CMStatistics 2016: Start Registration
View Submission - CFE
A1073
Title: Hidden leaders: Identifying latent lead-lag structures in multivariate ultra-high-frequency returns Authors:  Giuseppe Buccheri - Scuola Normale Superiore (Italy)
Fulvio Corsi - University of Pisa and City University London (Italy) [presenting]
Stefano Peluso - Catholic University of Milan (Italy)
Abstract: A test is proposed for the presence of latent lead-lag structures in the dynamics of multivariate ultra-high-frequency returns. To account for the non-synchronous trading and microstructure noise in the observed prices, the model is formulated in a state space representation with missing data, where the state vector of latent returns follows a VAR process. The likelihood is computed using the Kalman filter and optimized through an EM algorithm. In addition, by disentangling contemporaneous covariances from autocovariances in the latent VAR process, the proposed method is the first providing unbiased and consistent covariances estimates in presence of microstructure noise, asyncronicity and lead-lag dependencies. Extensive simulation analysis shows the accuracy of the estimator in recovery the true lead-lag structure in the latent return process and how neglecting the lagged dependencies causes severe distortions in the estimation of contemporaneous covariances. Finally, the empirical application to equity data provides useful information on latent lead-lag relationships among high-frequency returns.