# Testing for stock market herding over short periods

The literature has well established methods for testing stock market herding over a decent time window.

Are there any ways that have appeared in the literature to test for stock market herding over a short time span (e.g. 5 days)?

Perhaps something a bit more than just applying the identical time series models to intraday data (I don't necessarily see anything wrong with this, but would like to know if there's dedicated research on this topic).

-
I would imagine that over a short horizon you can't have decent information about the composition of others' portfolios. – John Oct 6 '12 at 5:11
Welcome to Quant.SE. You might consider registering as it may prompt some others to seek out an answer if they know you're a part of the community. – Louis Marascio Nov 18 '12 at 14:51
perhaps it would help by listing some of these well established pieces of literature first? – chollida Oct 5 '15 at 17:48

I guess the best way to test herding using intraday data is to use Hawkes modelling. Hawkes processes capture the fact an event is a consequence of a previous one (endogenous) or totally new (exogenous). A good start is Chapter V of Thibault Jaisson's PhD thesis: Market activity and price impact throughout time scales.

It is of course related to market impact. I would advice to read Market Impacts and the Life Cycle of Investors Orders by Bacry, Iuga, Lasnier and L (a preprint is available here). When you study market impact you often have a database with split metaorders, hence you know part of the herding.

-

As John mentioned, there are limits to short-term trading or portfolio composition data. However, there are a couple of relevant studies which feature intraday data that I transcribed from this pretty old IMF white paper:

Christie and Huang (1995) studied returns of US stock equities, finding that (controlling for clustering of correlated assets) dispersion on daily and monthly returns is higher at times of large stock movements. However, it's a pretty weak measure of "herding" because it's asset-specific, and overlooks assets of the same individual class/geographical region.

Kodres and Pritsker (1996) analysed public disclosures of intraday commodities trading data from the CFTC (i.e. only large participants), ran correlations/probit to see how likely participants will make the same trades when others are doing the same.

These studies seem to be very vanilla reg methods. I feel that the amount of "herding" is so market-specific and limited by paucity of data that I wouldn't dare generalise; I strongly suspect, however, that traders using technical analysis are implicitly trying to predict herding behaviors.

-