Monetary policy in the US is characterized by a substantial degree of inertia. While in principle this may well be the outcome of an optimizing central bank behavior, the ability of any derived policy rule to match the data relies on so large weights for interest rate smoothing into policy makers' preferences as to be theoretically flawed. In this paper we investigate whether such a puzzle can be interpreted as resulting from theconcern of monetary authorities for potential misspecifications of the macroeconomic dynamics. Accordingly, we propose a novel thick modeling approach that incorporates model uncertainty into the identification of central bank's preferences. The thick robust policy rule shows the kind of smoothness observed in the data without resorting to 'incredible' values for the preference parameters.