stock prices are not randomly generated values Instead they can be treated as discrete time series model which is based on a well defined numerical data Items collected it successive points at regular intervals off time Thus time series analysis can indeed be useful to predicts doctrines Understanding this we have come up with this tutorial on time Siri’s and Stock market Ooh Now before we guy’s head off to the session I’d like to inform that we have launched a completely free platform called as Great Learning Academy We have access to free causes such as a iCloud and digital marketing You can check out the details in the description below with us I would like you to walk you through this Grace session objective We would be having a brief overview of the capital market what this is all about Second we will also look at certain kinds of a forecasting methods like the exponential smoothing and the Arema After this we will also be looking at some kind of an accuracy measures for all these models that we have developed And then let’s question this other really good enough Do they really work Do they require any kinds of fine tuning or not So friends what exactly comes to your mind when I don’t speak about the markets Something like You know a place where something is sold and purchase in exchange of the currency isn’t it Well if I correlate this fool our financial market we are referring toe the capital markets So when I say capital markets they’re well organized and they are organized into two segments One is the primary market on the second one is the secondary market Primary market is basically a place where you know the new securities are issued to the investors The investors are nothing but they’re the companies who wants to raise the capital as they have certain kinds of expansion plans and growth Secondary market is a place where I can say it as in a follow up to the primary market where you have all these instruments like stocks bonds options futures They’re all traded when I say treated they are brought and sold in this market No All these transactions which happened they needs to happen at one common platform These platforms are called as an exchange in India We have to exchange uh two major exchange As you say there is the first the first major exchanges BSC Bombay Stock Exchange and the second one is the National Stock Exchange NSC There are other exchanges which also bills with the commodities and old But if you see these exchanges they have more than 2000 stocks with them And however no If I want to know how my markets are performing I need tohave some check I need to have some kind of an indicator right So these indicators and I can say it as an index The index is what measures the strength on performance off these markets If I get back to the major in nicest uh in India they are Sensex which is college and sensitive Index on uh nifty nifty Nifty 50 Soon a 50 50 Why do this in a few 50 years It is a combination off 50 stocks and sense It’s as around 30 stocks with which forms a part of this index When I compare this to of the world in dices which we will be looking at today there are several such world World Index is like you know doh Jones which is off America Nikki off Japan Hang Sang off Hong Kong Dax off Germany Out of this will be looking at three off the index’s today So with this basic information I like to share certain very interesting story with all of you This story is all often investment decision which Waas made by my friend So please tighten your seat belts There are certain interesting things which are now going to come in here on your screens You see this This is a nifty 50 daily chart I have taken this from way back in 2000 2018 February 2018 Now this is the story off my friend who was very very passionate about this talk investing in the early 2016 he started investing good Lamsam off money He kept investing His mantra was very simple Buy on dips movement The market is down by it So he was following this mantra for a long long time Don’t know what happened His confidence started picking up Come in this January 2020 Come in the January 2020 and he saw that the you know the markets were making high rise high rise then came in January 19 when the market started falling and it was a fall like a falling knife The moment you’ve tried toe casual bleed that is what happened here Next I have shown this with one’s vertical line This is relates to around 20 years off March If you guys remember 24th off March Waas a day when a log down was announced in India Right now What What is the thing that comes to your mind when it’s log Log down Entire economic is down Country comes to a standstill What will happen to the markets Nothing But you know it will be a case off Just sell That is what happened here My friend thought that now this is the point where it’ll start sliding here Movement He saw this The movement he heard this next day morning he started selling what was there in this portfolio He cleared up his portfolio completely Zero with lost See the fund What happened post 24th off march The markets started increasing All right This was the time when he started scratching his head pulling his ass and despair He was not able to believe what was happening Nothing worked out frustrated seeing you know this is out of his range Friends Why did they married the story to us Around 95% off The people who are trading and investing in the stock market have a same kind often story There is a moral off The entire story which I narrated to you is if you want to be a good investor or a trader you have to be first discipline here Everyone off us need tohave a good strategy in place Also we have need tohave some good understanding off this markets We should have some good methods and know that can work out in this field of stock markets Stock markets are fluctuating like anything and I will show you this in in my coming slides So let’s toe in today’s session let’s explore what really works what really doesn’t work and what is that we need to fine tune so that it works in the stock markets I’m not claiming any method will be the best method but this is a first step towards forecasting So with this basics let’s explore the following markets These are the world in dices which I’m showing you here on my left This is the Dow Jones of America And on my right it is Nifty 50 which is Indian Market See here Isn’t it a mirror image Do you see This is a replica of each other Look at this This Waas 2008 which I mentioned to you the subprime crisis 2000 and 8 to 2009 The fall was so steep deep But they did take some time But look at the same fall here from 2000 and 8 to 2009 in the nifty right same fall Then let’s look at this The crash which we have seen here in the month of February in March Exactly Like a knife falling down Same way the Indian markets behavior Look at this the way it failure Okay so let’s look at the other end ices Okay I have taken this off Nicky Nicky is off Japan And look at this This is off Hong Kong Hang sank Now if you look at this um you would see that uh see this is 2020 and this is somewhere around 2007 So if I take back here 13 years back Okay I would see that I am at this Almost at the same place in 2020 Look at the hang sang again It was somewhere around 20,000 points Here it is just just about 20,000 points but very close to it Even after 13 years they are at the same place So don’t you think India is a good place to invest Can we get into the Indian markets now With this thought I I thought no Let me explore The Indian market doesn’t 50 50 looking at this If I draw a straight line I can easily see that it has some kind off grinning movement If I If I would have invested somewhere here post 2008 somewhere I think I can take it us 2009 It would show me that Yes it has a good amount of trend It is increasing And certainly my money would um and no compound if I get into this So uh let’s analyze this from our data science Percy what we can do Well how we can look at this eyes What I have showcased you in the next slide This is it Applying my data science techniques off forecasting We have two techniques One is ah the decomposition on the other one is the crema So in the composition of what I did here is I applied a steal Ah method I applied a steel method on it and I started seeing a decomposing this I wanted to see what are the components I wanted to see What are the components in this nifty 50 What I could see here is there are trains The Nifty 50 chart has trained the seasonality and the reminder No The point here is if I uh and I’ll take Trend Plus added to the seasonality and added to the reminder I should get my original data here Now there is something interesting coming out of this chart and that is the reason why I thought Let me show you this decomposition output Look at this 2020 is sharp right A sharp line here A big line Here Look at the others Okay The residuals air not showing it No What happened is when we decompose this picture the when we compose this what it is saying to me here IHS this is an extraordinary movement which I cannot which I cannot forecast the Now if if I look at this data science and when I look at the time cities especially which will be using this for forecasting there is a basic assumption that we all go go ahead with The basic assumption here is that what has happened in the past can only be reflected in the near future This is what the times it he says What has not happened cannot be forecasted And if I know what has happened in the past then I will know what has what will be in the present If I know what is there in the prison then I will be able to forecast what will happen in the future This is what exactly happened here The big fall is waas It was not able to digest and it said I can’t see this as a season seasonal pattern I can’t see this as a trend pattern and hence I am throwing this out as a reminder Isn’t it interesting Let’s explore more and more Let’s look at this This is the decomposition output Uh we got this We will also go through the r code Um I will Cuba and I’ll walk you through this Our code where we have all these outputs Look at this The one which I have marked this in the red I have marked this because there was a financial crisis which happened somewhere It started somewhere in the month off If I remember it rightly it was June July 2000 and eight and a street play It started falling across and it you know it continued till April 2009 I wanted to compare this with March 2020 Look at this It says there are 2234 points that it couldn’t capture on either as a seasonal nor as a trend if I remember it rightly There was a sharp fall off around 4200 points and somewhere down the line what it did is I picked up some of these values here If I look at this the maximum value was somewhere around seven and 68 It was able to only digest this because this is what it saw in the past and try to extrapolate it in the future But nothing because the fall was around 4100 points That is why it was not ableto recognize this Onda It said This is something random behavior which has had But look at the other things He January 2000 and eight The seasonal component 3.17 When I compare this to January 2009 that is what it says 3.17 When I compare is through January 2020 the seasonal component remains the same 3.17 This is how it starts on calculating and forecasting Let’s mourn toe the other methods So the first method which I’m going to take it for forecasting easy simple the exponents moving in exponential smoothing We will now be looking at a simple exponential moving method The reason this is what was taken is um wanted to explore this whether this really works If it works toward extent if it doesn’t work towards what ex What extent and what is that we need to do if it doesn’t work OK so before we get into that let me just give you a basic idea off how simple exponential smoothing Zehr developed The idea here is what is the forecast if I’m trying to forecast for tomorrow I have to know what is my forecast today Plus I should also know how wrong I waas today in my forecasting Now if I translate this my English in tow Algebraic equation This is what it comes out Toby Why had Teplice one which is nothing but what I am forecasting for tomorrow will be a combination off What is my actuals today Plus what is what is my forecasted value today But before I know I get further Let me just give you a brief here that whenever you’re doing any data analytics hear any data comes out with certain kind off noise The noise has Toby either eliminated or reduced This is where this component is helping me out Toe dampen I can better explain this by an analogy during this This is a classic case Off logged on period Uh we can correlate this toe the log down period We’re all working from home home here And there is a lot of noise isn’t it Noises from our kitchen No he’s from uh the Children’s around and etcetera Now what will happen You will not be ableto work The productivity will go down And what will you do for this You will taken action right All of us will taken action to cancel this noise Either toe dampen it or to reduce that noise The actions that we all of us will take is isolate ourselves from getting into one particular closed room or putting on headphones This is exactly what this Alfa is doing here It tries to edges that errors which are coming out off it and it tries to smooth that out The Alfa value here boost within the range off 0 to 1 and not go deep into how this view No car We have come across this algebraic equation but it is an interesting point to note here if I replace this Alfa by one by two Okay If I replace this Alfa by one by two it comes out Toby are one by two This coefficient comes out to be won by two into one by two That is run by four The third coefficient comes out Toby one by two into one by four That is one by eight So if you look at this the coefficients are won by 21 by four and one by eight What do you see The raids which I am giving toe these coefficients This past observations is decreasing as and when I get into the past they are decreasing exponentially and then forecasting it Hence this model uh this method is called as exponential smoothing it is exponentially smoothing the noise that is being caused by the data that we have taken here Okay so with this brief overview I will walk you through certain uh you know the R codes which will talk to you about what is the analysis that were done on the Nifty 50 For the sake of simplicity I uh and uh took two files so that the files do not get over burdened or burdened And you know we can follow this very easily So look at this The nifty 50 The libraries which I have tow taken here Uh these are the very common libraries which we require the forecast mood graphics data sets t Siris on GG plots deep losses for the graphics that we’re going to see here F pp toe the imputed ears Okay Now there’s impute ears and the X t s You can ignore this for the time being They impute two years if we use it If there are certain kinds off data which are missing in my data series Okay So fortunately the data which I have for the Nifty 50 is complete in full sense There are no missing data there So we can ignore this on this extra years is a library which we use if we have data in the Excel format Okay I have taken it in the CSP format here so and let me just run this now These are the invoices These are the all the world in Nice is that we have have stored it in their respective variables These are all the data has been stored in the serious We filed each off the world in dices are in differences Three files I’ll run this Okay What it says is I’ve just looked at the data for the nifty 50 What it says is there are 1 51 observations and nine columns in this nine columns I’m not going to work on the all the nine columns I’m going to take only one data of one feature One feature is the closing so you will see all the closing prices here The time series that I’m going to take is for the closing price off Nifty Okay now the data which will have is in the raw format I have to convert this into my time Siri’s object So we have a time syriza function which takes the raw data 1950 And I put it there on my fifth column which is my closing price I start with uh the data My data starts from September 2007 and that is the reason why I give this start here so that my data cities understands where I need to start Okay assuming that it has a frequency I give a frequency off Well here And let’s see how the plot looks like Okay So all the plots which has been designed where which have bean used here is using the DJ plots The G plots are fantastic to look at It gives some new kinds of scales here on it is very easy to recognize on There’s a reason why the DJ plots have been used Fine Now we have already seen this chart We also discuss about it that It has a trained component dance Sorry Trained seasonally And the reminder Let’s mourn This is the plots which we have already seen for the world in dices We’ll just run it off the same techniques that we use here We use a time series function created as the time series object I am plotted here Well we have seen all this in Isis Okay Fantastic I wanted to see what the start off this data is I wanted to see what the data the end of this data is and then the frequency let me run this Also I wanted to build a regression uh line I wanted to see one of how the Transat this will be useful even during her arema And that’s the reason why it has been plotted So looking at this data what it has come what it says means it starts from September 2007 and the end off the data is 2020 Okay I would have even taken April but I’ve just limited it toe march Okay now analyzing this time cities this is something very interesting toe Analyze and we will spend some time on this Here I am plotting a monthly blot for many 50 50 I’m looking at what the frequency is And what is the maximum observation off this Andi I am also going to see it What is a seasonal plot Looks like this is something interesting Toe and explore Let’s look at this and run Okay Yeah Good Look at this No the movement You see at this monthly nifty chart you would be wondering the means of the same right It doesn’t move it All right Second is the maximum values for January February March April It doesn’t It’s almost same here Visible Now we have seen many many of the examples where were taken in the class for example The GDP India if I remember it right is GDP India The tractor sales What do you What did you see That in North from January to February March April It used to increase Come June July It was to get stabilized here And then we used to see from other September off October November It is decreasing Yep There was some kind of seasonality Why It is not happening here and nifty Why don’t I see that in here in nifty right That is a question that might be coming into your mind So friends nifty is a combination off 50 stocks When I see 50 stalks they are picked up from several sectors When it’s a sectors they are coming from pharmaceutical sector They’re coming from auto set sector They’re coming from Heidi sector then oil and natural gas sector the consumption sectors auto sectors All these sectors combined Some of these talks from these sectors are combined to form and nifty index Now imagine every sector has its own Seasonally If I look at auto most of the sales happens in the month off November December January and somewhere you know in between whenever there are festive seasons take an example off Titan Titan is in mainly into jewelry jewelries and watches right Today’s actually that’s their right Now This is where you know if I look at the Titans talks you will see in the next quarter you will see its rising The stock values arising come November December where you have the valley The festive season’s coming There you will see the Titan When you look at the Titan charts it will show an increasing trend that is a seasonally Now if I come down to the other stocks you know from the consumption industry consumption is always increasing here and there When if I look at the other stocks so they will have the other different kinds of seasonality Now when I mix all this seasonality together can I get a regular Can I get Can I predict something Some seasonality off it No right There will be a bit here and there There will be a small seasonality but the seasonality will not be visible if that is what is happening here Okay let’s look at the seasonality plots Exactly what I spoke about The seasonal teas are uh there are so against their season Lt’s but they are not very much identifiable If I compare this to the stalks off industry and you nearly were Maruti or Titans off that safe came Fantastic Now moving on Let’s see what the decomposition talks to me about If I’m sorry when I use this still function okay I think we have seen this There are three components Is uh what we have analyzed I’ll just run this good Okay Have used both the techniques attitude method as well as the multiple to get you met Look at this We have seen this The decomposition What it says is this Nifty 50 has trained seasonally and a reminder moving ahead These are the values which I have spoken to you in my earlier slides about how the total values are being formed Okay if I either of my seasonality plus train plus reminder it should give me the actual values for 2007 Moving ahead Uh is our single exponential smoothing Here comes certain interesting points Okay so before I uh show you this chart and how it white has been designed this way I just want to put up a caution The caution here is we should not forecast for more than two or three periods The reason is until the stock markets stock markets are subjected to lots and lots off fluctuations the one which we saw in the march toe the march 2020 Okay so if you are trying to forecast for say even more than six months more than six time periods into the future it is not going to give you the right data which is being expected right So we will be on Lee forecasting for only two or three periods Now you might be things of why Why are we taking the 24 I just wanted toe give you comfort because this is how we do it In our class we always take either most of the examples that we have come across is a 24 So I just kept it to 24 here and also I wanted to see how much it deviates So let’s run this We have SCS That is simple exponential smoothing which we apply on uh the nifty 50 This is my time series I plot the output and I also see what somebody tells to me about Okay Okay This is it If you see here something’s very strange right It shows me a straight line and there is no common sense no common sense here right So I’ll talk more about this when we um and get into the triple explanations moving because all these concepts will be very clear there But anyway what we can see here is why it has given a straight line Is the very first question Second as what is my confidence interval that is suggesting to me what will be my forecaster values Third is why has it started from this particular point Why has it not started here or why has it not start at this point Okay so now look at this That’s simple explanations Moving method doesn’t take any trend nor the seasonality It says I don’t recognize what is the seasonality in nifty 50 I don’t recognize What is the trend off the nifty 50 that is one second it says I know only where I can start from That is what we indicate by something called as a level So this is the level which I’m talking about It says my initial state Okay what does it mean Initial state 50 to 1 So let’s look at this chart It says I know where I need to start This is the value 50 to 1 It says I know where I need to start from and then it starts estimating by taking one point estimate and that is how it starts building up its levels It starts building up its level It it takes the one passed observation and then it for and then it forecast This is how it comes toe this point It sees this level it estimates and it says this is the point where I need to forecast on That is the reason why you know it starts drawing a straight line Let’s look at the other component we I think I told you about the Alfa Alfa is a constant of which tries to dampen the noise What is Alfa telling me about Look at this Alfa Alfa is point 9999 Isn’t it very close to one Why is it taking 0.999 Why it is not taking 0.1 or why is it not taking 0.5 Right So what it is saying is things method is telling me that I see a very sharp fall Okay I see a very sharp fall here and I need to take care of the shuffle in order to take care of his chart fall I need to take a very high value off Alfa when I take very high value of Alfa means I’m going to take the very first observation in the past So what is my very first observation in the past This is the value And hence it takes up 0.99999 almost close to one It becomes more clearer when we see this in our triple exponential methods So I’ll move on Let’s look at this Double exponential in double exponential smoothing Uh we take care off the trend Okay Now I know what the level is I need to take care of the trend because my nifty 50 this place a trend with it to take care of that trend We include one more constant called as a beat A here Now you did not remember this alpha beta and all but it’s an interesting point which comes out of all this analysis we already have The method hold which does it for us automatically brings up what should be I’ll find beat up you even you not do it manually Right So let’s run this and we create an auto blocked off it by using our DD method Onda we also come out with somebody off Okay let’s look at this Please do not worry about this Def function this We will be talking about this in our auto Rima When we go for an arena for forecasting Okay Look at this Better than the simple exponential smoothing Know what it does is it doesn’t dry straight horizontal line but it draws a straight line with certain kind office Look it is telling me okay I recognize some kind of a trend Okay I recognize that kind off a trend but I will not be ableto but my slope will be very very small The reason being I see a very high deep here coming out off it Okay so look at this Let’s look at this I’ll find better values here This is what it is in the exponential smoothing simple exponential smoothing we saw it does 0.99999 point 96 Okay And Vita it is almost equal to 0.1 Now what it says is this is the slope with which I am going to forecast This is the slope with which I am going toe move ahead in the near future This this is what it says Hence instead of the horizontal line it has showed me a a straight line with certain kind of Hislop here but is it The right met Easy predicting the right of values as the question Obviously not if I show you the nifty 50 where it is currently it is in 9000 If I remember it it is somewhere around 9900 levels Okay at the point estimate Do not match with it But the positive point which comes out of this is if you look at the confidence interval This is the 80% confidence interval Okay this one is 95% confidence interval My forecasted values lie between somewhere between this 80% confidence in travel If I look at the Nifty 50 where it is currently for the month of April it is closely as I said it is closely around mind those in 900 levels Fine So it discloses somewhere here Look at this Curse is somewhere here so it falls within my 80% confidence interval But I want to again Buddha precaution Here we are in the markets the stock markets We’re not forecasting any Julie piece We’re forecasting the price of the stock market When you have forecasting the price of the stock market it means you’re putting down your hard earned money there Can we go with this forecast Absolutely not Okay this needs some good good amount of fine tuning activities to be that Okay let’s see Now Will goto the um uh that exponential but yeah before we go that there was some some other explorations which was done here because of this sharp spike and of what I thought is let me take this off till December 2019 Let me take it data and reduce my data toe December 2019 and run and see what happens Is it able to predict What does it predict Can Let’s see around This is the same Um we we use the whole metal here with my started and they ended Now the end it is December 2019 Okay let’s plot the sensi That’s fine of what it say this if there was no corona kind off a thing Okay if there was no corona kind off a thing the markets would have mood upwards to an extent Yes but stocks are stock markets that always uh towards the you know subjected to lots of fluctuations here in that Still it doesn’t take any seasonality components I could have imagined that there this line could have you know slightly come down So that because I am seeing lots of fluctuations here this is how it learns It starts looking at all these fluctuations from these fluctuations It learns that there is some kind of a trend that is kinds of seasonality I need to apply those on the red lines and then start forecasting That is how it should be Right Let’s see when we do this for our triple exponential smoothing here I’m using an attitude method Uh huh Same methods Here we have hold winter We have this function a call assholes Winter We apply on nifty 50 uh for convenience sake and forecasting for 24 months Better to do it for two or three months I’m not more than that Let’s also look like look quiet They what the models is telling me about uh I would be interested to look on what is my alpha beta and gamma values being taken here Gamma is again a constant a damping factor toe Reduce the noises which are appearing Ludo Seasonally bitter is a component which is a constant which is taking care off the noises to dampen because off the trend component Okay let’s run this See her Fine Okay so look at this It shows Yeah it shows some kind off a slope It says I think this is somewhere around Uh I should say around six months in the I should say somewhere I think July again somewhere here it says again is going to come down to the levels off off around 8000 8000 8200 or so The good part off it What we can say here is at least my current value If I go and seeing the Nifty 58 falls with within this 80% conference in travel the other thing is it is taking a seasonally component But these are the point estimates it will be a very difficult to match toe what they actual values are because there is a high amount off of fluctuation As I told to you this fluctuations it is not recognizing in neither as a trained nor as it ah seasonality And it is just throwing away as a random or the reminders there And that’s the reason why you know it is not predicting predicting what it should be Uh predicting those actual values Let’s look at the Alfa Beta and gamma the alphabet and gamma again Here the Alfa increases 2.99 It takes a very high value here Bitta It takes uh I think it’s 0.1 and gamma is 0.1 So these are the damping factors If I translated into status takes These are the damping factors um which are being applied One this is for Vita is being applied for the trend This is a damping factor applied for the seasonality Why this damping factors air coming so low because off that high fluctuations that it has observed Look at this s value Okay There are 12 values here now Each of these values correspond to the seasonal allergies Okay this is for January February March April It starts applying this correction How do we read this It says I am going to apply 38.14% off correction on my trend component Okay It says I’m going to apply 1 45 0.76% correction on the trend component on this is negative It means I’m going toe Dampen that I am going to dampen it and trying to bring down my trend level So wherever fluctuations you see there in the charts this is what it is also speaking about If you see an increasing their it is because off the trend and this 1 45% being applied there If you see some downfalls there this is where the negative value is working on it Okay No all these models also look at the A K value A i C c values Why are these values and off How would these values picked up this values of Victor Um by ah the method I’m looking at this acai values e i c c values archive values the lowest their values It says this is the best and the optimum value that I can predict with that values It starts taking of Alpha Beta and then it starts forecasting off the point point forecast for the 24 months Okay this is what all my data says No Um let’s look at this Waas an exploration which was done If what would be my nifty 50 look like If I take a reduced the data till December 2019 Well as good as what we have seen in the hold Ah method But I know it says that Yeah this is what the deep futures are talking about It says there will be some kind of a small dip and then it will increase There will be a small LaPierre increase and then it will take a bit small lip So this is where our seasonal it is also working out This would have been a case if coronavirus kind of a thing wouldn’t have happened Okay this was just from an exploration per se No um this is all What we saw is how the exponential smoothing works around Let’s see the other aspect The other aspect in the sense the what is my what are the other methods that I can apply Can I look at Arema as one of the methods Will It works really well the way it is Simple exponential smoothing is working out Let’s see this Let’s analyze that So all this time we were you know talking about exponential smoothing and the mean a criteria there Waas It was focused on the trend and seasonality in the data If he compares this toe over Arema model it is exactly the opposite Second is it alls deals with the auto correlation on before we get into the court Auto correlation I’ll certainly explain to you what that that is a lot about But if I give this native 50 chart toe Arema you know what the Arima model will say to me is this nifty 50 child has trained seasonality I don’t recognize what the trend seasonality Czar If I if you give me tendencies and lt I will not be able to project it No Arema works on a simple concept off stationery Stationery is something which doesn’t change Stationary is something which remains the same Now how do I do this and apply on nifty 50 This is it Let’s look at this This is a straight line This shows me the trend off 1950 Let’s look at this 2000 and eight skill and compare this with 2000 and nine Can I see the points The price off nifty is increasing with a slope Now let’s look at 9 4010 Can I say nifty is increasing with the same slope right 2010 to 2011 Can I say here Yes Nifty is going increasing with the same slope isn’t it Right So what is constant The constant here is the rate off change The rate with which the raid with with nifty is moving is constant Now the thing is can I convert this into a time Cities Can I take the rate off Changes a time Siri’s and feed this to the Arima model It can be done right This is where we apply a technique called us Different thing when I applied is different thing What do I mean by that Different thing is I will take one data point of 2000 and eight and 2009 and take a difference off it Right This will be my one variable from 2009 to 2010 I’ll take a difference off it That will be my second variable The third variable will be 2010 to these other than 11 I’ll take a difference off it That will be my third variable All this variable will have a constant increase This is constant Something which is constant Can be forecasted by Ah the imam Okay so what is this We have something called is a different thing We have a function called as D I f in our code I will also walk you through the arc word in a few minutes from now But I just wanted to show you this chart This is how it looks like when we apply the difference in technique on Nifty 50 The order which was applied to it was one When I say one means I am taking only one difference for each of the time period variables and make this stationary also by definition we have that these fluctuations should revolve around the mean the mean off zero Here Now the interesting things comes out off it And why do we Why are we doing this First thing is we’re forecasting for Arema right And I wonder data to be station of stationary But you may question this Why can I say this This is stationary Where is it Is it revolving around the mean It’s showing me such a big spike isn’t it This is the thought might be coming into your minds No The thing here is we can consider this as Sai Click Sigh Click is a periodic When I say cycle is a periodic means A periodic means it doesn’t repeat in regular intervals It means it is a random cycles are considered to be stationary So with this I can say that I have produced a chart I have come down to a level where my data for 50 50 is stationary Hold on There is one more concept before we mourn Okay that is off auto co relation auto regulation I talked to you about a few minutes back about auto correlation right water What is this Auto correlation is nothing But if we look back toe the linear equations that we have multi linear equation and all that And we said that there is a relationship between the independent variables on my dependent variable which is my response There is also another kind of relationship The relationship is a relationship between the variables the independent variables And we also uh had learned this technique This uh uh And we also I think we have come across the multiple linearity So whenever we have those multiple unities and all that the first thing is we we said is whenever we’re going to develop any kinds of models on all that we have to clear off all the multiple inanity which is their apply the same concept here Auto regression also talks about the same thing It is regression Regression means I’m trying to find out the relationship between the variables Why it is serious auto regression because it’s trying to find out the relationship with with its same variable When we take the times there is uh when we design any kind of a time series models Okay we are working in Univerity We’re not working on a multi Vary it We’re using my uni Vary it when it’s single variable but they are spread across various past observations This is what is happening here we are now um this is the variables are trying to regress on itself Okay The variables they’re trying to find a relationship with itself and that is the reason why we say it isn’t or toe a regulation This is the more equation that we developed to use our auto regression If you want to build a model for the forecasting okay I do not get deep into this algebraic equations because it will be spending more time on this My uh my more interest is and how it works for the Nifty 50 years Now the second thing is before we get into all that we also need to know what the orders off the auto regulations are Simple concept I’ll put it in simple in simple plain English Here If I say it is an order off one it means I’m going to take only one variable in the past That is the menace Well if I say this is order to I’m going to do it Two variables into the past But please note every data in time series comes out with certain kind of an error Okay this error is denoted by an epsilon T This forms a basic foundation for over moving average When we create models using moving average see what the statisticians know when they were developing all these models what they came across is can I take errors which are in the past taking those past observations build up a linear equation and predict and forecast the values right This is the model that they came across with this is nothing but the moving average with auto regressive errors Nothing if I combine this with the auto regression That is what the statisticians came across They said the accuracy off the models is increasing If I add these errors and add this errors toe my auto regression model my accuracy increases Isn’t it interesting that model What we and we were talking about is nothing but Armagh e r m a simple as that Our total regulation plus moving average Both models combined together as a remmy Now if you add the difference in technique which you know we discussed a few minutes back If we add that to this model it becomes Arema simple as that So we have a model The second model is a me and the third one is a army and the fourth one is e R I m a arema We are going toe work only with Arema Okay we’re not going to use any other We’re going to use only the Arema model now before we deep dive into that there’s one more concept which I want toe put forth which will be easy for you toe understand Going ahead as I told you there is a core relationship which exist the core relationship That this we can also see it as a direct relationship between these variables is what we call this auto correlation It is given by an auto correlation function What I need to see is when I develop any kind of a stationary models the very first criteria is that the variables in interest should have their relations Zero There should not be any relationship between them When I find my a c f going down to zero it means I’m making it independent at what value it becomes Zero is what my cue value is when a dollar been arema model I have 33 components three parameters Sorry It is P D and cue that is the queue which I’m talking about here Now there is another relation that we need to take care off is a relationship between this whiting and Whiting minus one having all other variables fixed or constant This is what I’m trying So this is kind of an air in uh indirect relationship which exists with my response variable Here this is called S in simple terms This is called It’s a PC of partial auto correlation Partial co relationship between the variables Okay Now at what value I can remove or reduce those values is what my P value will be defined here The value p is coming from P D Q Right This is what the of parameters which we’re going to use when we’re window have the Arima models in place Now with this let’s go on toe our record and see how you know we can apply for our nifty 50 Look at this I think we discuss about this I’ll just load this Libraries I have taken a different data set just for the sake of clarity Eso that you know they don’t intermix with each other I have created one more day Does it with a different values I’ve stored it in a different variable for analyzing the Arema here I have taken 1 51 67 in my earlier nifty when I was analyzing it with 1 57 Okay so I think for four or five months more have taken here Now I’m converting that into the time series object We have seen this I quickly run through this as we have already seen this Now Great Now a zay said to you Arema It works only on the stationary time series How do I see Can I visually look at it and say yes This they this time series is stationary I have one method I can do in his program Okay I use a history um function on my nephew 50 time series and I brought it Let’s look at this Okay here comes Look at this Don’t you see that data is skewed here towards my right Right And um the less amount of data is here on the left Can I say this history Graham is not a normally distributed when we have such instagrams which are not normally distributed I say that data is not stationary This is one method This is a visualization Now let’s look at the statistical way off looking at that Um here it is I also talk about the coalition Here it is This is this statistical way off looking at this very it is stationary or not Is he Is he It stands for auto correlation function I applied that on the legs Okay I’ll also talk about how we got this legs But just for a moment let’s look look at this graph What does it What does it say to me here It says that the lag lines are decaying but they’re dicking at a smaller place When we have these kinds often graph Here we say that the time series that we are analyzing is not stationary If I had this time Syriza’s a stationary then I could have seen just one or two legs here and immediately the line would have decayed And it would have been within this statistical limits anything any black lines if they fall within this status statistical a limit In that case we say that the the core relationship between the variables have being reduced or minimized Or we can say Is that insignificant Okay now let’s look at how we got this legs And what of this legs Look at this I have taken the data from December 2006 I replied this on a window using windows function I’ve taken that legs and I have put it use it G lack plot This is something very interesting year Okay this is going to form form over foundation forever Analysis look at this It has plotted for 16 past observations The first one leg line This is my very first time period in the past This is my very last time period in the past No At the very first glance it may look like I’m not able to read this out Right There is 16 year So that is the reason why I know if I’ll just explain the one in the 16 which will make you very clear Do you see a small line here It’s an diagonal line It’s a thin line A diagonal line here Now look at this The data points are spread across these lines They are not on these lines What it is telling me is the relationships are weak If I look at the lax 16 it says Maya the relationship between the variables are weak Hence it is away from the diagonal line Look at this like one Look at this All the data points which I have are distributed on this diagonal line The diagonal line is not visible but it has been overshadowed by all these data points Right So this is where What can I influence from this it say’s by the relationship is very very strong Now let’s look at all this Whereas in when I move into the past as I when move I move into the past I see that the relationships are decreasing So I can say that for I 6 to 7 lacks the relationships are very very strong Okay this is where I have tow Apply my A c f m p s e f To reduce the relationship or eliminate in practicality we will not be able to eliminate but we can reduce Okay with this we mourn to do an statistical test This is a status statistical test that we apply for our um time series Now the null hypothesis here is the time series is not stationary The alternative happens exist A time series is stationary I apply ADF argument ID Dickey Fuller test Okay We also have KPs is but we usually work on the idea fear What it says is the value P value is 0.3334 Off With this I can say that I cannot reject my help null hypothesis which means that the time Siri’s is um is non stationary So I need to make it stationary How do I do it I use it Uh order difference off one So I use that difference di ff function on the nifty I guess the first order difference in year Okay Mhm This is what it is I think I even talk to you about this so very quickly Move across So the idea of test Okay here it is Now I need to again see whether my uh you know data is stationary or not Statistically this was visually Now I need to see it statistically Let’s run this This is what it is 0.0 point 01 which is less than my p value of 0.5 So I can easily say Here is my time Syriza’s long stationary Let’s apply the A C f and the p A c f to see whether what is the core relationship Okay for a d is going to one value Look at this easier values all the leg lines They fall within this statistical limits I can safely confirm I can safely say that and confirm that yes the relationship do not exist between the variables Look at the partial relationship between the correlations they fall within this statistical limits And I can easily say yes they are independent off each other Right now we apply we apply to split method We can also apply three split method water These methods to split thirties I’m referring to the train in the deaths 33 split method is train validation and testing so that have a different valuation Techniques are not getting into it It’s I’m using the most simplest one to split method dividing my data into train and test Okay we have seen this in our next 50 training and best have this data into my training data set and my does data set Now I run the auto dot arema Okay On the training data set the concept here is I built a model on my trading data set and then I compare it with my test data set With that I start looking at water My accuracy levels Okay so with that yes it says I the Arema model with 010200 will be a good fit for the nifty 50 What is the 010200 this component Okay the first one which I’m talking about is for the trained you need to apply P D Q Values These are the best values you can select for the trained component and this values t 200 p D Q Values off 200 This will be the best field values for this season Lt How does it come out with it Looks at the U S E C value to 193 and it says yes This is uh the most optimum value that I can come out with and give you n um the model with which you need to proceed and apply on it 50 50 Okay no there are two things here I applied Auto Auto Arema There is a manual way off also processing that now Why you may say you know when when we have all the wife I should we go for the manual one Okay that’s the question that may come in into your mind So see here in manual uh Arima model will give you me Give me some kind of a flexibility I can take the p d Q values uh and find unit This is where you know I’m going to find to my models toe work on the nifty 50 Okay just before that um I get into that concept first let me look at the accuracy levels what my accuracy levels are telling me about So I forecast that on my training data set I stored that in one of the variables Then I forecast what is my the fork I planned on my forecasted values Okay then I take this test value and bind it with my training data set Why am I doing this I wanted to see how actuals um how actors work with the forecasted values Okay How does that plot look like So this is it This data we had take until I think yeah before casted here before casted from 2000 19 That is a July 2019 Right This is what it is giving me here Now the interesting borders Why I did it from 2019 is I wanted to see whether the mass 2020 is it crediting that March 2020 but obvious I just wanted to explore that uh it can’t predict that kind often uh downfalls but it can just give you somewhere 18 95% confident that you really can’t tell you what can be the fall but the again the actual values will not be compared Actually well it’s cannot be compared with the forecasted values but the good It is just taking up their strengths It taking of those seasonal please I can see it on fall here and that’s That’s it Don’t fall even in the march right So it is predicting that it’s predicting some kind of a downfall Now let’s look like how the actuals are looking when I compare with my forecasted values So here it is not very close right The moment I say this it’s not very close Now how How do I find doing this This is where our data science concepts on my data analyst data techniques will come in tow Picture what I need to do here as I have to take an arema model start picking up the Cherokee values Second I need to take training datasets a multiple training datasets multiple testing datasets Start measuring their accuracy ease Start comparing them Accuracy’s start averaging those accuracies and come out with the most optimum value this is time consuming It really takes a lot off efforts And second thing is sorry And this again ding us You have to party accord with with certain kinds off loops involved Er you have to loop in with different values You have to take permutations off PDQ values Run a loop there take lots of various data sets apply on it um and then come out with the most optimal value Then you may get something a very closer value toe What do you have estimated the forecast It will be very closer I still I don’t claim that you know the values will match There is no such model Maybe you know we have to go further deep dive and see there But predicting this forecasting coming out with a forecasting more model which can be closely related to his possible but closely matching with it 1 to 1 matching may be very very difficult Task Okay with that let’s look what the accuracy off over model is no talking about I’ll also explain this accuracy what they mean Let’s look at this map values Okay Here I know I have manually calculated the our embassy Okay root mean square values And I have also taken of the accuracy function which gives me all the values it gives me the MSC map PMP me uh then m a s c rms e and all that Okay let’s look at the interest off for us Would be mapping maybe And our embassy I’ll explain about this How do we interpret map is It is always in the percentage Here it is 12.6 What it says is it is 12% away from the actuals Okay That is why I was saying that it needs to be fine tuned You can put in your hard earned money there and look at saying that you know it is okay if my value is 12% of a 12% is huge in the stock markets 10 12 4% 5% is also huge Maybe one person here and then there is Okay maybe 1.52% is also okay You find something greater than 4% 5% is huge Okay because we’re dealing with money Having said all this there is one more criteria which we also need to look at toe say that Yes my data is stationary As I said to you there is a component also which is called the residual Sarandon’s We also need toe Have those checks in place Are there really independent If I do not have those independence my forecast values will be You know Hayward here in death and their model may not work And that is the reason why we also need to see what Maris cereals are talking to me about So we have a method of a function Callers check residuals this You can find it in the forecast 8.0 package Let’s run this on See what the outputs are Yes Good Fantastic The it looks Tracing reads revolving around the means Second look at this Easy off It says yes The residuals of the A C s are independent off each other Then it says yes This is what I was talking to you about It says that the students are more or less and normally distributed when we have more or less normally distributed It means the data is stationary So here I can confirm my students are stationary Okay now this is it This is all the manual way of which this was a part of the exploration which I did how it works I take the same values of P D Q and apply it I also try toe do lots of some you know fine tuning activities here And uh I don’t try toe see how my actual values correlate with my forecasted values so I’ll just quickly run through this Okay Because this takes lots of efforts There is a lot of fine tuning activities that we need to do here This uh the next one is the box test that we applied box This is this’ll is just another way I wanted to show this how we can apply the box boxes on the receivers here We another hedge Your the new ally bodice is is that the cereals are independent and the cereals are alternative Possesses is the cereals are not independent This is just to showcase I know how things wrong Once you get the r code you can play with this Andi I just wanted to see what the accuracy levels are for my um arema manual view for working on that on and this is Ah yeah No the other parties Now if say my data has lots off variance Okay Fortunately I don’t see a lot of variance because we have taken a monthly data If you would have taken a daily data off 50 50 I could have seen that we’re friends But I just wanted to explore INSEE Okay so what I did here is I have taken the difference off it Then I applied a log first I took a longer They make a log off the times it is now Why am I taking the log of times There is is this is what we call it the transformation of the data when it when I transformed the data I am trying to reduce the variance transformation We have several methods One is log You can use square root squares you can use Uh there are the other methods there I’m just trying to recollect that but yes they are log You can use cube root square roots and all that Yeah so So I use log on it Then I started putting a one order off difference on it so that I could make it stationary Is it stationary or not The plot will tell me about it This is a visual representation This is as good as what We saw it earlier Yes it it says yes It is stationary I use an idea of test a CF test and the PCF test to see whether it is stationary to see whether the relationship between the variables are independent off each other or not I run on it I see Yes all the lines leg lines lie between the uh the significance levels And I can easily confirmed that they’re independent off each other Okay And what I do here is Iran A manual arema on the log transformed data This is required if you have the Ryans in your data Okay this is being done here just in case you require this goto work on with the Daily data daily later will have lots of parents Okay this is we again checked on on just checks to the serious for the method that we came across Okay Now with this thought let’s move on to this lives here I I wanted to put away thought before you guys we have seen there the simple exponential smoothing the double exponential smoothing on the triple exponential smoothing right So and the Arema here see the the accuracy levels Okay The accuracy levels are you may say uh and on the holes method works fantastically value because the map the value is eyes very less 5.2 When I compared it to Arima and all that now it also depends on the data that you’re taking Second It also depends on the public values that you have taken for the Arema It also depends on the Alfa Beta Gammas Gamma values that you have taken for your holes are the holes of intermodal That is where we need to do more fine tuning activities and then try to come off with it The best values which will be for the map e and e r m a c So this is a question which you know I like toe put forth before all off you here is which is the best matter that I can work with the answer My answer toe This will be friends This is a stock market where you know we have to find tune As I said we have toe come out with the PD cues values do lots off other algorithms work on it and then come out with the more accuracy levels So what we have been here all this The explanation that we have done here is it is just a tape Often iceberg methods as I said can be fine tuned And as I also said to you we have to work on those Arima models where we have toe finding them to find the best particular values there The caution is Are there any other methods that we can work out with Yes we can We can explore that even in the deep learning We can explore this with a random work methods Random walks method in my view may work much better have not explored that but we can explore it Why The reason is random work method doesn’t work on the times that it doesn’t work on the concept that it needs a time series Rigid stationery it says Give me any non stationary data I can even work on it Most off the financial data’s that we have Um the random walk is the matter that is being works uh better on the financial later Okay so we can also explore that So since the financial markets are very unpredictable and they are about toe have several fluctuations the fluctuations that we have seen like the fluctuations which we have seen during the surgical strikes the Brexit vote the coronavirus March 20 that we have seen that can bring in huge fluctuations which will be really really very difficult Do predict from any kind of forecasting methods So I leave you with the thought There is a lot a lot and lot to explore further And um and the hope of um you will continue this journey for a new exploring many other methods or fine tuning the same methods that we have come across today Thank you very much Off with this I end my session open for any questions on they used time cities cross validation to strengthen the mode Yes Yes yes yes You can use that cross relation So that is where I was talking about cross validation in the sense You know I even spoke about the uh three split method right The three split metal that I was for Ah talking to you about waas The training the test validation the validation did us it And the test data Is it also you can What you can do is you can also use a 10 full validation or any in full validation You can do that Do ah lots of turning in it and then come out within our embassy and the map value which is the lowest that also may work out with Well okay so uh the next question is is this an editor Decompositional Multiplication on also when do we use on When do we use money Okay so this is an idea to method that we have used Okay The second is when do we use attitude When do we use multiplication Um then the variance is constant across any across equal time periods We use Eddie do metal If the veterans keeps increasing with certain rate at equal intervals off time that is the time we use multiplication method Can you please tell the idea if the data needs to be split as training and testing Yeah That data has to be split into training and testing That is a to split method We can also record with a three split method What I was telling is take a training data set Take the validation data set on the testing data set that is called a tree split matter Okay you can work out with that Yeah And one more point here is you have to have a huge data now uh observed that divide I took data from 2000 and eight I took it purposely The reason is from 2000 and 8 to 2009 that was the time there was a big fall in the market The prime crisis So in my model should also be able to learn those kinds of big falls No you make ocean find that it was a big fall Why did not it learn and then pray try to project it for Mars 2020 Now look at this 2000 and eight and 2020 It’s almost 2012 years into the past So when we apply all this simple exponential sorry those exponential smoothing medals and that it starts averaging that and it starts giving the highest value toe the most recent past And it starts giving the low weight ege toe the to the time period which waas in tow The somewhere around 10 12 years back The waiting is very very low there That is the reason why it was not able to learn If there waas is such a kind of a fall say around the SE four or five months back it could have learned that And it could have projected very closer toe what we’re seeing currently So the next question is can you explain about the fine tuning methodologies that we implement on the variance found in the results Off training and test data fetched Let’s take this way I’ll get into the record I think that will be easier Know what you need to do here Us on this uh and off these are my PDQ values For these are my pedicure values for the trend component These are my pedicure values for my seasonal component What I will do here is I will create a function Yeah I will write the political values I will loop it when I say Lou Pate I will loop eight for different values off P D N Q What do I mean by that Is I’ll take 010 This is my one data point uh data parameters Then the second one B 011 Then I’ll take 100 Then I’ll take 111 I’ll take 210 200 210 I’ll keep doing this and I’ll run it in a loop with that And also compare it with the seasonal component This is one way The second way is look at the data sets the training and the test It does it What I will do here I will create a different sets off data sets Know what I mean to say yes This is my one day Does it training that does it Next I will do iss Um I’ll take values off four here Okay And my training data said no Well a testing data set will start from fight This is one data set Then I will increase I’ll go back into the past Say this is 2018 I start from here 2018 and I’ll run this I’ll get anonymous e values Third I’ll take one more data set Say from 2017 I’ll explain you this or even on the graph 17 Okay I take This is my third date Assert I’ll come out with another embassy value This is what I need what we have to do by taking different different data sets and coming out with my RMC values Look at this Why am I doing I’ll just show this on the graph So see what I’ll do this from 2000 Say this is 2009 to 2012 I’ll take this as one day does it training and best second from 2012 to 2000 and say this is 16 I’ll like one molded does it from 2016 I’ll take from years somewhere I think 2018 I’ll take one more data set What I’m trying to do here is I am telling the model I’m telling the model Take this a recent take this recent down force Learn this reader learn this recent bald downfalls and build a model for me Know what is happening I took entirely it Is it from here So this particular point it was not able to read this It was not able to learn this because all the exponential models and all that we have learned it is it starts giving the maximum weight age toe my recent past not which has happened 12 years back So by diverting mint as data sets in this way it starts learning these false this fall this fall this fall it will also learn these are prints This trained this up trained this up trained So if I have those kinds of data sets with me and build up a model I can come out with better accuracy So that is why I said this is a bit off time consuming It really needs a good amount of a foot But yes it can be that so moving on to the next question is auto Arema more accurate compared to the arema function Now let me explain this concept Auto Arema Auto Arema doesn’t understand the business objective Auto Arema only understands what is the best value I should give it from our core Per se okay or again A repeat auto auto Arema doesn’t understand Doesn’t understand the business objective It doesn’t understand the statistical objective That’s it What is the statistical objective Give me the most optimized value No coming down to Arema This is where you have the control in your hands You being as a data analyst a data scientist You know what the business objective is So here If I compared to many of the 50 my business objective here is I have to get it more accurate because of this downfalls And that is why I explained you here I have I know from the business objective that it is not taking those falls which had happened from 2000 to 2009 It was not taking those falls from 2015 to 2016 Isn’t it my business objective to learn even this downfalls Right Because my business objective is to come out with the best accurate model that can also see the fall which has happened from March 2020 My auto Reamer doesn’t understand my auto Reamer doesn’t understand You know I have to take this falls Though it is full years back It just understands I have to take this Have to give more Weight is to this I have to give less waiters to this I need to give lesser lesser lesser lesser waited to all this So where does the business of judo come in Dividing their data sets into smaller smaller data sets So that you pick up all the uh prints and the down traits So there is no direct answer to your question Whether or Taurima Eyes best or the manual Arema is best manual Arema the works better because it is good Ah I can say this is my view manual Arema Because you have the controls in your hands from the business objective Percy So you need to find unit So it’s like So when you in depth analysis who will be the stakeholder So that is the first part of the question on the conclusion that is generated from this analysis So is it more genetic or more specific Like will it be good or to buy or sell sell shares with this particular death Just No no no no Okay That is the reason why I came up with a disclaimer there Okay The disclaimers hiss do not try toe by himself This this is not an is an offer being made If we can you know show this slide again Let me But that is something that is something which is any which way so present in the day So I think we can want to the next question Yeah so just to answer that So why this is being presented here That it is the presented here to give you an idea half off the methods that we have learned in the class How that can be applied on this Whether it works best or not is a call that you need to take here And this is being given to you so that you can find unit Okay Because this is where we can give it a ready made Onda There are there are several analysis If I go back there are technical analysis Also there are so many methods there You know people have devised strategies there lots and lots of strategies People have spent 55 years devising those strategies And that is why I said you know it’s very difficult to predict but it is not impossible but the efforts needs to be made in that direction That is where people have developed a very complicated very difficult They have spending 55 years toe build up a strategy which is working for them So to answer your question this will be just a start It’s deep often iceberg as I said to you This can be modified learned And who will be the stakeholder ITT’s just for the educational purpose This has nothing to do with any kinds of investing or ah taking your calls in the market Yeah Thank you That I hope that question is answered So the next question So how can you be 20 rise Is there any way to keep updating it Let’s say we plead to get forecasted Oh Okay Good Here comes again That kind off time Time series Can we do it weekly Mhm The problem with the weekly You can do it You have to do some some kind off modifications Now why am I saying that is uh that when it comes to the weekly data when it comes to the weekly data you also need to take certain holidays into consideration Okay The weeks may differ when I say for the month off This is April So month of April I may have four weeks when I go down toe Uh April May I may have five weeks So the time series when he applied time series conserves the week periods should be constant I can have four time periods in one month I can have five time periods in the one month Okay so you need to do some kind of a fine fine tuning What What Find training will you do there You have to label that you have to come out with some kind of a label ings So you have to give completely from 1 to 52 weeks irrespective of the months But then in your internal coding what you need to do Take that up Onda Uh you know give someone to four if you don’t have some kind of accord code where you say 1 to 5 weeks will be my month of January then next If you come down to February you may find anything around 3 to 4 weeks So you have to the next four weeks If you’re going to come in say 1234 then comes in 567 seven So 567 and going to label for February that fine tuning you need to do and then started playing it as a weekly basis So the next person you know that we have received this how to train the same model with different set off training data Yeah So exactly the way we did it here Okay so I think I showed it on the R code Okay so this is that window Okay How do you take different different data sets So as I told you here this is for 2017 I’ll changes to 2016 Okay this will be my one data set My second day does it I’ll change it I can change both month and the year I’ll take it to 15 2015 I know I’ll changes my test data set from 2015 This will be my second date Asked what will be method that does it I’ll take it from 18 2018 I’ll take it one more from 2000 twenties Data set from 2018 This will be my third data set So I need to keep taking from data sets and then start finding out the RNC values So even I’m in this exploration process I may do it in the near future and see because this is a very interesting subject to make it happen Okay And look at the Nifty 50 charts before we uh you know wind up You have to look at this disclaimer Uh because this is not a mechanism toe go in grade in the market Okay So this is only for the educational purpose how we can implement whether it works if how we can find donate This is from the learning Percy They kept Okay fine Sorry Ah no no no So there’s this uh there’s actually one question you know just received So yeah Does the Arema have a real time business application or is it more often academy exercise No no no Arema has a business application It has a business application Otherwise there wouldn’t have been the smallest developed They have their replications This brings us to the end of the story A lawn time series in stock market Now before you guys sign off I like to inform that we have launched a completely free platform called That’s Great Learning Academy where you have access to free courses suggest a iCloud and it’s still marketing So thank you very much Frightening the session and have a great learning
Free Prescription Drug Cards Coupons
