Course contentsShow
← Data Science
Lesson 773 of 2,14520. Time Series ModelsFree lesson

Introduction to ARIMA: Components and Notation

Understand ARIMA(p,d,q) notation and how autoregressive, differencing, and moving average components combine.

Introduction to ARIMA: Components and Notation

What you'll learn: You'll understand what ARIMA models are and how their three components work together to forecast time series data.

What is ARIMA?

ARIMA stands for AutoRegressive Integrated Moving Average. It's a powerful forecasting method that combines three distinct components to model time series patterns. Think of ARIMA as a toolkit where you mix and match three tools depending on your data's characteristics.

The Three Components

AR (AutoRegressive) - p:
This component uses past values of the series itself to predict future values. The parameter p tells you how many past observations (lags) are used. It's like saying "today's weather depends on the weather from the past p days."

I (Integrated) - d:
This represents the number of times you need to difference the data to make it stationary (remove trends). The parameter d is the differencing order. If your data has a trend and you difference it once to remove that trend, then d=1.

MA (Moving Average) - q:
This component uses past forecast errors to improve predictions. The parameter q indicates how many past error terms are included. It's like adjusting your prediction based on how wrong you were in the previous q forecasts.

ARIMA(p,d,q) Notation

When you see ARIMA(1,1,2), it means:

  • p=1: Use 1 past observation (AR component)
  • d=1: Apply first-order differencing once
  • q=2: Use 2 past forecast errors (MA component)

The beauty of ARIMA is its flexibility. You can turn components "off" by setting them to zero. An ARIMA(2,0,0) is purely autoregressive, while ARIMA(0,1,1) combines differencing with moving average errors.

Key Takeaway: ARIMA(p,d,q) notation captures three key aspects of time series behavior: autoregression from past values (p), stationarity through differencing (d), and error correction from past mistakes (q).