Limitations of technical analysis: what the research really shows

Technical analysis promises something attractive: that the record of prices and volume contains clues about where prices go next. Generations of traders have built careers on that promise, and every charting platform, Finabase included, ships dozens of indicators and patterns to help. It is fair to ask the uncomfortable question directly: does it work?

The honest answer is "less than its fans claim, more than its harshest critics allow, and mostly not in the way people think." This guide lays out the case against technical analysis, the efficient-market argument and the problem of data snooping, summarizes what the best-known academic studies actually found, runs a small experiment with 1,000 random trading rules on SPY, and ends with the places where charts genuinely help: risk management and discipline.

The efficient-market argument

In 1970 Eugene Fama organized decades of research into the efficient-market hypothesis. Its weak form, the version that matters here, says current prices already reflect everything contained in past prices and volume. If that is true, a chart pattern cannot reliably predict returns, because anyone who could see the pattern would already have traded on it.

Early tests were not kind to chartists. In 1966 Fama and Marshall Blume tested "filter rules," buying after a rise of x% and selling after a fall of x%, on the Dow stocks. Some small filters beat buy and hold before costs, but the trading they required erased the gains once commissions were charged. Earlier still, in 1959, Harry Roberts showed that a series built from random numbers produced price charts with "head-and-shoulders" formations and trends that looked just like real ones.

The weak form is a strong claim, and it has well-known limits. Sanford Grossman and Joseph Stiglitz pointed out in 1980 that perfectly efficient markets are impossible: if prices already contained all information, nobody would be paid to gather it, and so nobody would. Andrew Lo's adaptive-markets view (2004) suggests efficiency varies over time as traders, strategies and conditions change. Both leave room for patterns that appear, get exploited and fade.

What the research found

The literature on technical trading is large. A few studies are cited often enough to know by name.

Brock, Lakonishok and LeBaron is the paper most often cited in favor. On almost a century of Dow Jones Industrial Average data, simple rules such as "price above its 200-day average" separated periods with higher average returns from periods with lower ones, and statistical models of random returns could not reproduce the difference. They tested the index itself, which cannot be traded directly, and did not focus on transaction costs.

Sullivan, Timmermann and White took that result seriously and asked how much of it came from the search. Using a bootstrap method that accounts for testing thousands of rule variants on the same data, they found the best rule's performance was not just luck in the original sample. But in the period after it, starting in 1987, the best rule no longer outperformed. The pattern was real once and then mostly disappeared.

Lo, Mamaysky and Wang replaced the human eye with an algorithm that detects patterns such as head-and-shoulders and double tops. Several patterns were associated with slightly different return distributions afterward, which suggests charts contain some information. That is a statement about statistics across thousands of stocks, not a trading system.

Park and Irwin surveyed the field in 2007. Most modern studies reported positive results, but the authors stressed that many suffered from data snooping, after-the-fact selection of rules and difficulty estimating costs. Their overall reading was that profits had been found in some markets and periods, and tended to shrink in more recent data.

A separate line of research is more favorable to one technical idea in particular: trend following. Moskowitz, Ooi and Pedersen (2012) found that an asset's own past 12-month return predicted its next month's return across dozens of futures markets, which they called time-series momentum. That is a systematic version of "trade with the trend," tested across many markets and decades, and it is closer to a moving-average rule than to reading candlestick patterns.

Data snooping: 1,000 random rules on SPY

The single most important limitation is not whether patterns exist. It is that anyone searching enough rules will find some that worked in the past by chance. Harvey, Liu and Zhu argued in 2016 that because hundreds of factors had been tested on the same data, a new result should clear a much higher statistical bar than the usual one, roughly a t-statistic of 3 instead of 2.

To see the problem directly, we generated 1,000 trading rules with no information at all. Each rule decided at random each month whether to hold SPY or cash, with the odds set so it was invested about 75% of the time and switched about once a year, much like the moving-average rule tested in Backtesting basics: biases, costs and an honest SPY test. Each switch costs 0.10%, trades happen at the next month's open, and cash earns nothing. We measured every rule against buy and hold (dividends included) from March 1995 to December 2009, then ran the same rules from January 2010 to October 2026.

None of these rules contains any market insight. Yet one in five beat buy and hold in the in-sample period, which included two deep bear markets in which merely being out of the market some of the time helped. The luckiest beat it by more than five points a year, more than the best moving-average look-back on the Backtesting basics: biases, costs and an honest SPY test page managed in the same years (+4.4 points). An investor shown only that rule's in-sample record would have had no way to tell it was luck. Out of sample, in a strong bull market, it lagged by almost four points a year, about the same as the average random rule.

This does not prove that moving-average rules are random. It shows that an in-sample record, even a long one, cannot by itself separate skill from luck once many alternatives have been tried. Every indicator setting, chart pattern and filter a trader experiments with adds to the count, whether or not they write it down.

Our brains are pattern detectors

Even a disciplined test cannot fix the way humans read charts.

• We see patterns in noise. Roberts's random charts made the point in 1959. A random walk wanders into trends, double bottoms and triangles regularly.
• Hindsight makes signals look obvious. On a finished chart the turning points are clear. In real time, the same bar could be the start of a reversal or another wobble.
• We remember hits. The divergence that called the top is memorable. The five that came before it, while price kept rising, are forgotten. See Divergence: regular and hidden, bullish and bearish, with RSI and MACD for a count of how rare strictly defined divergences are.
• Labels are flexible. "It broke support but held the 61.8% level" can explain any outcome. A method that can explain every outcome predicts none.

The Trading psychology: loss aversion, overconfidence and other biases page covers how these biases show up in real decisions.

Why some patterns may still work for a while

Not everything is noise. There are plausible reasons for some technical effects.

• Self-fulfilling attention. When many traders watch the same moving average or round number, their orders cluster there and can move price briefly. The effect is real but small and short, and it shrinks as more people try to exploit it.
• Slow information. Investors underreact to news and catch up over weeks or months, one explanation offered for momentum and trend-following returns.
• Forced and flow-driven trading. Index rebalancing, margin calls and fund flows move prices for reasons unrelated to value, sometimes in predictable ways.

Each of these is an argument for a small, unstable edge that has to be tested across many markets, survives costs only with care, and may disappear. None is an argument that a particular chart pattern on a particular stock predicts its next move.

Where technical analysis genuinely helps

The strongest case for technical tools is not prediction. It is structure.

• Defining risk before entry. A chart shows where a trade idea is proven wrong, the break of a swing low or a key level, which gives a logical stop. Combined with Position sizing: how many shares to buy for a set risk, that turns an opinion into a bounded bet. See also Stop-loss placement: volatility, structure and time stops.
• Discipline over emotion. A written rule, such as "exit if the weekly close falls below the 40-week average," removes the decision from the moment when fear or hope is strongest. Even a mediocre rule followed consistently can beat an inconsistent one.
• Trend filters for drawdown control. Simple trend rules have historically reduced exposure in long bear markets, at a cost in returns and whipsaws, as the SPY test in Backtesting basics: biases, costs and an honest SPY test shows. That is a risk preference, not free money.
• Execution. Volume, spreads and intraday ranges help choose when and how to trade, regardless of how the decision was made. See Liquidity and slippage: volume, depth, spread and market impact.
• A common vocabulary. Support, trend and breakout give traders a precise way to write down a plan and review it later in a journal.

How Finabase approaches this

Finabase gives you the same indicators and patterns as most platforms, and the Learn library tries to be honest about them. Indicator and pattern pages report real results from our own data, including failures: the Hammer candlestick pattern: meaning, rules and reliability page, for example, tests the pattern on 18 large US stocks and ETFs and finds that hammers were higher five sessions later about as often as any random day. The technical gauge on symbol pages and in the Scanner summarizes where each indicator stands; it is a description of the chart, not a forecast.

If you want to test an idea, use Backtesting, and read its results with the cautions in Backtesting basics: biases, costs and an honest SPY test: trades fill at the signal bar's close, costs are set to zero, and short ranges contain few trades. The most useful thing a backtest can do is show you a rule's worst periods before you live through them.

Frequently asked questions

Does technical analysis work?

Some technical effects, such as momentum and trend following, have support across many markets and decades, but they are modest, inconsistent and costly to trade. Most popular chart patterns and indicator signals have weak or unproven predictive power once data snooping and costs are accounted for. Technical tools are most clearly useful for managing risk and enforcing discipline.

If markets are efficient, why do professionals use charts?

Many use them for execution and risk control rather than prediction, to decide where to enter, where to exit and how much to trade. Some systematic funds trade trend-following strategies across many markets, where small effects can add up. And markets are not perfectly efficient, so small, temporary patterns can exist.

Is fundamental analysis better than technical analysis?

They answer different questions. Fundamental analysis estimates what a business is worth; technical analysis studies how its price behaves. Both face the same hurdle: public information is widely known, so any edge has to come from better analysis, discipline or patience. Many investors use fundamentals to choose what to own and charts to manage when and how much.

How can I tell whether a strategy is just lucky?

Count how many variations you tried, test on data you did not use while designing, check it on other markets and periods, charge realistic costs, and require many trades. A strategy that only works with one setting on one stock over one period is probably luck.

Summary

Technical analysis faces three serious limits: markets absorb public information quickly, searching many rules guarantees lucky winners, and human pattern recognition finds order in noise. The research is genuinely mixed: some simple rules showed real effects in older data, trend following has broad support, but many results shrink or vanish out of sample and after costs. Our 1,000 random rules on SPY show how convincing pure luck can look. Use charts to define risk, size positions and enforce discipline, test any predictive claim honestly, and treat every signal as a probability, not a promise.