Identifying the covariation between the diffusion parts and the co-jumps given discrete observations

In this paper we consider two semimartingales driven by diffusions and jumps. We allow both for finite activity and for infinite activity jump components. Given discrete observations we disentangle the {\it integrated covariation} (the covariation between the two diffusion parts, indicated by IC) from the co-jumps. This has important applications to multiple assets price modeling for forecasting, option pricing, risk and credit risk management. An approach commonly used to estimate IC is to take the sum of the cross products of the two processes increments; however this estimator can be highly biased in the presence of jumps, since it approaches the quadratic ovariation, which contains also the co-jumps. Our estimator of IC is based on a threshold (or truncation) technique. We prove that the estimator is consistent in both cases as the number of observations increases to infinity. Further, in presence of only finite activity jumps 1) \hat{IC} is also asymptotically Gaussian; 2) a joint CLT for \hat{IC} and threshold estimators of the integrated variances of the single processes allows to reach consistent and asymptotically Gaussian estimators even of the $\beta$s and of the correlation coefficient among the diffusion parts of the two processes, allowing a better measurement of their dependence; 3) thresholding gives an estimate of IC which is robust to the asynchronicity of the observations. We conduct a simulation study to check that the application of our technique is in fact informative for values of the step between the observations large enough to avoid the typical problems arising in presence of microstructure noises in the data, and to asses the choice of the threshold parameters.