Dismissing technical analysis wholesale as astrology is lazy; treating it as a basis for decisions is dangerous. There is only one correct question: price and volume are evidence — what can they prove, and what can they not? Four things this issue: the most stubborn anomaly, what volume is actually saying, how a rule gets arbitraged away, and which part a value investor should keep.
The Principle
Relative strength over the past 3–12 months positively predicts relative returns over the next 3–12 months — one of the very few pure-price signals that has yet to be falsified.
Origin · Quote
"Strategies which buy stocks that have performed well in the past and sell stocks that have performed poorly in the past generate significant positive returns over 3- to 12-month holding periods."
— Narasimhan Jegadeesh & Sheridan Titman, The Journal of Finance, Vol. 48, No. 1 (1993)
Reading It Closely
Three things are routinely misread. First, momentum is a cross-sectional ranking, not "buy what has gone up" — it compares the relative strength of every stock at a single point in time. Second, the time scale is the entire content of the claim: one month is reversal, 3–12 months is momentum, 3–5 years is reversal again (De Bondt & Thaler, 1985). Third, the cause remains unsettled: the behavioral camp attributes it to underreaction, the risk camp to compensation for some unidentified risk — and "unidentified risk" is just another name for a crash.
The Case
Daniel and Moskowitz documented the shape of that risk in Momentum Crashes (Journal of Financial Economics, 2016): in July–August 1932 the momentum portfolio lost roughly 91% in two months; in March–May 2009, as the market rebounded off the bottom, the shorted "losers" soared and momentum lost roughly 73% again.
Both crashes share the same signature: the tail end of a bear market + high volatility + a reversal in direction. The losses cluster precisely in the weeks when you most need to stay invested — that is not bad luck: shorting losers after a long decline is shorting a spring already compressed to its limit.
Limits · Decision Checklist
The negative skew of the return distribution makes the Sharpe ratio systematically overstate its appeal — small gains for years, then a loss you cannot recover from. Annual turnover often exceeds 100%, and spreads, market impact and short-term capital gains tax (Day 46) consume much of the paper return. More fundamentally, momentum is a portfolio-level statistical regularity, incompatible with concentrated holdings.
- Which window does my "trend" refer to? Switch to one month — does the conclusion flip?
- Across how many positions will I apply this signal? Fewer than 20 → it's a single bet.
- Net of turnover costs and tax, how much excess return is left? Can't compute it → don't use it.
- If the market reverses violently within three months, can I stomach this position's drawdown?
The Essence · This Week's Reflection
Momentum is statistically real and experientially unbearable — it makes you bleed at precisely the moment you can least afford to.
Review your purchases over the past 12 months and mark the ones whose real reason was "it had already gone up" — you may have been running momentum all along, without its discipline.
The Principle
Volume correlates with the magnitude of a price change, not its direction. It measures disagreement and the rate of information arrival, not "the balance of bulls and bears."
Origin · Quote
"Noise makes financial markets possible, but also makes them imperfect. … People sometimes trade on noise as if it were information."
— Fischer Black, "Noise," The Journal of Finance, Vol. 41, No. 3 (1986)
Reading It Closely
Black exposed what volume actually is: if everyone read information identically, volume would approach zero — trades happen because people disagree. Karpoff's 1987 survey confirmed the direction: volume is reliably positively related to the absolute value of the price change. So "heavy volume on the way up means institutions are accumulating" is self-consolation: in that same trade a buyer is building and a seller is distributing, and volume does not tell you which side is smarter.
Price up · Volume upNew information being widely absorbed — or distribution at the top, existing holders handing stock to new buyers.
Price up · Volume downMarginal buying is drying up. The advance continues; the participants are leaving.
Price down · Volume upUsually mechanical selling: redemptions, margin calls, index changes. The one quadrant that often creates entry points.
Price down · Volume downNobody cares. The value investor's usual hunting ground — and the usual address of a value trap.
All four quadrants answer "who is trading and why." None of them answers "buy or sell" — that requires a valuation.
The Case
On October 19, 1987 the Dow fell 508 points (−22.6%) in a single session and NYSE volume hit 604 million shares, about 1.8× the prior record (roughly 344 million on October 16). The Brady Commission's conclusion was not "panic" but mechanism: portfolio insurance sold into the decline by prearranged rule, the selling pushed prices lower, and lower prices triggered more selling.
The only information volume carried there was this: the sellers were mechanical, not judgmental. March 2020 was the same — margin calls and cash needs drove indiscriminate selling; even gold and Treasuries were sold. Separating mechanical selling from a fundamental re-rating is the one genuinely valuable use of volume.
Limits · Decision Checklist
Market structure keeps diluting what volume means: dark pools and broker internalization never reach public exchange data (often around 40% of US equity volume in recent years), and ETF creation/redemption, algorithmic order slicing and index rebalancing generate enormous trading unconnected to anyone's opinion (Day 56 opens up that plumbing).
- Does this volume spike have a mechanical explanation (index change, option expiry, a large redemption)?
- Am I looking at absolute volume or a turnover percentile? Only the latter compares across time.
- Is the seller forced or willing? Can I name who is selling and why they must?
- If this volume carried no information, would my conclusion change? No → I never needed it.
The Essence · This Week's Reflection
Volume answers "who is trading and why." It never answers "up or down next."
Recall a day one of your holdings collapsed: were the sellers mechanical (redemptions, margin calls, index changes) or judgmental? Which one did you react to?
The Principle
For any technical rule the question is not "does it work," but: on which sample, net of how much cost, and with how many people using it simultaneously — how much is left?
Origin · Quote
"The superior performance of the best trading rule is not repeated in the out-of-sample experiment covering the period 1987–1996."
— Ryan Sullivan, Allan Timmermann & Halbert White, The Journal of Finance, Vol. 54, No. 5 (1999)
Reading It Closely
This chain of papers is itself the whole lesson. In 1992 Brock, Lakonishok and LeBaron tested moving-average and trading-range-break rules on Dow data from 1897–1986 and reported statistically significant predictive power. In 1999 Sullivan and co-authors identified the fatal premise: those rules were the few that had survived decades of market practice, and the true denominator is the thousands tried and forgotten. They expanded the universe to 7,846 rules and used a bootstrap to correct for data snooping; in-sample significance shrank drastically and out-of-sample it vanished. The significance of any single rule must be divided by the number of rules tried to find it.
The Case
Two decades of trend following. In 2008 the SG Trend Index rose about 20% while the S&P 500 returned roughly −37% — its hedging value in a systemic crash is real and repeatable. But from 2011 to 2019 the same index went essentially nowhere, as fees and a low-volatility regime consumed nearly all of the return. Same rule, different regime, opposite conclusion — which is exactly why the word "effective" is nearly meaningless in technical analysis.
Limits · Decision Checklist
Conversely, dismissing technical analysis wholesale as astrology is also lazy: momentum and volatility clustering are independently replicated empirical facts, and Lo, Mamaysky and Wang (2000) found several formalized patterns do carry incremental information — it is just that "carries information" is not "profitable net of costs." The real dividing line is falsifiability: "the 50-day crosses above the 200-day" can be tested; "the fifth Elliott wave extended" cannot, because the wave count can be re-drawn after the fact to explain any outcome.
- Does this rule have an unambiguous entry and exit definition written down in advance?
- Are the reported returns net of commissions, spreads, market impact and tax?
- Is there out-of-sample performance from the years after publication? No → treat as unverified.
- How many rules and parameter sets did I try in total to find it?
The Essence · This Week's Reflection
A technical rule's real value = out-of-sample return − trading costs − everyone else running the same rule.
Write down the precise definition of whatever timing or stop-loss rule you use, then answer: did I have the rule before I looked at the data, or only after?
The Principle
Confine price signals strictly to risk and execution — sizing, leverage, liquidity. Never let them enter the valuation judgment of what the business is worth.
Origin · Quote
"One principle that applies to nearly all these so-called 'technical approaches' is that one should buy because a stock or the market has gone up and one should sell because it has declined. This is the exact opposite of sound business sense everywhere else, and it is most unlikely that it can lead to lasting success on Wall Street."
— Benjamin Graham, The Intelligent Investor, Chapter 1
Reading It Closely
Graham is right, but only under one premise: that your valuation is reliable. When the valuation rests on dressed-up statements or a stale industry assumption, a decline is not a discount — it means someone learned the truth before you did. The hard question becomes: how do you tell "the fall is an opportunity" from "the fall is information"? The answer is not in a chart but written in advance — "if my thesis is wrong, in what form will it first show up?" Price is the least reliable of those forms, and the earliest to arrive.
The Case
Both directions have to be held at once. Legg Mason Value Trust beat the S&P 500 for 15 consecutive years (1991–2005); through 2007–2008 Bill Miller kept averaging down in financial and housing-related holdings, and in 2008 the fund fell about 55% against roughly −37% for the S&P 500 total return. Miller did not misunderstand value — he mistook "cheaper" for "a new reason to buy."
The other direction: Berkshire Hathaway's share price fell about 49% from mid-1998 to March 2000; Amazon fell more than 90% from its late-1999 high into 2001. Any price-based stop would have taken you out at both of those points — precisely the two places you should not have been taken out.
Limits · Decision Checklist
One line reconciles them: a stop protects leverage and liquidity, not judgment. If you carry no leverage, your capital's horizon is long enough, and your thesis is falsifiable, a decline alone is not a reason to sell. But the moment leverage or redemption pressure exists, price must enter the decision — not because it is smarter, but because you may not survive until you are proven right.
- Can I explain this sale without mentioning price at all? No → price is thinking for me.
- Do I have leverage or a cash need within a year? Yes → price gets a veto.
- Is my reason for averaging down a new fact, or merely that it got cheaper?
- If it were suspended for a year and I could see no price, would my decision to hold change?
The Essence · This Week's Reflection
Price determines whether you survive to the day you are proven right. It can never determine whether you are right.
Pick a position sitting at a loss and write three sentences, in fundamentals language only, on why you still hold it. If one of them says "it has already fallen so much," price has crept into your thesis.
Going Deeper
If momentum is real, why doesn't a value investor just use it?
Because it is incompatible with your other constraints. Momentum demands dozens to hundreds of positions, turnover above 100% a year, and the nerve not to cut during a crash — while concentration, long holding periods and low tax drag are a long-term investor's three great structural advantages. Borrowing a strategy's signal without accepting its cost structure usually yields the worst of both: too little diversification to capture the statistical regularity, enough turnover to owe the tax. Only one thing transfers: don't keep averaging into 3–12 month relative weakness unless you can state a new fact.
In the AI era, has chart pattern recognition been eaten entirely?
The precisely definable part is disappearing fast: machines can enumerate a rule space far beyond human reach, so any replicable pattern decays faster than Sullivan's generation observed. But two things will not disappear. First, price series are non-stationary — participants change behavior because the models exist, so the training distribution differs from the future one, and no amount of data fixes that. Second, crowding is itself the new risk: when every model learns the same signal, the liquidity to exit vanishes for all of them at once. The stronger the models, the more the remaining alpha concentrates in what models cannot observe: private information, patience, and bearing risk others will not.
How much of technical analysis is self-fulfilling?
More than you'd think, and therefore unstable. The 200-day average "works" partly because enough people watch it — it becomes a Schelling point, a shared focal point for expectations, the same family as Soros's reflexivity (Day 33). But focal points get front-run: once enough traders anticipate the level, the signal is absorbed before it triggers, and the regularity drifts and then dies. A self-fulfilling regularity lives exactly as long as front-running it stays expensive.
If price carries no information, how does the market get priced at all?
This is precisely the Grossman–Stiglitz paradox (1980): if prices already reflected all information, nobody would have an incentive to pay the cost of gathering it; and if nobody gathers it, prices cannot reflect it. The market must therefore rest at an equilibrium in which some obtainable excess return always exists, exactly equal to the cost of collecting and analyzing information. The practical implication is concrete — your excess return does not come from reading charts, but from the work others are unwilling to do: finishing that annual report, thinking that industry through, and holding when others are forced to sell.