This paper presents an estimation of the output gap based on a multivariate filter that extracts the signal from output indicators and from a Phillips curve. The empirical findings demonstrate that as the specification of the model changes the estimates of output gap and potential output vary over a broad interval. The various models, which are assessed using multiple measures, are optimal only in relation to a specific measure or a time period, and none of them is better in absolute terms. In addition, models that were optimal prior to the recent economic crisis appear to have lost subsequently their explanatory power. The policy implications of these findings point to the need to base economic policy decisions on a model that is sufficiently flexible to incorporate different economic assumptions (such as anomalous cyclical conditions or hysteresis), and to accompany the estimates obtained with appropriate measures of error. We show that the median value of the different output gap estimates performs better, in terms of inflation forecast accuracy, than the individual models in the whole sample period. This measure of the Italian output gap results to be wider than the EC estimates and close to the IMF ones.