Skip to main content
AR(p) Model
(Autoregressive Model of order ‘p’)
Presented by: A V Chandana
Introduction
An Autoregressive (AR) model is a time series model where
the current value of a variable depends on its past values.
General form is :
Today’s value = combination of previous values + random error
This is called AR(p) because the model uses p previous
observations.
Z
Current value of the time series at time t.
Z ,Z​ ,...,Z ​
These are past values of the series.
ϕ ,ϕ ​
,...,ϕ​
These are AR coefficients (parameters).
They tell us how strongly past values influence the
present value.
p (Order of the Model)
The number p tells how many past values are used.
a
This is called the white noise error term.
It represents random factors we cannot predict.
t
t-1 t-2 t-p
1 2 p
t
Derivation of Autoregressive Model of Order p (AR(p))
Let Zt​be a stationary time series such that
E(Zₜ) = μ, Var(Zₜ) = σ ²
Let {at} be a white noise process associated with the time
series Zt which represents the random error.
Define the centered series
Z =Z −μ
so that the mean of Z​ becomes zero.
The autoregressive model of order p is defined by
where
ϕ ,ϕ ,…,ϕ​ are constants called autoregressive parameters, and
at​is white noise.
Now introduce the backshift operator B defined
as BZ =Z
Then BZ =Z and B Z =Z
Z
~
t
t
~
t
......................(1)
1 2 p
t-1
t
t
t-1 t
2
t-2
From equation (2),
Z =[ϕ(B)] a
Let
ψ(B)=[ϕ(B)] Then
Z =ψ(B)a
Thus the AR(p) model can be
expressed as a linear filter of the
white noise process.
There are p+2 unknown parameters
in the AR(p) model:
μ,σ ,ϕ ,ϕ ,…,ϕ
using the backshift operator, equation
(1) becomes
Z =ϕ BZ +ϕ B Z +⋯+ϕ B Z +a
Taking Z​ common,
(1−ϕ B−ϕ B−⋯−ϕ B )Z =a
Let
ϕ(B)=1−ϕ B−ϕ B −⋯−ϕ B
which is called the Autoregressive
operator.
Hence the AR(p) model can be written
as
ϕ(B)Z =a ..........................(2)
t t t
1 2
2 p
t t
p
~ ~
~ ~
~
t
1 2
2
p
p ~
t t
1 2
2 p
p
t t
t
~
t
-1
-1
~
t t
Z
2
1 2 p
Stationarity
AR(p) is stationary if the roots of ϕ(B)=0 lie outside the unit circle.
Variance
Autocovariance
This shows how values at different time lags relate.
Autocorrelation
Properties
Applications
Stock Market
prediction
Weather
Forecasting
Electricity
Demand
Forecasting
Sales
Forecasting
Traffic Flow
prediction
Economic
Analysis
Simple and easy to understand.
Useful for short-term forecasting.
Uses past data effectively.
Flexible by adjusting order p.
Requires stationary data.
Cannot predict sudden changes.
Difficult to choose correct order p.
Not good for long-term forecasting.
ADVANTAGES DISADVANTAGES
Example:Temperature
Prediction Suppose we want to predict today's temperature.
We observe that:
today's temperature depends on yesterday
yesterday depends on the day before
So we model:
𝑇𝑒𝑚𝑝 =0.6𝑇𝑒𝑚𝑝 +0.3𝑇𝑒𝑚𝑝 +𝑎
​
This is an AR(2) model.
Meaning:
Factor Influence
Yesterday's temperature 60%
Temperature 2 days ago 30%
Random weather variation 𝑎𝑡
Meteorologists use such models for short-term
forecasting.
t-1 t-2
t t
ThankYou