Relative importance of sectoral and aggregate sources of price changes

Abstract

This paper estimates a dynamic common factor model to assess relative importance of the aggregate and the sector-specific factors that determine changes in the prices of individual products. It also examines how aggregate price changes are affected by these factors. Two different specifications of the model are estimated: the baseline model with one aggregate factor, and a second specification with two aggregate factors. In the one-actor model, the aggregate factor contributes little to the movements of changes in prices, mostly of nondurable goods whereas it seems to have important contributions to the movements of changes in prices of commodity groups mainly used as intermediate or capital goods. In the specification with two aggregate factors, the additional factor has significant effects on changes in prices of ‘farm products’ and ‘processed foods and feeds’ only. Forecast-error variance decompositions of both aggregate and disaggregate price changes suggest that sectoral factors account for most of the variability at short horizons while the contributions of the aggregate factors increase as the time horizon lengthens. The results also show that sectoral factors are not only important for relative price changes but also have significant impact on aggregate inflation. The estimated common factors have statistically significant correlations with money growth and changes in the unemployment rate.

Acknowledgements

This paper is a revised version of Chapter 2 of the author's doctoral dissertation at the Department of Economics, Southern Methodist University, USA. The author is grateful to Nathan S. Balke for many helpful discussions, and would also like to thank Pinaki Chakraborty, Tricia Coxwell, Tom Fomby, Thomas Osang, Mark Wynne, and the audience at the 35th Annual Meetings of the Canadian Economic Association held in Montreal, and at the 71st Annual Conference of the Southern Economic Association held in Tampa, and the seminar participants at the Southern Methodist University and at the Sam Houston State University for their comments, and Jakovs Itkins for his help with the program codes.

Notes

One notable exception is Barsky and Kilian (2002), who argue that Fed's expansionary monetary policy bears much of the blame for the stagflation of the 1970s.

Although the Consumer Price Index (CPI) or the Personal Consumption Expenditure (PCE) Implicit Deflator has broader coverage of both goods and services––as opposed to PPI that includes goods only – and are more important from the monetary policy perspective, the reasons for using PPI in this paper are two-fold: first, to include the prices of capital goods which are not covered in previous studies (Bryan and Cecchetti, 1993) that consider consumer prices only. Secondly, previous research (Balke and Nath, 2003) indicates that prices of different types of goods such as consumption goods, capital goods etc., have interestingly different dynamics in terms of the effects of aggregate and sectoral shocks on price changes.

For a discussion see Hamilton (1994) and Harvey (1989).

In order to verify stationarity of these log-differenced price series, an Augmented Dickey–Fuller test is conducted for each of them separately. For each series, the null hypothesis of a unit root is rejected at 1% significance level. Since monthly data are used, in the test equation the augmented terms are included with a maximum lag of 12 and then based on the information criteria the appropriate lag lengths are selected. There is no uniformity of the lag lengths and they also vary according to whether Akaike Information Criterion (AIC) or Schwarz Criterion (SC) are used. The interested reader can obtain detailed test results from the author.

For example, for the baseline model with p = 1 and q = 1, AIC = 27.99; for p = 1 and q = 2, AIC = 36.38; for p = 2 and q = 1, AIC = 1.25; for p = 2 and q = 2, AIC = 1.03; for p = 3 and q = 2, AIC = 6.64; and for p = 3 and q = 3, AIC = 5.34.

Singleton (1980) uses a sequential chi-squared test procedure to determine the number of common factor. Norrbin and Schlagenhauf (1990), on the other hand, use likelihood ratio test to examine the importance of different sets of factors.

See Engle and Watson (1983).

We use multidimentional unconstrained minimization algorithm (Nelder-Mead) for ML estimation.

On the basis of the BLS classification scheme it is not possible to categorize a commodity group as entirely belonging to intermediate or capital goods category.

A similar methodology is used by Norrbin and Schlagenhauf (1990).

Since demeaned values of Y variables are used to estimate the model, one can ignore the intercept terms in the observation equation.

Note that this is not how aggregate inflation is calculated. An aggregate index is constructed by applying these weights to individual commodity prices (at levels and not at log differences). Then%age changes are calculated. However, since we calculate aggregate inflation from the real data using the same definition as from the estimated commodity price changes it makes actual and estimated aggregate inflation comparable. Also see Bryan and Cecchetti (1993).

It is well known that all these three broad measures of the money stock have a significant endogenous component, and the extent to which any of them reflects changes in exogenous monetary policy is controversial. Therefore, we use MB as an alternative measure.

Note that government purchases data are available at quarterly frequency. They are converted to monthly frequency using a method that assumes linear distribution of government purchases, provided by the DRI-Pro database from where the data have been extracted.