8. time series calculations
TRANSCRIPT
-
7/30/2019 8. Time Series Calculations
1/31
1
2
3
4
Step 1
Step 2
5
Year QuarterImports
(X)Trends (T)
Deviation from trend
OR Detrended values
: ( X - T) = S + R
19-2 1 114
2 1423 155 137.0 +18.0
4 136 138.25 -2.25
19-3 1 116 139.0 -23.0
2 150 139.25 +10.75
3 153 141.25 +11.75
4 140 143.75 -3.75
19-4 1 128 146.75 -18.752 158 151.125 +6.875
3 169 154.625 +14.375
4 159 158.5 +0.5
19-5 1 137 164.125 -27.125
2 180 168.625 +11.375
Calculate the trend value for each observation
For each observation, calculate the detrended value
The calculation of seasonal variation (Additive model)
(Pg 139 OJ)
Seasonal variation is a swing around the trend line
Using the Additive model where, X = T + S + R
Variation in any particular quarter will be : X - T = S + R
Calculation of Detrended values
How to calculate Seasonal Variation
-
7/30/2019 8. Time Series Calculations
2/31
3 192 171.25 +20.75
4 172 174.0 -2.00
19-6 1 145
2 194
Step 4: Calculate the total for each quarter
Step 5: Calculate the average for each quarter
Step 6: If sum of the average calculated for all 4 quarters is not equ
6
1 2 3 4
19-2 +18.0 -2.25
19-3 -23.0 +10.75 +11.75 -3.75
19-4 -18.75 +6.875 +14.375 +0.5
19-5 -27.125 +11.375 +20.75 -2.00
Total -68.875 +29.0 +64.875 -7.5
Average -22.958 +9.667 +16.219 -1.875
Adjust -0.263 -0.263 -0.263 -0.263
-23.221 +9.404 +15.956 -2.138
Seasonal variation -23 +9 +16 -2
7Assumptions
1
2 Each seasonal component ( ie each one of the four considere
3
8
What do these mean?
In a sufficiently long series, the residual influences tend to of
The seasonal variations are: Quarter 1 = -23, Quarter 2 = +9, Quarter 3 = +1
The sum of the 4 seasonals components for a given year is eq
Step 3: Arrange figures in the following order so as to ease the calc
YearQuarter
-
7/30/2019 8. Time Series Calculations
3/31
Knowledge of Seasonal variation is useful for firms to make planning
for eg. -23 means that economic conditions are such that the index of import
will be 23 points below
-
7/30/2019 8. Time Series Calculations
4/31
-
7/30/2019 8. Time Series Calculations
5/31
al to zero then we must adjust
=1.053
d in the example) is assumed to be constant over time
set each other
, Quarter 4= -2
ual to zero
lation of seasonal variation
=-68.96/3
-
7/30/2019 8. Time Series Calculations
6/31
will tend to fall below the trend, and on average it
-
7/30/2019 8. Time Series Calculations
7/31
-
7/30/2019 8. Time Series Calculations
8/31
-
7/30/2019 8. Time Series Calculations
9/31
1
2
3
Year Quarter Imports (X)Seasonal
variation
Series with seasonal
variation eliminated
19-2 1 114 -23 137
2 142 +9 133
3 155 +16 1394 136 -2 138
19-3 1 116 -23 139
2 150 +9 141
3 153 +16 137
4 140 -2 142
19-4 1 128 -23 1512 158 +9 149
3 169 +16 153
4 159 -2 161
19-5 1 137 -23 160
Data with Seasonal variation eliminated ( page 142 OJ)
How to Deseasonalise data (Additive model
Deseasonalised data = data with seasonal variation eliminated
Very easy to do: If seasonal variation is -23, meaning that our data is 23
below the trend line we must therefore increase it by 23 to eliminate se
Constant fluctuations tend to hide the underlying behaviour of the varia
comparisons ans assessment over time difficult to make hence the need
the data
-
7/30/2019 8. Time Series Calculations
10/31
2 180 +9 171
3 192 +16 176
4 172 -2 174
19-6 1 145 -23 168
2 194 +9 185
Important To read page 142 and 143 OJ
-
7/30/2019 8. Time Series Calculations
11/31
)
points on average
sonal influences
le making
o deseasonalised
-
7/30/2019 8. Time Series Calculations
12/31
1
2
Year QuarterImports
(X)Trend (T)
Seasonal
variation
(S)
Residual = X - T - S
19-2 1 114 -23
2 142 +93 155 137.0 +16 2
4 136 138.25 -2 -0.25
19-3 1 116 139.0 -23 0
2 150 139.25 +9 1.75
3 153 141.25 +16 -4.25
4 140 143.75 -2 -1.75
19-4 1 128 146.75 -23 +4.25
2 158 151.125 +9 2.1253 169 154.625 +16 -1.625
4 159 158.5 -2 +2.5
19-5 1 137 164.125 -23 -4.125
2 180 168.625 +9 +2.375
3 192 171.25 +16 4.75
4 172 174.0 -2 0
19-6 1 145 -23
2 194 +9
To read OJ page 144
Calculation of Residual ( page 144 OJ)
Residual = Original data -trend data- S
Residual = X - T - S
Calculation of Residuals (Additive model) ( page 144 OJ)
-
7/30/2019 8. Time Series Calculations
13/31
-
7/30/2019 8. Time Series Calculations
14/31
Year Quarter Trend (T)
19-2 19-2 1
19-2 2
19-2 3 137.019-2 4 138.25
19-3 19-3 1 139.0
19-3 2 139.25
19-3 3 141.25
19-3 4 143.75
19-4 19-4 1 146.75
19-4 2 151.125
19-4 3 154.62519-4 4 158.5
19-5 19-5 1 164.125
19-5 2 168.625
19-5 3 171.25
19-5 4 174.0
19-6 19-6 1
19-6 2
135136137138139140141142143144145146147148149150151152153154155156157158
159160161162163164165166167168169170171172173174175176177178
179180181182183184185
19-2
1
19-2
2
19-2
3
19-2
4
19-3
tr
end
value
-
7/30/2019 8. Time Series Calculations
15/31
119-3 219-3 3 19-3 4 19-4
1
19-4
2
19-4
3
19-4
4
19-5
1
19-5
2
19-5
3
19-5
4
19-6 119-6
quarter
Trend (T)
-
7/30/2019 8. Time Series Calculations
16/31
2
Trend (T)
-
7/30/2019 8. Time Series Calculations
17/31
1
2
Year Quarter Imports Moving annual total Moving pair total
Col. 1 Col.2 Col.319-5 2
681
3 192 1370
689
4 172 1392
703
19-6 1 145
8x = 8*176.5
= 1412
703 + w = 8x =1412
w= 709
2 194
8y = 8*178.75
= 1430
709+v = 8y =1430
v =721
3 a =198
4 b =184
The basis of the forecasting lies in our knowledge of the behaviour of tren
The trend will not suddenly change directioon
Forecasting from time series using observations which are read direc
trend line graph (ie estimating the position of the trend by eye and w
backward) page 146 OJ
FORECASTING METHOD
-
7/30/2019 8. Time Series Calculations
18/31
How to calculate a?
How to calculate b?
145 + 194 + 198 + b = 721
b = 721-145-194-198 = 184
172 + 145 + 194 + a = 709
a = 709 -172-145-194 =198
-
7/30/2019 8. Time Series Calculations
19/31
Trend
Col.4
171.25
174
x = 176.5
y = 178.75
ly from
rk
-
7/30/2019 8. Time Series Calculations
20/31
-
7/30/2019 8. Time Series Calculations
21/31
1
Year Quarter Trend (T)
19-2 3 137.0
4 138.25
19-3 1 139.0
2 139.25
3 141.25
4 143.75
19-4 1 146.75
2 151.125
3 154.625
4 158.5
19-5 1 164.125
2 168.625
3 171.25
4 174.0
2
3 Over the same period, the trend has changed 13 times
4 Average rate of change in trend is 37/13 = 2.846 units
5
Forecast trend for 19-6 Quarter 1 = 174.000 +2.846 = 176.846
Forecast trend for 19-6 Quarter 2 = 176.846 +2.846 = 179.692
Forecast trend for 19-6 Quarter 3 = 179.692 +2.846 = 182.538
Forecast trend for 19-6 Quarter 4 = 182.538 +2.846 = 185.384
Forecasting using average rate of change in tr
For the period over which the trend values are available, the trend has i
= 37 units
Future trend values can be predicted by adding the average rate of chavalues
The problem with projecting the trend by eye is that forecasts made usi
be consistent as opinions vary as to the future position of the trend
-
7/30/2019 8. Time Series Calculations
22/31
6
6.1
Year Quarter Forecast trendSeasonal
variationForecast index
19-6 1 176.846 -23 153.846 =154
2 179.692 +9 188.692 = 189
3 182.538 +16 198.538 = 199
4 185.384 -2 183.384 = 183
7
8
To read page 148 and 149 OJ
Many statisticians would argue that finding the average rate of change
whole period for which data is available may not be a valid method of p
value of the trend
The most recent data may be the more valid and some statisticians thinvalues only should be used to estimate the future trend
Forecasts for these periods can now be made by adding the appropriate
variation or by 'working backward'.
Adding appropriate value for seasonal variation to forecast import
-
7/30/2019 8. Time Series Calculations
23/31
nd
creased by : 174 - 137
ge to the previous
g this method will not
-
7/30/2019 8. Time Series Calculations
24/31
f the trend over the
edicting the future
that the last 3 trend
value for seasonal
-
7/30/2019 8. Time Series Calculations
25/31
1
1.1
2
3 Year Quarter Import Index Trend Actual/Trend
19-2 1 114
2 142
3 155 137.0 1.1314
4 136 138.25 0.9837
19-3 1 116 139.0 0.8345
2 150 139.25 1.0772
3 153 141.25 1.0832
4 140 143.75 0.9739
19-4 1 128 146.75 0.8722
2 158 151.125 1.04553 169 154.625 1.0930
4 159 158.5 1.0032
19-5 1 137 164.125 0.8347
2 180 168.625 1.0675
3 192 171.25 1.1212
4 172 174.0 0.9885
19-6 1 145
2 194
Year 1 2 3 4
19-2 1.1313 0.9837
4 Rearranging as below
MULTIPLICATIVE MODEL
The trend is calculated in exactly the same manner as for
additive model (using 4 Quarter , centred moving average)
X / T = S * R ( assumed to be insignificant)
X = T * S * R
Actual data = Trend * Seasonal * Residual
-
7/30/2019 8. Time Series Calculations
26/31
19-3 0.8345 1.0771 1.0831 0.9739
19-4 0.8722 1.0454 1.0929 1.0031
19-5 0.8347 1.0674 1.1211 0.9885
Total 2.5414 3.1899 4.4284 3.9492
Average 0.8471 1.0633 1.1071 0.9873
Seasonal
ratios
0.846 1.062 1.106 0.986
Adjusting by multiplying each average for each quarter by
4/4.0048, which gives :
-
7/30/2019 8. Time Series Calculations
27/31
-
7/30/2019 8. Time Series Calculations
28/31
-
7/30/2019 8. Time Series Calculations
29/31
1 To find a deseasonalised series, divide the actual data by the sea
Year QuarterImport
Index
Seasonal
ratio Deseasonalised index
19-2 1 114 0.846 135
2 142 1.062 134
3 155 1.106 140
4 136 0.986 138
19-3 1 116 0.846 137
2 150 1.062 141
3 153 1.106 1384 140 0.986 142
19-4 1 128 0.846 151
2 158 1.062 149
3 169 1.106 153
4 159 0.986 161
19-5 1 137 0.846 162
2 180 1.062 169
3 192 1.106 174
4 172 0.986 174
19-6 1 145 0.846 171
2 194 1.062 183
2 Using seasonal ratios for forecasting
3 Actual forecast = Forecast trend values are multiplied by the ses
Year QuarterForecast
trend
Seasonal
ratioForecast index
19-6 1 176.846 0.846 149.612 = 150
2 179.692 1.062 190.833 = 191
3 182.538 1.106 201.884 = 202
MULTIPLICATIVE MODEL ; DESEASONALISED SERIES pg 151 OJ
-
7/30/2019 8. Time Series Calculations
30/31
4 185.384 0.986 182.789 = 183
-
7/30/2019 8. Time Series Calculations
31/31
sonal variation
nal ratios
= 114 /0.846
=135