Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J...
Transcript of Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J...
![Page 1: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/1.jpg)
1
Forecasting:principles andpractice
Rob J Hyndman
2 ARIMA models
![Page 2: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/2.jpg)
Outline
1 Stationarity and differencing
2 Backshift notation
3 Autoregressive models
4 Moving Average models
5 Non-seasonal ARIMA models
6 Seasonal ARIMA models
7 Lab Session 3
2
![Page 3: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/3.jpg)
Stationarity
DefinitionIf {yt} is a stationary time series, then for all s, thedistribution of (yt, . . . , yt+s) does not depend on t.
A stationary series is:roughly horizontalconstant varianceno patterns predictable in the long-term
Transformations (e.g., logs) can help to stabilizethe variance.Differences can help to stabilize the mean.
3
![Page 4: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/4.jpg)
Stationarity
DefinitionIf {yt} is a stationary time series, then for all s, thedistribution of (yt, . . . , yt+s) does not depend on t.
A stationary series is:roughly horizontalconstant varianceno patterns predictable in the long-term
Transformations (e.g., logs) can help to stabilizethe variance.Differences can help to stabilize the mean.
3
![Page 5: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/5.jpg)
Stationarity
DefinitionIf {yt} is a stationary time series, then for all s, thedistribution of (yt, . . . , yt+s) does not depend on t.
A stationary series is:roughly horizontalconstant varianceno patterns predictable in the long-term
Transformations (e.g., logs) can help to stabilizethe variance.Differences can help to stabilize the mean. 3
![Page 6: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/6.jpg)
Stationary?dj %>% autoplot() +ylab("Dow Jones Index") + xlab("Day")
3600
3700
3800
3900
4000
0 50 100 150 200 250 300
Day
Dow
Jon
es In
dex
4
![Page 7: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/7.jpg)
Stationary?dj %>% diff() %>% autoplot() +ylab("Change in Dow Jones Index") + xlab("Day")
−100
−50
0
50
0 50 100 150 200 250 300
Day
Cha
nge
in D
ow J
ones
Inde
x
5
![Page 8: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/8.jpg)
Stationary?hsales %>% autoplot() +
xlab("Year") + ylab("Total sales") +ggtitle("Sales of new one-family houses, USA")
40
60
80
1975 1980 1985 1990 1995
Year
Tota
l sal
es
Sales of new one−family houses, USA
6
![Page 9: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/9.jpg)
Stationary?hsales %>% diff(lag=12) %>% autoplot() +xlab("Year") + ylab("Total sales") +ggtitle("Seasonal differences of sales of new one-family houses, USA")
−20
0
20
1975 1980 1985 1990 1995
Year
Tota
l sal
es
Seasonal differences of sales of new one−family houses, USA
7
![Page 10: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/10.jpg)
Stationary?hsales %>% diff(lag=12) %>% diff(lag=1) %>% autoplot() +xlab("Year") + ylab("Total sales") +ggtitle("Seasonal differences of sales of new one-family houses, USA")
−10
0
10
1975 1980 1985 1990 1995
Year
Tota
l sal
es
Seasonal differences of sales of new one−family houses, USA
8
![Page 11: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/11.jpg)
Electricity production
usmelec %>% autoplot()
200
300
400
1980 1990 2000 2010
Time
.
9
![Page 12: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/12.jpg)
Electricity production
usmelec %>% log() %>% autoplot()
5.1
5.4
5.7
6.0
1980 1990 2000 2010
Time
.
10
![Page 13: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/13.jpg)
Electricity production
usmelec %>% log() %>% diff(lag=12) %>%autoplot()
0.0
0.1
1980 1990 2000 2010
Time
.
11
![Page 14: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/14.jpg)
Electricity production
usmelec %>% log() %>% diff(lag=12) %>%diff(lag=1) %>% autoplot()
−0.15
−0.10
−0.05
0.00
0.05
0.10
1980 1990 2000 2010
Time
.
12
![Page 15: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/15.jpg)
Outline
1 Stationarity and differencing
2 Backshift notation
3 Autoregressive models
4 Moving Average models
5 Non-seasonal ARIMA models
6 Seasonal ARIMA models
7 Lab Session 3
13
![Page 16: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/16.jpg)
Backshift notation
Backward shift operator
Shift back one periodByt = yt−1
Shift back two periods:B(Byt) = B2yt = yt−2
Shift back 12 periodsB12yt = yt−12
14
![Page 17: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/17.jpg)
Backshift notation
Backward shift operator
Shift back one periodByt = yt−1
Shift back two periods:B(Byt) = B2yt = yt−2
Shift back 12 periodsB12yt = yt−12
14
![Page 18: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/18.jpg)
Backshift notation
Backward shift operator
Shift back one periodByt = yt−1
Shift back two periods:B(Byt) = B2yt = yt−2
Shift back 12 periodsB12yt = yt−12 14
![Page 19: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/19.jpg)
Backshift notation
First differencesy′t = yt − yt−1 = yt − Byt = (1− B)yt .
Second-order differences (i.e., first differences offirst differences):
y′′t = (1− B)2yt .
dth-order differences:(1− B)dyt.
Seasonal difference followed by first difference:(1− B)(1− Bm)yt .
15
![Page 20: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/20.jpg)
Backshift notation
First differencesy′t = yt − yt−1 = yt − Byt = (1− B)yt .
Second-order differences (i.e., first differences offirst differences):
y′′t = (1− B)2yt .
dth-order differences:(1− B)dyt.
Seasonal difference followed by first difference:(1− B)(1− Bm)yt .
15
![Page 21: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/21.jpg)
Backshift notation
First differencesy′t = yt − yt−1 = yt − Byt = (1− B)yt .
Second-order differences (i.e., first differences offirst differences):
y′′t = (1− B)2yt .
dth-order differences:(1− B)dyt.
Seasonal difference followed by first difference:(1− B)(1− Bm)yt .
15
![Page 22: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/22.jpg)
Backshift notation
First differencesy′t = yt − yt−1 = yt − Byt = (1− B)yt .
Second-order differences (i.e., first differences offirst differences):
y′′t = (1− B)2yt .
dth-order differences:(1− B)dyt.
Seasonal difference followed by first difference:(1− B)(1− Bm)yt .
15
![Page 23: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/23.jpg)
Outline
1 Stationarity and differencing
2 Backshift notation
3 Autoregressive models
4 Moving Average models
5 Non-seasonal ARIMA models
6 Seasonal ARIMA models
7 Lab Session 3
16
![Page 24: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/24.jpg)
Autoregressive models
Autoregressive (AR) models:
yt = c + φ1yt−1 + φ2yt−2 + · · · + φpyt−p + εt(1− φ1B− · · · − φpBp)yt = c + εt
where εt is white noise. This is a multiple regressionwith lagged values of yt as predictors.
8
10
12
0 20 40 60 80 100
Time
AR(1)
15.0
17.5
20.0
22.5
25.0
0 20 40 60 80 100
Time
AR(2)
17
![Page 25: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/25.jpg)
Stationarity conditions
We normally restrict autoregressive models tostationary data, and then some constraints on thevalues of the parameters are required.General condition for stationarityComplex roots of 1− φ1z− φ2z2 − · · · − φpzp lieoutside the unit circle on the complex plane.
For p = 1: −1 < φ1 < 1.For p = 2:−1 < φ2 < 1 φ2 + φ1 < 1 φ2 − φ1 < 1.More complicated conditions hold for p ≥ 3.
18
![Page 26: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/26.jpg)
Stationarity conditions
We normally restrict autoregressive models tostationary data, and then some constraints on thevalues of the parameters are required.General condition for stationarityComplex roots of 1− φ1z− φ2z2 − · · · − φpzp lieoutside the unit circle on the complex plane.
For p = 1: −1 < φ1 < 1.For p = 2:−1 < φ2 < 1 φ2 + φ1 < 1 φ2 − φ1 < 1.More complicated conditions hold for p ≥ 3.
18
![Page 27: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/27.jpg)
Outline
1 Stationarity and differencing
2 Backshift notation
3 Autoregressive models
4 Moving Average models
5 Non-seasonal ARIMA models
6 Seasonal ARIMA models
7 Lab Session 3
19
![Page 28: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/28.jpg)
Moving Average (MA) models
Moving Average (MA) models:
yt = c + εt + θ1εt−1 + θ2εt−2 + · · · + θqεt−qyt = c + (1 + θ1B + · · · + θqBq)εt
where εt is white noise. This is a multiple regressionwith past errors as predictors.
18
20
22
0 20 40 60 80 100
Time
MA(1)
−5.0
−2.5
0.0
2.5
0 20 40 60 80 100
Time
MA(2)
20
![Page 29: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/29.jpg)
Invertibility
Invertible models have property that distant pasthas negligible effect on forecasts. Requiresconsraints on MA parameters.
General condition for invertibilityComplex roots of 1 + θ1z + θ2z2 + · · · + θqzq lie outsidethe unit circle on the complex plane.
For q = 1: −1 < θ1 < 1.For q = 2:−1 < θ2 < 1 θ2 + θ1 > −1 θ1 − θ2 < 1.More complicated conditions hold for q ≥ 3.
21
![Page 30: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/30.jpg)
Invertibility
Invertible models have property that distant pasthas negligible effect on forecasts. Requiresconsraints on MA parameters.
General condition for invertibilityComplex roots of 1 + θ1z + θ2z2 + · · · + θqzq lie outsidethe unit circle on the complex plane.
For q = 1: −1 < θ1 < 1.For q = 2:−1 < θ2 < 1 θ2 + θ1 > −1 θ1 − θ2 < 1.More complicated conditions hold for q ≥ 3.
21
![Page 31: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/31.jpg)
Outline
1 Stationarity and differencing
2 Backshift notation
3 Autoregressive models
4 Moving Average models
5 Non-seasonal ARIMA models
6 Seasonal ARIMA models
7 Lab Session 3
22
![Page 32: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/32.jpg)
ARIMA models
Autoregressive Moving Average models:yt = c + φ1yt−1 + · · · + φpyt−p
+ θ1εt−1 + · · · + θqεt−q + εtφp(B)yt = θq(B)εt
Predictors include both lagged values of yt andlagged errors.φp(B) is a pth order polynomial in Bθq(B) is a qth order polynomial in B
Autoregressive Integrated Moving Average modelsCombine ARMA model with differencing.(1− B)dyt follows an ARMA model.
23
![Page 33: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/33.jpg)
ARIMA models
Autoregressive Moving Average models:yt = c + φ1yt−1 + · · · + φpyt−p
+ θ1εt−1 + · · · + θqεt−q + εtφp(B)yt = θq(B)εt
Predictors include both lagged values of yt andlagged errors.φp(B) is a pth order polynomial in Bθq(B) is a qth order polynomial in B
Autoregressive Integrated Moving Average modelsCombine ARMA model with differencing.(1− B)dyt follows an ARMA model.
23
![Page 34: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/34.jpg)
ARIMA models
Autoregressive Moving Average models:yt = c + φ1yt−1 + · · · + φpyt−p
+ θ1εt−1 + · · · + θqεt−q + εtφp(B)yt = θq(B)εt
Predictors include both lagged values of yt andlagged errors.φp(B) is a pth order polynomial in Bθq(B) is a qth order polynomial in B
Autoregressive Integrated Moving Average modelsCombine ARMA model with differencing.(1− B)dyt follows an ARMA model. 23
![Page 35: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/35.jpg)
ARIMA models
Autoregressive Integrated Moving Average modelsARIMA(p, d, q) modelAR: p = order of the autoregressive partI: d = degree of first differencing involved
MA: q = order of the moving average part.
White noise model: ARIMA(0,0,0)Random walk: ARIMA(0,1,0) with no constantRandom walk with drift: ARIMA(0,1,0) with const.AR(p): ARIMA(p,0,0)MA(q): ARIMA(0,0,q) 24
![Page 36: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/36.jpg)
Backshift notation for ARIMA
ARIMA(p, 0, q) model:yt = c + φ1yt−1 + · · · + φpyt−p + θ1εt−1 + · · · + θqεt−q + εtyt = c + φ1Byt + · · · + φpBpyt + εt + θ1Bεt + · · · + θqBqεt
or (1− φ1B− · · · − φpBp)yt = c + (1 + θ1B + · · · + θqBq)εt
ARIMA(1,1,1) model:(1− φ1B) (1− B)yt = c + (1 + θ1B)εt↑ ↑ ↑
AR(1) First MA(1)difference
Written out:yt = c + yt−1 + φ1yt−1 − φ1yt−2 + θ1εt−1 + εt
25
![Page 37: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/37.jpg)
Backshift notation for ARIMA
ARIMA(p, 0, q) model:yt = c + φ1yt−1 + · · · + φpyt−p + θ1εt−1 + · · · + θqεt−q + εtyt = c + φ1Byt + · · · + φpBpyt + εt + θ1Bεt + · · · + θqBqεt
or (1− φ1B− · · · − φpBp)yt = c + (1 + θ1B + · · · + θqBq)εt
ARIMA(1,1,1) model:(1− φ1B) (1− B)yt = c + (1 + θ1B)εt↑ ↑ ↑
AR(1) First MA(1)difference
Written out:yt = c + yt−1 + φ1yt−1 − φ1yt−2 + θ1εt−1 + εt
25
![Page 38: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/38.jpg)
Backshift notation for ARIMA
ARIMA(p, 0, q) model:yt = c + φ1yt−1 + · · · + φpyt−p + θ1εt−1 + · · · + θqεt−q + εtyt = c + φ1Byt + · · · + φpBpyt + εt + θ1Bεt + · · · + θqBqεt
or (1− φ1B− · · · − φpBp)yt = c + (1 + θ1B + · · · + θqBq)εt
ARIMA(1,1,1) model:(1− φ1B) (1− B)yt = c + (1 + θ1B)εt↑ ↑ ↑
AR(1) First MA(1)difference
Written out:yt = c + yt−1 + φ1yt−1 − φ1yt−2 + θ1εt−1 + εt 25
![Page 39: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/39.jpg)
R modelIntercept form
(1− φ1B− · · · − φpBp)(1− B)dyt = c + (1 + θ1B + · · · + θqBq)εt
Mean form
(1− φ1B− · · · − φpBp)(1− B)d(yt − µtd/d!) =(1 + θ1B + · · · + θqBq)εt
µ is the mean of (1− B)dyt.c = µ(1− φ1 − · · · − φp).R uses mean form.Including c equivalent to yt having dth orderpolynomial trend.
26
![Page 40: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/40.jpg)
US personal consumptionautoplot(uschange[,"Consumption"]) +xlab("Year") + ylab("Quarterly percentage change") +ggtitle("US consumption")
−2
−1
0
1
2
1970 1980 1990 2000 2010
Year
Qua
rter
ly p
erce
ntag
e ch
ange
US consumption
27
![Page 41: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/41.jpg)
US personal consumption(fit <- auto.arima(uschange[,"Consumption"]))
## Series: uschange[, "Consumption"]
## ARIMA(2,0,2) with non-zero mean
##
## Coefficients:
## ar1 ar2 ma1 ma2 mean
## 1.391 -0.581 -1.180 0.558 0.746
## s.e. 0.255 0.208 0.238 0.140 0.084
##
## sigma^2 estimated as 0.351: log likelihood=-165.1
## AIC=342.3 AICc=342.8 BIC=361.7
ARIMA(2,0,2) model:yt = c + 1.391yt−1 − 0.581yt−2 − 1.180εt−1 + 0.558εt−2 + εt,
where c = 0.746× (1− 1.391 + 0.581) = 0.142 and εt ∼ N(0, 0.351).
28
![Page 42: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/42.jpg)
US personal consumption(fit <- auto.arima(uschange[,"Consumption"]))
## Series: uschange[, "Consumption"]
## ARIMA(2,0,2) with non-zero mean
##
## Coefficients:
## ar1 ar2 ma1 ma2 mean
## 1.391 -0.581 -1.180 0.558 0.746
## s.e. 0.255 0.208 0.238 0.140 0.084
##
## sigma^2 estimated as 0.351: log likelihood=-165.1
## AIC=342.3 AICc=342.8 BIC=361.7
ARIMA(2,0,2) model:yt = c + 1.391yt−1 − 0.581yt−2 − 1.180εt−1 + 0.558εt−2 + εt,
where c = 0.746× (1− 1.391 + 0.581) = 0.142 and εt ∼ N(0, 0.351). 28
![Page 43: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/43.jpg)
US personal consumption
fit %>% forecast(h=10) %>% autoplot(include=80)
−1
0
1
2
2000 2005 2010 2015 2020
Time
usch
ange
[, "C
onsu
mpt
ion"
]
level
80
95
Forecasts from ARIMA(2,0,2) with non−zero mean
29
![Page 44: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/44.jpg)
Information criteria
Akaike’s Information Criterion (AIC):AIC = −2 log(L) + 2(p + q + k + 1),
where L is the likelihood of the data,k = 1 if c 6= 0 and k = 0 if c = 0.
Corrected AIC:
AICc = AIC +2(p + q + k + 1)(p + q + k + 2)
T − p− q− k− 2.
Good models are obtained by minimizing the AICc.
30
![Page 45: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/45.jpg)
Information criteria
Akaike’s Information Criterion (AIC):AIC = −2 log(L) + 2(p + q + k + 1),
where L is the likelihood of the data,k = 1 if c 6= 0 and k = 0 if c = 0.
Corrected AIC:
AICc = AIC +2(p + q + k + 1)(p + q + k + 2)
T − p− q− k− 2.
Good models are obtained by minimizing the AICc.
30
![Page 46: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/46.jpg)
Information criteria
Akaike’s Information Criterion (AIC):AIC = −2 log(L) + 2(p + q + k + 1),
where L is the likelihood of the data,k = 1 if c 6= 0 and k = 0 if c = 0.
Corrected AIC:
AICc = AIC +2(p + q + k + 1)(p + q + k + 2)
T − p− q− k− 2.
Good models are obtained by minimizing the AICc.
30
![Page 47: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/47.jpg)
How does auto.arima() work?
A non-seasonal ARIMA process
φ(B)(1− B)dyt = c + θ(B)εtNeed to select appropriate orders: p, q, d
Hyndman and Khandakar (JSS, 2008) algorithm:
Select no. differences d and D via KPSS test andseasonal strength measure.Select p, q by minimising AICc.Use stepwise search to traverse model space.
31
![Page 48: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/48.jpg)
How does auto.arima() work?
Step 1: Select values of d and D.
Step 2: Select current model (with smallest AICc) from:ARIMA(2, d, 2)ARIMA(0, d, 0)ARIMA(1, d, 0)ARIMA(0, d, 1)
Step 3: Consider variations of current model:vary one of p, q, from current model by±1;p, q both vary from current model by±1;Include/exclude c from current model.
Model with lowest AICc becomes current model.
Repeat Step 3 until no lower AICc can be found.
32
![Page 49: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/49.jpg)
How does auto.arima() work?
Step 1: Select values of d and D.
Step 2: Select current model (with smallest AICc) from:ARIMA(2, d, 2)ARIMA(0, d, 0)ARIMA(1, d, 0)ARIMA(0, d, 1)
Step 3: Consider variations of current model:vary one of p, q, from current model by±1;p, q both vary from current model by±1;Include/exclude c from current model.
Model with lowest AICc becomes current model.
Repeat Step 3 until no lower AICc can be found. 32
![Page 50: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/50.jpg)
Choosing an ARIMA model
autoplot(internet)
80
120
160
200
0 20 40 60 80 100
Time
inte
rnet
33
![Page 51: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/51.jpg)
Choosing an ARIMA model
(fit <- auto.arima(internet))
## Series: internet## ARIMA(1,1,1)#### Coefficients:## ar1 ma1## 0.650 0.526## s.e. 0.084 0.090#### sigma^2 estimated as 10: log likelihood=-254.2## AIC=514.3 AICc=514.5 BIC=522.1
34
![Page 52: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/52.jpg)
Choosing an ARIMA model
(fit <- auto.arima(internet, stepwise=FALSE,approximation=FALSE))
## Series: internet## ARIMA(3,1,0)#### Coefficients:## ar1 ar2 ar3## 1.151 -0.661 0.341## s.e. 0.095 0.135 0.094#### sigma^2 estimated as 9.66: log likelihood=-252## AIC=512 AICc=512.4 BIC=522.4
35
![Page 53: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/53.jpg)
Choosing an ARIMA model
checkresiduals(fit, plot=TRUE)
−5
0
5
0 20 40 60 80 100
Residuals from ARIMA(3,1,0)
−0.2
−0.1
0.0
0.1
0.2
5 10 15 20
Lag
AC
F
0
5
10
15
20
−10 −5 0 5 10
residuals
coun
t
##
## Ljung-Box test
##
## data: Residuals from ARIMA(3,1,0)
## Q* = 4.5, df = 7, p-value = 0.7
##
## Model df: 3. Total lags used: 10
36
![Page 54: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/54.jpg)
Choosing an ARIMA model
fit %>% forecast() %>% autoplot()
100
150
200
250
0 30 60 90
Time
inte
rnet
level
80
95
Forecasts from ARIMA(3,1,0)
37
![Page 55: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/55.jpg)
Outline
1 Stationarity and differencing
2 Backshift notation
3 Autoregressive models
4 Moving Average models
5 Non-seasonal ARIMA models
6 Seasonal ARIMA models
7 Lab Session 3
38
![Page 56: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/56.jpg)
Seasonal ARIMA models
ARIMA (p, d, q)︸ ︷︷ ︸ (P,D,Q)m︸ ︷︷ ︸↑ ↑
Non-seasonal part Seasonal part ofof the model of the model
wherem = number of observations per year.
39
![Page 57: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/57.jpg)
Seasonal ARIMA models
E.g., ARIMA(1, 1, 1)(1, 1, 1)4 model (without constant)
(1−φ1B)(1−Φ1B4)(1−B)(1−B4)yt = (1+θ1B)(1+Θ1B4)εt.
6 6 6 6 6 6(Non-seasonal
AR(1)
)(SeasonalAR(1)
)(Non-seasonaldifference
)(
Seasonaldifference
)(Non-seasonal
MA(1)
)(SeasonalMA(1)
)
40
![Page 58: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/58.jpg)
Seasonal ARIMA models
E.g., ARIMA(1, 1, 1)(1, 1, 1)4 model (without constant)(1−φ1B)(1−Φ1B4)(1−B)(1−B4)yt = (1+θ1B)(1+Θ1B4)εt.
6 6 6 6 6 6(Non-seasonal
AR(1)
)(SeasonalAR(1)
)(Non-seasonaldifference
)(
Seasonaldifference
)(Non-seasonal
MA(1)
)(SeasonalMA(1)
)
40
![Page 59: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/59.jpg)
Seasonal ARIMA models
E.g., ARIMA(1, 1, 1)(1, 1, 1)4 model (without constant)(1−φ1B)(1−Φ1B4)(1−B)(1−B4)yt = (1+θ1B)(1+Θ1B4)εt.
6 6 6 6 6 6(Non-seasonal
AR(1)
)(SeasonalAR(1)
)(Non-seasonaldifference
)(
Seasonaldifference
)(Non-seasonal
MA(1)
)(SeasonalMA(1)
)
40
![Page 60: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/60.jpg)
Seasonal ARIMA models
E.g., ARIMA(1, 1, 1)(1, 1, 1)4 model (without constant)(1−φ1B)(1−Φ1B4)(1−B)(1−B4)yt = (1+θ1B)(1+Θ1B4)εt.
All the factors can be multiplied out and the generalmodel written as follows:
yt = (1 + φ1)yt−1 − φ1yt−2 + (1 + Φ1)yt−4− (1 + φ1 + Φ1 + φ1Φ1)yt−5 + (φ1 + φ1Φ1)yt−6−Φ1yt−8 + (Φ1 + φ1Φ1)yt−9 − φ1Φ1yt−10+ εt + θ1εt−1 + Θ1εt−4 + θ1Θ1εt−5.
41
![Page 61: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/61.jpg)
European quarterly retail trade
autoplot(euretail) +xlab("Year") + ylab("Retail index")
92
96
100
2000 2005 2010
Year
Ret
ail i
ndex
42
![Page 62: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/62.jpg)
European quarterly retail trade
(fit <- auto.arima(euretail))
## Series: euretail## ARIMA(1,1,2)(0,1,1)[4]#### Coefficients:## ar1 ma1 ma2 sma1## 0.736 -0.466 0.216 -0.843## s.e. 0.224 0.199 0.210 0.188#### sigma^2 estimated as 0.159: log likelihood=-29.62## AIC=69.24 AICc=70.38 BIC=79.63
43
![Page 63: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/63.jpg)
European quarterly retail trade
(fit <- auto.arima(euretail, stepwise=TRUE,approximation=FALSE))
## Series: euretail## ARIMA(1,1,2)(0,1,1)[4]#### Coefficients:## ar1 ma1 ma2 sma1## 0.736 -0.466 0.216 -0.843## s.e. 0.224 0.199 0.210 0.188#### sigma^2 estimated as 0.159: log likelihood=-29.62## AIC=69.24 AICc=70.38 BIC=79.63
44
![Page 64: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/64.jpg)
European quarterly retail trade
checkresiduals(fit, test=FALSE)
−1.0
−0.5
0.0
0.5
1.0
2000 2005 2010
Residuals from ARIMA(1,1,2)(0,1,1)[4]
−0.2
−0.1
0.0
0.1
0.2
4 8 12 16
Lag
AC
F
0
5
10
−1.0 −0.5 0.0 0.5 1.0
residuals
coun
t
45
![Page 65: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/65.jpg)
European quarterly retail trade
forecast(fit, h=36) %>% autoplot()
80
90
100
110
1995 2000 2005 2010 2015 2020
Time
eure
tail level
80
95
Forecasts from ARIMA(1,1,2)(0,1,1)[4]
46
![Page 66: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/66.jpg)
Cortecosteroid drug salesH
02 sales (million scripts)
Log H02 sales
1995 2000 2005
0.50
0.75
1.00
1.25
−0.8
−0.4
0.0
Year
47
![Page 67: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/67.jpg)
Cortecosteroid drug sales
(fit <- auto.arima(h02, lambda=0, max.order=9,
stepwise=FALSE, approximation=FALSE))
## Series: h02
## ARIMA(4,1,1)(2,1,2)[12]
## Box Cox transformation: lambda= 0
##
## Coefficients:
## ar1 ar2 ar3 ar4 ma1 sar1
## -0.042 0.210 0.202 -0.227 -0.742 0.621
## s.e. 0.217 0.181 0.114 0.081 0.207 0.242
## sar2 sma1 sma2
## -0.383 -1.202 0.496
## s.e. 0.118 0.249 0.214
##
## sigma^2 estimated as 0.00405: log likelihood=254.3
## AIC=-488.6 AICc=-487.4 BIC=-456.1
48
![Page 68: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/68.jpg)
Cortecosteroid drug sales
checkresiduals(fit)
−0.2
−0.1
0.0
0.1
1995 2000 2005
Residuals from ARIMA(4,1,1)(2,1,2)[12]
−0.1
0.0
0.1
12 24 36
Lag
AC
F
0
10
20
30
−0.2 −0.1 0.0 0.1 0.2
residuals
coun
t
#### Ljung-Box test#### data: Residuals from ARIMA(4,1,1)(2,1,2)[12]## Q* = 16, df = 15, p-value = 0.4#### Model df: 9. Total lags used: 24
49
![Page 69: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/69.jpg)
Understanding ARIMA models
Long-term forecastszero c = 0, d + D = 0non-zero constant c = 0, d + D = 1 c 6= 0, d + D = 0linear c = 0, d + D = 2 c 6= 0, d + D = 1quadratic c = 0, d + D = 3 c 6= 0, d + D = 2
Forecast variance and d + DThe higher the value of d + D, the more rapidlythe prediction intervals increase in size.For d + D = 0, the long-term forecast standarddeviation will go to the standard deviation of thehistorical data.
50
![Page 70: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/70.jpg)
Prediction intervals
Prediction intervals increase in size withforecast horizon.Calculations assume residuals are uncorrelatedand normally distributed.Prediction intervals tend to be too narrow.
the uncertainty in the parameter estimates has notbeen accounted for.the ARIMA model assumes historical patterns willnot change during the forecast period.the ARIMA model assumes uncorrelated futureerrors
51
![Page 71: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/71.jpg)
Outline
1 Stationarity and differencing
2 Backshift notation
3 Autoregressive models
4 Moving Average models
5 Non-seasonal ARIMA models
6 Seasonal ARIMA models
7 Lab Session 3
52
![Page 72: Forecasting: principles and practice · 2018. 7. 26. · Forecasting: principles and practice Rob J Hyndman 2 ARIMA models. Outline 1 Stationarity and di˙erencing 2 Backshift notation](https://reader034.fdocuments.in/reader034/viewer/2022052617/60b0e32ae4d1f6398a1a2884/html5/thumbnails/72.jpg)
Lab Session 3
53