The Information Content of Order Executions at Low-Latency

This paper studies the liquidity of assets traded on NASDAQ as measured by the impact that order executions have on future market prices over very short horizons. After deriving a testable implication of Glosten and Milgrom (1985), high-frequency data from a six month sample of all stocks traded on NASDAQ are applied to the model. Empirical results characterize the average information content across asset executions during the sample period. The most active stocks fit the predictions of the GM model tested here, even at very low latency. Less active stocks do not reject the model, although there is a significant loss of power. Evidence is consistent with the hypothesis that some of this lack of power is due to the censoring that comes from the minimum tick size. Using the structure of the GM model, the information content of executed orders can be characterized. Average information content of trades decreases in the log of message frequency, consistent with the hypothesis that prices become more efficient and individual trades are less likely to alter market prices as the amount of activity in the asset increases. By this measure, frequency trade increases liquidity for the average trader in the average asset. Regression results suggest that the assets for which executions convey the most information are those that are infrequently messaged, have a high price and a low market capitalization. This statistical model has substantial predictive power for the 100 most frequently messaged stocks.