In a previous insight, we proposed a definition of the term « monetary policy shock. » Economists use several methods to identify these « shocks. » Here, we will focus on two methods: the statistical method and the narrative method.
The statistical method involves identifying shocks through the residuals of a regression in which the monetary instrument is the explained variable and the variables to which the central bank is supposed to respond are the explanatory variables. Strong assumptions must therefore be made here: the central bank’s instrument is known (the interest rate is generally chosen) and the variables monitored by the central bank are also known (it should be noted that most studies consider real variables, which is in line with the context in which we find ourselves here), so that an equation (otherwise generally referred to as a « reaction function » » or « Taylor rule »). From then on, all information that does not come from real variables but has an impact on the interest rate is found in the residuals of the equation in question. If, for certain episodes, we end up with a relatively large residual, we can conclude that there has been a « monetary policy shock, » insofar as monetary policy has not followed the evolution of real variables (1).
The narrative method consists of identifying shocks based on qualitative and quantitative data. The most widely cited method in recent years is that used by Romer & Romer (see link here). They identify shocks to the Fed’s monetary policy based on its deliberations. To put it simply, if the central bank changes its monetary policy without claiming to respond to movements in real variables, we can then speak of a purely exogenous reaction of monetary policy and therefore of a « monetary policy shock. »
The approach most often used in economic studies is the statistical approach. One limitation of the statistical method is that it presupposes both stability in the central bank’s preferences and knowledge of the central bank’s reaction function, which in reality has often proved difficult…
(1) We are simplifying things here in two respects. In reality, other assumptions must be made: the monetary policy shock does not affect real variables (GDP, inflation, etc.) contemporaneously (in more complex but more appropriate language for experts: restrictions will be imposed via Choleski decomposition in the SVAR estimated here).
Similarly, we present here the intuition behind the identification of a shock, but studies do not actually identify shocks a posteriori through these regressions in general: they use the results to simulate shocks (more on this in a future insight).
Julien P.
