US: Replicating Warsh’s Diffusion Index

TL;DR, we discuss the diffusion index recently highlighted by Fed Chair Kevin Warsh. We attempted to replicate the measure, got very close, examine the evidence we obtain, and discuss how these data should be interpreted.

Warsh stated that “over the past 12 months, 54% of goods and services in the PCE basket showed a price increase above 3%. This is well below the post-pandemic highs of about 77%, but it remains well above the level of 32% in the two decades that preceded the pandemic.”

Why 199 items. The first issue we encountered concerns the number of items included in the calculation, namely 199. We are somewhat puzzled by this number because the PCE basket contains more than 240 elementary items. Where, then, do the 199 components come from? In principle, the 243 elementary items can be aggregated in many different ways, depending on the level of aggregation chosen and the subindices that are constructed. The question is therefore how the staff selected precisely these 199 components.

In our view, the most plausible explanation is that the calculation starts from the aggregation used by the Cleveland Fed to construct its Median PCE inflation measure. The Cleveland Fed uses an aggregation that results in 200 components (see “Current components” link here). The difference of one item between the Cleveland Fed universe and the 199 components mentioned by Warsh may be explained by the wording of his statement. Warsh explicitly referred to “goods and services,” which suggests that the additional component included in the Cleveland Fed universe but excluded from his calculation could be nonprofit institutions serving households.

This is, of course, our assumption, but it also appears to be the most reasonable one. It is not straightforward to determine exactly how the 199 items were selected, and, looking at the available aggregation structures, following the Cleveland Fed classification seems to be the most plausible approach.

Weighted or unweighted? 199 items or all PCE components?

Another issue we had to resolve was whether Warsh was referring to a weighted distribution, using both the individual price indices and their weights in the PCE index, or to an unweighted distribution based solely on the individual price indices. After extensive investigation and several attempts to replicate the reported figures, we concluded that Warsh was most likely referring to the UNweighted distribution.

The reason is straightforward. The statistics he cited for the latest 12 months, the post pandemic peak, and the two decades preceding the pandemic are all closely matched by our unweighted calculations. The table below reports the share of components with inflation above 3% using both weighted and unweighted distributions, calculated for the 199 item aggregation as well as for the full sample of 243 items.

Table. Share of PCE items with YoY prices above 3%.

As the table shows, the unweighted columns provide by far the closest match to the figures cited by Warsh. Over the latest 12 months, our unweighted measure based on 199 items gives a share of 53%, very close to Warsh’s 54%. More importantly, our estimate for the two decades preceding the pandemic is exactly 32%, matching his figure, while the post pandemic peak of 77% is also very close to the 75 to 76% peak obtained using our methodology. The figure below shows the diffusion indexes for the 199 unweighted PCE items.

Figure. Diffusion indexes – 199 unweighted PCE items.

We would therefore tend to rule out a meaningful role for PCE expenditure weights in the calculation cited by Warsh and conclude that the statistics are most likely based on an unweighted cross-sectional distribution.

In any case, we have calculated the full set of statistics under all these alternative specifications. Users can download the Excel files we have prepared, which report the results for core components and for the full PCE basket, using both weighted and unweighted measures and both the 199 item and 243 item aggregations. Excel files are here for 199 unweighted, here for 199 weighted, here for 243 unweighted and here for 243 weighted.

As the figure below shows, however, the different diffusion indices are highly correlated with one another and therefore tend to convey a very similar signal over time. Going forward, we will continue to monitor the unweighted distribution closely, as we have already been doing for several years on our website.

The figure below shows the evolution of the diffusion indices using differen aggregations (an Excel file containing the numbers of the figure below is here).

Figure. Share of PCE items with YoY prices above 3%.

A word of caution about the diffusion indexes

We consider distributional measures and diffusion indices extremely useful tools for assessing inflation dynamics. However, we would caution against interpreting Warsh’s argument too literally. The reason is illustrated in the figure below, which shows the evolution of the share of items recording inflation above 3% together with core PCE inflation.

There have been several episodes in which the distributional signal generated false positives. In 2000 to 2001, for example, and again in 2004 to 2005, the share of items exceeding the 3% threshold increased from around 20 to 30% to more than 50%, without a corresponding significant acceleration in underlying inflation.

This is not an argument against using diffusion indices. Indeed, we would not have included these measures on our platform if we did not consider them informative. The point is simply that distributional measures should be interpreted with some caution and in conjunction with the broader set of information on underlying inflation dynamics.

On our website

We remind readers that all our distributional statistics, including the diffusion indices, are available on our website. We have generally preferred to monitor the unweighted distribution, a choice that turns out to be consistent with the approach used by Warsh. In the coming days, in light of the evidence discussed above, we will update our code to monitor the entire distribution, whereas until now we have focused primarily on core PCE components.

Contact

Want something more tailored?

We provide tailored consulting on ad-hoc projects.