A deeper dive into US core inflation

Kevin Warsh has argued for lower US interest rates by pointing out that a less commonly used measure of inflation, namely, the trimmed mean, is close to the Fed's 2% target. Putting aside the problems with this argument – viz, that it might have been overtaken by events in the Middle East, the difficulty in persuading other Fed officials, and damage to the Fed’s credibility from advocating a lower measure after five years of overshooting the 2% target – CCI's analysis suggests that Warsh’s claim has merit, with a couple of trimmed mean measures performing best in approximating the trend in inflation. These measures are close to the Fed target, but that might prove short-lived. They have picked up recently on an annualised basis and analysis suggests further gains would lift them towards other series showing higher inflation rates.

To get a better handle on the Fed Chair nominee Warsh’s public support for trimmed mean inflation as a superior measure of core inflation to the core PCE deflator, we have constructed a set of trimmed measures of the PCE deflator and assessed their performance against both the Dallas Fed trimmed mean measure and the widely-used core PCE deflator.

Recapping how these series are put together:

• The traditional core PCE simply excludes food and energy prices from the calculation of inflation. 
• The trimmed mean series takes a very different approach, “trimming” large price falls and large price rises from the ranked distribution of all price changes and then calculating the average inflation rate from the remaining price changes.

The decision on how much of the distribution of price changes to exclude from the calculation of the trimmed mean varies from one central bank/statistical agency to the next. As an example, a 60th percentile trimmed mean excludes 40% of price changes, comprising price rises in the top 20% of the distribution and price falls in the bottom 20% of the distribution. In the case of the 50th percentile, the trimmed mean is simply the weighted median inflation rate of the distribution of price changes.

The Dallas Fed trimmed mean measure is highly unusual in that takes a much more aggressive approach than other central banks in excluding more of the distribution, where the exclusions from the top and bottom of the distribution are not equal. It covers only 45% of the distribution of price changes, excluding 55% in an asymmetric calculation of taking 31% off the top and 24% off the bottom of the distribution.

As you might hope, though, the different measures usually move in line with one another, such that they tell broadly the same story on core inflation.

Unfortunately, that is not always the case, and there is currently a striking divergence between the indicators. Nearly all the trimmed mean measures place annual core inflation in a range of between 2¾% and 3%, which is modestly below the traditional core PCE inflation rate of about 3¼%. However, two trimmed mean measures – the Dallas Fed series and the 60th percentile trimmed mean – put the inflation rate much lower than the core PCE at around 2¼-2½%.  

Putting aside likely spillovers from the Iran war for the moment, these differences are important because the lower readings on inflation from the Dallas Fed and the 60th percentile trimmed mean would justify a lower policy interest rate given they are much closer to the Fed’s inflation target of 2%.

This begs the question, which measure of inflation has been more reliable in tracking the trend in inflation. This is a difficult question to answer because the trend in inflation is unobservable. As an approximation of the trend, though, central banks have often relied on a long-centred moving average of actual inflation.

A variety of tests, some of which are summarised in the table below, suggested that the trimmed measures all outperform the core PCE deflator in tracking the trend inflation rate, with the 60th percentile trimmed mean performing best, followed closely by the Dallas Fed measure.

COVID did not seem to affect the findings as splitting the sample between pandemic and pre-pandemic periods did not change the results, although it is hard to statistically differentiate between the 60th percentile trimmed mean and the Dallas Fed series and some of the tests suggested that there would be a small payoff from averaging the two series.  

Given that the Dallas Fed and 60th percentile trimmed mean series have historically been better at following the trend in PCE inflation, this supports the argument that inflation is currently lower than suggested by the core PCE measure and hence close to the Fed’s 2% target.

Importantly, though, the low readings for annual inflation on these measures will likely prove short-lived, suggesting that inflation had not sustainably returned to the 2% target prior to the Iran war. This is because annualised inflation recently picked up for both measures, increasing from about 2% late last year to around 2½% and approaching 2¾%.

    

This cautionary note is supported by further analysis showing on how gaps between the different measures tend to be closed over a one-year horizon. The analysis provides modest support for the view that gaps between both the Dallas Fed and the 60th percentile trimmed mean and other measures of core inflation, such as the other trimmed mean measures and the core PCE inflation rate, usually – but not always – tend to narrow through adjustment in the Dallas Fed and 60th percentile series.

This suggests that the other measures of inflation, while noisier than both the Dallas Fed and the 60th percentile series, still have some information about the trend in inflation, and that the current gaps between the series could be closed in part by a further pick-up in the Dallas Fed and 60th percentile measures.