Multilayer Probability of Informed Trading

Abstract

This paper discusses and documents the multilayer structure of informed trading in financial markets. While only 3.59% of 8,190 stock/quarter pairs have single information layer, 75% have two to five layers and 18% have six to eight layers. We develop a clustering algorithm which detects the number of information layers in simulated datasets with the accuracy rates between 95% and 97%. Extending the model of Easley et al. (1996), we introduce a new measure: the multilayer probability of informed trading, or MPIN. Remedial solutions for the computational problems arising from the maximum-likelihood estimation of the MPIN model are also provided. PIN model's estimates of the probability of informed trading as well as the five intermediate parameters are biased for datasets with multiple information layers. In contrast, MPIN model’s estimates for the same datasets are highly accurate, and unbiased. When compared to PIN, MPIN yields substantially higher probabilities of informed trading and information event occurrence, and much lower rate of informed traders. We obtain daily posterior multilayer probabilities conditional on the quarterly estimates of MPIN model. Around earnings’ announcements, posterior probabilities derived from MPIN model are approximately two-fold when compared to the ones from PIN model.

Keywords: MPIN, Multilayer Probability of Informed Trading, PIN, Cluster Analysis, Daily Posterior Multilayer Probabilities of Informed Trading, Market Microstructure

JEL Classification: C38, C46, G14, G17