Technical analysis is the most popular analytical framework in retail trading. Moving averages, RSI, MACD, Bollinger Bands, Fibonacci retracements, head-and-shoulders patterns — the toolkit is vast, the user base is enormous, and the debate about whether any of it works has been running for over sixty years.

We went through the research. Not blog posts or YouTube tutorials — the actual peer-reviewed studies, published in journals like The Journal of Finance, Econometrica, and Management Science. Here is what they found.

The academic starting point: markets are efficient

The modern case against technical analysis begins with Eugene Fama. In 1965, Fama tested whether past stock prices could predict future stock prices. His conclusion: serial correlations in daily returns were too small to exploit. The past history of a price series could not increase expected profits above a simple buy-and-hold strategy.

In 1970, Fama formalised this into the Efficient Market Hypothesis (EMH). Under even its weakest form — that current prices already reflect all past price information — technical analysis should have no predictive value. Chart patterns, moving average crossovers, and oscillator signals should be noise, not signal.

By the late 1970s, the academic consensus was clear: if markets were even weakly efficient, technical trading rules were seeing patterns where none existed.

The strongest counterargument: Brock, Lakonishok, and LeBaron (1992)

The most significant challenge to that consensus came from Brock, Lakonishok, and LeBaron in 1992. They tested two simple technical rules — moving average crossovers and trading-range breakouts — on nearly 90 years of Dow Jones Industrial Average data.

Their finding: buy signals produced average daily returns significantly higher than sell signals. The differences could not be explained by standard statistical models, including GARCH. The paper used bootstrap methods to rule out common null hypotheses.

It was a landmark result. For the first time, a rigorous study suggested that simple technical rules contained genuine predictive information.

But three problems would emerge.

Problem 1: data snooping

If you test 1,000 random trading rules on the same dataset, roughly 50 will appear "statistically significant" at the 5% level. That is not a sign of predictive power. It is arithmetic.

Sullivan, Timmermann, and White (1999) tested Brock et al.'s findings using a procedure called White's Reality Check — a bootstrap method designed to ask: does the best rule in a large universe of rules truly outperform, or is it just the luckiest survivor?

Their answer was nuanced. Some of the original rules survived the correction for the original DJIA sample. But the evidence weakened substantially when extended to more recent data.

Hansen (2005) refined the method further with the Superior Predictive Ability (SPA) test. Applications of the SPA test to technical trading rules found even less evidence of genuine predictive ability.

Problem 2: transaction costs

Bajgrowicz and Scaillet (2012) ran the most comprehensive test to date. They applied False Discovery Rate methodology to over 7,800 technical trading rules on DJIA data spanning 1897 to 2011.

Some rules appeared significant even after correcting for multiple testing. But persistence tests showed that an investor could not have selected the future best-performing rules in advance. And when even modest transaction costs were introduced — 0.1% per round trip — all apparent profitability disappeared.

This is the finding that matters most for practitioners. Technical strategies tend to trade frequently. Each trade incurs commissions, bid-ask spreads, slippage, and market impact. For most technical rules, these costs exceed the gross alpha the rule generates.

Problem 3: the rules stop working

Park and Irwin (2007) surveyed 95 modern empirical studies on technical analysis profitability. Of these, 56 reported positive results, 20 reported negative results, and 19 were mixed. That sounds encouraging — until you read the fine print.

Many of the positive findings were concentrated in earlier time periods. The rules that appeared to work on 1970s and 1980s data failed on 1990s and 2000s data. Park and Irwin noted that the positive studies were "subject to various problems in their testing procedures," including data snooping and difficulty estimating transaction costs.

Timmermann and Granger (2004) offered a theoretical explanation: any documented predictive pattern will attract capital until the pattern disappears. If a moving average crossover strategy generates excess returns, money flows in, the edge compresses, and eventually the strategy returns nothing. Lo (2004) formalised this as the Adaptive Markets Hypothesis — inefficiencies exist, but they are transient.

What about chart patterns?

Lo, Mamaysky, and Wang (2000) developed the first rigorous statistical framework for testing chart patterns. Using nonparametric kernel regression on US stocks from 1962 to 1996, they found that certain patterns (head-and-shoulders, double bottoms) showed statistically different return distributions compared to the unconditional distribution.

But there is an important distinction: statistical significance is not economic profitability. The patterns' predictive content was modest, unevenly distributed, and — when tested in real-time trading simulations — insufficient to cover transaction costs and slippage.

The broader problem: the factor zoo

The issues with technical analysis are not unique to technical analysis. Harvey, Liu, and Zhu (2016) catalogued 316 factors published in top finance journals that claimed to predict stock returns. They argued that the traditional statistical significance threshold (t-statistic above 2.0) was far too low given how many factors had been tested. They proposed raising the bar to 3.0 or higher.

Their conclusion: a substantial fraction of published "anomalies" in finance are likely false discoveries. The same logic applies to technical trading rules. If hundreds of indicator combinations are tested on the same dataset, finding a few "winners" by chance is expected, not surprising.

So why does everyone still use it?

Menkhoff (2010) surveyed 692 fund managers across five countries. The vast majority used technical analysis. At horizons of a few weeks, it was the most important analytical tool — more important than fundamental analysis.

The research suggests several reasons for this persistence:

• Self-fulfilling prophecy. If enough traders believe a support level is real and place orders there, the level may hold. Not because it has inherent significance, but because collective belief creates the order flow. This mechanism is real but fragile.
• Cognitive biases. Humans are wired to see patterns in random data. We remember the times a pattern "worked" and forget the times it did not. Technical analysis provides a narrative that satisfies our need for causal explanation. A random walk provides no narrative, which is psychologically unsatisfying even when it is more accurate.
• Risk management framing. Some practitioners use technical levels not as predictions but as reference points for stop-loss placement and position sizing. This descriptive use is not addressed by the profitability literature.
• Survivorship bias. Traders who succeed become visible through books and social media. Those who fail exit quietly. The visible survivors create a misleading impression of how well technical analysis works in aggregate.

What the evidence does and does not say

The evidence does say:

• Simple technical trading rules have not been shown to produce consistent, risk-adjusted profits after realistic transaction costs across multiple time periods and asset classes.
• The majority of positive findings are vulnerable to data snooping, multiple testing bias, and in-sample overfitting.
• Apparent profitability tends to decay over time.

‍

The evidence does not say:

• That no trader has ever profited from technical analysis. Individual skill, timing, and risk management can produce profits irrespective of the tools used.
• That price data contains no information. Volatility clustering, momentum, and mean reversion are real phenomena — but exploiting them after costs is a separate question.
• That all indicators are equally useless. Some describe observable market properties (like realised volatility). Others (like Fibonacci retracements) have no theoretical or empirical basis.

A different question: what can price data actually describe?

The research is clear that using price data to predict where price will go is, at best, unreliable. But there is a different question: can price data describe where price has been?

That is a lower bar, and one that is testable. A calibrated expected range does not claim to predict reversals, breakouts, or trend changes. It measures realised volatility and projects a statistical range around price. The range is descriptive — it shows where price has recently traded relative to its own behaviour.

The difference matters because a descriptive claim can be audited. You can count how often price stayed inside the range. You can publish the containment rate. You can show the misses. If the claim is wrong, the data will show it.

Most technical indicators do not do this. They make implicit predictions — a "buy signal," a "sell signal," "overbought," "oversold" — without specifying what those terms mean in testable, statistical terms. That is why the academic literature finds them wanting: they are not wrong in a falsifiable way. They are wrong in an unfalsifiable way, which is worse.

‍

The Behavioral Transform Model (BTM) is a calibrated expected-range indicator for TradingView. On the S&P 500 daily timeframe from 1928 to 2024, the inner band contained the next close about 71% of the time across roughly 24,000 trading days. The outer band contained about 94%. The full calibration data — including the misses and crisis-onset periods — is published at oisigma.com/proof.

‍

Past behaviour is not a guarantee of future results.

‍

30-day free trial. $15/month after that, cancel anytime. Have a look at the data and decide for yourself.

Sources

The studies cited in this article are listed chronologically below. All are peer-reviewed and published in major academic journals.

Fama, E.F. (1965). The Behavior of Stock-Market Prices. The Journal of Business, 38(1), 34–105.

Fama, E.F. and Blume, M.E. (1966). Filter Rules and Stock-Market Trading. The Journal of Business, 39(1), 226–241.

Fama, E.F. (1970). Efficient Capital Markets: A Review of Theory and Empirical Work. The Journal of Finance, 25(2), 383–417.

Brock, W., Lakonishok, J. and LeBaron, B. (1992). Simple Technical Trading Rules and the Stochastic Properties of Stock Returns. The Journal of Finance, 47(5), 1731–1764.

Sullivan, R., Timmermann, A. and White, H. (1999). Data-Snooping, Technical Trading Rule Performance, and the Bootstrap. The Journal of Finance, 54(5), 1647–1691.

Lo, A.W., Mamaysky, H. and Wang, J. (2000). Foundations of Technical Analysis: Computational Algorithms, Statistical Inference, and Empirical Implementation. The Journal of Finance, 55(4), 1705–1765.

White, H. (2000). A Reality Check for Data Snooping. Econometrica, 68(5), 1097–1126.

Lo, A.W. (2004). The Adaptive Markets Hypothesis. The Journal of Portfolio Management, 30(5), 15–29.

Timmermann, A. and Granger, C.W.J. (2004). Efficient Market Hypothesis and Forecasting. International Journal of Forecasting, 20(1), 15–27.

Hansen, P.R. (2005). A Test for Superior Predictive Ability. Journal of Business & Economic Statistics, 23(4), 365–380.

Park, C.-H. and Irwin, S.H. (2007). What Do We Know About the Profitability of Technical Analysis? Journal of Economic Surveys, 21(4), 786–826.

Savin, G., Weller, P. and Zvingelis, J. (2007). The Predictive Power of "Head-and-Shoulders" Price Patterns in the U.S. Stock Market. Journal of Financial Econometrics, 5(2), 243–265.

Menkhoff, L. (2010). The Use of Technical Analysis by Fund Managers: International Evidence. Journal of Banking & Finance, 34(11), 2573–2586.

Bajgrowicz, P. and Scaillet, O. (2012). Technical Trading Revisited: False Discoveries, Persistence Tests, and Transaction Costs. Journal of Financial Economics, 106(2), 473–491.

Neely, C.J., Rapach, D.E., Tu, J. and Zhou, G. (2014). Forecasting the Equity Risk Premium: The Role of Technical Indicators. Management Science, 60(7), 1772–1791.

Harvey, C.R., Liu, Y. and Zhu, H. (2016). ... and the Cross-Section of Expected Returns. The Review of Financial Studies, 29(1), 5–68.

‍