Efficient Market Hypothesis (EMH) is no longer a theory under debate; it is the framework modern investment is built on. Academia teaches it as a pillar, and the Sveriges Riksbank Prize has rewarded its proponents for decades, so the belief arrives credentialed. Regulators in both Europe and the US then turned that belief into rule, building fiduciary standards where deviating from the benchmark requires justification, making EMH the implied legal default. The result is visible at scale: passive assets crossed nineteen trillion dollars in 2025, over half the US fund market. And the framework polices its own perimeter. S&P’s SPIVA reports exist to prove that active management fails, while the data infrastructure to measure that failure costs at least thirty thousand dollars a seat and scales with AuM from there.
And yet the cracks are visible to anyone not looking away. In 2021 a failing video-game retailer ran from twenty dollars to nearly three hundred and fifty on no news whatsoever, moved there by a crowd on Reddit deciding it should; either the market is not efficient or the word means nothing. In 2025 a single Truth Social post erased two trillion dollars of value in a day, and another, “THIS IS A GREAT TIME TO BUY,” added four trillion back within hours. If prices already held all available information, a man typing in capital letters could not create it by the trillion. And the firms that actually win treat this as the point, not an embarrassment. Jane Street booked a record $39.6B in 2025 and, with Citadel Securities, moves roughly 30% of US equity flow; Renaissance compounded around 66% a year for three decades and made its staff sign permanent NDAs. They do not hire finance professors, and are not even counted as investment firms by the old guard. Everything EMH waves away as friction is exactly what they build their business inside.
The question is not whether EMH is right or wrong, but why it persists amid its faults. The answer is that EMH is not a theory but a paradigm: a framework embedded so deeply in institutional practice that counterevidence is filed as anomaly, and, more to the point, backed by enough power that disagreement carries a cost. What follows traces how that paradigm was assembled, who sustains it, and what it means that the people who actually move markets no longer believe it.
Efficient Retail Market Hypothesis
New York, November 1907. John Pierpont Morgan locks one hundred and twenty bankers in his private library and does not open the door until each has signed onto a rescue package. No regulator, no benchmark, no price signal: one man’s credibility was the settlement mechanism for the American banking system. The 19th century had run this way throughout. Two decades earlier, Bank of England Governor William Lidderdale had assembled a private consortium of the Rothschilds, Antony Gibbs, and most of the City to bail out Baring Brothers, all before the public knew a crisis had occurred. Victorian finance was a craft managed by guilds: you lent to a man because you knew his family, his school, his habits, and you vouched on his character. The theory to do otherwise already existed and went unused; in 1900 Bachelier modeled asset prices as random mathematics in his thesis, then died in obscurity, because a guild that trusts faces has no use for mathematics that trusts strangers.
The public came anyway, but on no firm footing. NYSE shareholders went from 1.5M in 1900 to 10M by 1929, and the Liberty Bond campaigns of 1917-18 pulled 20M Americans into securities markets for the first time, building the broker networks the 1920s bull market then exploited (Ott, 2011). This was mass entry without anything to make it rational: a public recruited by salesmanship and patriotism, holding stock on no better basis than the surrounding mania. The crash of 1929 is what that costs. The guild could vouch for a man it knew; it had no way to vouch for the market itself, to a public that would never know anyone inside it. That was the missing instrument, and the next forty years were spent building it.
The 1929 crash had discredited judgment itself, and the search for something to replace it ran for forty years through people who were not working together. Alfred Cowles III, a forecaster whose own service the crash had proven worthless, funded a commission in 1932 to test scientifically whether forecasting worked at all; it did not. Housed later in Chicago, the commission sheltered the foundational work of Arrow, Debreu, and Markowitz, whose 1952 portfolio theory, made computable by Dantzig’s linear programming at RAND, turned “own the diversified whole” from a slogan into a method. The Ford Foundation spent $46.3M rebuilding American business schools around quantitative economics, in part because Soviet central planning had to be answered with a social science that looked equally rigorous. Merrill Lynch, which sold equities to the public for a living, funded the CRSP database in 1960 to find out what stocks had actually returned; Fisher and Lorie answered in 1964, roughly 10% a year since 1926. Fama added the last piece in 1965, showing prices move randomly, and named it the Efficient Market Hypothesis in 1970. None of them set out to write a single sentence, but read together that is what they produced, addressed to a stranger with savings: you cannot be out-traded by a club with private information, you need to know no one, the market pays about 10%, so own it and hold. For the first time it was rational to invest in a market full of strangers, and the public came back to stay.
The Truth and the Index
Benchmarks are not facts of nature; they are conventions that hardened into infrastructure. Cap-weighting was not chosen for elegance. In 1957, when S&P expanded to 500 stocks, it was simply the easiest arithmetic to do with pen and paper. That accident embedded a permanent large-cap tilt and a self-reinforcing momentum loop: the biggest stocks draw the most inflows, their prices rise, their weight grows, more inflows follow. Better-designed alternatives exist, but none scale as cheaply, and cheapness, not correctness, decided the winner.
That same scalability rigs the contest built on top of it, and it does so in both directions at once. Active management’s costs are not just higher, they are inflated by regulation: benchmark license fees, pre-trade compliance engine, audit trails to prove prudent decisions… Passive escapes all of it, because tracking the index is the compliance answer; there is no discretionary tax to collect. Meanwhile the passive side carries a cost that no one counts: a fund that rebalances on predictable schedule telegraphs its intentions, and others trade ahead of it, so the loss shows up buried in tracking error rather than as a line-item fee. And beneath both, the benchmark defines the field of play itself. Short positions, CTAs, trend-following; every technique that does not fit a long-only cap-weighted vehicle is excluded from the comparison outright. Skill beyond long-only vanillas is treated as an illegal move, while infinite mechanical scalability is treated as the natural baseline. Who, then, is surprised that the selected active managers underperform?
Behind this scoreboard sits what Braun (2022) called asset manager capitalism. The Big Three (BlackRock, Vanguard, State Street) own more than 20% of the average S&P 500 constituent and cannot exit, because their portfolio is the market. Power shifted from exit to control, but the control is barely used, and that abdication is the exercise of it. Of roughly 4,000 shareholder proposals at Russell 3000 firms between 2008 and 2017, none came from the Big Three, and in North America they vote with management about 84% of the time. Holding a fifth of corporate America and declining to steer it is not neutrality; it is a standing vote for the status quo at a scale no one else can match. The wheel goes untouched because of how the money is made. In 2020 BlackRock drew about $29B from rising asset prices against only $5B from net inflows. The fee is a percentage of assets, so it harvests whatever the market hands over, regardless of whether capital was allocated well. The incentive is to gather assets, not to allocate them.
Which closes the loop, and is where the notion of the sovereign earns its place. The benchmark industry is itself an oligopoly: five providers (S&P Dow Jones Indices, CRSP, FTSE Russell, MSCI, NASDAQ) account for 95% of passive ETF licenses, and most active managers are required by law or by their investors to pay those fees and measure themselves against those indices. Because several providers are public, the circle joins: the largest license buyers are the largest passive providers, which are also among the largest owners of the licensing firms. A system that licenses itself, owns itself, and judges itself isn’t a market, it’s a sovereign.
Bureaucrats in Lab Coats
EMH presents itself as a description of markets. It is an instrument of regulatory power, and the disguise fails wherever it acts, because measuring the world and ruling it are different things. Each time it claims to measure something, it can be caught doing something a measurement cannot do: it sues, it overrides, it excludes.
The US Department of Labor built the legal meaning of fiduciary prudence on portfolio theory, and prudence so defined became measurable only against a benchmark. The law never names the index; it does something more effective and makes deviation actionable. A trustee who departs and underperforms can be sued, so tracking the benchmark became the only defensible move, not because anyone proved it correct but because disagreement became a liability. A description of reality has no plaintiffs. That you can be sued for departing from the benchmark is proof it stopped measuring markets and started governing them.
On an international level, Basel was the science meant to replace Morgan’s library: each version promising it had finally contained the danger, with AT1 bonds engineered as loss absorbers. Then the one crisis that tested it. When Credit Suisse collapsed in 2023, FINMA wrote $17B of AT1 bonds to zero while shareholders kept $3B, inverting the loss hierarchy the framework was built to guarantee; in 2025 a Swiss court overturned the decision, finding the regulator had acted with no legal authority. The rules-based framework dissolved, exactly when it was needed, into one regulator’s decree that a court later ruled had no ground. Underneath the technique was raw discretion.
Finally on a geopolitical level, a real measure of the market would be indifferent to politics, which is not the case at all. Iran has been outside benchmarks for decades; Russia was deleted from MSCI’s indices within days of the 2022 sanctions. If the index boundary is drawn upstream by state power, it cannot be a neutral measurement. The world’s market does not have a sanctions list. “Investable universe” means the part of the world American power permits you to own, and MSCI transmits that perfectly, precisely because it has no politics of its own to resist with. This is why pension funds in Canada, Europe, and Japan are herded into US equities to hit their targets, buying what is sold as a world portfolio but is mostly US.
Mandatory and Irrelevant
EMH was challenged for half a century and never changed, because every refutation got absorbed instead of landing. Black, Jensen and Scholes flattened the security market line in 1972; it became the “beta anomaly.” Roll proved in 1977 that CAPM cannot be tested at all, the market portfolio being unobservable; it became a debate about proxies. Grossman and Stiglitz showed in 1980 that perfect efficiency is logically impossible, since if prices already reflect all information no one is paid to gather it; it became “almost” efficient. Jegadeesh and Titman documented momentum in 1993; Fama eventually folded it in as a fourth factor. Each challenge was written in the language EMH built, which is exactly what converted it into a contribution rather than an exit. A paradigm does not die from anomalies piling up. It dies when a successor is built alongside it, and that is what has been happening for decades, on two levels.
The first is intellectual: practitioners stopped arguing and drew their own maps. Soros pitched reflexivity as a replacement outright: markets do not seek equilibrium but overshoot, because perception moves prices and prices move perception, and he made a billion dollars breaking the Bank of England with it rather than publishing. Falkenstein argued that when investors chase relative wealth, risk becomes deviation from the consensus and goes unpriced, a structural attack on CAPM; when he submitted the proof, the journal told him he was going crazy. The framework ignored the first and expelled the second, which is what paradigms do to evidence they cannot digest.
The second level is a business, not a theory. EMH’s method is to assume friction away: no latency, free information, costless execution. An entire class of firms lives inside exactly that discarded friction. Jane Street booked a record $39.6B in net trading revenue in 2025 and, with Citadel Securities, moves roughly 30% of US equity flow; Renaissance’s Medallion compounded around 66% gross a year for three decades and made its people sign permanent NDAs, so the cleanest refutation of efficiency exists as a trade secret rather than a paper. These are technology firms, low-latency stacks and physicists and multi-year strategic bets on arcane technologies such as OCaml. Everything EMH abstracts out as noise is what they implement and sell.
And the two halves meet here. A pension fiduciary cannot make those bets, not because they are unsound but because the prudence regime built around the benchmark makes deviation a liability; tracking is the legally safe move. Jane Street faces no such regime, and is not even counted as an investment firm. So EMH does not merely lose to the people who route around it. It forbids its own captives from joining them. The theory survives not because it describes the market — the people who actually move it discarded it long ago — but because it stays mandatory for the captured and irrelevant to the free. That is not a living theory. It is load-bearing, which is harder to remove.