The Limits of Chart Analysis: Confirmation Bias, Hindsight and Other Common Illusions

A chart made by flipping coins

The figure below is not a real market price. We made each candle by flipping a coin six times, moving the price up 1% for heads and down 1% for tails, and then strung those candles together. We generated about 20,000 such price paths and picked one that looked like a head and shoulders. It has a left shoulder, a head, a right shoulder and a neckline, and after dropping below the neckline, price keeps falling for a long while. It contains candlestick patterns too: using this course's definitions, the candles visible in the figure include 3 hammers and 7 bullish engulfing patterns. A coin has no memory of earlier results, so this shape has no cause and carries no information about what comes next.

On a chart of the past, you can find anything

On a finished chart, a pattern is plain to see. But while the shape is forming, the right side is empty. The figure below cuts the same prices off around the right shoulder. At this point there is no way to know whether price will drop below the neckline, rise back above the head, or drift sideways, and in the latter two cases the shape never earns the name head and shoulders. Just as the figure above was picked out of 20,000, if there are enough charts to look at, a textbook shape turns up somewhere. Scanning many coins and many bar lengths amounts to the same thing, and the pattern examples in books and articles also tend to be picked after the fact because they fit well. Explaining a past chart and predicting ahead of time are different things.

Confirmation bias: only the hits are remembered

Confirmation bias is the tendency to notice more readily, and remember for longer, information that fits what you already believe. Charts that moved as expected after a pattern stick in memory, while charts where nothing happened or price went the other way are quickly forgotten. With several indicators on screen, there are always reasons to expect a rise and reasons to expect a fall at the same time, which makes it easy to see only the side you want. One way to reduce this bias is to write down your call in advance and count hits and misses with equal weight.

Survivorship bias: you only hear the success stories

Stories of people who made a fortune from charts spread widely, but those who lost money with the same methods usually stay quiet. Judging a method's record only from the cases you hear about makes it look better than it really is. The same goes for coins: the coins whose charts you can look at today are the ones that survived. This course's measurements also used only 10 coins that are still traded, so they are not free of this bias either. The problems that arise when a strategy is tested without the coins that disappeared are covered in detail in the article on reading backtest results.

The idea that rules everyone knows get priced in first

If well-known rules such as the hammer or the golden cross really worked often, many people would have acted before the signal appeared. That would leave less to gain after the signal. This is not a claim that the market is always right, but the idea that a large edge is unlikely to last long in a rule that anyone can learn for free. Add fees and unfavorable fill prices, and a small edge disappears even more easily. It is not an idea proven to hold in every case, but it at least does not conflict with the candlestick pattern measurements described below.

What this course measured

This course measured prices after patterns and indicator signals on 29,946 daily bars of 10 coins on Binance (from each coin's listing to September 22, 2026). The baseline is the 5-bar up rate across all bars, 50.5%. After the nine candlestick patterns it ranged from 46.7% to 55.3%, all within 5 percentage points of the baseline, and of the four presented as bullish signals (hammer, bullish engulfing, morning star, three white soldiers), three came in below the baseline. Some indicator signals showed larger differences, but these include results that run opposite to the textbooks, such as a higher up rate after the death cross, which is supposed to be a bearish signal. Also, signals that fired on consecutive bars were each counted, so the samples are not independent of one another. The 5-day up rate for bars with RSI(14) below 30 also varied with the data: 62.3% on BTC daily bars (5 bars; baseline 53.1%) and 52.6% on BTC 4-hour bars (30 bars; baseline 52.8%). The list below is for the daily bars of the 10 coins, 5 bars later.

• 1,087 bars with RSI(14) below 30: 58.6% after 5 bars
• 1,516 closes below the lower Bollinger Band: 56.8% after 5 bars
• 88 death crosses (SMA50 crossing below SMA200): 69.3% after 5 bars
• 1,055 MACD crosses above the signal line: 50.5% after 5 bars

Where charts are still useful

Just because it is hard to predict what comes next from a chart does not mean charts are useless. A chart is the fastest summary of what has happened so far, and it lets you measure how much an asset usually moves. You can also use it to stick to a plan made in advance. If you mark on the chart the price at which you will admit your call was wrong, you can act on a promise made beforehand rather than on how you feel at the time.

• Summarizing the past: trend, range and key price levels
• Measuring risk: volatility (ATR) and the distance to your stop
• Keeping to a plan: comparing written conditions with what you actually did
• Defining where you are wrong: the price at which you drop your view

Compare your calls with a baseline

The simplest way to check whether your chart reading beats chance is to count the results. The Chart Prediction Quiz shows real historical charts with the coin name and dates hidden, has you pick whether the close 10 bars later will be higher or lower, and then shows your hit rate next to the hit rate you would have had by always picking the same answer (the guessing baseline). With only a few questions, the range of chance outcomes is wide, so it is better to play several rounds before judging. For a strategy that can be written down as rules, running it on historical bars in a backtester serves the same purpose.

What this article does not say

This article does not say that chart analysis is entirely useless, or that everyone who made money from charts was simply lucky. This course's measurements cover fixed definitions, 10 coins and one period; they did not test methods that combine several conditions, or other markets. The point is a narrow one. When a single shape or a single signal makes you feel you know what comes next, it helps to know how easily that feeling is produced, and to check with numbers whatever can be checked.