Estimation of trading costs: Trade indicator models revisited

It is a stylized fact that trade indicator models (e.g. Madhavan, Richardson, and Roomans (1997) and Huang and Stoll (1997)) underestimate the bid-ask spread. We argue that this negative bias is due to an endogeneity problem which is caused by a negative correlation between the arrival of public information and trade direction. In our sample (the component stocks of the DAX30 index) we find that the the average correlation between these variables is -0.193. We develop modified estimators and show that they yield essentially unbiased spread estimates.