AI forecast of U.S. recession

Daily Thoughts 25 Mar 2025 (Free)

I asked Grok to forecast the onset of U.S. recession. This is the answer:¹

Forecasting the exact start of a U.S. recession is inherently uncertain because it depends on a complex mix of economic indicators, policy decisions, and unexpected events, and the official determination is only made after the fact by the National Bureau of Economic Research (NBER). That said, I can provide a reasoned outlook based on current trends and historical patterns as of today, March 25, 2025.

Economic indicators right now show a mixed picture. The Atlanta Fed’s GDPNow model recently estimated Q1 2025 real GDP growth at +0.4% (as noted in posts on X), a sharp drop from earlier negative forecasts like -2.8%, suggesting the economy is slowing but not yet contracting. However, consumer sentiment is weakening—University of Michigan’s Consumer Sentiment Index dropped 10.5% in March 2025—reflecting fears over inflation and Trump administration policies like tariffs. Unemployment remains low at 4.1% with 151,000 jobs added in February, per the Bureau of Labor Statistics, but labor participation has slipped slightly, and layoff announcements are up. The yield curve, a classic recession signal, is no longer inverted (10-year Treasury at 4.318% vs. 2-year at 3.979% as of mid-March), reducing one immediate warning sign.

Expert forecasts vary widely. J.P. Morgan’s chief economist pegs a 40% chance of recession in 2025, up from 30%, citing Trump’s trade policies (e.g., 20% tariffs on Chinese imports, 25% on Canada and Mexico). Goldman Sachs raised their 12-month odds to 20% from 15%, while Polymarket and Kalshi betting markets show 41-43% probabilities for 2025. These align with rising concerns over tariffs disrupting manufacturing and supply chains, potentially triggering a contraction. Historically, recessions often follow Fed rate hikes with a lag of 1-2 years; the Fed began hiking in March 2022, pointing to a possible downturn window from late 2023 to early 2025—though aggressive cuts since September 2024 (current range 4.25%-4.50%) might soften that.

Given this, a plausible forecast is that a recession could begin in late 2025 or early 2026 if current slowdowns deepen—say, if GDP turns negative for two quarters (a common rule of thumb) starting in Q3 or Q4 2025, driven by tariff impacts, declining consumer spending, and tighter financial conditions. The NBER might date it to a peak around July or August 2025, reflecting when activity crests before a sustained drop. But this assumes no major policy reversals (e.g., tariff rollbacks) or external shocks (e.g., geopolitical crises) that could accelerate or delay it. The economy’s resilience so far—avoiding recession despite years of warnings—suggests it could also muddle through with slow growth instead, a “soft landing” some still hope for.

Bottom line: Late 2025 is a reasonable guess based on today’s trajectory, but it’s far from certain.

It is an interesting answer, but requires some fact-checking. He (let’s argue that Grok is a “he” for now; he does draw himself as a man) states that Atlanta Fed’s GDPNow would estimate a growth of 0.4% for U.S. Q1. This is interesting, because the update to the nowcast comes in only tomorrow. If the updated figure truly is 0.4%, then we ask from where did Grok get this (the data for the update is not public yet)? (Addition on 26 Mar:) I could not find the posts showing 0.4% growth for U.S. Q1 he mentioned.

The yield-curve un-inversion is something we commented last week. To note, the 10-year/3-month Treasury spread has just un-inverted from the re-inversion. This zig-zagging is a serious recession signal (like we explained).

The spread of U.S. 10y/3mo yield curve with points of re-inversion of the curve in red. Source: GnS Economics, St. Louis Fed, NBER.

Otherwise I think that his guess is rather good, as it also fits to our timetable. We are currently forecasting that the U.S. recession begins at Q3. AI is becoming a good tool also in forecasting and we will use Grok more in the future, as least to summarize things.

Truly abstract thinking (e.g., scenario forecasting) is still in its infancy with these models, but they will eventually become useful in that too, I am sure. Interesting times we live in.

Best,

Tuomas

---

Disclaimer:

The information contained herein is current as at the date of this entry. The information presented here is considered reliable, but its accuracy is not guaranteed. Changes may occur in the circumstances after the date of this entry and the information contained in this post may not hold true in the future.

No information contained in this entry should be construed as an investment advice. GnS Economics nor any of the authors cannot be held responsible for errors or omissions in the data presented. Readers should always consult their own personal financial or investment advisor before making any investment decision, and readers using this post do so solely at their own risk.

Readers must make an independent assessment of the risks involved and of the legal, tax, business, financial or other consequences of their actions. GnS Economics nor any of the authors cannot be held i) responsible for any decision taken, act or omission; or ii) liable for damages caused by such measures.

1. The figure of this post is a result of asking Grok to draw himself making an U.S. economic forecast.