What is technical analysis

What a chart records, and what it drops on the way

A chart is a record, not a forecast. For each period it keeps four numbers, the open, the high, the low and the close, plus a volume when the market is centralised. Everything else you see, the colour of the body, the length of the wicks, the shape of a pattern, is a drawing computed from those four numbers. There is nothing else in the data, and knowing exactly what it holds is your first protection against what people will make it say.

What a chart drops is at least as instructive. A candle does not say in which order the high and the low were reached: a session that dives then recovers, and a session that first climbs then collapses before coming back, produce the same drawing while telling two opposite stories. Dropped as well are the orders placed then cancelled without ever being executed, the size and identity of the participants, and the reason behind their decisions.

What remains is thin, and it is enough. What remains is the prices at which a move stopped and turned around, and how many times that happened. A very common confusion starts right here, so it is worth clearing up at once: the volume of a candle says how much changed hands during the period, never at what price inside it. Knowing where the trading concentrated takes a volume profile, which is built from the transactions themselves and cannot be deduced from four numbers. The working assumption of all technical analysis, stated honestly, comes down to this: a price where a move has already turned is a price a great many people can see and remember, and that visibility has a shelf life. No more than that. Not one of the tools you will meet later assumes anything else, they merely measure that visibility differently.

⚠️ The moment you leave the record for the story is easy to spot, because it always carries the same words. “The market wanted”, “sellers trapped”, “it went hunting for”. A record has no intentions. The test fits in one question: which number on your screen supports that sentence? If none does, you have just told a story, which is not forbidden but must not be treated as information.

Why a pattern works: because it is shared

Here is the question nobody asks early enough: through what mechanism would a drawing on a screen move a price? Put that way the idea sounds absurd, and it is. A shape holds no power. What holds power is the number of people looking at the same object at the same time and drawing the same conclusion from it.

Participants do not draw at random. They look at the last visible high, yesterday's low, the session open, round numbers, the most widespread moving averages. When thousands of people compute the same level, that level stops being an opinion and becomes a meeting point. Orders pile up there because everyone knows the others have seen it, and because it is enough that the others react for the reaction to happen.

Two bodies of theory describe this mechanism better than any trading manual does. John Maynard Keynes, in the General Theory of Employment, Interest and Money in 1936, compares the market to a beauty contest where the job is not to pick the prettiest face but the one the others will pick. Thomas Schelling, in The Strategy of Conflict in 1960, shows that two people who must meet in a city without being able to talk converge on the obvious landmark, which he calls a focal point. A chart level is exactly that, and it is a property of the crowd watching, not of the drawing.

The first consequence is practical and saves months of tinkering: an exotic setting is not better because it is rare, it is worse because it is rare. An average computed over thirty-seven periods interests nobody but you, so nobody will react alongside you. The genuinely crowded landmarks can be counted on one hand: yesterday's high and low, the session open and the weekly open, round numbers, and averages over twenty, fifty and two hundred periods. Test it tonight, with nothing to install: over three months, count the touches followed by a reaction on one of those averages, then on a neighbouring length nobody uses. Any gap you find does not come from the formula.

The second consequence is less comfortable, and it has two faces. A regularity that holds because it is shared wears out when the sharing changes, and it does change: populations of participants renew themselves, tools move on, a growing share of the orders is placed by machines. That is why a rule is re-measured rather than believed, and why a six-month record proves very little about the next six. The other face is that the orders a shared level attracts are predictable, starting with the protections everybody places in the same spot: the readability that makes a pattern useful is exactly the readability that makes it busy. The lesson on liquidity and stops draws the consequences.

Where the discipline comes from

The most quoted ancestor is Japanese. Munehisa Homma, a rice merchant from Sakata in the eighteenth century, kept a record of prices and a body of rules still credited to him. That has to be said with care, because part of the lineage was reconstructed long afterwards and candle charts as we know them are documented much later. The idea that matters is older than its tools: write prices down to see what repeats.

The western branch starts at the end of the nineteenth century with the editorials of Charles Dow, from which William Hamilton then Robert Rhea drew the body of principles known since as Dow theory. That is the subject of the market structure lesson, and it is where the notions of trend and confirmation come from, the ones you will meet everywhere else, often with no credit given.

The twentieth century added three contributions still visible in everyday vocabulary. Richard Wyckoff, early in the century, describes accumulation and distribution phases and invents the useful fiction of the composite operator. Ralph Nelson Elliott publishes his wave count in 1938. Robert Edwards and John Magee, in Technical Analysis of Stock Trends in 1948, fix the catalogue of patterns, head and shoulders, triangles, flags, which almost every manual still works from today.

The next shift is technical rather than intellectual. In 1978 Welles Wilder publishes New Concepts in Technical Trading Systems and brings the computed indicator into daily practice. What changes is not the sharpness of the reading, it is its distribution: a computation ships inside software, a hand drawing does not. The sharing on which the whole mechanism depends changes scale at that point.

A serious objection exists, and a school that hides it from you does you a disservice. Louis Bachelier described prices as a random walk as early as 1900 in his Théorie de la spéculation, and Eugene Fama formalised the efficient market hypothesis in 1970: if available information is already in the price, the past of the price predicts nothing. The measured answer came from Andrew Lo and Craig MacKinlay, who showed from 1988 onwards that some statistical dependencies do exist, that they are small, and that they vary over time. The honest position is therefore neither “it always works” nor “it is astrology”: these are coordination regularities, real, small and perishable, and that last property is exactly what makes measurement compulsory.

Three families of reading, three bets on the market

Everything you will read falls into three families, and each one makes a different bet on what a market does next. Knowing them by name saves you from collecting tools that say the same thing three ways, and from believing you diversified your reading when you merely repeated it.

Trend following bets on persistence: a move under way continues more often than chance would suggest. Its tools are moving averages, the break of previous extremes on a slow timeframe, and reading a run of higher lows. The shape of its record is very particular: many small losses while the market goes nowhere, a few long moves during the rare sustained phases. Running it means accepting a low strike rate without taking it personally. Its natural enemy is a market in balance, where it gets sliced session after session.

Mean reversion bets on excess: a price far from its recent reference comes back to it. Its tools are oscillators, bands, the distance to an average. The shape of its record is the mirror image of the previous one, frequent small favourable outcomes and rare large losses. It therefore demands a hard limit decided in advance, because the position looks right until the exact moment it turns out to be wrong, all at once and large. Its natural enemy is the sustained trend, the one that does not come back.

Breakout bets on a change of regime at a threshold: crossing that level does not extend the current state, it opens another one. Its tools are range boundaries, session extremes, compression followed by expansion. It pays the cost of false breaks up front, which makes it inseparable from a participation filter, real volume where real volume exists, session hours, the context of the timeframe above.

There is one reason to name them: they do not oppose each other two by two. Trend following and breakout both bet on continuation and often end up on the same side. Mean reversion bets against, and at the top of a range it sells the excess in the same second the breakout reader buys the push through, at the same price. Two opposite orders on one price, while a third reader has no business being there at all, because a range produces no rising run of lows. The most common incoherence in a beginner is not poor reading, it is running the three families at once and keeping the one that comforts the position currently open. Choosing your family according to the regime, before entering, removes that negotiation entirely.

The trap of reading after the fact

A closed candle tells the story both ways. The same one reads as “buyers took control” or “sellers absorbed the push”, and the choice between the two is almost always made after seeing what followed. A historical chart is crystal clear for that single reason: you already know how it ends, and your eye selects, effortlessly, the elements that were preparing that ending.

This bias is compounded by a second, nastier one that lives in the textbooks. A pattern only receives its name once it has completed. The forty shapes that started exactly the same way and produced nothing carry no name, are filed nowhere and appear in no illustration. The reader therefore sees a collection of successes and takes from it a feeling of reliability that nothing has measured. It is survivorship bias applied to drawings.

The first bias has a name and a date. Baruch Fischhoff documented it in 1975 under the term hindsight bias: once the outcome is known, you can no longer recover the uncertainty you held before it, and memory rewrites the former belief in the direction of what happened, without you noticing. This is not an opinion about trading, it is a laboratory result reproduced for fifty years.

The correction is not to read better, it is to change the order in which you write, which is what the last section is about. A second correction happens in the counting: when you survey a pattern, count the identical starts that never completed too. That number is the denominator missing from every pattern statistic you will ever read. ⛔ A success rate with no denominator is not a statistic, it is decoration.

Timeframe decides everything you see

A violent five-minute drop is an insignificant pause on the daily. Two people looking at the same asset on two timeframes see opposite markets and are both right. There is nothing to arbitrate between them, they are not describing the same object.

That obvious point carries a consequence people rarely follow through. A level, a trend, the value of an indicator, whether a setup counts as one at all, none of it exists outside a timeframe. A rule stated without its timeframe is not a rule, it is a sentence. The first reflex in front of any technical claim, your own included, is to ask which timeframe it was surveyed on.

Pick your decision timeframe, write it in your plan, and judge the position on it. It is the one that defines what a close is, therefore what a break is, therefore what proves you wrong. Without that choice, every candle of every timeframe becomes a contradictory opinion, and you will spend the session changing your mind by changing your zoom.

Here is the asymmetry that matters. Going up one step to place the context is legitimate, and it is done before entering. Going down one step while the position hurts is not, and it is almost never analysis, it is the search for a reason to stay. The test is chronological rather than technical: whatever you consult after entering that you had not planned to consult is not reading, it is a negotiation with yourself.

Record two separate pieces of information in your journal, the timeframe you had decided on and the one you actually watched. After a month, check whether the gap between them lines up with your worst exits. You do not have to take my word for it, you can measure it on your own record.

What technical analysis cannot know

It gives no cause. The chart shows that price fell, never why. Sticking an explanation on afterwards costs nothing as long as nobody extrapolates from it, and extrapolating from it is nonetheless what almost everybody does, including in the most serious market commentary.

It gives neither a target nor a date. A setup says “reactions have already happened here”, not “it will reach that price before Friday”. Any source announcing a numeric target with a date is selling you a certainty the data does not contain.

It gives no probability for the next candle. The only honest quantity is a frequency measured on a record, over a defined sample, with a defined rule. It is not a property of the pattern, it is a property of your rule applied to a given period. Change the period and the number changes, which is precisely what a serious backtest tests.

Finally, it knows nothing about what is not yet in the price, a central bank decision, a release, a liquidity accident. That is why position size is decided by the risk rule, as the lesson devoted to it sets out, and never by confidence in the reading: the reading can be excellent and the session unmanageable.

⛔ It also cannot tell a good decision from a good outcome. A trade taken outside your rules that ends up in profit is still a bad trade with a happy ending, and it is the one that will cost you most, because it teaches you to do it again. Only a dated piece of writing separates the two, which brings us to the exercise.

The forecast notebook: two weeks, ten minutes a day

Here is the exercise. Every day, before the close of the candle on your decision timeframe, you write one forecast, a single one, in a form that can be wrong. No position is needed, and that is the whole point: the exercise measures your reading alone, without the noise of money, and it costs nothing beyond ten minutes.

Four fields, not one more. One, what you expect, in a sentence. Two, the level that would prove you wrong. Three, the deadline, in candles or at a precise hour. Four, your confidence in three notches, low, medium or high. Without the second and third fields, a forecast always ends up being declared correct, because a price eventually passes everywhere if you give it enough time.

Those first three fields did not come from nowhere. This is Karl Popper's falsifiability, stated in 1934 in Logik der Forschung: a statement has value only if some observation could contradict it. Applied to a chart the rule is brutally effective, because it wipes out in one move the forecasts of the “it should go up, unless it goes down” kind, which are the majority of those we form in our heads and never write.

After two weeks you have a dozen lines and you count four things. How many forecasts were even checkable, and the first surprise is finding that half of them were not. How many turned out right. How many turned out right among the “high” ones compared with the “low” ones: if your high-confidence calls do no better than your low-confidence ones, your confidence carries no information at all, and that is the most useful thing you can learn about yourself this year. How many, finally, you rewrote after the fact, which you will know because your lines are timestamped.

The second fortnight adds one column, the regime named before the forecast, trend or range, and you cross your results by regime. The most common case by a distance is a reading that is sound in one regime and worthless in the other, which hands you a filter that costs nothing. Keep that notebook even once you move to real positions. A trade journal records outcomes, a forecast notebook records the reasoning that preceded them, and it is the only document your memory cannot rewrite. Together the two answer the question that decides how fast you progress: am I losing because I read badly, or because I read correctly and execute badly?