Precious metals — gold, platinum, and silver — provide a nice visual case study in how this plays out. All three blew past their long-term trends before collapsing (reverting) back to their 250-day moving-averages (mean).

These metals are a complicated story by themselves. Gold, in particular, carries a certain irreplicable1 aura: it’s part currency, part commodity, part emotional hedge against everything from inflation to geopolitical chaos, which makes its price moves as much about psychology as fundamentals. Central Bankers have been ditching US Treasuries (debt) for gold (which is a truly remarkable change).

Silver and platinum are more a story about industrialization and, yes, AI. Silver is the best conductor of electricity of any metal, and we need a lot more electricity for everything from data centers to electric vehicles. Platinum has lots of uses in industry and automobiles.

Whatever the underlying story, the pattern held up remarkably true. Parabolic moves upward tend to collapse back to longer-term averages, but, for these particular metals, the long-run price increases seem justified for good reasons.

This is not financial advice!

That’s interesting enough by itself, but even more interesting (in my opinion) is the sector2 driving that demand growth: commercial (which includes data centers).

Commercial and residential electricity sales tracked closely until diverging sharply during COVID. Notice that residential usage is much more volatile on a short-term basis, which mostly reflects weather. The chart marks ChatGPT’s late-2022 release as a reference point, and the timing lines up with when commercial’s climb steepens, consistent with data center buildout showing up as commercial load.

The chart alone can’t prove data centers are the cause because “commercial” is a broad category. But considering commercial office space is still heavily vacant post-COVID, the story seems fairly self-evident.

Share

How are other nowcasts/forecasts looking for Q2, 2026?

• Gemini (🤖 ChatGDP): ↑+2.2%

• New York Fed Staff Nowcast: ↑+2.7%

• Atlanta Fed GDP Now: ↑+1.7%

• St. Louis Fed Nowcast: ↑+2.0%

• Daniel Bachman’s The Nowcast: ↑+0.7%

• Median: ↑+2.0%

That blip at the end of the graph represents about 2.1 million people who are no longer in the workforce. This is also reflected in the labor force participation rate — the percentage of a population who are in the labor force — which fell to 61.5% in June 2026, the lowest outside the COVID period in 50 years. That said, the so-called “prime age” (ages 24-55) participation rate has mostly held steady at around 83%.

Why is this happening?

• Boomers retiring: 11,000 Boomers turn 65 every day, and many of them are leaving the workforce. The strong stock market is probably encouraging a lot of younger Boomers to retire early, as well.

• Immigration restriction: It’s unclear how many foreign-born people have been denied entry, detained, or removed in the last few years — the government statistics are not up-to-date — so I don’t want to speculate. What we do know is that foreign-born workers have a higher participation rate (65.6%) than native-born workers (61.0%), so their exit hits total labor supply disproportionately.

• Fewer young entrants: Younger people (ages 16-24) are struggling in this “low-hire/low-fire” job market. Normally, this group would be rushing into the labor market to fill new vacancies, but their labor force participation rate is falling faster than ever.

So why is this bad?

The main problem manifests in the U.S. federal budget. We spend a lot of money on elderly retirement benefits (Social Security, Medicare, and sometimes Medicaid), and those benefits are funded by payroll taxes. When people leave the labor force, the government loses revenue to cover those benefits. A crude estimate says that reducing the workforce by 2.1M people will lower tax revenue by $25-$35 billion. Boomers transitioning from the working world into the retirement benefit world is a double-hit to the budget because the government simultaneously stops collecting revenue from their paychecks and starts paying benefits. Many immigrants pay taxes without receiving any benefits, so losing them also deals a significant blow to our long-term fiscal trajectory.

ChatGDP is currently estimating +2.2% growth for Q2, 2026, with a range of +1.4% to +3.0%. That’s up +0.6% from where it started the quarter.

ChatGDP started Q2 around +1.6%, dipped slightly in early April, then ground sideways for most of May before jumping sharply higher in the last few weeks of the quarter. That late-quarter surge pushed the estimate to its current level. In addition to the numbers, Gemini also produces a daily set of key words to describe major themes in the economy.

How does it compare to other nowcasts? The Atlanta Fed’s GDPNow is currently tracking at +3.0% which is notably more optimistic than ChatGDP’s nowcast; Gemini may be weighing some downside risks that the Atlanta Fed’s model isn’t capturing.

The Prompt

Prompts matter a lot when working with AI. These models are also only trained on data through a specific date, so we have to point Gemini toward live news feeds and data sources to generate a real-time forecasts. In this case, Gemini is looking at Google RSS feeds covering economic news, the stock market, and jobs reports. Here is the full prompt:

You are an expert U.S. macroeconomic analyst and forecaster.

Using the following live economic headlines and indicators, generate an independent, data-driven forecast for U.S. real GDP growth (annualized SAAR) for the current quarter. Base your reasoning on qualitative trends from employment, inflation, ISM reports, consumer sentiment, spending data, and the stock market. Do not reference GDPNow or other official nowcasts.

Be explicit about uncertainty:
- Always provide a central forecast (most likely value).
- Include a plausible range that reflects confidence in your estimate.
- The range should normally be at least ±0.3 percentage points wide, and up to ±1.5 if recent data are mixed or volatile.

Output your response exactly in this format:
Forecast: [number]%  
Range: [low% – high%]  
Key Drivers: [short comma-separated list]  
Commentary: [2–4 sentences summarizing reasoning and risks]

Again, similar aesthetic and color scheme with date of bloom arranged left to right, but Pt. 3 of this experiment essentially collapses ~1200 years of data into two comparable distributions. Pre-1900 and post-1900 are shown as two stacked frequency density panels which allows the “earlier bloom ↔ later bloom” comparison direct and horizontal.

So where Pt. 2 prioritized showing the year-over-year trend as a continuous time series (with day-of-year as a secondary axis), Pt3 sacrifices the year dimension as a continuous axis in favor of making the earlier/later shift in bloom timing the primary, immediately legible takeaway.

What do you think? Pt. 1, Pt. 2, or Pt. 3?

At first glance, most people would probably assume this is just another story about large corporations dominating the economy, but that’s not quite what the data shows. Instead, the labor market increasingly resembles an hourglass economy: strength at the very small end and the very large end, while the middle struggles to grow.

So what’s going on? Here are some basic theories:

• Large firms are pulling away from the middle: Big companies have major advantages in recruiting, technology, financing, and compliance. They can offer better pay, benefits, remote work, and job security, making it increasingly difficult for mid-sized employers to compete for both workers and market share.

• Higher interest rates hit mid-sized firms hardest: Large corporations still have access to bond markets while small businesses can often survive through flexibility and owner labor. Mid-sized firms depend heavily on regional banks and commercial lending, leaving them especially exposed to tighter credit conditions after 2023.

• AI and automation are squeezing the middle market: Large firms can afford enterprise automation while very small businesses often operate in hands-on local services that are difficult to automate. Mid-sized firms are heavily concentrated in administrative, operational, and repeatable knowledge-work roles that AI tools are beginning to replace first.

• The pandemic accelerated consolidation trends: COVID strengthened e-commerce, centralized logistics, and software adoption, allowing large firms to gain even more scale advantages. Many small businesses adapted through contracting and local services, while mid-sized firms were often too complex to pivot quickly but too small to fully reinvent themselves.

• The “threshold problem”: Once firms grow beyond small-business size, costs rise sharply through HR requirements, compliance obligations, management overhead, and benefit expenses. Some business owners may now intentionally avoid scaling into the middle tier by relying more on outsourcing, contractors, and automation.

The general aesthetic and color scheme is the same, but the axes are flipped. As a general rule, a time dimension (i.e. year) should almost always run left to right. However, this dataset is unique in that it’s two time dimensions: year and day of the year.

In the prior chart from last month, the years run left/right which forces the day of the year to orient up/down. Mapping earlier in the year and later in the year to up/down is counterintuitive. Since the intent of this data is to visualize the bloom dates getting earlier in the year, it made sense to reorient the graph to make that earlier/later dynamic more intuitive.

What you think? This version or the prior version?

In almost every instance since the 1940s, +0% year/year employment growth meant a recession had already started.

More specifically, zooming in on the dates when year-on-year employment growth changes from positive to negative, it almost always means a recession started a few months prior.

Does that mean the U.S. is in a recession? Not technically. Despite this tepid employment growth, GDP growth came in at +2.0% for Q1, 2026. Not terrible! However, that growth was heavily dependent on nonresidential investment (i.e. business investment). Private consumption on goods and residential investment both declined, which tells us that businesses are keeping the economy out of recession, not consumers with higher paying new jobs.

In other words, we may have already entered a world where the employment growth ⇄ recession relationship no longer holds. If business investment — driven by the AI buildout and automation boom — becomes the dominant source of GDP growth, strong output and weak job growth could coexist indefinitely. The old recession indicators were built for a labor-driven economy. This one may not be.

It turns out that cherry blossom peak bloom dates offer an elegant way to visualize a changing climate because warmer winters produce earlier blooms. Japanese researcher Yasuyuki Aono spent years meticulously extracting cherry blossom peak bloom dates from over 1,000 years of written records in Kyoto, Japan.

Share

The Economist published a version of this graphic nearly 10 years ago, and the excellent Our World In Data team maintains it as an ongoing visual.

In Washington, DC (where I live), cherry blossoms are an equally big deal, but we don’t have records stretching millennia to observe deeper trends. Still, there is a general trend towards earlier blooms starting from the start of the easily available data.

The DC version of this trend is certainly less convincing (and far less elegant), but both datasets point in the same direction — earlier and earlier blooms caused by warmer and warmer winters.

🌸🌸🌸

I won’t weigh in on broader geopolitical strategy or the general gruesomeness of this situation, but financial markets’ sentiments are clear — markets do not like this. The S&P 500 is down by about 5% since Feb 27th, and the VIX volatility index is up almost 30%.

Share

Not only are stock prices sharply down and volatility is sharply up, but so-called “safe haven” assets (U.S. government bonds) are also oddly flashing warning signs. Market yield on 10-year government bonds increased from just below 4.0% on Feb 27th to 4.4% as of Mar 20th. Gold has also fallen by about 15% since Feb 27th. In other words, assets that typically protect against stock downturns are also in freefall.

Share

How are other nowcasts/forecasts looking for Q4, 2025?

• Gemini (🤖 ChatGDP): ↑+2.3%

• New York Fed Staff Nowcast: ↑+2.1%

• Atlanta Fed GDP Now: ↑+2.3%

• St. Louis Fed Nowcast: ↑+1.7%

• Daniel Bachman’s The Nowcast: ↑+2.8%

• Median: ↑+2.1%

Share

• Personal consumption, which is about 70% of total GDP, grew at +2.4% on an annual basis — solid, but not great compared to previous quarters.

• Private investment, which is about 15% of total GDP, picked up considerably in Q4. Business investment was solid and the residential investment remained weak.

• Government spending fell sharply in Q4 (remember the government shutdown?). State and local government spending grew +2.4%, but that wasn’t enough to offset the -16.6% reduction from the federal side.

• Overall trade added to GDP in Q4 because imports declined more than exports, but both actually shrank. Net trade or the “trade deficit” (exports minus imports) fell to -$950 Bn.

That said, I’m not here to resolve the broader “worth it” debate. What I do find interesting are the underlying labor market trends for young college graduates:

Share

For most of the period where data is available, young college graduates (ages 20-24) experienced unemployment rates within about ±1% of the overall unemployment rate. In other words, a newly minted graduate faced roughly the same risk of unemployment as the broader workforce. Starting around 2008, young graduates actually did better and enjoyed lower unemployment rates than the overall labor market through much of the post–Great Recession recovery.

That pattern began to reverse around 2014. Since then, unemployment among young college graduates never fell below 5%1, while the gap between their unemployment rate and the overall rate has steadily widened. Today, young college graduates face an unemployment rate of roughly 7%, notably higher than the broader workforce.

Is it AI?

Some observers are quick to attribute this trend to AI, but that explanation seems premature. ChatGPT was released in early 2023, yet the trend looks largely unchanged before and after. Long before the AI boom, hiring managers and human resources groups noted that young graduates aren’t prepared for work. Anecdotally, a friend of mine who works in a successful consulting firm recently lamented that all of the young men he hired over the last few years (all from prestigious universities) were “total disasters”. Factor in excessive grade inflation and it’s not surprising that hiring managers are less inclined to take a risk on a young new graduate.2

It’s also worth keeping the scale in mind. Depending on how you count, roughly 250,000 people ages 20–24 with a college degree or higher are currently unemployed. That’s only about 3% of the total unemployed population. The issue isn’t size—it’s what this growing gap may be signaling about how education, hiring, and early-career work are evolving.

Share

• In aggregate, the U.S. is vastly more productive and wealthy than it was in the early 1970s;

• And, huge portions of the population don’t feel like they’ve directly benefited from those gains.

On the other hand, you can find a lot of economic data series that tell a confounding story. Aggregate inflation-adjusted income is higher than ever and have grown faster than productivity (even excluding government transfers).

Share

But, aggregate statistics don’t apply to everyday experience and include confounding variables like population growth. The economic data that most closely tracks the productivity series is Real Disposal Personal Income, per Capita. Let’s break that one down:

• Personal Income: All income received by individuals (wages, dividends, interest, rents, proprietors’ income, transfers, etc.).

• Disposable: After personal taxes.

• Real: Adjusted for inflation (so it reflects purchasing power).

• Per Capita: Divided by population, to control for population growth.

I find this series a compelling comparison because it captures all income sources and subtracts taxes. Thus, it gives a broader view of how much real income households get to use.

Ultimately, comparing a median value (income) to an aggregate value (productivity) is not the best methodology, but most people’s lived experience is closer to the median. Unfortunately, we don’t have anything close to a “median worker productivity” statistic, but maybe we can hack something together in a later post.

So where do Wall Street experts think the market is headed in 2026? The consensus is “higher”, but not as high as recent years. Their median forecast is for the SP500 index to close 2026 at about 7,650 which would be an +11.8% gain. For comparison, a basic linear forecast based on the last 3 years of data expects the SP500 to hit 7,757 (+13.3% gain) by the end of 2026.1

Oppenheimer Asset Management has the most bullish forecast: 8,100 (+18.3% gain). They cite:

• Earnings growth: S&P 500 EPS expected to rise to about $305 in 2026, supporting higher valuations.

• Economic resilience: Solid corporate profits and a resilient U.S. economy.

• Broad market leadership: Gains driven by multiple sectors, not just mega-cap tech.

• Federal Reserve policy: Modest rate cuts expected, supporting equity valuations.

Ned Davis Research, on the other hand, has the most pessimistic prediction among the group: 7,100 (+3.7% gain). They cite:

• Slower earnings growth: Profit growth moderates versus recent years.

• High valuations: Limited room for further multiple expansion.

• Federal Reserve policy: Less aggressive rate cuts than markets expect.

• Market breadth: Narrow leadership increases downside risk.

Where do I stand? It’s really hard to bet against this market and the U.S. economy overall. I do worry about AI-driven “irrational exuberance”, but there are some easy policy levers the government can pull (i.e. reducing tariffs or lowering interest rates). That said, I don’t think the graph will hit +10% in a straight-line — there will be some rough patches along the way.

Share

How are other nowcasts/forecasts looking for Q4, 2025?

• Gemini (🤖 ChatGDP): +2.0%

• New York Fed Staff Nowcast: ↑+2.7%

• Atlanta Fed GDP Now: ↑+5.1%

• St. Louis Fed Nowcast: ↑+1.1%

• Congressional Budget Office: ↑+1.2%

• Daniel Bachman’s The Nowcast: ↑+6.3%

• Median: ↑+2.4%