Market Efficiency: A Framework for Judging Your Edge

A screen you built over the weekend just lit up. Across ten years of history it turned a modest stake into something that looks a lot like skill: the equity curve climbs at a steady angle, the drawdowns stay survivable, and the win rate is high enough to keep you interested. Market efficiency is the idea that decides whether you’ve found a real edge or simply fooled yourself. It reframes the excitement into one hard question. How much of that result is information the market hadn’t yet priced, and how much is risk, luck, or a cost you forgot to subtract?

I treat efficiency as a lens for reading evidence. Used that way, it turns a good-looking backtest into a set of checks you’ll need to clear before you trust your own numbers.

What market efficiency actually claims

The useful version of the idea asks a narrow question: how quickly and how completely does available information get absorbed into prices? That’s a statement about the speed and thoroughness of pricing, not a promise that prices are correct or that the future is easy to forecast. A market can be efficient and still be wrong often, because new facts keep arriving and no one prices tomorrow’s news today.

The tradition splits the idea into three forms, and the language matters. The weak form concerns information already contained in past prices and trading data, so it speaks to whether chart history alone gives you an ordinary profit. The semi-strong form concerns all public information, including filings, earnings, and news. The strong form is a demanding theoretical benchmark that folds in private information as well, and almost no one argues it holds in a strict sense. Most real debate lives in the middle.

Here’s a common misread. Weak-form efficiency gets read as a flat claim that price patterns never repeat and technical work is pointless. Its real reach is narrower. The ordinary, obvious pattern tends to get competed away once enough people can see it and act on it, which is a statement about the average return to a public rule after costs. That still leaves room for edges that are harder to find, harder to hold, or tied to risk most traders would rather avoid.

The mechanism that keeps prices honest

Prices stay roughly fair because a crowd of motivated people keeps pushing on them. When a company reports, quotes reprice within seconds as participants compete to act first. The reward for trading on public information shrinks as more traders chase the same signal, and the cost of trading, the spread, the commission, the slippage, eats whatever thin margin remains.

Risk does the rest of the work. Some returns look like free money until you notice you were paid to hold something painful: a position that bleeds in recessions, gaps against you overnight, or drops exactly when you most need the cash. New facts then reset the whole board, which is why a rule that’s worked for years can go quiet without warning. The mechanism only needs enough motivated traders competing for the easy money to get arbitraged toward zero. They don’t have to be perfect, or even right very often.

Why a pattern is not proof of an edge

A backtest that shows a pattern proves one thing for certain: the pattern existed in that data. Whether it’s an edge you can bank is a separate question, and there are at least five honest explanations for a pretty curve, only one of which pays you going forward.

It can be compensation for risk. Many documented effects, including factor momentum, may be real and still amount to a fee the market pays you for holding exposure that hurts at the wrong times. It can be a statistical accident, the kind of chance pattern that surfaces when you test enough rules against enough history. Nassim Taleb’s writing on randomness, drawn out in our lessons from Nassim Taleb, is a useful corrective here: run enough coin-flippers and some will look like geniuses on paper. It can be an implementation problem, an edge that lives on prices you could never actually trade. It can be temporary behavior that fades as the crowd learns. And it can be an effect that simply disappears after costs and wider adoption.

The first screen I ever trusted showed a Sharpe near 2 on the stretch of data I built it on, then slid under 0.5 the moment I pushed it onto years it had never seen. Nothing about the rule changed. I’d just been reading the noise in my sample as if it were signal. That gap between the fitted result and the honest one is the single most expensive lesson in this whole subject.

A checklist for judging a claimed anomaly

When someone shows you an anomaly, your own or a stranger’s, a short list of questions separates the durable from the decorative. None of them are exotic, and most edges die on the first two.

• Did it survive out-of-sample? A result that holds only on the data used to build it is a description of the past. Walk-forward analysis is the honest way to see whether a rule keeps working on data it was never fit to.
• Were survivorship and look-ahead bias handled? Testing only on companies that still exist quietly deletes the failures, which is why survivorship bias in a backtest flatters almost every long study that ignores it. Look-ahead bias, using data you wouldn’t have had in real time, does the same damage in reverse.
• Were turnover and trading costs included? A signal that trades often has to clear a real bill of spreads, commissions, and slippage before any of the paper return is yours.
• Did it hold across different periods and regimes? An edge that only appears in one bull market is a story about that market.
• Is there a plausible economic or behavioral mechanism? A reason you can state in a sentence is what separates a finding from a coincidence.

The costs question is the one that’s humbled me most. The first time I added realistic slippage and a round-trip commission to a high-turnover signal, its few points of annual edge turned negative. On paper it looked like an anomaly. In an account it was a way to pay the broker.

When the market watches itself back

There’s a newer wrinkle worth taking seriously. As forecasting systems and automated strategies grow, their outputs start to shape the very environment they were trained to predict. Research through 2026 on algorithmic feedback points at the same uncomfortable idea: once a model’s predictions drive real decisions at scale, the model stops being a neutral observer of the market and becomes one of the forces moving it.

This complicates simple historical evaluation. A backtest quietly assumes the world that produced your data will keep behaving the same way. When a popular signal attracts enough capital, the edge it measured can compress, invert, or cluster its losses into crowded exits that never appeared in the historical file. The misread to avoid is treating a clean backtest as a guarantee of future behavior, when at best it’s evidence about a market that may already be adapting to the strategy you’re about to run. Re-testing on fresh data, in the spirit of a three-phase backtesting protocol, is how you catch decay while it’s still cheap.

The test you can never run cleanly

Even with a spotless dataset, there’s a limit built into the question itself. You can never test market efficiency on its own. Every test also assumes a model of what returns and risk should look like, so a rejected result might mean the market is inefficient, or it might mean your model of fair return was wrong. The two are welded together, and no experiment fully separates them. This is the joint-hypothesis problem, and that’s why honest researchers rarely claim to have proven the market efficient or inefficient in general.

For a trader, that limit is oddly freeing. You don’t need a final ruling on efficiency. You need to know whether a specific edge survives out-of-sample testing, realistic costs, and a plausible reason for existing. Those are answerable questions, and answering them well is most of the job.

Trading as if the market is smart

The practical stance is to assume the market is hard to beat and then demand that any edge prove otherwise. That posture keeps you skeptical of your own backtests, forces you to subtract costs before you celebrate, and treats a decayed signal as normal rather than a betrayal. Efficiency works as a standard of evidence. It tells you how much to trust what you’ve found, and how loudly the numbers are allowed to speak.

Hold the frame in mind every time a fresh curve tempts you, and connect it to the rest of the library: the backtesting method that stress-tests it, the cost accounting that grounds it, and the behavioral reasons prices drift in the first place. Learn the pattern. Ride the trend. Keep the gains.

Educational content only. Not investment advice. Trading involves risk. You are responsible for your decisions.