Identifying Information Types in the Estimation of Informed Trading: An Improved Algorithm

The growing frequency of news arrivals, partly fueled by the proliferation of data sources, has made the assumptions of the classical probability of informed trading (PIN) model outdated. In particular, the model’s assumption of a single type of information event no longer reflects the complexity of modern financial markets, making the accurate detection of information types (layers) crucial for estimating the probability of informed trading. We propose a layer detection algorithm to accurately find the number of distinct information types within a dataset. It identifies the number of information layers by clustering order imbalances and examining their homogeneity using properly constructed confidence intervals for the Skellam distribution. We show that our algorithm manages to find the number of information layers with very high accuracy both when uninformed buyer and seller intensities are equal and when they differ from each other (i.e., between 86% and 95% accuracy rates). We work with more than 500,000 simulations of quarterly datasets with various characteristics and make a large set of robustness checks.