the pattern. The networks we form (whether among friends or funds) speak volumes about who we are. Mislove et. al’s research shows just how much can be inferred from these connections. Irithmics follows the same strands through capital markets, estimating how institutions mirror, mimic, and move together, using minimal, sometimes overlooked data points to uncover investor behaviour.

[1] Newman, M. E. J. (2004). Finding and evaluating community structure in networks. Physical Review E, 69: 026113. College Park: American Physical Society.

[2] Mislove, A. Viswath, B. Gummadi, K. P. Druschel, P. (2010). You Are Who You Know: Inferring User Profiles In Online Social Networks. WSDM '10: Proceedings of the third ACM international conference on Web search and data mining, pp. 251-260. New York: ACM.

[3] ‘Isomorphism is a constraining process that forces one unit in a population to resemble other units that face the same set of environmental conditions’ from Hawley, A. (1968). Human Ecology. Sills DL (ed.) International Encyclopedia of the Social Sciences, pp. 328-337. New York: Macmillan; Further Reading: DiMaggio, P. J. Powell, W. W. (1983). The iron cage revisited: institutional isomorphism and collective rationality in organisational fields. American Sociological Review, 48(2): pp. 147–160.Thousand Oaks: SAGE Publications.