Have the macroeconomic effects of QE programs been overestimated empirically? Using a large set of model specifications that differ in the degree of time-variation in parameters, the answer is yes. Our forecasting exercise suggests that it is crucial to allow for time-variation in parameters, but not for stochastic volatility to improve the fit with data. Having a more reliable specification, we find that the portfolio balance and signaling channels had sizable contributions to the transmission of QE programs. Finally, our identified structural shocks show that QE1 had larger macroeconomic effects than QE2 and QE3, but much smaller than usually found in the literature.