1
00:00:00,780 --> 00:00:05,400
In this video, we will discuss the moving average method of forecasting.

2
00:00:07,180 --> 00:00:12,700
But no, that it is very different than the moving average moving method that we discussed earlier.

3
00:00:13,480 --> 00:00:15,340
So don't confuse it with that.

4
00:00:16,410 --> 00:00:18,910
Now, let's see what moving average forecasting is.

5
00:00:21,820 --> 00:00:29,740
Let's say we have a time cities and we use persistance model or the auto regression model to make predictions.

6
00:00:31,330 --> 00:00:37,990
I hope you remember what these two models, the persistence model, is basically using the last available

7
00:00:37,990 --> 00:00:40,720
value as the forecast for next time period.

8
00:00:42,460 --> 00:00:49,330
And auto regulation is running a regression model on the lagged values and using this model to predict

9
00:00:49,330 --> 00:00:50,290
the future values.

10
00:00:52,180 --> 00:00:56,950
So suppose you have already found out the prediction using any of these methods.

11
00:00:59,070 --> 00:01:02,520
Now, let's say the actual value of the time P plus one is Y.

12
00:01:03,250 --> 00:01:06,910
And the predicted value or time P plus one is Y hat.

13
00:01:08,470 --> 00:01:12,160
So after applying the model on the input, we have the value y.

14
00:01:14,290 --> 00:01:22,450
Another difference between Y her and the actual value Y is called residual or the forecasting at a.

15
00:01:24,480 --> 00:01:30,990
We can find all the residuals where we know both the actual values and the predicted values.

16
00:01:32,980 --> 00:01:40,590
Ideally, these set of residuals, that is the time series of residues should be white noise.

17
00:01:41,610 --> 00:01:45,150
There should not be any pattern, indeed, residual values.

18
00:01:46,320 --> 00:01:49,800
Ideally, there should not be any information left.

19
00:01:49,920 --> 00:01:50,820
Indeed, as it was.

20
00:01:52,620 --> 00:01:55,620
But practically, this may not always happen.

21
00:01:57,090 --> 00:02:04,870
Our model may leave some information, indeed, as it was, to extract that information and improve

22
00:02:04,870 --> 00:02:05,650
our forecast.

23
00:02:06,970 --> 00:02:09,670
We don't not forecasting model on residuals.

24
00:02:11,650 --> 00:02:14,020
We tried to forecast the future residuals.

25
00:02:15,220 --> 00:02:22,330
If there is any pattern individuals, this second level model, we are calling it the second level model.

26
00:02:22,810 --> 00:02:25,110
First level was on the accurate time series.

27
00:02:25,750 --> 00:02:27,340
We found the residual from that.

28
00:02:27,640 --> 00:02:31,360
And now we are building these taking level model on these residuals.

29
00:02:32,860 --> 00:02:39,670
So if there is any pattern, indeed, residuals of second level model will identify it and capture that

30
00:02:39,670 --> 00:02:42,130
information in the forecast of residues.

31
00:02:42,370 --> 00:02:49,180
We can then add back these forecasted residuals to our original forecast to get better forecast.

32
00:02:50,350 --> 00:02:52,420
This is the concept of moving average.

33
00:02:53,240 --> 00:03:00,040
I've listed down all the steps that are used in the moving average model that is initially we use air

34
00:03:00,100 --> 00:03:01,960
or persistent model to build a model.

35
00:03:03,840 --> 00:03:05,280
We didn't find it as it was.

36
00:03:05,850 --> 00:03:09,660
They are more along these networks and forecast these residuals.

37
00:03:10,850 --> 00:03:15,410
And lastly, we use the forecastable tools to update our initial forecast.

38
00:03:16,490 --> 00:03:21,770
This moving average method is often used along with auto regression.

39
00:03:23,270 --> 00:03:27,530
That is, auto regression is used as the first prediction model.

40
00:03:28,700 --> 00:03:32,300
And then moving averages used to update the initial predictions.

41
00:03:34,300 --> 00:03:40,730
When auto regression and moving average model is used together, it is called auto momentum.

42
00:03:43,270 --> 00:03:52,120
This model is quite popular, although I do remember that Ottmar can only work on cities without print

43
00:03:52,240 --> 00:03:53,110
and seasonality.

44
00:03:55,630 --> 00:03:58,070
So do remember to remove that before playing.

45
00:04:00,520 --> 00:04:06,790
In the next lecture, we will see a variant of AAMA, which automatically handles trend and seasonality

46
00:04:06,790 --> 00:04:07,210
also.

47
00:04:10,220 --> 00:04:17,480
So that is all this is moving average method, if you like, residual values are used to forecast.

48
00:04:17,700 --> 00:04:19,340
There's Duell for the next time period.

49
00:04:20,150 --> 00:04:27,230
This forecasted residual is also ordered in the original forecast to get a new and improved forecast.

50
00:04:28,460 --> 00:04:31,210
Now, let's see how to implement this in our software.
