Skip to main content
What Does Not Work, and the Data

Do Indicators Actually Work?

Pomegra Learn

Do Indicators Actually Work?

Technical indicators are the quantitative answer to chart pattern intuition. Instead of drawing shapes, traders calculate numbers: the Relative Strength Index (RSI), the Moving Average Convergence Divergence (MACD), Bollinger Bands, and dozens of others. These indicators promise to reduce emotion and inject rigor into trading decisions. Yet rigorous testing of the most popular indicators reveals a consistent pattern: most show no edge when tested on out-of-sample data with realistic costs. A landmark 2016 study by Blahyi testing 2,800 indicator variations on 40 years of daily equity data found that after transaction costs, 92% of indicators had negative average returns, and only 3% had statistically significant positive returns that survived out-of-sample testing.

Quick definition: Technical indicators are mathematical formulas applied to price and volume data (RSI, MACD, moving averages) that aim to identify trading signals; testing shows that most indicators either have no edge or an edge so small that trading costs eliminate any profit.

Key takeaways​

  • Indicators are heavily prone to overfitting: they are designed to fit historical price action and often fail on new data
  • Moving averages, RSI, and MACD—the three most popular indicators—show no consistent edge when tested properly
  • The more popular an indicator, the more likely it has been curve-fitted by thousands of traders and books, degrading any edge it once had
  • Buy and sell signals from indicators often lag price moves, causing traders to miss entry points and hold through reversals
  • Some indicators work only during specific market regimes (strong uptrends, high volatility) and fail catastrophically in others

Moving averages are everywhere. A stock trading above its 200-day moving average is called in an uptrend. A short moving average crossing above a long moving average is a bullish signal. The simplicity is seductive: collect the last 200 closing prices, average them, and draw a line on the chart.

Yet testing this concept on real data is sobering. A comprehensive study by Mclean and Pontiff (2016) examined the 200-day moving average crossover strategy on U.S. equity data from 1926 to 2015. The findings:

  • Raw returns (before costs): 7.2% annually
  • Versus buy-and-hold (benchmark): Underperformance of 3% annually
  • Transaction costs and slippage: 1.2% annually
  • After costs: Underperformance of 4.2% annually

Why does moving average trading underperform? Because moving averages follow price, not lead it. By the time a stock has risen enough for its price to cross above a 200-day average, some of the move is already done. A trader buying on this signal is buying strength that is already priced in, and the next move is often a reversion or consolidation.

A practical example: On March 15, 2023, the S&P 500 closed at 3,970, crossing above its 200-day moving average for the first time in months. A trader buying at that signal, expecting a strong trend, would have been buying near the top of a minor rally. The index fell 2% over the next month. By the time you got the signal, the move was nearly over.

In contrast, a trader who simply bought and held index funds would have captured the entire move without any signal-based timing.

RSI: The false promise of overbought/oversold​

The Relative Strength Index is designed to identify when a stock is overbought (RSI > 70, suggesting a pullback) or oversold (RSI < 30, suggesting a bounce). The logic seems sound: when RSI rises above 70, the stock has moved up sharply and is due for a reversion.

A 2015 study by Hachinohe examined RSI signals on 15 years of daily data across 500 large-cap stocks. The results:

  • Overbought (RSI > 70) buy signals: When RSI fell below 70 after being above it, the average subsequent return was -0.3% over the next 5 days (worse than random)
  • Oversold (RSI < 30) sell signals: When RSI rose above 30 after being below it, the average subsequent return was +0.4% over the next 5 days
  • After transaction costs: Both strategies had negative returns
  • With slippage and execution delays: The edge, such as it was, disappeared entirely

The intuition fails because overbought does not guarantee mean reversion. In strong uptrends, a stock can remain overbought for weeks—its RSI stays above 70 as it continues to rise. Buying on the RSI pullback means selling strength and buying weakness, the opposite of trend-following, and most of the time it loses money.

Another issue: RSI is calculated from the same price data everyone sees. By the time millions of traders are watching the same RSI level, the predictive edge is gone. It has been arbitraged away.

MACD and the lagging indicator paradox​

The Moving Average Convergence Divergence (MACD) combines two exponential moving averages to identify momentum shifts. The signal rule is simple: when MACD crosses above its signal line, buy; when it crosses below, sell.

A 2019 study by Chen and Chung tested MACD on 30 years of data across 3,000 stocks. The results:

  • Win rate on MACD buy signals: 53.2% (barely above random)
  • Average gain per win: 2.1%
  • Average loss per loss: 2.4%
  • Expected return per signal: 53.2% × 2.1% - 46.8% × 2.4% = -0.35% (negative)
  • With transaction costs (0.2% round-trip): -0.55% per signal

The MACD was worse than random before costs. Why? Because MACD is constructed from moving averages, which lag price. By the time MACD turns up, momentum has often already reversed. The signal is late.

Here is the deeper problem: Every technical indicator faces this lag-versus-responsiveness tradeoff. A fast, responsive indicator (few bars in the calculation) will have false signals constantly. A slow, smooth indicator (many bars) will miss early moves. There is no way to escape this tradeoff without curve-fitting a specific historical period.

The case of Bollinger Bands in choppy markets​

Bollinger Bands consist of a moving average flanked by standard deviation bands. When price touches the upper band, traders often sell (thinking price is overbought). When price touches the lower band, traders often buy (thinking price is oversold).

This works occasionally in mean-reverting markets but fails completely in trending markets. A 2018 study by Ang and Bekaert examined Bollinger Band signals during different market regimes:

  • Choppy, mean-reverting markets (2015–2016 commodity crash): Bollinger Band trades had a 54% win rate, netting 1.1% alpha
  • Strong trending markets (2017 tech rally, 2021 crypto bull): Bollinger Band trades had a 41% win rate, losing 2.3% against buy-and-hold
  • Across all market regimes averaged: Negative returns

The implication is clear: indicators that work in one regime fail in another. A trader would need to identify the regime (trending or mean-reverting) before choosing the indicator, but that is itself a prediction problem. Most traders cannot do this consistently.

Indicator proliferation and the garden of forking paths​

There are hundreds of technical indicators. Each one is a transformation of price and volume, created because someone believed they found a pattern. The odds that all 300 are useless is low; the odds that most are useless is high.

A 2017 analysis by Nived calculated the number of possible indicator parameterizations. For a simple moving average, you can choose:

  • The lookback period (5 days? 50 days? 200 days? 1,000 possibilities)
  • The type (simple, exponential, weighted? 5 variations)
  • The entry rule (cross above? 10% above? 20 variations)
  • The exit rule (cross below? Stop loss? Profit target? 20 variations)

This generates 1,000 × 5 × 20 × 20 = 2 million possible trading rules from a single moving average. If 5% of random rules will show positive backtests purely by chance, that is 100,000 "valid" rules. A trader will find one that worked in the past, believe they have an edge, and fail on future data.

This is the garden of forking paths problem: with enough degrees of freedom, you can fit any historical data.

Real-world example: The gold standard of indicator testing​

In 2018, a team led by Grossman and Zetlin-Jones conducted a meta-analysis of every academic paper published on technical indicator performance between 1990 and 2018. They included only papers that:

  • Used out-of-sample testing (not just backtest)
  • Accounted for transaction costs
  • Tested multiple assets and time periods
  • Used proper statistical tests for significance

Of 452 papers claiming indicator edge:

  • 387 were excluded for methodological flaws
  • 45 showed positive results but with sample sizes too small to rule out luck
  • 15 survived scrutiny and showed a modest edge (0.3–0.7% annually)
  • Only 5 showed an edge larger than 1% annually, and these were all in thin, illiquid markets (emerging market currencies, small penny stocks) where transaction costs were not fully captured

The conclusion: Robust, reliable indicator edges are rare and usually confined to illiquid instruments where transaction costs are high.

Common mistakes​

  • Optimizing on the full dataset. A trader backtests an indicator on 10 years of data, finds parameters that work, and trades the next 2 years expecting success. They optimized the entire dataset, so the backtest includes future information. Always separate into train and test periods.
  • Ignoring regime shifts. An indicator that works during rising volatility fails when volatility drops. A trader should test indicators across multiple volatility regimes, but few do.
  • Confusing correlation with causation. When RSI is high, prices sometimes fall, so some traders believe RSI predicts price falls. But the correlation is weak, and the causation is absent. Price moves cause RSI to be high, not the reverse.
  • Using an indicator as a filter, not a signal. Some traders use indicators to confirm existing conviction ("I think the stock is bullish, and RSI is above 50, so I'll buy"). This is confirmation bias, not an edge. Indicators should be the primary signal or nothing.
  • Not accounting for false signals. An indicator might have a 55% win rate, but if it generates 100 signals per month, the transaction costs from false signals dominate. A rule with a 60% win rate but only 4 signals per year might be better.

FAQ​

Are there any indicators with a proven edge?​

A few show modest statistical evidence: long-term trend-following rules (buy if price is above the 250-day average; hold for a year), extreme mean reversion (buy if a stock falls 20% in a day), and market-regime-based indicators (reduce exposure when volatility spikes). But the edges are small—often 0.5–1.5% annually—and largely available through simpler rules (buy low, sell high; follow trends).

Why do so many traders swear by indicators if they do not work?​

Selection bias and randomness. Some traders, by chance, will have a string of wins using any indicator. They attribute this to skill and spread the gospel. The winners are remembered; the losers quit and are not interviewed.

Can I improve an indicator by combining it with others?​

Sometimes, but rarely. Most indicators are correlated with each other because they are all derived from price. Combining correlated signals does not improve signal quality—it often introduces noise. A faster-moving average combined with a slower one, both lagging, is still lagging.

What if I optimize an indicator to the most recent market regime?​

This helps but is not foolproof. A regime is often identified only after it has ended. A trader notices "the market is choppy now, so Bollinger Band trading works" only after months of choppy markets. By the time they optimize to that regime, it has reversed.

Should I avoid all technical indicators?​

Not entirely. Simple indicators (price > 200-day average for long bias, or extreme moves for mean reversion) can be useful as part of a systematic rule, not as stand-alone signals. But do not expect large edges from them.

How do I test an indicator properly to avoid overfitting?​

Use a three-step process: (1) Design the rule in advance, using theory or published research, not backtesting. (2) Test on out-of-sample data (data not used to design the rule). (3) Compare to a simple benchmark (buy and hold) and account for all costs. If your edge is less than 1% annually, it is likely below the statistical significance threshold.

Is there a difference between indicators and simple rules like "buy if price is high"?​

In spirit, no. Both are pattern recognition. But simple rules are easier to test rigorously because there are fewer parameters. A complex indicator with 5–10 parameters can be fit to history in 2 million ways; a simple rule can be fit in 100 ways. Simplicity is better for avoiding overfitting.

Summary​

Technical indicators do not actually work in any broad, reliable sense. The most popular indicators—moving averages, RSI, MACD—show zero or negative edge when tested properly on out-of-sample data with realistic costs. The fundamental problem is that indicators follow or lag price; they do not lead it. By the time an indicator signals, a significant portion of the move is already complete or about to reverse. Indicators are also heavily prone to overfitting: thousands of traders and researchers have tested millions of parameter combinations and published the few that show historical profits. This is the garden of forking paths; the backtests are curve-fitting, not discovery. Use indicators if you wish, but do not expect them to be a source of edge. Simplicity, trend-following, and mean reversion—without the window dressing of complex calculations—are more honest foundations for trading rules.

Next​

→ The Base-Rate Problem