You are on page 1of 4

Case: Testing for Granger Causality in the data

Your preliminary tests can be based on global daily stock index data between 940201 - 961231 (see globedat.xls in the directory
We present a set of Granger causality tests in the report factorgr . The Granger test is illustrated on the next work sheet.

Case: (1) Study the background and interpretation of Granger causality as presented in the report factorgr .
(2) Study the theory of Granger causality as presented in key literature.
(3) Using the data provided, carry out causality tests between two market indices of your own choice,
recognizing possible unit roots and cointegration.
(4) Test for possible Granger causality between the two series, recognizing possible cointegration between them (check th
(5) Report your results and findings in a written paper, following common practice in scientific reporting.

ality in the data


940201 - 961231 (see globedat.xls in the directory data/ ).

ger test is illustrated on the next work sheet.

resented in the report factorgr .

ket indices of your own choice,

nizing possible cointegration between them (check the report factorgr ).


mon practice in scientific reporting.

Consider the Granger causality test E t -> F t, i.e., the issue whether the European stock market returns have on impact on the Finni
market return for the current day (cf. the report factorgr , table 5.1). Formally, we want to test whether b <> 0 in the below regressio

Ft =Ft -1+b Et +t

The F -statistic for Granger causality has an asymptotic F -distribution with s and (T -s -r -1) degrees of freedom.
The test can be carried out using any OLS software. Below we show an example from a Matlab session (the Finnish market return a
three continental factors have been loaded in the matrix data having 752 observations from the database on this site).

y=data(2:752,1);
>> xr=[ones(751,1) data(1:751,1)];
>> xf = [xr data(2:752,2)];
>> br = inv(xr ' * xr) * xr ' * y;
>> bf = inv(xf ' * xf) * xf ' * y;
>> er = y-xr * br;
>> ef = y-xf * bf;
>> sser = er ' * er
sser =
1225.90
>> ssef = ef'*ef
ssef =
953.8711

Comments
The Finnish stock market return resides in column 1 of data()
We take the intercept + the lagged Finnish return as x -variables in the reduced mode
The European factor resides in column 2
the OLS-coefficients for the reduced model
the OLS-coefficients for the full model
residuals of reduced model
residuals of full model
sum of squares of reduced model

sum of squares of full model

>> T = 751;
>> r = 1;
The critical values for F (s,T-r-s-1) = F (1,748) are: 2.7123 ( 10%), 3.8539 (5%), 6.6689 (1%).
>> s = 1;
( SSEr - SSE f ) / s
>> F = ((sser-ssef)/s)/(ssef/(T-r-s-1))
The F -statistic for Granger-causality: F =
SSE f / ( T - r - s - 1 )
F=
Check the value reported in factorgr .
213.3251
>> xf = [xr data(2:752,3)];
>> bf = inv(xf ' * xf)*xf ' *y;
>> ef = y-xf * bf;
>> ssef = ef ' * ef
ssef =
1183.20
>> F = ((sser-ssef)/s)/(ssef/(T-r-s-1))
F=
27.0181

The Asian factor resides in column 3

Check the value reported in factorgr .

stock market returns have on impact on the Finnish stock


want to test whether b <> 0 in the below regression:

T -s -r -1) degrees of freedom.


rom a Matlab session (the Finnish market return and the
ions from the database on this site).

ides in column 1 of data()


Finnish return as x -variables in the reduced model.

( 10%), 3.8539 (5%), 6.6689 (1%).

F=

( SSEr - SSE f ) / s
SSE f / ( T - r - s - 1 )

You might also like