David Stark, Professor of Sociology at Columbia University and Director of the Center on Organizational Innovation, recently visited CEU accepting the invitation of Balázs Vedres, Director of the Center for Network Science.
At a lecture in the Quantum Room, he presented a novel paper in which he and his collaborators attempt to identify what kind of network structure might best support the accuracy of security analysts’ forecasts in highly volatile market situations. In a more abstract sense, the aim of the research was to investigate the old question about necessities in the balance of strong and weak ties in one’s network structure in networks, where ties are not signaling direct communication among actors, but rather standing for mutual observational activity interpreted as indirect interactions. The basis of the idea was the phenomenon that communication among market competitors is rather through observation than direct interaction.
Constructing bipartite networks of 9062 security analysts (set A) on the one hand and the totality of the securities with which they were working on (set B), ties were formed between each analyst and the markets which they were working with throughout 1993 and 2010. They introduced a concept in dyadic closure standing for the overlap of markets among 2 analysts according to which, the higher the overlap, the higher their dyadic closure is. They also measured triadic closure standing for the standard triadic closure concept, only applied to bipartite situations. Volatility stands for the variance of the return of a stock over a certain period of time: the higher the volatility of a stock, the riskier the decisions are made regarding its handling, but the more money can be made out of it. So the independent variables were the 3 above mentioned, stabilized with several control variables. With the constructed dependent variable they measured the accuracy of each decision and the level of return rates over a 3 month period of time.
The results showed that in their situation, in which connections are not understood as direct interaction among actors, rather through observation of one another (i.e. reading each other’s reports), the optimal setting for gaining the most knowledge on other relevant competitors knowledge schemas and consequently about the opportunities on the market is to have strong ties to your competitors’ who are themselves disconnected. In network terminology, this can be phrased as having strong ties within open triads or putting it another way, using the new concept introduced by Stark and his colleagues, having high dyadic closure but low triadic closure in bipartite networks. Actors within this position gained over 30% more on returns than expected, implying that this structure is outperforming both star structure, where both types of closures are low, whole networks, where both dyadic and triadic closure is high and also “donut” structures in which low dyadic closure of the ego is accompanied by high triadic closure (meaning that I do not share a lot of common markets with the ones who I compete with, but they – my primer competitors - share common markets with each other). As Stark interpret it; “the more the exposure [to competitors], the more the closure [with them], and the more open one becomes to others' decision-making schemas”.
Of course, if everyone would try to build up her contextual network based on this directive, the network structure would tend toward a complete graph thus losing the mentioned position through the structural change over time, shoving towards a structure with high closure both in a dyadic and in a triadic sense. As Stark argued, this possible evolutionary process may further emphasize that regarding our exact network position, not only our own initiatives matter when forming the structure around us, but that we are very much dependent on the decisions of others than ourselves. Professor Stark’s fascinating style and attentive attitude towards its audience made the lecture a thought-provoking event for many, concluding from the amount and diversity of the questions posed and comments raised.
Blog post by Júlia Perczel