Our forecast
High core, limited disinflation progress. We expect headline and core CPI to expand 41bps and 40bps (5.0% and 4.9% at annual rate) in February, respectively. Figure 1 below shows the sectoral breakdown of our MoM forecast for February. We expect core goods to expand 22bps and core services 49bps. Our forecast is close to consensus this month and we perceive the risks around it as well balanced. Conditional on our forecast, the disinflation progresses should be very limited, especially on the services front. Figure 2 shows our (MoM) forecast errors since July. The mean of our MoM forecast error is -1bps for total and -2bps for core, while the absolute mean is 14bps and 15bps, respectively.
Figure 1. February CPI forecast – details
Figure 2. Forecast errors
Implications for the “main” model
Conditional on our core CPI forecast, we continue to expect core PCE prices to grow 5.2% (QoQ ar) in the current quarter. Consequently, the “main” model delivers a forecast which is identical to the one at the time (Feb 24th) of the January PCE prices report. The model (Q4/Q4) forecast is: 4.5% in 2023, 4.1% in 2024, and 3.8% in 2025 (Figure below).
Figure 3. Main model forecast of core PCE price inflation (YoY)
Note: the figure shows the latest run of our “main” Phillips curve model. The confidence intervals (C.I.) are estimated using quasi-out-of-sample methods (estimate the model over a sub-sample, forecast, and calculate the root mean squared forecast errors). First quarter of forecast: 2023:Q2. The model is based on Detmeister et al. (2014).
Why did the model revise up significantly its forecast since December?
The model is about to “deanchor”. Compared to the time of the January FOMC (see slide #30 in our Pre January 2023 FOMC Meeting package here), the “main” model has revised up its forecast by about half of a percentage point throughout the entire medium-term horizon. The reason has to do with the mechanics of the model and the inclusion of the first quarter in sample. Technically, the model includes 6 lags of the dependent variable (core PCE price inflation, QoQ ar) and contains the typical restriction of augmented Phillips curve models: the sum of the estimated coefficients of the autoregressive components sums up to 1 with the coefficient of pi* (or “underlying inflation”, the “anchor”). If the estimation of the model is stopped before Covid (2019:Q4), the coefficient of pi* is 0.49: about half of the dynamic of core inflation is estimated to be driven by the “anchor”, and the other half by the most recent readings (the lags of the endogenous). However, what happened since Covid is that every time a new quarter was added to the sample, the model re-estimated a higher persistency of the process (that is, it cut the coefficient of pi*). The inclusion of 2023:Q1 has brought the pi* coefficient to .14 (that is, the model is progressively becoming “unanchored”). Not only but because the model includes 6 lags of core PCE prices, and because inflation has been very high for more than one year, the model is starting to consider high inflation as the new norm. Another way of thinking about this is that *despite* the fact that the econometrician has forced the forecast to converge back to 2% at some point in the forecast (by imposing the ”anchor” restriction), the model is almost rejecting it because the recent quarters are so strong that the coefficient of the anchor is converging to zero. Indeed, taken at face value, the confidence intervals in Figure 3 imply a probability close to 0 for core PCE price inflation to return to 2% by 2025.
Implications for the “main” model
Last warning for the Fed staff. In our experience, the Fed staff forecast included in the Tealbook remains judgmental and models like our “main” one are used to assess the risks around the forecast. According to the January FOMC minutes, last FOMC the staff revised down its judgmental forecast, a very puzzling downward revision which, in our view, was based on a bad judgmental call. The cold truth is that not only the incoming data have surprised to the upside but the models are sending very clear warning signals. The Fed staff is far from perfect, and can make mistakes. But generally, it should not make them twice in a row.