Bubbles Are Stories, Not Numbers: What Shiller Taught Me

011. A Book That Predicted the Crash

In March 2000, the Nasdaq Composite hit 5,000, the peak of the dot-com bubble. That same month, Robert Shiller published Irrational Exuberance.

Within weeks, the Nasdaq began to collapse. In two years, it fell from 5,000 to 1,100, evaporating 78%.

Shiller wasn't a prophet. He didn't see "tomorrow will crash." He saw that "this pricing system had detached from fundamentals and was held up only by stories."

What makes this book extraordinary is that it turns the vague word "bubble" into something you can dissect. Bubbles are not valuation phenomena — they are narrative phenomena.

Once you understand this, the way you look at markets is permanently changed.

022. The Shiller Model: Inventing CAPE

Shiller's most famous tool is CAPE (Cyclically Adjusted P/E) — using the average of the past 10 years of inflation-adjusted earnings as the denominator to calculate the price-to-earnings ratio.

Why 10 years? Because a single year's earnings are noise — economic cycles, industry cycles, currency cycles all distort that year's profit. A 10-year average reflects "normalized" earning power.

After inventing this tool, a stunning backtest result emerged — historically, when the S&P 500's CAPE is above 25, future 10-year real returns are terrible; when CAPE is below 15, future 10-year real returns are excellent.

Not 100% accurate. But the correlation is too significant to ignore.

A few historical highs:

• March 2000: S&P CAPE all-time high of 44. Real return over the next 13 years: near zero.
• October 2007: S&P CAPE of 27. Real return over the next 5 years: -2%/year.
• November 2021: S&P CAPE of 38. Real loss of 25% over the next 12 months.
• Early 2026: S&P CAPE around 35–37.

Valuation can't tell you when it will crash, but it can tell you "the downside space has been opened."

This is the most practical line Shiller left me. My own practice: check CAPE once a month, use it as one needle on my macro compass — not to time the market, but to adjust my portfolio structure.

033. The Real Insight: Bubbles Are Narrative Phenomena

CAPE is the book's marquee feature, but Shiller later fully unpacked what this book only hinted at in Narrative Economics — valuation is not the cause of bubbles; narrative is.

Every bubble comes with a "why this time is different" story.

In the late 1990s, the story was "the internet changes everything; traditional valuation methods don't apply." In 2006–2007, it was "home prices can never fall because the US has never had a nationwide housing crash." In 2020–2021, it was "zero rates are here to stay; tech stocks deserve a permanent valuation premium."

These narratives share a common structure:

1. A genuine core — the internet really does change everything; home prices really do rise over the long term; tech stocks really do command a premium.
2. An over-extrapolation — so traditional valuation no longer applies / home prices can never fall / the valuation premium is permanent.
3. A contagion mechanism — repeated by friends, media, social networks until everyone believes.

When these three elements align, a bubble begins.

The genuine core makes it hard to refute; the over-extrapolation makes people unwilling to refute; the contagion mechanism makes refutation heresy — that's the complete engineering of a narrative bubble.

The biggest takeaway from reading Shiller: whenever I look at a market, I now ask myself: What is the dominant narrative right now? Where is the over-extrapolation in this narrative?

If you can answer these two questions clearly, you have one more needle than the vast majority of traders.

044. The Narrative Most Worth Watching in 2026

I reread this book in 2026 — the third year of the AI wave.

The narrative around AI is extremely powerful: this time it's a true industrial revolution, a productivity leap, a need to rewrite valuation methods. The genuine core of these narratives certainly exists — AI is changing productivity, reshaping many industries.

But I'm now alert to the over-extrapolation part:

• "Top compute vendors can sustain 30%+ revenue growth forever" — likely wrong, because hyperscale customers are building their own chips (Google TPU, Meta MTIA, Amazon Trainium), which will erode Nvidia's share over time;
• "The application layer will see OpenAI-level winner-take-all dynamics" — likely wrong, because the open-source ecosystem (Llama, DeepSeek, Mistral) is rapidly catching up;
• "The Mag 7's valuation premium is justified because earnings will grow along an exponential curve" — likely wrong, because compute costs are also growing exponentially, and gross margins may not sustain expansion;
• "The AI capex cycle is unlike any previous cycle; there won't be overcapacity" — every infrastructure narrative in the past 100 years has used this "won't happen" argument, and in hindsight it never held.

I'm not saying these narratives are definitely wrong. I'm saying when a narrative is dominant and no counter-argument can enter mainstream discussion, it's worth being alert.

What Shiller taught me is not to judge right or wrong, but to judge "are we at the peak phase of the narrative?"

A simple way to tell: when every financial media outlet, every investment bank report, and every podcast host is using the same language to explain the market — that's the peak phase of the narrative. In 2026, the AI narrative is exactly in that state.

055. Where I Differ from Shiller

By my fifth read, I started to have reservations about some of Shiller's core arguments.

First, He Reads CAPE Too Conclusively. Shiller repeatedly uses CAPE to make long-term return forecasts — "real annualized returns of around 2% over the next 10 years" — which he has been saying since 2014. But in hindsight, the S&P 500's real annualized return from 2014 to 2024 was roughly 9% — 7 percentage points per year above CAPE's prediction. Over a decade, that's a 2x gap.

CAPE isn't wrong, but its predictive accuracy is overstated by Shiller. The reality: CAPE has stronger predictive power at extreme highs and extreme lows, but almost no predictive power in the middle range. Shiller himself admits this, but in public he still tends to give definitive forecasts. This "semi-honesty" is a built-in problem of the book.

Second, He Underestimates the Impact of the Interest Rate Regime on CAPE. The CAPE model implicitly assumes a stable risk-free rate. But in reality, the level of interest rates directly affects what a reasonable P/E should be — when rates are low, P/E should be high; when rates are high, P/E should be low.

From 2009 to 2021, the US 10-year Treasury yield was consistently below 2%, a period when the reasonable CAPE should have been 5–10 points higher than the historical average. Shiller's method made no adjustment for this, causing him to predict "the market is about to crash" from 2014 onward — it then rallied for another 7 years before finally pulling back.

Using a static historical average to forecast a market undergoing a structural shift in interest rates — this is the biggest blind spot in Shiller's approach.

Third, He Barely Studies the Narrative Reconstruction After a Bubble. All of Shiller's research focuses on "how bubbles form" and "how they pop." But the real market cycle has four phases: gestation → mania → crash → reconstruction. Shiller's tools work for the first three, but are almost silent on the fourth.

This blind spot was glaring in 2009–2013 — after the great crash, the market bottom worked through repeated reconstruction, but Shiller kept warning "it will fall again" during that period. In reality, that decade became one of the strongest ever for US stocks. He's good at seeing bubbles, but not good at seeing recoveries.

Fourth, He Lacks an Analysis of "Institutional Narratives." The narratives Shiller studied come mainly from media, retail investors, and street conversations. But after the 2010s, the dominant narrative-makers are institutions — investment bank research reports, hedge fund 13F filings, Fed officials' speeches.

The contagion mechanism of institutional narratives is completely different from retail narratives — it flows through research reports → sell-side calls → client portfolio rebalancing → ETF flows, completing contagion within hours. The model Shiller describes — "narratives slowly spreading among friends" — has little to do with the institutional narratives of 2026.

066. Shiller vs Soros: Two Rival Theories of Narrative

Shiller's "narrative economics" and Soros's "reflexivity theory" look very similar, but their cores are fundamentally different.

Shiller sees narratives as exogenous — a story is born in the media, then spreads, then affects prices. Narrative is the cause; price is the effect.

Soros sees narratives and prices as mutually reinforcing — price increases create narratives, which further push prices, until the reflexive loop breaks. Narratives and prices are co-causal.

These two theories lead to completely different methods for judging "when the narrative will reverse."

Shiller would look at "the degree of over-extrapolation" — how many people are repeating the same line. Soros would look at "the distance between price gains and fundamental improvement" — how many years of fundamentals the price has front-run.

My own approach is to use Shiller's method to judge "is the narrative at a peak?" and Soros's method to judge "is the turning point near?" The two methods answer different questions — Shiller answers "is it expensive?" and Soros answers "when will it change?"

077. Why This Book Is More Worth Reading in 2026 Than in 2000

When Shiller wrote this book in 2000, he faced a relatively simple market: a clear tech bubble, with clear valuation excesses, and a clear narrative (.com changes everything).

The market in 2026 is far more complex — the Mag 7's valuations are high, but their earnings are real; AI capex is enormous, but its conversion path is unclear; CAPE is in the 35–37 range, but the interest rate environment is different.

In this "hard to tell genuine from fake" state, Shiller's tools are needed even more. When judgment becomes difficult, narrative analysis is more valuable than valuation analysis. Because valuation is already pushed to extremes — any 1 P/E multiple difference could be a difference in model assumptions; but whether a narrative is at a peak is relatively objective — when everyone is telling the same story, it's a peak.

At the beginning of 2026, I made a "market narrative inventory," listing the 10 dominant narratives driving the market today, tagging each with "genuine core vs over-extrapolation." That inventory was more sobering than any DCF.

088. Final Thoughts

Many people read Irrational Exuberance expecting "how to predict bubbles." They finish disappointed — Shiller doesn't give a step-by-step timing system.

But that's a good thing. Because any step-by-step timing system that works is probably wrong.

What Shiller gives is a way of thinking:

1. Valuation can't precisely time, but it can tell you where the room is;
2. Narrative comes before valuation — those who understand narrative have one more step ahead;
3. When everyone believes "this time is different," odds are "this time is the same."

These three things hold in any market, any era.

Shiller once said something I've copied down many times: "People don't make decisions based on facts, but on the stories they believe."

Understand that, and the way you look at markets becomes slower, but also steadier.

That's the mark of a great book.

But remember — Shiller himself was repeatedly wrong in his public forecasts from 2014 to 2021. A good tool doesn't guarantee a good forecast. The book teaches not conclusions but a way of seeing. Conclusions must always be walked through again, in your own era, with your own judgment.