We estimate the impulse response function (IRF) of GDP to a banking crisis using an extension of the local projections method. We demonstrate that, though robust to misspecifications of the data‐generating process, this method suffers from a hitherto unnoticed bias which increases with the forecast horizon. We propose a correction to this bias and show through simulations that it works well. Applying our corrected local projections estimator to the data from a panel of 99 countries observed between 1974 and 2001, we find that an average banking crisis yields a GDP loss of just under 10% in 10 years, with little sign of recovery. Like the original local projections method, our extension of it is widely applicable. Copyright © 2013 John Wiley & Sons, Ltd.