| This dissertation investigates whether the approximating models, which are commonly used in monetary policy discussions, are quantitatively robust to model uncertainty. In the second chapter, we assume that the real world exhibits state-dependent pricing (SDP), as described in Dotsey, King and Wolman (1999) and generate two sets of time series data using the SDP model, with the same preferences and technology parameters but with different policy parameters and steady state inflation. Using Bayesian DSGE estimation methods, we estimate the parameters of the approximating time-dependent pricing (TDP) model and compare how the parameters change across subsamples. We find that the preferences and technology parameter estimates change significantly, which should not be the case according to our data generating process. Our results quantify the risk that the policymakers could be misled by their own model and end up making a wrong policy recommendation.;In the fourth chapter, we examine whether the approximation to the monetary policy itself has any consequence for the estimation of the Taylor rule (Taylor, 1993). We extend the original Taylor rule by including credit channel variables such as risk premium and the net worth/capital ratio. We find that the Taylor rule coefficient estimates of the credit channel variables are both significant. Furthermore, the degree of interest rate smoothing is reduced, compared to the original Taylor rule specification.;Overall we find positive evidence that researchers need to be cautious about the risk that a slight misspecification in the model could alter the quantitative implications of their research results. We conclude that taking the model uncertainty into account is crucial for getting the right monetary policy recommendation.;In the third chapter, we assume that the real world exhibits a credit friction in the loan market as in Bernanke, Gertler and Gilchrist (2000), then do the same experiment on the approximating TDP model. We again find that the preferences and technology parameter estimates change significantly, across subsamples. In addition, our analysis using the quadratic loss function indicates that the misspecification in the credit channel is costly in terms of welfare. |
