zeonta.dema() — EMA with roughly half the lag, by offsetting a single EMA with its own EMA.
What it measures
A single EMA always lags, because it is, by construction, still catching up to price. DEMA estimates that lag by smoothing the EMA a second time — the gap between EMA1 and EMA2 tells you roughly how far behind EMA1 has fallen — then adds that gap back once to cancel most of it out.
Formula
DEMA = (2 x EMA1) - EMA2, where EMA1 = EMA(Close, n) and EMA2 = EMA(EMA1, n)
Parameters
Required inputs: close
| Parameter | Default |
|---|---|
length |
20 |
Returns
| Column |
|---|
DEMA_20 |
Usage
Examples run against the 300-bar OHLCV fixture in tests/data/ohlcv.csv, loaded as df. The output shown is the real output.
import pandas as pd
import zeonta
df = pd.read_csv('tests/data/ohlcv.csv', parse_dates=['date']).set_index('date')
zeonta.dema(df['close'], length=20).tail(3)
date
2024-10-25 90.218397
2024-10-26 89.975636
2024-10-27 89.653624
Name: DEMA_20, dtype: float64
Accessor form: df.zta.dema(...)
How to read it
Read it exactly like ema — trend direction, support, crossovers — but expect turns sooner: on a straight-line move DEMA carries essentially zero lag, a property ema alone never has.
Pitfalls
Cancelling lag also cancels some of the smoothing that made moving averages useful in the first place — DEMA overshoots and whips around real reversals more than ema does, especially at short lengths. It also needs roughly twice the warm-up of a plain EMA (EMA2 needs a full window of already-warmed-up EMA1 values).