As of 2nd release for 2025Q1.
Figure 1: GDP (bold black), GDO (light blue), GDP+ based on 2022Q4 GDP (tan), final sales (red), all in bn.Ch.2017$ SAAR. Source: BEA 2025Q1 2nd release, Philadelphia Fed, and author’s calculations.
Final sales growth revised down 0.3 ppts, as consumption growth revised down 0.6 ppts. All in all, the outlook looks less rosy than in the advance release, as inventory growth are up revised up 0.3 ppts. Remember, final sales more closely approximates aggregate demand than does GDP.
Way, way off topic – for the statistically curious (including, I assume, Menzie’s students):
https://terryflynn.substack.com/p/how-any-randomised-controlled-trial
The gist, to the extent I have gotten it, is that much of our statistical modeling, all randomized control trial, relies on an unstated assumption; that assumption is that, for any individual, a future response is going to be the same as today’s response. So if, in a medical trial, drug X cures individual Y’s viral hiccups, then in any future trial, it would do the same. There is no room, in a randomized control trial, to assess the likelihood of a different response next time. The assumption is no variability in response.
The author, Terry Flynn, uses voter preference to make his point. If his imaginary poll respondent, Jo, reports a very strong likelihood to vote Tory today, there is not need to consider how strongly or weakly that preference is held. The preference for any given party is fixed.
Now, we readers of political handicapping journalism know that efforts are made to assess changeability – that’s the meat and potatoes of speculation about swing-voter behavior. But in political polling, results ignore the odds that some individual voter(s) will have a different preference ranking in the future. Those odds might explain how Hillary went from sure thing to also ran in 2016. Flynn gives Theresa May’s lost majority in 2017 as his example.
Those not obsessed with the math of political prognostication might want to consider that, according to Flynn, ALL randomized control trials suffer from this weakness, of necessity. It’s a “one equation, two unknowns” problem, insoluable in the context of such trials.