Statement
The slope of the New Keynesian Phillips Curve — the key parameter governing the inflation-output tradeoff in modern DSGE models — is tenuously identified in DSGE model-based estimation. Published estimates of the output coefficient (kappa) range over three orders of magnitude (from <0.001 to 4.15), and the variation is primarily driven by differences in the endogeneity correction implicit in different DSGE model specifications rather than by features of the raw data. OLS regressions of inflation on discounted future expected marginal costs yield near-zero slope estimates regardless of detrending method, indicating that the identifying information comes from auxiliary assumptions about household preferences, monetary policy, exogenous shock processes, and the treatment of marginal costs as observed or latent.
Evidence summary
- Schorfheide (2008): the primary source. Documents the three- orders-of-magnitude range in kappa estimates and traces the variation to the endogeneity correction. Shows analytically in a simple model that single-equation methods (OLS, IV) fail to identify kappa, and that DSGE model-based identification relies on contemporaneous correlations, impulse response dynamics, and autocovariance restrictions.
- The narrowing of lambda estimates when labor share is observed (0.005-0.135) supports the interpretation that much of the variation comes from the treatment of latent marginal costs.
Conditions and scope
- The claim is about Calvo/Rotemberg NKPC specifications with labor- share marginal costs. Alternative microfoundations (menu costs, state-dependent pricing) may have different identification properties.
- The claim is about U.S. data; international evidence may differ.
- The claim does not say the NKPC is unidentified — it says identification is sensitive to auxiliary modeling assumptions.
Counter-evidence
- Some studies with large-scale models and rich observation menus (Smets-Wouters 2003, 2007) obtain relatively stable estimates of lambda around 0.01-0.10, suggesting that richer models with more observables can stabilize identification.
Linked ideas
- (none yet)
Open questions
- Can micro evidence on price-change frequency be used to sharpen macro identification of the NKPC slope?
- How does regime switching in the NKPC parameters (as in Fernandez- Villaverde and Rubio-Ramirez 2007) affect the identification problem?
- Is the identification problem worse or better when asset prices (bond yields, cap rates) are included in the observation menu?