zeonta.mad() — Rolling median absolute deviation: a spread measure robust to outliers.
What it measures
A spread measure like stddev, but built from medians instead of means and squares at every step — the same robust-to-outliers idea behind using a median instead of a mean in the first place, applied twice over.
Formula
MAD = median(|Close - median(Close, n)|, n)
Parameters
Required inputs: close
| Parameter | Default |
|---|---|
length |
20 |
Returns
| Column |
|---|
MAD_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.mad(df['close']).tail(3)
date
2024-10-25 0.45280
2024-10-26 0.52970
2024-10-27 0.62205
Name: MAD_20, dtype: float64
Accessor form: df.zta.mad(...)
How to read it
Reads the same direction as stddev — rising means the window has gotten choppier — but a single wild bar barely moves MAD, while it can dominate stddev outright.
Pitfalls
Not the same thing as the mean absolute deviation cci uses internally, despite the similar name — that one averages the deviations, this one takes their median, and the two disagree whenever the window has any outliers at all.
Reference
Formula source: https://en.wikipedia.org/wiki/Median_absolute_deviation