Chart patterns
The well known chart patterns are vivid, but hard to define clearly. Where they've been rigorously tested, the results are mostly weak.
Learning objective: After this lesson you can place well known chart patterns in context without trusting them more than tests show.
Head and shoulders, double top, triangles, flags: there are whole books about these patterns, and in hindsight they always look impressive. You'll find twelve of the best known with sketches in our chart patterns overview.
The Problem of Ambiguity
The problem is clarity. When exactly is a shoulder a shoulder? How slanted may the neckline be? Give ten people the same chart and they'll draw ten different patterns into it.
Why Testing Fails
That's where testing fails. What can't be defined clearly can't be tested. And what can't be tested can't be called proven.
Where Patterns Make Sense
That doesn't mean it's all nonsense. Some patterns describe understandable processes, for example how the waiting orders are gradually used up in a sideways phase. But that's a description, not a forecast. If you overlook the difference, you pay for it.
If you want a computer to recognize patterns, you have to define them precisely. Usually the price history is smoothed, then the highs and lows are identified and conditions are set for how they must lie relative to each other. Studies that do this find weak anomalies for a few patterns. After trading costs, and once you account for how many patterns were tested, usually nothing is left.
Chance Findings
That second point matters. If you search many patterns, time frames and markets, you'll find some that look good by chance alone. If you test a hundred worthless rules, on average five of them look statistically significant at the usual level. Without a correction for the number of attempts, a single good finding therefore means nothing.
Patterns in Randomness
Then there's our mind. People often see patterns even in random series of numbers. In experiments with made up price charts consisting only of random numbers, participants found the same formations as in real charts. So being able to spot a pattern doesn't prove it means anything.
Summary
- What can't be defined clearly can't be tested.
- People see the same patterns in random price charts as in real ones.
- A single good finding among many attempts means nothing.
Did you get it?
Why are chart patterns hard to test?
Because they can hardly be defined clearly and spotting them remains subjective.
What happens if you test a great many patterns?
Some look good by chance alone. Without correcting for the number of attempts, a single good finding is worthless.
What do experiments with made up price charts show?
That people see the same patterns in them as in real charts.
Check your understanding
Your task
Draw a pattern you recognize into a chart of your choice. Have a second person look at the same chart without your pattern and draw one as well. Write two sentences on where the two drawings differ and which rule is missing for a computer to recognize the pattern unambiguously.
Sources and further reading
- The widely cited study by Lo, Mamaysky, and Wang, which was the first to define chart formations algorithmically and found only weak statistical associations. View source ↗
Related
- All candlestick patterns at a glanceCandlestick patterns
- All 12 chart patterns at a glanceChart patterns
- Fundamental vs. technical analysisStage 4
- Spotting trendsStage 4
- Moving averagesStage 4
- Matching questions for this stageQuestions