Bias correcting transaction cost models for characteristic function based estimators

Effective transaction cost is a necessary step for devising any competing financial trading strategy in the absence of quote data. Classically, Roll's model estimated the transaction cost using the first order autocovariance of the difference between daily transaction prices, while Hasbrouck's Bayesian approach used iterative posterior simulation. However, Roll's model fails when the first order autocovariance becomes positive, while Hasbrouck's estimator is computationally intensive, especially when many securities are involved. We develop empirical characteristic function methods with a first order bias correcting for spread estimation, and establish the asymptotic distribution of the corrected estimator under suitable conditions. Our numerical experiments illustrate the bias-variance tradeoff of characteristic-function methods compared to Roll's model, and their computational advantage over Hasbrouck's model.