EMA vs SMA
The exponential moving average weights recent bars more heavily, so it turns sooner. The simple version treats every bar in its window equally, so it lags more and moves less. Speed against stability is the whole of the difference between them.
Two ways of averaging the same prices. One gives recent bars more weight; the other treats every bar in the window the same. Everything that follows — speed, lag, how often it turns — comes from that one choice, and it matters less than the period does.
What each one is
The simple moving average adds the last N closes and divides by N. Every bar in the window counts equally, and a bar leaving the window has the same effect as one entering. Simple moving average covers it.
The exponential version weights recent bars more heavily. Older bars never fully drop out; their influence decays. Exponential moving average covers that calculation.
Both are lagging by construction. An average of past prices cannot lead price, and neither claims to — the question is only how far behind each one sits.
Where they differ
Speed. The exponential version turns sooner because the newest bars dominate. On this site’s shared series direction runs average 2.01 bars, so a faster line reacts to a great deal that reverses.
What happens when a bar leaves the window. On the simple average, an old extreme dropping out can move the line noticeably even if today was quiet. The exponential version has no such edge effect.
How widely watched each is. The long simple averages are the ones a great many people plot, which makes those specific levels partly self-fulfilling. The exponential equivalent has no comparable crowding behind it.
How often they signal. Faster means more crossings, more entries and more round trips — about 2% of the median bar range of 0.493 each, on this site’s shared series.
Where they agree
Both lag, and neither predicts. They describe where price has been. Any apparent forecast is the reader’s, not the indicator’s.
Both are dominated by the period. A 20 and a 200 differ far more from each other than an exponential 20 differs from a simple 20 — which is why the type is the smaller decision.
Both whipsaw in a range. Neither can distinguish a trend from a sideways market, so both produce a run of crossings when there is nothing to follow.
And neither contains a stop. The line is a statistic, not a level anybody is defending.
Which one to use
Use the simple average when the level itself is the point. If you are watching whether price holds a long, widely-plotted average, the crowding is the mechanism — the same argument that makes standard pivot points worth more than an unusual variant.
Use the exponential version when you want a faster directional filter and can accept more turns. Shorter holding periods, where being late costs more than being wrong occasionally.
Set the period first and argue about the type afterwards, if at all. Matching the period to your holding horizon changes the indicator’s behaviour far more than switching formulas ever will.
And run one, not both. Two averages of the same prices at the same period agree almost always, and the agreement reads as confirmation while being arithmetic.
What no setting fixes
The trade-off between earlier and wronger. Every increase in speed buys earlier signals and pays for them with more false ones. Hunting for a version that avoids this is fitting to whichever history you are looking at.
And neither sees participation. A clean crossing on an instrument nobody trades looks identical to one on an instrument everybody does.
Choosing the period, which is the real decision
Match it to how long you hold. A position measured in days is not informed by an average of the last two hundred, and a position measured in years is not informed by an average of the last ten.
The widely used lengths carry a crowding effect the unusual ones do not. The long round numbers are plotted by an enormous number of people, which is what makes those particular levels worth something beyond the arithmetic.
Test it on data you did not choose it from. Any period can be made to look excellent over a selected stretch, and that is the same fitting problem every parameter carries.
Then leave it alone. Changing the period after a losing run produces a setting fitted to the last few trades, and the next stretch of market will be the kind the new value handles worst.
The original data
Of the 24,971 unique videos in research/search-study-corpus.jsonl, 6 compare the two directly in the
title, at a median of 21,872 views across 6 channels. Moving averages generally appear in far more,
and the comparison outperforms most of them per video. The counts come from site/rank_compare.py and
site/corpus_count.py.
6 videos at a 21,872 median. A modest number of direct comparisons with a solid audience each — and almost none of them make the point that the period is the larger decision, which is the one thing that would change what a reader does next.
The answer to the question on that chart is that turning earlier is what it does, on every move including the ones that reverse. The faster line is not more informed — it has simply weighted the last few bars more heavily, and those bars have not finished being ambiguous.
When it fails
The failure is switching type after a losing run, and it never settles anything. The simple average was too slow, so the exponential one goes on. That version turns early on a choppy stretch and produces a string of false signals, so the simple one comes back. Each switch is made immediately after the weakness of the current setting has been demonstrated, which guarantees the new one will be tested by the opposite condition — and the period, which was doing most of the work throughout, never got looked at.
The second failure is running both. They are one measurement twice.
A third is arguing type before setting the period. That is the larger lever.
A fourth is stopping on either line. Neither is a level.
A fifth is expecting a faster line to lead. It weights recent bars; it does not anticipate.
And a sixth is using an unusual variant for a crowded level. The crowding was the point.
Related
Exponential moving average covers the weighted version. Simple moving average covers the equal-weighted one. And moving average crossover is the system built on either of them.
The period does more than the type ever will. Arguing about which average is better while both are set to whatever the platform suggested is arguing about the small decision — a 20 and a 200 behave completely differently whichever formula draws them.
— Michael Whitman
This page is educational, not financial advice. Test every idea on your own charts before risking money.