The views and opinions expressed in this post are solely those of the author and do not necessarily reflect the views of the Georgia Institute of Technology or the Georgia Board of Regents.
Like many students before me, when I was working on my Ph.D. in the early 2000’s, I had grandiose ideas about how one day I would write a dissertation that would change the science.
But eventually, life and reality step in, and the goal becomes just getting out of school rather than changing the world. So, instead, I wrote a short dissertation that brought into question the rational nexus between impact fees and actual government expenditures. No one cared.
But, at the time, one of the things that I thought would be VERY cool would be to build an economic model of the U.S. economy whose equations were derived by artificial intelligence. Of course, AI at that time was mostly a hypothetical and certainly nothing like the true AI we have today. But that was the dream.
It is no longer a dream.
If I am going to sit here each week and criticize the Atlanta Fed’s GDPNow model (which more often than not has a forecast that is somewhere between left field and the moon), I should offer an alternative.
So I have.
Introducing GDPcast. A full econometric AI-assisted model of the U.S. economy that uses sophisticated equations combined with my own expertise and outlook for the primary economic drivers. Like GDPNow, it shows my estimate of the current quarter, and that estimate changes as monthly source data are released. These are largely the same data series that I report on here each month. As each government release lands, the affected component updates and the change flows straight to the headline. This is what makes the forecast a “live,” GDPNow-style “now-cast,” and what makes attributing a forecast change to a specific data release possible.
Then, the model produces an 8-quarter forecast. And I will publish that quarterly forecast here, but ONLY for paying subscribers of this blog. The first one will drop at the beginning of the fourth quarter.
So, is my model any good? Well, head-to-head against GDPNow over 2016–2025 — 689 matched forecast dates, both models scored against BEA's first-published number — EIG GDPcast is statistically indistinguishable from GDPNow (RMSE 1.43 vs 1.40). But here is the kicker…my GDPcast performs BETTER early in the quarter than does GDPNow, which is when a forecast is actually the most valuable. In the first six weeks, my error runs about 15% below GDPNow's (1.64 vs 1.91). And the reason is structural. Before any hard data for a quarter arrives, GDPNow leans heavily on momentum; my model (GDPcast) disciplines its no-data estimates with economic structure — an inventory rule tied to sales, income-driven consumption, and tariff-aware trade equations that strip out the gold flows that have whipsawed other trackers since 2025.
But don’t take my word for it. On the site, I will publish my track record for everyone to see. That is something that GDPNow doesn’t do. And if I had their record, I wouldn’t either.
Currently, GDPNow is “nowcasting” third quarter GDP of 4.0%. There is no way on this beautiful green earth that third quarter GDP will be anywhere near 4.0%. I am sitting on 2.1%. We’ll find out who is more correct in October.
I can’t wait.
Home Builder Sentiment
Builder confidence inched up one point to 35 in August (full release here), a touch above the 34 consensus. Current sales conditions rose two points to 39, sales expectations for the next six months held at 43, and buyer traffic held at 23. That makes sixteen consecutive months with this index below 40.
The interesting number is buyer traffic at 23 — unchanged, and the same story builders told in July and June. The people who build houses for a living are reporting that hardly anyone is walking through the door. And that is consistent with the permits/starts/completions data discussed below.