Bid-Ask Spread Estimation for a Correlated Value Innovation Process

Contrary to predictions, previous studies of bid-ask spread estimators based on serial covariance in returns document surprisingly high proportions of positive serial covariances and therefore negative spread estimates. These findings may be due to the effects of time-variation in expected returns (George, Kaul and Nimalendran [1991]). Although purging the effects of time-varying expected returns yields more reasonable results, the bid-ask spread estimates from daily and weekly returns are still materially different. We present a method that avoids the need for removing the effects of time-varying expected returns by using a spread estimator developed directly for a correlated value innovation process. The new spread estimator not only yields more reasonable estimates of the bid-ask spread than the Roll [12] model, but the spread estimates from daily and weekly returns are almost equal, thus indicating an improvement over the George, Kaul and Nimalendran [1984] method