We compare theoretical and empirical forecasts computed by rational agents living in a model economy to those produced by professional forecasters. We focus on the variance of the prediction errors as a function of the forecast horizon and analyze the speed with which it converges to a constant (which can be seen as a measure of the speed of convergence of the economy to the steady state). We look at a standard sticky-prices-wages model, concluding that it delivers a strong theoretical forecastability of the variables under scrutiny, at odds with the data (professional forecasts). The flexible prices-wages version delivers a forecastability closer to the data and performs relatively better empirically (with actual data), but mainly because forecasts deviate little from the unconditional mean. These results can be interpreted in at least two ways: first, actual deviations from the steady-state are not persistent, in which case the implications of the specific formulation of nominal rigidities for short-run dynamics are unrealistic; second, exogenous (or unmodelled) steady-state shifts attributable to, e.g., changes in monetary-policy, taxation, regulation or in the growth of the technological frontier, occur in such a way as to strongly limit the performance of professional forecasters.