The Twelve-Dollar Prophets

Sometime in the spring of 2000, Eduardo Schwartz and Mark Moon sat down with a model that would make them look like fools and then, within eighteen months, like oracles. They were finance professors — Schwartz at UCLA Anderson, Moon at a stage of his career where building novel valuation frameworks for internet companies was either visionary or suicidal, depending on who was reading the working paper. The company they chose to value was Amazon.com, which at the time was either the future of all commerce or a spectacular bonfire of venture capital, and the market had decided, with the serene confidence of a crowd that has never been wrong, that it was worth $76.125 per share.

Schwartz and Moon said $12.42.

They were not short sellers. They were not polemicists. They did not write op-eds or go on CNBC to call Jeff Bezos a fraud. What they did was more dangerous and more precise: they wrote down two equations, made a series of assumptions that were, by the standards of financial modeling, conservative and transparent, fed the whole thing into a Monte Carlo simulation, and let the math run. The math came back and said that in the vast majority of possible futures — thousands of simulated revenue paths branching out over a 25-year horizon — Amazon’s equity was not worth anything close to what the market was paying for it.

The paper was published in 2000.¹ The NASDAQ peaked in March of that year and then began the long sickening slide that would erase $5 trillion in market value. Amazon itself fell from its split-adjusted high above $100 to roughly $6 by late 2001. The model’s $12.42 was not exactly right — the real bottom was lower — but it was so much closer to reality than $76 that the discrepancy between model and market tells you almost everything you need to know about how crowds deceive themselves with their own optimism.

This is a story about that deception. About the specific mechanism by which a market full of intelligent people convinced itself that a company burning cash at an extraordinary rate was worth six times what a careful stochastic analysis could justify. It is also, necessarily, a story about what the model got right and what it revealed about the hidden structure of high-growth equity — a structure that, twenty-five years later, most retail investors still do not understand.

The genius of the Schwartz-Moon framework was not computational sophistication. It was ontological honesty. Where a standard discounted cash flow analysis asks you to forecast revenues over some horizon and then discount them back at some rate — a process that buries enormous uncertainty inside point estimates — Schwartz and Moon asked a different question: what if we admit that we don’t know what Amazon’s revenue growth rate will be, and model that uncertainty explicitly?

They specified two coupled stochastic differential equations. The first described the evolution of Amazon’s sales revenue R:

The first equation says that revenue growth follows a geometric Brownian motion — it has some drift rate μ (the expected growth rate) and some volatility σ(t) that introduces randomness. So far, nothing exotic. The second equation is where the real insight lives. It says that the growth rate itself is not a fixed number but a stochastic variable that mean-reverts toward a long-run average μ̄ at a speed governed by κ. The term η(t) — the volatility of the growth rate — introduces uncertainty about uncertainty. The two Wiener processes dz₁ and dz₂ were assumed to be uncorrelated.

Read that again. They did not just model uncertain revenue. They modeled uncertain uncertainty. The growth rate could be high or low, but more than that, the trajectory of the growth rate — whether it was accelerating, decelerating, or oscillating — was itself random. This is the kind of thing that sounds like academic hairsplitting until you realize that the entire dot-com bubble was an argument about exactly this question: not “will Amazon grow?” but “how fast will Amazon’s growth rate change, and in which direction?”

The same structural question sits behind modern debates about companies like Tesla or Nvidia — not whether growth will occur, but whether markets are pricing the volatility of that growth correctly.

The calibration was straightforward and deliberately unflattering to anyone hoping for a high valuation. Cost of goods sold was set at 75% of sales. Variable expenses at 19%. Fixed expenses at $75 million per quarter. Initial sales of $356 million. Tax rate of 35%. An initial tax loss carry-forward of $559 million. The terminal value was pegged at ten times pretax operating profit at the 25-year horizon. The company was assumed to go bankrupt if cash turned negative. Starting cash position: $906 million.

None of these numbers were aggressive. None were pulled from a pitch deck. They were grounded in Amazon’s actual financials and publicly available data. The market price of risk for revenue was estimated from historical data. The market price of risk for the growth rate was assumed to be zero — a simplification, but one that biased the model in favor of the market price, not against it.

Then they ran Monte Carlo simulations. Thousands of possible revenue paths. In each simulation, the growth rate wandered stochastically, pulled back toward its long-run mean by the κ parameter but buffeted by the η(t) volatility term. For each path, the model computed cash flows, accounted for the possible exercise of convertible bonds and employee stock options, checked whether the company hit the bankruptcy boundary, and discounted whatever remained at the risk-free rate. The resulting distribution of equity values gave the answer: $12.42.

Here is the part that most accounts of the Schwartz-Moon paper miss, and the part that matters most for understanding how the market deceived itself.

The model’s output was extraordinarily sensitive to one parameter: η(t), the volatility of the revenue growth rate. Not σ(t), the volatility of revenue itself — that mattered, but within a manageable range. It was η(t) that swung the valuation wildly. A small increase in the volatility of the growth rate produced a large increase in the estimated share price. A small decrease cratered it.

This is not a quirk of the model. It is a deep structural truth about high-growth companies, and it explains why the dot-com bubble happened with such ferocity and why it collapsed so completely.

Think of it this way. When the growth rate itself is highly volatile — when the market genuinely doesn’t know whether Amazon will grow at 80% next year or 20% or negative 10% — that uncertainty creates optionality. Amazon had the option to keep investing if growth stayed high. It had the option to cut costs and retrench if growth slowed. It had the option to expand into adjacent markets if the core business plateaued. Each of those options has value, and the value increases with uncertainty. This is the fundamental insight of real options theory: a company facing a wider range of possible futures is worth more, all else equal, because it can exercise the favorable outcomes and abandon the unfavorable ones.

The misdirection was in how the market processed this optionality. Real options theory says that higher η(t) increases equity value through the option structure: more uncertainty means more valuable options. But the market in 1999 was not pricing Amazon as an option. It was pricing Amazon as a certainty. The crowd was not saying “Amazon’s growth rate is highly uncertain, which creates valuable optionality.” It was saying “Amazon’s growth rate is going to be enormous, which creates valuable cash flows.” The difference is subtle but fatal.

When you price a company on expected cash flows, you need the growth rate to actually materialize. When you price a company on optionality, you need the growth rate to be uncertain. These are different bets. The first requires being right about the future. The second requires only that the future be unpredictable. Schwartz and Moon’s model was making the second bet. The market was making the first. The market was wrong.

There is a particular kind of collective self-deception that occurs in speculative markets, and it operates through the mechanism of narrative collapse. The story of Amazon in 1999 was not a story about stochastic processes or mean-reverting growth rates. It was a story about the future of retail, the death of brick-and-mortar, the inevitability of e-commerce dominance. And that story was, in the broadest strokes, correct. Amazon did dominate e-commerce. It did transform retail. Bezos was not a fraud. The thesis was right.

But being right about the thesis is not the same as being right about the price. The crowd’s error was not directional but structural. It collapsed a distribution into a point estimate. It took the best-case revenue path out of Schwartz and Moon’s thousands of Monte Carlo simulations — the path where growth stayed high, costs came under control, and the company never hit the bankruptcy boundary — and priced the stock as though that path were the only one.

This is what collective illusion looks like in financial markets. The trick is never that the story is false. The trick is that the story is true enough to make you stop asking how much you should pay for it. The narrative replaces the distribution. The vision replaces the variance. And by the time you notice, you have paid $76 for something worth $12 and the mean-reversion parameter has begun to do its relentless, silent work.

Schwartz and Moon’s Monte Carlo runs contained paths where Amazon became everything its boosters promised. In those paths, the stock was worth far more than $76. But those paths were a minority. In the majority of simulations — the fat middle of the distribution — growth reverted to the mean, margins stayed thin, cash burned faster than revenue scaled, and the equity was worth a fraction of the market price. Some paths hit the bankruptcy boundary and the equity went to zero. The model’s $12.42 was the expected value across all those paths, weighted by probability. The market was pricing only the right tail.

There is a final twist that elevates this from a cautionary tale into something more complex, and it is the reason this case study belongs in the Ledger Domain archive rather than in a simple debunking of dot-com excess.

Schwartz and Moon were right about the price and wrong about the company. Amazon, of course, went on to become one of the most valuable enterprises in human history. The stock that traded at $6 in 2001 would eventually surpass $3,000 before a stock split. The bulls’ thesis — that Amazon would dominate e-commerce and then leverage that dominance into cloud computing, advertising, logistics, and everything else — was not just vindicated but exceeded.

But the model was never making a claim about Amazon’s destiny. It was making a claim about Amazon’s price at a specific moment, given what was knowable at that moment. And on those terms it was right. The market was paying for a certainty that did not yet exist. The model correctly identified the gap between what the crowd believed and what the math supported. That the company eventually grew into its valuation — and beyond — does not retroactively justify the price. It took Amazon nearly a decade to sustainably trade above $76 again. Anyone who bought at the peak and held for that decade would have been underwater for years, and most did not hold.

This is the essential Ledger Domain insight. The trick is rarely about wrong answers. It is about right answers misapplied — correct theses attached to incorrect prices, valid narratives stripped of their probability distributions, visions of the future traded as though the future were already here. Schwartz and Moon did not predict Amazon’s failure. They identified the precise mechanism by which the crowd was lying to itself about what it knew. The math did not say Amazon would fail. The math said the crowd had confused optionality with certainty, and that the price reflected a confidence the data could not support.

Twenty-five years later, the Schwartz-Moon paper is cited more as a curiosity than as methodology. Real options valuation remains a niche practice, largely confined to natural resource and pharmaceutical companies where the option structure is explicit — drill-or-don’t, develop-or-abandon. For technology companies, the industry still runs on DCF models that treat growth rates as point estimates and discount rates as fudge factors for risk that no one can precisely quantify.

This is, to put it mildly, a problem. Every major technology valuation debate of the past decade — Tesla, WeWork, the AI infrastructure buildout — has turned on exactly the question that η(t) captures: not whether growth will happen, but how uncertain the growth trajectory is, and whether the market is pricing that uncertainty correctly or merely pricing the narrative. The model Schwartz and Moon built in 2000 is a template for asking that question rigorously. That it remains underused is itself a kind of willful blindness — a collective agreement to keep the trick invisible.

The next time a company is priced at six times what a careful stochastic analysis can support, remember the two equations. Remember the mean-reverting growth rate. Remember that the crowd’s confidence is not the same as the crowd’s information. And remember that the volatility of the dream — η(t) — is always the question the market least wants to answer honestly.