We have updated our set of Phillips curve models (a.k.a. “thick modelling” approach). The models now include our nowcast of 2021:Q4 in the sample and the forecast extends to 2024:Q4. As mentioned yesterday, compared to the last run, the starting point of the forecast is slightly higher and we have a slightly tighter labor market.
(As a reminder: the total number of models included in this exercise is about 200,000. The models differ in one key assumption (anchored vs unanchored long-term expectations), as well as for type of slack, type of import prices, type of energy prices, number of lags of each variable, and type of (non)linearity in slack. Our “main” Phillips curve model -which mimics the Fed staff inflation framework as described in Detmeister et al. (2014)- is a particular case within the “anchored” models. Specifically, the “main” model assumes a flat trend (anchored) at 1.8 percent. Other “anchored” specifications included in this exercise do not assume a flat trend and let pi* evolve over time according to the behavior of actual inflation).
Starting with the “anchored” specifications (panel A below), the models took signal from the incoming data. Compared to the last run, the models have revised up each year (especially 2022 and 2023) and the dispersion around the mean/median of the distributions has also increased. The mean/median in 2022 is now above 3 percent (while it was around 2.5% percent in the last run). The mean/median in 2024 is little changed, although the uncertainty around it has increased. The main takeaway from our “thick modelling” is that the risks around the SEP (and the Fed Board staff) are skewed to the upside and that it might take longer than what the Fed is anticipating to return to target. The good news for the Fed is that the models forecasts are still converging -on average- back to target. Partially, this result is due to the “anchored” assumption. However, some of the models assume a time-varying anchor following the idea of “adaptive expectations”. Therefore, at least for the adaptive expectations models there is a chance of landing on a permanently higher -although still stable- inflation rate. In our view, the fact that we still do not observe such outcome implies that while the current shock is very large, it is hasn’t left a significant permanent print on pi* (which seems consistent with the behavior of some measures of long-term expectations).
Switching to the “unanchored” specifications, the models are currently forecasting core PCE price inflation to remain elevated (as any shock in the accelerationist world translates into a permanent shift) throughout the entire medium-term with the mean/median remaining between 4% and 5%. As usual, we see the “accelerationist” (“unanchored”) world as a tail event, as currently there are no signs that long-term expectations are de-anchoring to the upside.
Inflation forecast (Q4/Q4 of core PCE price inflation, %) from "thick modelling" approach – 2021:Q4 in sample
Phillips curve models with "anchored" long-term inflation expectations
Note: the figure shows the distribution of point forecasts of a large set of Phillips curve models with “anchored” expectations. Total number of models is about 100,000. This set of models restricts the coefficient of the long-term expectations to sum up to one with the lags of the endogenous variable (so that movements in long-term expectations eventually translate into equivalent movements in actual inflation). Whiskers show the 25th, 75th, 10th, and 90th percentile of the distribution. The red line and the light blue dot in each whisker show the median and the mean of the distribution, respectively.
B. Phillips curve models with "unanchored" long-term inflation expectations ("accelerationist" models)
Note: the figure shows the distribution of point forecasts of a large set of Phillips curve models with “unanchored” expectations (also called “accelerationist” models). Total number of models is about 100,000. This set of models does not restrict the coefficient of the long-term expectations to sum up to one with the lags of the endogenous variable (so that any shock to actual inflation results in permanent innovation). Whiskers show the 25th, 75th, 10th, and 90th percentile of the distribution. The red line and the light blue dot in each whisker show the median and the mean of the distribution, respectively.
Inflation forecast (Q4/Q4 of core PCE price inflation, %) from "thick modelling" approach – previous run
Phillips curve models with "anchored" long-term inflation expectations
Note: the figure shows the distribution of point forecasts of a large set of Phillips curve models with “anchored” expectations. Total number of models is about 100,000. This set of models restricts the coefficient of the long-term expectations to sum up to one with the lags of the endogenous variable (so that movements in long-term expectations eventually translate into equivalent movements in actual inflation). Whiskers show the 25th, 75th, 10th, and 90th percentile of the distribution. The red line and the light blue dot in each whisker show the median and the mean of the distribution, respectively.
B. Phillips curve models with "unanchored" long-term inflation expectations ("accelerationist" models)
Note: the figure shows the distribution of point forecasts of a large set of Phillips curve models with “unanchored” expectations (also called “accelerationist” models). Total number of models is about 100,000. This set of models does not restrict the coefficient of the long-term expectations to sum up to one with the lags of the endogenous variable (so that any shock to actual inflation results in permanent innovation). Whiskers show the 25th, 75th, 10th, and 90th percentile of the distribution. The red line and the light blue dot in each whisker show the median and the mean of the distribution, respectively.