5
$\begingroup$

I am using this R code:

library(lmtest)

TS <- read.table('Test.txt', sep='\t', header=TRUE)

grangertest(X ~ Y, order = 3, data = TS)

grangertest(Y ~ X, order = 3, data = TS)

I am getting these results:

Granger causality test

Model 1: X ~ Lags(X, 1:3) + Lags(Y, 1:3)
Model 2: X ~ Lags(X, 1:3)
  Res.Df Df      F    Pr(>F)    
1    637                        
2    640 -3 11.032 4.546e-07 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
> 
> grangertest(Y ~ X, order = 3, data = TS)  
Granger causality test

Model 1: Y ~ Lags(Y, 1:3) + Lags(X, 1:3)
Model 2: Y ~ Lags(Y, 1:3)
  Res.Df Df      F  Pr(>F)  
1    637                    
2    640 -3 3.5013 0.01527 *
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Could someone please help me to interpret the results? Let us say my level of significance is 0.05. It appears that I can reject the null hypothesis of no Granger causality. Does this meas that there is evidence that X and Y above are linked/influence each other? Is there any directionality? A nice blog about the granger test can be found here btw.

$\endgroup$

2 Answers 2

2
$\begingroup$

The answer is more or less already in the link you posted: The Granger Causality test fits a VAR model $$\begin{pmatrix} y_t\\ x_t \end{pmatrix}= \begin{pmatrix} \nu_y\\ \nu_x \end{pmatrix}+\begin{pmatrix} \alpha_{11,1}&\alpha_{12,1}\\ \alpha_{11,1}&\alpha_{12,1} \end{pmatrix}\begin{pmatrix} y_{t-1}\\ x_{t-1} \end{pmatrix}+... + \begin{pmatrix} \alpha_{11,p}&\alpha_{12,p}\\ \alpha_{11,p}&\alpha_{12,p} \end{pmatrix} \begin{pmatrix} y_{t-p}\\ x_{t-p} \end{pmatrix}+\begin{pmatrix} u_y\\ u_x \end{pmatrix} $$

and tests the NULL Hypothesis that X does NOT Granger cause Y, i.e. that the coefficients of the lags of X are not significant. $$ H_0:\alpha_{12,i}=0 \text{ for } i=1,...,p$$

The p-value you get comes from a Wald test. If it is lower than your significance level (in this case 0.05) you can reject the NULL, so X Granger causes Y. To see whether there is also Granger causality in the other direction you'd have to perfom a Wald test to check if $ \alpha_{11,i}=0\text{ for } i=1,...,p$. (which you did)
In your case both directions suggest Granger causality. The term causality is sometimes missleading, and could rather be thought as "predicatibility".
In your case: past values of x improve prediction of y (compared to only past values of y) and vice versa. There could also be a third variable z that influences both x and y.

$\endgroup$
0
0
$\begingroup$

Basically, if you see the first test, ( X ~ Lags(X, 1:3) + Lags(Y, 1:3) )you are proving if lags 1,2,3 of variable Y explain a X. Because you prove 3 coefficients, you use jont F test. The Pv is 4.546e-07, is under 0.05 then you can say that any Y lags explain X.

The second test have the same conclusion, in this case any X´lags explain Y.

Finally you have bidirectional causality.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.