The flash March core HICP report has generated different reactions. We offer 4 clarifying bullet points. In our estimates, the report was very solid. Some moderation might be coming, but the real challenge -disinflating core services with a high nominal wage growth- is ahead of us. We remain skeptical it can be achieved with minimal monetary effort and no help from fiscal policy.
Our research
Point #1: MoM in March was strong. As usual, different seasonal adjustment methods result in different answers. Figure 1, shows the evolution of the MoM (sa) using different techniques. As explained in a December note (here), we have a strong preference to use X-13 (which we show on flash day) or our own seasonal adjustment procedure (which we use in our CI-C model) because they result in much less volatile series. In March, they both point to a very strong MoM (about 48-49bps). Other methods, including (i) X-11, (ii) the ECB series, (iii) running X-13 on the flash series, or (iv) using the seasonals estimated until December 2022 (“2022 seasonals” in Figure 1) result in lower estimates but as mentioned, we do not take signal from them. Needless to say, the details in the final reading will provide more clarity.
Figure 1. MoM (sa) of core HICP according to different seasonal adjustment methods
Point #2: the March MoM sa can and will change next month. What a seasonal adjustment procedure does is to filter the raw series. Therefore, like any other filter, the methods used to adjust the published core HICP suffer from end of sample issues. What happens in practice is that every time a new observation is added to the series, the filter re-estimates the entire history (and even more the recent one), resulting in a slightly different MoM pattern. Can this be avoided? The answer is “yes” but it comes with a different cost. An alternative way of estimating the seasonal patterns is what the BEA/BLS do in the US: estimate the seasonals only once a year and use them for the upcoming year. In this way, the published (say) CPI seasonally adjusted figures do not change during a given year. However, when the seasonals are re-estimated at the end of the year, the new seasonals are applied back in history and they can result in a different MoM pattern (which is precisely what happened in the US 3 months ago – our note here). For the record, we have followed the BEA/BLS procedure in Figure 1. The resulting series (“2022 seasonals”) suggests a lower reading in March but higher readings in January and February with higher cumulative basis points in the first three months of the year (acquired inflation in 2023 using this series is 4.0%). In other words, there is no perfect method when estimating the seasonal factors and either option comes with some benefits and some costs.
Point #3. In April we expect 40bps for core HICP. Figure 2 shows the distribution of the 1-month delta of the MoM core HICP (i.e. how much the MoM changes from a month to the next) using different filters/techniques. One standard deviation of our own series is 8bps, while for the other series it is around 13-15bps. What this implies is that if we start with 48-49bps (MoM) in March using our series, it is reasonable to expect around 40bps next month (in which case (i) the filter would also revise down marginally recent history, and (ii) the YoY of core would remain at 5.7%). As for the other methods, they are more volatile and less predictable but our simulations (not shown) suggest that under reasonable assumptions they could print in the high 30bps MoM sa.
(Note: the evidence in figure 2 explains why, when contacted privately to discuss the forecast of sell-side investment banks, we have been puzzled to see that one house was forecasting 22bps MoM for March and another house was forecasting the YoY of core HICP to drop in March. Both forecasts implied a 3 standard deviation shock (!) to the delta of the MoM of core HICP, a pretty extreme event).
Figure 2. Distribution of 1-month delta of the MoM (sa) of core HICP using different methods.
Point #4. A simple way to reduce the seasonality issue suggests the YoY can remain elevated until the end of the year. A way to forecast the remaining part of the year based on the first three months is shown below. Figure 3 reports the correlation between the change (at annual rate) in each year between March and the previous December of the NSA core HICP (that is, how much prices have grown in non-seasonally adjusted terms in the first 3 months of the year) and the YoY at the end of the year. In other words, Figure 3 shows a possible forecast for the YoY at the end of 2023 (the orange dot), based on the price changes in the first 3 months of the year. While we do not pretend to base our forecast on a simple correlation, this exercise suggests that the YoY of core HICP can remain elevated in 2023, and surprise sell-side analysts, as it did recently, as well as the ECB staff.
Figure 3. Correlation between price changes (at annual rate) in the first 3 months of the year and at the end of the year.
Conclusion
The data suggests the Euroarea is similar to the US, we expect it to continue going forward. In our view and estimates, the evidence suggests that the Euroarea (core) inflation is quite similar to the US, just lagging a few months behind. In our judgmental forecast, which already accounts for a slowdown of core goods, core HICP ends the year a bit below 5%, well above the ECB staff forecast. If the US is of any guidance, as we think, there should be some moderation ahead, as the normalization of supply chains and energy prices will be reflected in the published statistics in the next few months. But the real challenge is to disinflate core services, with a stronger headwind (nominal wages) in the EA. In a nutshell: it is reasonable to expect core HICP to moderate to (say) 4%-5% “on its own”, but what will take us back to 2%?