A structural common factor approach to core inflation estimation and forecasting

Abstract

In the article we propose a new methodological approach to core inflation estimation, based on a frequency domain principal components estimator, suited to estimate systems of fractionally co-integrated processes. The proposed core inflation measure is the common persistent feature in inflation and excess nominal money growth and bears the interpretation of monetary inflation. The proposed measure is characterized by all the properties that an ‘ideal’ core inflation process should show, providing also a superior forecasting performance relative to other available measures.

Notes

¹ A current debate in monetary economics concerns the implications that monetary policy regimes have for the endogeneity of the inflation and money growth rates. According to Svensson (2003), nominal money growth would be exogenous relatively to inflation under strict money growth targeting, while inflation would be exogenous relatively to nominal money growth under strict inflation targeting. However, as argued by Nelson (2003), independently of the fact that money is not used as an instrument, monetary policy decisions have an impact on money growth dynamics. For instance, an open market operation which increases the policy rate aiming at a given level of inflation will slow money growth by reducing directly the monetary base. Hence, also in this latter case money growth can be regarded as a ‘quantity-side’ indicator of monetary conditions. Whether the final impact on inflation is then explained by aggregate demand changes deriving from the fall in money growth or the increase in interest rates depends on the characteristics of the transmission mechanism, with the Keynesian mechanism being just a special case of the monetarist one. As a consequence, the proposed definition of core inflation would seem to be useful independently of the monetary policy regime.

² The excess nominal money growth process has been computed by subtracting the output growth rate from the rate of nominal money growth. This is on the basis of the estimated unitary value of the output elasticity of real money balances. For reason of space we have omitted the results concerning the estimation of this latter parameter, which can however be found in the preprint version of the paper.

³ See the preprint version of the article for details.

⁴ A formal test partially supports this conclusion. The estimated co-breaking vector for inflation and excess nominal money growth is [1 − 0.807(0.040)], the p-value of the co-breaking test () is equal to 0.365. On the other hand, the null of homogeneity is rejected at the 1% significance level.

⁵ See the preprint version of the article.

⁶ See the preprint version of the article for details.

⁷ This correction ensures that the compressed excess nominal money growth process has the same standard deviation of the inflation process, and, as it will be shown later in the article, is justified on the basis of compliance with economic theory and forecasting perfomance of the core inflation process.

⁸ The p-value of the zero squared coherence at the zero frequency test is 0.008, while the p-value of the unitary squared coherence at the zero frequency test is 0.315. See Morana (2004b) for details concerning the implications of fractional co-integration for the value of the squared coherences at the zero frequency.

⁹ The MS-Cogley model has been constructed by applying the Cogley (2002) filter on the break-free inflation series, where the break process has been estimated by means of a Markov switching model. This allows to assess the improvement in forecasting provided by the use of a long memory filter.

¹⁰ The results are available upon request to the author.