Market Efficiency

Over the last eleven posts we’ve built a complete toolkit. We can price bonds and stocks, measure risk, construct diversified portfolios, estimate factor exposures, compute alpha, and pick the best available tradeoff between risk and return. Every one of those tools rests on the same assumption: that we can examine a security, judge it mispriced, and trade the gap at a profit.

But the gap may not be there. The market may have already incorporated every piece of analysis available, every earnings forecast, every chart pattern. That possibility is the subject of the efficient market hypothesis (EMH).

The Efficient Market Hypothesis

Eugene Fama formalized the idea in 1970: asset prices reflect available information. Not perfectly, not instantly, but well enough that nobody can systematically exploit the gaps after accounting for the costs of trying.

Fama defined three forms, each stronger than the last.

The weak form holds that prices reflect all past trading data: historical prices, volume, returns. If weak-form efficiency holds, technical analysis does not work. Charting patterns, moving-average crossovers, and support and resistance levels are all reading tea leaves. Past prices tell us nothing about future prices that is not already baked in.

The semi-strong form extends the claim to all publicly available information: financial statements, earnings announcements, macroeconomic data, analyst reports. If semi-strong efficiency holds, fundamental analysis does not work either. Reading 10-Ks more carefully than everyone else buys no advantage, because the information in those filings is already priced.

The strong form covers public and private information, including what corporate insiders know. This is the most extreme version, and under it even a CEO trading on knowledge of an unannounced merger cannot profit. Almost nobody believes it. Insider trading laws exist because insiders can profit from private information. The strong form is useful anyway, as a theoretical boundary.

Each form nests inside the next. Semi-strong implies weak, since public information includes past prices, and strong implies semi-strong. Most of the empirical debate is about the semi-strong form: whether publicly available information offers an edge or the market has already done the homework.

The Evidence

What Supports EMH

Active managers mostly underperform, and the SPIVA scorecard tracks the shortfall annually. Over 15-year windows, roughly 85-90% of actively managed U.S. large-cap funds underperform the S&P 500 after fees. The numbers are similar for international and small-cap funds. A few managers beat the benchmark in any year, but persistence is weak. Last year’s winners are not reliably next year’s winners.

Event studies show how fast the adjustment happens. When a company announces earnings that beat expectations, the stock price jumps within minutes, and by the time a retail investor reads the headline the adjustment is done. Mergers, dividend changes, and macro announcements adjust the same way. Information gets impounded quickly.

Chart patterns do not reliably outperform. Academic studies of moving-average crossovers, support and resistance levels, and other technical trading rules generally find that after transaction costs these strategies fail to beat buy-and-hold. Some show gross outperformance, but the trading costs eat the edge.

What Challenges EMH

The factor anomalies refuse to go away. The value premium from post 10, along with momentum and quality, persists across decades and geographies, shows up out of sample, and appears in markets that did not exist when the patterns were first documented. The debate is not whether the premia exist in the data. It is why they exist.

The risk-based account is that the premia compensate holders for exposure to pain that most investors cannot tolerate. Momentum has a severe left tail. It earned strong returns for years, then gave back several years of gains in a matter of months during the 2009 reversal when last year’s losers snapped back violently.

Value can go through protracted stretches of underperformance. From 2010 to 2020, value stocks trailed growth stocks so badly that many managers abandoned the strategy right before it rebounded. Quality holds up better, but even quality-tilted portfolios lag during speculative manias when the market rewards the opposite characteristics.

The behavioral account is that investors systematically misprice these characteristics. They overextrapolate recent trends (fueling momentum), overpay for exciting growth stories (creating the value premium), and undervalue boring, profitable companies (creating the quality premium).

Both explanations can be partially true. Either way, these anomalies challenge semi-strong efficiency. They persist after transaction costs, they survive across different time periods, and they are too large to dismiss as noise.

Behavioral biases push prices away from fundamentals. Investors overreact to dramatic news and underreact to subtle information. They anchor on round numbers, herd into popular names, and panic out of falling ones. Kahneman and Tversky documented dozens of systematic cognitive biases that affect financial decisions. If investors are not fully rational, prices can deviate from fundamentals.

Bubbles and crashes make the same point at scale. The dot-com bubble, the 2008 housing crisis, the meme-stock frenzy of 2021: in each case, prices moved far from any plausible estimate of fundamental value. Efficient-market defenders argue that bubbles are only identifiable in hindsight. Skeptics answer that a market pricing a pre-revenue company at a billion-dollar valuation is not reflecting available information.

Neither side has a knockout punch. Markets look mostly efficient most of the time, but “mostly” and “most” leave room for a lot of interesting debate.

Seeing Patterns in Noise

The weak form makes a claim about what a price chart contains, and our own eyes are worth testing against it first.

The chart below generates a random price path by geometric Brownian motion: a starting price of 100, a drift parameter (expected annual return), and a volatility parameter, combined with 252 days of random draws. There are no earnings surprises, no trends, no insider trades. The path is pure noise. Generate a few paths with the “New Path” button and read each one the way we would read a real chart.

The paths look uncomfortably like real stock charts. We find a rally, a consolidation, a breakout, a crash. We locate support levels and resistance zones, and it is tempting to narrate the year out loud: “it trended up through month 6, then hit resistance around 120 and pulled back.” Which of those features is real?

None of them. The path has no memory, and each day’s return is independent of the last. A price series built this way is a random walk. The human visual system detects patterns regardless, and it finds them in data that contains none.

This is the core argument for weak-form efficiency. If we cannot tell a random walk from a real stock chart by eye, then the “patterns” we find in actual price data may be equally meaningless. Technical analysts see trends. The random walk hypothesis says those trends are noise that happened to run in one direction for a while.

The Grossman-Stiglitz Paradox

Suppose we grant perfect efficiency and follow it through. If markets are perfectly efficient, nobody can profit from research. If nobody can profit from research, nobody does the research. And research is what keeps prices accurate, so granting perfect efficiency destroys the thing that produced it.

Sanford Grossman and Joseph Stiglitz pointed this out in 1980, and it remains the sharpest critique of pure efficiency. They concluded that markets must be just inefficient enough to compensate the analysts and traders who keep them in line. The profit from finding mispriced securities has to cover the cost of the research.

That resolution turns the question from a yes-or-no into a question of degree. A market’s place on that spectrum depends on how many people are trying to exploit mispricings and how costly their effort is. In heavily analyzed markets (U.S. large-cap stocks), the marginal mispricing is tiny. In less-scrutinized corners (small-cap emerging markets, obscure credit instruments), there may be more room. But the analysts flock to wherever the mispricings are, which narrows the gaps.

So the equilibrium corrects itself from both directions. If a market becomes too efficient the analysts leave, and as they leave mispricings widen until the returns pull new analysts back in.

Implications for Investors

Passive indexing is hard to beat. If the average dollar invested in the market earns the market return, and active managers charge higher fees than index funds, then the average active dollar underperforms the average passive dollar by the fee difference. This is arithmetic, not ideology. Whether some managers have skill is a fair argument, but the average active investor loses to the index after costs.

Alpha is competed away quickly. Post 10 defined alpha and post 11 showed how to tilt a portfolio toward it. Finding alpha, though, means identifying mispricings before the market corrects them, and in liquid markets that window is narrow. Quantitative hedge funds spend billions on data, infrastructure, and talent to capture mispricings that last minutes or seconds. An investor checking a brokerage app after work is not competing on the same playing field.

An edge also has to exceed its costs. Transaction costs, management fees, taxes on turnover, and the bid-ask spread all eat into returns, and even a small mispricing, correctly identified, might cost more to trade on than it returns. A strategy that earns 0.5% annual alpha but costs 0.8% in fees and trading costs is a losing proposition.

Diversification remains the only free lunch, and post 8 gave the reason: combining imperfectly correlated assets reduces risk without reducing expected return. No mispricing has to be identified for the benefit to arrive, and it arrives whether markets are efficient or not.

None of this means active management is pointless. It means the bar is high. An active strategy needs a clear thesis about what information advantage it rests on, why the market has not already priced that information, and whether the expected alpha exceeds its costs.

The Three Questions Again

Post 1 reduced every investment to three questions: how much money comes back, when it shows up, and what the odds are that it shows up at all. Amount, timing, risk. Everything since has been a tool for answering those three with more precision.

We answered the how much with bond pricing and the dividend discount model, where a contract spells the amount out for a bond and a growth forecast estimates it for a stock. We answered the when with present value and discounting, which turn a payment date into a price, and with duration, which compresses a bond’s whole schedule of dates into one weighted-average time.

We answered the what are the odds with standard deviation, correlation, beta, and the factor models. Then we put the three answers back together into a set of portfolio weights, and we defined alpha as the gap between what a security earns and what its risk exposure warrants.

The efficient market hypothesis explains why the toolkit exists: markets are not perfectly efficient.

Somebody has to do the analysis. Somebody has to price the bonds, value the stocks, and estimate the risk. The tools in this series are how that work gets done. And the EMH describes the result: the better everyone does that work, the harder it becomes for anyone to gain an edge from it.

The theory builds its own counterargument. The tools are worth learning because the market rewards the people who use them. The market rewards them just enough to keep them working, and not a dollar more.