The NY Fed has published an analysis (“Inflation Persistence: How Much Is There and Where Is It Coming From?”) by Martín Almuzara and Argia Sbordone. Please, find below our review, including how this analysis relates to the Fed Board staff approach.
To keep in mind
An analysis by the NY Fed suggests that trend inflation (defined as the “common” component across consumers price items) has been trending higher since late 2021 and it is now around 5%. In our view, the NY Fed estimate is probably upwardly biased but there is evidence that the pick-up in the trend started earlier than implied by the NY Fed paper. A more reasonable estimate of trend inflation for the US right now is around 3%.
What the paper does
The paper offers an estimate of trend inflation. In order to do so, the authors estimate a dynamic factor model on monthly data for the seventeen major sectors of the personal consumption expenditures (PCE) price index. The intuition is pretty simple: The model (which builds on Stock and Watson (2016)) estimates sector-specific trends and then aggregate them using spending shares to construct the “common trend” which the authors label “multivariate core trend”.
Results
The main result of the paper is reported in the figure below which shows the estimated common trend (the light blue line), and headline PCE price inflation (YoY the black line, and MoM the gray dashed line). According to the authors’ results, the trend hovers below but close to 2 percent from early 2016 through late 2020 and remains remarkably stable in the face of negative inflation numbers at the onset of the pandemic. Sometime in the fall of 2021, the common persistent component came to dominate the evolution of the trend and today it stands as a significant driver of inflation. The YoY of the common trend is estimated at about 5 percent now. Finally, in terms of sectoral contributions, the authors argue that the contribution of durables and services are now of comparable magnitude. In other words, according to our model, the increase in the persistence of inflation is not explained by a single sector or group of sectors but is pervasive.
How is Almuzara and Sbordone (2022) related to the Fed Board staff approach?
In our view, the Fed Board staff employs a similar model (described in Luciani (2020) and Luciani (2021)) to decompose monthly PCE readings. Therefore, Almuzara and Sbordone (2022) results are not surprising but there are some important differences.
What shocks to identify? Almuzara and Sbordone (2022) model includes one common component and one idiosyncratic term plus noise. The peculiarity of the Covid shock is that it is (an additional) common shock across (several) items, not a sector-specific shock and not an idiosyncratic term. Therefore, by including only one common component in the model, Almuzara and Sbordone (2022) implicitly assume that any Covid-related shock (i.e. used car prices, re-opening sectors effect, etc..) ends up in the unique “common” factor of the model. This issue raises the natural question of what shock(s) one wants to identify and how to treat Covid-related factors. A solution to this problem has been suggested by Luciani (2021), a Fed Board economist who was working on the inflation desk back then. What Luciani (2021) suggested was to add a second common component and restrict it to zero pre-March 2020 such that it identifies common shocks related to Covid. Below you see the latest results of our CI-C model (“Common-Idiosyncratic and Covid” model) which is contructed in the spirit of Luciani (2021). (For the record: These are the same charts we circulate on CPI day). The main message from our CI-C model is that even if you account for the very specific nature of the Covid shock, you still find that the common component (net of Covid effects) started to trend higher in mid-2021 and it is still trending higher now. Not only but real time estimates of this model started to show a significant pick-up in the common component around August 2021 (see figure 2 and 3 below), much earlier than suggested by the NY Fed analysis. As for the latest observation(s), our CI-C model suggests that the YoY of the common component is a bit below 3% in March 2022.
Figure 1 Contributions to YoY changes of CPI excluding food and energy items
Note: the Figure shows the decomposition of the YoY percent changes of CPI prices excluding food and energy items. The contributions are estimated using our CI-C model, a 2-stage OLS-LASSO regression model. The “Covid” effect is identified with price variations outside the 10th-90th percentiles of each item pre-Covid price change distribution.
Figure 2 Estimated “Common” component: YoY, 3m/3m a.r. and 6m/6m a.r. – latest run (March 2022)
Note: the Figure shows the 3m/3m at annual rate (green line), the 6m/6m at annual rate (red line), and the YoY (blue line) of the “common component” estimated using our CI-C model. Latest observation in-sample: March 2022.
Figure 3 Estimated “Common” component: YoY, 3m/3m a.r. and 6m/6m a.r. – real time estimate (August 2021)
Note: the Figure shows the 3m/3m at annual rate (green line), the 6m/6m at annual rate (red line), and the YoY (blue line) of the “common component” estimated using our CI-C model. Latest observation in-sample: August 2022.
What is the “true” level of the common component?
The answer to this question depends on what to identify in the data and whether one thinks that the Covid shock should be treated as a special event to be modelled separately. In our view Almuzara and Sbordone (2022) estimates are upwardly biased and subject to (possibly large) revisions if actual inflation moderates in the next few months (dynamic factor models revise entire history every time a new observation is added in the sample). On the other hand, the estimates of the CI-C model are probably underestimating (at least to some extent) the level of the common component by treating Covid-effects separately. If we combine the evidence of Almuzara and Sbordone (2022) with our CI-C model and with alternative measures of core/trend inflation (i.e. San Francisco Fed Shapiro (2020) model, Dallas Fed Trimmed mean index, our set of 11 models of trend inflation, etc..) we estimate that the “true” YoY of the common component is now in the 2.75%-3.25% range.
All told, we conclude that: (i) there is evidence that the common component (no matter how estimated) started to trend higher in 2021:H2, (ii) some models (including our CI-C model) were sending alarming signals already in June/August 2021, and (iii) we estimate the level of the common component around 3% in March 2022.
Implications for the Fed staff
As explained in the past, in our view, the Fed Board staff sets judgmentally the level of underlying inflation in the price medium-term “decomposition” based on three pillars: (i) measures of trend inflation, (ii) the behavior of long-term inflation expectations, and (iii) the evidence from the general equilibrium. Below we report the usual slide from our “Pre-FOMC meeting package”. The main takeaway is that all three pillars are now indicating upside risks around the Fed staff assumption (pi* = 2-ish%).
Unfortunately, this has been the pattern since mid 2021: the Fed staff has constantly been behind the curve. The good news is that (by reverse engineering the Fed staff forecast inferred from the latest FOMC minutes) it seems that the staff has revised up pi* again (the first time being back in December). But the bad news is that the upward revision (of about 1-2 tenths in 2023/2024) is probably too small given the evidence of the data.
For this reason, we re-iterate the message we have been sending for several months: in our view, the Fed staff is (finally) sending very hawkish signals to the FOMC by revising up pi*. But what they have done so far is, in our view, still at risk of not being enough.