PIN 추정을 위한 최적모형에 관한 연구

Abstract

This study compares two models for estimating the probability of informed trading (PIN) and investigates which model is superior in the Korean index options market. The study uses the method of maximum likelihood estimation (MLE) and Bayesian methods to calculate the PIN. Using the bid and transaction data (TAQ) of KOSPI 200 index options from May 2019 to September 2020, we obtain the following results. First, there were no cases where there was no solution in either method. Second, when estimating delta using the Bayesian method, no boundary solutions occurred, but with the MLE method, multiple boundary solutions occurred. Specifically, 10% of the total solutions in 5-minute tick data and 47% in 20-minute tick data were boundary solutions. Third, in the case of 20-minute tick data, the PIN estimated by the MLE method for ITM options significantly decreased compared to that of the 5-minute data. This is estimated to be caused by the previous boundary solutions inducing a downward bias in the PIN. Fourth, in the MLE method, the standard error of the estimate is almost zero, while the Bayesian method shows an error rate of around 2% based on PIN from 20-minute data. In conclusion, the Bayesian method is superior when there is no boundary problem, given a sufficient size of closing data, but when not, the magnitude of the error rate is considered to be an important factor in the choice of method.