Following the (Social Network) Herd

Here’s another academic take on the financial implications of social media: Earlier in a Risky Business contribution I discussed the finding by a team of researchers at Northwestern University’s Kellogg School of Business that traders acting in sync with one another, via instant messaging, are more likely to dodge losses. Now, a paper to be published in an upcoming Review of Economic Studies takes this line of inquiry a step further. It shows that large social networks in general often are quite efficient at aggregating the information that is so widely dispersed in society.

MIT professors Daron Acemoglu, Munther A. Dahleh and Asuman Ozdaglar teamed up with Ilan Lobel of NYU’s Stern School of Business to study “Bayesian” learning — a nineteenth century statistical theorem that shows how to predict the probability of any one of a set of possible causes of a given outcome from knowledge of each of their probabilities. They bring this concept to bear on more modern social networks, and they analyze the conditions under which these networks, as they become larger, are more likely to take “the right action.”

Previous research — not to mention common wisdom — have suggested that when people make decisions after observing each others’ actions, they often fall into “information cascades,” leading to counterproductive “herd” behavior. Think of asset price bubbles or the rush to stampede out of a burning theater.

But the MIT and Stern professors show that these cascades are unlikely to occur in a world in which people can observe the actions of their social network friends.

For example, let’s say a TV personality like Oprah Winfrey selects a technology to adopt, and a few of our friends rush out to buy it. We are likely to know that our friends’ actions were based on this single source of information and as a result are less likely to copy our friends’ decisions than if our friends had independently chosen the same action.

This type of “learning” is possible even when there are “influential agents” or “information leaders” who are observed by “many, most or even all agents, while others may be observed not at all or much less frequently,” they argue.

The paper concludes: “It is only when individuals are excessively influential — loosely speaking when they act as the sole source of information for infinitely many agents” — that social networks fail to deliver.