R and Time Series Analysis

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya

2022-11-29

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya

R and Time Series Analysis 2022-11-29 1 / 18
 Bibliography

Bartlett, M. S. 1946. “On the Theoretical Specification and Sampling

Properties of Autocorrelated Time Series.” Journal of the Royal
Statistical Society, Ser. B8 27:27–41.

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya

R and Time Series Analysis 2022-11-29 2 / 18
 Bibliography

Bartlett, M. S. 1946. “On the Theoretical Specification and Sampling

Properties of Autocorrelated Time Series.” Journal of the Royal
Statistical Society, Ser. B8 27:27–41.
Box, G. E. P., and G. Jenkins. 1976. Time Series Analysis: Forecasting
and Control. Second. San Francisco: Holden Day.

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya

R and Time Series Analysis 2022-11-29 2 / 18
 Bibliography

Bartlett, M. S. 1946. “On the Theoretical Specification and Sampling

Properties of Autocorrelated Time Series.” Journal of the Royal
Statistical Society, Ser. B8 27:27–41.
Box, G. E. P., and G. Jenkins. 1976. Time Series Analysis: Forecasting
and Control. Second. San Francisco: Holden Day.
Ljung, G. M., and G. E. P. Box. 1978. “On a Measure of Lack of Fit
in Time Series Models.” Biometrika 65:297–303.

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya

R and Time Series Analysis 2022-11-29 2 / 18
 Bibliography

Bartlett, M. S. 1946. “On the Theoretical Specification and Sampling

Properties of Autocorrelated Time Series.” Journal of the Royal
Statistical Society, Ser. B8 27:27–41.
Box, G. E. P., and G. Jenkins. 1976. Time Series Analysis: Forecasting
and Control. Second. San Francisco: Holden Day.
Ljung, G. M., and G. E. P. Box. 1978. “On a Measure of Lack of Fit
in Time Series Models.” Biometrika 65:297–303.
Racine, J. S. 2019. Reproducible Econometrics using R. Oxford
University Press, New York.

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya

R and Time Series Analysis 2022-11-29 2 / 18
 Introduction

Real time observations that are not survey datasets are mostly time
series nowadays.

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya

R and Time Series Analysis 2022-11-29 3 / 18
 Introduction

Real time observations that are not survey datasets are mostly time
series nowadays.
These data are synchronized or sampled at different high and low
frequencies ranging from 5-minutely,. . . , daily, . . . , annually.

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya

R and Time Series Analysis 2022-11-29 3 / 18
 Introduction

Real time observations that are not survey datasets are mostly time
series nowadays.
These data are synchronized or sampled at different high and low
frequencies ranging from 5-minutely,. . . , daily, . . . , annually.
R software allows for a robust time series analysis, with function
achieving different analyses in time domain and frequency domain.

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya

R and Time Series Analysis 2022-11-29 3 / 18
 Introduction (cont’d)

Some of these function are:

The ts ( ) function casts a data vector as numeric. This is a very
important step. It sets the data as time series vector.

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya

R and Time Series Analysis 2022-11-29 4 / 18
 Introduction (cont’d)

Some of these function are:

The ts ( ) function casts a data vector as numeric. This is a very
important step. It sets the data as time series vector.
plot( ) and many functions will automatically recognize time series
objects and plot them accordingly.

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya

R and Time Series Analysis 2022-11-29 4 / 18
 Introduction (cont’d)

Some of these function are:

The ts ( ) function casts a data vector as numeric. This is a very
important step. It sets the data as time series vector.
plot( ) and many functions will automatically recognize time series
objects and plot them accordingly.
for a linear model, the function lm ( ) or dyn$lm ( ) is used but the
latter requires a dyn package in R.

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya

R and Time Series Analysis 2022-11-29 4 / 18
 Some Useful R Functions for Time Series Analysis

During R installation, some basic functions are loaded/installed by

default while a lot of functions need to be loaded from optional
packages.

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya

R and Time Series Analysis 2022-11-29 5 / 18
 Some Useful R Functions for Time Series Analysis

During R installation, some basic functions are loaded/installed by

default while a lot of functions need to be loaded from optional
packages.
accuracy( ) : summary measures of the forecast accuracy

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya

R and Time Series Analysis 2022-11-29 5 / 18
 Some Useful R Functions for Time Series Analysis

During R installation, some basic functions are loaded/installed by

default while a lot of functions need to be loaded from optional
packages.
accuracy( ) : summary measures of the forecast accuracy
acf( ) : autocorrelation function

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya

R and Time Series Analysis 2022-11-29 5 / 18
 Some Useful R Functions for Time Series Analysis

During R installation, some basic functions are loaded/installed by

default while a lot of functions need to be loaded from optional
packages.
accuracy( ) : summary measures of the forecast accuracy
acf( ) : autocorrelation function
adf.test( ) : augmented Dickey-Fuller test

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya

R and Time Series Analysis 2022-11-29 5 / 18
 Some Useful R Functions for Time Series Analysis

During R installation, some basic functions are loaded/installed by

default while a lot of functions need to be loaded from optional
packages.
accuracy( ) : summary measures of the forecast accuracy
acf( ) : autocorrelation function
adf.test( ) : augmented Dickey-Fuller test
AIc( ) : Information Criterion

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya

R and Time Series Analysis 2022-11-29 5 / 18
 Some Useful R Functions for Time Series Analysis

During R installation, some basic functions are loaded/installed by

default while a lot of functions need to be loaded from optional
packages.
accuracy( ) : summary measures of the forecast accuracy
acf( ) : autocorrelation function
adf.test( ) : augmented Dickey-Fuller test
AIc( ) : Information Criterion
ar( ) : fit an autoregressive time series model

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya

R and Time Series Analysis 2022-11-29 5 / 18
 Some Useful R Functions for Time Series Analysis
(cont’d)

arima( ) : fit an ARIMA model

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya

R and Time Series Analysis 2022-11-29 6 / 18
 Some Useful R Functions for Time Series Analysis
(cont’d)

arima( ) : fit an ARIMA model

checkresiduals( ) : check residuals for white noise

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya

R and Time Series Analysis 2022-11-29 6 / 18
 Some Useful R Functions for Time Series Analysis
(cont’d)

arima( ) : fit an ARIMA model

checkresiduals( ) : check residuals for white noise
cycle( ) : gives the positions in the cycle (period) of each observation

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya

R and Time Series Analysis 2022-11-29 6 / 18
 Some Useful R Functions for Time Series Analysis
(cont’d)

arima( ) : fit an ARIMA model

checkresiduals( ) : check residuals for white noise
cycle( ) : gives the positions in the cycle (period) of each observation
diff( ) : difference the time series d time

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya

R and Time Series Analysis 2022-11-29 6 / 18
 Some Useful R Functions for Time Series Analysis
(cont’d)

arima( ) : fit an ARIMA model

checkresiduals( ) : check residuals for white noise
cycle( ) : gives the positions in the cycle (period) of each observation
diff( ) : difference the time series d time
partial autocorrelation function (stats)

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya

R and Time Series Analysis 2022-11-29 6 / 18
 Example 1: ts ( ), plot ( ) and autoplot ( ) functions
## Gold
## 1 289.30
## 2 302.70
## 3 299.40
## 4 301.20
## 5 306.90
## 6 293.20
## 7 297.20
## 8 286.70
## 9 275.80
## 10 297.20
## 11 292.80
## 12 293.60
## 13 288.50
## 14 286.40
## 15 287.30
## E.16Olubusoye,
Olusanya 280.30
Isaac Essi & OlaOluwa S. Yaya
R and Time Series Analysis 2022-11-29 7 / 18
 ts (), plot () and autoplot () functions (cont’d)
## Jan Feb Mar Apr May Jun Ju
## 1997
## 1998 302.70 299.40 301.20 306.90 293.20 297.20 286.7
## 1999 286.40 287.30 280.30 286.80 270.80 261.25 256.2
## 2000 283.90 292.70 279.40 273.60 272.60 289.30 277.5
## 2001 266.55 267.40 258.20 263.80 266.10 271.10 267.1
## 2002 282.95 296.80 302.75 308.70 326.80 314.70 303.8
## 2003 368.15 349.95 336.55 338.55 364.85 345.45 355.3
## 2004 401.60 396.00 426.70 386.70 395.10 393.30 390.6
## 2005 422.10 435.50 427.80 434.50 416.80 435.00 429.3
## 2006 568.70 561.20 583.50 653.70 642.60 615.20 635.5
## 2007 652.70 669.10 663.20 677.70 660.60 649.10 663.5
## 2008 925.00 973.50 915.80 876.70 886.30 924.60 913.3
## 2009 926.50 941.60 917.40 885.50 978.57 926.35 953.5
## 2010 1080.67 1117.30 1112.89 1178.72 1216.00 1242.13 1180.5
## 2011 1332.43 1410.96 1431.83 1563.30 1535.37 1499.67 1626.7
## E.2012
Olusanya 1737.28
Olubusoye, 1696.48
Isaac Essi & OlaOluwa R 1667.56
S. Yayaand Time Series1664.46
Analysis 1560.422022-11-29
1597.07 1614.2
8 / 18
 ts (), plot () and autoplot () functions (cont’d)
2000
1500
Gold

1000
500

2000 2005 2010 2015 2020

Time
Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya
R and Time Series Analysis 2022-11-29 9 / 18
 ts (), plot () and autoplot () functions (cont’d)
2000

1500
US Dollar/Troy

1000

500

2000 2005 2010 2015 2020

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya
R and Time Series
Date Analysis 2022-11-29 10 / 18
 ts ( ), plot ( ) and autoplot ( ) functions (cont’d)
Figure 14: London Bullion Gold Price

2000

1500
US dollar/troy

1000

500

2000 2005 2010 2015 2020

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya
R and Time Series
Year Analysis 2022-11-29 11 / 18
 Example 2: diff( ) function
This is used to find the unit differences (integration/unit root) in time
series. for a nonstationary series, the series requires 1 or 2 unit differences
to make it stationary
Recall,

yt = (1 − B)d Xt
where yt is the resulting stationary series, and Xt is the (original)
nonstationary series. For a series requiring 1 unit difference, i.e. d = 1, then,
we have,

yt = (1 − B)1 Xt = (1 − B)Xt = Xt − BXt = Xt − Xt−1

Note, in the case of fractional or decimal value of d, this leads to fractional

unit root often known as fractional integration.
Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya
R and Time Series Analysis 2022-11-29 12 / 18
 Example 2: diff( ) function (cont’d)
## Jan Feb Mar Apr May Jun Ju
## 1998 13.40 -3.30 1.80 5.70 -13.70 4.00 -10.5
## 1999 -2.10 0.90 -7.00 6.50 -16.00 -9.55 -5.0
## 2000 -5.10 8.80 -13.30 -5.80 -1.00 16.70 -11.8
## 2001 -5.95 0.85 -9.20 5.60 2.30 5.00 -4.0
## 2002 3.75 13.85 5.95 5.95 18.10 -12.10 -10.9
## 2003 20.10 -18.20 -13.40 2.00 26.30 -19.40 9.8
## 2004 -14.00 -5.60 30.70 -40.00 8.40 -1.80 -2.7
## 2005 -15.69 13.40 -7.70 6.70 -17.70 18.20 -5.7
## 2006 52.70 -7.50 22.30 70.20 -11.10 -27.40 20.3
## 2007 17.10 16.40 -5.90 14.50 -17.10 -11.50 14.4
## 2008 93.50 48.50 -57.70 -39.10 9.60 38.30 -11.3
## 2009 60.60 15.10 -24.20 -31.90 93.07 -52.22 27.1
## 2010 -17.63 36.63 -4.41 65.83 37.28 26.13 -61.5
## 2011 -85.76 78.53 20.87 131.47 -27.93 -35.70 127.1
## 2012 172.81 -40.80 -28.92 -3.10 -104.04 36.65 17.1
## E.2013
Olusanya 0.95
Olubusoye, Isaac -84.06
Essi & OlaOluwa R and17.24
S. Yaya Time Series-120.37
Analysis -88.712022-11-29
-153.50 1390.7
/ 18
 Example 2: diff( ) function (cont’d)

Note, in diff (seriesname, 1), “1’ ’ is the lag difference, i.e. from xt to xt−1 .
if it is xt to xt−2 , it means it will be diff (seriesname, 2).
Thus, for first difference of a time series, it is usually diff (seriesname, 1).

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya

R and Time Series Analysis 2022-11-29 14 / 18
 Example 2: diff( ) function (cont’d)
200
100
Gold

0
-100
-200

2000 2005 2010 2015 2020

Time
Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya
R and Time Series Analysis 2022-11-29 15 / 18
 Example 3: acf() function
Gold
1.0
0.8
0.6
ACF

0.4
0.2
0.0

0.0 0.5 1.0 1.5 2.0

Lag
Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya
R and Time Series Analysis 2022-11-29 16 / 18
 Example 5: pacf() function
Series data3
0.15
0.10
0.05
Partial ACF

0.00
-0.05
-0.10
-0.15

0.5 1.0 1.5 2.0

Lag
Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya
R and Time Series Analysis 2022-11-29 17 / 18
 Example 6: adf.test() function
The ADF (Augmented Dickey-Fuller) unit root testing in R requires the
library “tseries’ ’.
##
## Augmented Dickey-Fuller Test
##
## data: data3
## Dickey-Fuller = -5.595, Lag order = 6, p-value = 0.01
## alternative hypothesis: stationary
Note, the ADF test has the null hypothesis, H0 that the series is
nonstationary (has a unit root) against the alternative, H1 that is it
stationary (no unit root).

Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya

R and Time Series Analysis 2022-11-29 18 / 18
 Example 6: adf.test() function
The ADF (Augmented Dickey-Fuller) unit root testing in R requires the
library “tseries’ ’.
##
## Augmented Dickey-Fuller Test
##
## data: data3
## Dickey-Fuller = -5.595, Lag order = 6, p-value = 0.01
## alternative hypothesis: stationary
Note, the ADF test has the null hypothesis, H0 that the series is
nonstationary (has a unit root) against the alternative, H1 that is it
stationary (no unit root).
Selected lag length is for the augmentation component. Such a lag
length is determined in the ADF regression based on minimum
information criterion used.
Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya
R and Time Series Analysis 2022-11-29 18 / 18
 Example 6: adf.test() function
The ADF (Augmented Dickey-Fuller) unit root testing in R requires the
library “tseries’ ’.
##
## Augmented Dickey-Fuller Test
##
## data: data3
## Dickey-Fuller = -5.595, Lag order = 6, p-value = 0.01
## alternative hypothesis: stationary
Note, the ADF test has the null hypothesis, H0 that the series is
nonstationary (has a unit root) against the alternative, H1 that is it
stationary (no unit root).
Selected lag length is for the augmentation component. Such a lag
length is determined in the ADF regression based on minimum
information criterion used.
β2022-11-29
The t-statistic value is obtained by computing t =
Olusanya E. Olubusoye, Isaac Essi & OlaOluwa S. Yaya
R and Time Series Analysis . 18 / 18