Scribd
Upload
22 views30 pages

Chapter 2

This document discusses autoregressive (AR), moving average (MA), and autoregressive moving average (ARMA) time series models. It shows how to generate simulated AR, MA, and ARMA processes in R and examines their autocorrelation (ACF) and partial autocorrelation (PACF) functions. The document also demonstrates how to estimate AR, MA, and ARMA models using maximum likelihood in R and check the fit through residual analysis. Model selection can be based on comparing Akaike information criterion (AIC) and Bayesian information criterion (BIC) values for different models. Poorly fitting models will exhibit non-random patterns when checking standardized residuals, sample ACF of residuals, normal Q-Q

Uploaded by

Abdellah Chaoui
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
22 views30 pages

Chapter 2

This document discusses autoregressive (AR), moving average (MA), and autoregressive moving average (ARMA) time series models. It shows how to generate simulated AR, MA, and ARMA processes in R and examines their autocorrelation (ACF) and partial autocorrelation (PACF) functions. The document also demonstrates how to estimate AR, MA, and ARMA models using maximum likelihood in R and check the fit through residual analysis. Model selection can be based on comparing Akaike information criterion (AIC) and Bayesian information criterion (BIC) values for different models. Poorly fitting models will exhibit non-random patterns when checking standardized residuals, sample ACF of residuals, normal Q-Q

Uploaded by

Abdellah Chaoui
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 30

AR and MA models

ARIMA MODELS IN R

David Stoffer
Professor of Statistics at the University
of Pittsburgh
AR and MA Models
x <- arima.sim(list(order = c(1, 0, 0), ar = -.7), n = 200)
y <- arima.sim(list(order = c(0, 0, 1), ma = -.7), n = 200)
par(mfrow = c(1, 2))
plot(x, main = "AR(1)")
plot(y, main = "MA(1)")

ARIMA MODELS IN R
ACF and PACF
AR(p) MA(q) ARMA(p, q)
ACF Tails off Cuts off lag q Tails off
PACF Cuts off lag p Tails off Tails off

ARIMA MODELS IN R
ACF and PACF
AR(p) MA(q) ARMA(p, q)
ACF Tails off Cuts off lag q Tails off
PACF Cuts off lag p Tails off Tails off

ARIMA MODELS IN R
ACF and PACF
AR(p) MA(q) ARMA(p, q)
ACF Tails off Cuts off lag q Tails off
PACF Cuts off lag p Tails off Tails off

ARIMA MODELS IN R
ACF and PACF
AR(p) MA(q) ARMA(p, q)
ACF Tails off Cuts off lag q Tails off
PACF Cuts off lag p Tails off Tails off

ARIMA MODELS IN R
Estimation
Estimation for time series is similar to using least squares for
regression

Estimates are obtained numerically using ideas of Gauss and


Newton

ARIMA MODELS IN R
Estimation with astsa
AR(2) with mean 50:
Wt = 50 + 1.5(Xt−1 − 50) − .75(Xt−2 − 50) + Wt

x <- arima.sim(list(order = c(2, 0, 0),


ar = c(1.5, -.75)),
n = 200) + 50
x_fit <- sarima(x, p = 2, d = 0, q = 0)
x_fit$ttable

Estimate SE t.value p.value


ar1 1.5429 0.0435 35.4417 0
ar2 -0.7752 0.0434 -17.8650 0
xmean 49.6984 0.3057 162.5788 0

ARIMA MODELS IN R
Estimation with astsa
MA(1) with mean 0:

Xt = Wt − .7Wt−1

y <- arima.sim(list(order = c(0, 0, 1), ma = -.7), n = 200)


y_fit <- sarima(y, p = 0, d = 0, q = 1)
y_fit$ttable

Estimate SE t.value p.value


ma1 -0.7459 0.0513 -14.5470 0.0000
xmean 0.0324 0.0191 1.6946 0.0917

ARIMA MODELS IN R
Let's practice!
ARIMA MODELS IN R
AR and MA together
ARIMA MODELS IN R

David Stoffer
Professor of Statistics at the University
of Pittsburgh
AR and MA Together: ARMA

ARIMA MODELS IN R
AR and MA Together: ARMA

ARIMA MODELS IN R
AR and MA Together: ARMA

x <- arima.sim(list(order = c(1, 0, 1),


ar = .9,
ma = -.4),
n = 200)
plot(x, main = "ARMA(1, 1)")

ARIMA MODELS IN R
ACF and PACF of ARMA Models
AR(p) MA(q) ARMA(p, q)
ACF Tails off Cuts off lag q Tails off
PACF Cuts off lag p Tails off Tails off

ARIMA MODELS IN R
ACF and PACF of ARMA Models
AR(p) MA(q) ARMA(p, q)
ACF Tails off Cuts off lag q Tails off
PACF Cuts off lag p Tails off Tails off

ARIMA MODELS IN R
Estimation
Xt = .9Xt−1 + Wt − .4Wt−1

x <- arima.sim(list(order = c(1, 0, 1),


ar = .9,
ma = -.4),
n = 200)
x_fit <- sarima(x, p = 1, d = 0, q = 1)
x_fit$ttable

Estimate SE t.value p.value


ar1 0.9083 0.0424 21.4036 0
ma1 -0.4458 0.0879 -5.0716 0
xmean 49.5647 0.4079 121.5026 0

ARIMA MODELS IN R
Let's practice!
ARIMA MODELS IN R
Model choice and
residual analysis
ARIMA MODELS IN R

David Stoffer
Professor of Statistics at the University
of Pittsburgh
AIC and BIC

AIC and BIC measure the error and penalize (differently) for
adding parameters

For example, AIC has  k = 2 and BIC has  k = log(n)


Goal: find the model with the smallest AIC or BIC

ARIMA MODELS IN R
Model Choice: AR(1) vs. MA(2)
gnpgr <- diff(log(gnp))
sarima(gnpgr, p = 1, d = 0, q = 0)

$AIC $BIC
-8.294403 -9.263748

sarima(gnpgr, p = 0, d = 0, q = 2)

$AIC $BIC
-8.297695 -9.251712

ARIMA MODELS IN R
Residual Analysis
sarima() includes residual
analysis graphic showing:

1. Standardized residuals

2. Sample ACF of residuals

3. Normal Q-Q plot

4. Q-statistic p-values

ARIMA MODELS IN R
Bad Residuals

ARIMA MODELS IN R
Bad Residuals

ARIMA MODELS IN R
Bad Residuals

ARIMA MODELS IN R
Bad Residuals

ARIMA MODELS IN R
Bad Residuals

ARIMA MODELS IN R
Bad Residuals

ARIMA MODELS IN R
Bad Residuals

ARIMA MODELS IN R
Let's practice!
ARIMA MODELS IN R

You might also like