How to Use the EMA
To use the exponential moving average, choose a length from your holding period and then decide whether it is acting as a directional filter, a signal line, or a level price reacts to. It weights recent bars more heavily than a simple average, so it turns sooner and it still lags.
An exponential moving average is an average that pays more attention to recent bars. That single property explains why people prefer it and why it does not solve the problem they hoped it would.
Before you start
A length chosen from your holding period rather than from a popular number. 20, 50 and 200 are conventions. If you hold for five bars, a 200-length average is describing something you will never experience.
A decision about whether it is a filter, a signal or a dynamic level. Those are three different uses with three different failure modes, and running all three at once means always having a reason.
The instrument’s ordinary bar range, so a touch can be compared to noise. On this site’s shared series the median bar range is 0.493 with a ninetieth percentile of 1.101.
The steps
1. Pick the length from how long you hold
A rough rule: the average should span several of your typical holding periods. Holding for 5 bars, a 20-length average describes the context; a 200-length one describes somebody else’s.
2. Use one average as a directional filter
Long only above a rising average, short only below a falling one. That removes most setups taken against the prevailing direction and requires no second line.
3. Check the slope, not just the side
Price above a flat average means nothing. The filter only works while the average is sloping, and a flattening line is the signal that the condition it filters for has ended.
4. Add a second average only if you want a signal
A crossover is an event you can act on. It is also produced constantly in a range, so it needs the first average’s slope as a filter or it fires on nothing.
5. Treat a touch as a location, not a level
Price returning to a rising average is a common place for a trend to resume. It is not a level anyone placed an order at, so it needs confirmation from something structural.
6. Do not change the length after a losing run
Adjusting the length until recent trades would have worked is fitting the indicator to the past. The length is chosen from the holding period and then left alone.
7. Take the stop from structure, not from the line
An average recalculates continuously, so a stop on it drifts with the indicator. Where the idea is wrong is fixed, and that distance sets the position size.
How to tell it worked
You can state the length and why it matches your holding period. If the answer is “it is the common one”, the length was inherited rather than chosen.
Trades taken against the average’s slope are 0 for the period. That is the filter working.
Crossovers that fired within 3 bars of each other produced no trades. The slope condition suppressed them.
And every stop came from a level rather than from the line, so it did not move when the average recalculated.
How it differs from a simple average
The weighting decays rather than cutting off. A simple average drops the oldest bar entirely; this one reduces its weight and never quite discards it, which is why the two behave differently after a large bar leaves the window.
Faster turning means more signals, and on this site’s shared series a round trip measures about 2%
of the median bar range of 0.493. More signals is not more edge — it is the same edge divided into
more pieces, each costing the same to act on. The figures are in research/series-measurements.json.
In a flat market it sits inside the noise. Price crosses it constantly, and every crossing looks identical to the one that mattered in a trend.
The efficiency ratio is the measurement that separates the two conditions. On this site’s shared series it has a median of 0.34 over ten bars with 30% of readings above 0.5 — most stretches wander rather than travel, which is exactly when an average sits inside the noise rather than beside it.
The original data
Of the 24,971 unique videos in research/search-study-corpus.jsonl, 109 have an instruction-shaped
title about this average, at a median of 18,862 views across 85 channels. Moving averages generally
appear in 105 videos at 1,681 and the crossover specifically in a subset of those. The counts come
from site/rank_howto.py, which deduplicates by video id.
109 videos at 18,862 against 105 at 1,681 for moving averages in general. Naming the specific variant carries eleven times the audience per video, which is the same pattern this site finds everywhere: specificity is what people search for.
The answer to the question on that chart depends entirely on the slope. A cross below a rising average is a pullback; the identical cross below a flattening one is a change of condition. The indicator produces the same visual event in both cases, which is why the slope check belongs before the cross rather than after it.
When it fails
The failure is a range, and it fails by producing continuous signals. Price oscillates across a flat average, every crossing looks like the one that worked in a trend, and a strategy taking them is stopped out repeatedly by ordinary movement. The average is behaving correctly throughout — it is averaging, which is all it ever claimed to do — and the condition that made its signals meaningful simply is not present.
The second failure is expecting the weighting to remove the lag. It reduces it and cannot remove it.
A third is inheriting the length. 20, 50 and 200 are conventions.
A fourth is running three uses at once. One of them will always agree with you.
A fifth is placing a stop on the line. It moves every bar.
And a sixth is changing the length after a bad run. That fits the indicator to the past.
Related
Exponential moving average explains the weighting and why old bars never fully leave. MA crossover covers the two-line signal and the filter it needs. And moving average is the general form and how the variants differ.
The useful version for me is one line as a filter rather than two lines as a signal. One average answers which side of the market I am allowed to be on, which is a question worth answering. Two averages crossing answers a question about themselves, and in anything other than a clean trend the answer changes every few bars.
— Michael Whitman
This page is educational, not financial advice. Test every idea on your own charts before risking money.