Autobox can allow users to specify a regressor
with a “1” where there is an outlier or a “0” where there is not.
If the user does not specify the outlier, Autobox
does a search using different combinations of 0’s and 1’s in order
To detect the point where an intervention takes
place.
|
|
||
TIME |
Y |
X |
|
1 |
.1 |
0 |
|
2 |
.2 |
0 |
|
3 |
.1 |
0 |
|
4 |
2.0 |
1 |
|
5 |
.1 |
0 |
|
6 |
.3 |
0 |
|
7 |
2.4 |
1 |
|
8 |
.2 |
0 |
|
9 |
.1 |
0 |
|
10 |
.1 |
0 |
|
11 |
.2 |
1 |
|
In the case where X is not suggested by the user there
are 2 outliers or 2 unusual values ( time period 4 and 7 ). Thus the model
would contain 2 dummy variables. However in the case when the user knows
about a causal variable ( point in time of a promotion for example ) AUTOBOX
would detect an outlier at time period 11. This would lead to a totally
different set of coefficients and a different comprehension of the nature of
the data.
|
|