Academic journal article Federal Reserve Bank of Atlanta, Working Paper Series

The Role of Commodity Prices in Forecasting U.S. Core Inflation

Academic journal article Federal Reserve Bank of Atlanta, Working Paper Series

The Role of Commodity Prices in Forecasting U.S. Core Inflation

Article excerpt

1 Introduction

It is a widely accepted fact that U.S. core inflation is notoriously difficult to forecast. Numerous studies have reported the elusive predictability of U.S. inflation over the post-war period using a standard predictive framework and the recent literature has shifted to unobserved component models (Stock and Watson, 2007) and disaggregated data (Stock and Watson, 2015). The lack of robust predictors for trend inflation is rather unfortunate given the importance of core inflation for conducting and communicating monetary policy. In this note, I document some surprising success at forecasting 12-month ahead core inflation by exploiting particular transformations of futures commodity prices. More specifically, I construct (interest-adjusted) spreads of commodity futures prices (convenience yields, CYs) which reflect market expectations about future economic conditions and are believed to contain information about aggregate (excess) demand in the economy. As an added advantage, these CYs tend to purge the noise in raw prices and induce stationarity (with persistence similar to that exhibited by inflation). Given the highly heterogeneous nature of the different commodities, I try to identify individual commodities, not necessarily from the non-food/non-energy category, whose sources of variations appear to have inflationary consequences. I then isolate the common variation in the CYs of these commodities by averaging and smoothing. The resulting series proves to possess systematic predictive power for the annualized core inflation by tracking closely its future movements, especially during the most recent period (since the early 2000s).

As described above, the objective of this note is fairly narrow in scope: to document and characterize the predictive information contained in futures commodity prices. It extends the work in Gospodinov and Ng (2013) by focusing the analysis on specific commodities and core inflation. While the source of the forecasting success of these commodities is believed to be related to their ability to encompass information about future global and domestic excess demand, this conjecture has not been thoroughly investigated. Furthermore, a better aggregation of the information from these commodities and models is also left for future research. Various ways of forecasting improvement and robustification are discussed in the concluding remarks. In what follows, I will provide some motivation for the proposed predictor and present the out-of-sample forecasting results.

2 Data and Heuristics

The sample period for the analysis is January 1988-December 2015 and is dictated by availability (and sufficient liquidity) of the commodity price data. I focus on two measures of core inflation: (i) CPI (excluding food and energy, source: BLS) and (ii) PCE (excluding food and energy, source: BEA), both of which are seasonally adjusted. In this note, I use only annual (12-month) inflation rates computed as

[inf.sub.t] = 100 x ([P.sub.t] - [P.sub.t-12])/[P.sub.t-12],

where [P.sub.t] is either the CPI or PCE index. Similar results are obtained when inflation is constructed as [inf.sub.t] = 100 x [ln([P.sub.t]) - ln([P.sub.t-12])]. The forecasting horizon is also 12-month (1-year) ahead.

Figure 1 plots the core and headline inflation rates based on CPI and PCE indices. By construction, the core inflation is much less volatile than its headline counterpart (the variance ratio of headline to core inflation is 1.7 (1.6) for CPI (PCE) index) while their unconditional means are very similar. In general, core inflation is more persistent than headline inflation with first-order autocorrelation of 0.982 and 0.988 for CPI and PCE, respectively.

The commodity (nearest and next-to-nearest) futures price data is from Bloomberg and is sampled at monthly frequency as the last available observation for the month. While data for this sample period is available for 22 commodities, in this note I use data only for 3 commodities: copper (standardized (25,000 lbs. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed


An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.