Getting a Clear Signal on Inflation
From time to time, components comprising the Consumer Price Index exhibit some idiosyncratic price changes, obscuring the inflation signal in the data. Examples of this “noise” range from mismeasurement and holidays that are not linked to calendar dates (causing unanticipated seasonal variation), to one-time changes in excise taxes (like the recent increase in tobacco taxes). Often researchers and analysts tend to explain away (or exclude) these peculiar “one-off” price movements, and rightly so. Or, since the effect of such idiosyncratic changes on the CPI dissipates over time, it can be greatly minimized by looking at the data over a longer time period, though this technique comes at the expense of a near-term read on the data. An alternative is to use cross-sectional trimming techniques, such as the median CPI or 16 percent trimmed-mean CPI. They offer a way to reduce noise in a much more consistent manner, with no sacrifice in timeliness.
Recently, a couple of examples of these idiosyncratic price changes have shown up in the data. First, the price of used autos spiked. The spike, which began in August 2009 and seems to have receded this past April, corresponds to the duration of the CARS program (commonly referred to as “Cash for Clunkers”). During that program, used cars that were traded in were destroyed instead of making their way to used auto dealer lots. So some of the recent price change likely reflects an artificial reduction in supply. That said, given the tightness in credit conditions, it is also possible that some of the increase is due to a shift away from higher-priced new vehicles, which are usually purchased with a loan. Still, the correspondence between the price movements of used autos and the CARS program is striking: prices fell roughly 10 percent between the start of the recession and the month before the CARS program began, but since that time they have jumped 12 percent.
Another example of monthly noise in the CPI occurred in April, as club membership dues and fees for participant sports posted their largest monthly increase on record (the series goes back only to 1998). Prices spiked at an annualized rate of 31 percent, following a 16.4 percent decrease in March. Such a large increase immediately following a substantial decrease is indicative of a seasonal adjustment or mismeasurement issue.
A trimmed-mean approach with these sorts of price anomalies may be more useful than ad hoc exclusions. The median CPI and 16 percent trimmed-mean CPI eliminate much of the overall monthly noise by excluding the highest and lowest price changes—those that are usually symptomatic of idiosyncrasies. In fact, research shows that trimmed-mean measures are better predictors of future inflation than the headline CPI or the CPI excluding food and energy.
Recent trends in the trimmed-mean estimators have been decidedly disinflationary. The median CPI was virtually unchanged in April, rising at an annualized rate of 0.1 percent, and has been flat for the past six months. That pattern is much the same for the 16 percent trimmed-mean measure, which is up at an annualized rate of just 0.7 percent over the past six months. As for used auto prices, they were in the upper tail of the price-change distribution for seven consecutive months (August 2009 through February 2010), thus trimmed away in the calculations.
|Percent change, last|
|1 mo.a||3 mo.a||6 mo.a||12 mo.||5 yr.a||2009 average|
|Consumer Price Index|
|Less food and energy||0.6||0.6||0.3||0.9||1.9||1.8|
|16% trimmed meanb||0.2||0.3||0.7||0.9||2.3||1.3|
- a. Annualized.
- b. Calculated by the Federal Reserve Bank of Cleveland.
- Sources: U.S. Department of Labor and Bureau of Labor Statistics.