Concept

DEMA

DEMA, also known as double exponential moving average, is a Trend concept. A reference entry: the Library explains it rather than implements it.

What is DEMA?

DEMA, the double exponential moving average, is a lag-reduced average rather than an average applied twice. Patrick Mulloy introduced it in a 1994 Technical Analysis of Stocks & Commodities article, following up with the triple-stage TEMA the same year. The formula is DEMA = 2 × EMA − EMA(EMA): take the EMA of price, take a second EMA of that EMA, and combine. The gap between the single and double-smoothed EMA estimates how far the EMA trails its own input; the combination adds that gap back onto the single EMA, pulling the line closer to current price while keeping most of the smoothing.

The trade-off is the usual one for lag reduction. A DEMA of a given length turns with price sooner than the same-length EMA, but the correction term amplifies whatever price just did, so it overshoots at sharp reversals and whipsaws more readily in chop. It sits one rung below TEMA on the lag-versus-stability ladder: responsive enough to matter, stable enough to still read as a trend line.

Mulloy's construction seeded a family. Tim Tillson's T3 applies a softened version of the same correction repeatedly, ZLEMA instead de-lags the input (price plus its change over roughly half the length) before a single EMA pass, and the Hull MA reaches a similar goal with weighted averages. Adaptive designs attack the problem differently, varying effective length with conditions as an adaptive-lookback MA does, while filters such as the Ehlers SuperSmoother approach smoothing from signal processing. DEMA endures in this company because it is the simplest fix that visibly works: one extra EMA pass and one subtraction.

Two properties follow from the arithmetic. First, the correction gives some older prices negative weight, so DEMA is not confined to the recent price range: after an abrupt move it can overshoot beyond any traded price, the flip side of its speed. Second, its effective smoothing is lighter than the stated length suggests, closer to a shorter SMA or EMA, so a 20-period DEMA should be judged against faster baselines, not the 20 EMA it superficially resembles. Neither is a flaw; both are the price of lag reduction, and they matter most when markets turn hardest.

How to calculate DEMA

Two EMA passes and one combination produce it:

  1. 1Compute EMA1 = EMA of price with the chosen length N.
  2. 2Compute EMA2 = EMA of EMA1 with the same length, a double-smoothed and therefore slower line.
  3. 3Combine: DEMA = 2 × EMA1 − EMA2. The subtraction cancels a large part of the single EMA's lag.
  4. 4Plot and read like any moving average, remembering the effective smoothing is lighter than the stated length suggests.

How it's calculated

A reduced-lag moving average built from a single and a double smoothed EMA of the same price series.

α=2n+1\alpha = \frac{2}{n + 1}
EMA1⁡t=α×Pt+(1−α)×EMA1⁡prev\operatorname{EMA1}_t = \alpha \times P_t + (1 - \alpha) \times \operatorname{EMA1}_{\text{prev}}
EMA2⁡t=α×EMA1⁡t+(1−α)×EMA2⁡prev\operatorname{EMA2}_t = \alpha \times \operatorname{EMA1}_t + (1 - \alpha) \times \operatorname{EMA2}_{\text{prev}}
DEMA⁡t=(2×EMA1⁡t)−EMA2⁡t\operatorname{DEMA}_t = (2 \times \operatorname{EMA1}_t) - \operatorname{EMA2}_t
P_t: source price at bar t (default close)
t: current bar index
n: smoothing length (commonly 9, varies by platform)
alpha: EMA smoothing factor
EMA1_t: exponential moving average of price
EMA1_prev: EMA1 on the prior bar
EMA2_t: exponential moving average of EMA1
EMA2_prev: EMA2 on the prior bar
DEMA_t: double exponential moving average

Introduced by Patrick Mulloy in 1994 to reduce the lag of a standard EMA.

It is not an EMA applied twice; doubling EMA1 and subtracting EMA2 cancels most of the lag added by smoothing.

Both EMAs are commonly seeded with the first price or an n-bar SMA.

How traders use it

  • As a lower-lag trend filter: price above a rising DEMA, below a falling one, with the reduced lag surfacing regime turns a few bars earlier than the equivalent EMA and, in exchange, producing more false turns in ranges.
  • In crossover systems: fast/slow DEMA pairs or price-DEMA crosses fire earlier than their EMA counterparts; the extra signals need a trend, volatility, or structure filter to be tradable.
  • As the smoothing engine inside other tools: a MACD or Supertrend computed on DEMA inherits the earlier turn, a common way builders tune a parent indicator's responsiveness without changing its logic.
  • As a trailing reference in fast trends, where a standard EMA of comparable smoothness trails too far behind price to act as a useful dynamic level.
  • As a pullback line in established trends: touches of a DEMA are traded in the manner of dynamic S/R via MA, though a low-lag line hugs price, so touches are frequent and individually weaker as evidence.

DEMA vs neighboring moving averages

EMA: The building block. A same-length EMA is smoother but trails turning price further; DEMA subtracts an estimate of that lag, buying earlier turns at the cost of overshoot.

SMA: The equal-weight baseline: smoothest ride, longest delay, trailing a steady trend by roughly half its length. DEMA spends extra computation specifically to shrink that delay; the two bracket the responsiveness spectrum.

Ehlers SuperSmoother: A different route to low lag: a two-pole filter designed to suppress short-period noise rather than to cancel an EMA's lag after the fact. It tends to stay calmer in chop, while DEMA turns harder at genuine reversals.

Adaptive-lookback MA: Adapts the length to conditions instead of correcting a fixed length's lag: fast in trends, slow in ranges. DEMA is always fast; an adaptive average tries to be fast only when it pays.

Concept family

Trend

100 concepts mapped · 100 in the Library

DEMA FAQ

Is DEMA just an EMA applied twice?

No. An EMA of an EMA is smoother and slower than the original. DEMA uses that double-smoothed series the opposite way: it subtracts it from twice the single EMA, treating the gap between the two as a lag estimate and adding it back. The result is faster than the plain EMA, not slower, which is the entire point of Mulloy's construction.

Is DEMA better than EMA?

It is faster, not better. For the same length DEMA hugs price more closely and flags turns earlier, but it overshoots sharp reversals and generates more whipsaw in sideways markets. Which side of that trade-off helps depends on the strategy; lag reduction is a property, not an edge, and extra speed only helps a system that can absorb the extra noise.

What length should a DEMA use?

There is no canonical setting. Traders commonly reuse familiar EMA lengths, but because DEMA cancels much of the lag, it behaves roughly like a shorter EMA; some lengthen it to recover smoothness. Match the length to the swing duration being traded and test it rather than trusting a default.

What is the difference between DEMA and TEMA?

One more smoothing stage and a stronger correction. TEMA combines three EMA passes as 3 × EMA1 − 3 × EMA2 + EMA3, canceling more lag than DEMA at the cost of larger overshoot and more whip in sideways markets. DEMA is the moderate choice, TEMA the aggressive one; both trace to Mulloy's 1994 work.

Does DEMA repaint?

No. It is computed from closed-bar EMAs, so historical values are fixed once their bars close; only the live bar's value moves with incoming ticks. It does need warm-up: values near the start of the data are unreliable until both EMA stages have converged.

Is DEMA good for day trading?

It is popular on intraday charts precisely because lag hurts most there, but the same speed multiplies false turns when the session ranges. Intraday users usually pair it with a higher-timeframe trend regime label or similar filter and take signals only in that direction. Test it against a plain EMA before assuming the speed helps.

Turn DEMA into a trading strategy.

Describe your DEMA idea to Quant. It builds the strategy with you and backtests it on real data.