This is the second part of the note, Part I is here. In Part II, we explain what drives core inflation in FRB-US and we compare it to a set of alternative models. We also discuss under what circumstances core inflation is expected to re-accelerate. The conclusion is that the Fed has been successful in taking the “disaster inflation” risk off the table, but it is very hard to take the “moderately above 2%” scenario off the table. The conclusion is the following: we suspect that, without large surprises, from here the path for lower rates is via lower growth rather than lower inflation.
We proceed by Q&As.
What are the drivers of the core PCE price inflation forecast in FRB-US?
It is complicated, but ultimately it is (only) about long-term expectations. There are two equations governing the dynamics of core PCE price inflation in FRB-US. These equations can be seen here (we briefly discussed them last March in this note). When we started working with FRB-US, we made a few changes so that the embedded Phillips curve could be closer to the Fed staff framework described in Detmeister et al. (2014). The changes we have implemented are orthogonal to the points we make here. There are 3 key ingredients in the two equations governing core PCE price inflation in FRB-US: (i) an error-correction mechanism in deviation from the 2% target, (ii) the estimated persistency of the process (which is hard coded), and (iii) the flatness of the Phillips curve (“slack” is captured by the unemployment gap). In plain English this means the following: “FRB-US forces core PCE prices to converge back to 2%. The speed of the convergence is determined by the hard-coded persistency of the process. The labor market does not matter”.
(While this might sound a bit rough, please note that it is a feature of any New Keynesian model. What drives inflation, the “trend” of the series is pi* (expectations), not the labor market).
Can you show us these dynamics?
Figure 1 plots the contributions to core PCE price inflation forecast. In the model, disinflation is achieved by keeping expectations (the blue bars) anchored, through a negative contribution of core import prices (the orange bars) which has been typical in the last 20 years in the US, and by “taking time” that is considering the estimated persistency of the process (the red bars). (The reader can think about “persistency” in terms of CPI rents/OER: a shock today does not show up immediately in published statistics..).
Figure 1. Contributions to YoY of “inconsistent” FRB-US core PCE price inflation forecast
Ok, but are you really saying that the labor market plays no role?
Yes and no. It is no mystery that the Phillips curve (PC) estimated on the pre-Covid sample is pretty flat. No matter what model one uses, the slope of the PC is around 10-15bps per 1pp of unemployment (or unemployment gap) on that sample. Therefore, it is impossible to explain what happened during Covid using the unemployment gap. In a nutshell, this is what is going on with FRB-US, given how the PC is specified in the model.
What about a non-linear PC?
Indeed, this works better. Several people have pointed out that the ratio between vacancies and unemployed (V/U) is a better measure of “slack”. Unfortunately, it is not possible to use V/U in FRB-US as the number of vacancies does not enter the model. However, we maintain a partial-equilibrium model to assess the risks. Figure 2 shows the current forecast using V/U as a measure of “slack”. The left panel of Figure 2 shows the baseline in which we currently assume that V/U falls to 1 at the beginning o 2024 and stays there. The right panel shows a scenario in which we assume that V/U stops falling now. For the record, the slope of PC using V/U is estimated around 0.7. This put the contribution of the labor market during Covid to about 1.5pp of core inflation (please note that in a PC, this is the contribution on top of expectations). In other words, if one believes in the framework, the labor market explains a bit less than half of the deviation from target (with the remaining part attributed to factors outside the framework, such as supply chain issues).
In any case, the takeaway from Figure 2 is that according to the models we need additional help from the labor market to go back to target. Specifically, the estimates suggest that V/U should fall at least another 1pp-1.5pp.
Figure 2. Model-based forecasts using V/U as a measure of slack
Assuming V/U falls to 1
Assuming V/U stops falling now
What about any other possible specification of the PC?
Same results. We run a (very) large set of alternative specifications before each FOMC. Specifically, we consider any possible permutation for a total of about 50,000 models. Each model differs in one dimension (different type of “slack”, different measure of “imported inflation”, etc..). Figure 3 shows the slide #24 of our Pre-July 2023 FOMC meeting package. Figure 3 shows the distribution of point forecasts of all models in each year. The takeaway is that no matter what type of PC one considers, on average the forecast is close to the baseline and remains above the Fed target.
Figure 3. Thick modelling approach
How sensitive is FRB-US to the assumed persistency?
A lot. For instance, if we assume a slightly more persistent process (simulations are here, orange line is the baseline and green line is this scenario), core PCE remains above 3% in 2025. For this reason, we are extra careful when working with the model and we re-estimate the persistency of the process every time.
What about the “other factors” not captured in a PC?
We generally do not discuss them because they tend to dissipate and because they do not have much predictive power in the pre-Covid sample (although some enter in some of the sectoral PCs we maintain). The list is long and it includes PMIs, PPIs, supply chain issues indexes, etc.. If anything, these indicators point to downside risks around the baseline.
What about non-PC models?
Again, downside risks right now. There are two alternative models: the monetarist (MV=PQ) and the fiscal theory of the price level. We never show them because the PC remains the workhorse model in central banks, but we monitor them. If anything, they point to downside risks around the FRB-US baseline (for instance, the correlation between core CPI and M2 can be seen here.
Can inflation re-accelerate from here?
It is hard, although not impossible. No matter what model one has in mind, in order for core PCE price inflation to re-accelerate, we need to see: (i) a rise in expectations, or (ii) a rise in V/U, or (iii) renewed “other factors” issues (i.e. renewed supply chain issues), or (iv) renewed acceleration in wage growth, or (v) a new fiscal stimulus, or (vi) renewed endogenous acceleration in M2, or (vii) a large depreciation of the Dollar. While the probability of each is non-zero, our judgment (as well as the evidence from the models) is that it is pretty low.
Can inflation get stuck at 3% and/or how confident can we be to go back to the 2% target?
That is the problem, Mr. Watson. The models we maintain have been estimating a very low probability of going back to target at the end of the medium-term since the end of 2021. At some point, that probability was close to zero (conditional on the back-then monetary/fiscal policy). In recent months, we have seen some downward revisions to the point forecasts. Our “main” model currently put the probability of hitting 2% at the end of 2025 around 15% (see yellow bands here). However, most models continue to estimate a non-negligible probability (as high as 40%) of remaining around 3% net of a recession (see for instance Figure 2 above). The main issue is that uncertainty at that horizon is extremely elevated and consequently the confidence bands are very wide.
Can the models revise down these estimates?
Absolutely, the model can be surprised down by the incoming data. In this case, we re-run them and inform you promptly.
Conclusion
An interesting and dangerous spot for the Fed staff and the FOMC. We waited to circulate this note to see how markets reacted to yesterday’s CPI report. Monetary policy in the last 2 years has taken the “disaster inflation” risk off the table. But we are not sure it can take the “a bit above 2%” scenario off the table. This happens not only because the models continue to suggest that going back to target requires additional slowdown (or a recession), but also because it might not be optimal for the Fed (in FRB-US the Fed does not force inflation to converge back to 2% because the sacrifice-ratio is too high). The conclusion is the following: from now, it seems that the path for lower rates is via lower growth rather than lower inflation, especially (as we think) if the Fed will stop being aggressive with the short-end of the curve.