Granger Causality Statistical Test for Time Series

#datascience #machinelearning #timeseries
The Granger causality test is a statistical hypothesis test for determining whether one time series is useful in forecasting another
Idea is to measure the ability to predict the future values of a time series using prior values of another time series
When time series X Granger-causes time series Y, the patterns in X are approximately repeated in Y after some time lag. Thus, past values of X can be used for the prediction of future values of Y.
A time series X is said to Granger-cause Y if it can be shown, usually through a series of t-tests and F-tests on lagged values of X (and with lagged values of Y also included), that those X values provide statistically significant information about future values of Y

[@SkyRiderJavelin]

Well presented content clear and concise, Thanks for posting

[@jaypdevaney]

Another great video, thanks!

[@siddhant17khare]

Hello Sir,
Thanks a lot for providing this wonderful demonstration!
Just a small query : should both the variables (x and y ) be made stationary before applying Granger causality (of x on y) Or only making the x variable stationary would suffice the need ?

[@AIEngineeringLife]

In time series y is a feature as well as y used for forecasting. So ideally if you see that way it is target as well

[@siddhant17khare]

@@AIEngineeringLife Another related question Sir -
a) When a t-test at lag 2 gets evaluated for Granger causality, then I guess the p-value is only for that specific lag (i.e 2) w.r.t the target variable, isn't it?
b) However, when an F-test gets evaluated for Granger causality at lag =2 , does it mean that the p values is for both lag 1 and lag 2 put together in the equation? (since it is F-Test)
c) Also Sir, rejecting the null hypothesis based on p value of F-Test (for lag=2) only suggests that atleast one of the two coefficients of lags is not equal to 0 . Therefore, ideally shouldn't we also look at t-test results of Granger causality once we have ascertained the p-value of F-Test, in order to conclude which lag(either 1 or 2 or both) play a role in predicting the target variable or not? Because I feel this method of Granger causality is no different from how we interpret OLS model.
Please let me know your suggestions.

[@dimplechutani2768]

Hi Sir , Have you uploaded your code notebooks somewhere ?

[@sriadityab4794]

Hi. Your lecturers are really helping me. Thanks a lot.
I have a question on this topic. Can we use this test to select the features for an LSTM model?

[@AIEngineeringLife]

Yes you can do that if you have lot of features. But again any statistical test we have to take it with some caution and not rely completely on it

[@sriadityab4794]

@@AIEngineeringLife Thanks. You mean to say results given by the test to be considered from a domain perspective?

[@drmearajuddin2334]

If data is stationary is at first difference.. Shall. We have to run granger causality test on differenced data or level data?

[@hanooltari]

Are the ‘p values’ the same as ‘granger causality magnitude’?

[@siddarama2284]

Could you please share me the code for this video?

[@AIEngineeringLife]

Here it is - github.com/srivatsan88/End-to-End-Time-Series/blob/master/Granger_Causality_Test_For_usefulness_of_TimeSeries_in_Forecasting.ipynb

[@bmebri1]

Right...