知情交易概率的估计方法比较

A Comparative Study on PIN Estimations

知情交易概率PIN模型的极大似然估计,由于似然函数形式复杂,在最优化过程中很容易出现计算溢出的问题。本文提出了PIN模型的广义矩估计,并通过数值模拟比较了这一新方法和以往文献提出的原始极大似然估计、改进极大似然估计在不同情况下的估计精度。模拟结果表明,在某些情况下,广义矩估计比极大似然估计更容易计算得出也更具有精度优势。本文还提出了用bootstrap方法对广义矩估计结果进行误差修正,进一步提高了广义矩估计方法对于知情交易概率PIN的估计精度。

The popular MLE methods of the probability of informed trading（PIN） often generates infeasible and biased estimator because of the complicated mixture Poisson form of the likelihood function. We propose the GMM estimation of PIN as an effective supplemented method. Results in simulation analysis suggest that the GMM estimation has lower bias and lower RMSE than MLE method suggested by previous studies in most situations,especially in active markets. Furthermore, we apply bootstrap algorithm to correct GMM estimation error and improve the performance of GMM for more accurate PIN.