Time Series Talk : ARMA Model

[@th2315]

Your explanation and summary is much better and cleaner than my professor’s two-hour long lecture, much appreciated!

[@ritvikmath]

Happy to help!

[@sarthaksharma5860]

Bro i am indian nobody teached us these topics on any platform thanks i am watching your videos❤❤❤

[@damianjalaksa1518]

i love how you are explaining this topic with real world examples.

[@ritvikmath]

Thanks!

[@anishd7187]

Your really good at explaining difficult things, thank you!

[@vijayantmehla7776]

Thank you a lot for helping me understand this well.. I plan to see this entire series, its really well explained & in simpler terms. I wish you were my professor. Thanks again!

[@Blue17918]

You have the best TS course on KZbin! THANK YOU SO MUCH!

[@YueHuang_Olivia]

Thanks for the explanation!! Better then a lot of university lecturers!!

[@chaitanyabisht]

One of the most simple and concise explanation of ARMA model!!

[@ricmatestudiante3856]

Thanks!! With the pandemic, my time series analysis classes are getting very complicated, but here I am getting a good understanding of the ARIMA model. Thank you !!

[@vithaln7646]

oh my god , after lot of videos this is the clear explanation,

[@dirtyrickington]

hey ritvikmath i have a forecasting final tomorrow and its 2AM rn and im binge watching all ur videos.....i love u....love from Toronto Canada

[@sebastiancabrera7035]

You, my friend, are a lifesaver

[@TADIWANASHEMAKWANGUDZE]

how do i like this more than once......thanks man

[@surinderdhawan1061]

You made the things easy peasy for me. Thank you....

[@MrSocialish]

Bro is doing God's work in Crayola

[@esubah1]

wow, you broke this down so nicely. Thank you.

[@ritvikmath]

Glad it was helpful!

[@pan19682]

Your presentations are as clear as fine water. Thanks a lot for your help. Gongratulations. Would you mind presenting more videos in econometrcs models GLS models and more advanced.

[@ziwenwang1561]

Thank you for your explanation!

[@datax-analytviews8009]

I really love this, thank you

[@marccervera1896]

Great video and great explanation!

[@ritvikmath]

Glad you liked it!

[@tiantingwang2365]

OMG u r doing god work thanks

[@justusmzb7441]

I'd be very interested in how the regression of such a model is made. Probably not that crazy, but I am a little startled because the errors would probably be dependent on the coefficients.

[@mathijsgoethals4631]

You will never replace Ben! But, decent examples lad

[@HenriqMK]

who is Ben?

[@adrienl.6581]

You are my hero Thank-you !!

[@myprojectsdhaval701]

GREAT EXPLANATION

[@ritvikmath]

Glad it was helpful!

[@j.r.3049]

One thing that I didnt quite understand: Does the Order describe A.: HOW FAR you can look back (e.g. to the t-Pth value) or B.: HOW MANY TIMES you can look back (so e.g. Order 3 means there are 3 lags in the ACF/PACF that are different from 0)

[@niccolatartaglia3016]

Excellent explanation!! However, one note: I think the language you are using to describe the epsilon is not quite correct. In particular, in your MA model video (which is also excellent) you describe the epsilons as a white noise process but here you describe them as deviations from our previous estimation. I believe they are a white noise process (as you said in the other video) and not deviations from our estimate (since that estimation does not exist yet). Please, correct me if I am wrong.

[@jacobm7026]

Notice the way he defines the estimation. It is merely the model without the epsilon. That is, the estimated # light bulbs this month is equal to our estimation plus some error (which happens to be modeled as white noise). The estimation exists as soon as we decide to calculate it because all the information on the right side that lends itself to our expected value of the # of light bulbs for this month is known. The white noise is the epsilon/deviations

[@boxu2148]

Thanks for the video. What if the PACF show sig for 1 and 4, but not 2 and 3? What order should we give to AR?

[@ritvikmath]

Good question, it would be order 4 in that case, but you would not have terms for 2 and 3 :)

[@fatimetouhadramy2405]

thank you ❤❤❤❤❤❤❤❤ u'r life saver

[@milliekim5072]

Thank you so much!

[@reginacheong4596]

At the 6:00 minute, if the 2 ACF spikes are at interval 1 and 3, would the ARMA still be (1,2)? Are the input based on the number of spikes above the red dotted lines?

[@ritvikmath]

Good question, the order of the AR or MA part is based on the *last* significant lag in the PACF / ACF respectively.

[@prateeksharma9455]

Hi Ritvik, excellent video. Can we infer that AR part behaves like mu of MA (as you mentioned in previous vid) to get the baseline for which we want to smooth the errors ??

[@bts-be1sg]

brilliant

[@sanchitgoyal6720]

This is a basic question on Box-Jenkins MA models. As I understand, an MA model is basically a linear regression of time-series values Y against previous error terms et,...,et−n. That is, the observation Y is first regressed against its previous values Yt−1,...,Yt−n and then one or more Y−Y^ values are used as the error terms for the MA model. But how are the error terms calculated in an ARIMA(0, 0, 2) model? If the MA model is used without an autoregressive part and thus no estimated value, how can I possibly have an error term?

[@thebongscookbook2273]

nicely explained ! if you add same with real data on excel and then explain ARMA(1,1) it will be amazing !

[@Raaj_ML]

Nice explanation. But what about pre-requisites for ARMA like stationary , removal of trend and seasonality etc ?

[@williamstan1780]

I have a question, for a time series to make use of ARMA model, the time series has to be stationary right? If it is stationary, It means it fulfill the requirement of there is no correlations between current t to any previous time which means there would be near 0 for ACF. Then there wouldn't be any instant that it would be higher than the blue dot line right? Or am I missing something?

[@familienolte1501]

I think, when the model is stationary it just has a constant mean. Correlation can still be existent. Think of a sinus curve. It has a constant mean, so it is stationary, while it still has lots of autocorrelation.

[@FindMultiBagger]

Thanks 🙏

[@precisemeetaws112]

Question - is the L t-1 (the AR part of model) should be what I predicted for last period or what was the actual demand at last period??

[@TimelyTimeSeries]

I think it is the actual value. Firstly, the L t-1 does not have the hat notation. Secondly, we kinda assume that we already have our time series; we have a sequence of light bulb demand. From that sequence, we want to model the demand at time t.

[@pujasaxena8417]

Really good one

[@shadrackdarku8613]

great

[@florencee5407]

Thank you!

[@ritvikmath]

No prob!

[@AadityaMankar-sc1ux]

This will be ARMA(1,0) Model because ACF is decaying and PACF has a strong lag at 1

[@jokubasp6824]

The only thing I cannot understand: Why are there only error terms in the MA part of the model, where is the actual moving average? (as given in your previous video on the MA model, as μ). Do we assume it to be = 0? Thank you.

[@ritvikmath]

that's a good question! Notice the constant term beta_0. You can explicitly add a mu to this model but you can also assume that this mu is already incorporated into the constant term beta_0.

[@jokubasp6824]

Makes sense. Awesome, Thank you! Earned a sub today, was really helpful!

[@CharlieBingen]

Last video you talked about the invertability, so based on that, ARMA(1, 1) is equivalent to ARMA(infinity, infinity)?

[@omsonawane2848]

No, as here we are taking absolute values of previous lag values and their errors and not the infinite sum.

[@feng125]

Question about that last part of the video: 1) Are you running the ACF and PACF on the observed data or on the residuals data? 2) If my PACF shows a spike at 12 (eg: a certain month of the year has seasonally high demand), do i then set ARMA(12,1)?

[@Japuta666]

Ricky Chua hi Ricky, adjust the serie. In other words, you need a seasonally adjusted serie.

[@ritvikmath]

1) on your observed data 2) you probably want to use a seasonal model in this case!

[@martinak1723]

I fuckin love watching signal processing while high

[@space_ace7710]

Nicce!

[@ranitchatterjee5552]

When forecasting values, does value of e_(t-1) remains constant, if not how do we determine its value?

[@dr_ugly4497]

Yes it should since e_{t-1} represents the error made in time period t-1. If the lightbulb production volume was off by `e` last year, then it should still be off by `e` two years from now.

[@wirawoo]

good one, thanks

[@nemes1s_aoe]

Any way to calculate the suitable error threshold for ACF/PACF plots?

[@chloe-mariek2488]

"a time series in the wild" gets me every time

[@MiMi-zm2uc]

Thanks!

[@goodyonsen77]

Dude why the MA() order is 2 but not 1? What singles out 2?

[@AnirudhJas]

From what I understand, the moving average up to 2 terms is significant as shown in the ACF plot. Hence, the order is 2. Also, there could be cases where 1st, 2nd and 4th terms are outside the error line but 3rd term is inside. In that case, the order is still 4, the last significant lag. The coefficient for the 3rd term becomes zero, because it is inside the error line.

[@schopsell4299]

dude you are fucking awesome!

[@ritvikmath]

Thanks!!!

[@elpapi031]

In the example, if a good model, according to the given ACF & PACF, would be an ARMA(1,2), so, there is missing a term such as "phi_2 x Epsilon_(t-2)", ¿right?

[@omsonawane2848]

No, the term missing would be the current error . The term mentioned above will be contributes towards the equation.

[@AkashSingh-ed6vo]

For the MA(1) part why didn't you include mu value to calculate lsubt?

[@omsonawane2848]

Because the MA(1) model assumes average mean to be zero. Hence the term is eradicated.

[@CRockaell]

is there any way to use ARMA((1,3), 1) processing in R?

[@iskalasrinivas5640]

I think you swapped meaning of acf and pacf?

[@Jamesvandaele]

I think I once saw a time series in the wild. But I am not sure... I am not good at math and can't understand anything here... why am I here ...

[@spytheman]

University lecturers need to dissect ARIMA to AR and MA before diving to ARMA and ARIMA.

[@mikelmenaba]

You are a G

[@vadimkorontsevich1066]

1:40 the clue

[@calvinraab8798]

Thank you!