Hello all,
I'm currently undertaking a SCM study looking at the impact of an institution's framework change on inflation using the generic synth and synth-runner packages.
Issue is, I can't find any sources that talk about some practical issues like whether the data (outcome and predictors) needs to be stationary, will cointegration between outcome variable and predictors cause an issue (and would this be mitigated by using all the variables at their first-difference? I think practically it should be okay to use first differences and then generate the values for the synthetic pre and post-treatment and then back these onto the original data to create the canonical SCM charts.
Just a bit unsure about all of it considering lots of authors including Abadie et al. use variables that must be non-stationary (e.g. GDP) but there's very little talk of this, as well as the relationship between the outcome variable and predictors.
Any help from you kind folks would be greatly appreciated.
I'm currently undertaking a SCM study looking at the impact of an institution's framework change on inflation using the generic synth and synth-runner packages.
Issue is, I can't find any sources that talk about some practical issues like whether the data (outcome and predictors) needs to be stationary, will cointegration between outcome variable and predictors cause an issue (and would this be mitigated by using all the variables at their first-difference? I think practically it should be okay to use first differences and then generate the values for the synthetic pre and post-treatment and then back these onto the original data to create the canonical SCM charts.
Just a bit unsure about all of it considering lots of authors including Abadie et al. use variables that must be non-stationary (e.g. GDP) but there's very little talk of this, as well as the relationship between the outcome variable and predictors.
Any help from you kind folks would be greatly appreciated.
Comment