A1807
Title: How do buys and sells interact: A copula-based PIN model with zero-inflated Poisson distributions
Authors: Chu-Lan Kao - National Yang Ming Chiao Tung University (Taiwan)
Emily Lin - SJU (Taiwan) [presenting]
Shan-Chi Wu - Institute of Statistical Science, Academia Sinica (Taiwan)
Abstract: Classical PIN models assume that, given the information scenario, the number of buy and sell order flows are independently Poison distributed, which imposes an assumption on the probability of no-trades. However, empirical data shows that the implied probabilities of no-trades do not match the aforementioned Poison and independent assumptions. Therefore, we propose a new PIN model that better fits the data by using zero-inflated Poison distributions and copula functions, which allow us to match the probability of no-trades. The expectation conditional maximization (ECM) is further proposed to tackle the parameter fittings, which is verified by simulation studies. The empirical studies show that this model outperforms the original PIN models, with significant parameters on the zero-inflations as well as copulas. In particular, we find that it is possible for an information to simultaneously increase the probability of no trade and boost up the average number of transactions, which contradicts to the intuition.