Granger Causality in Python : Data Science Code

[@gabriellaurin2717]

You're simply the best ! I've learned so much from your videos and they have helped me get through my time series and forecasting courses.

[@ritvikmath]

I'm so glad!

[@stets43]

Your videos are informative, concise, and life saving. Thank you!

[@bibah6]

But the important question here is how do you select for the optimal lag? lets say you do the analysis for 30 lags and 15 of them are statistically significant, then which methodology could you use to pick the best result

[@mostrotorino]

Great video! Do we need both the time series be stationary? Thanks.

[@vickyyang8659]

your video is the best to present Granger causality, thank you for sharing.

[@conradobittencourt662]

Thank you very much for this concise and high quality video!

[@apoorvgupta07091989]

Do we need the series to be stationarised

[@ফকিরতালিব]

Problem faced and solution derived: Uning this code with alpha of 0.5, I get the second lag significant at less than 2% levels in almost all runs. Ritvik did not run the code in the video which could not make me confident. Using an alpha of 0.9, I get all the lags significant with p values close to zero. I think this is because the data has unit root, which we should avaoid. So, if we take the first difference of the data to make it stationary and then run the same causality test on the differenced data, it gives consistent results always. If the lag used is 3, it show the third lag is important and the first two lag are not. This is consistent with econometrics rules.

[@권-w5k]

Hi Thank you so much for this awesome tutorial. I have a quick question about the data. For granger causality, you just used the raw, non-stationary data whereas you used the stationary data in VAR. Is there any criteria whether I can use stationary or non-stationary for different types of time series analysis (Granger causality, VAR, IRF, etc)? Thanks always :)

[@cleansquirrel2084]

Another beautiful video!!

[@ritvikmath]

Thanks!

[@victorgonzalezreyes16]

Do we need all p-values to be less than 5% to consider there is granger causality? What if 3 out of 4 p-values are less than 5%?

[@ResilientFighter]

Curious to know if there is any sort of regularization that can be applied to SARIMA and how that would be interpreted.

[@Juan-Hdez]

Very useful. Thank you!

[@programmingwithjackchew903]

for grander test do we need to make sure the data is stationary and free of seasonality?

[@FluffyTashiLai]

how do I input my own data? Do I name t1 = TSLA or t2 = RAY.. something like that?

[@ফকিরতালিব]

kzbin.info/www/bejne/jpTdZJSkmt2sja8

[@qiguosun129]

Great video and clear codes!

[@izzatii]

Thank you so much so the video. If you don’t mind, can you show us how to do the augmented toda yamamoto granger causality test using python?

[@ca20041]

I ran the code few times and for few runs p_value at lag =1 is < 0.05, could it be possible ? because by construction you shifted the time series by 3, so I expected to need at least lag=3 to t1 Granger causes t2.

[@ফকিরতালিব]

I see the same thing. I get the second lag significant at 2% or lower levels in almost all runs. Ritvik did not run the code in the video which does not make me confident. He uses an AR coefficient of 0.5 but he did not either explain how a larger or smaller value may affects the outcome.

[@user-or7ji5hv8y]

Great video

[@ritvikmath]

Thanks!

[@alvinanil]

How is GC different from the cross correlation?

[@RaymondPeckIII]

So... if you do a Vector Auto Regression you get Granger Causality essentially for free. No?

[@camotsuchuoi]

awesome, thank you

[@ritvikmath]

You're welcome!

[@hamedmajidian4451]

Where the code is?

[@giorgosmarinos7205]

Awsome job man!

[@ritvikmath]

Hey thanks!!