An Extended MRR Model for Transaction-Level Analysis of High Frequency Trading Processes

Transaction-level analysis of security price change due to Madhavan, Richardson and Roomans (1997, hereafter MRR) has been a useful framework in financial analysis. The one order Markov property of the trade indicator variables is a key assumption in the MRR model, which contradicts the information lag empirically evidenced in high frequency trading processes. In this paper, a non-parametric test is employed, showing that the Markov property of the trade indicator variables is rejected in most of trading days.Based on the spread decomposed structure, an extended MRR model is proposed with a moving average structure adopted to absorb the information lag as an extension. Empirical results show that the information lag plays an important role and the difference of the adverse selection risk parameter between the original and the extended is significant.Further, our analysis suggests that the information lag parameter can be a useful measure of the average speed at which the information integrates into the price