Everyone seems to know how the HICP works. Are you really sure?
This is a didactical note directed to the analysts/economists working with the EA HICP index in which we answer some of the most recurring questions we get about the mechanics of the index. The list includes: (i) what does it mean that the HICP is “chain-linked?” (ii) how does the “chain-linking procedure works?”, (iii) “why the weighted average of core goods and core services does not sum up to core HICP itself as for the US CPI?”, etc.. (In part II we evaluate the distortions of the index and suggest how to use the “unchained” HICP to assess where we are in the Euroarea)
We provide an Excel file (here) that shows how to chain and unchain an HICP index and how to construct the published core HICP series starting from core goods (NEIGs) and core services.
This exercise should also clarify why we are so careful right now with our EA call. As we show (in Part II), the “unchained” core HICP suggests that the situation in 2023 is, if anything, marginally worse than in 2022 at this point of the year. We discuss alternative chain-linking procedures and what can happen going forward.
What the HICP is NOT
The HICP is constructed with a procedure which is very different compared to national CPIs. Most people think that every month Eurostat observes the prices of N items, calculates the MoM changes, updates the level of each item adding the new observation (level of the new month = level of the previous month times the observed MoM change), and combines the items into the main aggregates with a weighted average using the consumption shares as weights. While this is (very roughly) what happens with most CPIs (for details about US CPI see here), the procedure to construct the HICP is very different. The key difference is a specific “chain-linking” procedure which we discuss in the next section.
The HICP chain-linking
HICP “chain-linking” implies that the value of an index in a given month is calculated by multiplying the level of the “unchained” index with the value of the chained index in the previous December. Yes, you read it correctly: it is tricky and it can introduce distortions. Let us start by understanding what Eurostat measures, how does it, and why. Figure 1 (from the HICP handbook –page 177 and following here) visualizes the idea. The main issue is that the HICP is meant to capture “pure price changes” only. This implies that the movements of the index should reflect only movements of prices, not of the consumption weights (which are updated annually, although not for every country). For this reason, every new years’ eve the price index of every item is re-set to 100. Over the course of the year, Eurostat observes the monthly price changes and updates the price level. This continues until the next January when, again, the price level is reset to 100. This procedure is shown in the top part of Figure 1 and results in a series with a “jump” between each December and the following January. This series -the light green one in the top part of Figure 1- is called the “unchained” index. Clearly, a problem arises: how do you chain the level of the series in each December with the following January? In theory, there are different possibilities (which we will discuss in Part II). In practice, Eurostat has opted to use December as the linking-month. As explained in details in the next section, this means that published (chained) HICP indexes that you download from your data providers (exemplified by the bottom line in Figure 1) are calculated from the “unchained” index (the line at the top of Figure 1) by multiplying each month in a given year by the level of the chained series in the previous December. Mechanically, this procedure “boosts” the level of the current year up to (around) the level of the previous December and eliminates the “jump” observed in the “unchained” index.
Figure 1. Annual chain-linking of HICP
Constructing the published (chained) core HICP index starting from core goods and core services – a step by step example
It is easy, if you know how to do it. Suppose that you want to create an exclusion index in HICP space by aggregating N items. For instance, suppose you want to create the core HICP index starting from core goods (NEIGs) and core services published (chained) indexes. Can you do it? If so, how? Here is a step-by-step guide using our didactical Excel file (download it here). The procedure involves 3 steps.
First step: unchain the published indexes. Because the published indexes are chained, the first step is to unchain them. This is done by dividing each observation by the level of the series in the previous December. Figure 2 shows how this is done in the Excel file. Columns B to F in the Excel file contains published data for core goods (NEIGs), core services, their weights, and the index of core HICP (which is what we would like to re-construct). Columns H and I contains the “unchained” NEIGs and core services indexes that we have constructed. For instance, as shown in Figure 2, cell H81 (the level of unchained NEIGs in April 2022) is calculated by dividing the chained/published index by its own level in December 2021.
Figure 2. The unchaining procedure – Excel file.
Second step: create the unchained aggregate index of interest. This is the most intuitive step as it is precisely what most people try to do with the chained indexes. Indeed, once we have the unchained indexes, we can work as we would do with the US CPI and take a weighted average. The unchained core HICP (column J in Excel) is nothing but the unchained NEIGs level times its weight in core HICP plus the unchained core services level times its weight in core HICP. The same would apply (say) to the calculation of the contributions. Put it simply, we can summarize it in this way: “Everything you have tried to do and failed (aggregating, calculating the contributions, etc..) with the published HICP indexes it is because the indexes are chain-linked using December as a reference. You just need to work with the unchained indexes”. The reason why it cannot work with published indexes is pretty clear: every item index is chained using its own level in the previous December. When an analyst/economist tries to come up with an aggregate index starting from elementary items, (s)he is ignoring this crucial part (which is, in fact, not necessary with unchained indexes).
Figure 3. The aggregation procedure – Excel file.
Third and last step: chaining the constructed index. Once we have the unchained index of interest (in this case core HICP), we have to chain-linking it, otherwise it would look like the line at the top of Figure 1. The procedure, as explained in this note, involves multiply the level of the unchained index in each month by the level of the chained index in the previous December (with the first December, the starting point, set to an arbitrary level – say 100). In the Excel file, this is done in column K. In this column we have also re-based the index to compare it to the published one. Finally, in column L of the Excel file we have checked that the published core HICP index matches our reconstructed index.
Figure 4. The aggregation procedure – Excel file.
We have now constructed an index starting from published (chained) series. The procedure involves unchaining the items series, taking a weighted average of the unchained indexes, and re-chain the constructed index of interest. The same procedure can be applied to construct whatever other index in HICP space.
Conclusion
Only in the Euroarea they (we) could come up with something like the HICP. In part two of this note, we will discuss the distortions introduced by the HICP chain-linking procedure and how to deal with them. We will also discuss how to use the “unchained” indexes to gauge where we are and the concept of “acquired inflation” in unchained space. In the meanwhile, the only possible conclusion is the following: only in the Euroarea they (we) pretend that the HICP is a nice and intuitive index to work with. Copy and pasting the BLS CPI-U methodology would have been a much better idea.