The Atlanta Fed's macroblog provides commentary on economic topics including monetary policy, macroeconomic developments, financial issues and Southeast regional trends.

Authors for macroblog are Dave Altig and other Atlanta Fed economists.

« Price stability and the monetary base | Main | Private sector forecasts at variance »

June 18, 2009

CPI: The left and the right of it

Updated 11:30 a.m.

We got a good reading of May's inflation numbers this week. On both the producer and the consumer sides, price measures for the month came in well short of market expectations. The prospect of deflation has been getting a good deal of coverage in the blogosphere; see Andy Harless' blog, Economist's View, and Paul Krugman's column.

Greg Mankiw, however, points out that a trimmed mean estimate of the consumer price index (CPI), which removes the large relative price changes in each month, makes the deflation story seem a bit, uh, exaggerated.

"As every grade school student learns when the teacher reports results of the latest test, the average of any data set can be thrown off by a few extreme outliers; the median is a more robust statistic to estimate the central tendency in the data.

"Right now, the two measures of inflation are diverging substantially. The standard CPI shows deflation over the past year, but that average is due to a few anomalous sectors, such as energy. If you look at the median CPI, which shows what a more typical price is doing, the inflation rate does not look very unusual."

While the median is certainly a valuable way to look at inflation, there is also some interesting information that can be gleaned from breaking down the whole distribution of prices.

The chart below (hat tip to Brent Meyer at the Cleveland Fed) shows another interesting feature of yesterday's CPI release. Notice the clear downward shift in the distribution of CPI component price changes. Over half of the prices within the CPI market basket posted growth at or below 1 percent last month, up from an average of 29 percent in 2008, with a whopping one-third of the price index posting declines in May.


Of course, one month does not a trend make, but the month's price numbers were nonetheless noteworthy.

By Laurel Graefe, economic research analyst at the Atlanta Fed

June 18, 2009 in Inflation | Permalink


TrackBack URL for this entry:

Listed below are links to blogs that reference CPI: The left and the right of it:


The median approach would be correct and helpful only in the case when all components of the CPI were random realizations of some stochastic process. Then, for short time series, the median provides a robust estimate of the distribution parameters. In the case when all prices in the CPI are inter-related and the total (amount of money, i.e. all prices times all quantities) in constrained, the median value just suppresses valuable information about the inherent interaction. The evolution of a weighted mean value is superior in presenting real processes.

Posted by: Ivan Kitov | June 19, 2009 at 09:47 AM

Neither trimmed-mean nor median CPI statistics have any physical meaning. Energy cost is increasing now just like last summer - and that needs to be reflected substantially within the CPI, not zeroed out. The CPI must reflect everything actually spent by a citizen, including both energy cost and home mortgage price (not OER). Yes, the CPI can be thrown off by wild swings in any one component - similarly, overall consumer habits will be vastly altered by wild price increases in any one component of CPI. I don't have the capacity to whimsically choose to not spend any money tomorrow on energy - dwelling electric and car fuel bills are essentials.

The reason the US is/was the economic center of the world is the transparency and reliability of our data. It's time to return to basic principles.

Posted by: Unsympathetic | June 19, 2009 at 04:07 PM

Mankiw's test score example is inappropriate. The effect of the mean-trimmed approach described is more analagous to dropping the Math grade when calculating a student's overall GPA.

Also, this problem is compounded when component values of the index are related to each other in a lagging manner. For example, a large decrease in owner equivalent rent may cause a large decrease in other price components in subsequent months; while OER may be trimmed one month, other components will be trimmed subsequently and the overall price decrease may never show up in the mean-trimmed or median index.

The mean-trimmed and median approaches are more appropriately applied to the raw data used to calculate the component values. For instance, if one family reported that the price of bread had doubled in one month, this would be a true outlier and deleting it would likely yield better reults.

Posted by: Paul | June 20, 2009 at 08:13 PM

I forgot to mention, I *do* think that the change distribution chart is a helpful way of looking at things, particularly if OER is broken up into 4 components to better even out the weightings as suggested by the Cleveland FED. Considering the large number of components declining, I suggest adding "-1 to 0", "-2 to -1", and "< -2" columns to the chart.

Posted by: Paul | June 20, 2009 at 08:22 PM

Commenter Unsympathetic argues that consumers do not have the option to cut spending on energy because they are essentials, and therefore one should not set aside energy prices when looking at the CPI. The experience in 2007 - 2008 doesn't support this view. Oil prices spiked, and consumers reduced demand, and then oil prices fell. The recent experience of Juneau Alaska a few years ago is also relevant. They lost something like 30% of their electricity supply overnight (avalanche took out a transmission line) and consumer electricity rates spiked because of the expense of providing power using standby generators. Consumers very quickly changed their habits and substantially reduced their electricity consumption (I forget the percentage but it was something like 20-40%). The point being that energy price shocks do get blunted by consumers reducing demand, and I think this effect would be magnified in our current recessionary environment, so I think it is appropriate to set aside energy prices when looking for trends in the CPI.

Posted by: AndyfromTucson | June 21, 2009 at 07:20 AM

It WOULD BE noteworthy if some Economist somewhere demonstrated enough concern for her chosen field of study to argue the societal benefits from correlating inflationary measurements to a population sampling - any sampling. Have FED Economists learned nothing from the unmasking of mpt?

Posted by: bailey | June 22, 2009 at 06:31 PM

There isn't much more room to expand the money supply without increasing inflation (for the entire year).

However c. Aug-Sept real-gdp will begin dropping again. I.e., stagflation (business stagnation accompanied by inflation) will overtake the landscape.

Posted by: flow5 | June 23, 2009 at 11:15 AM

the data for 1966 is absolutely clear proof

Posted by: flow5 | June 26, 2009 at 03:13 PM

CPI is traditionally one of the most volatile numbers to watch.

I think you need to look at other factors to see if we are in a deflationary cycle or inflationary one. CPI on a stand alone basis is not clear enough.

The savings rate of the US has gone up. This is clearly deflationary.
Home values are still declining. This is deflationary. Unemployment continues to rise, deflationary.

On the other hand, fiscal policy is certainly inflationary, and Fed policy is inflationary. Fed policy will have more to do with inflation than fiscal in the long run.

I think that velocity of money has slowed to the point where it will take significantly longer for Fed policy to make things happen. Credit markets still are not functioning at optimum.

There is a lot of fear in the marketplace, and this keeps inflation in check, for now.

Posted by: Jeff | June 27, 2009 at 10:47 AM

What I think the inflation measures should reflect:

(a) accurate calibration of sums, because effect of inflation is important in accumulation, so that change of long term measurements is the appropriate accumulation of short term measurements.

But it is also the case that certain sectors have significantly more variance month to month than others; and this difference in variance persists.

(b) So the inflation estimate ought to have lower variance, which is what the trimmed mean attempts to do in an unsatisfactory way.

To me (I'm a physicist with experience in signal processing and machine learning) the appropriate response then is to have different time-scales for averaging (slower for high variance vs faster for low variance) for the different sectors, then generate the composite statistic and report changes of this.

Also, the inflation statistic would be more publically honest if economists et al used one which does include food & energy instead of the "core inflation" BS---which induces contempt by the public.

I might suggest Kalman filters of the deflator for each sector with varying smoothing parameters. Lots of smoothing for food and energy, less smoothing for other things. But importantly there should be no long-term growing mismatch between smoothed and raw deflators for any sector. This is because food and energy are truly important components of people's expenses and accumulation of changes are critical for actual inflation and perceptions of it.

Posted by: Matthew Kennel | July 03, 2009 at 12:18 AM

Post a comment

Comments are moderated and will not appear until the moderator has approved them.

If you have a TypeKey or TypePad account, please Sign in

Google Search

Recent Posts

December 2014

Sun Mon Tue Wed Thu Fri Sat
  1 2 3 4 5 6
7 8 9 10 11 12 13
14 15 16 17 18 19 20
21 22 23 24 25 26 27
28 29 30 31      



Powered by TypePad