1
00:00:11,160 --> 00:00:14,080
Everyone, and welcome back to TIME series analysis.

2
00:00:14,530 --> 00:00:20,700
So this is the next section of the course and we are now going to focus on Arima in this lecture.

3
00:00:20,710 --> 00:00:24,040
We'll introduce this section and we'll give you an outline of what you'll learn.

4
00:00:25,440 --> 00:00:31,530
So the basic strategy we are going to follow in this section is this we'll start by learning the basics

5
00:00:31,530 --> 00:00:37,800
of the model, the model and the full Arima model is designed this way to get you coding as quickly

6
00:00:37,800 --> 00:00:38,550
as possible.

7
00:00:39,120 --> 00:00:43,800
Once you understand each kind of model, you'll be able to apply it to real data right away.

8
00:00:44,340 --> 00:00:46,400
Now, of course, that's not the end of the story.

9
00:00:51,070 --> 00:00:57,520
The remainder of this section will focus on the details and applications, so using a remote effectively

10
00:00:57,520 --> 00:01:00,370
is all about choosing the right Arima orders.

11
00:01:00,850 --> 00:01:05,590
Basically, you need to specify how many parts data points your model should depend on.

12
00:01:06,340 --> 00:01:09,900
The classic way of doing this is to use the ATF and the ATF.

13
00:01:10,390 --> 00:01:14,650
These are special plots that you can look at to directly choose your model orders.

14
00:01:15,250 --> 00:01:20,380
Now, one thing I really dislike about most times the resources is that they only tell you the rules

15
00:01:20,380 --> 00:01:22,810
for how to use the ATF and PKF.

16
00:01:23,200 --> 00:01:25,780
But they miss the most important part, in my opinion.

17
00:01:26,260 --> 00:01:31,270
And that is proving to you that these rules really work and that there is real logic behind why we should

18
00:01:31,270 --> 00:01:31,850
use them.

19
00:01:32,800 --> 00:01:38,020
So in this course, we're going to do an exercise that I've never seen in any time series course, which

20
00:01:38,020 --> 00:01:42,880
is to demonstrate that the rules for how to use the ATF and the ATF really work.

21
00:01:43,720 --> 00:01:49,030
This will involve running some simulations, just like how you do when you want to simulate stock prices.

22
00:01:50,230 --> 00:01:55,900
Doing this exercise will give you a better understanding of how to use the AKF and the passive compared

23
00:01:55,900 --> 00:01:58,170
to if you would just memorize rules for no reason.

24
00:02:02,890 --> 00:02:08,950
So in modern times, while using the Akef in the pickoff are useful, we can also make use of computational

25
00:02:08,950 --> 00:02:11,410
power in order to do model selection.

26
00:02:11,980 --> 00:02:16,720
We'll learn how to use auto arima and we'll see how that stacks up against more classical methods like

27
00:02:16,720 --> 00:02:17,410
the Akef in the.

28
00:02:19,650 --> 00:02:26,190
Now, Ottowa Rimma exposes us to a few more extensions of the basic Arima model specifically, we can

29
00:02:26,190 --> 00:02:29,240
also add seasonality as well as exoticness data.

30
00:02:30,570 --> 00:02:35,240
At this point, it's worth trying to understand how auto arima actually works.

31
00:02:35,640 --> 00:02:37,290
How does it select the best model?

32
00:02:37,530 --> 00:02:40,400
What criteria does it use to define what is best?

33
00:02:40,890 --> 00:02:42,470
So we'll discuss this as well.

34
00:02:43,230 --> 00:02:48,720
In statistics and specifically regression analysis, it's common to use the AC in the back.

35
00:02:49,860 --> 00:02:54,930
For some reason, statisticians really like to have lots of options all at once, even though they can

36
00:02:54,930 --> 00:02:55,760
only pick one.

37
00:02:56,130 --> 00:02:57,960
So we'll just have to learn about both.

38
00:03:02,560 --> 00:03:07,600
Once we've completed our study of how a remote works, well then apply it to even more data sets.

39
00:03:07,930 --> 00:03:10,340
We'll look at both sales data and stock prices.

40
00:03:11,260 --> 00:03:16,240
Now, stock prices are an interesting one because Arima gives us the tools to really understand what's

41
00:03:16,240 --> 00:03:19,420
going on from an objective and statistical point of view.

42
00:03:20,200 --> 00:03:24,220
So there are several important things we'll do in the section regarding stock prices.

43
00:03:25,360 --> 00:03:30,580
First will apply Otto Arima in order to see what the best model will be, at least according to the

44
00:03:30,580 --> 00:03:31,990
Otto Arima package.

45
00:03:32,830 --> 00:03:36,710
We'll also compare those Arima predictions with the new forecast.

46
00:03:37,180 --> 00:03:43,480
Remember that this is one of the most important things to do when evaluating your models will then apply.

47
00:03:43,480 --> 00:03:49,360
What we learned about the AKF and the pickoff to stock returns, that is, will apply the classical

48
00:03:49,360 --> 00:03:53,090
method to see what they have to tell us about the nature of stock prices.

49
00:03:53,800 --> 00:03:56,500
This will bring us back to the random walk hypothesis.

50
00:03:56,830 --> 00:03:59,260
We'll learn what our new analysis methods have to say.

51
00:04:03,920 --> 00:04:10,190
Another important topic in this session is learning about how a rhema makes its forecasts, this kind

52
00:04:10,190 --> 00:04:14,870
of detail is hidden from you because when you use a library, all you're doing is calling the function

53
00:04:14,870 --> 00:04:15,970
to get forecast.

54
00:04:16,250 --> 00:04:19,830
It doesn't tell you anything about how those forecasts are really made.

55
00:04:20,180 --> 00:04:24,710
If you've been taking my courses for long enough, you know that I often speak out against bloggers

56
00:04:24,710 --> 00:04:30,220
and marketers and Udemy instructors who make content about predicting stock prices with storms.

57
00:04:30,800 --> 00:04:33,140
This is mostly because they do so incorrectly.

58
00:04:33,410 --> 00:04:38,600
And the mistake that they make is that they don't understand how to forecast their plots cheat most

59
00:04:38,600 --> 00:04:44,220
of the time because they use true values from the future, when in reality those values would be unavailable.

60
00:04:44,750 --> 00:04:46,890
And if you don't understand why this is bad.

61
00:04:46,910 --> 00:04:48,740
Well, here are two simple reasons.

62
00:04:49,850 --> 00:04:55,310
Number one, if you take any code like that seriously and you try to use it in the real world, you

63
00:04:55,310 --> 00:04:56,330
will lose money.

64
00:04:57,200 --> 00:05:02,540
Number two, you'll be wasting time learning things which are incorrect and even worse, including those

65
00:05:02,540 --> 00:05:07,280
projects in your portfolio, which will make you look very bad in front of potential employers.

66
00:05:07,820 --> 00:05:13,160
So this lecture should help you do your due diligence to determine what is a legitimate forecast and

67
00:05:13,190 --> 00:05:13,810
what is not.