Aiolfi, M. and A. Timmermann, 2004, “Structural Breaks and the Performance of Forecast Combinations,” mimeo, Bocconi University.
Altissimo, F., L. Bilke, A. Levin, T. Mathä, and B. Mojon, (2005) “Sectoral and Aggregate Inflation Dynamics in the Euro Area,” Paper prepared for the 20th Annual Congress of the European Economic Association, Amsterdam.
Blix, Mårten, and P. Sellin, 1999, “Inflation Forecasts with Uncertainty Intervals,” Sveriges Riksbank Quarterly Review 2: 12–28.
Bryan, M. F., and S. G. Cecchetti, 1994, “Measuring Core Inflation,” in N. Gregory Mankiw (ed.) Monetary Policy, Chicago: The University of Chicago Press, 1925-215.
Coats, Warren, 2000, “Inflation Targeting in Transition Economies: The Case of the Czech Republic,” ed. By Warren Coats, (Prague: Czech National Bank).
Coletti, D., B. Nunt, D. Rose and R. Tetlow, 1996, “Bank of Canada’s New Quarterly Projection Model. Part 3, The Dynamic Model: QPM,”Bank of Canada Technical Report No. 75 (Ottawa: Bank of Canada).
Cristadoro, R., M. Forni, L. Reichlin, and G. Veronese, 2001, “A Core Inflation Index for the Euro Area,” CEPR Discussion paper series No. 3097.
Faruqee, H., 2005, “Declining Money Velocity in the Euro Area: Implications for the ECB’s Monetary Analysis,” IMF Country Report No. 05/259.
Forni, M., M. Hallin, M. Lippi, and L. Reichlin, 2000, “The Generalized Factor Model: Identification and Estimation,” Review of Economics and Statistics, Vol. 82, No. 4, pp. 540–54.
Forni, M., M. Hallin, M. Lippi, and L. Reichlin, 2003, “The Generalized Factor Model: One-Sided Estimation and Forecasting,” manuscript.
Hahn, E. (2002), “Core Inflation in the Euro Area: An Application of the Generalized Dynamic Factor Model,” Center for Financial Studies Working Paper No. 2002/11.
Vega, J. L. and M. A. Wynne, 2002, “A First Assessment of Some Measures of Core Inflation for the Euro Area,” Federal Reserve Bank of Dallas Working Paper No. 0205.
Prepared by Emil Stavrev.
In this note core inflation denotes indicators by permanent exclusion, while underlying inflation stands for the unobservable component of inflation driven by fundamental factors.
Each indicator has specific advantages. In particular, GDFM measures are good coincident and leading indicators—see, for example, Cristadoro and others (2001), Hahn (2002), and Forni and others (2003); core inflation and trimmed means can be computed in real time; trimmed means are superior estimators of the central tendency if excess kurtosis of the sectoral distribution of prices is an issue—see Bryan and Cecchetti (1997).
There are indicator-specific disadvantages. For example, the static nature of both permanent and variable exclusion indicators is a drawback, as their leading indicator properties could vary over time, depending on the nature of the shocks. Also, the exclusions in deriving core inflation are significantly based on subjective criteria—the results in Table I.2 as well as several other studies, among them Vega and Wynne (2002), show that the excluded components are not always the most volatile ones.
Two estimation methods were used: (i) an expanding window—the initial point of the sample remains fixed, while the end point is extended each time by one month, and, (ii) a 4-year rolling window—both the initial and the end points are moved forward by one month each time a new observation is added—the results from both methods are similar.
The sample period is January 1997–November 2005 for the estimates with year-on-year data and February 1996–November 2005 for the estimates using seasonally adjusted month-on-month data. To eliminate the effect of the sample period on forecast evaluation, the forecasting performance of the measures was assessed over a common sample. As a result, the length of the sample period was restricted by the GDFM, as 4-digit disaggregated HICP data used to estimate the model are available only since January 1996.
An alternative is to estimate the model with the right-hand side variables lagged 6, 12, 18, and 24 months, respectively. However, this not only quickly reduces the degrees of freedom in the estimation, but also results in worse goodness of fit indicators (root mean square error and bias) for the simulated out-of-sample forecast.
The output gap, the exchange rate, and the oil prices are forecast with the spectral density filter.
This assumption implies a vertical long-run Phillips curve and provides the necessary identification restriction for the SVAR coefficients (see Quah and Vahey, 1995, for further details).
The results in Table I.4 are for estimates using year-on-year data. In that case, a central estimate of 2 percent and a RMSE of 0.4 percentage points suggest that with 70 percent probability year-on-year inflation is forecast to be in the range of 1.6-2.4 percent.
The large number of disaggregate information used in the GDFM could be behind its superior performance over the sample period used here—as Hendry and Hubrich (2006) show, disaggregate information should, in theory, help forecasting the aggregate. However, they also find that including disaggregate information does not always improve forecasts of the aggregate inflation for the euro area, in particular at longer forecast horizons, as changing collinearity among the components undermines the performance of disaggregated models.
The performance of the SVAR model is similar to that of the Phillips curve. However, their usefulness as a tool for monetary policy analysis is questionable, as the probability of measurement error exceeding 0.5 percentage points is in the range of 40-60 percent—see Folkertsma and Hubrich (2001) for details.
While providing useful insights about the driving forces of inflation, the reduced form Phillips curve model is missing an important component, namely, monetary policy. To gauge what is its contribution over the sample period, a structural model with monetary policy reaction function would have to be used.
Notice that he exchange rate, oil prices, and the industrial production-based output gap are forecast with an ARMA process (Figure I.8). Using WEO projections for oil and the exchange rate and replacing the industrial-production based output gap with the WEO output gap for the whole economy would yield a lower inflation forecast.
The estimates from a 4-year rolling AR1 process (Figure I.13) suggest declining coefficient on lagged inflation. Given the persistence of the oil shocks since early 2004, this decline of the coefficient could suggest falling inflation persistence in the euro area over the past several years (perhaps reflecting increased competition due to globalization).