ACF & PACF Code Example : Time Series Talk

We look at how to generate ACF and PACF plots for ice cream data and stock price data!
Code and data available at my GitHub:
github.com/ritvikmath/Time-Se...

[@JawadLion]

You are ridiculously helpful and so underrated.

[@ritvikmath]

thanks!

[@hassamdaudi]

I second that! You're saving my assignments man, and 100x better than my professor in your explanation

[@shrutipandey8696]

Thank you so much for uploading these lectures, I didn't understand time series this clearly up until now

[@NFZ602]

Thank you for posting such practical videos and examples. Appreciate you!!!

[@alithmajok]

Very significant to my thesis thanks for your presentation.

[@mariasmoczynska8937]

Thank you so much for these videos! incredibly helpful

[@user-mw7uj8rj1c]

HI recently I started my study about time series and I did not find any material where multiple kinds of ACF and PACF have been discussed. I found it very difficult to corelated my ACF and PACF plot but after reading and watching things from multiple places, I am able to figure out about my plot. I request you to bring more example so that people can corelate there problem because the way you explained the plot is very easy.

[@MrCEO-jw1vm]

I started this playlist out of curiosity to get an introduction to time series analysis since it's a term I have been hearing for a while now. Now I feel like I will finish this playlist in no time. You explain things really well, and you have just showed me awesomely how to do the code of time series analysis which I understood very well. Great job, man!! We very much appreciate it!

[@bjoernaagaard]

Do you know how one would remove multiple seasonalities in Python? For, say, electricity consumption. Thanks!

[@zsomborveres-lakos]

I like your content and I leave a comment for the algorithm :)

[@nagusameta366]

I got one line each for the PACF and ACF who are above the error bands, while the other ones are literally drowning in the blue area, is that good enough to do AR(1) and MA(1)? It's a dataset which have dates for project implementation through the months of a year, where some are just 0 meaning there was no project in that month

[@acb_gamez]

Worth noting that pandas as a diff() function to simplify making a difference column. To replace what was done in the notebook: df["FirstDifference"] = df.Close.diff(periods=1). This will leave you with the first row containing Nan values, so either call dropna() or fillna(0.0) on the output.

[@PeterElverson752]

Is there a reason why this throws a SettingWithCopyWarning? I don't see where there is chained indexing.

[@saip7258]

how to identify seasonality in timeseries data because when we make data stationary we are removing trend and seasonality so acf and pacf how they are use ful in determing seasonality or any other methods to identify seasonslity in data. thanks in advance

[@BBB_025]

You mention tools other than ACF and PACF to better define the order of the model(s). These other tools were not covered in previous videos correct?

[@hosseinmortazavi5547]

@ritvikmath thanks for great movie it clearify the previous video alot. I am wondering as the Ice Cream Dataset is not stationary why in the video you did not statinarize it first like the stock price dataset first?

[@Juan-Hdez]

Very useful. Thank you!

[@varunyavagal579]

Where is the csv available?

[@algorithmo134]

@ritvikmath import yfinance as yf does not work? Do you know why? I installed pip and everything but still it doesn't work

[@qiushiyann]

Great video!

[@vipoplekhachinabutr57]

You’re so helpful.

[@garyatate]

SUPER helpful!!

[@redcat7467]

YF gives error. Anyone knows the way around the issue?

[@wangjessica1275]

When do I know I need to use both AR and MA, how will the curve for ACF and PACF look like?

[@sohailhosseini2266]

Thanks for the video!

[@harjotsodhi1806]

Very helpful!

[@Gauloi007]

Hey very nice video, my question about this your model you use y(t)-y(t-1) to make the series stationary, but why not normalized the series by (y -ym)/ sigma(y) use a Indice sp500 normalized in the same way and make the différence and try predict this difference... And now we have portefeuille of Y and short sp500... And we do our time series stuff... Fama did even better with his line market and the zero risk interest,but in fact it's hard what is the zero risk interest,but find the hyperplan of some economy cotation, like us1y, us2(bond), brent, gold, gaz,dxy why a PCA method keep the biggest eegs , and keep the normal with the plan as stationary. variables....

[@karannchew2534]

Why does the autocorrelation diminishing? It look likes the 'pattern' keeping repeating beyond the initial periods i.e. stay corrleated. Shouldn't the autocorrelation be as high as the beginning?

[@_isDev]

before plot ACF or PACF we must make the time series stationary? Awesome videos!

[@walterreuther1779]

10:56 Seems very non stationary with regards to covariance, probably this is why we're getting significant lags in acf and pacf, right?

[@godse54]

Nice work super helpful

[@ritvikmath]

Glad it helped

[@user-wr4yl7tx3w]

But if you take the first difference, then won’t you model be predicting the first difference?

[@anirudhm7058]

very well done

[@hongkyulee9724]

Thank you for the wonderful insight ! >_< !!!

[@rajarams3722]

Why did you not make the Ice cream data stationery ? It had seasonality.

[@user-cy9zf4oz2i]

it's like partial derivation concept, pacf is disregarding the confounding effects

[@ggpopa1319]

Amazing!!!

[@nickkoprowicz4831]

Why didn't you have to remove the seasonality before plotting the acf and pacf in the ice cream example?

[@bahmanrostami-tabar6733]

Because the time series is stationary, if your time series is non-stationary you should look at ACF and PACF after making it stationary by differencing or transformation

[@kunal_tajane]

Amazing video and explanation. One query - when you said their ACF is going to zero means that it is the AR model and MA model. Is that always the case or since we could see the seasonality in the original chart.

[@user-mw7uj8rj1c]

As per my understanding if you are observing sinusoidal and decreasing graph in your ACF then you can consider AR(p) model

[@cosmicbandido8186]

actually I would have prefered a interpretation of the acf and pacf plots. What does it mean that at lag 1, 3, 10, or what ever has a higher correlation ?

[@user-qk5jd3os8i]

It implies that either lag 1,3 or 10 has more predictive power than other lags. Though you should also be aware of the fact that the correlation between different time points of one stock can be caused by randomness. The higher the value of ACF or PACF the better.

[@mansivarshney7998]

I've a question... At 2:35 why you used pd.datetime again here for start date even when you used it before on whole data?

[@sgpleasure]

That step is to only show time-series data from 2010.

[@chiragbakhai6918]

Hey rithvik, How is your PACF lag 22 is going to -1.5. PACF should be bound by 1 to -1??

[@florianwendler375]

I have the same question

[@chaitanyapatil6839]

Hi..You mentioned in your MA model video that to use MA model, we should check ACF and it should be decaying. It will decay for k > q. But here you mentioned that *since we are observing decaying ACF we will be using Autoregressive*.. am I missing something?

[@dhruvpatel4948]

Yes, that part was quite confusing. Does anybody have any explanation?

[@asankapradeep]

The ACF and PACF plots should be considered together to define the process. For the AR process, we expect that the ACF plot will gradually decrease and simultaneously the PACF should have a sharp drop after p significant lags. To define a MA process, we expect the opposite from the ACF and PACF plots, meaning that: the ACF should show a sharp drop after a certain q number of lags while PACF should show a geometric or gradual decreasing trend. On the other hand, if both ACF and PACF plots demonstrate a gradual decreasing pattern, then the ARMA process should be considered for modeling. Ref: towardsdatascience.com/identifying-ar-and-ma-terms-using-acf-and-pacf-plots-in-time-series-forecasting-ccb9fd073db8

[@Yan-vt4sh]

你是我的神！

[@abinandhans5386]

Thank you

[@Didanihaaaa]

Hi, thanks for your great channel. so what I learned here is as follows: my question comes up at the third question. 1 if the data is stationary then subtract one lag to eliminate the stationary data. 2 apply ACF and PCAF to find out dominant lags 3 once the dominant lags were found we can apply AR or ARIMA model to original data (before subtracting lags, Non-stationary data) for fitting and prediction?

[@mikelmenaba]

Fire