Does Technical Analysis Work? What the Research Says

Technical analysis, forecasting prices from past prices and volume, works in some tested forms and fails in others. The strongest evidence supports slow, diversified trend following across futures markets, where trends have persisted for 1 to 12 months, while simple chart rules and most intraday setups lose their apparent edge once researchers correct for data snooping (testing many rules and keeping the lucky winner) and trading costs.

Key takeaway

Technical analysis has no single verdict, but a specific rule does once it is tested out of sample, after costs and against the number of rules tried. Slow trend following across dozens of futures markets passes that test in the academic record, while most simple daily chart rules fail it and a test of 7,846 intraday rules on an S&P 500 ETF found none profitable after data-snooping correction.

What does the research say about technical analysis?

The research says technical trading rules earned profits in many speculative markets until at least the early 1990s, and that the result weakens as tests get stricter about data snooping and costs. Park and Irwin's survey in the Journal of Economic Surveys (2007) counts the modern evidence: "Among a total of 95 modern studies, 56 studies find positive results regarding technical trading strategies, 20 studies obtain negative results, and 19 studies indicate mixed results." The split is 58.95% positive, 21.05% negative and 20.00% mixed.

The 56 positive studies are weaker evidence than the count suggests, because Park and Irwin found most of the literature exposed to data snooping, rules chosen after the fact, or hard-to-estimate risk and transaction costs.

The twelve studies at the core of this article are listed in publication order; years of data are end year minus start year plus one.

Study	Market and data	Period (years of data)	Finding	Main caveat
Brock, Lakonishok & LeBaron (1992), Journal of Finance	Dow Jones index, daily; 26 moving-average and range-break rules	1897-1986 (90)	For: buy signals beat sell signals	No trading costs in headline tests
Bessembinder & Chan (1998), Financial Management	Dow, the same rules	Full sample; since 1975	Against after costs: break-even one-way cost 0.39% (0.78% round trip), 0.22% since 1975 (0.44%)	Tested the index, not a traded contract
Sullivan, Timmermann & White (1999), Journal of Finance	Dow, daily, plus S&P 500 futures; 7,846 rules	1897-1986 (90) in sample; 1987-1996 (10) out of sample	Against: best rule passes in sample, fails out of sample	Equity index data only
Lo, Mamaysky & Wang (2000), Journal of Finance	U.S. stocks; computer-detected chart patterns	1962-1996 (35)	Mixed: some patterns carry incremental information	Information, not net-of-cost profit
Osler (2000), FRBNY Economic Policy Review	Intraday FX; support and resistance levels from six firms	1996-1998 (3)	For: levels predict intraday trend interruptions	Varies by firm and currency; not a profit test
Park & Irwin (2007), Journal of Economic Surveys	Survey of 95 modern studies	Survey	Mixed: 56 positive, 20 negative, 19 mixed	Snooping, after-the-fact rule choice, cost and risk estimates
Marshall, Cahan & Cahan (2008), Journal of Empirical Finance	Intraday U.S. equity (5-minute S&P 500 ETF); 7,846 rules	2002-2003 (2)	Against: none profitable after snooping correction	One instrument, two years, not futures
Park & Irwin (2010), Journal of Futures Markets	17 U.S. futures markets, daily rules	1985-2004 (20)	Against: best rules significant in 2 of 17 markets (11.76%)	Data ends 2004
Moskowitz, Ooi & Pedersen (2012), Journal of Financial Economics	58 liquid futures; time-series momentum	1985-2009 factor data (25)	For: significant in all 58 instruments	Monthly, diversified, not intraday
Bajgrowicz & Scaillet (2012), Journal of Financial Economics	Dow, daily	1897-2011 (115)	Against: best rules not selectable in advance; low costs erase gains	Equity index only
McLean & Pontiff (2016), Journal of Finance	97 published stock return predictors	Varies by predictor	Decay: returns 26% lower out of sample, 58% lower after publication	Stock anomalies, not chart rules
Hurst, Ooi & Pedersen (2017), Journal of Portfolio Management	Trend following across global markets	Since 1880	For: positive average returns in every decade	Long-horizon diversified portfolio

Read in date order, the studies show support for simple Dow rules first (1992), stricter retests of simple rules next (1998, 1999, 2008, 2010, 2012), and post-2010 positive results (2012, 2017) only for slow, diversified trend following.

The early positive results on simple rules mostly failed later retests once data snooping and costs were counted. The positive results that held up after 2010 are about slow trend following across many futures markets.

Do moving average crossover rules work?

Simple moving-average rules looked profitable on 90 years of Dow data before costs, then failed the stricter tests that followed. Brock, Lakonishok and LeBaron tested 26 rules, moving-average crossovers (buy when a short average of prices rises above a long one) and trading-range breaks (buy when price breaks above a recent high), on the Dow Jones index from 1897 to 1986, and their Journal of Finance abstract (1992) concludes: "Overall, their results provide strong support for the technical strategies." Returns after buy signals beat returns after sell signals, but the headline tests charged no trading costs.

Sullivan, Timmermann and White (1999, Journal of Finance) asked whether those 26 rules were merely survivors of decades of experimentation. They built a universe of 7,846 rules and applied White's Reality Check, a bootstrap test of whether the best rule beats what luck alone would produce among that many tries. The best rule survived in the 1897-1986 sample, then showed no significant superior performance in 1987-1996 or in S&P 500 futures data. Bajgrowicz and Scaillet (2012, Journal of Financial Economics) extended the Dow data to 2011 and found that an investor could not have picked the future best rules in advance.

Testing many rules manufactures winners by chance. If every rule has zero true edge and each is tested independently at the 5% significance level, the chance that at least one looks like a winner is 1 - 0.95^N for N rules, and 0.05 x N false winners are expected: 5.00% for one rule, 40.13% for 10, 73.65% for the original 26 (1.3 expected), 99.41% for 100 (5 expected) and a near-certainty for 7,846 (392.3 expected). The figures are illustrative, because real rules are correlated, which lowers the effective count; the Reality Check bootstraps the actual rule universe instead. The same trap catches a trader tuning parameters on one chart, covered in how backtest overfitting misleads futures traders.

This is why the 1999 retest mattered: testing 7,846 rules guarantees hundreds of lucky winners, so the best one must beat what luck alone would produce. Correlated rules lower the effective count, but not the logic.

Do chart patterns and support and resistance work?

Chart patterns and support and resistance levels carry some measurable information, but the two best-known studies measured prediction, not profit after costs. Lo, Mamaysky and Wang (2000, Journal of Finance) used an algorithm (nonparametric kernel regression) to detect patterns such as head-and-shoulders and double bottoms in U.S. stocks from 1962 to 1996. Several patterns shifted the distribution of later returns, and the authors concluded that some technical indicators provide incremental information and may have practical value; the study did not measure whether trading the patterns beat costs.

Osler (2000, Federal Reserve Bank of New York Economic Policy Review) tested support and resistance levels that six foreign-exchange firms provided to customers in 1996-1998. The levels helped predict intraday trend interruptions (price stalling or reversing at the level) in dollar-mark, dollar-yen and dollar-pound rates for at least five business days after release, with predictive power varying by firm and currency. A level where price tends to pause is not yet a profitable strategy, a gap that level-based frameworks such as ICT and smart money concepts also have to close; whether large orders deliberately run those levels is covered in whether stop hunting is real.

Untraceable success rates

Figures claiming a chart pattern succeeds a fixed percentage of the time could not be traced to any peer-reviewed source during research for this article, so treat them as unverified.

Does trend following work in futures?

Trend following has the strongest evidence of any technical approach, but only in its slow, diversified form: positions held for months across dozens of futures markets. Moskowitz, Ooi and Pedersen's Time Series Momentum paper (2012, Journal of Financial Economics) reports: "We document significant 'time series momentum' in equity index, currency, commodity, and bond futures for each of the 58 liquid instruments we consider." Time-series momentum, a market's own past return predicting its next return, persisted for 1 to 12 months and partly reversed later; a diversified portfolio earned substantial abnormal returns over 25 years of factor data (1985-2009) and did best in extreme markets. Hurst, Ooi and Pedersen (2017, Journal of Portfolio Management) extended the evidence back to 1880, finding positive average returns in every decade and good performance in 8 of the 10 largest crisis periods, defined as the worst drawdowns of a 60/40 stock/bond portfolio.

Daily rules tested one futures market at a time fare far worse. Park and Irwin tested 17 U.S. futures markets from 1985 to 2004 with data-snooping corrections, and their Journal of Futures Markets paper (2010) concludes: "This evidence suggests that technical trading rules generally have not been profitable in the U.S. futures markets." The best rules were significant in only 2 of 17 markets (11.76%). The two results need not conflict: Park and Irwin searched many daily rules per market and penalised the search, while Moskowitz, Ooi and Pedersen measured one pre-specified effect, a market's own past return, across 58 instruments at once.

The evidence does not say "technical analysis works in futures". It says one slow, pre-specified, diversified effect works across many markets, while fast rules selected market by market mostly do not.

For a day trader the catch is the holding period: a 1-to-12-month signal does not transfer to an account that must be flat by the close, and a single-market trend signal lacks the diversification behind the published returns.

Does technical analysis work for day trading?

The evidence says no for simple intraday rules picked from a large menu, with a documented exception in futures. Marshall, Cahan and Cahan tested 7,846 filter, moving-average, support-and-resistance, channel-breakout and on-balance-volume rules on 5-minute bars of an S&P 500 ETF in 2002-2003, and their Journal of Empirical Finance abstract (2008) is blunt: "none of the 7846 popular technical trading rules we test are profitable after data snooping bias is taken into account." The sample (one instrument, two years, not a futures contract) makes the result a warning rather than a verdict on every setup.

Market intraday momentum is the documented exception. Baltussen, Da, Lammers and Martens (2021, Journal of Financial Economics) studied more than 60 futures on equities, bonds, commodities and currencies from 1974 to 2020 and found that the return from the previous close to the last 30 minutes of the session positively predicts the return in those final 30 minutes, an effect they link to hedging by option market makers and leveraged ETFs. The intraday result that survives has the same shape as trend following: one pre-specified effect with an economic cause, tested across many markets, not the best of many rules on one chart. The paper documents predictability; whether a version of it beats a given trader's costs is still that trader's test.

Most tools day traders use, such as VWAP, volume profile value areas and the opening range breakout, have far less peer-reviewed testing on futures than the daily Dow rules. Missing tests are not evidence of failure; they mean the only evidence for a specific setup is the trader's own.

Costs weigh most at intraday horizons, where a small per-trade edge pays them on every round trip. Net expectancy per round trip = (win rate x average win) - (loss rate x average loss) - round-trip cost. An illustrative rule that wins 55% of trades at +4 ticks and loses 45% at -4 ticks has a gross expectancy of 0.40 tick, so it breaks even at a round-trip cost of 0.40 tick. With commission, exchange fees and slippage totalling 0.25 tick it nets +0.15 tick per trade (+150 ticks per 1,000 trades); at 0.50 tick, -0.10 (-100 per 1,000); at 0.75 tick, -0.35 (-350 per 1,000). A single tick of slippage on one side exceeds the whole edge.

A small real edge and a small real cost are the same size. Half a tick of slippage and commission per round trip is enough to turn this rule from a winner into a loser without changing a single signal.

Why do trading rules stop working after they are published?

Published trading rules stop working for three measurable reasons: some edges were never real (data snooping), some were real but smaller than trading costs, and some are traded away once investors read about them. McLean and Pontiff measured the first and third across 97 published stock return predictors, and their study, later published in the Journal of Finance (2016), reports: "Portfolio returns are 26% lower out-of-sample and 58% lower post-publication." For an illustrative predictor earning 1.00% a month in its original sample, the decline means 0.74% after the sample ends and 0.42% after publication, a further 43.24% fall. The authors treat the 26% decline as an upper bound on data mining and attribute the extra 32 percentage points (58% minus 26%) to publication-informed trading, and predictors with higher in-sample returns declined more after publication. The predictors were stock anomalies, not chart rules, though the same mechanisms can act on any published signal.

Part of the drop is the original result being partly luck; the larger part is other people trading it once it is public. A rule you read about in a book has usually already paid out to someone else.

Trading costs consumed the classic Dow result. Bessembinder and Chan (1998, Financial Management) put break-even costs at 0.39% per one-way trade for the full sample and 0.22% since 1975 (0.78% and 0.44% per round trip), small compared with estimates of actual trading costs. Bajgrowicz and Scaillet (2012) found that even low transaction costs completely offset the rules' in-sample performance on Dow data through 2011.

How do you test whether a technical rule works?

Test the exact rule you trade on data it has never seen, with realistic costs, and count every variant you tried along the way. For an intraday futures trader, the sequence is:

Write the rule so a computer could execute it: entry, exit, stop placement, session hours and position size.
Build it on one period, then run it once on a later period you have never looked at.
Charge commission, exchange fees and a slippage allowance on every round trip, and simulate intraday fills with tick data rather than bar data.
Record how many variants you tested; the more you tried, the stronger the out-of-sample result the winner needs.
Forward test on a simulator or at minimum size before adding contracts or accounts.

Collecting enough out-of-sample trades to judge a rule is slow, and that pace sets a floor under how long it takes to become a profitable trader.

A trade copier cannot supply the evidence a rule lacks. Thor, this blog's own product, copies one master account's trades to many accounts, so copying a rule that has not passed an out-of-sample, cost-inclusive test multiplies its losses across every account instead of diluting them.

Go deeper

Frequently asked questions

Is technical analysis a scam?

No, technical analysis is a family of forecasting methods with mixed evidence, not a fraud in itself. The weak point is untested claims: a method sold without an out-of-sample, cost-inclusive record offers no evidence either way.

Is technical analysis a self-fulfilling prophecy?

The self-fulfilling explanation is a plausible hypothesis, not an established finding. Price may react where many traders watch the same level, but McLean and Pontiff's evidence on published stock return predictors points the other way: wider knowledge of a signal tended to shrink its returns, not strengthen them.

Do indicators work better on lower timeframes?

No study found during research for this article shows simple indicator rules doing better on 1-minute or 5-minute charts than on daily charts. Lower timeframes generate more signals, which raises total trading costs and the number of rule variants a trader can try, and both work against a real edge.

Is technical or fundamental analysis better for futures trading?

Neither has a proven advantage for short-term futures trading. Surveys cited by Marshall, Cahan and Cahan (2008) show traders place more emphasis on technical analysis the shorter their horizon, while the literature they reviewed had focused on long-horizon rules, so head-to-head evidence at intraday horizons is thin.

Can you make money with technical analysis?

Some traders do, but a method's popularity is not evidence that it pays for any particular trader. In a 2019 University of Sao Paulo working paper, Chague, De-Losso and Giovannetti found that 97% of people who began day trading Brazilian equity futures in 2013-2015 and persisted for at least 300 days lost money, and only 0.4% earned more than a bank teller (US$54 per day), though the study measures day traders, not technical analysis itself.

Why do chart patterns look so convincing on old charts?

Chart patterns look convincing because they are spotted after the outcome is already visible. Lo, Mamaysky and Wang (2000) automated pattern detection because identifying shapes by eye is subjective, and CFTC rule 17 CFR 4.41 requires any simulated performance of a commodity trading advisor or pool operator to carry a warning that simulated programs are designed with the benefit of hindsight.

How many trades does it take to know a rule works?

Thousands, when the edge is small. For an illustrative rule that wins 55% of trades at +4 ticks, loses 45% at -4 ticks and nets 0.15 tick after costs, the standard deviation per trade is 8 x sqrt(0.55 x 0.45), about 3.98 ticks, so (2 x 3.98 / 0.15)^2, about 2,816 trades, are needed before the average sits two standard errors above zero. A bigger edge per trade or steadier results cut that number sharply.

Does the efficient market hypothesis mean technical analysis cannot work?

No, but the weak form of the efficient-market hypothesis, which says past prices cannot earn excess returns after costs and risk, is consistent with most of the corrected results on simple rules. Time-series momentum in futures is the most durable exception, and Moskowitz, Ooi and Pedersen (2012) found speculators profiting from it at hedgers' expense.

Do professional traders use technical analysis?

Yes: Taylor and Allen (1992, Journal of International Money and Finance) reported a 1988 Bank of England survey in which at least 90% of chief foreign-exchange dealers in London placed some weight on technical analysis at one or more horizons, and systematic trend following of the kind Hurst, Ooi and Pedersen (2017) tested is a core managed-futures strategy. Widespread use is not evidence of profit, which is what the tests in this article measure.