Skip to content
AmiBroker TA Course

Candlesticks and Chart Patterns

Patterns are the most seductive part of technical analysis, and the part with the widest gap between how confidently they are taught and how well they have been measured. Every trading bookshop carries a volume of named shapes, each with a picture, a story about crowd psychology, and a claimed outcome. Almost none of them carry the two numbers that would let you decide whether the shape mattered: how often the claimed outcome followed the pattern, and how often it followed anyway.

This part supplies the vocabulary, because you need it — patterns are how a large part of the trading world talks, and you cannot read a research note, a forum thread or a colleague’s chart annotation without it. It also supplies the thing that vocabulary usually arrives without: a method for finding out whether any particular pattern was associated with anything at all in your data, over your period, on your instruments.

The problem this part is really about

Section titled “The problem this part is really about”

There is a specific reason pattern research goes wrong, and it is worth stating before you meet a single candle.

A pattern is a shape, and shapes can be invented without limit. Classify each bar by three properties — say, how big its body is, where it closed within its range, and whether it gapped — and you already have dozens of single-bar categories. Chain two bars together and you have hundreds of two-bar sequences. Three bars gives you thousands. Add “but only in an uptrend”, “but only on high volume”, “but only near a prior high”, and the count runs into the tens of thousands.

Now test them all against a few decades of daily data and rank them by how well the next ten bars went. Something will come top. Something always comes top. The question that decides whether you have found anything is whether the top result is better than what you would expect from the same search run on data with no structure in it at all. Most published pattern claims never ask that question, and the ones that do tend to report smaller effects than the pictures suggest.

That is not an argument for ignoring patterns. It is an argument for treating each one as a claim with a testable form, which is exactly what the rest of the course is built to do.

What changes for you here

Section titled “What changes for you here”

Three things should be different by the time you finish this part.

  • You will be able to turn a picture into arithmetic. “A hammer” becomes a statement about the ratio of the lower wick to the body, with numbers you chose and can defend. Once a pattern is arithmetic, two people can implement it identically, and a computer can count it.
  • You will always ask “compared with what?”. A pattern followed by a rise 55% of the time is unremarkable if the instrument rose 55% of the time regardless. The comparison set — the base rate — is the whole test, and choosing it badly is the most common way an honest person fools themselves.
  • You will know how many things you looked at. The number of patterns you tried before you found the good one is part of the result. Keeping that count is a habit, and this is where it starts.

What you need first

Section titled “What you need first”

You need AmiBroker with a database of daily bars for a handful of liquid instruments — the workspace from Part 3 — and enough history for the counts to mean anything. Ten years of daily data on twenty symbols is a reasonable starting point; one year on one symbol is not.

You need the chart-reading vocabulary from Part 4, particularly the swing definition, and the honest treatment of drawn levels from Part 5. The reality check at the end of this part builds directly on the base-rate comparison you first met in Part 6, and adds the harder version of it.

Nothing here needs the Professional edition, a real-time feed or a paid data subscription. Everything is end-of-day work.

You do not need to write AFL yet. Four of the five pages ship a complete formula that you run rather than compose. They are there so that a claim about a pattern can be checked against a count on your own screen within a minute, and so that Part 8 arrives as notation for something you have already watched work.

How the part is arranged

Section titled “How the part is arranged”

The first two lessons build the candlestick vocabulary from the bottom up: what a single candle measures, then what happens when you start chaining bars together and the number of possible patterns explodes. The third moves up a scale to the classic chart patterns — double tops, head and shoulders, triangles, flags — and to the awkward fact that the tools most commonly used to detect them can see the future. The fourth lesson is the methodological core: base rates, conditional versus unconditional comparison, and how to close the “it only works in the right context” escape hatch instead of hiding behind it.

The reality check at the end does the whole thing for real. You encode one candlestick pattern in AFL, define a forward measurement window, compute the unconditional base rate and the within-context base rate, and read what comes back without flattering it. Whatever the answer is on your data, the procedure is the bridge into Parts 8 and 9, where the language you have been running becomes the language you write.

Part7 of 36Level1 — Chart ReaderPages5Estimated time2.3 hours

0 / 5 lessons in this part completed