In this note we explain why in July the risks are skewed to the upside in core inflation space. The reason is that the seasonal adjustment filters expect a drop in the NSA level which is unlikely to happen in the current environment. Consequently, the filters can deliver large MoM readings in seasonally adjusted terms and revise up recent readings. We expect this effect in some EA countries, in the aggregate EA core HICP, and possibly in Japan (in the western-style core index). The US should not be affected, while in the UK the strength of the data can easily mask this effect. At the end of this note we summarize our view about core inflation and monetary policy across regions.
(Stop here or skip to the end if not interested in the details)
Previous research
Our research correctly anticipated the upward revisions in seasonally-adjusted data in recent months. Back in May, we wrote a note (here) to warn that the EA data was not as weak as many were suggesting. Since then, the continued upward trend in short-term government yields in the EA vindicated our call. In the note, we pointed out three things using Spanish core CPI as an example. First, the NSA level was trending higher with no visible deviation from the post-2021 trend. Second, the NSA level lost its pre-Covid seasonal pattern. Finally, because of the previous two points, the traditional seasonal-adjustment filters were prone to (much) larger than usual revisions. In real-time the filters estimated the MoM (ar) of Spanish core CPI in April at -1% but we explained why that estimate was downwardly biased and prone to large upward revisions (in the note we forecasted a 5pp upward revision on the MoM saar). Indeed, the latest estimate the MoM of Spanish core CPI in April is at 3.1%. The issue is that next month (July) is, again, particularly prone to misinterpretations/upward surprises/revisions. We explain why in the next sections and why we are so careful with our EA call in this moment.
(For additional research on seasonal issues in the EA the reader can see our previous notes here and here).
The NSA level issue
Most series have lost their seasonal pattern; this is a big issue. We use once again Spain as an example. We then extend the analysis to 10 EA countries, the EA aggregate, the US, the UK, and Japan. Figure 1 shows the NSA level of core CPI (index excluding unprocessed food and energy items) in Spain. As pointed out a few months ago two evidences emerge: it is unclear whether the NSA level has deviated from the post-2021 trend, and seasonality appears to be much less pronounced (which is typical of a high inflation environment).
Figure 1. NSA level of Spanish core CPI
Figure 2 shows the MoM (ar) of the index in Figure 1 in the 3 years before Covid. The red lines indicate the month of July. As indicated, the level used to drop in the months of January and July. This is a crucial feature because while in a low inflation environment a change in relative prices requires a drop in the level (i.e. a discount), in a high-inflation environment the level does not need to drop (at least not as much) to achieve the same result. Indeed, and unsurprisingly so, in 2022 the NSA level in July was flat.
Figure 2. MoM (ar) of NSA level of Spanish core CPI pre-Covid
If the level does not drop in July, the filters will interpret it as a large SA increase and will revise up recent history. Figure 3 shows the MoM saar (sa in-house) of Spanish core CPI under the assumption that the change in the level in the next two months will be the same as in the correspondent months of 2022 (a scenario which we evaluate as likely). The gray bars in Figure 3 shows the forecast (Jul and Aug). The takeaway is that the filter ends up revising up recent history (i.e. MoM in April would be 3.5% ar) and would deliver very high readings for July (and August). In a sense, this is like stating the obvious but this is where we are, especially in the Euroarea.
The bottom line is the following: for several CPI/HICP series/countries the risks are to the upside in July in seasonally adjusted space. For this reason, we suspect that most built-in seasonal adjustment procedures in common data providers (Haver, Bloomberg, Macrobond) will show up large MoM (sa) readings in July. Consequently, if one is relying (say) on X-13 to seasonally adjust the data and on the 3m changes to gauge the momentum of core CPI, it is quite likely to observe an acceleration from the current 3.7% (ar) to around 6.3% (ar) with little or no signs of deceleration.
(Please also note that looking at the evolution of the distribution of price changes mitigates this issue but does not solve it because in any case the data have to be seasonally-adjusted first).
Figure 3. MoM saar of Spanish core CPI
How widespread is this issue?
The issue is very widespread, each series in each country requires a specific treatment. To illustrate this point, consider Figure 4 which shows the NSA level of core CPI and core HICP for Italy. The two series show a radically different seasonal pattern, with the HICP being way more volatile than the corresponding CPI index (another evidence that the creation of the HICP can be considered a negative productivity shock..). For this reason, in Table 1 we have reported for each country/index the months more prone to upward surprises in seasonally-adjusted terms (that is, months in which the NSA level used to drop pre-Covid). The bottom line is that the months more prone to this end-of-sample distortion are January, July, and November.
After several simulations with different countries/filters, we have concluded that: (i) the countries most prone to this effect in July are Spain and Italy (and somehow France), (ii) the eurozone is prone to upside risks, although less than any individual country (of course), (iii) the US should not be affected, (iv) the UK is somehow prone to this effect but it can be masked by the strength of the data, (v) Japan is also prone but only the western-style index of core. (we can discuss each country in details via email)
Figure 4. NSA level of Italian core HICP and core CPI
Table 1. “The minefield” – months more prone to upward surprises in SA space
Conclusion
Here are our conclusions, region by region:
US: we think that the risks are skewed to the downside starting from the July CPI report (in case you missed it, in this note we explained why we think it is time to bet against the hawkish Fed).
Euroarea: for some countries the risks are skewed to the upside in the near-term due to end-of-sample distortions. For the Euroarea as a whole, we have been arguing for a terminal rate of at least 4% when the market was pricing it below 3% (we see with pleasure that some big sell-side houses have changed their calls to 4% in the last few days.. always a bit late). Right now, we are more careful because the distance between the ECB forecast and our own has narrowed. Risks remain to the upside but we think they are small.
UK: seasonal distortions are small compared to the strength of the data. Unfortunately, if history is of any guidance, a country does not get to 7% of core inflation (8-9% sequential) without going through some recessionary pain to restore price stability. It is unclear whether a 7% terminal rate can achieve that.
Japan: risks are to the upside in the near-term, including some seasonal end-of-sample distortions, especially for the western-style core index. We continue to think that a revision of the YCC will be necessary to bring core back to target.