The identificaton of news shocks in Beaudry and Portier (2006) is only unique for their bivariate VECMs: Together with my colleague Andre Kurmann, I tried to replicate the influential AER paper by Beaudry and Portier, which finds that news shocks are an important source of business cylce fluctations. Their estimated impulse response have been regarded as a challenge to "standard" models of the business cycle However, as Andre and I found, the empirical strategy has been flawed and the identfication scheme does not have a unique solution. The problem arises from using long-run restrictions in a VECM. As discussed by King, Plosser, Stock and Watson in their important AER (1991), the cointegration relationships embedded in a VECM already restrict long-run responses. In the case of Beaudry and Portier, this makes some of their restrictions redundant and leaves the system underidentified. Below is a picture comparing their point estimates with the set of all possible solutions. (please click to enlarge) |
News >