News and correlations: an impulse response analysis

We proceed to an impulse response analysis on the conditional correlations between three stock indices returns: the Nikkei, the FTSE 100 and the S&P 500. As a first step, we estimate an extension of the general asymmetric dynamic conditional correlation (GADCC) model proposed by Cappiello, Engle and Sheppard (2006) to model the possible interactions between conditional correlations. In a second step, we apply the definition of the impulse response function in nonlinear models of Koop, Pesaran and Potter (1996) to the conditional correlations matrix. The estimates of the GADCC model are used to estimate the impact of an innovation on the conditional correlations for different forecast horizons through boostrapped simulations. For each forecast horizon, we estimate the density of the impulse response function with non parametric kernel estimator. These densities show that the impacts of shocks on conditional correlation are most often asymmetric and depend on history. They disappear as the forecast horizon increases. In a first step, we estimate the unconditional correlation impulse response functions with random shocks and histories. In a second step, we estimate these impulse response function with the observed history for several recent dates. We end by computing the impulse response function for an observed shock and history.