CODEWITHSUNDEEP
pythonpandasalgorithmmoving-average
Ian Ash
Ian Ash

Reputation: 1212

Calculate exponential moving average as per Pandas ewm()

I am trying to build an exponential moving average algo which produces the same output as the Pandas ewm() function. In particular, I am trying to implement this approach:

enter image description here

The code below works correctly up until the point that the moving average window starts to move beyond the initial dataset, at which point I start to get different results versus the Pandas calculation.

I've been looking at this for hours now and am stumped. Can anyone point out to me how I have mis-implemented the above formula?

import numpy as np
import pandas as pd

class MovingAverages(object):
    def __init__(self, **kwargs):
        self.measures = []
        self.lookback_period = 5 
        ema_multiplier = 2 / (self.lookback_period + 1)
        self.lookback_alphas = []
        for i in range(1,self.lookback_period+1):
            self.lookback_alphas.append((1 - ema_multiplier ) ** i)

    def insert_bar(self):
        self.measures.insert(0, 0)

    def on_calculate(self, c):
        index = min(len(c), self.lookback_period+1)
        y = c[0]
        for i in range(1,index):
            y += self.lookback_alphas[i-1] * c[i]
        y /= 1 + sum(self.lookback_alphas[0:index-1])
        self.measures[0] = y

if __name__ == "__main__":
    data = [5.00,7.00,4.00,3.00,4.00,
            5.00,6.00,7.00,9.00,13.00,
            15.00,14.00,14.00,12.00,
            11.00,10.00,9.00,8.00,
            8.00,8.00,10.00,11.00,
            13.00,16.00,18.00,20.00]

    # Manually calculate exponential MA and write into list 
    ma_online = MovingAverages()
    series = []
    for d in data:
        series.insert(0, d)
        ma_online.insert_bar()
        ma_online.on_calculate(series)

    # Calculate a reference exponential MA using Pandas
    df = pd.DataFrame({'close': data})
    ma_pd = list(df.close.ewm(span=ma_online.lookback_period, adjust=True, ignore_na=True).mean())

    # Compare the two lists
    ma_online.measures.reverse()
    for i in range(len(data)):
        assert round(ma_pd[i], 2) == round(ma_online.measures[i], 2)

Upvotes: 3

Views: 4058

Answers (1)

Vincent Saulue-Laborde
Vincent Saulue-Laborde

Reputation: 1457

I'm not familiar with Python or Panda, but from my understanding of the documentation, when you use span in df.close.ewm(span=ma_online,...), you just define the decay (coefficient) in the formula, not the size of a sliding window. By default, there doesn't seem to be a fixed side window.

From another documentation of DataFrame.ewm :

span : float, optional
Specify decay in terms of span, α=2/(span+1), for span≥1

And:

See also:
rolling: Provides rolling window calculations

The rolling method is probably what you need.

Upvotes: 1

Related Questions