Does MACD Work? What the Evidence Says About 12/26/9

📅 Originally Published: May 21, 2026 · Last Updated: July 23, 2026

MACD is simple enough to feel objective. A charting platform subtracts a 26-period exponential moving average from a 12-period average, smooths that difference with a 9-period signal line, and marks the crossovers. TradingView and MetaTrader both describe 12, 26, and 9 as the standard inputs for the indicator.

The calculation is straightforward. Trouble starts when a visible crossover is treated as an investable edge. A preset tells you how the software is configured, while a fair benchmark, transaction costs, a new sample, and a different market determine whether the rule deserves confidence.

For standard MACD, this guide separates three questions that are often blended together. A standard MACD preset is only the starting specification: whether standard MACD has worked in direct tests, what broad technical-rule research can and cannot say about MACD, and what evidence you would need before letting a crossover change your portfolio.

What Happened When 12/26/9 Was Tested Across Five Markets

The most relevant paper in this article is a direct test of MACD(12,26,9), not a study of technical analysis in general. Chong, Ng, and Liew examined five developed-market indices from 1976 through 2002. The standard setting showed no predictability in Italy, Canada, the United States, or Japan. Germany’s DAX produced a statistically significant negative buy-minus-sell spread.

The table gives a narrower answer than either side of the MACD debate usually wants. It does not show that every MACD application loses money. It shows that the common 12/26/9 crossover did not produce a consistent signal across these five indices and this historical sample. Four results were statistically indistinguishable from zero. The only significant result went in the wrong direction.

The result rejects the idea of a universal default while leaving other MACD designs, assets, frequencies, and confirmation methods open to separate testing.

Primary source: Terence Tai-Leung Chong, Wing-Kam Ng, and Venus Khim-Sen Liew, “Revisiting the Performance of MACD and RSI Oscillators,” Journal of Risk and Financial Management, 2014. Read the paper.

My Reproducible SPY Check: One Rule, One Window, No Tuning

I also ran a deliberately simple replication on a pinned SPY close series from February 23, 2010 through December 30, 2019. The rule held SPY when the prior day’s MACD line was above the prior day’s signal line and otherwise held cash at 0%. With no parameter search and a one-day execution lag, buy-and-hold compounded at 13.80% a year. The MACD rule compounded at 6.15% before costs and 5.08% after a 5-basis-point charge on each position change.

The split windows matter. A strategy can look weak because one market regime dominates the sample. Here, the long-or-cash rule lagged in both halves as well as the full period. It spent about half the trading days invested, so it gave up a large share of a strong equity advance. Lower drawdowns did not make up for the return gap in this particular test.

Treat this original analysis as a reproducibility check rather than universal proof. It covers one U.S. ETF, one decade, one signal definition, and no interest on cash. The script omits taxes, bid-ask spreads, market impact, and alternative MACD rules. Its value is showing the full rule and its limits instead of turning one favorable chart into a law.

For a fuller explanation of execution lags, costs, benchmarks, and data snooping, see how to build a technical-analysis backtest that can survive basic scrutiny. The difference between simple and exponential averages also matters when you define a rule, as explained in SMA versus EMA crossover testing.

The 7,846-Rule Study Is Strong Evidence, but Not a MACD Test

Bajgrowicz and Scaillet tested 7,846 simple technical rules on the Dow Jones Industrial Average from 1897 to 2011. Their modern subperiods, 1962–2011, produced no positive abnormal performance even before transaction costs. That strongly undercuts claims of easy, persistent alpha from common public rules, while leaving MACD(12,26,9) untested by this particular study.

The distinction is important because the rule universe contained five named categories: filter rules, moving averages, support and resistance, channel breakouts, and on-balance-volume averages. The paper did not claim to test every indicator built from moving averages, and it did not list standard MACD as one of the 7,846 rules.

Its main benchmark also was not simply buy-and-hold. The authors focused on excess returns and Sharpe ratios relative to being out of the market and earning the risk-free rate. They discussed buy-and-hold as an alternative comparison, but that was not the paper’s primary statistic.

What the paper does establish is still useful. When researchers examine thousands of rules, some will look impressive by chance. The false discovery rate procedure was designed to separate genuine performance from those lucky results. In the more recent samples, the researchers found no positive performance under zero transaction costs, and their out-of-sample rule-selection exercise was negative in most recent periods.

The paper’s own conclusion is appropriately limited: investors should be wary of common technical indicators presented as obvious money-making tools. It also states that the results say little about other markets, different frequencies, or more sophisticated rules.

Primary source: Pierre Bajgrowicz and Olivier Scaillet, “Technical Trading Revisited: False Discoveries, Persistence Tests, and Transaction Costs,” Journal of Financial Economics, 2012. Read the paper.

Volatility-Sorted Portfolios Are a Different Strategy

Han, Yang, and Zhou found that a moving-average timing strategy performed strongly on portfolios sorted by volatility, especially high-volatility portfolios, even after transaction costs. That result is relevant because it shows that technical-rule performance can depend on the assets to which a rule is applied. It does not validate standard MACD or turn high-volatility stock selection into a simple retail fix.

The study used a moving-average strategy on volatility-sorted portfolios. MACD uses two exponential moving averages plus a signal line, so the trading rule is different. The portfolio construction is also part of the result. A trader would need a defined stock universe, a volatility-ranking method, reconstitution rules, transaction-cost assumptions, and an out-of-sample test. Choosing a few volatile stocks by reputation is not the same experiment.

The useful question is narrower: which rule, on which assets, against which benchmark, and under what costs? Change the rule, universe, sample, or rebalancing schedule and the result can change with it.

Primary source: Yufeng Han, Ke Yang, and Guofu Zhou, “A New Anomaly: The Cross-Sectional Profitability of Technical Analysis.” Read the paper record.

Before MACD Can Earn a Place in Your Process

A MACD signal should not control a trade until you can state the rule in advance, compare it with the right benchmark, include realistic frictions, and show that it survives data it was not tuned on. If those conditions are missing, the honest status of 12/26/9 is “untested for this use,” not “proven” or “disproven.”

1. Define the action, not only the indicator

“Use MACD” is not a strategy. Decide what happens after a bullish crossover, a bearish crossover, or a zero-line break. Specify whether the portfolio goes long, moves to cash, shorts, reduces position size, or merely records the signal. Specify when the trade executes. A same-close signal and trade usually contains look-ahead bias.

2. Match the benchmark to the decision

A long-term investor considering whether to exit an index fund should compare the rule with buy-and-hold, including dividends and the return on cash. A tactical manager may also care about drawdown, volatility, and exposure. A short-term trader needs spreads, slippage, and taxes. One benchmark cannot answer every decision.

3. Keep the test outside the tuning sample

If you try dozens of settings and publish the best one, you have created the same multiple-testing problem the 7,846-rule study warns about. Choose the rule before the final test. Use a separate validation period or walk-forward process. Do not keep changing the parameters after each disappointing result.

4. Test stability, not one winning chart

Break the result into different periods and market regimes. Check nearby settings such as 10/24/8 and 14/30/10. A genuine effect should not disappear when one input moves slightly. A strategy that works only at one exact parameter combination is more likely to be fitted to noise.

5. Decide how much authority the signal receives

A chart overlay can be observational, advisory, or binding. Write that distinction into your process. For many long-term investors, the safest role is observational: MACD can describe recent momentum without automatically changing the allocation. A written investment policy statement can prevent an attractive chart signal from overriding a plan that was built for a different horizon.

MACD FAQ

Bottom Line: Decide What the Signal Is Allowed to Do

The default MACD setting is a reasonable chart convention and a weak basis for an automatic trade. The direct five-market evidence does not show a consistent edge, the broad 7,846-rule study does not directly test MACD, and the reproducible SPY check in this article lagged buy-and-hold in both halves of its sample.

Before the next crossover, write down four items: the exact action, the benchmark, the cost assumptions, and the out-of-sample period. If any one is missing, keep the signal observational. That boundary is more useful than changing 12/26/9 to another preset and hoping the chart improves.

Educational quantitative analysis based on published data. Not investment, tax, or legal advice. Consult a licensed professional before acting on any calculation. About TheFinSense.