Course contentsShow
← Data Science
Lesson 757 of 2,14520. Time Series ModelsFree lesson

Introduction to Exponential Smoothing

What exponential smoothing is, why it's useful, and how it differs from simple moving averages.

Introduction to Exponential Smoothing

What you'll learn: How exponential smoothing improves on simple moving averages by giving more weight to recent observations.

The Core Idea

Exponential smoothing is a forecasting technique that creates predictions by calculating a weighted average of past observations, where recent values matter more than older ones. Unlike a simple moving average that treats all observations in the window equally, exponential smoothing applies weights that decrease exponentially as you go back in time.

Think of it like your memory: yesterday's events are vivid and detailed, last week is a bit fuzzy, and last year is even hazier. Exponential smoothing mimics this by making recent data points influence your forecast much more than ancient history.

Why It's Useful

Exponential smoothing solves a key limitation of moving averages. With a simple moving average, a 10-period window treats the value from 1 period ago the same as the value from 10 periods ago—then suddenly drops that 10th value completely when it leaves the window. This creates a "cliff effect."

Exponential smoothing is smoother and more responsive because:

  • No arbitrary cutoff: All past data contributes (just with declining weight)
  • More responsive to recent changes: The most recent observation gets the highest weight
  • Computationally efficient: You only need the last forecast and the new observation

Key Differences from Moving Averages

Simple Moving AverageExponential Smoothing
Equal weights within windowDeclining weights (exponential decay)
Ignores data outside windowUses all historical data
Requires storing multiple valuesOnly needs last forecast

Key Takeaway: Exponential smoothing improves forecasting by emphasizing recent data through exponentially declining weights, making it more adaptive and efficient than simple moving averages.