Estimation of Monetary Policy Preferences in a Forward-Looking Model: A Bayesian Approach

Abstract

In this paper we adopt a Bayesian approach towards the estimation of the monetary policy preference parameters in a general equilibrium framework. We start from the model presented by Smets and Wouters (2003) for the euro area where, in the original set up, monetary policy behaviour is described by an empirical Taylor rule. We abandon this way of representing monetary policy behaviour and assume, instead, that monetary policy authorities optimize an intertemporal quadratic loss function under commitment. We consider two alternative specifications for the loss function. The first specification includes inflation, output gap and difference in the interest rate as target variables. The second loss function includes an additional wage inflation target. The weights assigned to the target variables in the loss functions, i.e. the preferences of monetary policy, are estimated jointly with the structural parameters in the model. The results imply that inflation variability remains the main concern of optimal monetary policy. In addition, interest rate smoothing and the output gap appear to be, to a lesser extent, important target variables as well. Comparing the marginal likelihood of the original Smets and Wouters (2003) model to our specification with optimal monetary policy indicates that the latter performs only slightly worse. Since we are faced with the time-inconsistency problem under commitment, we initialize our estimates by considering a presample period of 40 quarters. This allows us to approach, empirically, the timeless perspective framework.

Generalized method of moments and inverse control

“…That is, given the loss function (4), the true data generating process is determined by commitment to a simply policy rule whose coefficients are chosen to minimize expected loss. If instead we assumed that policy is the outcome of optimal commitment (e.g., Ilbas, 2007) or optimal discretion (e.g., Dennis, 2004;So¨derlind et al, 2005), then we could still preserve the GMM framework by augmenting the system with appropriate first-order conditions from the Lagrangian as shown by So¨derlind (1999). Taking that step here, however, would amount to imposing false restrictions on the model given our assumption about the nature of policy.…”

Estimating Central Bank Preferences under Commitment and Discretion

“…9 9 An alternative strategy adopted by Ilbas (2010) is to partition the sample to include an initialization period that precedes estimation. She finds that a presample period of 20 quarters is sufficient to eliminate any effects on parameter estimates of setting the multipliers equal to zero in the initial period. …”

Estimating Central Bank Preferences Under Commitment and Discretion

“…Values of κ near zero are evidence 9. An alternative strategy adopted by Ilbas (2010) is to partition the sample to include an initialization period that precedes estimation. She finds that a presample period of 20 quarters is sufficient to eliminate any effects on parameter estimates of setting the multipliers equal to zero in the initial period.…”