Yesterday evening, I started to think that maybe I should write a piece on the defense of President Trump. I think he’s really trying, but wrong staffing choices and enormous pressures (both outside and inside of his administration) are pushing him towards deepening errors, of which the next one can lead to cataclysmic consequences. Maybe later.
Today, I want to correct and update the U.S. inflation forecasts. Their latest version included a data error, where I had mixed seasonally adjusted and not seasonally adjusted data. This was just a careless error on my part. So, let’s fix them with updated data.
Like in the previous setting, I assumed that the month-over-month changes of the U.S. Consumer Price Index (CPI), core inflation, and super-core inflation would be unit root processes. In other words, and like detailed in the Weekly Forecasts 5/2026, I approximate the very long memory of the three series by assuming that they are I(1) nonstationary processes.
The Johansen Trace Test indicated that three series would have two stationary cointegrating vectors between them. Assuming the existence of two stochastic trends in the three-dimensional system of the time series makes sense, because we can plausibly assume that the core inflation and super-core inflation are linked to the CPI through somewhat differing statistical processes. This implies that the three-dimensional system of (approximated) I(1) nonstationary processes existing between them would have two (stationary) linear combinations linking the time series together through time.
I inserted the variables into a vector error-correction (VECM) model assuming a cointegration rank of two. I included a constant and 30 lags of the endogenous variables in the model to control for autocorrelation.1 The forecasts the model yielded are presented in Figure 1.
