Primer: Core Versus Headline CPI

The distinction between headline and core inflation is commonly made in market and economic commentary. This primer briefly introduces these concepts, and why they are used.

Definitions

Analysts often refer to "core" measures of inflation in different countries; the exact definition can vary. This article will use the American definition.

• Headline Inflation is the inflation rate for all items in the price index basket (such as Consumer Price Index - CPI).
• Core Inflation is the inflation rate which results if we exclude food and energy from the price index calculation basket. (Some countries will exclude other items which have highly volatile prices.) 

The chart above shows the annual inflation rate for the core and headline CPI inflation rates for the United States since 1970. In the early years of the period, the two measures were high and tracked each other. In the post-1990 period, we see that core inflation has been fairly stable, while the headline inflation rate has swung both above and below the core (mainly the result of gasoline prices).

Why Use Core Inflation?

From the perspective of the cost of living, it makes no sense to exclude food and energy prices from consideration. (One of the stranger theories floating around on the internet is that the CPI calculations always exclude food and energy.) Why do economists look at the core measure?

The justification for the modern use of core inflation is that it is less volatile, and captures the underlying trend in prices.

The chart above compares average hourly earnings in the United States versus core and headline measures of CPI inflation. If we look at the bottom panel (headline inflation) near the Financial Crisis, we see that wage growth was completely decoupled from the wild swings in headline CPI inflation (due to the oil price spike then collapse). Meanwhile, core inflation was more stable, resembling the trend in wages.

The chart above indicates that the two measures have not greatly diverged in the long run. I chopped down the period shown to be the low inflation period of 1994 to present. The secular rise in energy prices since the mid-1990s shows up with the All Items CPI rising by more than the core measure. That said, the total divergence is only 2.5% over the two decades (as shown in the bottom panel; the normalised ratio goes from 100 to 102.5). This is negligible when compared to the absolute change in the price level.

In other words, much of the hysterics about economists ignoring energy and food prices was misplaced.

Other Measures -- Trimmed Mean, Median

If all we are interested in is reducing the volatility of inflation, it is easy argue that arbitrarily eliminating food and energy from the CPI may not be the best way forward. Economists have studied this issue, and have proposed a statistical means of creating a low volatility inflation trend series -- trimmed mean CPI measures, or the median CPI. The chart above shows the median CPI from the Cleveland Federal Reserve bank. (Note: the series from FRED is a monthly series; I took the annual average to create the equivalent of the annual rate of change series. As long as the rate of inflation is low, the divergence between this approximation and the true annual rate of change is very small.)

(I summarise how these measures are calculated in a short appendix.)

Researchers have studied the properties of these inflation measures, and argued that they have better predictive properties for headline inflation than core inflation. This seems reasonable, but the core inflation measure is actually more useful for analysts making short-term inflation forecasts in practice. In particular, for those who are making forecasts of inflation-linked bond carry.

We can decompose inflation forecasting into two steps:

1. forecast core CPI; and
2. forecast food and energy CPI. 

There are a great many economic forecasters who focus on core CPI, so we have a great many methodologies available for the first task. We then need to look at the second task, which mainly depends upon energy prices. (Food prices are increasingly stable, as populations have moved to eat in restaurants and manufactured "food products.")

A trimmed mean CPI does not offer the ability to make such an analytical decomposition. As a result, I have deeply studied the literature discussing those measures.

Appendix: Summary of Trimmed Mean CPI Calculation

The calculations for a trimmed mean and median CPI are fairly straightforward, but the difficulty with them is that they are based on weighted averages. (All of the groupings in the CPI basket have different weights, based on consumption patterns.) Dealing with the weightings makes the calculations look more complicated than they really are.

Within a trimmed mean CPI, the monthly inflation rate for the index is calculated as follows.

1. Arrange all of the monthly percent changes of the groupings within the CPI in order.
2. Exclude a certain percentage (the Cleveland Fed calculates an index using 16%) of groups (based on the their CPI weighting) with the highest and lowest monthly changes. For example, if the two groupings with the highest monthly percent changes both have weights of 8% of the total CPI, they would be both excluded from the calculation. If the top 16 groups each have a weight of 1%, we would exclude all 16.
3. We take the weighted average of all of the remaining groups.

For a median CPI, the calculation is simpler -- we just take the median monthly percentage change, taking into account the group weightings. That is, 50% of the weighted groups will have a percentage change greater (or equal to) the weighted median, and 50% of the weighting will have a percentage change less than (or equal to) the weighted median.

In both cases, we may need to "split" a grouping in order to do the calculation. For example, if we are calculating a 16% trimmed mean CPI, and the grouping with the least percentage change has a weighting of 20%, we would include that group within the mean calculation, but with a (non-normalised) weight of 4% -- that is, we lopped off 16% of the 20% of that group.

(c) Brian Romanchuk 2016