Learn Additive vs Multiplicative Decomposition

Swipe to show menu

Understanding how to break down time series data into its underlying components is crucial for effective analysis. Two primary approaches are additive decomposition and multiplicative decomposition. In an additive model, the components—trend, seasonality, and residual—combine through simple addition. The relationship is described as:

\text{Time series} = \text{Trend} + \text{Seasonality} + \text{Residual}

This model is appropriate when the seasonal fluctuations and noise remain roughly constant over time, regardless of the trend. For instance, if monthly sales increase by about 100 units every December, regardless of whether the overall sales trend is rising or falling, the additive approach is suitable.

In contrast, a multiplicative model assumes the components interact through multiplication:

\text{Time series} = \text{Trend} × \text{Seasonality} × \text{Residual}

Use the multiplicative model when the magnitude of seasonal variation grows or shrinks in proportion to the trend. For example, if sales double during the holidays only when the overall sales trend is high, and the seasonal effect scales with the level of the series, a multiplicative decomposition better captures this pattern.


              12345678910111213141516171819202122232425262728293031323334353637
            
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

# Simulate time series data
np.random.seed(42)
periods = 72  # 6 years of monthly data
time = np.arange(periods)
trend = 10 + 0.5 * time

# Additive seasonal component
seasonal_additive = 5 * np.sin(2 * np.pi * time / 12)
# Multiplicative seasonal component
seasonal_multiplicative = 1 + 0.3 * np.sin(2 * np.pi * time / 12)

# Additive model: constant seasonality
additive_series = trend + seasonal_additive + np.random.normal(0, 2, periods)

# Multiplicative model: seasonality grows with trend
multiplicative_series = trend * seasonal_multiplicative * (1 + np.random.normal(0, 0.05, periods))

# Plot both series
fig, axs = plt.subplots(2, 1, figsize=(10, 8), sharex=True)

axs[0].plot(time, additive_series, label="Additive Series")
axs[0].set_title("Additive Seasonal Pattern")
axs[0].set_ylabel("Value")
axs[0].legend()

axs[1].plot(time, multiplicative_series, label="Multiplicative Series", color="orange")
axs[1].set_title("Multiplicative Seasonal Pattern")
axs[1].set_xlabel("Time (Months)")
axs[1].set_ylabel("Value")
axs[1].legend()

plt.tight_layout()
plt.show()

Choosing between additive and multiplicative decomposition depends on the characteristics of your data. If the amplitude of the seasonal pattern remains steady as the trend changes, the additive model is usually best. If the seasonal effect becomes more pronounced as the data grows or shrinks, the multiplicative model captures this scaling behavior more accurately. Always visualize your data and consider the nature of both the trend and seasonality before selecting a decomposition approach.

Everything was clear?

Thanks for your feedback!

Section 1. Chapter 7

Ask AI

Ask anything or try one of the suggested questions to begin our chat

Section 1. Chapter 7