An econometric specification of monetary policy dark art

The classical Taylor rules usually do not yield the same estimation error when working in a monthly or a quarterly framework. This brings us to the conclusion that there must be something that monthly Taylor rules can capture and that the quarterly one cannot: we postulate that it simply boils down to the fact that the target rate's changes are irregularly spaced in time. So as to tackle this issue, we propose to split the target rate chronicle between changes in the target and the associated durations, that is the time spending between two changes in the target rate. In this framework, we propose to consider that changes in rate can be regarded as a real monetary policy decision, whereas the duration period between two changes can be related to a "wait and see" position or some fine tuning problematic. To show that both these features of monetary policy do not react to the same fundamentals, we propose an econometric understanding of the Fed's reaction function using a new model derived from financial econometrics that has been proposed by Engle and Russell (2005). We propose to model the changes in target rates with a classical ordered probit and the durations with an autoregressive conditional duration model. We extracted the Fed anticipations regarding inflation and activity using some factor based method, and used these factors as explanatory variables for the changes in rates and the related durations. We show that the target rate level, the scale of the change in target rate and the associated duration do not necessarily react to the same factors and if they do, the impact can be different. This empirical result supports the idea that durations and scale of the change in target rate deserve equal attention when modeling a Central Bank reaction function.