Forecasting Inflation with the Hybrid New Keynesian Phillips Curve: A Compact-Scale Global VAR Approach^*

* This article is a revised version of my MSc in Economics and Econometrics dissertation at the University of Nottingham, UK, and the Working Paper 785 of the Central Bank of Chile. The views and ideas expressed in this paper do not necessarily represent those of the Central Bank of Chile or its authorities. Any errors or omissions are the author’s sole responsibility.

ABSTRACT

In this article, the multihorizon predictive power of the Hybrid New Keynesian Phillips Curve (HNKPC) is analysed by making use of several close- and open-economy specifications for the headline inflation of six developed countries. The key element is the use of direct measures of inflation expectations – Consensus Forecast – embedded in a compact-scale Global VAR (GVAR) environment, becoming the baseline open-economy HNKPC (OE-HNKPC) specification. These OE-HNKPC point forecasts are evaluated using the Root Mean Squared Forecast Error (RMSFE) statistic and statistically compared with several benchmarks, including traditional atheoretical models. Several OE-HNKPC as well as a closed-economy HNKPC (CE-HNKPC) specifications are also analysed. The results indicate that in four out of six countries, the CE-HNKPC is the best forecasting model, whereas for the same countries, a parsimonious OE-HNKPC is the second-best alternative, and in most cases, outperforming traditional statistical benchmarks. The RMSFE is obviously affected by the unanticipated effects of the Great Financial Crisis (GFC), spoiling out the performance of a number of competing forecasts. However, when considering an evaluation sample just before the crisis, both the CE-HNKPC and the parsimonious OE-HNKPC still come out as the best forecasting models. Furthermore, these preferred models also do an excellent job tracking inflation better than the best atheoretical models during the GFC.

KEYWORDS:

• New Keynesian Phillips curve
• inflation forecasts
• out-of-sample comparisons
• survey data
• global VAR
• time-series models

JEL-CODES:

• C22
• C26
• C53
• E31
• E37
• E47

Acknowledgements

I thank the suggestions, comments, and help of Rodrigo Caputo, Kevin C. Lee, Pablo Medel, James H. Stock, two anonymous referees, and seminar participants at the Central Bank of Chile. I also thank to Consuelo Edwards for editing services. Nevertheless, I exclude them for any error or omission that remains at my own responsibility.

Disclosure statement

No potential conflict of interest was reported by the author.

ORCID

Carlos A. Medel http://orcid.org/0000-0001-9621-1440

Notes

* This article is a revised version of my MSc in Economics and Econometrics dissertation at the University of Nottingham, UK, and the Working Paper 785 of the Central Bank of Chile. The views and ideas expressed in this paper do not necessarily represent those of the Central Bank of Chile or its authorities. Any errors or omissions are the author’s sole responsibility.

1 Some theoretical derivations of the NKPC can be found in Smets and Wouters (2003, 2005), Christiano, Eichenbaum, and Evans (2005), Erceg and Levin (2003), and Collard and Dellas (2004), among others.

2 Some forecasting exercises using an expression similar to equation (1) can be found in Nason and Smith (2008) for the US, and Jean-Baptiste (2012) for the UK case, among others reviewed later.

3 Notice that I refer to “compact” rather than “small”-scale because despite considering just a few countries, together they represent a substantial share of world trade and financial markets.

4 These countries are the US (acting as a reference country), Canada (CAN), the Euro Zone (EUR; henceforth considered as one country), Japan (JPN), Switzerland (SWI), and the United Kingdom (UK; also considered a single country).

5 Nevertheless, it is virtually impossible to support that it is truly a closed economy estimation. Despite that the specification is defined for closed economies, headline inflation contains, embedded in virtually all its components, prices set or affected by international markets. This is specially the case of commodity prices in food and energy CPI components.

6 A recent survey of the many inflation forecasting methods can be found in Faust and Wright (2014).

7 An interesting exercise is conducted in Granger and Jeon (2011) which studies how the original Phillips Curve paper could be estimated with the time-series econometrics known 50 years later. This is made using the same original variables and sample, and providing some extensions for robustness.

8 It is relevant to statistically test the hypothesis of the NKPC as some research suggests that must be flat in the  plane. See Kuester, Müller, and Stöling (2009) and the references therein for details.

9 However, an unrestricted VAR(1) version of the CE-HNKPC is also analysed to have a fairer comparison with the VAR-alike OE-HNKPC. These VAR are estimated with Ordinary Least Squares (OLS) and the lag order is set to one.

10 Empirical results do not deliver substantial parameter differences between GMM and OLS.

11 This issue is typically raised when the model is used to test the theory (i.e. through impulse response functions) in the presence of the so-called empirical puzzles. Notice that this problem is a recognised shortcoming of the NKPC which has led to an intense debate noticeable in Rudd and Whelan (2005) and Galí et al. (2005).

12 The authors make use of a dynamic stochastic general equilibrium (DSGE) model to forecast key macroeconomic variables of the US anticipating the GFC, to prove DSGE suitability for forecasting purposes.

13 Note that I focus on the output gap instead of the unemployment gap following the recommendations of Staiger, Stock, and Watson (1997a, 1997b).

Additional information

Notes on contributors

Carlos A. Medel

Carlos A. Medel is Senior Economist and Economic Adviser to the Governor at the Central Bank of Chile. He has worked as Economist in the Macroeconomic Analysis, Economic Research, and Financial Stability Areas of the same Central Bank. His research interests are applied macroeconometrics and time series econometrics. He has published over twenty articles in journals such as Applied Economic Letters, Czech Journal of Economics and Finance, International Finance, Journal of Business Cycle Research, and in several Latin American journals. He holds an MSc in Economics and Econometrics from the University of Nottingham, UK, and an MA and BA in Economics from the Pontifical Catholic University of Chile.