1
00:00:01,080 --> 00:00:09,180
Now, let's discuss the Atima model are stands for auto regression, integrated moving average.

2
00:00:10,410 --> 00:00:13,860
You know, about auto regression and the moving average part of it.

3
00:00:15,090 --> 00:00:16,740
What does integrated stand for?

4
00:00:18,330 --> 00:00:23,220
This method is called integrated because it includes the method or different thing.

5
00:00:24,870 --> 00:00:27,010
You may remember different thing from previous Lichter's.

6
00:00:27,420 --> 00:00:32,130
Different thing was used to be seasonal seasonals and data and data.

7
00:00:34,290 --> 00:00:40,560
And I've been repeating this several times that auto regression and moving others models should not

8
00:00:40,560 --> 00:00:44,490
be applied on cities, which has trend and seasonality.

9
00:00:45,900 --> 00:00:52,110
So to handle Trinian's seasonality, we integrate are Almo model with the different method.

10
00:00:54,090 --> 00:00:54,980
The idea's simple.

11
00:00:56,370 --> 00:00:59,490
We will first remove trend using the defensing method.

12
00:01:01,710 --> 00:01:05,860
We may have to do defensing several times if we are trying to remove non-linear trend.

13
00:01:05,970 --> 00:01:14,670
Also, once we have been rendered, we apply auto regression on this new series to find the initial

14
00:01:14,670 --> 00:01:15,660
set of forecast.

15
00:01:17,660 --> 00:01:23,280
Using these initial forecast, we find it is a duals or forecast errors.

16
00:01:24,770 --> 00:01:27,350
Then we apply moving average method on these days.

17
00:01:27,350 --> 00:01:29,240
Dools to update our forecasts.

18
00:01:31,010 --> 00:01:34,180
In the end, we reintroduce the effect of trend.

19
00:01:34,580 --> 00:01:42,260
By doing these different thing that is adding Bagdadi lag values to our forecast to include deterrent

20
00:01:42,260 --> 00:01:42,590
effect.

21
00:01:45,320 --> 00:01:51,650
Now, when we will be applying Atima in our software, we will have to specify three parameters.

22
00:01:53,300 --> 00:01:58,970
These three are parameters, letter to air, that is auto regression, ie.

23
00:01:59,570 --> 00:02:00,710
That is integration.

24
00:02:01,490 --> 00:02:03,500
And Emmy, that is moving average.

25
00:02:04,130 --> 00:02:07,700
These are denoted by B, B and Q.

26
00:02:10,070 --> 00:02:12,280
The first parameter is B.

27
00:02:13,580 --> 00:02:16,850
It is also called the order of auto regression.

28
00:02:18,320 --> 00:02:24,290
This basically denotes the number of blagged values we are going to use in our auto regression model.

29
00:02:26,180 --> 00:02:34,580
If we are using only P and T minus one values to forecast P plus one values, then B is equal to two.

30
00:02:35,900 --> 00:02:39,020
If we use only P value, then B is equal to one.

31
00:02:41,240 --> 00:02:42,660
Second parameter is D..

32
00:02:43,640 --> 00:02:45,770
This is the order of different thing.

33
00:02:47,450 --> 00:02:53,160
This will tell our model how many times different thing is to be done to remove trend.

34
00:02:54,200 --> 00:03:01,720
If we are seeing a quadratic print, we may do defensing ways, in which case B will be equal due to

35
00:03:03,520 --> 00:03:05,840
bodily nutrient single.

36
00:03:05,900 --> 00:03:08,000
Different thing is sufficient.

37
00:03:08,480 --> 00:03:09,980
In which case we have these equal do.

38
00:03:09,980 --> 00:03:16,760
One last parameter is Q which is called the order of moving average.

39
00:03:18,410 --> 00:03:25,950
This is basically the window size of the desert world that we will be continuing to forecast future

40
00:03:25,960 --> 00:03:32,870
residuals if we are forecasting residuals on the basis of last three residual values.

41
00:03:33,470 --> 00:03:34,750
Then you is equal to three.

42
00:03:35,720 --> 00:03:39,230
If we are using last two residual values, then Q is equal to two.

43
00:03:41,240 --> 00:03:49,400
So by specifying these three parameters, our software will know what exactly is to be done to implement

44
00:03:49,520 --> 00:03:49,860
Atima.

45
00:03:52,250 --> 00:03:56,120
But how do we know what are the right values of BD and Q?

46
00:03:58,430 --> 00:04:03,680
One way to know is by using our experience and understanding of the problem at hand.

47
00:04:05,330 --> 00:04:13,010
For example, if we know that our product shows a quadratic print, we can straightaway said B is equal

48
00:04:13,010 --> 00:04:13,600
to two.

49
00:04:15,680 --> 00:04:24,080
The second way is by running a software iteration and finding which set of values of B, B and Q give

50
00:04:24,200 --> 00:04:26,210
the maximum estimate accuracy.

51
00:04:28,220 --> 00:04:34,970
When we do not have large data and there are limited number of options for BD and Q, this can also

52
00:04:34,970 --> 00:04:35,290
be done.

53
00:04:38,230 --> 00:04:39,530
So that's all about Rhema.

54
00:04:40,660 --> 00:04:42,730
See you in the political class of Atima.
