US, EA: V/U Falls – An Update

Our previous research

Our models have provided useful guidance. In the last 12 months, we circulated 3 US research notes using JOTLS series as measures of slack. The reader can find them here (March 2022), here (August 2022), and here (April 2023). We remind the reader that the results shown in those notes are conditional on the information set (including the assumptions on monetary/fiscal policies) at the time of the note. Nevertheless, in retrospective we can argue that the models served us very well. For instance, the August 2022 note reported a table (Table 1 here) in which we showed the forecasts of our “main” model using the ratio between the number of vacancies and the unemployed (V/U) as a measure of slack. A year ago, the model forecasted core PCE price inflation (Q4/Q4) at 3.6% in 2023:Q4 conditional on V/U reaching 1.5 at the end of 2023 (an incredibly accurate forecast!). In general, the main takeaway from our previous research on this point was that while some measures of JOLTS were showing signs of disinflation, the road to 2% required an additional (non-trivial) drop. Overall, we think that this call has been vindicated.

Updating the models for the US

We are getting close, but we are not there yet. Table 1 shows the forecast of our “main” model for core PCE price inflation using different measures of slack. The top panel of Table 1 shows the assumptions we are currently working with. For instance, in our baseline we assume a gentle rise of the unemployment rate (reminder: in our baselines we always try to be as close as possible to the central bank staff assumptions as inferred from public documents). As for the other measures of “slack”, we currently assume that the jobs opening rate falls to 3.5% by 2025, that the vacancy per unemployed falls to 0.8, and that the quits rate falls to 1.8%. Under these assumptions, the models continue to see a gentle disinflation ahead (bottom panel in Table 1). In our baseline, core PCE price inflation falls but remains above 2% at the end of the medium-term. The model using V/U delivers a forecast very similar to the baseline, while the other two measures of slack deliver a more benign picture. In terms of revisions, compared to the previous notes, the models are now a bit more optimistic, as the incoming data of both the left-hand-side variable as well as the exogenous have been a bit softer than expected.

The bottom line of this exercise is that the situation has improved a bit recently. However, ruling out a scenario in which core inflation runs persistently above target is complicated, unless one is ready to assume a more favorable “slack” than shown in Table 1 (which essentially means assuming some form of recession).

Note: the table shows core PCE price inflation forecast of our “main” model using different measures of “slack”. The upper panel shows the assumptions of each scenario in Q4 in each year. “Ugap” stands for “unemployment gap”.

Figure 1. Model-based forecast of core PCE price inflation (YoY)

Using the Jobs Opening Rate

Using Vacancies per Unemployed (V/U)

Using Quits

What about the EA?

The EA seems to be where the US was 6-9 months ago. Table 2 shows the same exercise for the Euro area. While in the latest ECB macroeconomic projections there is no significant slowdown and the unemployment rate remains around 6.4% (as in our baseline), for this exercise we are assuming that the job opening rate falls to 2% and the vacancies per unemployed falls to 0.2. The idea of this exercise is to assess what would need to happen on the labor market to reach target. Under these scenarios, core HICP falls more than in the baseline, as the labor market cools off but remains above target in 2025 (Note: the YoY in 2025:Q4 is just a touch above 2%. Table 2 shows the average of the YoY in each year, not the Q4/Q4). The figures of the two scenarios are: here for the vacancy rate, and here for V/U.

The takeaway of this exercise is that to our eyes, the Euro area looks like the US at the time of our previous notes on this point. In all scenarios core HICP does moderate going forward because some transitory factors dissipate and the expectations remain anchored. But based on today’s information set, going back to 2% requires quite an additional help from the labor market and a significant fall of the vacancies.

Note: the table shows the core HICP price inflation forecast of our “main” model using different measures of “slack”. The upper panel shows the assumptions of each scenario in Q4 in each year. “Urate” stands for “unemployment rate”. Core HICP price inflation is expressed as average of the YoY in each year.

Ok but PPIs (and PMIs) are falling.. then what?

PPIs have some contemporaneous predictive power but low power at lower frequencies. We often get questions about PPIs (and PMIs) falling. How much signal should one take? How much should one extrapolate going forward? As an example, Figure 2 shows the correlation between core goods CPI (excluding used cars and trucks) and PPI final demand, finished core goods. As Figure 2 makes clear, there is a clear correlation between the two series, with PPI anticipating CPI by 1-2 quarters. Having said so, once this correlation is properly tested, the evidence suggests something different. In this case, we have considered a simple model that includes PPI and 2 lags of it on the right-hand-side, and we have compared it to some alternatives: an MA(4) process, the unconditional mean of the CPI, and a random walk. In other words, we have tested if PPIs have better forecasting properties than recent history of CPI itself or a random process. The conclusion of this exercise is that the root mean squared forecasting error (RMSFE) of the PPI model is not bad (for instance it is 1.16 at 1-quarter horizon, and 1.24 at 4-quarters horizon) but it does not beat the MA(4) process. In other words, looking at PPI or recent history of CPI itself to forecast CPI is pretty much the same thing. The reason, which is also the takeaway, is pretty clear by looking at Figure 2: precisely because the correlation between PPI and CPI is mostly contemporaneous, PPIs are very useful to nowcast CPI (and possibly the next quarter or so) but they are not useful to forecast at the horizons most relevant for monetary policy.

Figure 2. US CPI and PPI – QoQ ar

Conclusion and implications for the Fed/ECB

We (as well as the Fed/ECB) are aware of the slowdown. But it is not enough to rule out the real risk: a world in which inflation remains above target. Compared to 6 months ago, the situation has improved somewhat in the US and much less in the EA. The current slowdown should be welcomed by policymakers, not feared. The models continue to suggest that in order to reach target in the medium-term, some help is needed from the labor market. The fall in some measures of economic activity (PPIs, PMIs, etc..) suggest some downside risk. The models also continue to suggest that going forward inflation should moderate but, net of a recession, it is very hard to rule out a new equilibrium in which core inflation is persistently higher than target. As for policy, we have started to bet against the hawkish Fed at the end of June and we have not changed our mind. The ECB should recognize that the inflation problem in the EA is bigger than they think, but the suspect is that after September the concerns about growth will dominate, at least for a while (more on this tomorrow after the preliminary HICP).

Contact

Want something more tailored?

We provide tailored consulting on ad-hoc projects.