Here is how I test contagion effect between six FX series (deleted two due to VAR specification).
I first split the whole period into two subperiods exogenously suggested by Forbes Rigobon. By observation of the data, I decided to split the data by 2009 week 35.
I start by specifying the correct VAR. In order to find the correct form of VAR, unit root tests are performed and four I(1) and two I(2) processes are found. After attempts are made using both or either of the difference of I(2) processes. Correct VAR can’t be specified at given optimal information criteria (results are not shown here.) Thus the four I(1) process are discussed here and correct model is specified as VAR(3). The LM test for the model VAR(3) is passed.
(1)My questions is that why FR test we accept all cases while testing coskewness we reject all?
(2)Is my logic in specifying the VAR correct?
(3)Am I correct to say that the series USDGBP to USDJPY have lower correlation but not significant in FR test maybe you have excessive interdependence or maybe there is nothing has changed.
Thanks in advance!