Scribd
Upload
26 views19 pages

ARIMA-X Time Series Analysis

The document discusses analyzing time series data to build ARIMA models. It reads in time series data, checks it for stationarity and seasonality, builds some initial ARIMA models, analyzes the residuals to check model fit, and selects a final ARIMA(1,0,0)(0,1,2) model. It then generates forecasts from the final model and plots the results.

Uploaded by

santu
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
26 views19 pages

ARIMA-X Time Series Analysis

The document discusses analyzing time series data to build ARIMA models. It reads in time series data, checks it for stationarity and seasonality, builds some initial ARIMA models, analyzes the residuals to check model fit, and selects a final ARIMA(1,0,0)(0,1,2) model. It then generates forecasts from the final model and plots the results.

Uploaded by

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

Batch 1 Anmole

ARIMA-X
[Link]

26/09/2019
data=[Link]("ARIMA-X/[Link]")# Read the file
df<-data #ts
head(df)

## Year Mth TS agri_cdt WPIT CPIAL IIP RainDev


## 1 2007 Apr 22305 1941.546 109.5 394 128.2 -7.57
## 2 2007 May 23488 1931.055 109.5 395 136.9 -16.08
## 3 2007 Jun 29631 1941.359 109.5 399 136.7 22.13
## 4 2007 Jul 20594 1926.283 109.5 404 136.6 17.33
## 5 2007 Aug 20051 1908.914 110.3 408 134.6 21.34
## 6 2007 Sep 27466 1982.069 110.5 410 134.0 45.33

names(df)

## [1] "Year" "Mth" "TS" "agri_cdt" "WPIT" "CPIAL"


## [7] "IIP" "RainDev"

Things to check
AIC
MAPE
RMSE
RMarkdown
Residual
Batch 1 Anmole

Undertanding the data


tsdisplay(df$TS, lag=50)

[Link](df$TS, alternative="stationary", k=4)

##
## Augmented Dickey-Fuller Test
##
## data: df$TS
## Dickey-Fuller = -3.9679, Lag order = 4, p-value = 0.0149
## alternative hypothesis: stationary

[Link](df$TS, alternative="stationary", k=6)

##
## Augmented Dickey-Fuller Test
##
## data: df$TS
## Dickey-Fuller = -2.6434, Lag order = 6, p-value = 0.3117
## alternative hypothesis: stationary

[Link](df$TS, alternative="stationary", k=12)


Batch 1 Anmole

##
## Augmented Dickey-Fuller Test
##
## data: df$TS
## Dickey-Fuller = -2.0832, Lag order = 12, p-value = 0.5418
## alternative hypothesis: stationary
Batch 1 Anmole

Checking for trend


tsdisplay(diff(df$TS,1),lag=50)

[Link](diff(df$TS,1), alternative="stationary", k=4)

## Warning in [Link](diff(df$TS, 1), alternative = "stationary", k = 4): p-


## value smaller than printed p-value

##
## Augmented Dickey-Fuller Test
##
## data: diff(df$TS, 1)
## Dickey-Fuller = -7.4469, Lag order = 4, p-value = 0.01
## alternative hypothesis: stationary

[Link](diff(df$TS,1), alternative="stationary", k=10)

## Warning in [Link](diff(df$TS, 1), alternative = "stationary", k = 10):


p-
## value smaller than printed p-value

##
## Augmented Dickey-Fuller Test
Batch 1 Anmole

##
## data: diff(df$TS, 1)
## Dickey-Fuller = -6.9717, Lag order = 10, p-value = 0.01
## alternative hypothesis: stationary

__There is no trend as it is trend stationary


Batch 1 Anmole

Checking for seasonality


#seasonal differencing
tsdisplay(diff(df$TS,12),lag=50)

[Link](diff(df$TS,12), alternative="stationary", k=13)

##
## Augmented Dickey-Fuller Test
##
## data: diff(df$TS, 12)
## Dickey-Fuller = -3.6518, Lag order = 13, p-value = 0.03497
## alternative hypothesis: stationary

1,0,0 1,1,0 or 0,1,1 ->>> seasonality


Batch 1 Anmole

Model 1
m3 <-arima(df$TS, order=c(1,0,0), seasonal = list(order = c(1, 1, 0), period
= 12))

summary(m3)

##
## Call:
## arima(x = df$TS, order = c(1, 0, 0), seasonal = list(order = c(1, 1, 0),
period = 12))
##
## Coefficients:
## ar1 sar1
## 0.8081 -0.3335
## s.e. 0.0692 0.1207
##
## sigma^2 estimated as 20934141: log likelihood = -729.92, aic = 1465.83
##
## Training set error measures:
## ME RMSE MAE MPE MAPE MASE
## Training set 943.6982 4244.195 2875.019 1.560307 6.915774 0.3667942
## ACF1
## Training set -0.02000167

[Link](residuals(m3), lag=10, fitdf=2, type="Ljung")

##
## Box-Ljung test
##
## data: residuals(m3)
## X-squared = 13.008, df = 8, p-value = 0.1116

m1 <-arima(df$TS, order=c(1,0,0), seasonal = list(order = c(0, 1, 1), period


= 12))

summary(m1)

##
## Call:
## arima(x = df$TS, order = c(1, 0, 0), seasonal = list(order = c(0, 1, 1),
period = 12))
##
## Coefficients:
## ar1 sma1
## 0.7884 -0.2125
## s.e. 0.0726 0.1024
##
## sigma^2 estimated as 21966635: log likelihood = -731.24, aic = 1468.48
##
Batch 1 Anmole

## Training set error measures:


## ME RMSE MAE MPE MAPE MASE
## Training set 983.0864 4347.599 2910.808 1.68214 6.973157 0.3713601
## ACF1
## Training set -0.0445108

[Link](residuals(m1), lag=10, fitdf=2, type="Ljung")

##
## Box-Ljung test
##
## data: residuals(m1)
## X-squared = 12.619, df = 8, p-value = 0.1256

Errors are random


Batch 1 Anmole

Model 2
m2 <-arima(df$TS, order=c(1,0,0), seasonal = list(order = c(1, 1, 0), period
= 12))

summary(m2)

##
## Call:
## arima(x = df$TS, order = c(1, 0, 0), seasonal = list(order = c(1, 1, 0),
period = 12))
##
## Coefficients:
## ar1 sar1
## 0.8081 -0.3335
## s.e. 0.0692 0.1207
##
## sigma^2 estimated as 20934141: log likelihood = -729.92, aic = 1465.83
##
## Training set error measures:
## ME RMSE MAE MPE MAPE MASE
## Training set 943.6982 4244.195 2875.019 1.560307 6.915774 0.3667942
## ACF1
## Training set -0.02000167

[Link](residuals(m2), lag=10, fitdf=2, type="Ljung")

##
## Box-Ljung test
##
## data: residuals(m2)
## X-squared = 13.008, df = 8, p-value = 0.1116

[Link](residuals(m2), lag=10, fitdf=2, type="Ljung")

##
## Box-Ljung test
##
## data: residuals(m2)
## X-squared = 13.008, df = 8, p-value = 0.1116

fcast.m1 <- forecast(m2, h=12)


fcast.m1

## Point Forecast Lo 80 Hi 80 Lo 95 Hi 95
## 87 61966.61 56103.02 67830.20 52999.02 70934.20
## 88 43587.62 36048.89 51126.35 32058.12 55117.11
## 89 35212.02 26756.70 43667.34 22280.72 48143.32
## 90 59804.46 50800.84 68808.09 46034.61 73574.32
## 91 83255.80 73911.49 92600.11 68964.91 97546.69
## 92 52428.79 42868.57 61989.01 37807.69 67049.89
Batch 1 Anmole

## 93 38525.03 28826.41 48223.65 23692.28 53357.79


## 94 44742.57 34954.64 54530.50 29773.22 59711.92
## 95 42353.28 32507.47 52199.10 27295.41 57411.16
## 96 48108.88 38225.45 57992.31 32993.48 63224.28
## 97 51781.12 41873.20 61689.03 36628.27 66933.96
## 98 54835.64 44911.77 64759.50 39658.39 70012.88

summary(fcast.m1)

##
## Forecast method: ARIMA(1,0,0)(1,1,0)[12]
##
## Model Information:
##
## Call:
## arima(x = df$TS, order = c(1, 0, 0), seasonal = list(order = c(1, 1, 0),
period = 12))
##
## Coefficients:
## ar1 sar1
## 0.8081 -0.3335
## s.e. 0.0692 0.1207
##
## sigma^2 estimated as 20934141: log likelihood = -729.92, aic = 1465.83
##
## Error measures:
## ME RMSE MAE MPE MAPE MASE
## Training set 943.6982 4244.195 2875.019 1.560307 6.915774 0.3667942
## ACF1
## Training set -0.02000167
##
## Forecasts:
## Point Forecast Lo 80 Hi 80 Lo 95 Hi 95
## 87 61966.61 56103.02 67830.20 52999.02 70934.20
## 88 43587.62 36048.89 51126.35 32058.12 55117.11
## 89 35212.02 26756.70 43667.34 22280.72 48143.32
## 90 59804.46 50800.84 68808.09 46034.61 73574.32
## 91 83255.80 73911.49 92600.11 68964.91 97546.69
## 92 52428.79 42868.57 61989.01 37807.69 67049.89
## 93 38525.03 28826.41 48223.65 23692.28 53357.79
## 94 44742.57 34954.64 54530.50 29773.22 59711.92
## 95 42353.28 32507.47 52199.10 27295.41 57411.16
## 96 48108.88 38225.45 57992.31 32993.48 63224.28
## 97 51781.12 41873.20 61689.03 36628.27 66933.96
## 98 54835.64 44911.77 64759.50 39658.39 70012.88

plot(fcast.m1)
Batch 1 Anmole
Batch 1 Anmole

Final Model
m4 <-arima(df$TS, order=c(1,0,0), seasonal = list(order = c(0, 1, 2), period
= 12))

summary(m4)

##
## Call:
## arima(x = df$TS, order = c(1, 0, 0), seasonal = list(order = c(0, 1, 2),
period = 12))
##
## Coefficients:
## ar1 sma1 sma2
## 0.7753 -0.2965 0.4028
## s.e. 0.0726 0.1256 0.1662
##
## sigma^2 estimated as 18719145: log likelihood = -727.4, aic = 1462.81
##
## Training set error measures:
## ME RMSE MAE MPE MAPE MASE
## Training set 784.0267 4013.385 2774.736 1.280259 6.771766 0.3540001
## ACF1
## Training set 0.01828299

[Link](residuals(m2), lag=10, fitdf=2, type="Ljung")

##
## Box-Ljung test
##
## data: residuals(m2)
## X-squared = 13.008, df = 8, p-value = 0.1116
Batch 1 Anmole

ARIMA X
library(TSA)

## Registered S3 methods overwritten by 'TSA':


## method from
## [Link] forecast
## [Link] forecast

##
## Attaching package: 'TSA'

## The following objects are masked from 'package:stats':


##
## acf, arima

## The following object is masked from 'package:utils':


##
## tar

#covariate_x<-[Link](df$agri_cdt, df$WPIT, df$CPIAL, df$IIP, df$RainDev)


armx1<-arimax(df$TS, order=c(1,0,0), seasonal = list(order = c(0, 1,2),
period = 12),
xtransf = df$agri_cdt, xreg=df$agri_cdt,method = c("ML"))

summary(armx1)

##
## Call:
## arimax(x = df$TS, order = c(1, 0, 0), seasonal = list(order = c(0, 1, 2),
period = 12),
## xreg = df$agri_cdt, method = c("ML"), xtransf = df$agri_cdt)
##
## Coefficients:
## ar1 sma1 sma2 xreg
## 0.6481 -0.3969 0.4320 15.9824
## s.e. 0.0892 0.1239 0.1768 4.0690
##
## sigma^2 estimated as 16227994: log likelihood = -722.5, aic = 1453
##
## Training set error measures:
## ME RMSE MAE MPE MAPE MASE
## Training set 27.38449 3736.794 2677.295 -1.133999 6.703173 0.3415686
## ACF1
## Training set 0.1217091
Batch 1 Anmole

ARMAX Model2
library(TSA)
covariate_x<-[Link](df$agri_cdt, df$WPIT, df$CPIAL, df$IIP, df$RainDev)
armx2<-arimax(df$TS, order=c(1,0,0), seasonal = list(order = c(0, 1,2),
period = 12),
xtransf = df$WPIT, xreg=df$WPIT,method = c("ML"))

summary(armx2)

##
## Call:
## arimax(x = df$TS, order = c(1, 0, 0), seasonal = list(order = c(0, 1, 2),
period = 12),
## xreg = df$WPIT, method = c("ML"), xtransf = df$WPIT)
##
## Coefficients:
## ar1 sma1 sma2 xreg
## 0.7126 -0.3321 0.3506 373.1313
## s.e. 0.0875 0.1238 0.1691 245.7976
##
## sigma^2 estimated as 18535767: log likelihood = -726.5, aic = 1461
##
## Training set error measures:
## ME RMSE MAE MPE MAPE MASE
## Training set 496.1127 3993.672 2802.908 0.3296495 6.837369 0.3575943
## ACF1
## Training set 0.07525393

[Link](residuals(m4), lag=18, fitdf=3, type="Ljung")

##
## Box-Ljung test
##
## data: residuals(m4)
## X-squared = 19.191, df = 15, p-value = 0.2052
Batch 1 Anmole

ARMAX Model2 - Forward way


library(TSA)
covariate_x<-[Link](df$agri_cdt, df$WPIT)
armx3<-arimax(df$TS, order=c(1,0,0), seasonal = list(order = c(0, 1,2),
period = 12),
xtransf = covariate_x, xreg=covariate_x,method = c("ML"))

summary(armx3)

##
## Call:
## arimax(x = df$TS, order = c(1, 0, 0), seasonal = list(order = c(0, 1, 2),
period = 12),
## xreg = covariate_x, method = c("ML"), xtransf = covariate_x)
##
## Coefficients:
## ar1 sma1 sma2 df.agri_cdt [Link]
## 0.6492 -0.3967 0.4367 16.7313 -48.3394
## s.e. 0.0891 0.1242 0.1778 5.4527 233.9849
##
## sigma^2 estimated as 16194130: log likelihood = -722.48, aic = 1454.95
##
## Training set error measures:
## ME RMSE MAE MPE MAPE MASE
## Training set 58.06771 3732.893 2665.362 -1.047675 6.662 0.3400462
## ACF1
## Training set 0.1205346
Batch 1 Anmole

ARMAX Model3 - Forward way


library(TSA)
covariate_x<-[Link](df$agri_cdt, df$CPIAL)
armx4<-arimax(df$TS, order=c(1,0,0), seasonal = list(order = c(0, 1,2),
period = 12),
xtransf = covariate_x, xreg=covariate_x,method = c("ML"))

summary(armx4)

##
## Call:
## arimax(x = df$TS, order = c(1, 0, 0), seasonal = list(order = c(0, 1, 2),
period = 12),
## xreg = covariate_x, method = c("ML"), xtransf = covariate_x)
##
## Coefficients:
## ar1 sma1 sma2 df.agri_cdt [Link]
## 0.6316 -0.3937 0.3980 13.0172 18.2128
## s.e. 0.0971 0.1246 0.1841 7.5953 39.2499
##
## sigma^2 estimated as 16360677: log likelihood = -722.4, aic = 1454.79
##
## Training set error measures:
## ME RMSE MAE MPE MAPE MASE
## Training set -66.47704 3752.04 2686.913 -1.370133 6.74367 0.3427957
## ACF1
## Training set 0.1204256
Batch 1 Anmole

ARMAX Model4 - Forward way


library(TSA)
covariate_x<-[Link](df$agri_cdt, df$IIP)
armx5<-arimax(df$TS, order=c(1,0,0), seasonal = list(order = c(0, 1,2),
period = 12),
xtransf = covariate_x, xreg=covariate_x,method = c("ML"))

summary(armx5)

##
## Call:
## arimax(x = df$TS, order = c(1, 0, 0), seasonal = list(order = c(0, 1, 2),
period = 12),
## xreg = covariate_x, method = c("ML"), xtransf = covariate_x)
##
## Coefficients:
## ar1 sma1 sma2 df.agri_cdt [Link]
## 0.6041 -0.4008 0.4454 14.1400 109.7743
## s.e. 0.1026 0.1242 0.1754 4.0567 101.8850
##
## sigma^2 estimated as 15927998: log likelihood = -721.93, aic = 1453.87
##
## Training set error measures:
## ME RMSE MAE MPE MAPE MASE
## Training set 19.00905 3702.099 2654.414 -1.201959 6.642797 0.3386495
## ACF1
## Training set 0.1330893
Batch 1 Anmole

ARMAX Model5 - Forward way


library(TSA)
covariate_x<-[Link](df$agri_cdt, df$RainDev)
armx6<-arimax(df$TS, order=c(1,0,0), seasonal = list(order = c(0, 1,2),
period = 12),
xtransf = covariate_x, xreg=covariate_x,method = c("ML"))

summary(armx6)

##
## Call:
## arimax(x = df$TS, order = c(1, 0, 0), seasonal = list(order = c(0, 1, 2),
period = 12),
## xreg = covariate_x, method = c("ML"), xtransf = covariate_x)
##
## Coefficients:
## ar1 sma1 sma2 df.agri_cdt [Link]
## 0.6456 -0.3984 0.4118 15.7887 5.5575
## s.e. 0.0913 0.1249 0.1743 4.0916 12.2218
##
## sigma^2 estimated as 16492586: log likelihood = -713.14, aic = 1436.28
##
## Training set error measures:
## ME RMSE MAE MPE MAPE MASE
## Training set 37.3044 3763.539 2698.115 -1.111362 6.752929 0.3442248
## ACF1
## Training set 0.1268562
Batch 1 Anmole

ARIMA X - Final only with agricdt


library(TSA)
#covariate_x<-[Link](df$agri_cdt, df$WPIT, df$CPIAL, df$IIP, df$RainDev)

summary(armx1)

##
## Call:
## arimax(x = df$TS, order = c(1, 0, 0), seasonal = list(order = c(0, 1, 2),
period = 12),
## xreg = df$agri_cdt, method = c("ML"), xtransf = df$agri_cdt)
##
## Coefficients:
## ar1 sma1 sma2 xreg
## 0.6481 -0.3969 0.4320 15.9824
## s.e. 0.0892 0.1239 0.1768 4.0690
##
## sigma^2 estimated as 16227994: log likelihood = -722.5, aic = 1453
##
## Training set error measures:
## ME RMSE MAE MPE MAPE MASE
## Training set 27.38449 3736.794 2677.295 -1.133999 6.703173 0.3415686
## ACF1
## Training set 0.1217091

MAPE is less then ARIMA Final model


[Link](residuals(m4), lag=18, fitdf=3, type="Ljung")

##
## Box-Ljung test
##
## data: residuals(m4)
## X-squared = 19.191, df = 15, p-value = 0.2052

Errors are random as well

You might also like