The last 3 readings of the Average Hourly Earnings (AHE) might be biased by seasonal adjustment issues and shifts in the industry-mix. A model suggests that the “unbiased” MoM has averaged 37bps per month against the published 26bps. The model also suggests that in April the MoM could be 0.4%-0.5%. We explain why.
The March MoM is downwardly biased by seasonal adjustment issues
A “properly” seasonally adjusted AHE would have been close to 0.4% MoM in March. Figure 1 shows the MoM of AHE (total private, the black bar) and 10 sub-sectors. In March, AHE grew 0.27% MoM, a touch above the average of the previous two months but below pace in Q4. The March MoM was influenced by a large drop in one sector (“Other services”) which contracted almost 1 percent. Excluding the “Other services” sector (or alternatively assuming a flat reading in March for this sector – see next paragraph about this assumption) would have resulted in around 0.32% expansion of the AHE index. Because “Other services” never posted a large drop (except in June 2020 during the first phase of Covid), it is natural to ask why it dropped so much and how much signal we should take from it.
Figure 1. Published MoM of AHE (total private) and AHE sectors
Figure 2 shows the published level of AHE “other services” non-seasonally adjusted (the blue line), seasonally adjusted by the BLS (the red line), and seasonally adjusted by UnderlyingInflation (the thick black line). As for other series, we tend to favor our own seasonal adjustment procedure because it results in a smoother series. The NSA series went sideways in March, as well as our own series, while the BLS SA series dropped. We conclude that the March reading in “other services” is probably an anomaly due to the new seasonals (For the record: in March 2017 the NSA level went sideways and the BLS SA series did not drop but moved sideways as well. This is another reason why we are suspicious about the March 2023 reading).
Figure 2. Level of Average Hourly Earnings in “Other services”.
The composition effect in the first three months of the year
A model controlling for industry mix suggests AHE is moving sideways. AHE does not control for shifts of workers (and hours) across sectors (see Chapter 2 of BLS handbook). Constructing a fixed-weight index requires access to unpublished data and it is beyond the scope of this note. Nevertheless, we have set up a model to explain the MoM of AHE in terms of its own recent past, a measure of slack, a couple of Covid dummies, a variable capturing the “residual calendarity” effect (see here, here, and here for a discussion about “residual calendarity” in AHE), and the shares of employment across sectors. Put it differently, with some limitations, the model is an attempt to answer the question: “what would the AHE MoM have been if the number of weekend days was constant across months and there was no shift in employment across sectors?”. Figure 3 shows the MoM of AHE (the orange line) together with the fitted values of the model (in the figure, the extreme values above 1 and below -0.5 recorded during Covid have been omitted to help visualizing the fit). The R^2 of the model is 0.5.
Figure 3. AHE MoM (sa) and model fitted values.
The model suggests that composition issues biased the MoM of AHE upward between May 2021 and February 2022, as the orange line was constantly above the blue line. The bias disappeared in the last part of 2022 before turning negative in the last few months. The fitted values of the model in Jan-Mar 2023 are .38, .32, and .43 respectively (against the published .30, .21, and .27). Overall, the model suggests that net of composition issues the MoM of AHE remains elevated and it is moving sideways.
Ok, but what is going on at the sectoral level and where does the model result come from?
The intuition behind the model result is shown in Table 1: workers are shifting into sectors with lower AHE level and lower AHE growth. In the first 3 months of the year, the average MoM growth of AHE is 26bps. As usual, some sectors have posted stronger growth while other posted weaker growth. However, on average, sectors above the aggregate AHE figure are those with a negative change in employment share. For instance, “trade, transportation, and utilities” posted large AHE gains but its share of employment dropped. The opposite happened in sectors that have gained employment shares (such as “leisure and hospitality” and “private education and health services”): AHE in these sectors has grown less than the average, putting downward pressure on the aggregate number. Therefore, after controlling for employment shares, the model estimates that without shift across sectors, AHE in the last 3 months would have been higher than published by the BLS.
(Note #1: because of the way AHE is calculated, what matters for the bias is the level of AHE in each sector more than its growth rate. However, it happens that in the first three months of the year, the sectors that have lost shares of employment are those with higher level of AHE and higher growth of it, while the sectors that have gained shares are those with lower level of AHE and lower growth rates. An “easy” way to see this effect is to plot the level of AHE and its growth rate across sectors; the reader can see it here. In the figure, the red (blue) dots are the sectors that gained (lost) shares of employment. Crucially, all red (blue) dots lie below (above) the best fit because on average they have lower (higher) AHE level and lower (higher) AHE growth rates. This is the source of the downward bias in the published AHE in the first three months of the year.)
(Note #2: the changes in employment shares reported in Table 1 look “small” but they are not. In terms of pre-Covid 3-month changes, the shifts in the last 3 months are equivalent to 2-3 standard deviation shock in some sectors)
Table 1. AHE and sectors details.
The “residual calendarity” effect in April
April MoM of AHE will be boosted by “residual calendarity” and could round to 0.5%. As explained in previous notes (see here and here), months with higher number of weekend days are associated with stronger MoM of AHE, a feature known as “residual calendarity”. March 2023 has 8 weekend days (Sat-Sun), the lowest possible while April 2023 has 10 of them, the highest possible. In our model, the effect of one extra weekend day is estimated around 6-7bps on the MoM. Therefore, one should expect the MoM in April to be boosted by about 12-14bps, all else equal. For this reason, it is reasonable to expect a solid 0.4% in April (MoM, sa), but given all the above (that is, adding up the “other services” effect which should not repeat in April) it would not be surprising to get a 0.5% in the end.
Conclusion and implications for the Fed staff
The conclusion of this exercise is that we doubt there is much deceleration in wages at the moment. AHE is a useful monthly indicator that suffer from several biases. The construction of a true “fixed—weight” index is beyond the scope (and possibilities) of this note. Having said so, the message from our exercise is in line with other wage indicators (i.e. Atlanta Fed wage tracker) and suggests prudence in interpreting the data. In our experience, the Fed staff uses AHE only as a way to anticipate the wage and salaries part of the ECI. Putting everything together, it would not be surprising to see the ECI moving roughly sideways in Q1 and post another 4% (QoQ ar) growth rate. This would imply that wage growth remains above the level consistent with the Fed target and that monetary policy needs to remain restrictive.