Herding Versus Splitting In Order Flow

When a stock shows a burst of same-direction trades in a short window, there are two very different underlying causes that look identical in a raw trade tape: herding, where many independent traders each separately decide to buy (say, reacting to the same news), and splitting, where a single large trader (an institution executing a big order) breaks it into many smaller child trades over time to reduce market impact. Both produce a run of same-side trades; only one reflects genuinely new independent information.

The distinction matters because the correct response differs: genuine herding is informative — many independent agents converging on the same view is real evidence the price should move — while a single split order is not new information at all, just one decision spread thin, and following it as if it were a fresh signal from many traders is a mistake.

The standard decomposition looks at trade-size and timing statistics that differ between the two: split orders from an execution algorithm tend to have unusually regular timing and similar sizes (since they're generated by one scheduling logic, e.g. VWAP-style slicing), while herding from independent traders produces more irregular arrival times and heterogeneous sizes, since each trader is acting on their own schedule. Autocorrelation in trade signs that decays slowly, combined with low size-and-timing dispersion, points toward splitting; irregular dispersion with fast-decaying autocorrelation points toward herding.