Last week, the release of the Spanish CPI for April generated some optimism as the data were taken as “weak”. This note shows why, on the other hand, the data essentially confirmed the existing trend. The intuition is the following: the pre-Covid seasonal pattern is much weaker now, possibly because firms are re-pricing more frequently. Therefore, comparing the data with previous years can be misleading. Not only, but traditional filtering techniques do not help because the end-of-sample distortion is larger than usual (that is, the real-time seasonally adjusted figure is likely to be revised up significantly going forward). In an environment where real-time data can send conflicting signals, in our experience the best advice is to monitor the evolution of the distribution of price changes. Without a clear improvement in this dimension it is very hard to be optimistic. A similar reasoning could apply to core HICP in case of a “weak” reading tomorrow with the same characteristics.
The April Spanish (core) CPI
The trend is intact, hard to be optimistic. Figure 1 shows the level of the (NSA) Spanish CPI excluding unprocessed food and energy items (left panel) and its MoM percent change at annual rate (right panel). Two main takeaways: first, the trend of the non-seasonally adjusted series is intact, (ii) the MoM changes are still trending higher, and (iii) the data do not have a seasonal pattern anymore (or it is much weaker than pre-Covid). Therefore, any optimism about the marginal data is at least questionable.
But the data were lower than expected. How can that be?
Pre-Covid, the NSA level displayed a clear seasonal pattern. Seasonality is now harder to see, or at least it is masked by the strength of the data. Therefore, if consensus is formed assuming that nothing has changed, then we are at risk of getting downward “surprises” in months in which the NSA level is assumed to be stronger (based on the pre-Covid seasonal pattern, as in April), and upward surprises when it is supposed to be softer. Put it simply: we can be in an economy in which the distribution of price changes has shifted higher and it has less seasonal variation than pre-Covid. If that is the case (which seems very much possible in an environment in which firms re-optimize much more frequently), the surprise in April could be reversed entirely in the next few months (in July in particular when the NSA level is expected to drop).
Figure 1. Level (left) and MoM ar (right) of NSA Spanish CPI index excluding unprocessed food and energy
What about seasonally adjusted data? And the YoY?
Seasonal adjustment methods can lead to (very) distorted results in real-time. Figure 2 shows the estimated SA MoM (left chart) using Census X-13. According to the filter, in seasonally-adjusted terms, “core” prices in Spain contracted about 1% in April, a result which is hard to believe in the current environment. Should we trust this result? And how distorted is it? In order to understand the severity of the distortion, we have simulated the level of the series in the next 4 months, assuming an NSA growth rate lower than in 2022 (average 30bps per month vs 40bps), and we have re-run the filter. The resulting SA MoM is shown in the right panel of Figure 2 (light blue bars indicate the forecasted months). Under this (admittedly generous) scenario, the SA MoM in April of the series would be re-estimated from -1% in real-time to 3.9%, that is a 5pp (!) upward revision. (The reader can visually compare the last dark blue bar in each panel) This is why we invite prudence in commenting real-time data and suggest to never take much signal from a single print.
Figure 2. Estimated SA MoM of Spanish “core” CPI
Note: the charts show the estimated SA MoM of Spanish “core” CPI. The left panel shows the real-time estimate ending the sample in April 2023. The right panel shows how the MoM would look like in August 2023, assuming that the NSA level of the series in the next 4 months will growth a bit less than it did last year.
The YoY helps only partially. By definition, the YoY does not suffer from seasonal distortion. However, changes in the YoY are a valid indicator if (i) seasonal patterns do not change (which is not the case right now), and/or (ii) the distribution will not shift again (which cannot be ruled out right now). Indeed, in the forecasted scenario of figure 2 (right panel), the YoY would remain well above 6% in August 2023 with no signs of improvements for more than one year (if this scenario reminds the UK to the reader, the reader is right).
Conclusion: the ECB and how to navigate the next few months
The next few months will be tough, ECB likely to get it wrong in our view. In a situation in which the data are distorted by N effects, the last thing a policy-maker should do is to take signal from a single print (remember the UK two months ago). For this reason, we are pretty confident that the ECB will get it wrong (that is, it will compromise on 25bps immediately). On our side, we continue to stress what has worked nicely in the last two years: it is all about the distribution of price changes. Until the distribution of core HICP will start normalizing, we will not be optimistic. We are not in the ‘70s. But we could easily land on an inflation rate persistently above target.