|
|
Hi,
I'm trying to do forecasting given some initial data. The data that i
have is a little bit tricky and i'm not sure if i can use it for AR.
The data is as follow:
at time t=0 my value is always 1.
At time t=1 I compare my data to time t=0, and get some number that
describe precision, say 0.9.
at time t=2 again i compare my data to time t=0 and get the precision
of say 0.87
i'm doing the same for t=3 and t=4.
In addition, i can get also the precision when t=1 is the base of
comparison. so i can get the precision at time t=2 comparing to time
t=1 (say 0.95), the precision at time t=3 comparing to time t=1 (0.91)
etc....
To conclude i have different starting times t, and for each t, i can
get a set of values (precisions) comparing to the specific t.
as an example, this shows the precision for 4 different starting
times.(t=1,2,3 and t=4)
Time precision
1->1 1
1->2 0.9
1->3 0.87
1->4 0.85
1->5 0.81
Time precision
2->2 1
2->3 0.95
2->4 0.91
2->5 0.86
Time precision
3->3 1
3->4 0.92
3->5 0.89
Time Precision
4->4 1
4->5 0.97
Now, what i want is to forecast the value at time t=6. As you can see,
first i need to fixed the starting time. so let's say that i want to
forecast the value comparing to t=4. How can i do it?
Notice that I can't use the regular AR model (as much as i understand)
since in the AR model of say AR(2) the function is like
Y(t)=a+bY(t-1)+cY(t-2) , so it means that the value of t=5 for example,
is a function of t=3 and t=4, where in my data the value of t=5 can be
comparing to 4 or comparing to t=3, but they are from different
series....
any ideas? suggestions?
thanks a lot,
Kiwi
|
|