What Is a Double Exponential Moving Average?
Double exponential moving average is a moving average that reduces lag by taking an exponential average, measuring how far it trails price, and adding that gap back. It responds faster than an ordinary exponential average, which makes it earlier on real turns and earlier on false ones.
Every moving average lags, because every moving average is an average of the past. DEMA is an attempt to measure that lag and subtract it, and understanding what it actually removes matters more than the formula.
How it works
Start with an ordinary exponential moving average. It sits behind price, and the smoother you make it, the further behind it sits.
Now take an exponential average of that average. The gap between the two is a measurement of how far the first one is trailing.
Add that gap back on. Twice the first average minus the second — the line now sits much closer to price than the smoothing period alone would put it.
The trade nobody states plainly
Lag is not a defect. Lag is the smoothing. The reason a moving average is readable at all is that it ignores most of what price just did.
Remove the lag and you put the noise back. DEMA reacts to moves an ordinary average would have absorbed, and most of those moves go nowhere.
Which is why a faster average is not a better one. It is a different setting on one dial, and the right end of that dial depends on what you are trading, not on the indicator.
A worked example
Take this site’s shared series. Direction runs average 2.01 bars with a longest of 11, and the median bar range is 0.493 against a ninetieth percentile of 1.101.
A two-bar average run is the whole problem. An average fast enough to catch a two-bar move is fast enough to fire on every two-bar move, and most of them reverse.
DEMA catches the eleven-bar run earlier than an EMA does. That is a real gain and it is the case people remember.
It also fires on the two-bar runs, which is the case people forget, and there are far more of them. Each one costs a round trip — 0.0098 on this site’s series, about 2% of the median bar range.
Where it genuinely helps
In a strong, sustained trend. When a move actually runs, getting in earlier is worth real money and the false signals are rare.
As a filter rather than a trigger. Using it to decide which side of the market to look at, while taking entries from something slower, keeps the speed and drops most of the cost.
On higher timeframes. A faster average on a daily chart is still slow in absolute terms, so the noise it admits is a smaller share of the move.
And where the alternative is a very long average. If a 200-bar average is too slow to be useful, a DEMA of 50 may sit where you actually want it — which is a legitimate use of the construction rather than a claim about superiority.
The original data
On this site’s shared series: median bar range 0.493, ninetieth percentile 1.101, largest bar 2.338. Direction runs average 2.01 bars with a longest of 11. A round trip costs 0.0098, about 2% of the median bar range.
Put the run length against the cost and the arithmetic is unforgiving. An indicator that produces one extra signal per average run pays 0.0098 each time, and at two bars per run those signals arrive constantly.
And this site’s fee measurement shows the same shape over time: 5 basis points costs 1.5% of a thirty-year balance, 75 costs 20.2%. Small recurring costs decide outcomes; a faster average is a decision to pay more of them.
How it compares to the alternatives
Against a simple moving average, DEMA is dramatically faster and dramatically noisier. They are not substitutes.
Against an exponential moving average, it is moderately faster for the same period, which is the comparison most people actually mean.
Against a Hull moving average, it is slower and less aggressive — Hull applies a similar lag-reduction idea more forcefully, with the same trade amplified.
And against simply shortening the period, DEMA is smoother for the same responsiveness. That is the honest claim for it: not less lag at no cost, but a better shape for a given amount of lag removed. A shortened ordinary average becomes jagged as it speeds up, reacting to individual bars; DEMA keeps a cleaner line at the same reaction speed, which is a genuine difference and a modest one.
When it fails
The characteristic failure is choosing it for the backtest and holding it through the chop. The strategy is tested over a period containing a few strong trends, DEMA catches them earlier than the alternatives, and the results look decisive. Then the market ranges — which on this site’s series is most of the time, with direction runs averaging two bars — and the same responsiveness that won the trends produces a signal every few bars, each one paying a spread. Nothing about the indicator changed; the market stopped supplying the only condition it is better at.
A second failure is treating “less lag” as strictly better. Lag is the smoothing, and removing it returns the noise.
A third is running it at a short period, which compounds the speed until the line is close to price itself.
A fourth is comparing it to an EMA of a different length and concluding it is better, when the two were never set to the same responsiveness.
And a fifth is stacking it with other fast indicators. They all respond to the same moves, so agreement between them is not confirmation, it is repetition.
Related
Exponential moving average covers the average this is built from. Triple exponential moving average covers the same idea pushed further. And moving average covers what all of them are doing underneath.
Every indicator that claims to remove lag has made the same trade, and DEMA is the clearest place to see it. You are not getting a faster average for free — you are getting an average that is wrong sooner as well as right sooner, and which of those you notice depends entirely on the market you are in.
— Michael Whitman, from this video
This page is educational, not financial advice. Test every idea on your own charts before risking money.