852 Macroeconometric Modeling for Policy
17.4.3 Aspects of optimal policy: the impact of model specification on optimal
monetary policy 890
17.4.4 Theory evaluation: the New Keynesian Phillips curve 892
17.4.4.1 The New Keynesian Phillips curve 893
17.4.4.2 The equilibrium correction implications of the NKPC 895
17.4.4.3 Testing the equilibrium-correction implications of the
NKPC 896
17.4.5 Forecasting for monetary policy 897
17.4.5.1 Assumptions about the forecasting situation 898
17.4.5.2 Real-time forecast performance 900
17.4.5.3 Ex postforecast evaluation and robustification 903
17.5 Conclusion 907
17.6 Appendix: Data definitions and equation statistics 910
“I think it should be generally agreed that a model that does not generate
many properties of actual data cannot be claimed to have any ‘policy
implications’ ...” (Clive W.J. Granger (1992, p. 4))17.1 Introduction
Depending upon its properties, a macroeconometric model can highlight vari-
ous aspects of economic policy: communication of policy actions, structuring
of economic debate, policy simulations, testing of competing theories, forecast-
ing, stress testing, etc. From an academic perspective, the desired properties of
a model are also legion, but the end result will depend upon the preferences
for coherence along dimensions such as: theory foundations (microfounda-
tions/aggregation/general/partial), econometric methods (Bayesian/frequentist),
and model properties (size/robustness/nonlinearities/transparency/dynamics).
Satisfying the different needs and desires of policy making could therefore entail a
collection of models. Such a model collection could include more or less calibrated
theory models, structural and Bayesian vector autoregressions (VARs), simultane-
ous equation models (SEMs), and dynamic stochastic general equilibrium (DSGE)
models (see Pagan, 2003, for an overview). A choice of model(s) for the event at
hand could then be made on the basis of strengths and weaknesses of the various
candidates. The inherent weaknesses of the main candidates are well known. If one
were to follow Ambrose Bierce, and write a “Devil’s Dictionary” of macroecono-
metrics, some of the entries could read: Structural VARs: how to estimate models
inefficiently; SEMs: estimates of something; Bayesian estimation:seecalibration;
DSGEs: sophisticated naivety. Therefore, a choice of model(s) for the event at hand
should be made on the basis of strengths and weaknesses of the various model
classes.
The profession’s collective understanding of the causes and possible remedies
of model limitations, both in forecasting and in policy analysis, has improved