The Instrument and Its Assumptions
Every measurement instrument encodes a theory of the world. A thermometer assumes that temperature is a stable, continuous property of matter. A seismograph assumes that the ground is normally still. The Cyclically Adjusted Price-to-Earnings ratio — CAPE, developed by Robert Shiller and John Campbell in the 1980s — assumes that corporate earnings, averaged over a decade, are a reliable proxy for the long-run earnings power of the economy. At a CAPE of 40, the market is priced at forty times that smoothed earnings figure, roughly twice its long-run historical average of around 17. The standard interpretation is unambiguous: at such levels, future returns are likely to be low, and the risk of a significant correction is elevated.
That interpretation has been correct, on average, over the past century. It was correct in 1929, approximately correct in 2000, and partially correct in 2007. The question this article examines is not whether CAPE is a useful instrument — it is — but whether the theory of the world embedded in it still holds when the economy is undergoing a structural transformation of the kind that occurs once or twice per century. The answer is not "yes" or "no." It is: under what conditions does each interpretation dominate, and what would we need to observe to know which world we are in?
A Ratio Built for a Stable Economy
To understand what CAPE can and cannot tell us, it helps to be precise about what it measures. The ratio takes the current price of the S&P 500 and divides it by the average of the past ten years of inflation-adjusted earnings. The ten-year average is designed to smooth out the business cycle — to prevent a recession-year earnings trough from making the market look artificially expensive, and a boom-year peak from making it look artificially cheap. The logic is elegant: over a full business cycle, earnings should reflect the underlying productive capacity of the economy, and price should bear a stable relationship to that capacity.
The stability assumption runs deep. It assumes that the relationship between corporate earnings and GDP is roughly constant over time — that the profit share of the economy does not undergo permanent shifts. It assumes that the composition of the index does not change so dramatically that past earnings become a poor guide to future ones. And it assumes that the discount rate — the rate at which investors convert future earnings into present value — does not undergo a structural decline that would justify permanently higher multiples.
Each of these assumptions has been challenged by the past three decades of economic history. The profit share of US GDP has risen from roughly 6 percent in the 1950s and 1960s to approximately 11 to 13 percent today. The composition of the S&P 500 has shifted dramatically toward capital-light, high-margin technology businesses whose earnings dynamics differ fundamentally from the industrial and consumer companies that dominated the index when CAPE's historical baseline was established. And the long-run real interest rate declined substantially through the 2010s, which, all else equal, justifies higher price-to-earnings multiples through basic discounted cash flow arithmetic.
None of this proves that CAPE is wrong. It suggests, rather, that the instrument is measuring something real, but that the interpretation of its readings requires more contextual judgment than a simple comparison to the historical average.
Two Frameworks, One Number
The most clarifying way to approach CAPE at 40 is to recognize that the same number is consistent with two radically different theories of the world, and that the theories make different predictions about what happens next.
Framework A holds that the economy is fundamentally cyclical and mean-reverting. Corporate profit margins are high today because of a combination of low interest rates, globalization, and a favorable regulatory environment — all of which are either reversing or subject to political pressure. The technology sector's dominance reflects a temporary concentration of rents that will be competed away as AI tools become commoditized and antitrust enforcement intensifies. The historical average CAPE of 17 reflects something real about the long-run relationship between capital, labor, and earnings in a competitive economy, and the current deviation from that average will eventually correct, either through falling prices or through a prolonged period of earnings growth that outpaces price appreciation. Under this framework, a CAPE of 40 implies expected real returns of roughly 2 to 3 percent annually over the next decade — low, but not catastrophic unless accompanied by a sharp repricing event.
Framework B holds that the economy is undergoing a structural regime shift of the kind that occurs when a general-purpose technology — steam power, electricity, the internet, and now artificial intelligence — diffuses through the productive base of the economy. In such transitions, the relationship between past earnings and future earnings power breaks down, because the technology raises productivity in ways that are not yet visible in the historical earnings record. The ten-year average in CAPE's denominator includes years of pre-AI earnings that systematically understate the earnings power of an economy in which AI has begun to compound productivity gains. Under this framework, CAPE is not measuring expensive; it is measuring a denominator that has become stale.
The critical observation is that both frameworks are internally coherent. Neither can be dismissed on logical grounds. The question is empirical: which description of the world is more accurate, and what evidence would distinguish between them?
What the Technology Argument Actually Requires
The regime-shift argument is often stated loosely — "AI will change everything" — in a way that obscures the specific and demanding conditions it requires to justify current valuations. It is worth being precise about what Framework B actually needs to be true.
First, it requires that productivity gains from AI be captured disproportionately by capital rather than distributed to workers or consumers through competition. This is not guaranteed. In previous technological transitions, the gains from productivity improvements were eventually competed away in product markets, reducing prices for consumers rather than expanding corporate margins. The question is whether the winner-take-all dynamics of platform businesses and AI model development create a more durable form of market power than previous technologies. There are reasons to think they might — network effects, data advantages, and the capital intensity of training large models all create barriers to entry — but this is a structural bet, not a certainty.
Second, it requires that the productivity gains be large enough and arrive quickly enough to grow earnings into the current valuation. At a CAPE of 40, earnings would need to roughly double, relative to their current level, to bring the ratio back to 20 — a level that would still be above the long-run average but within a range that most analysts would consider defensible. Doubling earnings in real terms over a decade requires sustained earnings growth of approximately 7 percent annually above inflation. That is not impossible — it is roughly what happened during the most productive phases of the 1990s technology expansion — but it requires that the AI productivity story deliver at scale, across the broad economy, within a compressed timeframe.
Third, and most subtly, it requires that the concentration of earnings in a small number of dominant platforms be durable rather than temporary. The six or seven companies that account for a disproportionate share of S&P 500 earnings today are genuinely extraordinary businesses. But the history of technology is also a history of disruption — of dominant platforms that seemed permanent being displaced by the next wave of innovation. If AI itself disrupts the current AI leaders, the earnings concentration that justifies high multiples may prove transient.
| Condition | Required for Framework B | Current Evidence |
|---|---|---|
| Productivity gains captured by capital | High and durable profit margins | Margins elevated; durability uncertain |
| Earnings growth ~7% real annually | Sustained for a decade | Early AI adoption; broad diffusion pending |
| Platform dominance durable | Barriers to entry persist | Strong today; historically impermanent |
| Discount rate remains low | Real rates stay structurally depressed | Rates have risen; trajectory uncertain |
The Denominator Problem
There is a specific technical argument embedded in the current debate that deserves careful examination: the claim that CAPE may overstate expensiveness because its ten-year earnings average includes years in which AI had not yet materially raised productivity, making the denominator a poor estimate of current and future earnings power.
This argument has genuine force, but it cuts in both directions. It is true that if earnings are about to accelerate dramatically, a backward-looking average will make the market look more expensive than it is on a forward basis. But the same logic applies to any period of anticipated earnings acceleration — and the history of financial markets is littered with cases in which anticipated earnings acceleration failed to materialize at the scale and speed that valuations implied. The internet boom of the late 1990s was accompanied by precisely this argument: that traditional valuation metrics were backward-looking and failed to capture the earnings power of a networked economy. The argument was not wrong in principle — the internet did eventually transform corporate earnings — but the timing was off by a decade, and investors who bought at 2000 valuations experienced a lost decade even as the underlying technology thesis proved correct.
The denominator problem is real. But it is a reason to hold CAPE more lightly, not a reason to dismiss it. A more precise formulation would be: CAPE tells us that the market is pricing in a substantial acceleration of earnings growth. Whether that acceleration materializes is a separate question, and one that CAPE, by construction, cannot answer.
Instruments for an Uncertain Transition
If CAPE is neither a reliable warning signal nor a reliable endorsement of current valuations, what instruments are better suited to navigating a period of structural economic transition? This is not a question with clean answers, but several approaches are worth examining.
Forward earnings multiples with scenario weighting. Rather than relying on a single valuation ratio, investors and analysts can construct explicit scenarios for earnings growth under different assumptions about AI diffusion, profit share stability, and competitive dynamics, and weight those scenarios by probability. This approach makes the assumptions visible and debatable, rather than embedding them silently in a historical average. Its limitation is that scenario probabilities are themselves uncertain, and the exercise can become a sophisticated way of confirming prior beliefs.
Profit share monitoring. The most important single variable in the CAPE debate is the profit share of GDP. If AI raises productivity and that productivity is captured by capital, the profit share should rise further — toward 15 to 18 percent of GDP, levels that would justify higher structural multiples. If the profit share stabilizes or declines — because competition intensifies, labor bargaining power recovers, or regulatory intervention redistributes rents — the regime-shift argument weakens. Monitoring corporate profit share as a fraction of GDP, rather than focusing on the price-to-earnings ratio in isolation, provides a more direct test of the underlying thesis.
Concentration risk assessment. The current S&P 500 is unusually concentrated in a small number of companies. The top ten holdings account for a share of the index that has few historical precedents. This concentration means that the index's aggregate valuation is heavily influenced by the valuations of a handful of businesses whose earnings dynamics may not be representative of the broader economy. Disaggregating the index — examining the valuation of the median stock rather than the market-cap-weighted average — reveals a more moderate picture. The median stock in the S&P 500 trades at a CAPE closer to 20 to 25, which is elevated but not extreme. The question of whether the concentration at the top is justified or represents a specific bubble within a broadly reasonable market is analytically distinct from the question of whether the index as a whole is overvalued.
Regime-change indicators. The regime-shift argument implies specific observable predictions: rising profit share, accelerating productivity growth, declining capital intensity in leading sectors, and increasing earnings dispersion between AI-adopting and non-adopting firms. These are measurable. If the data over the next two to three years shows sustained productivity acceleration and rising profit share, the Framework B interpretation gains credibility. If productivity growth remains in its historical range and profit margins compress, Framework A reasserts itself. Treating CAPE as one input in a broader dashboard of structural indicators — rather than as a standalone verdict — is more epistemically honest than either dismissing it or treating it as definitive.
The Bubble That Arrives on Schedule
There is a second-order effect in the technology-valuation relationship that the current debate raises but does not fully develop: the historical pattern by which general-purpose technology waves produce bubbles not despite their transformative potential but partly because of it.
The railroad bubble of the 1840s occurred because railroads were genuinely transformative — they did reshape the economy, reduce transportation costs dramatically, and create enormous wealth. The bubble was not a mistake about the technology; it was a mistake about the timing and distribution of returns. Most of the value created by railroads accrued to shippers and consumers, not to railroad investors, who competed away their returns in a frenzy of overbuilding. The electricity bubble of the 1920s followed a similar pattern: the technology was real, the productivity gains were real, but the financial claims on those gains were mispriced because investors assumed that the companies deploying the technology would capture the value rather than pass it through to customers.
The internet bubble of 2000 is the most recent and most instructive case. The internet did transform the economy. The companies that dominated in 2010 and 2020 — Amazon, Google, Facebook — were barely visible or nonexistent in 2000. The investors who lost money in the bubble were not wrong about the technology; they were wrong about which companies would capture the value, and they were wrong about the timeline. The lesson is not that transformative technology cannot justify high valuations; it is that the relationship between technological transformation and financial returns is mediated by competitive dynamics, regulatory environments, and the distribution of productivity gains — all of which are genuinely uncertain.
The current AI moment shares structural features with each of these historical episodes. The technology is real. The productivity potential is real. The question of who captures the value — and over what timeline — is open. A CAPE of 40 is the market's answer to that question. It is an answer, not a fact.
What the Ratio Cannot Resolve
The deepest limitation of CAPE — and of any valuation instrument — is that it cannot distinguish between a market that is expensive because investors are irrational and a market that is expensive because the future is genuinely better than the past. Both produce the same number. The instrument cannot tell you which world you are in.
What CAPE can tell you is that the margin of safety is thin. At a ratio of 40, the market is pricing in a future that is substantially better than the historical average. If that future materializes, returns will be moderate but positive. If it does not — if the productivity gains are slower, more broadly distributed, or captured by a different set of companies than the current index leaders — returns will be poor. The asymmetry is not symmetric: the upside from being right about the regime shift is modest (the market is already pricing it in), while the downside from being wrong is substantial.
This is not an argument for selling equities or for treating CAPE as a market-timing signal. It is an argument for holding the uncertainty explicitly — for recognizing that the current valuation level represents a specific and demanding bet about the future of the economy, and for sizing that bet in proportion to one's conviction and one's capacity to absorb the downside if the bet is wrong.
The Map and the Territory
Every generation of investors encounters a moment when the standard instruments seem to be measuring the wrong thing. Sometimes they are right — the instruments are calibrated for a world that no longer exists, and the new world genuinely requires new tools. Sometimes they are wrong — the instruments are measuring something real, and the conviction that "this time is different" is the most expensive belief in finance.
The honest position is that we do not yet know which of these is true. The structural arguments for a higher equilibrium CAPE are genuine and not easily dismissed. The historical arguments for mean reversion are also genuine and not easily dismissed. The appropriate response is not to choose a side and defend it with certainty, but to hold both possibilities in view, monitor the indicators that would distinguish between them, and make decisions that are robust to being wrong about which world we are in.
CAPE at 40 is not a verdict. It is a question posed by the market to the economy: can you grow into this price? The economy is in the process of answering. The answer will take years, not months, to become legible. In the meantime, the ratio sits at 40, patient and ambiguous, waiting to be proven right or wrong by a future it cannot see.