Over the last twenty-five years, there has been a lot of interest in herd behavior in financial markets—that is, a trader’s decision to disregard her private information to follow the behavior of the crowd. A large theoretical literature has identified abstract mechanisms through which herding can arise, even in a world where people are fully rational. Until now, however, the empirical work on herding has been completely disconnected from this theoretical analysis; it simply looked for statistical evidence of trade clustering and, when that evidence was present, interpreted the clustering as herd behavior. However, since decision clustering may be the result of something other than herding—such as the common reaction to public announcements—the existing empirical literature cannot distinguish “spurious” herding from “true” herd behavior.
In this post, we describe a novel approach to measuring herding in financial markets, which we employed in a recently published paper. We develop a theoretical model of herd behavior that, in contrast to the existing theoretical literature, can be brought to the data, and we show how to estimate it using financial markets transaction data. The estimation strategy allows us to distinguish “real” herding from “spurious herding,” or the simple clustering of trading behavior. Our approach allows researchers to gauge the importance of herding in a financial market and to assess the inefficiency in the process of price discovery that herding causes.
Let’s give an overview of the model that we brought to the data and try to explain why herding would arise. In the model, an asset is traded over many days; at the beginning of each day, an event may occur that changes the fundamental value of the asset. If an event occurs, some traders (informed traders) receive (private) information on the new asset value; although this information may be imprecise, these traders do know that something occurred in the market to alter the value of the asset. The other traders in the market trade for reasons not related to information, such as liquidity or hedging motives. If no event occurs, all traders only trade for non-informational reasons.
How does herding occur in this market? That is, when is it ever rational for a trader who has information on the asset value to trade against her own information in order to follow the behavior of the crowd?
Let’s say a series of sells arrives to the market. What would an informed trader whose information points to an upward movement in the asset’s fundamental value do? First of all, she knows that something happened in the market to change the asset’s fundamental value; otherwise she would not be informed. Additionally, she realizes that although her own private information says otherwise, the sells that have already occurred likely reflect the fact that other informed traders received negative news about the asset value. Since many sells arrived to the market, presumably reflecting the fact that many people received bad information about the asset value, the trader also realizes that the negative information reflected in those trades is likely to be more valuable than her own. Therefore, it is rational for her to sell and follow the crowd by going against her own information. All subsequent traders with good information on the asset value will be in a similar position as she is, thus starting a herd.
Because traders who herd rationally decide not to follow their own information, the aggregation of private information in the market is impaired. As a result, price discovery—the convergence of an asset’s price to its fundamental value—is slower.
It is possible to estimate the model outlined above using stock market transaction data from a transaction dataset. Transaction datasets (such as the New York Stock Exchange TAQ dataset) contain all posted bids and asks and all transaction prices in each day of trading. That is, they collect all trading activity and prevailing quotes. These data are the empirical counterpart to the series of buys and sells that appear in our model. Because of this, they can be used to directly infer the prevalence of herding in the market.
As an illustration, in the paper, we measure the importance of herd behavior using transaction data of Ashland Inc., an NYSE-traded stock, in 1995. We find that herding on Ashland Inc. occurred quite often: on average, the proportion of herd buyers was 2 percent and that of herd sellers was 4 percent. Additionally, we find that not only did herding occur but also it was at times misdirected (that is, herd buying in a day when the asset’s fundamental value declined and herd selling in a day when the asset’s fundamental value increased). On average, in a bad-event day, the proportion of herd buyers was 1 percent; in a good-event day, the proportion of herd sellers was 2 percent. Because agents herd and do not follow the information they have, the process of price discovery slows down: on average, we find that the price was 4 percent further away from its fundamental value than it would otherwise have been.
The views expressed in this post are those of the authors and do not necessarily reflect the position of the Federal Reserve Bank of New York or the Federal Reserve System. Any errors or omissions are the responsibility of the authors.
Marco Cipriani is a Research Officer in the Federal Reserve Bank of New York’s Research and Statistics Group.
Antonio Guarino is a professor of economics at University College London.