We Asked GMO’s Head of Asset Allocation Why This Bubble is Easy — But Investors Will Get it Wrong

Finance-focused summary

Bubble framework: “easy” vs “hard” bubbles

• Investment bubble definition (GMO view): an important asset is priced to feel “close to unownable.”
• Easy bubble: you can avoid most downside while still taking a normal amount of risk by rotating into less-overvalued risk assets.
  • Example: Internet bubble (late 1990s/2000)
    • S&P 500 described as the most expensive in history (on typical valuation metrics).
    • Navigation was feasible because investors could own some risk assets without buying the most overvalued ones.
    • Suggested “still-own-risk” alternatives:
      • small caps
      • REITs
      • emerging equity
      • emerging debt
• Hard bubble: avoiding downside requires giving up risk assets broadly.
  • GFC bubble (2007–2008): described as overpricing across “all risk assets everywhere”, so diversifying away was difficult.
• Even harder bubble (end of 2021):
  • Stocks and bonds both overvalued.
  • Key mechanism: avoiding losses required avoiding duration (i.e., moving out of interest-rate-sensitive assets), potentially toward cash/cash-like.
  • Practical issue: clients don’t like holding cash.

AI bubble thesis (current market context)

• GMO frames the AI boom as a bubble in US stocks “writ large”:
  • US large caps: part of the bubble.
  • US small caps: “still look pretty overvalued” as well.
  • Non-US markets: better opportunities because they are “priced to deliver a decent return.”
• Main portfolio implication: it should be possible to build US-and-non-US stock + bond portfolios that avoid most bubble peril without an “insane” portfolio.

---

Risk-reward valuation method (how GMO interprets charts)

They describe risk-reward scatterplots built from client forecasts at past points (not backtested data-mining).

Core logic

• Horizontal axis: increasing risk assets (low-risk near origin outward).
• Vertical axis: higher expected return.
• The slope of the regression line indicates how much you’re paid for risk.
  • Rational market expectation: positive slope
  • Normal slope expectation: around ~0.7
• Historical episode interpretations:
  • 2000: slope described as ~0.4 (still positive → paid for risk, just less than normal)
  • 2007: “turned on its head”—investors were paying for the privilege of taking risk (negative/very unfavorable pricing; described as a “truly global bubble”)
  • 2021: slope positive but “below zero” expected returns in real terms (“everything… negative expected return in real terms”)
    • Emphasis: avoiding duration is key because repricing of long-duration assets can be devastating.
  • Later transition:
    • US equities overvalued; rest of world initially looked fine
    • After non-US risk assets did well, the slope fell:
      • 0.4 → 0.1 now (or still ~0.4 if excluding US equities)

Duration caution

Avoid assets with large valuation repricing risk due to long duration, such as:

• stocks
• long-duration bonds
• real estate
• infrastructure
• much of private assets

---

Earnings bubble risk (vs “price bubble”)

• GMO’s concern: today may be an earnings bubble, not just a valuation bubble.
• Analogies / comparisons:
  • 2000: “mostly evaluation bubble,” but some earnings bubble elements.
  • Europe (mid/late 2000s): earnings “went up 100% over 4 years,” then later struggled to return to those levels on an index basis.
  • EM (~2012): similar pattern.
• Mechanism: earnings can look elevated by ramping investment ahead of depreciation.
  • Microsoft example: spending on $200B data centers; builds take ~3 years; depreciation begins later → earnings boosted mechanically while depreciation lags.
• Semiconductor/capital cycle caution:
  • AI demand can mask cyclical fundamentals.
  • They warn memory/cycle assets may look cheap later:
    • SK Hynix, Micron
    • Even if trailing P/E looks low, they can be lousy investments if the capital cycle turns.

---

AI CapEx and capital cycle: what can go wrong

• Scale context (data centers):
  • Forecast: ~$700B data center spending
  • Framed as ~2.2% of US GDP
  • Compared to past infrastructure buildouts (fiber in the late 1990s; electricity and railroads historically).
• Transformational tech thesis: world-changing tech can be real, but profits don’t accrue to builders automatically.
  • Early strong ROI often draws in capital, then destroys ROI via overcapacity.
• Financing risk (Ponzi/circular financing risk):
  • Bubble-like deals may increasingly use sketchy financing structures.
  • Examples:
    • OpenAI → AMD GPUs
      • OpenAI receiving warrants on AMD stock, described as about worth half the deal value.
      • Framed as improving presentation/economics (“looks more profitable”).
    • Anthropic → leasing $36B of Alphabet TPUs
      • Broadcom offers buyback in a default scenario to improve rating/terms.
• Balance sheet / leverage note:
  • Hyperscalers reportedly doubled debt ratios in the last 9 months (GMO claim).
  • GMO does not predict defaults, but highlights rapid leverage build.

---

IPO/supply overhang: market issuance as a tailwind/headwind

• GMO expects more supply into US stocks over the next ~12 months than in living memory.
• Amount: ~5–6% of aggregate US market cap may arrive as “supply” via:
  • IPOs
  • share sales
  • lockups expiring
• Timing dynamics: supply doesn’t hit immediately at IPO date; it ramps as lockups expire and shares trade.
• Rule of thumb cited: 1% increase in supply → ~7.5% worse return over the subsequent year (linear assumption noted; not guaranteed).
• Link to “bubbly behavior”:
  • 2000 and 2021 were the two highest supply points in the last 50 years—both also frothy times.
• Flow sensitivity note: even if demand/supply is “passive” (e.g., 401(k) flows), price impact depends on:
  • who holds which stocks
  • price sensitivity (growth owners less price-sensitive than value owners)

---

Asset-class return forecasting (GMO model ingredients + key numbers)

Framework / methodology (step-like description)

• Use a 7-year forecast horizon (average time for convergence back toward fair value).
• Build expected returns from:
  • income
  • growth
  • a valuation shift term to revert to fair value
    • valuation shift assumed to revert ~1/7 each year

“Fair value” anchored to interest-rate environment

• GMO argues fair value is easier to define relative to alternatives than absolutely.
• Anchor from cash, then add required premia:
  • Bond term premium: assumed ~100 bps (bond vs cash)
  • Stocks vs high-quality bonds: assumed ~4.5% (stocks ~4.5 percentage points more than cash; ~3.5% relative to high-quality bonds as stated)

Interest-rate scenario targets

• Low interest rate environment (GMO best guess):
  • Real return to cash ~0%
  • Stocks need to deliver ~4.5% better than cash
  • Implies normalized P/E ~21x
• If cash return is higher by 1 to 1.5%:
  • normalized equity fair value falls to ~16x

Conviction themes implied by the chart discussion

• Non-US equities vs US:
  • GMO highlights a “big gap” between US and rest of world.
  • US traded at a premium vs non-US only in recent ~15 years (no such history as of 2010).
  • US dollar described as overvalued vs 2012 → potential tailwind for non-US currency exposure for US investors.
• Value and small caps:
  • “Small looks cheaper than large pretty much everywhere.”
  • Value described as quite cheap to extraordinarily cheap.

Path dependency / drawdown

• Expected returns aren’t a straight line.
• Example: 2022
  • stock market down 17% nominal
  • in real terms fell “by a lot more” due to hot inflation
  • valuations can become cheaper quickly after sharp bear markets

---

“Benchmark-free” portfolio concept

• GMO contrasts two types of client complaints:
  • Tracking error complaint (benchmark-relative underperformance)
  • Absolute-return complaint (wasting portfolio in assets with negative expected real return)
• Benchmark-free objective:
  • Similar risk to a 60/40 portfolio, but without “wasting” allocation in assets held only due to fear they might do well.
  • You won’t feel obligated to own:
    • US stocks if not attractively priced
    • bonds if not attractively priced
• Implementation details:
  • Forecasts are value-driven and move more slowly.
  • Some assets can’t be forecast with 7-year return models:
    • example: merger arbitrage (resolves in 3–12 months)
  • Still ask: “are you getting paid adequately for blow-up risk?”

---

Private equity research takeaway (bias toward small-cap “junk”)

• Research scope mentioned: analysis of ~700+ leveraged buyouts from 1981 onward.
• Key findings:
  • LBO universe skews heavily small/mid-cap (mega-cap “trillion-dollar” firms not typically in scope; RJR Nabisco singled out as mega-cap-ish at the time).
  • Over decades, large-cap profitability rose; small-cap profitability trended less favorably.
  • LBO targets averaged less profitable and higher debt (“kind of junk”) before going private.
• Allocator implication:
  • Many endowments/foundations may have ~half their equity exposure via private vehicles → embedded bet toward:
    • small
    • junky/low-quality
  • GMO argues junky companies have never outperformed (though they can be volatile/high beta and sometimes outperform).
• Suggested mitigation:
  • Public offset example: long S&P 500 / S&P 100, short Russell 2000 to balance size/quality biases (may tie up capital; may not have positive expected returns by itself).
  • Prefer a quality tilt:
    • bias toward quality is argued to do better.
    • For expensive junk, GMO suggests private might be preferable to public (public expensive junk is “a disaster”).

---

“Why you’re getting paid” (risk premium & avoiding mismatched payoffs)

• GMO says understanding why an investment pays is crucial to avoid stupid mistakes.
• Skepticism examples:
  • Tail-risk hedging marketed as “cash-like return”
    • If crash protection comes with cash-like returns, the counterparty must absorb horribly nasty correlated return—unlikely unless you’re being paid appropriately.
  • Buying call options as “limited downside, unlimited upside”
    • Someone sells the calls; long-call returns may disappoint over the long run.
• Diagnostic:
  • If a thesis says a group underperforms, you should see:
    • valuation/pricing consistent with low future fundamentals
    • and whether pricing remains unfavorable at each point in time
  • Overvalued value stocks are singled out as particularly unhelpful.

---

Disclosures / disclaimers noted

• “No information on this podcast should be construed as investment advice.”
• Also noted: securities discussed may be holdings of hosts’ firms or clients.

---

Tickers / companies / instruments mentioned

Indices / portfolios (implicit instruments)

• S&P 500
• S&P 100
• Russell 2000
• 60/40 portfolio

Equities / organizations

• Tesla
• Microsoft
• Palantir
• SpaceX
• OpenAI
• Anthropic
• Alphabet
• AMD
• Broadcom
• SK Hynix
• Micron
• RJR Nabisco
• GM (mentioned in context of relative supply/impact)
• Nvidia

Assets / sectors / instruments

• REITs
• emerging equity
• emerging debt
• private equity
• LBOs
• merger arbitrage
• data centers / AI CapEx
• warrants
• TPUs
• high-quality bonds
• duration
• cash/cash-like
• infrastructure
• real estate

Macro / currency

• US dollar

---

Key numbers explicitly cited

• Europe earnings bubble: ~100% increase over 4 years (2007–2008 era described)
• Microsoft / data centers: $200B data center spending example
• Data center spending forecast: ~$700B (~2.2% of US GDP)
• Anthropic deal example: $36B worth of Alphabet TPUs
• Supply overhang:
  • ~5–6% of aggregate US market cap over ~12 months
  • 1% supply → ~7.5% worse subsequent-year return (rule of thumb)
• Forecast mechanics:
  • 7-year horizon
  • valuation shift reverts 1/7 per year
• Valuation anchors:
  • term premium: ~100 bps
  • stock excess return over cash: ~4.5% (and ~3.5% vs high-quality bonds)
  • normalized P/E fair value in low-rate case: ~21x
  • higher cash-return case: normalized P/E ~16x (cash higher by ~1 to 1.5%)
• Risk-reward slope examples:
  • normal: ~0.7
  • 2000: ~0.4
  • 2021: slope transitions noted as 0.4 then → 0.1 now (excluding US equities: ~0.4)
• Market drawdown example:
  • 2022: stock market -17% nominal

---

Presenters / sources

• Ben (GMO): described as GMO’s Head of Asset Allocation (surname not provided in subtitles)
• Jeremy Grantham: referenced (including a quote attributed to him about “every single risk asset”)
• Hosts/sources named in subtitles:
  • “Ben”
  • “Jeremy” (Grantham)