We divide the full sample around the third quarter of 1979 which corresponds to the
appointment of Paul Volcker as Fed Chairman. This lines up with a number of empirical
studies that demonstrate a significant difference in the way monetary policy was conducted
pre- and post-1979 (see Clarida, Gali' and Gertler, 2000, and Favero and Rovelli, 2003 among
many others). Moreover, we remove from the second sub-sample the period 1979:3-1982:3
when, as documented by Bernanke and Mihov (1998), the operating procedure of the Fed
temporarily switched from federal funds rate to non-borrowed reserves targeting.
We estimate a version of the central bank Euler equation using the Generalized Method of
Moments (GMM) with an optimal weighting matrix that accounts for possible heteroskedas-
ticity and serial correlation in the error terms (see Hansen, 1982). In practice, we employ a
four lag Newey-West estimate of the covariance matrix. Starting from date t - 1, four lags of
the explanatory variables, the federal funds rates and the measure of inflation left out from
the regression are included as instruments corresponding to a set of 19 overidentifying restric-
tions that can be tested for. The null hypothesis of valid overidentifying restrictions is never
rejected. Moreover, the multiple endogenous regressor analog of the F-statistics from the first
stage regression is always close to (and exceeds in most cases) the critical value that ensures
that the bias is no more than 10% of the inconsistency of OLS (see Stock and Yogo, 2003).
Given the conservative nature of the test, we consider reliable the GMM inference.
3.1 Preliminary Analysis
The nonlinear terms in (5) stem from asymmetric central bank preferences but we cannot ex-
clude in principle that some alternative source like a nonlinear Phillips curve might also return
evidence of asymmetry in the policy rule (see Schaling, 1999). A simple way to discriminate
between nonquadratic ob jectives and nonlinear constraints is to perform the REgression Spec-
ification Error Test (RESET), which is designed to detect incorrect functional forms, on the
New-Keynesian Phillips curve. To this end, we estimate equation (1) over the full sample
using Instrumental Variables and a twelve-lag Newey-West variance covariance matrix. The
set of instruments dated at time t - 1 includes four lags of the GDP deflator inflation, the CBO
only shocks affecting the target level of output. When applied to an estimated New-Keynesian model for the
Euro area, they find that the counterfactual flexible-price level of output, which is the one responding to all
non-monetary shocks in the economy, is indeed extremely volatile, whereas the target level of output, which is
the one only affected by supply and demand disturbances, actually follows a relatively smooth path.