Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Solving simultaneous causality in panel data regression

    Hello,

    For my Bachelors thesis I am running a regression on panel data consisting of 21 countries (id), with data over 60 quarters (time).
    I am measuring the effect on total trade between two parties, so my dependent variable Y is defined as (import+export).
    I am using 5 dependent variables, which vary over time and over country, those being:
    1. GDP
    2. CPI
    3. NEER (nominal effective exchange rate)
    4. exchange rate (formulated by S in the equation picture)
    5. FTA (this is a dummy variable wether or not the two parties have a Free Trade Agreement)

    I made my own time dummies (varying over time but not over country) and my own trend dummies (varying over time and country) and added those in my regression too to correct for any time or trend specific effects.
    Then I used a fixed effects (after running a Hausman test) panel data regression with robust standard errors (to correct for heteroskedasticity and serial correlation).

    As I found out yesterday, some of my independent variables are simultaneous causal with my dependent variable. For example, as total trade (Y) is dependent on GDP (X1), GDP is also dependent on total trade.
    I did some research on how to solve this and the most common answer I find is to use lagged variables. As this is beyond the scope of my studies (in which we only covered very basic Stata and econometrics), I am a bit unsure of how and which method to use. More specifically, I found the Arellano Bond method which proposes an IV-estimator to solve for this causality. They propose to take lagged effects and differences of the dependent and independent variables to correct for this.

    By reading articles I found out I have to use the Arellano Bond method (xtabond) and then use GMM-style instruments for the endogenous variables, so in my case for example total trade and GDP. Then I would have to specify the number of lags to use, and here is where I loose track.
    Is there any test which I can run to see how many lags I would have to use?
    Is there also a test which I can run to see which variables are exogenous and which are causal with the dependent variable, or do I have to find this out with theoretical research?
    Then lastly, is the one-step GMM instruments enough in my case or should I use the two-step version?

    I hope anyone can help me out, if that is by advising me some literature/articles to read which explain the A-B method in an understandable way I would appreciate that too.

    (I tried for a long time to add the regression equation in this question but I couldn't find how, so instead I added is as a jpg, any tips on this are welcome too).




Working...
X