There were massive revisions to the Job Openings and Labor Turnover Survey time series in the last release — going all the way back to the start in December of 2000. These were more extensive than the revisions last year. However, the old model fit is consistent with the new data, and the forecasts are pretty much unchanged (this is actually pretty astounding). Here is the data release from February (blue) and the new one from March (red) for hires and job openings:
And here are the models (JOLTS openings continues to fall a bit below the expected trend):
Actually, the revision made the hires model slightly better. If you look back to the previous post on JOLTS data, you can see the latest points are now completely in line with the forecast compared to being a bit above.
Update
Per anonymous comment below, the counterfactual has gotten a bit smaller over the past several months because the counterfactual recession date was fixed to 2019.7 based on the yield curve data. If we push the date out further — say, to 2022 — the counterfactual shock size gets bigger (and more uncertain):
With openings still falling a bit below modeled trend, is the red line/gray band merely the same model but with a shock that best fits the most recent below-trend data? Can't help but notice that the last shock was the GFC and obviously much bigger. The question is whether the red line/gray band is really meant to represent a realistic trajectory or is it simply for the time being falling in between the previous non-shock trajectory and the trajectory we would actually get in the event of an economic shock (i.e., if you were going to predict/model a recession, you'd get a much lower trajectory than seen in the red line/gray band)?
ReplyDeleteHope you can make sense of that or else perhaps simply explain in your own words how you interpret the most recent DIEM for JOLTS Openings as depicted in this post. Thanks, as always, and very interesting update.
The gray band is a counterfactual with a recession with a fixed center at 2019.7 (mid-September, per the yield curve inversion estimate). I did this because the fits to shocks that are just beginning are highly uncertain without any additional information (such as the yield curve data).
DeleteAs that date approaches without a larger deviation from the no-shock path, the fit shock will get smaller — essentially telling us we should push the date out further. Since the original estimate based on JOLTS data alone was late 2019 or 2020 (and the yield curve data is actually for the onset, not the center and so should have a later center), I will probably push the counterfactual date out a bit. However, as of now, I don't have a good number to put in there that isn't just guesswork. Once we have the yield curve inversion, I can then use that date to estimate a new recession center (and I'll do the center instead of the onset).
But I will add a new graph in an update here to give some idea of the effect.
Thanks for the clear response! The update is very helpful.
DeleteI guess still surprised that the upper end of the uncertainty band shows a path where openings never actually fall (in fact, rise fairly steeply). Is it because the JOLTS data only goes back to the early 2000s or is your sense that in milder recessions/shocks, openings managed to achieve such a feat?
Feel free to ignore, though, as I know it's just a model and perhaps conservative in some of its assumptions at that, but if you thought there was an interesting explanation or dynamic at work, would love to hear any final thoughts.
Again, thanks so much. Continued interesting work.
There are lots of factors that come into play that affect the counterfactuals, one of which is that the early estimates nearly always under-shoot the magnitude — and subsequently over-shoot. For example, see this estimate of the Great Recession:
Deletehttps://informationtransfereconomics.blogspot.com/2018/09/forecasting-great-recession.html
Another is the fact that the shocks are not actually symmetric — often steeper one side or the other. This will make a symmetric shock appear either too shallow or too steep compared to reality. It's a small effect, but it's amplified when looking at the leading edges of recessions. This can lead to additional under- and over-shooting behavior of the estimate. This latter piece I think has more to do with the kind of recession: a sudden surprise shock versus a gradual weakening of conditions until some threshold hits. I am thinking 2008 was the former, and 2019-2021 recession will be the latter (more like the 2001 recession). The former will overestimate the size early on, while the latter will underestimate it. But those are just my hunches — I haven't done the work that bears that out.