Clustering Structure of Microstructure Measures

Abstract

This paper investigates popular market microstructure measures for stock returns prediction and builds a clustering model for them to study their correlation and the best measures to use as representatives. Using high-dimensional statistical methods, we build the clustering dendrogram and select 20 representatives from all measures. Furthermore, we provide several interesting insights of the market microstructure measures from our clustering results. We found that the time-weighting technique is only useful for Herfindahl-Hirschman Index (HHI) related measures. HHI measures on the number of trades are always redundant. However, the HHI measures on quotes are very important. Also, we find a strong relationship between quote prices and quote shares.