What a moving average actually is

A moving average is the average price over the last N periods, recalculated each bar. Its only job is to strip out noise so the underlying direction is visible.

Two variants dominate:

• Simple moving average (SMA) weights every one of the last N prices equally. A 200-day SMA is just the mean of the last 200 closes.

• Exponential moving average (EMA) weights recent prices more heavily and older ones progressively less, so it turns sooner after a change in direction.

Traders argue endlessly about which is better, but both are averages of prices that have already printed, so both lag by construction: the SMA rolls old prices out of a fixed window, the EMA simply fades them. Choosing EMA over SMA sets how fast you react, not which one is right.

The conventional lengths carry conventional meanings:

• 9 and 20 for short-term momentum, used by intraday and swing traders.

• 50 as the medium-term trend, the classic "is this pullback normal" line.

• 100 and 200 as the long-term regime filter. Institutions watch the 200 closely enough to make it a widely shared reference point.

Why it is always late, and the failure mode nobody warns you about

A moving average is an average of prices that have already printed, exponential ones included, so it cannot turn until the underlying move is already underway. The lag is arithmetic rather than a tuning problem: shorter averages and EMAs turn sooner, which is why traders treat length as a speed-versus-noise dial, but the signal always arrives after the move it describes. The research, though, does not endorse the usual assumption that longer is therefore safer, as the next section shows.

That trade-off produces the practical failure mode:

A moving average is a trend-following tool, so a crossover system's results vary with both the settings you choose and the stretch of market you measure them over. The documented reasons backtested moving-average performance gets overstated, though, are methodological rather than a matter of which market mood you happened to sample: data-snooping, ignored transaction costs, and in at least one published case a backtest run with look-ahead bias. That is the subject of the next section.

What a century of testing actually found

Moving averages have been studied more rigorously than any other indicator, and the story arrives in three acts.

Act one: it worked. In 1992 Brock, Lakonishok and LeBaron tested moving-average and trading-range rules on the Dow from 1897 to 1986 and concluded that "our results provide strong support for the technical strategies" (Brock, Lakonishok & LeBaron, 1992). Days after buy signals returned more than days after sell signals, and were less volatile too. Returns following sell signals were actually negative. Note carefully: those are raw signal returns, quoted with no deduction for the cost of trading them.

Act two: it survived the obvious objection, then failed the real one. The objection is data-snooping: test thousands of rules and a few will look brilliant by chance alone. Sullivan, Timmermann and White expanded the original 26 rules to 7,846 and re-ran them over a century of daily Dow data with a bootstrap built to price in exactly that problem (Sullivan, Timmermann & White, 1999). Over the original 1897 to 1986 window the finding held: the best-performing rules still beat the benchmark after the data-snooping correction, in all four of the original subperiods. That is narrower than it sounds, since most rules do not clear such a correction, but it is a genuine pass.

What they could not do was keep working. Tested out of sample on the following decade, 1987 to 1996, the best rule was no longer statistically significant. The rule that looked best as of the end of 1986, a five-day moving average, went on to earn 2.8% with a p-value of 0.322. And where the snooping adjustment did bite, it bit hard: a rule earning almost 10% a year on S&P 500 futures scored a p-value of 0.04 assessed on its own and 0.90 once you accounted for the universe of rules it had been selected from.

Act three: costs, and the modern record. The best rule over the full century earned 17.17% annualised, but it did so across 6,310 trades, roughly 63 a year. That pace implies a break-even transaction cost of just 0.27% per trade, a threshold the authors concede was probably exceeded in the early part of the sample.

Later work on the mature indices is bleaker. Hsu and Kuan widened the search to 39,832 rules over 1990 to 2000 and found no statistically significant profitable rule for either the Dow or the S&P 500, though the same search did find significantly profitable rules in the younger NASDAQ Composite and Russell 2000 data (Hsu & Kuan, 2005). Valeriy Zakamulin, who wrote the scholarly book on this (Zakamulin, 2017), puts it flatly: "the performance of market timing strategies is highly overstated, to say the least," and attributes the gap to data-mining plus ignored market frictions (Zakamulin). In one case he traced a celebrated set of "too good to be true" moving-average results to a backtest run with look-ahead bias (Zakamulin, 2018).

The golden cross problem

Here is the detail that should change how you read the famous signal. When researchers searched tens of thousands of rules, the winners were not the 50/200 golden cross. They were extremely short averages: a two-day moving average for the Russell 2000, a two-day average with a small band for the NASDAQ, and two-day or five-day rules for the Dow. The lengths that dominate financial media are not the lengths that won the tests.

The popular golden-cross statistics deserve the same scepticism. Figures like "the market gains an average of 9.9% in the year after a golden cross" circulate constantly, but no peer-reviewed primary source for them surfaced in the research behind this piece, and the pages carrying them are blog-tier. That absence is the point: unless you can trace a figure to a source that states its sample, period and baseline, a bullish-sounding average means very little.

Reputation versus evidence

Using moving averages honestly

• Use them as a regime filter, not a trigger. "Price is above a rising 200" is a useful statement about context. "The 50 crossed the 200, therefore buy" is a mechanical rule with a poor out-of-sample record.

• Match the length to your holding period. A day trader reading a 200-day average is reading a line with nothing to say about their timeframe.

• Expect whipsaws and count them. Before trading a crossover, ask what it would have done through a three-month range, not just through the trend on your screen.

• Charge yourself realistic costs. The academic edge died at 0.27% per trade. If your rule trades often, fees and slippage are not a footnote, they are the outcome.

• Do not stack averages and call it confirmation. Five moving averages agree because they are five transforms of one price, exactly as Part 13 argued. That is one opinion, repeated.

Common mistakes

• Trading the cross mechanically - the signal arrives after the move, and its record out of sample is weak.

• Assuming 50/200 are special - they are conventions, and the tested winners were far shorter.

• Reading a backtest as a promise - the best-looking rule on history is the one most likely to have been fitted to it.

• Treating the average as support - it is a calculated line, not a level where orders actually sit.

• Ignoring the market regime - the same rule that prints money in a trend bleeds steadily in a range.

Moving averages on ApeX Omni

ApeX Omni's charts carry the standard moving averages alongside the price action this series has taught you to read. The sensible use on a perpetuals venue is as a regime filter: let a longer average tell you whether the market is trending or ranging, then size and place your risk accordingly, taking the actual entry from structure, levels and volume. A crossover system that whipsaws six times in a range is an irritation at 1x and an account event at 20x. Trade the regime on ApeX Omni, and keep your stop where the structure says, not where the average happens to sit.

The bottom line

A moving average is the most honest indicator on the chart and the most over-promoted signal in trading, which is a strange thing to be at once. As a description it is excellent: one line that answers "which way has this market been going" without pretending to know anything else. As a trigger it carries a century of testing that reads like a cautionary tale, an edge that was real in the sample where it was discovered, survived the statistical correction that should have killed it, and then quietly stopped working in the years that followed, with transaction costs waiting underneath the whole time. Use the line to see the trend. Do not ask it to tell you when.

Next in this series: RSI, the momentum oscillator everyone reads as an overbought and oversold signal, and what happens to that reading in a market that refuses to cool down.

Frequently asked questions

What is a moving average in trading? It is the average price over a set number of recent periods, redrawn on each new bar, plotted to smooth out noise so the trend direction is easier to see.

What is the difference between SMA and EMA? A simple moving average weights all periods in its window equally; an exponential moving average weights recent prices more heavily, so it reacts sooner. Both still lag, because both average prices that have already happened, so the choice is about reaction speed rather than correctness.

Is the golden cross a reliable buy signal? The evidence is much weaker than its fame suggests. The best performers in the large academic searches were very short averages of two to five periods, not the 50/200 crossover, and the widely quoted golden-cross return statistics trace to secondary write-ups rather than peer-reviewed studies.

Do moving-average trading rules actually work? They worked on Dow data from 1897 to 1986 and survived correction for data-snooping over that window. They were not statistically significant out of sample in 1987 to 1996, and later work found nothing significant for the Dow or S&P 500 in the 1990s. Realistic transaction costs erode what remains.

Which moving average period is best? There is no universally best length. Shorter averages react faster and generate more false signals; longer ones are more reliable and later. Match the length to your holding period.

Why does price seem to bounce off the 200-day moving average? Partly attention, since many participants watch it, and partly coincidence in a trending market. The peer-reviewed order-flow evidence for price reacting at levels concerns round numbers, not moving averages, so treat the bounce as a tendency rather than a rule.

Explore more from this series: Part 4: Support and Resistance | Part 5: Trend and Trendlines | Part 13: Do Indicators Actually Work? | Part 15: RSI Explained (Coming Soon)

This article is for educational and informational purposes only and is not financial, investment, or legal advice. Do your own research before making any trading decision.