Market tops do not begin with "numbers" but with "sloppy stories"—3 signs Goldman is wary of

"The top of a bull market always begins with a 'story'."—Goldman Sachs Chief Economist Jan Hatzius and U.S. Equity Strategist Ben Snider discussed both the "numbers" and "behavioral psychology" needed to forecast 2026 on Bloomberg's "Odd Lots" podcast (recorded December 19, 2025). The moment the reason for rising stock prices begins to be explained not by earnings, but by a "contrived narrative," is the moment to be on alert. This conversation provides a fairly concrete blueprint of that warning sign.

1. While 2025 was called a "bubble," it was actually a market with "low euphoria."

Snider describes the recent market as follows: "It's often called a bubble, but it's one of the markets with the least amount of euphoria in recent history." The point is that rather than stock prices being disconnected from earnings and bought on "dreams," at least up to this point, it is largely characterized by the fact that "stock prices rose because earnings grew."

1-1. Signs that "prices are running ahead of earnings" appear in conversation

Symbolic of this is the "leap" in logic surrounding GLP-1 (obesity drugs).
Snider recalls, "There was a moment when people said, 'Drugs reduce weight, so fuel consumption will decrease, so buy airline stocks,'" and states, "When these kinds of narratives start appearing in conversation, it's a signal that prices are beginning to run ahead of earnings."

In other words, it is a behavioral finance observation that it is dangerous when investors' "desire to explain" precedes quantitative data.

2. What supported the stock price rise in 2025 was "earnings growth"

It is often viewed that the "AI market = only a few megatechs," but Snider emphasizes the "breadth." He explains that S&P 500 earnings growth was about 12% in the most recent quarter, and even excluding megacaps, the median was about 10%, noting that the "non-top" S&P (S&P 490/493) has also delivered returns of roughly 15% for three consecutive years.
The important thing here is the assessment that the market was not just lifted by "expectations alone," but that corporate earnings actually followed through.

2-1. The reason megatech slowed down is "deceleration in AI investment" and the "scent of debt"

On the other hand, Snider cites the following as the background for the recent weakness in some AI stocks.

• Growth in AI investment (CapEx) "looks like it will slow down next year"

• To expand further, "it looks like more debt will be needed"

This is the view that these two points cooled investor sentiment. It means we have entered a phase where the "funding of investments" is being questioned, rather than a peak of euphoria.

3. "AI boosted half of GDP" might be a misunderstanding

The most piercing counterargument in this dialogue is Hatzius's strong denial of "AI as the driver of GDP."
He cites two reasons for his view against the idea that AI investment has significantly boosted U.S. GDP.

1. AI-related investment goods are largely importedso it is a mistake to talk about GDP contribution by looking only at investment

2. Semiconductors are easily treated as intermediate goodsand are not directly reflected in investment statistics

As a result, he states that the contribution over the past 3-4 years is "at most about 20bp in total," and for the most recent year, it is "close to zero."
The lesson to be drawn from this is clear: "AI excitement" and "growth observed in macro statistics" do not move at the same speed.

4. The 2026 economy: Growth is solid, but will the unemployment rate not fall?

According to Goldman's outlook, even if the 2026 real GDP growth rate is about 2.6%, the assumption is that the unemployment rate will remain flat at around 4.5%.

Hatzius explains this seemingly contradictory combination through 'accelerated productivity.' Underlying productivity for the five years post-pandemic is about 2%, higher than the previous cycle (about 1.5%). Furthermore, he states that 'the AI factor may not yet be fully incorporated into that acceleration,' and that AI could add to it over the next five years.

4-1. Productivity is 'good news,' but it could increase frictional unemployment in the short term

Regarding the question of whether AI will destroy jobs, Hatzius acknowledges that while historically 'in the long run, there is no visible relationship between rising productivity and permanent worsening of unemployment,' in the short term, 'frictional unemployment (time taken to find a job)' could increase. In short, the assessment is that 'the cost of job mobility' becomes more of an issue than 'total unemployment'.
the cost of job mobility becomes more of an issue than total unemployment, is the summary.

5. The 2026 market: S&P 500 target and the 'triggers' to watch out for

Goldman's S&P 500 target is '7676.' However, what they prioritize more than 'hitting the number' is what might trigger a collapse.

5-1. Ben's watchlist: Unemployment claims, Fed pivot, and 'sloppy narratives'

Snyder specifically lists the following:

• Weekly initial jobless claims ('worrying if they rise for several consecutive weeks')

• Signals that the Fed is leaning toward rate hikes rather than rate cuts

• Signs that investors are starting to justify things with 'poorly reasoned narratives' (leaps like GLP-1 to airline stocks)

These three are a framework for capturing 'top formation' through a combination of fundamentals (employment/monetary policy) and psychology (narrative).

6. Inflation and tariffs: How to read the numbers, including 'statistical quirks'

Regarding the background of why CPI came in weaker than expected, Hatzius suggests the possibility that the treatment of rent (shelter) was underestimated, explaining that it cannot be read as straightforwardly as a month-on-month comparison.

Also, regarding tariffs, while they are seen as contributing about 50bp to 2025 inflation, the view is that this is a VAT-like 'one-time increase in price levels' that could drop off as a 'cyclical factor' in 2026.

6-1. The '3 levers' companies used to absorb tariffs

Snyder states that the reason margins did not collapse due to tariffs is that companies used

1. price pass-through

2. pushing back on suppliers/supply chain restructuring

3. cost reduction/efficiency improvements

The point here is that this also connects to the discussion on productivity.

Summary: The key to reading 2026 is 'earnings' + 'employment' + 'quality of narrative'

The core of 2026 as indicated by this dialogue is simple.

• Stocks are ultimately discounted future cash flows = earnings as the foundation

• What shakes the premise of those earnings is employment (unemployment rate/jobless claims) and monetary policy

• And the scent of a market top appears in the 'sloppy narratives' coming from investors' mouths before it shows up in the data

That is why what you should really watch in 2026 is not flashy AI news, but Thursday morning jobless claims, whether more companies can quantify the effects of AI in their earnings reports, and whether the conversation among market participants has shifted from 'talk of earnings' to 'talk of far-fetched justifications'—that is the change to watch.

Recommended Articles

Next Big Wave (Growth Stocks, Seeds of Ideas, Deep Dives into Trends)