By Michael Owyang, Assistant Vice President and Economist
Economists use price indexes such as the personal consumption expenditures price index (PCEPI) and the consumer price index (CPI) to measure inflation.1 Policymakers often focus on “core” inflation, which is based on the price of a subset of the full basket of items that go into the main (often called “headline”) index. Core inflation is meant to represent long-run inflation trends and is relevant to policymakers because it is thought to be a good forecaster of future inflation. However, the way core inflation is currently measured means it might not be as good of an indicator as it could be.
A common measure of core inflation is headline less food and energy prices, a measure proposed in a paper by Robert Gordon in 1975.2 The rationale behind excluding two items that comprise a large portion of the average American's consumption bundle is that food and energy prices were volatile (as can be seen in the chart below) and generally poor predictors of future inflation.
At the time, food prices tended to be volatile because of weather, and energy prices depended largely on OPEC. In other words, both items were thought to depend on short-term factors outside of policymakers’ control. Core inflation was thought to be more representative of the long-run inflation trends that central banks are meant to monitor.
However, the factors affecting food prices in the U.S. have changed dramatically since Gordon’s paper was released. As a result, food prices have become less volatile.
A Regional Economist article by William Gavin and Rachel Mandal explored this issue in detail several years ago. The authors suggested a few reasons for growing stability in food prices. For instance, improvements in technology have nearly erased geographic barriers in trade for food commodities. As food became increasingly available from a variety of locations, prices became less sensitive to localized short-term supply shocks. Ultimately, food prices were more and more driven by demand instead of supply.
Consumers have also changed their eating habits from cooking at home to buying prepared meals. This transition removed consumers farther from the more volatile commodity prices.
As food prices stabilize, there may be value in adding them back into core inflation measures. Food is, after all, a large portion of consumer purchases. And as the prices are more driven by demand factors, the current price inflation for food may be indicative of broader inflation trends.
As a back-of-the-envelope test of the predictive content of food prices, we conducted a simple forecasting exercise to assess the ability of various “core” measures to predict inflation over the next two years. We used headline PCEPI as our measure of future inflation, and we forecasted it using the past year’s inflation rate, measured by four combinations of the PCEPI:
Our initial estimation sample is 1983:Q1 through 1998:Q1. Every time a quarter was added, we dropped a quarter off the beginning, thus keeping the sample an even 15 years. Our last estimation sample ends in 2013:Q1.
Accuracy of a forecast is measured by the root mean squared forecast error (RMSE). A lower RMSE means the associated forecast was more accurate in predicting future inflation. We found that the RMSE did not change much regardless of which trend inflation measure was used. However, the indicator with the lowest RMSE is the headline PCEPI itself, followed by the PCEPI that includes food but does not include energy.
These results suggest two things. First, the best indicator of future inflation is past inflation. Second, unlike in the past, food prices appear now to add predictive content to inflation, although only a marginal amount. Overall, this forecasting exercise suggests that the concept of core inflation as it stands may not be the best indicator of inflation over the long term.
1 For the rest of this post, we’ll focus on the PCEPI.
2 Gordon, Robert J. “Alternative Responses of Policy to External Supply Shocks,” Brookings Papers on Economic Activity, 1975, No. 1, pp. 183-206.