The science and art of DSGE modelling II. Model comparisons, model validation, policy analysis and general discussion

The previous chapter described the construction, calibration and the Bayesian estimation of DSGE models with a particular focus on the New Keynesian model. This chapter shows how model comparisons can be made and how the model's success in fitting data can be assessed by comparing second moments and by a comparison with a benchmark DSGE-VAR. We then demonstrate how the estimated model can be used for computing optimal monetary policy. Our two chapters as a whole will then describe a seamless construction, estimation and policy analysis methodology for macroeconomics summarized by the followingsteps. 1. The construction of a DSGE model describing the first-order conditions for economic agents in the form of a set of non-linear difference equations; 2. the solution of the steady state to be used for both solution and calibration; 3. Bayesian estimation of the linearized model; 4. model comparisons between different models or variants of the same model; 5. model validation by comparison with second moments and a benchmark DSGE-VAR; 6. optimal policy analysis with a) optimal commitment (the 'Ramsey problem') b) optimal policy under discretion c) optimized simple commitment Taylor-type rules. The previous chapter has covered steps 1 to 3. This chapter proceeds from step 4 through to step 6 in sections 2 to .4 The final two sections of this chapter switch from a practical toa more reflectivemode.Although we have claimed that the DSGE approach to macroeconomic modelling enjoys a reasonable consensus at the moment, there is a growing debate that is either demanding new types of DSGE model or a totally different approach. The final two sections 5 and 6 join this debate.