This paper analyses installation entries and exits in the EU ETS, over 2005–2013. Patterns of entries and exits are identified across countries, sectors, and time. The limitations of the EUTL as a data source for research purposes mean that only genuine exits (reflecting production capacity reduction), and not genuine entries (reflecting production capacity increases), can be systematically identified. Exits are found to be relatively frequent events, more so in manufacturing sectors with small average installation size. Moreover, exits were concentrated in 2007 and in 2012, the final years of Phase I and Phase II. We investigate whether the perverse incentives of closure provisions, in free allocation, explain such time pattern. A discrete-time hazard model for the exit event is estimated using a three-tier dataset combining installation-, firm-, and macro-level information. The results indicate that, most likely, closure provisions delayed installation exits, especially in Phase II.