The Limits of Chart Analysis: Bias, Hindsight and 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. The reason for an experiment like this is that the human eye is built to find order even in randomness. Just as we see faces in clouds, we see shoulders and heads in a jagged line. So the mere fact that you saw a pattern on a chart says nothing about whether the pattern carries information. To find out, you have to count many charts with the same definition and compare the result with a baseline that has no conditions at all.

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. In practice this illusion shows up as the feeling that "it's that shape again." The moment you name a shape that is still forming, you start waiting for the ending that name promises, and signals pointing the other way become harder to see. Setting your criteria in advance helps: write down first the condition that completes the pattern, for example a close below the neckline, and do not name it until that condition is met.

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. Having several indicators on screen is not a problem in itself, but choosing them in reverse order is. If you settle on a conclusion first and then look for indicators to back it, the indicators become decoration rather than evidence. Hearing "this time is an exception" often is another common sign. If every wrong call gets an excuse attached, you end up counting only the hits. Writing down the reasoning, the price that would invalidate it and when to check in a trading journal lets you count fairly later.

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. Profit screenshots in communities and videos are the sample where this bias bites hardest. The people posting pick their good trades, and viewers remember only the ones that stand out. Without knowing how many people used the same method and how many of them made money, success stories tell you nothing about a method's record. The example charts in indicator manuals are of the same nature, since they were picked after the outcome was known.

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. Here is how this idea plays out in actual behavior. If many people are watching the same resistance line, some will sell just below it in advance, so price sometimes turns before it reaches the line. When many people buy on a breakout, the pullback right after it can be larger. So a level everyone watches is a place where a reaction is likely, not a place where the direction is settled.

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%). These figures count only the direction 5 bars after the signal, so they leave out the size of the rises and falls, and costs. 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. Measuring risk in particular is a different kind of task from calling direction. You may not know whether price will rise tomorrow, but you can measure from past bars how much this asset usually moves in a day, and knowing that range lets you set the stop distance and quantity. If you use charts not as a tool that tells you the future but as a tool that helps you make and keep a plan, the illusions above have less room to creep in. Also, price levels many people watch, such as support and resistance, are clearly useful as a reference for where to place stops and targets rather than as grounds for predicting what comes next.

• 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. The reason to compare with a baseline is that 50% is not always the mark of chance. If the questions came mostly from long rising stretches, always picking "up" scores above 50%. For example, even with a 60% hit rate, if the guessing baseline for the same questions is 58%, the share due to skill is small.

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. If you want to challenge or extend these measurements, do it the same way: define the rules in advance, count every case and compare with a baseline. If the measurements change, the conclusions of this article can change too, and that is the strength of checking with numbers. By the same token, one or two striking examples are hardly grounds for changing a conclusion. The statistics in the other articles of this course can be read by the same standard.

What looks different in crypto and stocks

The illusions form the same way, but the conditions that feed them differ by market. Crypto piles up bars 24 hours a day without a break, so there are far more charts to look at, and if you scan hundreds of coins and many bar lengths, a textbook shape always turns up somewhere. It has the same structure as the experiment above, where one path was picked out of 20,000. With stocks, the chart stops when the regular session ends and the next day's open gaps, and these gaps are often caused by information from outside the chart, such as earnings releases or news. Reading the gap the day after an earnings release from Samsung Electronics or SK Hynix as the result of a chart pattern misidentifies the cause. The daily price limit on Korean stocks (±30%) clips moves and changes the shape of candles, and US indexes and large tech stocks trade in extended hours, so a regular-session chart misses part of the move. Volume means something different too: crypto volume is tallied separately on each exchange, so one exchange's chart does not represent the whole market. And this course's measurements were taken on daily crypto bars, so there is no basis for assuming they carry over to stocks unchanged.

A practical checklist

Going through the steps below when you form a view from a chart filters out the illusions above one by one. Live charts add one more illusion. A bar that is still forming has no settled shape yet, so a bar that looked like a hammer intrabar can close as an ordinary bearish candle, and an indicator signal that lit up can vanish when the bar closes. A chart that shows only past signals never records these vanished signals, so signals look more reliable than they really are. Making and recording calls on closed bars is more accurate. When counting your calls with the quiz or a journal, include only calls made on closed bars; that keeps you from looking at a past chart and believing you knew it at the time. Even if you cannot follow every step below each time, keeping at least ④ and ⑤ leaves you something to review later. Count first whether you followed the steps you set, before counting whether the call was right.

• ① Can this shape be explained by a definition fixed in advance?
• ② Has the condition that completes the pattern been met, or is it still forming?
• ③ Did you write down the evidence for the other direction with equal weight?
• ④ Have you set the price at which you would consider your call wrong?
• ⑤ Did you record the call so it can be counted later?
• ⑥ Did you confirm the signal on a closed bar rather than a forming one?