What’s in the Core of the CPI?

This morning, BLS reported that the consumer price index increased 2.1 percent over the last 12 months. The index for all items less food and energy, commonly known as the “core CPI,” rose 2.2 percent over the same 12 months.

The CPI is used for many purposes; when should the core CPI be used instead? Cost-of-living adjustments, such as those given most years to Social Security beneficiaries, are designed to escalate the payments to maintain their level after accounting for inflation. Because the escalated payment needs to account for inflation in all the items bought by beneficiaries, the CPI is the appropriate measure. Similarly, when an index is used for deflating a nominal value to remove the effects of inflation (for example, deflating a nominal wage to derive a “real” wage), the CPI is more appropriate than the core CPI.

Where the core CPI is potentially more useful is when the user is interested in the index as a general indicator of inflation. The idea here that the Fed is interested in controlling future inflation. The CPI is an average of inflation rates for many components (that is, individual goods and services). Some component prices are more volatile than others, and are thus less informative about trends or future inflation.

For example, prices of fruits and vegetables change frequently; their price changes often reflect temporary supply factors, such as the effects on agriculture of unusual weather. The Fed, however, may want to focus more on price changes related to the demand-side of the economy, because that’s what it controls through monetary policy. So, it may want to focus on prices that change less frequently—prices of things like medical treatment, tuition, or rent. The core CPI is the BLS index that removes food and energy prices that tend to be more volatile than other products.

The idea of a core index is pretty old. Checking historical CPI reports available from the St. Louis Fed’s Fraser on-line library, the BLS has published a CPI excluding food at least since 1957. The CPI excluding food and energy was introduced in 1978 in the aftermath of the 1973 energy crisis.

The next advance in measurement of core prices came from research in the early 1990s by Bryan and Cecchetti. They observed that while food and energy prices are sometimes volatile, they are not the only prices that can be volatile. For example, in December 2016 “watches & jewelry” and “public transportation” showed unusually large price changes. Why not remove the most volatile prices each period, rather than pre-deciding that food and energy prices are the only ones to be removed? This approach is called the “trimmed mean” – a certain percentage of outlying observations are removed before the mean is calculated. A similar approach uses the median, or central price change, as a measure of core inflation.

The Cleveland Fed has carried on this work and each month releases the “median CPI” as well as the “16% trimmed mean” (that is, the average after removing 16% of the sample with the most outlying price changes). In the 12 months ending December 2016, both measures increased 2.5%.

Another variation on core CPI comes from the Atlanta Fed, the “sticky price CPI.” They note that the justification for looking at a core CPI is to focus on prices that are sticky, in the sense that they change infrequently, responding to overall inflation rather than to supply shocks. But the frequency of price changes can be observed directly from the microdata. So the Atlanta Fed researchers split the CPI components into “sticky prices” (prices that typically go more than 4½ months between price changes) and “flexible prices” (those that change more frequently than once every 4½ months). The sticky-price and flexible-price inflation rates are then calculated from each subsample. In the 12 months ending December 2016, the sticky-price CPI increased 2.6%.

The chart shows that all the core measures tend to move together, and are all much smoother than the all-items CPI. Fed researchers also find some evidence that the median CPI does somewhat better than the traditional core CPI (that is, the CPI excluding food and energy) at forecasting future inflation. The forecasting performance of the sticky-price CPI is mixed, but it generally does at least as well as the traditional core.

However, there are some problems with the statistical comparisons used by these researchers. First, as James Bullard, President of the St. Louis Fed, points out, the tests of core measures have all been based on assuming the forecaster wants to select a single measure to forecast future inflation. He argues, however, that the forecaster should really be looking at using all available information. In that context, we should be interested in the marginal predictive value of adding a core measure to supplement other available information, rather than in running a horse race between alternative measures.

Second, there’s a point that Nick Rowe calls “Milton Friedman’s thermostat.” The idea is that the Fed is already trying to control inflation, and if they’re doing a good job of it they are already taking account of past price inflation in setting monetary policy. The analogy is that the Fed is acting like a thermostat, adjusting monetary policy up and down to try to keep future inflation on target. This analogy suggests that correlations between past inflation and future inflation must be difficult to interpret because in addition to reflecting inertia in prices, they also reflect past actions by the Fed to offset that inertia.

I’d like to close this post with a suggestion for BLS. When looking at the Atlanta Fed’s analysis of CPI components with sticky prices and flexible prices, one thing that jumps out is that prices of “food away from home” (that is, restaurant food) are clearly sticky. That shouldn’t be surprising, because food is a relatively small part of the cost of restaurant food, and it’s costly to change menus. And looking at price changes, restaurant food is not especially volatile. In 2009, BEA decided to stop treating restaurant food as part of the “food” that’s excluded from its core PCE price index, and I suggest that BLS might also consider reclassifying food away from home to be part of its core CPI.