Everyone seems to know how the HICP works. Are you really sure?
This is the second part of a didactical note about the mechanics of the HICP (Part I is here). In Part II, we discuss the bias in the HICP and how the “unchained” index can be used to gauge where we are in terms of disinflation. We show that, net of seasonal adjustment and chaining biases, 2023 is still marginally worse than 2022. We also show that, when comparing apples to apples, unsurprisingly the US seems to be way ahead of the EA. We proceed by Q&As.
Is the YoY of the published (chained) core HICP biased?
Yes, absolutely and it is no mystery (see for instance this ECB paper on the point). The intuition is pretty simple: the level of the series in a given month is “boosted” by the level of the previous December. Therefore, the YoY is biased. Another chain-linking procedure would return a different YoY.
Are there other chain-linking options that could be applied in the construction of the HICP?
Yes, there are two other options in theory. The published HICP indexes are chain-linked using December as the overlap or linking month. This technique is called “one-month overlap technique”. The other two options are the “annual overlap technique” (overlap with the annual average of the previous year), and the “over-the-year technique” (overlap with same month of the previous year). In the other two options, the index is constructed using the average level of the previous year and the same month of the previous year, respectively.
What distortions does each chain-linking procedure introduce?
Every chained HICP index you can construct is distorted in its own way. The general answer to this question is.. “it’s complicated”. The reality can be summarized as “each technique is biased in its own way”. The list of distortions is quite long and includes biases of the MoM changes, biases of the YoY changes, amplifications of the seasonal variations, etc.. A non-exclusive comparison table of the properties of chain-linking techniques can be found in this ECB paper which also discusses in details the statistical distortions due to chain linking over December. Table 1 below shows a summary of whether the MoM, YoY and annual averages are distorted across the three methods. The reader should note that the technique which results in an unbiased YoY is the “over-the-year technique” (that is, NOT the one used for the published core HICP index). Therefore, what virtually everyone does (taking the published YoY changes of HICP indexes as a “true” or “unbiased” inflation rate) is completely wrong and comes from the fact that they do not understand how the HICP is constructed. This is also why we generally do not give importance to acceleration/deceleration of the YoY changes, unless it is the YoY in December.
Table 1. Properties of chain-linking techniques
How does the other two chain-linking options look like right now?
Alternative measures are running lower (of course). Figure 1 shows the level of published NSA (chained) core HICP (the blue line) together with two other indexes we have constructed. The yellow line in Figure 1 shows the index constructed using the “annual overlap technique”, while the light-blue line shows the index constructed using the “over-the-year technique”. As a reminder: all indexes in Figure 1 are constructed starting from the same unchained MoM changes. Two evidences emerge from Figure 1. First, in a low inflation environment the difference between the 3 indexes is small but it is not anymore in this high-inflation economy. Second, the level of the blue line is higher than the alternative indexes, especially now. This should not come as a surprise: it is the result of the specific chain-linking procedure chosen by Eurostat. Because the level of the series in December is generally higher than the average of the year (and in this environment of each month of the year), the “one-month overlap technique” of Eurostat is boosting the level of the following year. This is one of the reasons why we have been so “inflationistas”/hawkish and out of consensus about the EA: we do understand how the HICP is constructed and what biases it suffers from. (Please note that this upward bias is expected to continue for a while. Not only, but the alternative indexes also suffer from biases and the bias can flip sign across methodologies during the disinflation phase).
Figure 1. Published NSA core HICP level and alternative chain-linked indexes
Can we construct a US-style HICP index?
A “US-style” core HICP index is also lower right now. Figure 2 shows the level of published NSA (chained) core HICP (the blue line) together with an index (the red line) we have constructed with a simple technique: cumulate the NSA MoM changes of the unchained index. This is nothing but what happens in the US with the CPI index. Once again, unsurprisingly the red line is below the blue as the level of the red line is not boosted by the “one-month overlap technique” of Eurostat. For the record, the YoY of the red line is currently at 4.84%.
Figure 2. Published NSA core HICP level and “US-style” HICP index
Note: the chart shows the published core HICP index (blue line) and a core HICP index (red line) constructed by cumulating the NSA MoM changes of the unchained core HICP index.
Does this mean that inflation is decelerating in the Euroarea?
NO, in fact it is almost the opposite. See below.
At the end of the day, is there an “easy” way to survive to all the above?
Yes, and it is called “acquired inflation”. When we talk about inflation in the EA there are generally two sources of confusion. The first one is that the data are published only NSA and everyone has his/her own way of adjusting for seasonality (in most cases, a terrible way that introduces more noise than anything – see our ad-hoc notes here and here). The second one is that virtually everyone ignores that the published indexes are chain-linked with a specific methodology that introduces biases. If follows that the easiest way to gauge the evolution of prices is, in fact, the simplest: cumulate the NSA MoM of the unchained index over the year and compare it to previous years.
Figure 3 shows the comparison EA-US of the cumulative (“acquired”) inflation across years (in the chart “NYE” stands for “New Year’s eve”). For the EA we show the evolution of the “unchained” NSA index, while for the US the NSA CPI published by the BLS. In both cases the index is re-set to 1 on NYE. This is the only unbiased apple-to-apple comparison EA-US. Everything else, it cannot be compared across the Atlantic either because the data are seasonally adjusted in different ways or because the published HICP is chained.
The main takeaway from Figure 3 is that the red line (cumulative NSA inflation in 2023) in the EA is higher than in 2022. At the opposite, 2023 acquired inflation in the US is lower than in 2022 and, under our forecast for June (last point of red line in Figure 3, right panel), it is expected to cross the 2021 acquired inflation. (For the record: Figure 3 for the EA is driven by core services – figure here).
Figure 3. Acquired inflation in EA (left) and US (right) across years
Euroarea (NSA unchained core HICP)
US (NSA core CPI)
Figure 3 is the main reason why we see signs of improvement in the US while we remain skeptical for the EA. To the best of our efforts, inflation in the EA is neither accelerating nor decelerating.
Conclusion
Never underestimate how the data are put together.
P.S. The reader can now understand why, with respect, we tend to be skeptical of any HICP analysis going around. The good news for the reader is that (s)he is in the right place.