Discussion Paper

No. 2008-4 | February 11, 2008
The New Keynesian Phillips Curve Tested on OECD Panel Data


In their work, Galí, Gertler and Lopez-Salido, GGL, assert that the hybrid New Keynesian Phillips curve (NPC) with dominance of forward-looking behavior and real marginal costs is robust to choices of estimation procedure, details of variables definitions and choice of data samples. In an estimation on panel data from OECD countries we replicate the typical empirical NPC from country studies. However, we also test an alternative economic interpretation of the empirical correlations. Specifically we find that the expected rate of future inflation and real marginal costs serve as replacements for equilibrium correction terms that are implied by the general imperfect competition model of wage and price setting. As a explanatory model of OECD inflation, the NPC is encompassed by an existing theory.

Paper submitted to the special issue "Using Econometrics for Assessing Economic Models" edited by Katarina Juselius.

JEL Classification:

C23, C52, E12, E31



Cite As

Roger Bjørnstad and Ragnar Nymoen (2008). The New Keynesian Phillips Curve Tested on OECD Panel Data. Economics Discussion Papers, No 2008-4, Kiel Institute for the World Economy. http://www.economics-ejournal.org/economics/discussionpapers/2008-4

Comments and Questions

Sunil Sapra - Comment on “The New Keynesian Phillips Curve tested on OECD panel data” by Roger Bjornstad and Ragnar Nymoen
March 09, 2008 - 03:15

Comment on “The New Keynesian Phillips Curve tested on OECD panel data” by Roger Bjornstad and Ragnar Nymoen

Comments by Sunil Sapra, Department of Economics and Statistics, California State University, Los Angeles, USA.

E-mail: ssapra@calstatela.edu

I have a few comments with respect to the methodology of this ...[more]

... interesting paper.

1. As the authors acknowledge, macro panels generally exhibit cross sectional dependence and panel unit root and cointegration tests, which ignore this dependence suffer from severe size distortions. Nevertheless, the panel unit root tests such as the Im, Pesaran and Shin, Levin-Lin-Chu, Fisher ADF and Fisher PP tests and panel cointegration tests such as the Pedroni tests used by the authors assume cross sectional independence. Alternative panel unit root tests due to Bai and Ng (2004), Choi (2002), Chang (2002, 2004) and Pesaran (2004) take account of cross sectional dependence. The authors take account of this dependence in estimation by employing time dummies. In contrast, cross section correlation has been explicitly modeled through a dynamic factor model by Bai and Ng (2004), Bai (2004), Moon and Perron (2004) and Phillips and Sul (2003) among others. While the authors’ approach is easier to implement, some discussion of the relative merits and drawbacks of these approaches is desirable.

2. Given that the models are estimated by GMM assuming spherical errors, the standard errors should be computed using the Newey-West or the Andrews or the Robinson estimators of the limiting covariance matrix since the form of heteroscedasticity and autocorrelation is unknown.

3. How serious is the problem of weak instruments? The quality of instruments could affect the size and power properties of tests of significance as demonstrated Stock-Watson and Han-Hausman studies among others.


Bai, J. and S. Ng (2004) A PANIC attack on unit roots and cointegration, Econometrica 72, 1127-1177.

Chang, Y. (2002) Nonlinear iv unit root tests in panels with cross section dependency, Journal of econometrics 110, 261-262

Chang, Y. (2004) Bootstrap unit root tests in panels with cross section dependency, Journal of econometrics 120, 263-293

Choi (2002) Instrumental Variable estimation of a nearly stationary heterogeneous error component model, Journal of Econometrics 109, 1-32

Moon, H. R. and B. Perron (2004) Testing for unit roots in panels with dynamic factors, Journal of Econometrics 122, 81-126

Pesaran, H. (2004) General Diagnostic tests for cross-section independence in panels,Working Paper, Trinity College, Cambridge.

Philips and Sul (2003) Dynamic panel estimation and homogeneity testing under cross section dependence, Econometrics Journal 6, 217-259

Roger Bjørnstad, Ragnar Nymoen - Response to comments by Sunil Sapra
April 09, 2008 - 09:34

We find these comments both relevant and important.

1. The econometrics in macro panels has advanced rapidly the last years, not least within the field of dealing with cross-sectional dependence. In our paper we use a panel of 20 OECD-countries to test the New Keynesian Phillips curve, NPC, against ...[more]

... an alternative model of price-inflation.

From earlier research on country data, our own and others, we have formulated the hypothesis that the dominance of the leading inflation term in particular may be due to omission of other explanatory variables and equilibrating mechanisms, which are in fact formulated and estimated in existing models of ‘price setting’, e.g., the ICM models that we refer to in the paper. If that view is supported by the evidence, macroeconomics is in the (somewhat strange) situation that a new and widely accepted model does in fact not encompass an existing one; hence there is loss of information and learning instead of the opposite. If we are on the right track, estimation of a NPC should give more ore less the same result regarding the main parameters of interest, the coefficient of the forward inflation rate in particular, on any country data set, and also on a panel of macro data. This is the motivation of the paper.

Since we use a macro panel we do address the issue of cross-sectional dependence in Section 4. However, we do not model the cross sectional correlation explicitly. The reason is partly, as Sapra points out, that our approach is more convenient to implement, but also that we want to keep the estimation and the presentation of results close to the format that readers familiar with the literature on the NPC (and the earlier literature on the ICM) can recognize. Moreover, our ambition at this stage has not been to estimate "the best" possible price-inflation model on this dataset. Our more limited aim has been to test one model against another, and since our estimates fairly well corroborates the findings forin several single-country analyseis, we take it as a strong indications that the treatment of the two estimated models are reasonably balanced, and that the assumption of the statistical model we use for inference are not rejected by the misspecification test (see 2). That said, the more refined estimation methods that Sapra mentions are cleary relevant to us in future work, on this or other macro panels.

2. We have conducted several estimators, i.e. correcting for cross-section specific heteroskedasticity, period specific heteroskedasticity, contemporaneous covariances and between period covariances. The various estimators show very similar standard errors in our panel. The standard errors we have chosen to report are robust to arbitrary serial correlation and time varying variances in the disturbances.

3. Thank’s for reminding us of this issue. Although there are three endogenous ‘regressors’ in Table 1, all the results are robust to treating Δulc and Δ(pi- p) as exogenous with the inflation lead as the only endogenous explanatory variable. In that setting, the strength of the instruments can be tested by their predictive power in the first stage regression. Note that we estimate two different specifications, namely M1 versus M2, and it is M2 which is most likely to be subject to weak instrument problems, since there are two instruments less in M2 than in M1 (and M1’), i.e. the NPCs. Hence we test the strength of the other 12 instruments. The result is F=57, which is much higher than the rule of thumb of 10. By implication, it is therefore even less evidence for weak instrument problems for M1, since the ‘first stage regression’ for that model includes two more significant predictors, namely the lagged values of (ulc-pi) and (ulc-p), which are jointly significant with Χ2 (2) =22.36. In line with this, the test statistic for instrument irrelevance is F=76 for M1 and M1'. Incidentally, Bårdsen et al (2004) report F=71 for their re-estimation of the euro area NPC.

Anonymous - Referee Report
March 27, 2008 - 11:22

See attached File

Anonymous - Referee Report
April 01, 2008 - 12:31

See attached file

Roger Bjørnstad - Response to referee reports
May 16, 2008 - 20:18

The response to the two referee reports are attached

Roger Bjørnstad - Revised paper
May 20, 2008 - 20:47

Revised paper attached