Does technical analysis work? What a chart can and cannot tell you

The honest answer is that academic evidence runs in both directions, that the favourable evidence carries a serious methodological problem, and that what worked in the 1980s worked worse in the 1990s. Neither "it's astrology" nor "it's a science": it's a tool with abundant literature and conclusions uncomfortable for both camps.

What technical analysis is and what it isn't

Technical analysis studies past price and volume to try to say something about the future. That's the whole definition, and from it come the two usual confusions.

What it is

A way of reading how the people already inside an asset are behaving: where they bought, where they sold, with how much conviction, and in what quantity.

It doesn't try to value whether a company is good, which is what the previous chapter did. It tries to describe the behaviour of those who have already made their decision about it.

What it isn't

It isn't predicting the future. No pattern says what will happen. At best it says that in similar past situations, one thing happened more often than another.

It isn't independent of everything else. A chart doesn't know the company reports tomorrow, or that a regulator just opened an investigation. Everything you learned in previous chapters still governs.

And it isn't an alternative to looking at the accounts. They're two different questions: one is what the company is worth, the other is what the price is doing. Confusing "the chart looks good" with "the company is good" is the underlying error most of the others come from.

What the academic evidence says

There's abundant literature on this, and it's worth reading before picking a side. The reference review is by Cheol-Ho Park and Scott Irwin, published in 2007 in the Journal of Economic Surveys, which went through 95 modern studies on the profitability of technical strategies.

The count

Result	Number of studies
Positive	56
Negative	20
Mixed	19

At first glance it looks like a clear win. And leaving it there would be dishonest, because the authors themselves don't.

The asterisk, which is half the conclusion

Park and Irwin note that most of those empirical studies suffer from problems in their testing procedures. They name three specifically:

Data snooping. Trying many combinations on the same data until one works. With enough attempts, finding something that looks brilliant is practically guaranteed, whether or not a real edge exists.

Ex post rule selection. Choosing which rule to test after seeing the data, which is an elegant way of cheating without noticing.

Difficulty estimating risk and transaction costs. Many studies calculate gross profits. With the per-trade costs we covered in Chapter 3 — spread included — a strategy that trades heavily can go from profitable to ruinous without a single signal changing.

The authors conclude that future research correcting those deficiencies is needed to provide conclusive evidence. Meaning: after 95 studies, the honest answer is still "unresolved."

The finding that orients most: the edge erodes

There's a result in that same review that says more than the count. Modern studies find consistent profits in speculative markets at least until the early 1990s, and in stock markets profitability decreases from the late 1980s onward.

That pattern matches exactly what the efficient market hypothesis predicts, as formulated by Eugene Fama in 1970: if prices absorb available information, any documented and published edge destroys itself. Once enough people know and exploit a pattern, the exploitation eliminates it.

It isn't that technical analysis never worked. It's that what worked stopped working as it became well known.

So what conclusion to draw

Three, and none is comfortable for anyone.

Dismissing it entirely isn't supported by the evidence. Fifty-six positive studies are too many to treat it as superstition.

Treating it as a formula that works isn't either. The methodological asterisk is enormous and the erosion over time is documented.

What survives all of this is the descriptive part. A chart shows, beyond argument, where the price has been, how much it moved, and on what volume. That isn't a prediction: it's a fact. And from those facts you can ask better questions.

This chapter deliberately stays in that part. It teaches you to describe a chart precisely and to know the tools, without promising any of them anticipates anything.

How to read a candlestick

A candle summarises four figures from a period: open, high, low, and close. It's the world's most-used format because it packs those four numbers into a shape you read at a glance.

The anatomy

The body is the rectangle, running from open to close. If the close lands above the open it's coloured one way; below, another.

The wicks are the thin lines extending above and below, marking the high and low reached during the period.

Those four figures — the OHLC — are all a candle contains. Everything else is interpretation.

What each part contributes

A long body means the price moved a lot between open and close and ended far from where it started. It indicates one side dominated the whole period.

A very small body means it opened and closed in nearly the same place. Whatever happened in between, the period ended in a draw.

A long wick means rejection. The price got there and didn't stay: someone pushed back hard enough to return it. A long lower wick says a move down was attempted and couldn't hold; a long upper one, the same upward.

A warning about candlestick patterns

Dozens of named patterns exist — hammer, engulfing, shooting star — and an entire industry of material promising they anticipate reversals.

They deserve exactly the criterion from the previous section: they're descriptions of what already happened, and their predictive capacity is subject to the same methodological problems as everything else. Memorising forty pattern names isn't learning to read a chart.

What does help, and what this chapter teaches, is knowing how to describe what a candle says: where it opened, where it closed, how far it got, and what happened to that attempt. That's a verifiable fact and the basis of any subsequent reading.

Why the timeframe changes everything

The same asset, on a weekly chart, a daily one, and a one-hour one, can tell three different stories at once. None is lying: each answers a different question.

A stock can be, simultaneously:

• Rising on the weekly chart, in a sustained trend for months.
• Falling on the daily, mid-way through a two-week correction.
• Turning up on the hourly, forming a short-term floor right now.

All three readings are correct and describe different scales of the same movement.

What each is for

The wide frame — weekly, daily — answers where the asset stands fundamentally and where the levels most people watch sit.

The short frame — hourly, fifteen-minute — answers what's happening right now within that context.

Using only the short one is reading a stray sentence from a book. Using only the long one is knowing what the book is about without knowing which page you're on.

The error it produces, and it's about honesty

The problem isn't looking at several timeframes. It's switching frames until you find one that confirms what you already wanted to believe.

If the daily says the price is falling and you don't like that reading, there will always be a timeframe where it looks like it's rising. With enough frames available, any thesis finds support in one of them.

The way to avoid it is deciding beforehand which frame defines context and which defines the moment, and not switching mid-decision. It's the same discipline we'll see at the chapter's end with overfitting (tuning a rule until it fits the past perfectly, which is the surest way to make it fail in the future): the problem is almost never the tool, it's the order it gets used in relative to the conclusion.

Linear or logarithmic scale: which to use and when each one misleads

Logarithmic for any long history, or any asset whose price has multiplied. Linear for the short term, where it makes no difference. The distinction is that linear measures distance in currency and logarithmic measures it in percentage, so across decades the linear scale flattens the past and inflates the present.

What each one does

Almost every platform draws the price axis linearly by default: the distance between 10 and 20 on screen equals the distance between 100 and 110, because both span ten units.

The problem is those two moves aren't comparable. From 10 to 20 is +100%. From 100 to 110 is +10%. A linear scale paints them identically.

The logarithmic scale does the opposite: the same vertical distance always represents the same percentage change. On it, 10 to 20 and 100 to 200 occupy the same space, which is correct, because both are doubling.

The example that settles it: the 1929 collapse

The figures come from Federal Reserve History, the US Federal Reserve System's historical archive, covering the Dow Jones Industrial Average.

The index closed at 381.17 on September 3, 1929. It fell to 41.22 on July 8, 1932, 89% below its peak, and did not return to that level until November 23, 1954: twenty-five years later.

Now draw that on a Dow chart running from 1920 to today, with the index in the tens of thousands of points.

On a linear scale, the most devastating collapse in the history of the US market is a scratch in the bottom-left corner. Three hundred and forty points are invisible when the axis reaches forty thousand. The picture suggests not much happened in 1929.

On a logarithmic scale, that same stretch takes up the height an 89% fall deserves, which is enormous. And at the same time, any recent 10% correction appears the size it actually is: small.

Both images show the same index and the same data. One of the two is telling the story backwards.

When the logarithmic scale misleads

Worth saying, because it isn't that one is correct and the other false: each answers a different question.

Logarithmic misleads when what matters to you is distance in money. Looking at a short stretch and wanting to see how many euros separate the current price from a specific level, the logarithmic scale compresses that distance at the top and stretches it at the bottom, and the visual impression stops matching the arithmetic.

For a chart covering days or weeks, where the price doesn't change order of magnitude, the difference between the two is imperceptible and linear reads more directly.

The practical rule

Long history or multiplied price: logarithmic. Anything spanning years, or any asset that has gone from tens to hundreds.

Short term: either works, and linear is easier.

Almost every platform switches it with one button, and almost nobody touches it. It's the cheapest correction available on a chart: it requires learning nothing, only knowing the button is there.

What to remember

• Of 95 modern studies reviewed by Park and Irwin, 56 find positive results, but the authors themselves warn of serious methodological problems in most.
• The documented profitability of technical strategies erodes over time, exactly as the efficient market hypothesis predicts.
• A candle contains four figures and nothing more: open, high, low, close. The rest is interpretation.
• Switching timeframes until you find the one confirming your thesis isn't analysis, it's confirmation bias.

Related: what actually happens when you hit buy · why price history changes after a split · what each indicator measures and how people fool themselves