How to Use ewm in pandas for Exponential Weighted Calculations

Use pandas.DataFrame.ewm() or pandas.Series.ewm() to create an exponential weighted object, then apply functions like mean() to get exponentially weighted moving averages. This method gives more weight to recent data points, useful for smoothing time series data.

Syntax

The ewm() method creates an exponential weighted object on a Series or DataFrame. You can specify parameters like span, com, or halflife to control the decay of weights.

• span: Specify the decay in terms of span, higher means slower decay.
• com: Center of mass, another way to specify decay.
• halflife: Time for the weight to reduce by half.
• adjust: If True (default), weights are adjusted for imbalance.
• ignore_na: If True, ignores missing values in calculation.

After calling ewm(), use aggregation functions like mean(), std(), or var() to get results.

python

df.ewm(span=span_value, adjust=True, ignore_na=False).mean()

Example

This example shows how to calculate the exponential weighted moving average (EWMA) of a simple numeric series using span=3. The EWMA gives more weight to recent values, smoothing the data.

python

import pandas as pd

# Create a simple data series
data = pd.Series([1, 2, 3, 4, 5, 6, 7, 8, 9, 10])

# Calculate EWMA with span=3
ewm_result = data.ewm(span=3, adjust=False).mean()

print(ewm_result)

Output

0     1.000000
1     1.500000
2     2.250000
3     3.125000
4     4.062500
5     5.031250
6     6.015625
7     7.007812
8     8.003906
9     9.001953
dtype: float64

Common Pitfalls

Common mistakes when using ewm() include:

• Confusing adjust=True and adjust=False. adjust=True calculates weighted averages with weights normalized, while adjust=False uses recursive calculation which is faster but slightly different.
• Not handling missing values properly. Use ignore_na=True to skip NaNs in calculations.
• Using incompatible parameters together, like setting both span and halflife at the same time.

Example of wrong and right usage:

python

import pandas as pd

s = pd.Series([1, None, 3, 4, None, 6])

# Wrong: default adjust=True includes NaNs affecting results
wrong_ewm = s.ewm(span=2).mean()

# Right: ignore_na=True skips NaNs
right_ewm = s.ewm(span=2, ignore_na=True).mean()

print('Wrong EWMA with NaNs:')
print(wrong_ewm)
print('\nRight EWMA ignoring NaNs:')
print(right_ewm)

Output

Wrong EWMA with NaNs:
0    1.000000
1         NaN
2    3.000000
3    3.500000
4         NaN
5    6.000000
dtype: float64

Right EWMA ignoring NaNs:
0    1.000000
1    1.000000
2    2.000000
3    3.000000
4    3.000000
5    4.500000
dtype: float64

Quick Reference

Parameter	Description	Example
span	Decay in terms of span; higher means slower decay	ewm(span=3)
com	Center of mass, alternative decay control	ewm(com=0.5)
halflife	Time for weight to reduce by half	ewm(halflife=2)
adjust	If True, weights are normalized; if False, recursive calculation	ewm(adjust=False)
ignore_na	If True, skips NaN values in calculation	ewm(ignore_na=True)

Key Takeaways

Use pandas ewm() to calculate exponential weighted statistics giving more weight to recent data.

Choose decay parameter like span, com, or halflife to control how fast weights decrease.

Set adjust=False for faster recursive calculation or adjust=True for normalized weights.

Use ignore_na=True to handle missing data properly in calculations.

After ewm(), apply aggregation functions like mean(), std(), or var() to get results.