Technical Analysis: What It Is and What the Evidence Says

Scope. This article and the three that follow it describe what technical analysts actually do and what the research has found. They do not teach a method. No signal, entry or exit rule, parameter setting or backtested result appears in any of the four, and none of them concludes that the practice works or that it does not. A reader should finish understanding the practice and the state of the evidence, and should not finish holding something to apply.

That is the whole of the definition. Everything else — charts, patterns, indicators, levels — is apparatus built on top of it, and the apparatus is what most descriptions of the subject start with, which is why the underlying claim so often goes unexamined.

The claim is unusually testable, which is the reason this pillar can treat it seriously rather than dismissively. It says that the record of past prices contains information about future prices. That proposition can be stated formally, and it has been tested for six decades.

The three classical premises

The tradition running from Charles Dow through Edwards and Magee rests on three assertions that practitioners generally state openly.

Price discounts everything. Whatever is known about a company, an economy or a sentiment is already reflected in the price, so the price is a sufficient statistic and studying anything else is redundant.

Prices move in trends. Movements persist rather than being independent from one period to the next, and a trend in place is more likely to continue than to reverse.

History repeats. Participants respond to similar conditions in similar ways, so recognisable configurations recur.

These are premises, not findings, and the distinction is the whole article. Each is an empirical claim that could be true or false, and none is established by being stated at the front of a textbook. The first premise, taken strictly, is the semi-strong efficient-market hypothesis — and if it holds, the second premise cannot, because a price that fully reflects available information moves only on news, and news is by construction unpredictable. Premises one and two are in tension with each other, and a reader who notices that has understood something most introductions to the subject skip past.

What practitioners actually do

The practice is far broader than the caricature of a lone chartist drawing triangles. Technical methods are used by foreign-exchange dealers, futures traders, systematic trend-following funds and a large population of individual traders, and the professional end of it is quantitative, rule-based and risk-managed rather than impressionistic.

A working process typically involves selecting a timeframe — the same instrument produces entirely different pictures on a five-minute and a weekly chart, and practitioners treat the timeframe as a choice rather than a given; characterising the trend on that timeframe; marking levels where price previously turned, covered in Support and Resistance; overlaying indicators that transform the price series into a smoothed or bounded derivative of itself; looking for confluence, meaning several of these apparatus agreeing; and defining exposure and exit in advance.

That last element deserves emphasis, because it is frequently the part that carries the practical weight. Much of what a technical framework supplies is not a forecast but a discipline: a predetermined point at which a position is judged wrong, a size fixed before the position is taken, and a rule that removes discretion at the moment discretion is least reliable. Those are risk-management properties, and they are logically separable from any claim that the chart predicts anything. A framework can be useful for imposing consistency while being uninformative about direction, and the two questions are routinely conflated by both advocates and critics.

What the evidence says

The formal question is weak-form efficiency: whether past price and volume data can be used to earn returns above a passive alternative after costs. Pillar 22's article on the efficient-market hypothesis sets out the three forms and what each rules out. Technical analysis is a direct test of the weakest of them.

The literature does not deliver a verdict, and reporting it honestly means reporting that.

The most comprehensive survey of the field reviewed the empirical work in two groups. Early studies found technical trading strategies profitable in foreign-exchange and futures markets but not in stock markets. Among 95 modern studies, 56 reported positive results, 20 negative and 19 mixed — and the same survey qualified that tally heavily, noting that most of the studies are exposed to data snooping and to selection of the rules after seeing the data, and that the positive evidence runs at least until the early 1990s rather than through to the present.

The data-snooping correction is the most important methodological development in the area. A widely cited 1992 study of moving-average and range-breakout rules on the Dow found support for them. A 1999 paper re-ran the exercise across a universe of 7,846 rules drawn from five families — filter rules, moving averages, support and resistance, channel breakouts and on-balance-volume averages — applied to a century of daily Dow data, using a bootstrap that adjusts for the fact that the best rule was selected from a large search. Correcting for the search substantially changed what the earlier result meant.

The other direction is equally real. A 2000 paper attacked the subjectivity problem directly by defining patterns algorithmically through nonparametric kernel regression and applying the definitions to a large sample of US stocks from 1962 to 1996. It found that several technical indicators do provide incremental information relative to the unconditional distribution of returns. That is a peer-reviewed finding in a leading journal, and it is not compatible with blanket dismissal.

Two pieces of arithmetic that hold regardless of the debate

The first is the size of the search. Consider only the simplest family — a rule that compares a fast moving average to a slow one. With the fast length running from 2 to 50 days and the slow length from one day longer up to 200, there are 8,526 distinct pairs. Allow three choices of price input and four confirmation lags and the count reaches 102,312.

The second is friction. Any rule that generates activity incurs spread, commission and market impact on every leg, and that cost is certain while the return is not.

Canonical data — and why this cluster does not use Wexford Instruments. Every other article in this pillar works from Wexford Instruments, the portal's fictional statement company. Wexford has financial statements but no price history, and inventing one would create a competing canonical figure. The four technical articles therefore use synthetic price series generated as driftless random walks with stated parameters, which is not a limitation but the point: a synthetic series with no mechanism in it is exactly the benchmark against which any claim about chart structure has to be judged. All parameters, definitions and counts are stated where used.